Research Articles, Behavioral/Cognitive

Ultrafast Cortical Gain Adaptation in the Human Brain by Trial-To-Trial Changes of Associative Strength in Fear Learning

Melissa Yuan, Tamara Giménez-Fernández, Constantino Méndez-Bértolo and Stephan Moratti
Journal of Neuroscience 19 September 2018, 38 (38) 8262-8276; DOI: https://doi.org/10.1523/JNEUROSCI.0977-18.2018

Abstract

In fear conditioning, more efficient sensory processing of a stimulus (the conditioned stimulus, CS) that has acquired motivational relevance by being paired with an aversive event (the unconditioned stimulus, US) has been associated with increased cortical gain in early sensory brain areas (Miskovic and Keil, 2012). Further, this sensory gain modulation related to short-term plasticity changes occurs independently of aware cognitive anticipation of the aversive US, pointing toward implicit learning mechanisms (Moratti and Keil, 2009). However, it is unknown how quickly the implicit learning of CS–US associations results in the adaptation of cortical gain. Here, using steady-state visually evoked fields derived from human Magnetoencephalography (MEG) recordings in two experiments (N = 33, 17 females and 16 males), we show that stimulus-driven neuromagnetic oscillatory activity increases and decreases quickly as a function of associative strength within three or four trials, as predicted by a computationally implemented Rescorla–Wagner model with the highest learning rate. These ultrafast cortical gain adaptations are restricted to early visual cortex using a delay fear conditioning procedure. Short interval (500 ms) trace conditioning resulted in the same ultrafast activity modulations by associative strength, but in a complex occipito-parieto-temporo-frontal network. Granger causal analysis revealed that reverberating top-down and bottom-up influences between anterior and posterior brain regions during trace conditioning characterized this network. Critically, in both delay and trace conditioning, ultrafast cortical gain modulations as a function of associative strength occurred independently of conscious US anticipation.

SIGNIFICANCE STATEMENT In ever-changing environments, learned associations between a cue and an aversive consequence must change under new stimulus–consequence contingencies to be adaptive. What predicts potential dangers now might be meaningless in the next situation. Predictive cues are prioritized, as reflected by increased sensory cortex activity for these cues. However, this modulation also must adapt to altered stimulus–consequence contingencies. Here, we show that human visual cortex activity can be modulated quickly according to ultrafast contingency changes within a few learning trials. This finding extends to frontal brain regions when the cue and the aversive event are separated in time. Critically, this ultrafast updating process occurs orthogonally to aware aversive outcome anticipation and therefore relies on unconscious implicit learning mechanisms.

Introduction

Humans have evolved to prioritize sensory information associated with possible negative consequences. In the realm of Pavlovian fear conditioning, electrophysiological and human imaging studies have shown that, as subjects learn the relationship between the conditioned stimulus (CS) and unconditioned stimulus (US), activation of the sensory cortex related to the fear relevant CS increases (Knight et al., 2004; Moratti and Keil, 2005; Miskovic and Keil, 2012; McTeague et al., 2015). To what extent this ability depends on conscious awareness of CS–US contingencies is still debated (Morris et al., 2001; Knight et al., 2003; Williams et al., 2004; Pessoa et al., 2005), but likely reflects enhanced cortical gain (also known as neuronal sensitivity; Cardin et al., 2008) due to short-term changes in local neural networks (Letzkus et al., 2011) and probably do not depend on a conscious representation of CS–US contingencies (Knight et al., 2003; Moratti et al., 2006; Moratti and Keil, 2009). However, the speed at which these changes can occur is unknown.

Moratti and Keil (2009) found that sensory cortex activity evoked by the CS is linearly related to the associative strength (the number of CS–US pairings), but independent of conscious US expectancy. The investigators separated the effects of associative strength and US expectancy by using the gambler's fallacy in a fear-conditioning paradigm first implemented by Perruchet (1985). The results of Moratti and Keil (2009) suggest that the cortical facilitation of fear cue processing is determined by previous exposure to learning contingencies rather than by cognitive anticipation of the US. However, it was not evaluated how quickly the sensory gain in early visual cortex adapted to the previous learning experience. Therefore, in the present study, we recorded trial-by-trial fluctuations of visual cortex activity during a delay fear-conditioning paradigm inducing a gambler's fallacy on US expectancy to dissociate associative strength and expectancy ratings. Single-trial visual cortex activity was fitted based on the learning history using a computational implementation of the Rescorla–Wagner (RW) model (Wagner and Rescorla, 1972; Thorwart et al., 2009) and US expectancy ratings. We hypothesize that short-term cortical plasticity in early visual cortex changes quickly across trials as predicted by a RW model with the highest learning rate and will vary orthogonally to US expectancy (Moratti and Keil, 2009).

Previous fMRI research has provided evidence that delay and trace conditioning are different phenomena that may be driven by distinct brain regions (Knight et al., 2004; Cheng et al., 2008). Critically, delay and trace conditioning have been associated with nondeclarative and declarative memory processes, respectively, whereby trace conditioning implicates the hippocampus along with frontoparietal brain structures (Knight et al., 2004; Cheng et al., 2008; Haritha et al., 2013; Harnett et al., 2016). However, it has been demonstrated in humans that trace conditioning involves the hippocampus and associated declarative memory processes only when the trace interval is ≥1000 ms; shorter trace intervals still allow for implicit learning (Woodruff-Pak, 1993; Gerwig et al., 2008).

Because frontal areas and their top-down influences on sensory areas become critical in the representation of task contingencies when events are separated in time (Miller and Cohen, 2001), it is of interest to determine whether frontal areas can also modulate their activity as fast as sensory areas based on prior learning experience rather than on conscious US expectancy. We hypothesize that trace-conditioned ultrafast trial-by-trial activity changes in higher-order brain areas as a function of associative strength orthogonal to US expectancy can be observed using short trace intervals. Further, these higher-order brain areas should exert top-down influences on sensory areas to bridge the temporal gap (Fuster, 1990). Therefore, the same experiment as described above was repeated, but with a temporal gap (500 ms) between CS offset and US onset introduced to assess fast cortical gain modulations by associative strength or US expectancy on a trial-by-trial basis in short interval trace conditioning.

Materials and Methods

Experiment 1 (delay fear conditioning)

Participants.

After giving written informed consent, 20 right-handed subjects (11 females) participated in the delay-conditioning paradigm (Experiment 1). Three subjects had to be excluded due to excessive noise in the MEG data. Therefore, the total sample size consisted of 17 volunteers (11 females). The mean age of the sample was 24.8 years (range 19–32). All subjects had normal or corrected to normal vision and no family history of epilepsy. Subjects received 20€ or class credit for participation. The local ethics committee approved the study following the Declaration of Helsinki.

Based on previous research (Moratti et al., 2017), the effect size of cortical source localized steady-state visually evoked field (ssVEF) differences between CS–US paired and CS-alone presentations was d = 0.67. Given the vast literature of increased steady-state visual evoked potential (ssVEP) or ssVEF responses for CS–US pairings versus CS-alone presentations (for review, see Miskovic and Keil, 2012) a one-sided paired t test was assumed (CS–US > CS-alone) for a power analysis. Given an acceptable power of 0.80, a sample size of n = 16 would be needed to detect these differences. Therefore, a sample size of n = 17 was considered to be sufficient. The sample size estimation was completed using G-power (version 3.1.9.3: http://www.gpower.hhu.de; RRID:SCR_013726).

Experimental design.

The visual CS was a gray-shaded 45° grating (0.31 cycles/cm, square wave) that was projected centrally (visual angle 8° both horizontally and vertically) on a screen. The US was a 95 dB sound pressure level white noise with instantaneous onset delivered binaurally by an air tube system attached to a sound amplifier. The CS was presented for 4000 ms with luminance modulation at 12.5 Hz to evoke an ssVEF response. To this end, a 12 Hz on-off flicker (50 80 ms cycles each containing a 40 ms on and a 40 ms off period) as in our previous reports on ssVEF modulation by fear learning was used (Moratti et al., 2009). During CS–US pairings, the US was presented at 3040 ms after CS onset, then overlapped for 960 ms and co-terminated with the CS.

