The role of the subthalamic nucleus in perceptual decision making and surprise. Thomas Meindertsma (5810493)
Supervisor: Tobias H. Donner & Niels Kloosterman, Co-assessor: Mike X. Cohen Brain and Cognition, Department of psychology, University of Amsterdam 31 EC - February 2012-July 2014
Research project 1, MSc Brain and Cognitive Sciences – Cognitive Neuroscience track
Introduction
Bistable perception is a process in which a static stimulus is perceived as two temporally alternating percepts. Over the past decades, several bistable visual phenomena have been studied, including binocular rivalry (BR) and motion induced blindness (MIB) (Leopold and Logothetis, 1999; Bonneh et al., 2001). Binocular rivalry is a phenomenon that occurs when dissimilar monocular stimuli are presented to corresponding retinal locations of the two eyes. Rather than perceiving a stable combination of the two stimuli, subjects alternatingly perceive one or the other stimulus (Blake and Tong, 2008). In motion induced blindness a small, salient target, surrounded by a global moving pattern temporarily disappears from visual awareness, only to reappear after a few seconds (Bonneh and Donner, 2011). These phenomena share many features, including their temporally random pattern of perceptual changes and the exclusivity of the perceived state (Leopold and Logothetis, 1999). It is probable that analogous neural mechanisms underlie these different perceptual phenomena (Brascamp et al., 2007; Bonneh et al., 2014).
It was long thought that perceptual bistability is a result of antagonistic connectivity and adaptation in low level sensory areas. An abundance of research shows this antagonistic activity and adaptation in the visual cortex during BR (Blake, 1989; Blake and Logothetis, 2002). An fMRI study with MIB also showed results that can support low level antagonistic activity, namely a slight BOLD decrease after the target disappeared from visual awareness and an increase of BOLD after the target reappeared (Donner et al., 2008).
Leopold and Logothesis suggested a radically different explanation in 1999. In their view, alternating perceptual states reflect the effect of selection processes occurring in higher areas involved in action-selection (Leopold and Logothetis, 1999). An animal study by Libedinsky and Livingstone
supports this idea particularly strongly (Libedinsky and Livingstone, 2011). In this study single cell activity was recorded from the frontal eye fields (FEF) of monkeys, while they reported target disappearances and reappearances in an MIB task. They found a modulation of spiking activity in the FEF that strongly correlated with the perceptual reports of the monkeys. This modulation preceded the latency of perception-related activity in the visual cortex, supporting the hypothesis that frontal (i.e. non-sensory) areas are involved in generating the contents of visual perception.
The cortical areas involved in action selection (including prefrontal and motor areas) form loops with the basal ganglia. These loops give regulatory input back to the cortex and are thought to be a core mechanism in action-selection (Cohen and Frank, 2009). One loop that seems to be particularly important is the “hyper-direct” pathway, of which the subthalamic nucleus (STN) is an important link. This pathway has an inhibitory nature and seems to regulate the threshold for motor responses in an unspecific fashion (Frank, 2006). It has already been suggested that this modulatory role of the STN might not be confined to motor planning. In a study that combined intracranial recordings from the STN with scalp EEG, Cavanagh et al. found that high-conflict decisions were accompanied by increased low-frequency activity in the STN (Cavanagh et al., 2011). Additionally, the authors found a relation between the EEG signal from frontal electrodes on the scalp and decision threshold, which was reversed when the STN was stimulated. This evidence pleads for a causal role of the STN in this decision process.
The STN might have a similar function during bistable perception, modulating the decision threshold for perceptual information. A low decision threshold means that only little evidence is needed for a decision to be made. In bistable perception, there is continuous evidence for both perceptual states of the stimulus, but the strength of this evidence (randomly) fluctuates. When the threshold for perceptual decisions is low, the fluctuations in evidence strength will often be enough to reach the threshold, thus the percept will frequently change from one to the other state. When the decision threshold is high, enough evidence to reach the threshold would only be reached occasionally, so only few perceptual changes are expected. Thus, a correlation between the level of activity in the STN and the frequency of perceptual changes in bistable perception would be expected.
Here, we studied how the spectral characteristics of bistable perception in the STN relate to those observed in the visual and motor cortex. We had the rare opportunity to measure local field potentials from the human STN, while the subjects performed MIB and control tasks. We used these tasks to be able to compare the results to those found in the visual cortex with fMRI and MEG (Kloosterman et al., submitted; Donner et al., 2008).
