The neural mechanisms related to first and second order decision-making: a pilot study

The neural mechanisms related to

first and second order

decision-making: a pilot study

Diane Helena Maria Roozendaal (10001424)

Abstract

The underlying processes of first (the ability of distinguishing between stimuli) and

second (the evaluation of one’s own correct and incorrect responses) order

decision-making are currently hotly debated. We stimulated MT/V5, highly associated with the

visual processing of motion, to manipulate first order decision-making. To aim at

metacognition performance, we manipulated activity in aPFC with a recently

developed theta-burst (TBS) protocol. In this pilot study we found no significant effects

of TBS, but provided new insights in methods and techniques that can be used to

unravel the neural mechanisms related to first and second order decision-making.

February 5th _{2015 – November 9}th _{2015, 26 ECTS}

MSc in Brain and Cognitive Sciences, Cognitive Neuroscience, University of Amsterdam Supervisor: Martijn E. Wokke

Co-assessor: H. Steven Scholte

1 Imagine coming home at dusk, when you suddenly

notice that you are not alone. You vaguely see the face of a man. When he spots you, he flees. You decide immediately to call the police. They ask you to give a description of the man and ask you how certain you are that this is what you saw. Since the environment was dark and you only saw him a split second, you are not confident with the answers you give.

The above example involves an assessment of the quality (am I confident about what I saw) of a previously made decision (yes, the man had a beard). This second order judgement is often referred to as metacognition, or simply: cognition about cognition (Flavell, 1979). Metacognition is the ability to evaluate the success of a cognitive process, and can be assessed with subjective ratings that reflect confidence about first order performance (whether a stimulus is a face or not, or whether the man had a beard or not).

The neural networks underlying metacognition are currently hotly debated. In general, there are three potential models that could describe the neural mechanisms of metacognition (Maniscalco, 2014). The first model is supported by animal studies and can be described as a single channel model, in which first order (the ability to distinguish between stimuli) and metacognitive decision-making share the same underlying neural mechanisms (Fetsch et al., 2014). For instance, single cell recordings in rhesus monkeys revealed that choice and the degree of certainty in that choice are represented by the same neural correlates (Kiani & Shadlen, 2009). By offering a sure-bet option in a visual discrimination task, monkeys could chose to opt out indicating a certain degree of choice uncertainty. Recorded firing rates in LIP neurons revealed similar activity during opt out choices and perceptual choices. These findings indicate that first order and metacognitive decisions are represented by the same neural correlates. Hebart and colleagues (2014) support this theory with evidence that metacognition in humans can be seen as a direct derivative of the decision variables underlying first order decisions. On the contrary, other studies postulate that metacognition can be described as a hierarchical model, in which metacognitive decisions can be manipulated independently from first order decisions (Timmermans et al., 2012). Rounis et al. (2010) stimulated the prefrontal cortex, an area highly associated with visual awareness, during a visual discrimination task.

Stimulation did not directly affect task performance, but specifically affected one’s subjective visibility. Thus, participants were less aware of the quality of their visual processes, or less able to evaluate previous made decisions. These results indicate that metacognitive processes can be manipulated, without affecting first order decision-making. As such, metacognition can be seen as an additional (higher-order) process. Along the reasoning of this hierarchical model, the prefrontal cortex (PFC) is thought to play an important role in metacognitive functioning (Fleming & Dolan, 2012; Fleming et al., 2014). However, the exact functional and causal role of the PFC in metacognition remains unclear. A third model that we will not discuss further assumes strictly separate processing routes for first and second order decision-making (Del Cul et al., 2009).

In order to come to understanding brain-behaviour relationships, lesions studies in humans have proven to be extremely valuable in defining brain regions necessary for cognitive functioning, providing causal evidence regarding PFC in metacognition. In addition, this field of research is able to include subjective measures often lacking in for instance animal studies. Several studies examined metacognitive functioning in patients with prefrontal lobe lesions or damage. Although alternations in self-evaluation and confidence are reported, findings are not consistent (Pollen, 1995; Schmitz et al., 2006). One recent study reports domain-specific metacognitive impairment in patients with anterior frontal lobe lesion (Fleming et al., 2014). Results demonstrate a decrease in metacognitive accuracy while leaving perceptual performance unaffected. These findings implicate a necessary role of aPFC in metacognition but not in first order performance, supporting the hierarchical model for metacognition.

Recently, this same causality between brain and behaviour can be assessed by inducing a so-called “reversible lesion” by means of brain stimulation. A recently developed protocol coined Theta Burst Stimulation (TBS) provides an optimal solution to noninvasively mimic lesions (Huang et al., 2005). TBS involves the injection of a train of magnetic pulses (50 Hz) at a theta frequency (5 Hz) to locally depress cortical areas for a prolonged amount of time. Studies using TBS to investigate the underlying mechanisms of metacognition, showed that TBS of the PFC resulted in specific impairment of metacognition, while task performance remained unchanged (Rounis et al.,

2 2010; Chiang et al., 2014). Fleming and colleagues

(2012) used fMRI to elaborate these findings and reported converging evidence for prefrontal areas involved in metacognition. These studies provide evidence in favour of the hierarchical model in which first order decision-making and metacognition initially share the same visual information, but higher order processes to evaluate first order performance are needed to produce metacognition.

