**METHODS** **Participants** Twenty-six healthy volunteers (20 male, 6 female) aged between 19 and 42 years (M = 25.2, SD = 6.8) participated in this experiment. All participants were right‐handed, reported no psychiatric or neurological disorders, no current use of any psychoactive medication and no previous experience with a fear-conditioning procedure. The ethical committee of the University of Amsterdam approved the study protocol (2012-KP-2346). All participants signed informed consent and received financial compensation for their participation. **Experimental design** The experimental design consisted of two conditions: fear conditioning with and without acoustic startle probes (Figure 1). Four different pictures served as a to-be conditioned stimulus (CS), two of which were paired with an acoustic startle probe on every presentation (CSProbe) and the other two were never paired with a startle probe (CSNo probe ). One of the CSProbe stimuli (CS+Probe) was, in addition to the startle probe, also paired with the unconditioned stimulus (US) and the other was never paired with the US (CS-Probe). Likewise, of the two CSNo probe stimuli, one was paired with the US (CS+No probe) and the other was never paired with the US (CS-No probe). **Stimuli and conditioning procedure** The experiment consisted of one session of fMRI scanning during which we used a classical fear-conditioning paradigm, with delay conditioning and partial reinforcement. Four pictures of neutral faces served as CSs. The images (obtained from Shutterstock, Inc.) were selected according to age (adult vs. child) and sex (male vs. female). All faces were cropped, scaled to equal height, and separated from their background using Adobe Photoshop software (CS5). Images were then placed on a gray background and adjusted on mean brightness. Two pictures were followed by an acoustic startle probe (100dB, 40 ms burst of broadband noise administered binaurally through headphones) on every presentation (CSsProbe) and the other two were never followed by the probe (CSsNo probe). To reduce initial startle reactivity and to familiarize participants with the pictures, fear acquisition was preceded by a habituation phase consisting of six acoustic startle probes presented alone (i.e., noise alone (NA) and a pre‐exposure phase consisting of the first three presentations of each of the four to-be CSs (CS+Probe, CS-Probe, CS+No probe and CS-No probe) and the NA. During the pre‐exposure phase, acoustic startle probes, but not yet the USs were administered. Then, throughout fear acquisition, each CS was presented 13 times, with 46% of the CS+ presentations being reinforced with the US. Again, CSProbe trials, but not CSNo Probe trials, co-terminated with the presentation of the acoustic startle probe. In addition, 13 NA acoustic startle probes were presented to obtain baseline startle responses. An electrical shock served as the US and was delivered twice for 2 ms, with a delay of 300 ms, by a Digitimer DS7A through MRI-compatible carbon electrodes attached to the participant’s right wrist. The intensity of the electrical shock was individually adapted at a level that the participant described as very uncomfortable but not painful (intensity range 8 –70 mA, M = 25, SD = 14.8). The other two stimuli were never reinforced (CS-Probe and CS-No probe). The two CSs+, as well as the two CSs‐, were opposites (i.e., male adult and female child, or female adult and male child, etc.; Figure 1) in order to counteract generalization on basis of sex or age. Assignment of the images as CS+, CS‐ and Probe, No probe was counterbalanced across participants. Visual stimuli were backward-projected onto a screen that was viewed through a mirror attached to the head coil. Participants were further instructed to look carefully at the images, as some of the images would be followed by the electrical shock. Duration of CS presentation was 8 s. The acoustic startle probe was presented 7 s after CS onset and the US was presented 7.5 s after CS onset (Figure 1). The relatively long interval between CS onset and acoustic startle probe and US was used to minimize their influence on the neuroimaging and pupil dilation data. The inter-stimulus- interval (ISI) varied between 12, 15, and 18 s (M = 15 s). Trial order and ISIs were semi‐random, such that no more than two consecutive trials or ISIs were of the same type. After scanning, awareness of the CS-US contingency was first