Eighty-eight healthy, TMS-naïve volunteers (43 men) were recruited from a large sample collected at a Collaborative Research Center (SFB-TRR 58) of the Universities in Münster, Würzburg and Hamburg, Germany, as well as internet announcements. They were screened for current mental health and right-handedness by using the Mini International Neuropsychiatric Interview (M.I.N.I., Sheehan et al., 1998) and the Edinburgh Inventory (Oldfield, 1971). All female volunteers were additionally screened for a regular menstrual cycle and the non-usage of any hormonal contraceptives for at least 3 months prior to measurement. In order to account for facilitating effects of estrogen on extinction learning (Glover et al., 2012) and extinction recall (Milad et al., 2010; Zeidan et al., 2011), women only participated in the experiment during their early follicular phase (defined as the first 5 days of a 28-day cycle) when estradiol and progesterone levels are low. Contraindications regarding the TMS safety guidelines (Wassermann, 1998) such as epilepsy, use of pacemakers or pregnancy were assured. Participants gave written informed consent in accordance with the Declaration of Helsinki in its latest version from 2008. All procedures were approved by the ethics committee of the University of Würzburg.

Three female subjects dropped out due to the experience of discomfort while receiving TMS application and were thus not considered for further data analysis. Demographic data of the remaining N = 85 participants are presented in Table 1. None of the reported variables reached statistical significance for group comparisons between active and sham TMS (student's t-test, p > 0.05). Group differences can therefore be interpreted in terms of TMS effects.

The paradigm consisted of four phases divided into familiarization, fear acquisition and extinction learning on day 1 and a test for extinction recall on day 2 (see Figure 1). Two male neutral faces served as conditioned stimuli (CS; Tottenham et al., 2009) and an aversive scream of 95 dB served as unconditioned stimulus (UCS; IADS, Bradley and Lang, 1999). Volunteers were first familiarized with both CS by presenting each face eight times without the UCS. During the following fear acquisition phase consisting of 32 CS presentations one neutral face (CS+) was randomly followed by the UCS in 50% of trials whereas the other face (CS−) never preceded the UCS. Both extinction phases (day 1 and 2) consisted of 40 trials in total (20 CS+, 20 CS−) without UCS presentations. CS stimuli were presented for 6000 ms duration separated by jittered inter trial intervals (ITI) of 5000–8000 ms displaying a fixation cross. The UCS lasted 1380 ms and followed CS+ offset after a jittered temporal gap of 0–1000 ms (Guhn et al., 2012). The assignment of CS+ and CS− was counterbalanced across subjects and stimuli were presented in a pseudo-randomized order such that maximally three similar faces followed each other. Presentation® version 12.2 software (Neurobehavioral Systems, Inc., Albany, CA, USA) was used for presenting the paradigm.

Following the fear acquisition phase, subjects received one offline session of either active or placebo (sham) rTMS prior to performing extinction learning on day 1. Stimulation was applied via a round coil (MMC-140 Parabolic) of a Medtronic MagPro X100 stimulator (Medtronic MagPro, Düsseldorf, Germany) to a cluster within the medial prefrontal cortex. The coil was positioned in the middle of the cluster which was identified by marking channel 26 of the NIRS probe set corresponding to the MNI coordinates x = 14.5, y = 68.3, z = 21.3 (according to http://www.jichi.ac.jp/brainlab/virtual_registration/Result3x11_E.html. This channel represents the center of the mPFC activation cluster for which we found an increase in oxygenated hemoglobin concentration over the time course of extinction learning in a prior study (Guhn et al., 2012). Inter-subject variance was considered by assigning Fpz according to the 10-20 EEG system (Jasper, 1958). Emanating from Fpz, channel 26 was marked resulting in slightly varying positions for TMS coil positioning based on the participants individual head sizes. The upper edge of the coil was tilted 2 cm away from the scalp in order not to stimulate the premotor cortex; the handle of the coil was pointed upwards. The rTMS protocol was adapted from Baeken et al. [2010; stimulation intensity of 110% of the individual resting motor threshold (RMT), 10 Hz stimulation frequency, 40 trains of 4 s duration (1560 pulses), inter train intervals of 26 s], who found an amygdala attenuation in response to negative stimuli after one rTMS session. For the present study, this protocol was selected corroborating the intention that it should impact the fear circuit in the same way, i.e., the proposed increased prefrontal top–down modulation of subcortical systems, in particular the amygdala. Sham rTMS was applied using a placebo coil (MC-P-B70 Placebo) which appeared similar in placement and acoustic properties to the active coil but had a magnetic shield embedded limiting the amount of the magnetic field. In order to control for the proposed facilitatory effects of active rTMS, fNIRS was used to monitor blood oxygenation as an index of functional brain activity in the mPFC directly following the stimulation, i.e., during extinction learning, and during extinction recall on day 2 (see below). The TMS protocol and the subsequent attachment of the NIRS probeset resulted in a time lag of approximately 25 min between the fear acquisition and extinction learning phase.