Above all, we wanted to know if the STN is actively involved in perception. If there is a relation between the local field potential in the STN and the effects found in the visual cortex, it would plead for role of the STN in perception. Furthermore, it would also support the idea that bistable perception is not only explained by local competition and adaptation in the visual cortex (Leopold and Logothetis, 1999). Secondly, we were interested in the specific role that the STN has in bistable perception. By combining basal ganglia models (Cohen and Frank, 2009) with the literature on bistable perception, we hypothesized a correlation between the local field potential in the STN and the frequency of perceptual changes during MIB.
The basal ganglia circuitry is also known to be involved in monitoring and predicting the timing of external events. The unpredictable occurrence of behaviorally relevant visual targets in an “oddball” detection task have been found to induce the P300 event-related potential, a marker of surprise (Nieuwenhuis et al., 2005), in the STN in humans (Baláz et al., 2008; Moll et al., personal communication). Here, we did studied if, and in what way, basal ganglia activity is modulated by surprise about the timing of perceptual changes.
Materials and methods Subjects
A total of three subjects participated in this experiment, performing different tasks. All subjects were diagnosed with Parkinson’s decease (PD) and had bilateral electrode implantations in the subthalamic nuclei as treatment for their decease (known as deep brain stimulation). The experiments were conducted 3-4 days after the electrodes were implanted, just before surgery in which the battery was implanted and connected to the intracranial electrodes. During this period the electrodes can be used to measure local field potential (LFP) in the STN, instead of electrically stimulating it. The subjects were off drugs (L-dopa) while performing the experiment.
The subjects participated in the experiments in between two major surgeries, which impaired their physical fitness and alertness. This influenced our experiment in several ways. First, most potential subjects were not able to participate because they were still in recovery of their first surgery. We had hoped for more subjects, but this was beyond our control. Second, we recorded data for as long as the subjects were able to do the tasks, resulting in a different number of trials in every subject. And third, due to limited availability of subjects we were not able to pilot our task and stimulus before we started our experiment. We made some changes in the experimental design after the data collection had already started. So different subjects performed slightly different tasks.
We chose to use MIB and a corresponding physical replay condition (see Subject 1 & 2 for more detail) in this experiment for several reasons. One important reason was that MIB tasks are more natural than BR tasks, and therefore easier for our subjects to perform. Second, we have MEG data of the visual and motor cortices from healthy subjects performing the same tasks in our lab, which can be beneficial when interpreting the data from this current experiment (Kloosterman et al., submitted). Third, we can study perceptual effects by comparing the illusory absence of the target with the veridical presence of the target. Unfortunately, the MIB tasks proved to be hard for the subjects. The intervals between disappearances and reappearances of the target are often short, making it ill-suited for subjects with motor impairments. We switched to a non-illusory task without short intervals after the first two subjects (see Subject 3 for more detail).
Experimental protocol
Subject 1 & 2
The first and second subject performed the same set of conditions, to (i) determine the factors driving transient modulations of local field potential in the subthalamic nucleus around reports of perceptual switches and (ii) identify the impact of these transient modulations on subsequent neural activity and perception. The task set consisting of (i) a motion induced blindness condition (MIB), (ii) a physical replay condition (Replay) and (iii) a physical replay condition without mask (Replay-no-mask, figure 1A).
Stimulus
The target was a salient yellow disc (full contrast, diameter: 0.12 or 0.2° of visual angle) surrounded by a moving mask (square, equally spaced grid of 9 by 9 blue crosses, 17° width/length), both superimposed on a black background and centered on a fixation mark (red outline, white inside, 0.8° width and length) (Figure 1A). The target was located on one of the four visual field diagonals at an eccentricity of 3°. Target size and location (visual field quadrant) were individually selected for each subject prior to MEG, to yield a percentage of target invisible time of at least 20%. The mask rotated around the fixation square (speed: 120°/s). The target was separated from the mask by a black “protection zone” subtending about 2° around the target (Bonneh et al., 2001). Stimuli were presented using the Presentation Software (NeuroBehavioral Systems, Albany, CA, USA) on a computer screen (60 Hz refresh rate) placed in front of the subject.
Behavioral Tasks and Design
MIB condition: The MIB stimulus was continuously presented for several runs of 2 min duration each. The subjects’ task was to maintain stable fixation and report the spontaneous disappearance and re-appearance of the target, by pressing or releasing a response button with their index finger (left or right, counterbalanced across subjects). The subjects pressed the button to report disappearances and released the button after reappearances.