assessed verbally. Thereafter, participants gave written retrospective US-expectancy ratings and evaluated the US and startle probe. ![*Figure 1* Task design showing the 4 conditions (CS+Probe, CS-Probe, CS+No probe and CS-No probe) and the timing of the acoustic startle probe and US onsets][1] [1]: http://osf.io/ja39v/files/designfig.png **MEASURES** **Fear-potentiated startle reflex** The conditioned fear response was measured as potentiation of the acoustic startle reflex. The eye blink component of the reflex was acquired through electromyography (EMG) measurements of the right orbicularis oculi muscle33. We used disposable MR compatible carbon electrodes (Kendall H135TSG) with a vinyl tape backing. To prevent possible warming of the electrodes in the scanner we used carbon electrode leads with carbon current-limiting resistors serially connected between the lead and electrode34. Electrode leads were twisted to minimize gradient artefacts on the EMG recordings and connected to an MRI‐compatible amplifier (Geodesic EEG system 300 MR, EGI, USA) that was placed outside the scanner bore and grounded through a RF filter. Data were sampled at 1000 Hz and recorded using NetStation software (version 4.5.2, Electrical Geodesics, Inc. (EGI), Eugene, USA). We tried to offline correct the raw EMG signal for scanner artefacts. Multiple different filter options from NetStation software and Brain Vision Analyzer software (version 1.05, Brain Products GmbH, Munich, Germany) were applied. Scanner artefacts were removed by subtracting an artefact template from the EMG data using a sliding average of 2, 5, 10, 15, 20 and 25 consecutive volumes. In addition we removed the scanner artefact with all other available options in NetStation; by applying Gaussian removal, averaging over all volumes, and calculating an exponential average with a 0.5, 0.10 or 0.15 smoothing factor. Every filter successfully removed the MRI-artefact from the EMG data, however none was able to distinguish between FPS signal and MRI-artefact and therefore the FPS signal was removed from the data together with the MRI-artefact and only a low frequency (<28 Hz) signal component of the FPS signal remained, which likely reflects a motion artefact 33. **Pupil dilation** Pupil dilation responses were recorded continuously using a non-ferromagnetic eye tracker with fiber optic camera upgrade (EyeLink 1000, SR Research Ltd., Mississauga, Ontario, Canada). Data were sampled at 250 Hz. Subsequently, pupil dilation responses were calculated as the peak in pupil diameter during stimulus presentation (i.e., 0‐6.5 s after stimulus onset) from baseline (i.e., 500 ms prior to stimulus onset). Data samples that were obscured by eye blinks were discarded. Furthermore, trials containing substantial signal loss, affecting more than 50% of the baseline or the stimulus presentation, were discarded (M = 3.1%, SD = 4.2%) and these trials were replaced using the linear-trend-at‐point method. However, if more than 25% of the trials within one condition had to be discarded, the participant was excluded from pupil dilation analysis. Finally, pupil dilation responses were converted to t-scores within participants. **Neuroimaging** Imaging was conducted using a 3T MRI scanner (Philips, Achieva XT) with a 32-channel head‐coil. Whole‐brain functional MRI images were acquired: GE‐EPI, TR = 2000 ms, TE = 27.6 ms, FA = 76.1°, FOV = 240 mm; matrix = 80 x 80; slice thickness = 3 mm; 37 axial slices sequentially acquired). Additionally, a T1-weighted anatomical image was obtained for each participant (TR = 8.2 ms; TE = 3.7 ms; FA = 8°; FOV = 240 x 188 mm; matrix = 240 x 240; slice thickness = 1 mm; 220 axial slices sequentially acquired). **Retrospective US-expectancy and aversion ratings of US and acoustic startle probe** Participants retrospectively rated the likelihood of US delivery for each stimulus type for the beginning (pre‐exposure), mid (early acquisition), and end (late acquisition) of the fear conditioning task on an 11‐point Likert scale (‐5: certainly not, 0: maybe, and 5: certainly). In addition, the US and acoustic startle probe were evaluated for averseness by means of the arousal and valence dimensions of the Self-Assessment Manikin35. Participants rated the stimuli on the pictorial dimensions of arousal and valence on a 9-point Likert scale (1: excited and positive, 9: calm and negative). **ANALYSIS** **Pupil dilation and retrospective US-expectancy