On the day of stimulation (day 1), subjects were first familiarized with the experimental design and asked to answer questionnaires concerning mood (Positive and Negative Affect Scale, PANAS; Krohne et al., 1996) and anxiety (State-Trait Anxiety Inventory, STAI; Laux et al., 1981). Subsequently, they were introduced to the TMS machine by identifying the individual RMT defined as the lowest stimulation intensity capable of inducing a visible finger movement at least 5 times out of 10 single pulses over the right hand area of the primary motor cortex. TMS application and all measurements were conducted in a sound-attenuated, electrically shielded and air-conditioned cabin. Subjects were prepared for the experiment by attaching headphones and electrodes for startle potentiation and skin conductance recordings (see below). They were instructed about the separation of the experiment into three parts: (1) in the first half of the experiment they are confronted with two neutral faces on the computer screen as well as two auditory sounds (familiarization and fear acquisition), (2) subsequently the rTMS application to their forehead while sitting still on a chair, and (3) immediately after the stimulation the second half of the experiment again consisting of faces and auditory stimuli (extinction learning). Subjects were not instructed about the CS+/UCS contingency or the UCS absence during the extinction phase. At the end of day 1 the PANAS was assessed a second time to evaluate a potential rTMS impact on mood (Tupak et al., 2013).

On day 2, subjects had to answer a self-construed questionnaire concerning rTMS side effects based on Wassermann (1998) (“Did you experience any adverse side effects after the rTMS yesterday? If yes, please mark which kind of discomfort you experienced and how long it lasted.”). They were prepared for physiological recordings and underwent the test for extinction recall while the instruction resembled that of day 1. TMS was not applied a second time. Afterwards subjects were unblinded to the rTMS condition and were paid for participation.

Conditioned fear responses (CR) were assessed by FPS, SCR, fNIRS, and subjective valence and arousal ratings for CS+ and CS−.

The eyeblink component of the startle reflex was measured by recording electromyographic (EMG) activation of the right orbicularis oculi muscle. Two 5 mm Ag/AgCl disc surface electrodes were positioned approximately 1 cm below the pupil and 1 cm below the lateral canthus of the right eye (impedance <5 kΩ). A third electrode was placed at the right mastoid and served as isolated ground. The acoustic startle stimuli consisted of a 50 ms burst of white noise with 40 ms plateau and 5 ms rise and fall time at intensities of 100 dB (sound pressure level, SPL) delivered binaurally via in-ear headphones. No background sound was presented. Startle probes were delivered in half of the trials (4000 ms after CS onset) and ITI (randomly between 3000 and 5000 ms). EMG activity was recorded via a 72-channel amplifier (QuickAmp, Brain Products GmbH, Munich, Germany) and sampled at 1000 Hz. Data was acquired, saved and analyzed with Vision Recorder/Analyzer Version 2.0 (Brain Products GmbH, Munich, Germany). The EMG-signal was filtered with a 28 Hz high-pass and a 500 Hz low-pass filter (time constant 0.0057 s, 24 dB per octave). A notch filter was applied to control for components caused by (electro-)magnetic interference. After rectification signals were smoothed using a 50 ms moving average filter. Each segment was baseline-corrected 50 ms prior to the startle probe onset. Startle amplitudes were further defined as peak magnitudes (in microvolt) from the corrected EMG signal between 21 and 200 ms following probe onset. Artifact rejection was performed manually for every single peak. Startle non-responders on either one or both days were identified by mean magnitudes of less than 5 μV per day and excluded accordingly (n = 14). Another male subject had to be excluded due to a nystagmus, which made startle blink recording impossible. In order to allow for inter-individual differences, absolute blink magnitudes were normalized using z-standardization (Blumenthal et al., 2005). ITI startle probes were further utilized as control condition for CS+ and CS− by converting startle magnitudes during each CS presentation (X) into Z scores using the ITI mean and standard deviation per phase (Z_CS = (X_CS – M_ITI)/SD_ITI); (e.g., Bonnet et al., 1995; Blumenthal et al., 2005).