In the present study, we aim to investigate the neural mechanisms related to first and metacognitive (second order) decision-making. We aim to provide evidence for a hierarchical model in which metacognition can be seen as an additional, higher-order process, distinguishable from first order decision processes. TBS will be used to causally manipulate brain areas associated with metacognition and first order decision-making, after which participants perform a face detection task. During this task participants need to distinguish between face and non-face stimuli, which are moving in opposite direction from a moving noisy background, known as form-from-motion (Wokke et al., 2012; 2014).

Metacognition can be behaviourally measured with a recently developed signal detection theoretic approach (Maniscalco & Lau, 2012). In contrast to previously used measurements of metacognition, such as a simple phi correlation between confidence and accuracy or a Goodman-Kruskall gamma coefficient (Nelson, 1984), this recent approach is considered “bias-free”, since it is able to distinguish between response bias (the consistency of choosing one answer over the other) and objective performance. The authors named this metacognitive measure: metacognitive sensitivity, which reflects how well a participant can distinguish between one’s own correct and incorrect responses. This evaluation can be assessed with confidence ratings and is calculated in units of meta d’. Meta d’ is derived from a classical signal detection theory (SDT) analysis, in which d’ reflects one’s ability to discriminate between stimuli (Macmillan & Creelman, 1991). According to SDT, when one performs a stimulus discrimination task an internal signal for each stimulus, on each trial, is evaluated along a decision axis. This internal signal is used to discriminate whether the stimulus will be classified as stimulus 1 (S1) or stimulus 2 (S2). Therefore, two internal response distributions are produced corresponding to S1 and S2. The stimulus is classified as S1 if the signal exceeds criterion 1 (c1;

separating the two internal response distributions), and is classified as S2 if it fails to exceed c1. The larger the distance between these two distributions along the decision axis, the better one can distinguish S1 from S2. Thus, a better sensitivity for discriminating S1 from S2, which is called type 1 d’. When assigning confidence ratings to this stimulus discrimination, classifying the accuracy of responses, one can calculate type 2 sensitivity or meta d’ (Maniscalco, 2012). A single type 2 classification (one trial) can result in four possible outcomes when evaluating a specific response (S1 or S2): correct and incorrect responses with low or high confidence. With these variables one can produce a response specific ROC (Receiver Operating Characteristic) curve, in which all false positives (incorrect responses with high confidence) and hit rates (correct responses with high confidence) are plotted for all confidence criteria (criteria that separate the different confidence distributions). This response specific type 2 ROC curve represents the expected metacognitive sensitivity for a theoretically metacognitive ideal participant, given a constant d’ and c. To calculate the actual observed metacognitive sensitivity, meta d’, one can measure the distance from C1 to the internal response distributions produced for each confidence rating. The larger the distance from C1 along the decision axis, the more confidence one can assign to a stimulus discrimination. Given the observed metacognitive sensitivity, one can produce an ROC curve. The direct comparison of this observed ROC curve to the expected type 2 ROC curve, the value of meta d’ relative to d’ is defined as metacognitive efficiency. Meta d’/d’ reflects how much first order information from d’ is left for metacognition, in signal-to-noise units. A theoretically metacognitive ideal participant will have a value of 1 (Fleming & Lau, 2014).

The direct comparison between first and metacognitive decision-making in combination with TBS provides an optimal solution to investigate the neural mechanisms that are related to first and second order processes. In this study we therefore manipulated neural activity strongly related to first order decision-making by stimulating MT/V5, an area highly associated with the visual processing of motion. We hypothesized a decrease in first order performance, d’, without affecting metacognitive sensitivity. To aim at metacognitive performance, we manipulated activity in aPFC with TBS. We predicted a decrease in metacognitive sensitivity, while leaving task

3 performance unaffected. Thus, by directly

manipulating the brain areas associated with first and metacognitive performance combined with the direct comparison of d’ versus meta d’, we aim to investigate the neural mechanisms related to first and metacognitive decisions and study the functional role of aPFC in metacognition.

Methods

Participants

Ten healthy volunteers (6 females, mean age = 23.5, SD = 2.12) were recruited from earlier transcranial magnetic stimulation studies to participate in this study for financial compensation. Therefore, participants were already familiar with stimulation and the chance of drop-out was minimized. All had normal or corrected-to-normal vision and were screened for both TBS and MRI according to international guidelines (Wasserman, 1998; Rossi et al., 2009). Written informed consent was obtained from all participants and the study was approved by the local ethics committee.