The experimental design was modeled after Perruchet (1985) and was identical to our previous study (see also Clark et al., 2001; Moratti and Keil, 2009): The 156 conditioning trials were presented as runs of one, two, thee, or four CS–US pairings and as runs of one, two, thee, or four CS-alone trials. For each subject, the run order was pseudorandomized with restrictions to meet the conditions described in Table 1. Within each order, the probability of an US was 50% and thus independent of the run length. After each trial, subjects were asked to rate their US expectancy for the next trial by moving a circle horizontally using a visual analog scale controlled by a left and right response button. An interval of 2 s was introduced between CS offset and the subjects' US expectancy rating. Subjects' response triggered an intertrial interval varying randomly between 4 and 8 s.

View this table:
Table 1.

Organization of trials

After having given informed consent, the subjects' head shapes and five additional index points (nasion, left and right periauricular points, and two additional forehead positions) were digitized to obtain the relative head position to the MEG sensors. Then, two Ag/AgCl EOG electrodes were attached near the left and right outer canthi and two above and below the right eye. An electrode at the right mastoid served as ground. Finally, subject's heads were placed in the MEG helmet.

After US expectancy rating training, the experimental session and MEG recording began. First, participants were shown 10 CS-alone trials while instructed that they would not receive any US (habituation trials). Then, subjects were instructed that, from that point on, the US would be presented in 50% of the trials and that, between each trial, they would have to evaluate their US expectancy for the upcoming trial. After the experiment, subjects were informed about the intended provocation of the gambler's fallacy and were paid or given class credit for participation.

MEG data preprocessing.

MEG data were recorded continuously (1000 Hz sample rate, 0.1–330 Hz online filter) using a 306-channel system (Elekta, VectorView). The MEG sensors consisted of 102 magnetometers and 102 pairs of orthogonal planar gradiometer pairs. However, because our previous report on visual cortex activity modulation by CS–US and CS-alone presentation was based on a magnetometer system (Moratti and Keil, 2009), only magnetometer data entered in the analysis. Continuous MEG data were noise filtered using a temporal signal space separation algorithm (correlation coefficient 0.90, time windows of 10 s) as implemented in the Neuromag MEG software (Taulu and Hari, 2009). Eye blink artifacts were detected and corrected in the data using a signal space projection (Uusitalo and Ilmoniemi, 1997). Then, peristimulus epochs of 3240 ms (200 ms baseline and 3040 ms poststimulus interval) were extracted. The last 960 ms of trial length were not analyzed because of the possibility of introducing artifacts by US presentation, which started at 3040 ms after stimulus onset. Epochs containing artifacts were detected using peak-to-peak detection, with magnetometers at 5 pT thresholds. Further, trials were also visually inspected for movement artifacts and excluded if necessary. Subsequently, a band-pass filter of 1–40 Hz was applied followed by a moving window averaging procedure in which a 0.8 s window containing 10 cycles of a 12.5 Hz oscillation was shifted across the epoch in steps of 0.08 s (one 12.5 Hz cycle). The moving window averaging procedure started at 400 ms after stimulus onset to avoid contamination of the ssVEF signal with the transient VEF response. The resulting single-trial moving window averages were submitted to a cortical source estimation procedure (see below).

Cortical source modeling and extraction of the cortically localized 12.5 Hz ssVEF response.

The underlying current source density of the magnetometer-based ssVEFs was estimated for each time point, experimental condition, and subject using a depth-weighted l2-minimum norm estimation (12-MNE) as implemented in Brainstorm (http://neuroimage.usc.edu/brainstorm; Tadel et al., 2011; RRID:SCR_001761). A dipole mesh derived from a template brain of 7307 vertices (Collins et al., 1998) was used to calculate the forward solution using a boundary element head model as implemented in OpenMEEG (Gramfort et al., 2010; RRID:SCR_002510). Before calculating the forward solution, the head and sensor positions of each subject were co-registered with the template brain by realigning the individual with the template brain's fiducials and further refined by minimizing the mean distance between the individual head shape points and the template brain's scalp surface. During this step, the template brain was size scaled based on the size of the individual head shapes. After this, the l2-MNE was applied to each single-trial moving window average for each subject and transformed into the frequency domain by applying a fast Fourier transform, resulting in single-trial 12.5 Hz ssVEF amplitude estimates at each cortical vertex. Finally, the single-trial ssVEF amplitude distributions on the cortical surface were interpolated into 3D volumes in MNI space and smoothed with a Gaussian kernel (8 mm full-width-half-maximum) for further statistical parametric analysis and model fitting.

Although ssVEF conditioning effects have been reported on higher-order harmonics of the steady-state response (Lithari et al., 2015), we have chosen to analyze the fundamental frequency for the following reasons: (1) our previous reports on the modulation of ssVEF responses by fear learning were based on the fundamental frequency (Moratti and Keil, 2005, 2009; Moratti et al., 2006, 2017), (2) the results to be replicated in the current study were based on the fundamental frequency (Moratti and Keil, 2009), (3) cortical source localized ssVEFs of the fundamental frequency are more confined to the visual cortex (see Fig. 2 in Lithari et al., 2015), and (4) acquisition and extinction effects are best observed at the fundamental frequency (see Fig. 5 in Moratti et al., 2017).

Statistical analysis.

The aim of this study was to fit single-trial ssVEF amplitude modulations to predicted single-trial associative strengths derived from the RW model, as well as subjects' own conscious US expectancy (intertrial ratings). To this end, the ALTSim MATLAB simulator (Thorwart et al., 2009) was used to estimate the single-trial associative strengths based on the previous reinforcement history (CS–US pairing or CS-alone presentation) as predicted by the classic RW model (Wagner and Rescorla, 1972). Because we had hypothesized that we would see ultrafast gain modulations in early visual cortex as a function of previous learning experience, the fastest learning rate was used (α = 1). The associative strength values ranged between 0 and 1, with 0 indicating no associative strength between the CS and the US and 1 representing the maximum association. The ALTsim simulator offers a wide range of learning models (e.g., Pearce–Hall, Replaced Elements Model, etc.). However, because our simple learning paradigm with only one CS resulted in almost identical learning curves across the trials for the different models, the RW model was chosen due to its simplicity. The RW model assumes that the trial-by-trial change of the associative strength ΔV is a decreasing linear function of the difference between the accumulative associative strength V of the stimulus and a predetermined fixed associative value λ as follows: Embedded Image where φ represents the US intensity. Then, these trial-by-trial changes of modeled associative strength ΔV based on the previous learning history and the empirical expectancy ratings (see “Experimental design” section above) were entered as regressors into a general linear model (GLM) fitting ssVEF single-trial amplitude changes contained in the 3D ssVEF volume data for each individual subject (SPM12: http://www.fil.ion.ucl.ac.uk/spm/software/spm12/; RRID:SCR_007037). This resulted in individual 3D images in MNI space containing the β values estimating the fit between cortically localized ssVEF responses and the predicted trial-by-trial associative strength changes ΔV and the trial-by-trial expectancy ratings. Because both regressors entered into the GLM, the variance due to the presence of one against the other predictor was thus accounted for. Figure 1 summarizes these analysis steps.

Figure 1.
Figure 1.

Basic analysis steps up to the first-level statistical analysis. a, Single-trial MEG time series. b, An extracted ssVEF response from that single-trial using a moving window procedure. c, Sensor space topography of a ssVEF peak. d, Fast Fourier transform-derived 12.5 Hz amplitude of the extracted single-trial ssVEF in cortical source space (minimum norm solution). e, Cortical source space solutions were interpolated into a 3D volume in MNI space. f, Design matrix for the first-level analysis in SPM with weights derived from the RW model with learning rate α = 1 (RW1) and expectancy ratings (expect) as regressors.

Thereafter, these individual β values were tested against zero at the group level (second-level analysis). The resulting group-level statistical parameter maps were thresholded at p < 0.05 and corrected for multiple comparisons using the false discovery rate (FDR) at the cluster level (pFDR/cluster < 0.05). The resulting significant voxel clusters served as masks containing RW-β and expectancy-β values significantly different from zero for further analysis.