Replay conditions: In the two different Replay conditions, the target was physically removed from the screen in the same temporal sequence as it had previously disappeared during one of several previous MIB runs completed by the corresponding subject. The physical offsets and onsets of the target were always instantaneous, mimicking the typically abrupt quality of the perceptual switches in MIB for the small target used here. The general purpose of the Replay conditions was to test whether report-related, transient modulations in the local field potential during MIB may have prompted the spontaneous perceptual switch (then they should be specific to MIB), or were driven by the perceptual switch (then they may also occur during Replay-active).
Replay: This condition was identical to MIB in all respects except for the changes in the physical presence of the target.
Replay-no-mask: This condition was identical to Replay-active, except that the target was presented without mask. The purpose of this condition was to test whether potential report-related modulations during MIB and Replay may have reflected a contextual effect in the sensory processing of the mask stimulus (i.e., differing between the perceived or physical presence and absence of the target stimulus). Specifically, given the close relation the STN has to motor processing, this was intended to differentiate between motor related and perception related activity. Additionally, this condition served as a control to verify if the reaction times of our subjects were within the range to be suitable for the tasks in this experiment.
Subject 3
The third subject performed a different task, tailored to the question if and how the STN is involved in surprise about the timing of perceptual events. The stimulus and task were the same as in the Replay condition that the other subjects performed, but the temporal structure of disappearances and reappearances of the target was different. The task for this subject consisted of two Replay conditions, of which the interval durations between disappearances and reappearances of the target was altered to manipulate the predictability of the perceptual change (target disappearance or reappearance) or, in other words, the (temporal) surprise.
In these two Replay conditions, surprise about the timing of perceptual events (defined as a violation of temporal expectation about the target on- and offsets) was manipulated by sampling interval durations (target present/absent) from duration distributions based two hazard function that differed in broadness (figure 1C). A hazard function describes the probability that an event will occur at a particular point in time, given that it has not occurred yet (Luce, 1986). Varying the broadness of the hazard function allowed us to parametrically manipulate surprise about the time of occurrence of an event, since broader hazard functions yield more unpredictable event occurrences. A hazard function is computed as follows (eq. 1):
Eq. 1 where λ(t) is the value of the hazard function at time point t, f(t) is the value of distribution f on time point t, and F(t) is the area under the curve of distribution f from -∞ to time point t. In this experiment, instead of calculating the hazard function from a distribution of intervals, we defined hazard functions and calculated interval distributions from those (eq. 2).
Eq. 2 Both interval duration and the variance of the timings of perceptual events have been found to affect surprise (Fiorillo et al., 2008). We chose the parameters of our hazard function to optimize the difference in predictability between the two conditions and keep the pace of the task within the limits of our subjects’ motor capabilities. For the surprising condition (will be referred to as ‘Replay-high-surprise’ for now on) we used a flat hazard function (mean = 6s), meaning a perceptual event (target dis- or
reappearance) has an equal chance of happening at every point in time, hence surprise is high on average. For the non-surprising condition (‘Replay-low-surprise’) we used a narrow Gaussian hazard
)
(
1
)
(
)
(
t
F
t
f
t
))
(
1
(
)
(
)
(
t
t
F
t
f
function (mean = 3.5s, SD = 0.35s), resulting in a regular, predictable distribution of interval durations. We truncated the distributions of both conditions to only include interval durations longer than 3s, because short intervals proved to be difficult in the first two subjects.
Figure 1. Experimental design. A. The stimulus and percepts of the 3 conditions. During MIB the stimulus (depicted on the bottom) is always the same, but the percept (depicted as thought clouds) alternate between ‘target visible’ and ‘target invisible’. During Replay and Replay-no-mask the percept matches the stimulus. B. Subjects 1 & 2 responded to perceptual switches by pressing a button with their right or left index finger when the target disappeared and released it when the target reappeared. Subject 3 pressed his index finger to report disappearances and switched to pressing his middle finger to report reappearances. C. The interval durations between stimulus changes for the replay condition of subject 3 were drawn from two different distributions that corresponded to the hazard rates depicted here.
Depth recordings of local field potentials
We recorded local field potentials using bilateral microelectrodes in the left and right STN. We recorded bipolar derivations (online referencing to neighboring contacts) from each quadropolar electrode, resulting in three channels per hemisphere. The most ventral contacts were located near the ventral end of the STN, near the substantia nigra. The medial contacts were located centrally in the STN. The uppermost contacts recorded from the superior border of the STN with the lower thalamic base. Additionally, we recorded from a vertical and a horizontal EOG to be able to point out and exclude data containing eye movement and blink artifacts.
Data analysis
The data was analyzed in Matlab (The Mathworks, Natick, MA, USA) using the Fieldtrip toolbox (Oostenveld et al., 2011) and custom made software.