SCR was assessed by using two Ag/AgCl electrodes attached to the thenar eminence of the subjects' left palm. Measurements were acquired via a 72-channel amplifier and a Galvanic Skin Response (GSR) sensor which constantly delivered a 0.5 V current (Brain Products GmbH, Munich, Germany). The sampling rate was set to 1000 Hz. SCR recording and analyses were performed with Vision Recorder/Analyzer Version 2.0 (Brain Products GmbH, Munich, Germany). Offline, raw data were first high-pass filtered with 1 Hz and a notch filter of 50 Hz and afterwards segmented into CS+ and CS− trials that were baseline-corrected 1000 ms prior to CS onset. SCR were characterized by peak responses in a time window of 1 to 5 s after CS onset. Artifact rejection was performed manually for every single trial. Similarly to the FPS analyses SCR data were z-transformed across both days without the first four respective CS trials in order to account for inter-individual differences. Six non-responders had to be excluded and were thus not considered for further analysis.

Functional NIRS is based on near-infrared light of different wave lengths that is emitted to the cortical surface by means of sensors attached to the participant's forehead and thereby measures local changes of blood oxygenation. A detailed description can be found elsewhere (Obrig and Villringer, 2003). Oxygenation concentration was measured with the continuous wave system ETG-4000 (Hitachi Medical Co., Tokyo, Japan) using a 3 × 11 array which covered the prefrontal cortex. The interoptode distance was set to 3 cm. Signals were acquired with a sampling rate of 10 Hz. The method was included in order to discuss FPS, SCR, and rating results in the light of rTMS induced mPFC activation within the targeted fNIRS channels. We hypothesized that if rTMS modulates the processing of conditioned fear, it will correlate with higher mPFC activation in the cluster for which we found a signal increase from early to late extinction learning in a previous study (Guhn et al., 2012). Accordingly, we time-locked the onset of the signal to the jitter mean, i.e., 6500 ms after CS onset, and manually screened for artifacts due to head movement or technical problems. Signals were further processed by applying a cosine filter of 0.5 Hz correcting for low-frequency signal drifts. The four regressors (CS+ early, CS+ late, CS− early, CS− late) were modeled as delta functions and convolved with a gaussian hemodynamic response function at 6.5 s peak time. Time series for blood oxygenation (O₂Hb) during both extinction sessions were then assessed by applying a general linear model approach. Beta estimates for stimulus (CS+, CS−) by phase mean (extinction learning early, extinction learning late, extinction recall early, extinction recall late) between groups (active, sham) were investigated by using repeated measures analyses of variance (ANOVA).

Subjective CS+ and CS− ratings were assessed through self-assessment manikins (SAM; Bradley and Lang, 1994) for valence and arousal at different time points during the experiment: after familiarization and twice during/after fear acquisition, as well as during/after both extinction sessions. Subjects were asked to indicate whether a face was perceived as pleasant or unpleasant and whether it induced arousal or not on a 9-point Likert Scale.

Demographic data such as age and years of education were compared between groups with student t-tests. Psychometric data (UCS-intensity, PANAS, and STAI scores) were analyzed by using the Mann–Whitney-U-test, rTMS side effects by using Fisher's Exact Probability Test.