Then, we tested β values associated with the RW model against expectancy ratings related betas (third-level analysis). However, differences between model fits may be consistent at group level but not reflect β values significantly different from zero. Therefore, these β values actually do not represent brain areas where ssVEF modulations are best explained by the RW-model or expectancy ratings. Therefore, the RW-expectancy model contrast was restricted to voxel clusters that contained RW-betas different from zero. Expectancy-betas were not restricted for this planned contrast to maintain the entire range of expectancy-betas (representing betas that explain ssVEF's variance and noise). Consistent with this, the reverse expectancy-RW contrast was restricted to voxel clusters that contained expectancy-betas different from zero as indicated by the second-level analysis. RW-betas were not restricted for this planned contrast to maintain the entire range of RW-betas (explaining ssVEF's variance and noise). By applying the β value restrictions in both directions for both contrasts, we did not bias our results to one of the two contrasts. However, to safeguard against any possible biasing effects, we also include the results of the same analysis using unrestricted β values. The final resulting statistical parametric map (SPM) contrast images (third-level analysis) were thresholded at p < 0.05 and corrected for multiple comparisons using the FDR at the cluster level (pFDR/cluster < 0.05). Mean β values of significant voxel clusters together with their 95% confidence intervals (95% CIs) are reported below (Cumming, 2014).

Finally, mean single-trial ssVEF responses for each subject derived from the significant voxel clusters based on the previous analysis were extracted for each trial after one, two, three, and four CS-alone and CS–US presentations. The same was done for the expectancy ratings. Both the sorted single-trial ssVEF responses and expectancy ratings were submitted to a specific linear contrast analysis with CS alone and CS–US sequences as within-subject factors. This specific contrast is hypothesis driven and was based on our previous publication to test for a linear increase of occipital ssVEF amplitude with increasing associative strength and a linear decrease of expectancy ratings following the gambler's fallacy (Moratti and Keil, 2009). Note that in our previous study, whole-brain specific contrast analysis resulted in linear effects in the occipital cortex.

Experiment 2

Participants.

Seventeen right-handed subjects (nine females) for the trace-conditioning paradigm (Experiment 2) participated in the study after having given written informed consent. One subject had to be omitted due to excessive noise in the MEG data and 16 volunteers were submitted to analysis (eight females). The mean age of the sample was 26.5 years (range 20–39). All subjects had normal or corrected to normal vision and no family history of epilepsy. Subjects received 20€ or class credit for participation. The participants did not differ from the participants of Experiment 1 with respect to age (t(31) = 1.12, p = 0.270). The local ethics committee approved the study following the declaration of Helsinki.

Experimental design.

All aspects of the stimuli, procedure, data acquisition, MEG data preprocessing, cortical source, and statistical analysis were identical to the delay-conditioning paradigm of Experiment 1 except that a 500 ms interval between CS offset and US onset was introduced during stimulus presentation. However, here, a complex occipito-parieto-temporo-frontal network emerged (see Results) that indicated a superior model fit for the RW model in several voxel clusters. Therefore, mean RW-β values for these clusters were compared using repeated-measures ANOVA with a within-subject factor cluster to test for systematic differences between brain regions. Because mean RW-β values did not vary across different voxel clusters (see Results), these β values were collapsed across the different brain regions for further analysis. As before, mean β values of significant voxel clusters together with their 95% CIs will be reported.

Statistical analysis between the two experiments.

In both delay (Experiment 1) and trace (Experiment 2) conditioning, the RW model best explained the ssVEF modulations in sensory and higher-order cortical networks, respectively (see Results). Therefore, the RW-related β values were submitted to one-way ANOVA with a between-subjects factor experiment (delay vs trace). Thereby, a nondirected F and directed t contrasts will be reported. The final resulting SPM contrast images were thresholded at p < 0.05 and corrected for multiple comparisons using the FDR at the cluster level (pFDR/cluster < 0.05). Mean β values of significant voxel clusters together with their 95% CIs are reported.

Granger causal (GC) interactions within the occipito-parieto-temporo-frontal cluster and statistical analysis.

Analysis of the trace-conditioning paradigm resulted in ssVEF amplitude modulation patterns following the associative strength as predicted by the RW model within an occipito-parieto-temporo-frontal network. To characterize the neuronal dynamics within this cortical network, we estimated GC interactions at the 12.5 Hz driving stimulus frequency. Thereby, the occipito-parieto-temporo-frontal network was subdivided in anatomically meaningful subregions (see Fig. 10). Then, the source waveforms for each vertex within each subregion were extracted for each trial. The complex and complex-conjugate spectral coefficient sequences for each trial, each vertex within a subregion, and each subject were estimated by a fast Fourier transform applied over a time interval between 400 and 3040 ms after stimulus onset (not including the transient evoked field). Then, the Fourier coefficients were averaged for each subregion. The scalar products between these coefficients of the cortical subregions yielding the cross-spectral density matrix were factorized, resulting in nonparametric GC spectral estimates between all possible pairs of cortical subregions (Dhamala et al., 2008). Finally, the 12.5 Hz GC estimates for each pair of subregions and GC interaction direction were extracted for analysis. These analysis steps were done using the fieldtrip toolbox (http://www.fieldtriptoolbox.org; RRID:SCR_004849).

The same GC computations were repeated for randomized source wave epochs for each cortical subregion to control for spurious GC interactions. Thereby, the empirical GC interactions were compared with the GC estimates based on the randomized epochs across all participants using a one-sided paired t test (GC empirical > GC random). To account for multiple comparisons (10 × 10 cortical subregions; see Fig. 10) the GC estimates of empirical and randomized source waveform epochs were randomly switched for each pair of cortical subregions and participant. Then, paired t tests (10 × 10) were calculated. This step was repeated 10,000 times. During each step, the maximal t value was entered into a permutation-based distribution. Finally, initial empirical t values ≥95th percentile of the permutation-based distribution (see Fig. 10) were considered as significant GC interactions. Significant GC interactions between a pair of cortical subregions in both directions were interpreted as bidirectional GC coupling. Mean GC estimates of significant GC interactions are reported together with their 95% CI.

Results

Experiment 1 (delay conditioning)

Induced gambler's fallacy

The aim of using the experimental design of Perruchet (1985) and our previous report (Moratti and Keil, 2009) was to induce a gambler's fallacy. Figure 2a depicts the expectancy ratings after 4 to 1 CS-alone and 1 to 4 CS–US trial sequences. Expectancy ratings declined linearly (F(1,16) = 15.69, p = 0.001, η2 = 0.50; Fig. 2a) with increasing associative strength (Fig. 2b), thus reflecting the successful induction of a gambler's fallacy with respect to US expectancy.

Figure 2.
Figure 2.

Delay-conditioning experiment: a, US-expectancy ratings after 4 to 1 CS-alone and 1 to 4 CS–US trials are shown. Error bars indicate SEs. b, Associative strengths (weights) after 4 to 1 CS-alone and 1 to 4 CS–US trials as predicted by the RW model.

Cortically localized single-trial ssVEF amplitude changes as explained by the RW model or expectancy ratings

One of the main objectives of the study was to demonstrate changes in stimulus-driven ssVEF amplitudes during the course of the experiment as a function of previous learning history (associative strength) as predicted by the RW model or the US-expectancy ratings. The successful induction of the gambler's fallacy effect (Fig. 2) allowed for the evaluation as to whether ssVEF amplitude modulations are driven by conscious US expectancy ratings or by orthogonally established associative strengths between the CS and US, thus contrasting aware and implicit learning processes. In this study, we contrasted β values explaining more of the variance of cortically localized ssVEF amplitude modulations by the RW model than by expectancy ratings and vice versa. Figure 3a shows significant voxel clusters indicating a better single-trial ssVEF model fit in favor of the RW model (threshold pFDR/cluster < 0.05; left occipital cluster pFDR/cluster = 0.010; tmax(16) = 2.70, puncorrected = 0.008; right occipital cluster: pFDR/cluster = 0.025; tmax(16) = 3.64, puncorrected = 0.001). Because the left and right occipital cluster differences between the RW model and expectancy ratings were not significantly different, both clusters were collapsed for further analysis [interaction model (RW vs expectancy) by hemisphere (left vs right: F(1,16) = 0.01, p = 0.923, η2 = 0.001]. Mean β values of the left and right occipital clusters were greater than zero (mean β: 0.0884, 95% CI [0.0362 0.1406]; 13 (76.5%) of 17 subjects showed positive β values; Fig. 3b), whereas mean β values derived from the expectancy model fit distributed around zero (mean β: 0.0001, 95% CI [−0.0008 0.0010]; Fig. 3c). Mean β values of the RW model exceeded mean β values associated with the expectancy ratings in the significant occipital voxel cluster [t(16) = 3.33, p = 0.004, with 13 (76.5%) of 17 subjects showing greater β values in the occipital cluster for the RW model than for the US-expectancy ratings]. Figure 3d shows z-scored mean ssVEF single-trial amplitude changes of the occipital voxel cluster (red line) together with single-trial z-scored predicted associative strengths (blue line) for a representative participant. The β values associated with more explained variance by the expectancy ratings compared with the RW model did not yield any significant voxel clusters (one voxel cluster pFDR/cluster = 0.837).