Trial extraction
For all conditions, we extracted trials of variable duration, centered on subjects’ button presses or releases, from the 2 min runs of continuous stimulation. Thus, in the case of MIB, the term “trial” refers to an epoch of constant stimulation and is solely defined based on subjects’ subjective reports of target disappearance and re-appearance. The following constraints were used to avoid mixing data segments from different percepts when averaging across trials: (i) The maximum trial duration ranged from -1.5 s to 1.5 s relative to report; (ii) when another report occurred within this interval, the trial was terminated 0.5 s from this report; (iii) when two reports succeeded one another within 0.5 s, no trial was defined; (iv) for the Replay conditions, trials were only included when the appropriate response was given within 0.2s and 2s after target on- and offsets.
Trial rejection and line noise removal
Trials containing eye blinks, saccades, or muscle artifacts were rejected from further analysis using standard automatic methods. The signal time courses were band-pass filtered in the specific frequency range that contained most of the artifact. These ranges were as follows: 1 to 15 Hz (EOG channel only) for blinks and 110 to 140 Hz (STN depth electrodes only) for muscle activity. Filtering was followed by z-transformation. Trials exceeding a predefined threshold z-score were removed completely from analysis.
Line noise was removed by subtracting the 50, 100, 150 and 200 Hz Fourier components from the raw MEG time course of each trial.
Analysis of MEG power modulations
Spectral analysis of local field potential
We used sliding window Fourier transform (Mitra and Pesaran, 1999) (window length: 400ms, step size: 50ms) to calculate time-frequency representations of the local field potential (spectrograms) for the six channels and each single trial. We used a single Hanning taper for the frequency range 3 - 35 Hz (frequency resolution: 2.5 Hz, bin size: 1 Hz) and the multi-taper technique for the frequency range 36 - 100 Hz (spectral smoothing: 8 Hz, bin size: 2 Hz, five tapers).
Power modulations (denoted as M(f) in the figures) were characterized as the percentage of power change at a given frequency bin, relative to a “baseline” power value for each frequency bin. The baseline was computed by averaging the power values across all time points within each trial (maximum time range -1.5 to 1.5 s around the event), and then across all variable-length trials (i.e., pooling across disappearance and re-appearance trials), but separately for each condition (i.e., MIB, Replay, Replay-no-mask). As a result, the modulations around perceptual reports were expressed as percentage of power change relative to the mean power across all artifact-free trials of each condition. Before performing statistical tests we averaged all power modulations across the six STN channels (3 left, 3 right).
Statistical tests of power modulations
We used a two-tailed permutation test (1000 permutations) (Efron and Tibshirani, 1994) to test the significance of the overall power modulations. In subjects 1 & 2 (MIB, Replay and Replay-no-mask), we tested the overall power modulation for (i) significant deviations from zero, and (ii) significant differences between disappearance and reappearance. In subject 3 (Replay-high-surprise, Replay-low-surprise), we also tested the overall power modulations for (i) significant deviations from zero, and (ii) significant differences between disappearance and reappearance. Additionally, (iii) we tested for differences between the Replay-high-surprise and Replay-low-surprise condition. For all these tests, we used a cluster-based procedure (Oostenveld et al., 2011) across all time-frequency bins to correct for multiple comparisons.
Correlation of STN power with percept stability
We wanted to investigate if the role of the STN as a threshold modulator for action planning and decision making also holds for perceptual decisions. If the activity level in the STN would modulate the perceptual decision threshold, there should be a correlation between “baseline” STN activity levels and the frequency of the perceptual switches during MIB. We used the replay conditions as a control in this analysis. Since the perceptual switches in these conditions were triggered by the stimulus, no correlation between switch dynamics and the activity in the STN would be expected.
To test if this relation exists, we cut the data up into segments with equal length. We used different segment lengths (1, 2, 4 and 8 seconds) to check the robustness of the possible effects. For every segment we computed the mean oscillatory power in the STN over the whole segment, resulting in one value per frequency bin per segment. We then averaged the frequency dimension over conventional frequency bands: theta (4-7 Hz), alpha (8-12 Hz), beta (13-30 Hz), gamma (31-100 Hz), and one extra custom made pooling (9-30 Hz) that was based on the power modulations around reports. We correlated the average power in these frequency bands to the number of perceptual switches in the corresponding segments. We tested the significance of the resulting correlation using permutation statistics. We randomly shuffled the number of perceptual switches over the segments and recalculated the correlation 1000 times. Then, we checked how many times this artificial correlation was higher than the observed correlation and divided this number by the total number of permutations; this was our p-value.