Experimental design

Participants were asked to come to the Spinoza Centre for Neuroimaging for a minimum of six sessions. The first session included explanation and familiarization of the task and one hour in the MRI scanner. During scanning, a structural T1 image, four blocks of the task, a functional LO/PPA/FFA and a MT/V5 mapper were acquired. The latter for specific determination of MT/V5 coordinates, necessary for neuronavigation (see theta-burst stimulation). The LO/PPA/FFA mapper and the four blocks of the task were used for other purposes. The second and third session included individual optimization of the task, in such way that participants achieved threshold level performance (~75% correct). Performance had to be stable over 4 blocks of practice, without a learning effect or response bias towards face or non-face stimuli. Optimization included adaptation of the stimulus duration, with a minimum duration of 3 steps and a maximum of 6 steps (see visual discrimination task). If threshold level performance was not achieved within two sessions (one session existed of 4 blocks with 120 trials each), participants had to return at least one week later, in which the stimulus duration was adapted. The fourth, fifth and sixth session comprised theta-burst stimulation of right aPFC, bilateral MT/V5 and sham stimulation (order counterbalanced across participants).

Visual discrimination task

Participants were asked to perform a binary forced-choice visual discrimination task (figure 1A) in front of a BenQ 3D monitor (1920 x 1080 pixels) with a 60Hz refresh rate. Participants were seated approximately 85 centimetres from the centre of the screen in a dimmed room. The stimuli were made in such way that certain parts of the screen shortly moved in opposite direction from a noisy background. This way form-from-motion was created that tests one’s ability to discriminate between visual stimuli and random noise (Wokke et al., 2012; 2014). Each trial started with a blank grey screen (2000 ms) followed by a static rectangle filled with black and white dots (1024 x 768 pixels; 3000 ms). Each dot had the size of one pixel and an equal probability of being black or white. In the centre of the screen a red dot was located, indicating the participant to fixate at the middle of the screen. When the dot turned green, certain parts of screen were displaced five steps towards the right bottom corner of the screen (315°) with steps of one pixel per screen refresh (16 ms). Simultaneously, the entire background was displaced five steps to the left bottom corner of the screen (225°). Form-from-motion face or non-face stimuli were created due to these displacements (respectively left and right column in figure 1B). Non-face stimuli existed of the same moving elements and amount of pixels as the face stimuli, although scrambled. Two levels of difficulty were constructed, one easy condition in which participants performed near ceiling (figure 1B; upper row) and one near threshold (figure 1B; bottom row). Easy and medium stimuli were made in Adobe Photoshop (Adobe Systems Incorporated: www.adobe.com). Easy stimuli contained complete (compassing the entire form of a face) facial contours of about 7 pixels width, whereas medium stimuli had partial contours of approximately 3 pixels width. Medium stimuli also existed of less elements (for example no eyebrows) compared to easy stimuli. Furthermore, contours were similar in medium face and non-face stimuli, although the content differed.

After stimulus presentation a blank screen (1500 ms) was presented to avoid impulsive responses. Participants were asked to respond with button presses whether they saw a face (left index finger) or not (right index finger), followed by a question how they visually experienced the stimulus with a four-point PAS-scale in which 1: no experience and 4: clear experience (Overgaard, 2006). Additionally, they were asked how

4 A

5 C

Figure 1. (A) Task design. Participants were instructed to fixate at the red dot in the middle of the screen. A trial started with a blank grey screen, followed by a rectangle filled with black and white dots. When the fixation dot turned green, certain parts (dark coloured for visualization) of the screen moved 315° while the background moved 225°. As a result, form-from-motion face or non-face stimuli ‘popped out’. The moving parts were displaced five steps in total, with steps of one pixel per screen refresh (16 ms). Participants were instructed to respond with button presses whether they saw a face or not, how their visual experience was of the stimulus and how confident they were of their discrimination choice. (B) Two conditions were made, face (left) and non-face stimuli (right), with two difficulty levels; easy (upper row) and medium (bottom row). (C) TBS conditions. Upper row shows unilateral, right aPFC coordinates. Second row displays bilateral MT stimulation. Third row represents the position of the coil (angled 45° on electrode Cz) in sham condition.

confident they were of their discrimination choice on a four-point scale (1: very uncertain, 2: uncertain, 3: certain and 4: very certain). The instruction was to respond within 5 seconds after stimulus offset. Participants were encouraged to use the entire scales and to keep in mind that 50% of the trials existed of face stimuli and the other of non-face stimuli. In total 80 different stimuli were created from 20 unique faces, randomised within each block with equal probability. After theta-burst

stimulation, participants performed four blocks of the visual discrimination task. Each block contained 60 different stimuli, in which 20 easy trials (10 faces and 10 non-faces) and 40 medium trials (20 faces and 20 non-faces) were shown. Stimuli were presented using Presentation (Neurobehavioral Systems: www.neurobs.com).

fMRI

A functional mapper was obtained to determine specific MT/V5 coordinates necessary for theta-burst stimulation. Participants were presented with 10 blocks of coherently moving dots and 10 blocks of randomly appearing stationary dots. These blocks were randomized and each block had a duration of 16 seconds. During a block of coherent movement, dots moved towards and away from the fixation cross at the centre of the screen every 2 seconds. Changes in blood oxygen level dependent (BOLD) responses during the coherently moving and stationary dots were analysed using FSL (FMRIB’s Software Library: www.fmrib.ox.ac.uk/fsl). Events were convolved with the canonical hemodynamic response function, followed by first level analysis within the general linear model.