Figure 3.
Figure 3.

Delay-conditioning experiment. a, First and second columns represent sagittal and axial slices of the MNI template brain with its corresponding MNI coordinates, respectively. The color scale indicates the t values for the RW-expectancy contrast. The threshold is set at p < 0.05 and FDR corrected at the cluster level. b, Bar plots depict mean β values of the significant voxel clusters in a for the RW model and expectancy ratings (EXP). Error bars indicate the corresponding 95% CIs. The black dots indicate the single-subject β values. The gray shaded circle indicates the bar plot that will be shown amplified in the next panel. c, Bar plot, its 95% CI (error bars), and single-subject β values (black dots) for the EXP condition only. d, Mean ssVEF activities across significant voxel clusters (a) for all trials together with the varying learning weights of the RW model for a representative participant. Note that <156 trials are depicted as some trials were omitted due to artifacts. e, RW-expectancy contrast in a “glass brain” representation but without restricting the β values before analysis (pFDR/cluster < 0.05). f, Expectancy-RW contrast without restricting the β values before analysis (pFDR/cluster < 0.05). Color bar indicates the t values for the expectancy-RW contrast. g, Bar plot depicting mean β values of the significant voxel cluster in f for the RW and EXP conditions. Error bars indicate the corresponding 95% CIs. The black dots indicate the single subject β values.

To suppress nois,e this analysis was based on β values restricted to be significantly different from zero (see “Statistical analysis” section). Figure 3e shows the SPM in a “glass brain” representation of the RW-expectancy contrast without any restriction of β values, resulting in the same occipital source clusters (pFDR/cluster < 0.05). However, the expectancy-RW contrast using unrestricted β values yielded a frontoparietal significant voxel cluster (Fig. 3f). Critically, this voxel cluster did not emerge because the expectancy ratings explained more variance of the ssVEF amplitude modulation across trials, as the expectancy β values clustered around zero (mean β: 0.0003, 95% CI [−0.0007 0.0012]), but because of negative RW β values (mean β: −0.04, 95% CI [−0.008 −0.07], with 14 (82%) of 17 subjects showing negative β values).

Finally, for a better comparison with our previous report (Moratti and Keil, 2009), mean ssVEF amplitudes derived from the significant occipital voxel cluster of the better RW model fit were sorted along trials that occurred after 4 CS-alone to 1 CS-alone and from 1 CS–US to 4 CS–US sequences (Fig. 4). Whereas expectancy ratings decreased (Fig. 2a), ssVEF amplitudes increased linearly (F(1,16) = 11.23, p = 0.004, η2 = 0.41; Fig. 4a), replicating our previous results in an independent sample (Moratti and Keil, 2009). Figure 4b depicts linked paired observations between the mean ssVEF amplitudes across all CS-alone and CS–US trials showing that 16 (94.1%) of 17 subjects increased ssVEF amplitudes for reinforced trials.

Figure 4.
Figure 4.

Delay-conditioning experiment: a, Mean ssVEF amplitudes across the significant occipital voxel clusters (see Fig. 3a) for the trial sequences after 4 to 1 CS alone and 1 to 4 CS–US presentations. Error bars indicate SEs. Note that the SEs reflect between-subject variability. b, Paired observations of mean ssVEF amplitudes across all CS-alone and CS–US presentations for each subject.

Experiment 2 (trace conditioning)

Induced gambler's fallacy

Figure 5a depicts the expectancy ratings after 4 to 1 CS-alone and 1 to 4 CS–US trial sequences during trace conditioning. Expectancy ratings declined linearly (F(1,15) = 8.67, p = 0.010, η2 = 0.37; Fig. 5a) with increasing associative strengths (Fig. 5b), thus reflecting the successful induction of a gambler's fallacy with respect to US expectancy during trace conditioning.

Figure 5.
Figure 5.

Trace-conditioning experiment: a, US-expectancy ratings after 4 to 1 CS-alone and 1 to 4 CS–US trials. Error bars indicate SEs. b, Associative strengths (weights) after 4 to 1 CS-alone and 1 to 4 CS–US trials as predicted by the RW model.

Cortically localized single-trial ssVEF amplitude changes as explained by the RW model or expectancy ratings

Here, we contrasted β values explaining more of the variance of cortically localized ssVEF amplitude modulations by the RW model than by expectancy ratings and vice versa, but for the trace-conditioning experiment. A complex occipito-parieto-temporo-frontal network showed a significantly better single-trial ssVEF model fit in favor of the RW model (threshold pFDR/cluster < 0.05). Figure 6 depicts the significant voxel clusters and its corresponding MNI coordinates. Table 2 shows the cluster statistics and the maximum t values for each brain region.

Figure 6.
Figure 6.

Trace-conditioning experiment. a–e, Sagittal and axial slices of the MNI template brain with its corresponding MNI coordinates. The color scale indicates the t values for the RW-expectancy contrast. The threshold is set at p < 0.05 and FDR corrected at the cluster level. f, Because there were no differences in β values between the clusters, mean β values across all clusters are shown for the RW model and expectancy ratings (EXP). Error bars indicate the corresponding 95% CIs. The black dots indicate the single subject β values. The gray shaded circle indicates the bar plot that will be shown amplified in the next panel. g, Bar plot, its 95% CI (error bars), and single-subject β values (black dots) for the EXP condition only. h, Mean ssVEF activities across the significant voxel clusters (a) for all trials together with the varying learning weights of the RW-model for a representative participant. Note that <156 trials are depicted as some trials were omitted due to artifacts. i, RW-expectancy contrast in a “glass brain” representation but without restricting the β values before analysis (pFDR/cluster < 0.05). j, Expectancy-RW contrast without restricting the β values before analysis (pFDR/cluster < 0.05). The color bar indicates the t values for the expectancy-RW contrast. k, Bar plot depicting mean β values of the significant voxel cluster in j for the RW and EXP conditions. Error bars indicate the corresponding 95% CIs. The black dots indicate the single-subject β values.

View this table:
Table 2.

Cluster statistics and the maximum statistics within each cluster are shown

However, because the differences between the model fits did not vary across the different brain regions, β values were collapsed across the significant clusters for further analysis [interaction model (RW vs expectancy) by brain regions (Table 2): F(4,60) = 0.86, p = 0.492, η2 = 0.05, ε = 0.74]. Mean β values of the occipito-parieto-temporo-frontal cluster were different from zero [mean β: 0.0885, 95% CI [0.0411 0.1305]; 13 (81.3%) of 16 subjects showed β values greater than zero; Fig. 6f], whereas mean β values derived from the expectancy model fit were distributed around zero (mean β: 0.0005, 95% CI [−0.0001 0.0011]; Fig. 6g). As shown by the whole-brain SPM contrasts, collapsed mean β values of the RW model exceeded mean β values associated with the expectancy ratings in the significant occipito-parieto-temporo-frontal cluster [t(15) = 4.07, p < 0.001; with 13 (81.3%) of 16 subjects showing greater β values for the RW model than for the US-expectancy ratings; Fig. 6f,g]. Figure 6h shows z-scored mean ssVEF single-trial amplitude changes of the occipito-parieto-temporo-frontal voxel cluster (red line) together with single-trial z-scored predicted associative strengths (blue line) for a representative participant. The expectancy ratings did not yield a superior model fit compared with the RW model (two voxel clusters at pFDR/cluster = 0.844 for both clusters).

To suppress noise this analysis was based on β values restricted to be significantly different from zero (see “Statistical analysis” section). Figure 6i depicts the SPM in a “glass brain” representation of the RW-expectancy contrast without any restriction of β values resulting in the same complex occipito-parieto-temporo-frontal source cluster (pFDR/cluster < 0.05). However, the expectancy-RW contrast using unrestricted β values yielded a left frontal significant voxel cluster (Fig. 6j). Critically, this voxel cluster did not emerge because the expectancy ratings explained more variance of the ssVEF amplitude modulation across trials, as the expectancy β values clustered around zero (mean β: 0.00006, 95% CI [−0.0004 0.0003]), but because of negative RW β values (mean β: −0.03, 95% CI [−0.005 −0.05], with 11 (69%) of 16 subjects showing negative RW β values).