For FPS and SCR analyses, subjects were first characterized by CS+ and CS− responses during the acquisition phase. We analyzed paired (CS-UCS) as well as unpaired (CS-noUCS) CS+ trials since UCS followed the CS with a short temporal gap, i.e., the analyzed segment did not include the actual UCS delivery. Subjects who did not show higher responses for CS+ than CS− were not considered for further TMS group comparisons due to non-successful fear conditioning (e.g., Phelps et al., 2004). Likewise 22 subjects (13 women) had to be excluded. Potential group differences on a descriptive or psychometric level (age, UCS-intensity, STAI-T, STAI-S; PANAS) were accounted for and did not reveal any significant results. CS trials were averaged for each stimulus (CS+, CS−) per phase (acquisition, extinction learning, extinction recall) and statistically evaluated using repeated measures ANOVA with stimulus and phase mean as within-subject factors and group (active, sham) as between-subject factor. A p-value < 0.05 was considered significant; Greenhouse Geisser correction was applied in case of non-sphericity. Post-hoc t-tests were used when (1) stimulus × phase × group interactions proved to be significant or (2) stimulus × phase interactions proved to be significant without significant group effects; in the second case post-hoc t-tests were conducted within groups. Additionally, we analyzed gender effects for FPS and SCR data and tested for significant interactions between gender and TMS group. A short theoretical background and discussion of these results is provided in the supplement.

The final sample consisted of n = 21 (13 women) subjects receiving active and n = 24 subjects (12 women) receiving sham stimulation. ANOVA revealed significant main effects for stimulus [F_{(1, 43)} = 15.35, p < 0.001], the interaction of stimulus × phase [F_{(1.5, 66.5)} = 5.7, p = 0.009] and a trend-wise significant stimulus × phase × group interaction [F_{(1.5, 66.5)} = 2.92, p = 0.074].

As expected, t-tests revealed significant differences between CS+ and CS− trials during acquisition within both groups (p < 0.001), but revealed sustained CS+/CS− discrimination for sham only, i.e., higher FPS responses for CS+ than for CS− for both extinction learning [t₍₂₃₎ = 2.3, p = 0.031] and extinction recall [t₍₂₃₎ = 2.44, p = 0.023; Figure 2]. CR for both groups in time course are provided in Figure 3. In order to statistically analyze these group differences during the experimental phases we continued to separate each extinction session into an early and a late phase consisting of 10 trials each for which we used the CS+/CS− differences. A one-way ANOVA examining the effects of phase (acquisition, early extinction learning day 1, late extinction learning day 1, early extinction recall day 2, late extinction recall day 2) on FPS magnitudes revealed a trend-wise significant main effect of phase for the active group [F_{(2.3, 46.7)} = 2.98, p = 0.054]. This is composed of a negative linear trend [F_{(1, 20)} = 4.19, p = 0.054]: FPS responses decreased proportionately through all phases while the sham group neither showed a significant main effect of phase (p > 0.79) nor significant trends. Figure 4 shows the time course of the difference scores (CS+ minus CS−) throughout the five phases.

The final sample for SCR analyses consisted of 47 subjects, n = 26 active (15 women) vs. n = 21 sham group (9 women). The three-way ANOVA revealed significant stimulus [F_{(1, 45)} = 26.28, p < 0.001], phase [F_{(1.6, 74)} = 7.62, p = 0.001] and stimulus × phase interaction effects [F_{(2, 90)} = 14.84, p < 0.001]. Group did not influence main or interaction effects (p > 0.1). Both groups showed successful discrimination during acquisition (p < 0.001). Notably, the sham group still showed the CS+/CS− discrimination sustained during extinction learning [t₍₂₀₎ = 2.11, p = 0.047] while the active group displayed no significant CS+/CS− differences (p > 0.9; Figure 5).

In order to keep the sample constant we examined self-reports only for subjects who were analyzed either for FPS or SCR data (n = 62, see Table 2). This sample did not differ from the non-conditioners (n = 23) in any of the assessed descriptive or psychometric measures.