Results
Please note that this project is still ongoing. As mentioned before, data collection was slower than anticipated because of difficulties with subject recruitment. The size of the dataset at the current stage is not sufficiently large to base meaningful conclusions on. The small size and number of the individual datasets, and the relatively noisy data, yielded statistical low power. The results presented here are exploratory.
Figure 2. Power modulations (in percent signal change) pooled across subject 1 & 2 around report of perceptual switches. The top row shows the time frequency representation around disappearance reports for MIB (A), Replay (B) and Replay-no-mask (C). The middle row shows the time frequency representations around reappearance reports for MIB (D), Replay (E) and Replay-no-mask (F). The number of trials per condition and switch type are shown in the top right corner of every plot. The bottom row shows the differences between reappearance and disappearance for MIB (G), Replay (H) and Replay-no-mask (I). Fully saturated colors highlight significant clusters tested across trials (p<0.05, two-sided permutation test, cluster-corrected).
The reaction times were not in a normal range
In the MIB condition, the subjects’ reports are the only measure we have detect perceptual switches. It is therefore important that the subjects are able to respond within a reasonable range after the perceptual switch happened. In the Replay conditions we also knew the timing of the on- and offsets of the target. We analyzed the reaction times in these conditions to check if our subjects were able to respond within an acceptable range after the perceptual switch. We excluded trials with reaction times that fell outside a 0.2-2s range after stimulus change (target on- or offset), because we were not able to say if these were responses to the stimulus change or not. These trials were counted as ‘misses’. The subjects were only able to respond within the response range in 36% of the Replay trials (Replay = 28%, Replay-no-mask = 47%). We used only the trials with a response within the reaction time range for further analysis. Within this range, the mean RT pooled across subject 1 & 2 was 1086ms (SD=85ms) (Replay mean = 922 +/-124ms, Replay-no-mask mean = 1068 +/-116ms). These reaction times are slower those found in healthy subject performing the same tasks (500-600ms, Kloosterman et al., submitted). This means it might be harder to find switch-related activity in this experiment.
No clear transient power modulations in the STN around perceptual switches
First, we wanted to test if oscillatory power in the STN is modulated around perceptual switches during MIB. We pooled all trials of the MIB condition across subjects 1 and 2 and separated disappearances from reappearances (figure 2). There was a significant decrease of beta band power (10-30 Hz) around the report of disappearances in MIB (figure 2A). A decrease in beta power was also present around reappearance reports, although not significantly (figure 2D). There were no significant differences between disappearance and reappearance reports (figure 2G). The Replay condition (figure 2B, E and H) and the Replay-no-mask condition (figure 2C, F, I) show similar results, with a significant decrease of beta power around disappearance reports in the Replay condition. There were no significant power modulation around reappearance reports, nor were there significant differences between disappearance and reappearance in any condition. The beta decrease around reports may be related to the report of the perceptual change (Donner et al., 2009; Joundi et al., 2013).
Power in the STN around report is slightly modulated by global temporal structure of the task
With the task that subject 3 performed we wanted to study the effect of surprise about the timing of perceptual changes on power modulations in the STN. In studies that used oddball stimuli the P300 component of the event-related potential, a marker of surprise (Nieuwenhuis et al., 2005), has been recorded in the STN (Baláz et al., 2008; Moll et al., personal communication). Here we manipulated the global temporal predictability of the perceptual switches in two conditions. In both the Replay-high-surprise and Replay-low-Replay-high-surprise conditions there is a significant suppression of beta power (13-30 Hz) around report of disappearances (figure 3A-B). This beta suppression is also present during reappearance reports in both conditions (figure 3D-E). This result is consistent with what we found in our other subjects and may be related to the motor report of the perceptual change (Donner et al., 2009; Joundi et al., 2013). There was also a significant difference in the strength of the beta power decrease between reappearance and disappearance (figure 3G-H), with a stronger beta suppression around disappearance.
This difference may be caused by a perceptual effect (the perceptual change being a disappearance or a reappearance), or reflect a difference in report (i.e. finger used for button press or the preparation of the motor response). Additionally, there is a significant increase in alpha power (5-12 Hz), present in both conditions during both disappearances and reappearances (figure 3A,B,D and E).
Figure 3. Subject 3. STN power modulations (in percent signal change) during high and low surprise. The top row shows the time frequency representation around disappearance reports for high surprise (A), low surprise (B) and the difference between high and low surprise (C). The middle row shows the same conditions around reappearance reports (D-F). The bottom row shows the difference between disappearances and reappearances for high surprise (G) and low surprise (H). The number of trials per condition is shown in the top right corner of every figure A, B, D and E. Fully saturated bins depict significant clusters (p<0.05, two-sided permutation test, cluster-corrected).