As we had done for the delay-conditioning experiment for a direct comparison with our previous report (Moratti and Keil, 2009), mean ssVEF amplitudes derived from the significant occipito-parieto-temporo-frontal voxel cluster of the better RW model fit were sorted along trials after 4 to 1 CS-alone and 1 to 4 CS–US sequences (Fig. 7). Whereas expectancy ratings decreased (Fig. 5a), ssVEF amplitudes increased linearly (F(1,15) = 9.72, p = 0.007, η2 = 0.39; Fig. 7a), extending our previous results indicating the implication of a distributed occipito-parieto-temporo-frontal network during trace conditioning (Moratti and Keil, 2009). Figure 7b depicts linked paired observations between the mean ssVEF amplitudes across all CS-alone and CS–US trials showing that 13 (81.3%) of 16 subjects increased ssVEF amplitudes for reinforced trials.

Figure 7.
Figure 7.

Trace-conditioning experiment: a, Mean ssVEF amplitude across all significant voxel clusters (see Fig. 6a–e) for the trial sequences after 4 to 1 CS alone and 1 to 4 CS–US presentations. Error bars indicate SEs. Note that the SEs reflect between-subject variability. b, Paired observations of mean ssVEF amplitudes across all CS-alone and CS–US presentations for each subject.

Comparison between delay- and trace-conditioning experiments

Both experiments (delay and trace conditioning) showed a superior model fit for the RW model compared with the expectancy ratings in different brain regions. However, a simple comparison between the two null hypothesis tests for the delay and trace conditioning did not allow any conclusion as to how ssVEF amplitudes were modulated as a function of learning history in distinct brain regions for each experiment. Therefore, mean β values across the significant voxel cluster for each experiment were compared. Figure 8a shows an SPM of a nondirected F contrast between the RW β values of the delay and trace-conditioning experiments (pFDR/cluster < 0.05). The trace- versus delay-conditioning contrast resulted in an almost identical left frontoparietal significant voxel cluster (pFDR/cluster < 0.05) as the nondirectional F contrast. The delay versus trace contrast did not yield any source clusters that were significant (smallest pFDR/cluster = 0.735). Whereas in trace conditioning, ssVEF amplitude modulation across trials followed the predicted RW associative strengths in this frontoparietal voxel cluster (mean β: 0.03, 95% CI [0.012 0.042], 14 (88%) of 16 subjects showed β values greater than zero), in delay-conditioning, ssVEF modulation was inversely related to the predicted RW model weights [mean β: −0.02, 95% CI [−0.006 −0.039], 14 (82%) of 17 subjects showed negative β values]. Only three (18%) of 17 subjects from the delay-conditioning experiment were characterized by β values in the range of those observed in the trace-conditioning experiment. Conversely, only 2 (13%) of 16 subjects from the trace-conditioning experiment had β values in the range of the delay-conditioning experiment.

Figure 8.
Figure 8.

Comparison of delay and trace conditioning. a, Nondirectional F contrast comparing the RW β values of the delay and trace-conditioning experiments (threshold pFDR/cluster < 0.05). b, Trace-delay contrast indicating greater RW β values for the trace-conditioning experiment (pFDR/cluster < 0.05). c, Mean β values across the significant frontoparietal voxel cluster from b are shown for the delay and trace-conditioning experiments. Error bars indicate 95% CIs. The black dots indicate single-subject β values.

GC interactions during trace conditioning

Whereas in delay-conditioning, fast (within three to four trials) cortical gain modulations by associative strength across trials were restricted to early visual cortex, trace conditioning resulted in ssVEF amplitude modulations by associative strength in a distributed occipito-parieto-temporo-frontal network, whereas differences between the two experiments were most pronounced in frontoparietal brain regions. Therefore, GC interactions between plausible cortical subregions derived from the significant clusters of the RW model fit (see Table 2 for the significant clusters and Figs. 9, 10 for the derived subregions) within this network were characterized. To control for spurious GC interactions, the empirical observed GC values between pairs of the subregions were compared with GC influences of randomized epochs. Figure 9c shows the distribution of the maximum t values obtained by the permutation statistics. GC differences were considered significant when the associated t values were equal or superior to the 95th percentile of the permutation distribution (tpermutation = 3.47, ppermutation = 0.05). Figure 9a depicts the mean GC values across subjects together with the 95% CIs for the empirical and random GC interaction observations that were significantly different from each other.

Figure 9.
Figure 9.

Granger Causal interactions during trace conditioning. a, Mean GC estimates for empirical (e) and randomized (r) epochs that were significantly (permutation statistics) different from each other. The error bars indicate 95% CIs. In a third column of the graph, single-subject GC differences between the empirical and randomized epochs are depicted as black dots. dorso-front, dorsolateral frontal; calc, calcarine; pariet, parietal; front, frontal; occip, occipital; temp, temporal; front-centr, frontocentral; occip-lat, occipital lateral; L, left hemisphere; R, right hemisphere. b, Directed GC connectivity matrix. Only significant (see above) GC interactions are depicted in color. The color bar indicates the associated t values that exceed the critical t values derived from the permutation distribution shown in c. The red dotted line indicates the 95th percentile of this distribution (tpermutation = 3.47, ppermutation = 0.05).

Figure 10.
Figure 10.

Directional connectivity during trace conditioning. a, Left, GC interactions between cortical subregions (colored regions) are shown for the left hemisphere (smoothed template brain as has been used for the source estimation). Right, The same interactions are shown in a schematic representation. Red dots indicate the center of gravity of the cortical subregions. The blue arrows indicate the GC directionality. The same abbreviations for the cortical subregions are used as in Figure 9. b, The same information is shown for the right hemisphere. c, The same information is shown for the interhemispheric GC interactions.

The left column of Figure 10 shows the GC interactions between the cortical subregions on the smoothed template brain as has been used for the cortical source estimation procedure. The right column shows a schematic representation of the same networks for better visualization. Further left, right, and interhemispheric GC interactions are shown separately (Fig. 10a–c).

In the left hemisphere, ssVEF responses from the calcarine area were fed forward to the occipital region and to a frontal network whereby frontal and frontocentral brain regions were GC connected bidirectionally with the parietal cortex. The frontal network exerted a top-down influence onto the occipital area via the frontocentral node. The right hemisphere was characterized by top-down GC influences from parietal to calcarine, occipital lateral, and temporal brain areas. Interhemispheric connections were dominated by top-down interactions from left frontal to right temporal, right parietal brain regions. The left frontocentral cortex exerted top-down GC influences onto the right occipital lateral area. In posterior brain areas, the left calcarine area GC influenced the right occipital lateral and calcarine area that in turn predicted ssVEF amplitudes in left occipital cortex.

Discussion

Delay fear conditioning

In order for organisms to survive in rapidly changing environments, fast-adapting sensory systems that discriminate efficiently between fear relevant and irrelevant stimuli confer a great advantage. In this study, we show that fast single-trial changes in associative strength between the CS and US as determined by a RW model with a maximum learning rate predict ultrafast ssVEF amplitude fluctuations in early visual cortex during delay fear conditioning. Thereby, these fast changes of local short-term plasticity in visual cortex were characterized by upregulations and downregulations of ssVEF amplitudes driven by a visual stimulus that changed its associative strength with an aversive US within a few trials. Critically, US expectancy did not better explain activity changes in visual cortex across trials, thus replicating previous results (Moratti and Keil, 2009) and supporting the idea that fast changes in sensory gain due to varying CS–US contingencies are independent of the cognitive anticipation of the US.