All subjects indicated successful fear acquisition as evident from significant main effects for stimulus [valence: F_{(1, 60)} = 8.16, p = 0.006; arousal: F_{(1, 60)} = 27, p < 0.001], phase [valence: F_{(2.6, 157.7)} = 18.31, p < 0.001; arousal: F_{(3, 180)} = 42.22 p < 0.001] and significant stimulus × phase interactions [valence: F_{(2.3, 137.7)} = 11.8, p < 0.001; arousal: F_{(1.9, 114)} = 15.67, p < 0.001]. CS+ and CS− were equally evaluated during familiarization [valence: t₍₆₁₎ = 0.27, p = 0.790; arousal: t₍₆₁₎ = −0.18, p = 0.857] but self-reports diverged significantly during fear acquisition, in that CS+ was rated as more unpleasant [t₍₆₁₎ = 4.79, p < 0.001] and evoked higher arousal [t₍₆₁₎ = −5.57, p < 0.001] than CS−. This significant discrimination persisted over both extinction learning [valence: t₍₆₁₎ = 2.05, p = 0.044; arousal: t₍₆₁₎ = −4.76, p < 0.001] and extinction recall [valence: t₍₆₁₎ = 2.33, p = 0.023, arousal: t₍₆₁₎ = −4.19, p < 0.001] although CS+ valence increased [t₍₆₁₎ = −6.88, p < 0.001] and CS+ arousal decreased in the course from acquisition to extinction [t₍₆₁₎ = 7.67, p < 0.001] again resulting in familiarization-like levels (p > 0.1).

In order to account for rTMS induced group differences we conducted a three-way ANOVA examining effects of stimulus by phase with only two levels (extinction learning, extinction recall) between groups. We found a significant stimulus × phase × group interaction for arousal [F_{(1, 60)} = 4.33, p = 0.042]. The active group (n = 32) discriminated significantly less between CS+ and CS− while the sham group (n = 30) persisted to evaluate CS+ as more arousing than CS− [t_(53.8) = −2.01, p = 0.043] resembling the FPS and SCR results (Figure 6). Valence ratings revealed no group differences. The three-fold interaction did not reach statistical significance (p > 0.2) for the whole sample (N = 85), including conditioners and non-conditioners.

We neither found significant group differences during extinction learning nor during extinction recall in the sample of n = 62 which was used for the subjective ratings. Exploratorily we analyzed the subsample of volunteers fulfilling the requirements for the analysis of both FPS and SCR (n = 12 active and n = 13 sham, two data sets were not included into the analysis due to an insufficient signal quality) since those participants were believed to have the strongest conditioning response regarding the consistency across measurements. However, we are well aware that the results have to be regarded cautiously. For the cluster reported in our pilot study (10 medial prefrontal channels expanding to the right hemisphere: 5, 16, 24, 26, 27, 35, 36, 37, 45, 47) the active and sham group differed in the amount of O₂Hb in response to CS+ during the early extinction learning phase for which the active group displayed a higher signal than the sham group [student t-test: t₍₂₃₎ = 2.65, p_one−tailed = 0.008]. While there was no signal change from the early to the late phase in the active group, the sham group showed a trend-wise significant signal increase [t₍₂₃₎ = −1.61, p_one−tailed = 0.067] resembling the signal increase reported in the previous study. During the extinction recall on day 2 there were no within or between-group differences (see Figure 7).

Side effects were assessed using a questionnaire which contained previously published rTMS side effects such as headache, neck pain, dizziness, drowsiness, nausea, speech, or sleep problems, problems to concentrate, paraesthesia, seizures, muscle contraction, faint, local discomfort at the stimulated site and ear noise (Wassermann, 1998). Subjects were asked to evaluate these side effects in their intensity and duration before unblinding them regarding the TMS group. For completeness, the n = 3 females who dropped out due to rTMS discomfort were included in the analysis (N = 88). Overall, rTMS was well tolerated. Twenty-two subjects (25%) reported side effects, therefrom 10 subjects of the sham group. Type of side effects per group are depicted in Table 3, no other side effects were quoted. Headaches as the most prominent side effect lasted less than 1 h in 11 subjects; 5 subjects complained about headaches for less than 6 h and 2 for less than 12 h. There was neither a significant group difference concerning the overall frequency of side effects nor the type of side effects (p > 0.49), except for neck pain which was trend-wise quoted more frequently by the sham group (p = 0.056). Altogether, the results demonstrate that subjects were actually TMS-naïve.

Possible mood changes caused by rTMS were evaluated using PANAS × group repeated measures ANOVA. Positive affect showed a significant main effect [F_{(1, 86)} = 45.94, p < 0.001] indicating that subjects rated their affect prior to the experiment as more positive than afterwards. Negative affect did not change. The group interaction did not reach statistical significance, i.e., rTMS did neither induce negative nor positive mood changes (p > 0.1).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.