The variable of interest in this analysis is the difference in power between the high and low surprise condition. Although the transient power in these two conditions is qualitatively similar, as would be expected, the strength of the modulations differs slightly between high and low surprise. The beta band shows a trend towards stronger suppression in the Replay-low-surprise condition (figure 3C and 3F). This is true for both disappearance and reappearance. The low frequency (5-12 Hz) power increase, on the opposite, is significantly stronger in the Replay-high-surprise condition around disappearance reports (figure 3C) and shows the same trend around reappearance (figure 3F). This low frequency effect is not known to be related to responses; instead it may reflect surprise about the timing of the perceptual changes.
Correlation between sustained power in the STN and the frequency of perceptual switches
We were also interested in the role of the STN in perceptual decisions. If the model by Cohen and Frank (Cohen and Frank, 2009) generalizes to perceptual decision making, one might expect a relationship between STN activity and the frequency of perceptual switches. Specifically, we predicted a correlation between baseline power in the STN and the frequency of perceptual changes during MIB.
We cut the data of subject 1 and 2 up in segments and correlated mean power in different frequency bands to the number of perceptual switches per segment. We predicted a correlation only in the MIB condition, since this is the only condition in which the frequency of perceptual switches depended on the subjects endogenous processes and not on the external stimulus. The Replay conditions, in which the perceptual switches were determined externally by the stimulus, served as a control. We tried different segment lengths (1, 2, 4 and 8 s) to check the robustness of any relationships to be. The results for the different frequency bands of subject 1 are depicted in figure 4A-E, the results for subject 2 are depicted in figure 4F-J.
In subject 1 there were significant positive correlations between STN power and the number of perceptual switches in the Replay condition in the theta and gamma band (p<0.05, Bonferroni corrected, figure 4A and D). These correlations were not present for all segment lengths. Overall, the correlations in this subject in all conditions do not show a coherent structure and are therefore hard to interpret.
The results for subject 2 show a different picture than those of subject 1 (figure 4F-J). In subject 2 there are positive correlations only in the MIB condition. These correlations are significant for all frequency bands, sparing out the alpha band. In most robust correlation was found in the beta band, were all segment lengths were significant (figure 4H), while in the theta band, gamma band and custom pooling (9-30 Hz) only part of the segments were significant (figure 4F, I and J). The correlations in subject 2 are consistent with our hypothesis that the role of the STN in perceptual decisions that is analogous to the motor threshold modulation proposed by the model by Cohen and Frank (Cohen and Frank, 2009).
Figure 4. Correlation between STN power and the frequency of perceptual switches (subjects 1 & 2). The top row shows the correlation values (r) of subject 1 between the number of perceptual switches per time segment (on the x-axis) and the mean power in that time segment in the frequency bands theta (A), alpha (B) beta (C), gamma (D) and a custom pooling (9-30 Hz) (E). The bottom row shows the same for subject 2 (F-J). The horizontal bars in the bottom of the plots represent significant correlations per condition (p<0.05, Bonferroni corrected for multiple comparisons). Error bars are 95% confidence intervals.
Discussion
We studied local field potential oscillatory power in the STN around perceptual changes in bistable perception. We found that transient power modulations around report of perceptual changes were affected by surprise about the timing of these changes in a band-specific manner (figure 3C and F). Beta power suppression was weaker in the high surprise condition, while simultaneous alpha increase was stronger in this condition. Additionally, sustained power in all conventional frequency bands except the alpha band is positively correlated to the frequency of perceptual switches in one subject (figure 4F-J). Discussion of the results of this study will be kept to a minimum, because the data set is too small to be conclusive.
Besides the small amount of data, the interpretability of these results is also tempered by some prominent inconsistencies. The correlation between sustained power in the STN and the frequency of perceptual changes during MIB was found quite reliably in subject 2, but was not at all present in subject 1. The results in subject 2 suggest that the STN plays the role of decision threshold modulator that we hypothesized, but more subjects need to be tested before we can really interpret this finding.
The sign of the correlation in the MIB condition in subject 2 is positive, meaning that increased power in all frequency bands except alpha in the STN is accompanied by more perceptual switches. For lower frequencies, increased synchrony is a sign of inhibition of neuronal processing, at least in cortical areas (Jensen and Mazaheri, 2010). This is consistent with our hypothesis. More inhibition in the STN (i.e. less activity) was hypothesized to lower the threshold for perceptual decision; hence positive correlation would be expected. The correlation in the higher frequency band (gamma) is also positive, which is less consistent with our hypothesis. Different from the lower frequencies, gamma band power increases are thought to reflect increases in neuronal processing (Donner et al., 2009). Thus, for this frequency band, a negative correlation would be expected.