To functionally map areas PPA, FFA and LO, participants were presented every 2 seconds with blocks containing stimuli of faces, houses, objects (chairs, bottles and scissors) or phase-scrambled versions of the objects. Each block lasted 16 seconds and was presented four times, in randomised order.

Magnetic resonance images were collected with a 3 Tesla Philips Achieva dStream MRI scanner using a 32-channel head coil. Functional, ascending, GE-EPI scans were performed with voxel size = 2 mm, TR = 2000 ms, TE = 28 ms, FOV = 200 mm, slice thickness = 2.5, slice gap = 0.3, 28 slices, sense factor = 2.5 and a matrix size of 112 x 112. Data were motion and slice time corrected, spatially smoothed with a FWHM of 5 mm Gaussian kernel and high-pass filtered (0.01 Hz). The functional images were normalised and registered with a structural image (T1 turbo field echo, 182 coronal slices, TR = 9.7, TE = 4.6, flip angle = 8, FOV = 256 mm, slice thickness = 1.2 and a matrix of 256 x 256).

Theta-burst stimulation

We used an fMRI-guided navigation system (Visor2tm ANTneuro) to precisely target aPFC and MT/V5. For aPFC, a mask with a 4 mm sphere was made around MNI stereotactic coordinates (-24,

6 50 55 60 65 70 75 80 85 90 95 100 1 2 3 5 6 7 9 10 Perc en ta ge correct Participant

65, 18) (Fleming et al., 2010). Stimulation of unilateral aPFC was based upon a study of Volman and colleagues (2011), who induced bilateral aPFC blood flow reductions by applying cTBS over left aPFC. Based on these findings and the possible unpleasant location (forehead, approximately 5 centimetres above the eye), we decided to stimulate unilateral right aPFC. For MT/V5, individual bilateral MT/V5 masks were created with a 4 mm sphere of the voxels with the highest activation. Stimulation was applied with a pause of 5 minutes between each stimulated site. In the sham stimulation condition the coil was placed approximately at electrode location Cz with an angle of 45°, so the magnetic pulse did not run through the cortex. All participants were stimulated with a fixed stimulator output of 38% for all three locations, ensuring that we did not activate motor regions (Stewart et al., 2001).

Stimulation of bilateral MT/V5 and right aPFC was achieved with a Magstim Rapid2 stimulator (Magstim Company, UK) and a 70-mm figure-of-eight coil. Participants were instructed to sit as comfortable as possible with their head placed in a chinrest. At first, they received one train (3 pulses) at 30% stimulator output to familiarize the participants with theta-burst stimulation, at each specific location. If participants felt comfortable, they received one train of pulses at actual stimulator output; 38%. If stimulation was unpleasant, we re-positioned the coil, although keeping it within a 2.5 millimetre range of the target, and applied test trains until a comfortable position was found. For the actual experiment, trains of three pulses with an interval of 20 milliseconds (50 Hz) were delivered every 200 milliseconds (5 Hz, theta frequency) with a total duration of 40 seconds (Huang et al., 2005). After each stimulation session, a pause of 5 minutes before task onset was given to aim for equal distributed TBS effects, for approximately 45 minutes, throughout the entire experiment.

Analyses

To test the association between the probability of a correct answer and the PAS experience scale, a 2 (difficulty level) x 3 (TBS condition) x 4 (PAS level) repeated measures ANOVA was performed. Our hypothesis was tested with a 2 (difficulty level: easy and medium) x 3 (TBS condition: aPFC, MT and sham) repeated measures ANOVA, separately for within-subjects variables accuracy, first order

performance, metacognitive sensitivity and metacognitive efficiency. Accuracy was calculated with total percentage correct for both face and non-face stimuli. First order performance was measured in type 1 d’, in signal-to-noise units, and reflects one’s ability to discriminate between stimuli or distinguish signal from noise (Galvin et al., 2003). Metacognitive sensitivity, meta d’, is in the same units as d’ and therefore easily to compare (Maniscalco & Lau: http://www.columbia.edu/~bsm2105/type2sdt/). The comparison of meta d’ relative to d’ is defined as metacognitive efficiency. A theoretically ideal metacognitive efficiency will have a value of 0, meta d’ – d’ (Fleming & Lau, 2014). In addition, we tested for overall TBS effects on response bias, separately for both d’ and meta d’ with a 2 (difficulty level: easy and medium) x 3 (TBS condition: aPFC, MT and sham) x 3 (criterion for metacognitive responses) repeated measures ANOVA, to test whether TBS might cause one to become more conservative or liberal. For first order performance, we tested only for medium trials, since participants performed near ceiling on easy trials, leading to little bias towards face or non-face stimuli.

Figure 2. Accuracy scores of each participant in sham condition for medium trials, with 95% confidence interval. All participants performed near threshold level (blue line: 75%). No significant differences were found between accuracy and threshold level, p > 0.05.