There are several possible mechanisms that could drive such fast adaptive trial-to-trial ssVEF amplitude modulations in fear learning. First, the early visual cortex receives input from the amygdala (Amaral et al., 2003), a core structure in fear learning (LeDoux, 2000). The afferents from the amygdala may modulate the excitability of early visual cortex, thus increasing the cortical gain for acquired fear relevant stimuli. Supporting this notion, neuroimaging studies have shown a neuromodulatory role of the amygdala with respect to the extrastriate visual cortex during fear conditioning (Büchel et al., 1998; Morris et al., 2001). In contrast, early visual cortex ssVEF amplitude decreases should involve the ventral medial prefrontal cortex (vmPFC), which inhibits the amygdala and its influence on the visual cortex (Morgan et al., 1993; Maren and Quirk, 2004; Milad et al., 2005). However, no ssVEF activity modulations were observed in vmPFC in the current study even though vmPFC ssVEF amplitude modulations have been observed during extinction in MEG recordings (Moratti et al., 2017).

Second, cholinergic input from the nucleus basalis (Froemke et al., 2013) triggering inhibition of inhibitory interneurons in the sensory cortices (Letzkus et al., 2011) has recently been suggested to control short-term local gain changes of sensory cortex during association learning. In this manner, increased cholinergic input leads to disinhibition of local inhibitory cortical circuits, resulting in neuronal excitation in response to the CS+. In contrast, decreased cholinergic input results in the inhibition of local neural circuits in sensory cortex. Although the majority of studies supporting this notion have been animal studies in the auditory domain (Weinberger, 2015), Bentley et al. (2003) showed the implication of cholinergic input for visual cortex gain modulation in humans.

However, as it is difficult to record brain activity from deep structures such as the amygdala using MEG, our data cannot provide a clear answer as to whether amygdalo-pedal influences, cholinergic neuromodulations, or a combination of both are the mechanism behind these ultrafast changing cortical gain increases and decreases across trials following changing CS–US associations as determined by the RW model. Trial-by-trial increases of visual cortex activity driven by amygdala afferents are plausible, but decreases of ssVEF amplitudes imply other structures that exert inhibitory influences on the amygdala, such as the vmPFC, which was not observed in the present study.

Because ssVEFs or potentials (EEG) are sensitive to attention (Morgan et al., 1996; Müller and Hübner, 2002; Müller et al., 2003), it is also possible that attention related top-down influences on early visual cortex could explain the ssVEF trial-by-trial fluctuations. However, attention would imply awareness-driven processes such as US expectancy. Due to the orthogonal manipulation of the CS–US association strengths and US expectancy, in addition to the superior model fit of the RW model explaining ssVEF amplitude fluctuations in early visual cortex, it is unlikely that top-down influences of conscious US expectancy drive fast trial-by-trial changes in sensory cortical gain. Together, amygdalo-pedal and aware attentional top-down processes are less likely mechanisms underlying single-trial sensory cortical gain modulations. Therefore, we speculate that cholinergic neuromodulatory input into early visual cortex may be the best candidate for explaining ultrafast trial-by-trial sensory gain modulations in early visual cortex driven by associative strength (Froemke et al., 2013; Weinberger, 2015). Future studies manipulating pharmacologically cholinergic input into sensory visual cortex (Bentley et al., 2003) have the potential to provide a more definitive answer.

Trace fear conditioning

In the trace-conditioning experiment, ultrafast trial-by-trial fluctuations of the CS–US associations orthogonal to US expectancy modulated ssVEF amplitude as in the delay-conditioning experiment, but within an extended and complex occipito-parieto-temporo-frontal network. At first, this appears to stand in stark contrast to the findings obtained by Clark et al. (2001), who showed that eye-blink responses are predicted by US expectancy in trace conditioning using the same Perruchet design. There is evidence that trace conditioning involves hippocampus-dependent declarative learning mechanisms (McGlinchey-Berroth et al., 1997; Clark and Squire, 1998; Carter et al., 1999). However, it has been shown that, in humans, trace conditioning depends on these hippocampus-associated declarative learning processes only when trace intervals are 1000 ms or longer (Woodruff-Pak, 1993; Gerwig et al., 2008). Further, there is evidence that the cortical gain in sensory cortex as reflected by ssVEPs is not modulated by fear learning when using very long trace intervals (3 s; Miskovic and Keil, 2013). Therefore, our original hypothesis was that by using a short trace interval of 500 ms, ssVEF amplitude modulations should be better explained by associative strength orthogonal to conscious US expectancy in brain regions relevant for trace fear conditioning. Accordingly, we did not observe a superior model fit for US expectancy over changes in associative strengths as predicted by the RW model. However, because the length of the trace interval seems to be critical for which learning mechanisms (implicit vs explicit) are engaged and if cortical gain modulation occurs, future studies should investigate the relationship between trace intervals and the cortical gain modulations by parametrically manipulate the trace intervals.

The anatomical distribution of the implicated brain regions in our trace-conditioning experiment is consistent with prior findings regarding the interaction of anterior and posterior brain regions when task contingencies are related to stimuli that are separated by time (Fuster, 1990; Miller and Cohen, 2001). Indeed, GC interactions within this occipito-parieto-temporo-frontal network were characterized by top-down influences from frontal onto posterior and temporo-lateral brain areas that in turn projected back to frontal areas. Further, the implication of an occipto-parieto-temporo-frontal network during trace conditioning is consistent with previous human neuroimaging studies on trace and temporal conditioning (Knight et al., 2004; Cheng et al., 2008; Haritha et al., 2013; Harnett et al., 2016). Interestingly, the unrestricted expectancy-RW contrast revealed an inverse relationship between ssVEF amplitude and association strength in a left frontoparietal source cluster during delay conditioning, whereas in the same brain region, ssVEF amplitudes followed the predicted associative strengths across trials during trace conditioning. By omitting the US in delay conditioning the task contingency has to be maintained over time, possibly implicating similar frontoparietal brain structures as during trace conditioning. This inverse relationship in the left frontal cortex during trace conditioning may be triggered by similar mechanisms.

To what extent the amygdala as a core structure in fear conditioning (see above) plays a role in the observed MEG-derived ssVEF modulations within the occipto-parieto-temporo-frontal network is difficult to ascertain, as discussed for the delay-conditioning experiment above. Alternatively, the same cholinergic mechanism (see above) as in delay conditioning may act, not only on local sensory cortices, but also on higher-order brain areas in short interval trace conditioning.

Conclusion

In this study, we show that, by inducing a gambler's fallacy with respect to US expectancy, stimulus-driven neuromagnetic oscillatory activity modulated by the associative strength can be separated from cognitive processes related to US expectancy. Critically, these activity modulations covary with ultrafast changing CS–US associative strengths within three to four trials. During delay conditioning, these fast ssVEF modulations as a function of previous learning experience are limited to early visual cortex, replicating a previous finding using exactly the same experimental design in an independent sample and on a different MEG machine (Moratti and Keil, 2009). In contrast, short interval trace conditioning (500 ms) evokes ssVEF activity changes as predicted by the RW model in an extended occipto-parieto-temporo-frontal network characterized by reverberating GC influences between frontal and posterior brain regions. By using a short trace interval, it was possible to observe ssVEF modulations as a function of associative strength independent of aware US expectancy also in frontoparietal brain regions, consistent with the notion that this network comes into play when task contingencies relate to events that are separated in time (Fuster, 1990; Miller and Cohen, 2001). Our findings further illustrate the extensive flexibility of the brain to prioritize motivationally relevant information by ultrafast (within a few trials) and efficient learning mechanisms likely mediated by short-term plasticity changes (Song and Keil, 2014) in sensory cortex in delay and within extended occipto-parieto-temporo-frontal networks in trace conditioning.

Footnotes

  • This work was supported by the The Spanish Ministry of Science and Innovation (Grants PSI-2009-12702, RYC-2009-04974, and PSI-2014-52205-R).

  • The authors declare no competing financial interests.