However, this is not necessarily a strong inconsistency. First, oscillations in low frequencies do not reflect neuronal inhibition itself, but rather affect the excitability of neurons slightly by changing local field potential (Fries et al., 2007). This has been found to influence neuronal processing in some cases, but this is not a direct process based on inhibitory and excitatory connections between neurons. The same goes for higher frequency synchrony and excitation. Second, the STN is a subcortical area with a very different local structure then the cortex. It does not have the layers that are systematically found in the cortex. Synchronous neural activity might also have a different effect here.
We found an effect of the predictability of the timing of perceptual changes on the power modulations in the STN. Part of this effect consisted of a trend towards stronger beta suppression around report in the low surprise condition, compared to the high surprise condition (figure 3C and 3F). This beta suppression, present in both conditions, is likely to reflect the motor response (Donner et al., 2009; Joundi et al., 2013). The beta suppression in the high surprise condition is more sustained, for disappearances there is even a significantly stronger beta suppression around 1s after report (figure 3C). Since these results are locked to the report, the weaker and more prolonged beta suppression in the high surprise condition cannot be due to a larger variability in response latency. However, it could be that it does not reflect surprise about the timing of perceptual changes, but just less preparatory motor processing or less decisive responses. Or, in terms of decision making, in the high surprise condition evidence might be integrated less efficiently (because it was not anticipated) and a decision is reached slower (Gold and Shadlen, 2007; Donner et al., 2009).
Besides the weaker beta suppression in the high surprise condition, we also found a stronger alpha power increase in this condition around the same time. This effect is not known to be specifically motor-related. The sign of this effect, being stronger in the more surprising condition, cannot be accounted for by the same alternative explanation. Instead, it might reflect the surprise signal we hypothesized. Further research is needed to tease apart this effect of surprise and motor activity, for
example by omitting the motor response and letting subjects count the number of perceptual switches instead (Meindertsma et al., 2014).
In the analyses we did so far, we focused on the effect of the average predictability of the timing of perceptual changes, which we manipulated in two separate conditions. However, another possibility is to use a model-based approach to capture the trial-to-trial level of surprise, taking into account the recent history of interval durations between perceptual changes (Mars et al., 2008). By fitting a model of trial-to-trial surprise level to the STN time frequency spectrum we can study the role of the STN in surprise processing more precisely. We aim to do these analyses in the future.
Another interesting direction for future research would be to simultaneously measure STN activity with intracranial electrodes and whole head cortical activity with scalp EEG or MEG. This would provide the opportunity to look further into the transient effects around perceptual reports in the STN and see how they relate to other parts of the brain, in particular the visual cortex. From previous fMRI and MEG research using the same stimulus and task we know that disappearance and reappearance reports are accompanied by opposite modulations of visual cortex activity (Kloosterman et al., submitted; Donner et al., 2008), this opposite signature is not present in the STN. However, beta power suppression around target disappearances is significantly stronger than around target reappearances (figure 3G-H). This effect might be perception-related. By comparing the timing and strength of STN and visual cortex modulations, and by looking at connectivity measures between them, we can distinguish between motor-related, decision-related and perceptual signals in the STN.
Besides the first indication about the role of the STN in perception, we gained important knowledge about the difficulties of recording from depth electrodes in human subjects. We learned several important lessons about this over the course of this experiment. First, data collection prognoses are unreliable. Even when there is a steady supply of potential subjects (i.e. PD patients undergoing DBS electrode implantation surgery), many of them will not be able to participate because they have not recovered enough from surgery. Second, the patients in our study had difficulty reporting the MIB illusion. Illusory percepts in MIB are often short and the timing of target disappearance and reappearance is typically unpredictable (Bonneh et al., 2001). The slow reaction times and high number of missed perceptual changes (no button press) in the Replay conditions indicated that this was hard for our subjects. We changed the task halfway through the experiment to accommodate the subjects’ difficulties with reporting perceptual changes by (i) focusing on Replay conditions only and (ii) using only interval durations (between target on- and offsets) of more than three seconds.