7 60 62 64 66 68 70 72 74 76 78 80 aPFC MT Sham Perc en ta ge correct

Medium

In total, eight participants (two men, mean age = 23, SD = 1.93) were included in the analyses. One participant was excluded because his accuracy and first order performance scores deviated more than 2 SD from the mean at his first session. We aimed for each participant to perform near threshold level (~75%), to optimally test the effects of TBS. Three participants performed near chance level at medium trials during sham stimulation. Two participants were willing to return for a second sham session to achieve threshold performance. One participant was unable to return, leading to exclusion of this participant. A one-sample t-test to test whether accuracy in sham stimulation differed from threshold level yielded no significant differences, t(7) = -0.452, p > 0.05 (figure 2).

Stimulus experience

Participants were asked to rate their perceived experience of the stimulus with a four-point PAS scale (Overgaard, 2006). To test the association between the probability of a correct answer and one’s perceived experience, a 2 (difficulty level: easy and medium) x 3 (TBS condition: aPFC, MT and sham) x 4 (PAS scale) repeated measures ANOVA was performed with within subjects variable accuracy. A main effect of experience (F(3,24)=

81.39, p < 0.001) and an interaction effect of experience x TBS condition were found (F(3, 24)=

12623, p < 0.001). These findings indicate that the better one perceives the stimulus, the higher the probability of a correct answer (see supplementary data, figure S1). Post-hoc analyses revealed a significant difference between the first and second, and second and third PAS level (p < 0.05).

For the interaction effect, post-hoc analyses showed a significant difference between the first two PAS levels. However, this is likely due to the fact that these minor responses were correct guesses.

To further explore the differences between the second and third PAS level for each TBS condition, we conducted an exploratory t-tests. This revealed significant differences between the second and third PAS level within each TBS condition for all medium trials (p < 0.05), whereas easy trials revealed no significant differences. We looked further into these medium trials and tested whether the value of the second PAS level relative to the third PAS level was significantly different between TBS conditions. These paired t-tests revealed no significant differences, p > 0.05. In sum, a significant difference between PAS 2 and PAS 3 were found within each TBS condition, but this difference was not significant between TBS conditions. Thus, TBS does not have a significant effect on perceived stimulus experience. However as argued, this analysis does not take response bias into account and is therefore not an optimal, bias-free approach.

Accuracy

The effect of TBS on one’s accuracy, overall percentage correct for both face and non-face stimuli, was tested with a 2 (difficulty level: easy and medium) x 3 (TBS condition: aPFC, MT and sham) repeated measures ANOVA with dependent variable accuracy. A significant main effect of difficulty level was found (F(1,5182) = 169, p < 0.001).

No significant effect of TBS on accuracy was found, p = 0.163. Results are displayed in figure 3.

Figure 3. Overall accuracy scores per TBS condition for easy (left) and medium (right) trials with 95% confidence interval. A significant main effect of difficulty level was found, p < 0.001.

80 82 84 86 88 90 92 94 96 98 100 aPFC MT Sham Perc en ta ge correct

Easy

8 Despite the fact that we did not found an

interaction effect between difficulty level and TBS condition, we performed some exploratory analyses. A one-way ANOVA for accuracy did not reveal any significant effects of TBS (p>0.05). Interestingly, although not validated nor confirmative, three separate t-tests to test for differences between TBS locations revealed a significant difference in accuracy for sham (mean = 93.80, SD = 2.86) and MT (mean = 91.41, SD = 2.63) conditions (t(7) = -3.8, p < 0.01). This indicates an effect of MT stimulation on accuracy, however these exploratory findings should be treated with caution.

First order performance

To test the effect of TBS on one’s ability to discriminate between stimuli, a 2 (difficulty level: easy and medium) x 3 (TBS condition: aPFC, MT and sham) repeated measures ANOVA with d’ as a within subjects dependent variable was performed. A significant main effect of difficulty level was found (F(1, 42) = 157, p < 0.001). No

significant effect of TBS on first order performance was found (p = 0.224). Results are displayed in figure 4.

Although the ANOVA did not reveal any significant interactions, we conducted several exploratory analyses. Firstly, a one-way ANOVA for independent variable TBS condition, which revealed no significant effects of TBS on d’.

However, when performing three separate t-tests between TBS conditions, a significant difference in d’ for sham (mean = 3.32, SD = 0.62) and MT (mean = 2.93, SD = 0.40) conditions (t(7) = -2.69, p < 0.05) for easy trials was found. This might indicate an effect of MT stimulation on first order performance for easy trials. However, these findings are exploratory and should be interpreted with great caution.

Metacognitive sensitivity

A 2 (difficulty level: easy and medium) x 3 (TBS condition: aPFC, MT and sham) within subjects repeated measures ANOVA was conducted to test for effects of TBS on the ability to distinguish between one’s own correct and incorrect responses, meta d’. A significant difference between easy and medium trials was found (F(1, 49)

= 110.7, p < 0.001). No significant effect of TBS on metacognitive sensitivity was found, p = 0.907 (figure 5).

Metacognitive efficiency

The effects of TBS on metacognitive efficiency, the amount of first order information available for metacognition, was tested with a 2 (difficulty level: easy and medium) x 3 (TBS condition: aPFC, MT and sham) repeated measures ANOVA with dependent variable meta d’ – d’. No significant effects of difficulty level nor TBS on metacognitive efficiency were found, p = 0.452 (figure 6).