  • Correspondence should be addressed to Dr. Stephan Moratti, Departamento de Psicología Experimental, Procesos Cognitivos y Logopedia, Facultad de Psicología, Universidad Complutense de Madrid, Campus de Somosaguas, 28223 Pozuelo de Alarcón (Madrid), Spain. smoratti{at}ucm.es

References

  1. ↵
    2. Amaral DG,
    3. Behniea H,
    4. Kelly JL
    (2003) Topographic organization of projections from the amygdala to the visual cortex in the macaque monkey. Neuroscience 118:1099–1120. doi:10.1016/S0306-4522(02)01001-1 pmid:12732254
    OpenUrlCrossRefPubMed
  2. ↵
    2. Bentley P,
    3. Vuilleumier P,
    4. Thiel CM,
    5. Driver J,
    6. Dolan RJ
    (2003) Cholinergic enhancement modulates neural correlates of selective attention and emotional processing. Neuroimage 20:58–70. doi:10.1016/S1053-8119(03)00302-1 pmid:14527570
    OpenUrlCrossRefPubMed
  3. ↵
    2. Büchel C,
    3. Morris J,
    4. Dolan RJ,
    5. Friston KJ
    (1998) Brain systems mediating aversive conditioning: an event-related fMRI study. Neuron 20:947–957. doi:10.1016/S0896-6273(00)80476-6 pmid:9620699
    OpenUrlCrossRefPubMed
  4. ↵
    2. Cardin JA,
    3. Palmer LA,
    4. Contreras D
    (2008) Cellular mechanisms underlying stimulus-dependent gain modulation in primary sensory cortex neurons in vivo. Neuron 59:150–160. doi:10.1016/j.neuron.2008.05.002 pmid:18614036
    OpenUrlCrossRefPubMed
  5. ↵
    2. Carter CS,
    3. Botvinick MM,
    4. Cohen JD
    (1999) The contribution of the anterior cingulate cortex to executive processes in cognition. Rev Neurosci 10:49–57. pmid:10356991
    OpenUrlCrossRefPubMed
  6. ↵
    2. Cheng DT,
    3. Disterhoft JF,
    4. Power JM,
    5. Ellis DA,
    6. Desmond JE
    (2008) Neural substrates underlying human delay and trace eyeblink conditioning. Proc Natl Acad Sci U S A 105:8108–8113. doi:10.1073/pnas.0800374105 pmid:18523017
    OpenUrlAbstract/FREE Full Text
  7. ↵
    2. Clark RE,
    3. Squire LR
    (1998) Classical conditioning and brain systems: the role of awareness. Science 280:77–81. doi:10.1126/science.280.5360.77 pmid:9525860
    OpenUrlAbstract/FREE Full Text
  8. ↵
    2. Clark RE,
    3. Manns JR,
    4. Squire LR
    (2001) Trace and delay eyeblink conditioning: contrasting phenomena of declarative and nondeclarative memory. Psychol Sci 12:304–308. doi:10.1111/1467-9280.00356 pmid:11476097
    OpenUrlCrossRefPubMed
  9. ↵
    2. Collins DL,
    3. Zijdenbos AP,
    4. Kollokian V,
    5. Sled JG,
    6. Kabani NJ,
    7. Holmes CJ,
    8. Evans AC
    (1998) Design and construction of a realistic digital brain phantom. IEEE Trans Med Imaging 17:463–468. doi:10.1109/42.712135 pmid:9735909
    OpenUrlCrossRefPubMed
  10. ↵
    2. Cumming G
    (2014) The new statistics: why and how. Psychological science 25:7–29. doi:10.1177/0956797613504966 pmid:24220629
    OpenUrlCrossRefPubMed
  11. ↵
    2. Dhamala M,
    3. Rangarajan G,
    4. Ding M
    (2008) Estimating granger causality from fourier and wavelet transforms of time series data. Phys Rev Lett 100:018701. doi:10.1103/PhysRevLett.100.018701 pmid:18232831
    OpenUrlCrossRefPubMed
  12. ↵
    2. Froemke RC,
    3. Carcea I,
    4. Barker AJ,
    5. Yuan K,
    6. Seybold BA,
    7. Martins AR,
    8. Zaika N,
    9. Bernstein H,
    10. Wachs M,
    11. Levis PA,
    12. Polley DB,
    13. Merzenich MM,
    14. Schreiner CE
    (2013) Long-term modification of cortical synapses improves sensory perception. Nat Neurosci 16:79–88. doi:10.1038/nn.3274 pmid:23178974
    OpenUrlCrossRefPubMed
  13. ↵
    2. Fuster JM
    (1990) Prefrontal cortex and the bridging of temporal gaps in the perception-action cycle. Ann N Y Acad Sci 608:318–329; discussion 330–336. pmid:2127512
    OpenUrlCrossRefPubMed
  14. ↵
    2. Gerwig M,
    3. Esser AC,
    4. Guberina H,
    5. Frings M,
    6. Kolb FP,
    7. Forsting M,
    8. Aurich V,
    9. Beck A,
    10. Timmann D
    (2008) Trace eyeblink conditioning in patients with cerebellar degeneration: comparison of short and long trace intervals. Exp Brain Res 187:85–96. doi:10.1007/s00221-008-1283-2 pmid:18253726
    OpenUrlCrossRefPubMed
  15. ↵
    2. Gramfort A,
    3. Papadopoulo T,
    4. Olivi E,
    5. Clerc M
    (2010) OpenMEEG: opensource software for quasistatic bioelectromagnetics. Biomed Eng Online 9:45. doi:10.1186/1475-925X-9-45 pmid:20819204
    OpenUrlCrossRefPubMed
  16. ↵
    2. Haritha AT,
    3. Wood KH,
    4. Ver Hoef LW,
    5. Knight DC
    (2013) Human trace fear conditioning: right-lateralized cortical activity supports trace-interval processes. Cogn Affect Behav Neurosci 13:225–237. doi:10.3758/s13415-012-0142-6 pmid:23263840
    OpenUrlCrossRefPubMed
  17. ↵
    2. Harnett NG,
    3. Shumen JR,
    4. Wagle PA,
    5. Wood KH,
    6. Wheelock MD,
    7. Baños JH,
    8. Knight DC
    (2016) Neural mechanisms of human temporal fear conditioning. Neurobiol Learn Mem 136:97–104. doi:10.1016/j.nlm.2016.09.019 pmid:27693343
    OpenUrlCrossRefPubMed
  18. ↵
    2. Knight DC,
    3. Nguyen HT,
    4. Bandettini PA
    (2003) Expression of conditional fear with and without awareness. Proc Natl Acad Sci U S A 100:15280–15283. doi:10.1073/pnas.2535780100 pmid:14657356
    OpenUrlAbstract/FREE Full Text
  19. ↵
    2. Knight DC,
    3. Cheng DT,
    4. Smith CN,
    5. Stein EA,
    6. Helmstetter FJ
    (2004) Neural substrates mediating human delay and trace fear conditioning. J Neurosci 24:218–228. doi:10.1523/JNEUROSCI.0433-03.2004 pmid:14715954
    OpenUrlAbstract/FREE Full Text
  20. ↵
    2. LeDoux JE
    (2000) Emotion circuits in the brain. Annu Rev Neurosci 23:155–184. doi:10.1146/annurev.neuro.23.1.155 pmid:10845062
    OpenUrlCrossRefPubMed
  21. ↵
    2. Letzkus JJ,
    3. Wolff SB,
    4. Meyer EM,
    5. Tovote P,
    6. Courtin J,
    7. Herry C,
    8. Lüthi A
    (2011) A disinhibitory microcircuit for associative fear learning in the auditory cortex. Nature 480:331–335. doi:10.1038/nature10674 pmid:22158104
    OpenUrlCrossRefPubMed
  22. ↵
    2. Lithari C,
    3. Moratti S,
    4. Weisz N
    (2015) Thalamocortical interactions underlying visual fear conditioning in humans. Hum Brain Mapp 36:4592–4603. doi:10.1002/hbm.22940 pmid:26287369
    OpenUrlCrossRefPubMed
  23. ↵
    2. Maren S,
    3. Quirk GJ
    (2004) Neuronal signalling of fear memory. Nat Rev Neurosci 5:844–852. doi:10.1038/nrn1535 pmid:15496862
    OpenUrlCrossRefPubMed
  24. ↵
    2. McGlinchey-Berroth R,
    3. Carrillo MC,
    4. Gabrieli JD,
    5. Brawn CM,
    6. Disterhoft JF
    (1997) Impaired trace eyeblink conditioning in bilateral, medial-temporal lobe amnesia. Behav Neurosci 111:873–882. doi:10.1037/0735-7044.111.5.873 pmid:9383510
    OpenUrlCrossRefPubMed
  25. ↵
    2. McTeague LM,
    3. Gruss LF,
    4. Keil A
    (2015) Aversive learning shapes neuronal orientation tuning in human visual cortex. Nat Commun 6:7823. doi:10.1038/ncomms8823 pmid:26215466