In future studies, we will use what we have learned and avoid fast or difficult tasks involving ambiguous stimuli. Additionally, we plan to measure STN local field potential and scalp EEG simultaneously. The relation between STN and visual and motor cortex power modulations will help to disentangle motor, perceptual and decision-related signals in the STN. Here, we set a first step in understanding if and how the STN processes surprise about the timing of perceptual changes. We plan to continue this work by employing model-based analyses of surprise, aimed to capture trial-to-trial variability in surprise about the timing of the perceptual changes.
References
Baláz M, Rektor I, Pulkrábek J (2008) Participation of the subthalamic nucleus in executive functions: An intracerebral recording study. Mov Disord 23:553–557
Blake R (1989) A neural theory of binocular rivalry. Psychol Rev 96:145–167 Blake R, Logothetis NK (2002) Visual competition. Nat Rev Neurosci 3:13–21 Blake R, Tong F (2008) Binocular rivalry. Scholarpedia 3:1578
Bonneh Y, Donner T (2011) Motion induced blindness. Scholarpedia 6:3321
Bonneh YS, Cooperman A, Sagi D (2001) Motion-induced blindness in normal observers. Nature 411:798– 801.
Bonneh YS, Donner TH, Cooperman A, Heeger DJ, Sagi D (2014) Motion-induced blindness and Troxler fading: common and different mechanisms. PLoS One 9:e92894
Brascamp JW, Knapen THJ, Ee R Van, Berg AV Van Den (2007) Flash suppression and fl ash facilitation in binocular rivalry. J Vis 7:1–12.
Cavanagh JF, Wiecki T V, Cohen MX, Figueroa CM, Samanta J, Sherman SJ, Frank MJ (2011) Subthalamic nucleus stimulation reverses mediofrontal influence over decision threshold. Nat Neurosci 14:1462–1467.
Cohen MX, Frank MJ (2009) Neurocomputational models of basal ganglia function in learning, memory and choice. Behav Brain Res 199:141–156.
Donner TH, Sagi D, Bonneh YS, Heeger DJ (2008) Opposite neural signatures of motion-induced blindness in human dorsal and ventral visual cortex. J Neurosci 28:10298–10310.
Donner TH, Siegel M, Fries P, Engel AK (2009) Buildup of choice-predictive activity in human motor cortex during perceptual decision making. Curr Biol 19:1581–1585.
Efron B, Tibshirani RJ (1994) An Introduction to the Bootstrap. CRC Press.
Fiorillo CD, Newsome WT, Schultz W (2008) The temporal precision of reward prediction in dopamine neurons. Nat Neurosci 11:966–973.
Frank MJ (2006) Hold your horses: a dynamic computational role for the subthalamic nucleus in decision making. Neural Netw 19:1120–1136
Fries P, Nikolić D, Singer W (2007) The gamma cycle. Trends Neurosci 30:309–316
Jensen O, Mazaheri A (2010) Shaping functional architecture by oscillatory alpha activity: gating by inhibition. Front Hum Neurosci 4:186
Joundi R a, Brittain J-S, Green AL, Aziz TZ, Brown P, Jenkinson N (2013) Persistent suppression of subthalamic beta-band activity during rhythmic finger tapping in Parkinson’s disease. Clin Neurophysiol 124:565–573
Kloosterman NA, Knapen T, Hillebrand A, Dijk BW Van, Lamme V, Donner TH (submitted) State Change in Human Visual Cortex Predicts the Stability of a Perceptual Illusion.
Leopold D, Logothetis N (1999) Multistable phenomena: changing views in perception. Trends Cogn Sci 3:254–264.
Libedinsky C, Livingstone M (2011) Role of prefrontal cortex in conscious visual perception. J Neurosci 31:64–69.
Luce RD (1986) Response Times: Their Role in Inferring Elementary Mental Organization, 1st ed. Oxford University Press, USA;
Mars RB, Debener S, Gladwin TE, Harrison LM, Haggard P, Rothwell JC, Bestmann S (2008) Trial-by-trial fluctuations in the event-related electroencephalogram reflect dynamic changes in the degree of surprise. J Neurosci 28:12539–12545
Meindertsma T, Kloosterman NA, Nolte G, Engel AK, Donner TH (2014) A correlate of covert decisions in the human visual cortex. Poster presented at Human Brain Mapping, Hamburg, Germany.
Mitra PP, Pesaran B (1999) Analysis of dynamic brain imaging data. Biophys J 76:691–708Nieuwenhuis S, Aston-Jones G, Cohen JD (2005) Decision making, the P3, and the locus coeruleus-norepinephrine system. Psychol Bull 131:510–532.
Oostenveld R, Fries P, Maris E, Schoffelen J-M (2011) FieldTrip: Open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput Intell Neurosci 2011:156869.