Figure 4. Overall first order performance per TBS condition for easy (left) and medium (right) trials with 95% confidence interval. A significant main effect of difficulty level was found, p < 0.001.

2 2,2 2,4 2,6 2,8 3 3,2 3,4 3,6 3,8 4 aPFC MT Sham d'

Easy

Medium

9 -1,5 -1 -0,5 0 0,5 1 1,5 aPFC MT Sham Me ta d ' -d'

Medium

Easy

First, we tested whether TBS had an effect on participants’ first order response bias for medium trials only. A one-way ANOVA with independent variable TBS (aPFC, MT and sham) was performed. No significant main effect of TBS was found,

F(2,21)=1.005, p = 0.383 (see supplementary data;

figure S2).

A separate ANOVA was performed to test for TBS effects on metacognitive response biases (supplementary data; figure S3). A 2 (difficulty level: easy and medium) x 3 (TBS condition: aPFC, MT and sham) x 3 (criterion) repeated measures ANOVA for metacognitive decisions (four-point confidence scale) towards faces were analysed between TBS conditions. As expected, significant

main effects of difficulty level and criterion were found and therefore an interaction effect of difficulty level x criterion, p < 0.05. This interaction effect revealed a significant difference for easy and medium trials between the second and third criterion. We did not expect to find a difference of the first criterion, since this criterion represents the difference between very uncertain and uncertain, so responses made with low confidence were not expected to be affected by difficulty level. Discussion

The aim of this study was to investigate the neural mechanisms related to first and second (metacognitive) order decision-making. We therefore manipulated activity in MT/V5 and aPFC, Figure 5. Overall metacognitive sensitivity per TBS condition for easy (left) and medium (right) trials with 95% confidence interval. A significant main effect of difficulty level was found, p < 0.001.

Figure 6. Overall metacognitive efficiency per TBS condition for easy (left) and medium (right) trials with 95% confidence interval. No significant effects of difficulty level nor TBS condition were found, p > 0.05.

10 highly associated with first order and

metacognitive decision-making respectively, with a recently developed theta-burst stimulation protocol. As expected, an overall significant effect of difficulty level was found, in which participants had lower accuracy, first order performance and metacognitive sensitivity scores on medium difficulty level trials compared to easy difficulty level trials. First, we tested whether perceived experience of the stimulus was associated with the probability of a correct answer. We found a significant effect of experience level, indicating that a clear perceived experience resulted in a higher probability of giving a correct answer. However, as reasoned before, this is not a valid approach since confounding factors, such as response bias, were not taking into account. When allowing for confounding factors while testing the effect of TBS on accuracy, first and metacognitive performance, no significant differences were found. Finally, no significant effects of TBS on response bias for both first and metacognitive decisions were found either. This indicates that participants did not become more or less conservative nor liberal due to TBS. Taken all these findings together, our hypothesis that metacognition is an extra, higher order process, independent from first order decision-making is not supported by the data.

Form-from-motion stimuli

The lack of TBS effects found for accuracy, first and metacognitive performance could have been the result of various confounding factors. Exploratory analyses of first order performance and accuracy revealed a significant difference between sham and MT stimulation, for easy trials only, indicating that there was an effect of TBS on MT/V5. Previous studies have shown that stimulation of MT/V5 does affect first order performance in a motion detection paradigm using similar stimuli (Wokke et al., 2012; 2014). However, we did not found such an effect on medium trials. An explanation can be found in the mechanisms that underlie form-from-motion processing. The traditional view is that form-from-motion stimuli are processed via three different stages, starting in LO which is selective for human object recognition (Grill-Spector et al., 1998), then MT/V5 is selective for motion (Newsome et al., 1988; Stoner & Albright, 1992; Britten et al., 1996), followed by inferior temporal cortex (IT) for shapes (Sary et al., 2003). Therefore, partial stimulation of this processing stream by stimulating only one mechanism (MT/V5) could

have led to insufficient manipulation of form-from-motion processing. However, Wokke et al. (2014) demonstrated that stimulation of LO and MT/V5 had opposing effects on task performance, suggesting competitive visual processing streams. In our study, activity in LO could have influenced the form-from-motion processing, while MT/V5 activity was inhibited. This might indicate that MT/V5 alone might not be the optimal location to manipulate first order decision-making in this paradigm.

However, this does not clarify the exploratory effect we found of MT/V5 stimulation on easy trials for first order performance. An explanation could be that medium stimuli were differently processed than easy stimuli. For instance, participants could have used a different strategy for responding to medium stimuli. This could be caused by the limited amount of stimuli we used. Twenty easy faces were diversified into medium face, easy non-face and medium non-face stimuli. Despite the fact that we programmed the task in such way that the same faces did not occur in one block, it could have been the case that participants recognized certain faces in the medium condition. Participants performed the task at least six times (two practice sessions and three TBS sessions). These repetitions could have led to the phenomenon that participants memorized some of the medium stimuli and therefore always made the same errors and correct responses. This could resulted in a strategy switch for discriminating between medium face and non-face stimuli. Thus, form-from-motion processing was less efficient used during medium trials, which could explain the differences in stimulation effects found between medium and easy trials. This explanation is supported by the large variability found in MT condition for accuracy, first and metacognitive performance on medium trials, which can be explained by the errors made in recalling the stimuli.