ratings** Pre‐exposure (trials 1-3) was compared to late acquisition (trials 10‐16). Accordingly, pupil dilation responses and US-expectancy ratings were separately computed for the pre‐exposure and late acquisition phase. To examine fear acquisition, we analyzed the differential increase in responding by means of repeated measures ANOVA (SPSS, version 23) with Stimulus (CS+ vs. CS‐) and Trial (pre-exposure vs. late acquisition) as within-subjects variables. Subsequently, to statistically examine the effect of acoustic startle probes on physiological and behavioral (i.e., pupil dilation, US‐expectancy ratings) fear learning, we used repeated measures ANOVAs with Stimulus (CS+ vs. CS‐), Trial (pre-exposure vs. late acquisition), and Condition (Probe vs. No probe) as within-subjects variables and with a significance level of p < .05. Where appropriate, Greenhouse-Geisser corrections were applied to control for the violation of the sphericity assumption. In addition, arousal and valence ratings of startle probes and US were analyzed using paired samples t-tests (two-tailed). Post-hoc t-tests were tested one‐tailed with a significance level of p < .05. **Neuroimaging** Imaging data were processed and analyzed using FSL (FMRIB’s Software Library: www.fmrib.ox.ac.uk/fsl) software. First, functional images were motion corrected (MCFLIRT)36, spatially smoothed (5-mm full‐width‐at‐half‐ maximum Gaussian kernel), and high‐pass filtered (cutoff = 100 s). Structural images were brain extracted (BED)37. Subsequently, for each participant, functional images were aligned to the structural image and transformed, on the basis of this structural image, to standard space (MNI) using a 7-degree‐of‐freedom affine registration followed by linear warping. Thereafter, functional MRI data were analyzed using general linear models. Stimulus onsets for the four conditions (CS+Probe, CS‐Probe, CS+No Probe and CS‐No Probe) were modelled as double Gamma functions with a duration of 0 ms, leaving a time window of 7 s until onset of the startle probe in the Probe condition and 7.5 s until the onset of the US in the reinforced CS+ trials. Furthermore, the USs and acoustic startle probes (as well as all temporal derivatives and motion parameters in six directions) were included in the model as regressors of no interest to ensure estimation of the variance uniquely explained by the CSs. Contrasts for the four conditions versus baseline were specified on the first single-subject level. Then, on the second single-subject level, the main effect of Stimulus (CS+/CS-) across the Probe and No probe condition (CS+Probe and CS+No Probe > CS‐Probe and CS‐No Probe), the main effect of Condition (Probe/No probe) across CS+ and CS- (CS+Probe and CS‐Probe <> CS+No Probe and CS‐No Probe), and the interaction between Stimulus and Condition were specified. Voxel-wise statistical tests were family-wise error rate corrected for multiple comparisons (p < 0.05) for the whole brain or the ROIs using threshold free cluster enhancement (TFCE)38 with 5000 permutations. The six ROIs were defined on the basis of a recent meta-analysis of fear-conditioning studies16 and included the dACC, insula, ventral striatum, thalamus and midbrain/dorsal pons. In addition, we included the amygdala because of its hypothesized involvement in fear-potentiated startle4,8. The subcortical regions (ventral striatum, amygdala and thalamus) were anatomically defined using the Harvard‐Oxford subcortical structural atlas as implemented in FSL (probability ≥ 25%). The cortical regions (dACC and insula) and midbrain/dorsal pons were created centered around peak activation coordinates obtained from the literature. We used 15‐mm spheres for dACC left/right: x = 8/-10, y = 18/6, z = 42/44 and for insula left/right: x = ‐40/40, y = 18/16, z = -2/2 16. The midbrain/dorsal pons was defined as two box masks (10 x 10 x 22 mm) centered around the periaqueductal grey (PAG) x = 0, y = -30, z = -1239 and the LC x = 2, y = -38, z = -2840. All coordinates are reported in MNI standard space. We performed additional ROI analyses to obtain non-biased effect sizes. Therefore, we extracted the mean z-value on single subject level from the individual ROIs (FSL Featquery) and conducted repeated measures ANOVA (SPSS, version 23) to calculate partial eta-squared. For both types of fMRI analyses, we used Bonferroni correction for six ROIs, resulting in an adjusted significance level of .0083.