    OpenUrlCrossRefPubMed
  26. ↵
    2. Milad MR,
    3. Quinn BT,
    4. Pitman RK,
    5. Orr SP,
    6. Fischl B,
    7. Rauch SL
    (2005) Thickness of ventromedial prefrontal cortex in humans is correlated with extinction memory. Proc Natl Acad Sci U S A 102:10706–10711. doi:10.1073/pnas.0502441102 pmid:16024728
    OpenUrlAbstract/FREE Full Text
  27. ↵
    2. Miller EK,
    3. Cohen JD
    (2001) An integrative theory of prefrontal cortex function. Annu Rev Neurosci 24:167–202. doi:10.1146/annurev.neuro.24.1.167 pmid:11283309
    OpenUrlCrossRefPubMed
  28. ↵
    2. Miskovic V,
    3. Keil A
    (2012) Acquired fears reflected in cortical sensory processing: a review of electrophysiological studies of human classical conditioning. Psychophysiology 49:1230–1241. doi:10.1111/j.1469-8986.2012.01398.x pmid:22891639
    OpenUrlCrossRefPubMed
  29. ↵
    2. Miskovic V,
    3. Keil A
    (2013) Visuocortical changes during delay and trace aversive conditioning: evidence from steady-state visual evoked potentials. Emotion 13:554–561. doi:10.1037/a0031323 pmid:23398582
    OpenUrlCrossRefPubMed
  30. ↵
    2. Moratti S,
    3. Keil A
    (2005) Cortical activation during Pavlovian fear conditioning depends on heart rate response patterns: an MEG study. Brain Res Cogn Brain Res 25:459–471. doi:10.1016/j.cogbrainres.2005.07.006 pmid:16140512
    OpenUrlCrossRefPubMed
  31. ↵
    2. Moratti S,
    3. Keil A
    (2009) Not what you expect: experience but not expectancy predicts conditioned responses in human visual and supplementary cortex. Cereb Cortex 19:2803–2809. doi:10.1093/cercor/bhp052 pmid:19304914
    OpenUrlCrossRefPubMed
  32. ↵
    2. Moratti S,
    3. Keil A,
    4. Miller GA
    (2006) Fear but not awareness predicts enhanced sensory processing in fear conditioning. Psychophysiology 43:216–226. doi:10.1111/j.1464-8986.2006.00386.x pmid:16712592
    OpenUrlCrossRefPubMed
  33. ↵
    2. Moratti S,
    3. Giménez-Fernández T,
    4. Méndez-Bértolo C,
    5. de Vicente-Pérez F
    (2017) Conditioned inhibitory and excitatory gain modulations of visual cortex in fear conditioning: effects of analysis strategies of magnetocortical responses. Psychophysiology 54:882–893. doi:10.1111/psyp.12841 pmid:28169431
    OpenUrlCrossRefPubMed
  34. ↵
    2. Morgan MA,
    3. Romanski LM,
    4. LeDoux JE
    (1993) Extinction of emotional learning: contribution of medial prefrontal cortex. Neurosci Lett 163:109–113. doi:10.1016/0304-3940(93)90241-C pmid:8295722
    OpenUrlCrossRefPubMed
  35. ↵
    2. Morgan ST,
    3. Hansen JC,
    4. Hillyard SA
    (1996) Selective attention to stimulus location modulates the steady-state visual evoked potential. Proc Natl Acad Sci U S A 93:4770–4774. doi:10.1073/pnas.93.10.4770 pmid:8643478
    OpenUrlAbstract/FREE Full Text
  36. ↵
    2. Morris JS,
    3. Büchel C,
    4. Dolan RJ
    (2001) Parallel neural responses in amygdala subregions and sensory cortex during implicit fear conditioning. Neuroimage 13:1044–1052. doi:10.1006/nimg.2000.0721 pmid:11352610
    OpenUrlCrossRefPubMed
  37. ↵
    2. Müller MM,
    3. Hübner R
    (2002) Can the spotlight of attention be shaped like a doughnut? Evidence from steady-state visual evoked potentials. Psychol Sci 13:119–124. doi:10.1111/1467-9280.00422 pmid:11933994
    OpenUrlCrossRefPubMed
  38. ↵
    2. Müller MM,
    3. Malinowski P,
    4. Gruber T,
    5. Hillyard SA
    (2003) Sustained division of the attentional spotlight. Nature 424:309–312. doi:10.1038/nature01812 pmid:12867981
    OpenUrlCrossRefPubMed
  39. ↵
    2. Perruchet P
    (1985) A pitfall for the expectancy theory of human eyelid conditioning. Pavlov J Biol Sci 20:163–170. pmid:4069791
    OpenUrlPubMed
  40. ↵
    2. Pessoa L,
    3. Japee S,
    4. Ungerleider LG
    (2005) Visual awareness and the detection of fearful faces. Emotion 5:243–247. doi:10.1037/1528-3542.5.2.243 pmid:15982091
    OpenUrlCrossRefPubMed
  41. ↵
    2. Song I,
    3. Keil A
    (2014) Differential classical conditioning selectively heightens response gain of neural population activity in human visual cortex. Psychophysiology 51:1185–1194. doi:10.1111/psyp.12260 pmid:24981277
    OpenUrlCrossRefPubMed
  42. ↵
    2. Tadel F,
    3. Baillet S,
    4. Mosher JC,
    5. Pantazis D,
    6. Leahy RM
    (2011) Brainstorm: a user-friendly application for MEG/EEG analysis. Comput Intell Neurosci 2011:879716. doi:10.1155/2011/879716 pmid:21584256
    OpenUrlCrossRefPubMed
  43. ↵
    2. Taulu S,
    3. Hari R
    (2009) Removal of magnetoencephalographic artifacts with temporal signal-space separation: demonstration with single-trial auditory-evoked responses. Hum Brain Mapp 30:1524–1534. doi:10.1002/hbm.20627 pmid:18661502
    OpenUrlCrossRefPubMed
  44. ↵
    2. Thorwart A,
    3. Schultheis H,
    4. König S,
    5. Lachnit H
    (2009) ALTSim: a MATLAB simulator for current associative learning theories. Behav Res Methods 41:29–34. doi:10.3758/BRM.41.1.29 pmid:19182121
    OpenUrlCrossRefPubMed
  45. ↵
    2. Uusitalo MA,
    3. Ilmoniemi RJ
    (1997) Signal-space projection method for separating MEG or EEG into components. Med Biol Eng Comput 35:135–140. doi:10.1007/BF02534144 pmid:9136207
    OpenUrlCrossRefPubMed
  46. ↵
    2. Wagner AR,
    3. Rescorla RA
    (1972) Inhibition in Pavlovian Conditioning: application of a theory. In: Inhibition and learning (Boaks RA, Halliday MS, eds), pp 301–336. London: Academic.
  47. ↵
    2. Weinberger NM
    (2015) New perspectives on the auditory cortex: learning and memory. Handb Clin Neurol 129:117–147. doi:10.1016/B978-0-444-62630-1.00007-X pmid:25726266
    OpenUrlCrossRefPubMed
  48. ↵
    2. Williams MA,
    3. Morris AP,
    4. McGlone F,
    5. Abbott DF,
    6. Mattingley JB
    (2004) Amygdala responses to fearful and happy facial expressions under conditions of binocular suppression. J Neurosci 24:2898–2904. doi:10.1523/JNEUROSCI.4977-03.2004 pmid:15044528
    OpenUrlAbstract/FREE Full Text
  49. ↵
    2. Woodruff-Pak DS
    (1993) Eyeblink classical conditioning in H.M.: delay and trace paradigms. Behav Neurosci 107:911–925. doi:10.1037/0735-7044.107.6.911 pmid:8136067
    OpenUrlCrossRefPubMed
Back to top

In this issue

Email
Print
View Full Page PDF
Citation Tools
Respond to this article
Ultrafast Cortical Gain Adaptation in the Human Brain by Trial-To-Trial Changes of Associative Strength in Fear Learning
Melissa Yuan, Tamara Giménez-Fernández, Constantino Méndez-Bértolo, Stephan Moratti
Journal of Neuroscience 19 September 2018, 38 (38) 8262-8276; DOI: 10.1523/JNEUROSCI.0977-18.2018
del.icio.us logo Digg logo Reddit logo Twitter logo Facebook logo Google logo Mendeley logo

Jump to section

Keywords

Responses to this article

Respond to this article

Jump to comment:

No eLetters have been published for this article.