Theta-burst stimulation of prefrontal cortex

The fact that we did not find any effects of TBS on metacognition could have been the result of inefficient stimulation of aPFC. We stimulated right unilateral aPFC, based upon the study of Volman and colleagues (2011) who induced bilateral blood flow reductions by stimulating only right aPFC. We combined this procedure with functional based MNI coordinates from an fMRI study who investigated metacognitive sensitivity (Fleming et al., 2010). It could be possible that due to

11 anatomical differences, we insufficiently

stimulated the entire aPFC. Or that stimulation did not reach the cortex at all. Wagner et al. (2009) showed that the effectiveness of TBS stimulation depends on the alignment of sulci and gyri. Therefore, the angle and position of the TBS coil is crucial for the magnetic pulses to effectively reach the cortex. Due to individual differences, for example a large frontal sinus, it is possible that aPFC of some participants in our study was ineffectively stimulated. It is more accurate to create individual bilateral aPFC masks based on functional mappers instead of MNI coordinates. This way, one can precisely target individual aPFC and account for anatomical differences, for example the folding of the cortex. In addition, other individual differences such as grey matter volume or white matter microstructure have been demonstrated to correlate with metacognitive performance (Fleming et al., 2010). These individual differences in prefrontal regions could have affected the varied metacognitive scores found in this study (Kanai & Rees, 2011)

Despite the evidence of aPFC involved in metacognition (Fleming et al., 2012; 2014), other studies stimulated DLPFC and did found significant effects of TBS on metacognition (Rounis et al., 2010). It is plausible that, taken the highly interconnected frontal cortex into account, stimulation of DLPFC indirectly affected aPFC and thus impaired metacognitive processes. It is well-studied that DLPFC is involved in multiple executive functions such as attention, working memory, inhibition and planning (Beck et al., 2001; Curtis and D’Esposito, 2003; Turatto et al., 2004). The significant effect of DLPFC stimulation found in metacognition, could have been the result of impairment of one of the previous mentioned processes instead of targeting metacognitive processes.

In the last ten years, metacognition is hotly debated and therefore new methods and protocols are developed to optimally investigate the neural correlates of metacognition. This pilot study provides new insights in stimulation protocols and task designs that can be used to unravel the neural mechanisms related to first and second order decision-making. Our findings should be considered as very preliminary results and interpreted with great care because of the above mentioned factors. Future studies are needed to elaborate these findings and methods.

References

Beck, D. M., Rees, G., Frith, C. D., & Lavie, N. (2001). Neural correlates of change detection and change blindness. Nature neuroscience, 4(6), 645-650. Britten, K. H., Newsome, W. T., Shadlen, M. N., Celebrini, S., & Movshon, J. A. (1996). A relationship between behavioral choice and the visual responses of neurons in macaque MT. Visual

neuroscience, 13(01), 87-100.

Chiang, T. C., Lu, R. B., Hsieh, S., Chang, Y. H., & Yang, Y. K. (2014). Stimulation in the Dorsolateral Prefrontal Cortex Changes Subjective Evaluation of Percepts.

Curtis, C. E., & D'Esposito, M. (2003). Persistent activity in the prefrontal cortex during working memory. Trends in cognitive sciences, 7(9), 415-423.

Del Cul, A., Dehaene, S., Reyes, P., Bravo, E., & Slachevsky, A. (2009). Causal role of prefrontal cortex in the threshold for access to consciousness.

Brain, 132(9), 2531-2540.

Fetsch, C. R., Kiani, R., Newsome, W. T., & Shadlen, M. N. (2014). Effects of cortical microstimulation on confidence in a perceptual decision. Neuron, 83(4), 797-804.

Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cognitive– developmental inquiry. American psychologist, 34(10), 906.

Fleming, S. M., Weil, R. S., Nagy, Z., Dolan, R. J., & Rees, G. (2010). Relating introspective accuracy to individual differences in brain structure. Science,

329(5998), 1541-1543.

Fleming, S. M., & Dolan, R. J. (2012). The neural basis of metacognitive ability. Philosophical

Transactions of the Royal Society of London B: Biological Sciences, 367(1594), 1338-1349.

Fleming, S. M., & Lau, H. C. (2014). How to measure metacognition. Frontiers in human neuroscience, 8. Fleming, S. M., Ryu, J., Golfinos, J. G., & Blackmon, K. E. (2014). Domain-specific impairment in metacognitive accuracy following anterior prefrontal lesions. Brain, 137(10), 2811-2822.

12 Grill-Spector, K., Kushnir, T., Edelman, S., Itzchak,

Y., & Malach, R. (1998). Cue-invariant activation in object-related areas of the human occipital lobe.

Neuron, 21(1), 191-202.

Hebart, M. N., Schriever, Y., Donner, T. H., & Haynes, J. D. (2014). The relationship between perceptual decision variables and confidence in the human brain. Cerebral Cortex, bhu181.

Huang, Y. Z., Edwards, M. J., Rounis, E., Bhatia, K. P., & Rothwell, J. C. (2005). Theta burst stimulation of the human motor cortex. Neuron, 45(2), 201-206.

Kanai, R., & Rees, G. (2011). The structural basis of inter-individual differences in human behaviour and cognition. Nature Reviews Neuroscience, 12(4), 231-242.

Kiani, R., & Shadlen, M. N. (2009). Representation of confidence associated with a decision by neurons in the parietal cortex. Science, 324(5928), 759-764.

Macmillan, N. A., & Creelman, C. D. (1991). Detection theory: A user's guide. Cambridge University Press. New York.

Maniscalco, B., Rounis, E., Rothwell, J. C., Passingham, R. E., & Lau, H. (2009). Theta-burst transcranial magnetic stimulation to the prefrontal cortex impairs metacognitive visual awareness. Journal of Vision, 9(8), 764-764. Maniscalco, B., & Lau, H. (2012). A signal detection theoretic approach for estimating metacognitive sensitivity from confidence ratings. Consciousness

and cognition, 21(1), 422-430.

Maniscalco, B. (2014). High-level cognitive and

neural contributions to conscious experience and metacognition in visual perception (Doctoral

dissertation, Columbia University).

Nelson, T. O., Gerler, D., & Narens, L. (1984). Accuracy of feeling-of-knowing judgments for predicting perceptual identification and relearning. Journal of Experimental Psychology:

General, 113(2), 282.

Newsome, W. T., & Pare, E. B. (1988). A selective impairment of motion perception following lesions

of the middle temporal visual area (MT). The

Journal of Neuroscience, 8(6), 2201-2211.

Overgaard, M., Rote, J., Mouridsen, K., & Ramsøy, T. Z. (2006). Is conscious perception gradual or dichotomous? A comparison of report methodologies during a visual task. Consciousness

and cognition, 15(4), 700-708.

Pollen, D. A. (1995). Cortical areas in visual awareness. Nature, 293 -294.

Rossi, S., Hallett, M., Rossini, P. M., Pascual-Leone, A., & Safety of TMS Consensus Group (2009). Safety, ethical considerations, and application guidelines for the use of transcranial magnetic stimulation in clinical practice and research. Clinical neurophysiology, 120(12), 2008-2039.

Rounis, E., Maniscalco, B., Rothwell, J. C., Passingham, R. E., & Lau, H. (2010). Theta-burst transcranial magnetic stimulation to the prefrontal cortex impairs metacognitive visual awareness. Cognitive Neuroscience, 1(3), 165-175. Sary, G., Vogels, R., & Orban, G. A. (1993). Cue-invariant shape selectivity of macaque inferior temporal neurons. Science, 260(5110), 995-997. Schmitz, T. W., Rowley, H. A., Kawahara, T. N., & Johnson, S. C. (2006). Neural correlates of self-evaluative accuracy after traumatic brain injury.

Neuropsychologia, 44(5), 762-773.

Stewart, L. M., Walsh, V., & Rothwell, J. C. (2001). Motor and phosphene thresholds: a transcranial magnetic stimulation correlation study.

Neuropsychologia, 39(4), 415-419.

Stoner, G. R., & Albright, T. D. (1992). Neural correlates of perceptual motion coherence. Nature, 358(6385), 412-414.

Timmermans, B., Schilbach, L., Pasquali, A., & Cleeremans, A. (2012). Higher order thoughts in action: consciousness as an unconscious re-description process. Philosophical Transactions of

the Royal Society B: Biological Sciences, 367(1594),

1412-1423.

Turatto, M., Sandrini, M., & Miniussi, C. (2004). The role of the right dorsolateral prefrontal cortex in

13 visual change awareness. Neuroreport, 15(16),

2549-2552.

Volman, I., Roelofs, K., Koch, S., Verhagen, L., & Toni, I. (2011). Anterior prefrontal cortex inhibition impairs control over social emotional actions.

Current Biology, 21(20), 1766-1770.

Wagner, T., Rushmore, J., Eden, U., & Valero-Cabre, A. (2009). Biophysical foundations underlying TMS: setting the stage for an effective use of neurostimulation in the cognitive neurosciences. Cortex, 45(9), 1025-1034.

Wassermann, E. M. (1998). Risk and safety of repetitive transcranial magnetic stimulation: report and suggested guidelines from the International Workshop on the Safety of Repetitive Transcranial Magnetic Stimulation,

Electroencephalographic Clinical Neurophysiology,

108, 1–16.

Wokke, M. E., Sligte, I. G., Steven Scholte, H., & Lamme, V. A. (2012). Two critical periods in early visual cortex during figure–ground segregation. Brain and behavior, 2(6), 763-777. Wokke, M. E., Scholte, H. S., & Lamme, V. A. (2014). Opposing dorsal/ventral stream dynamics during figure-ground segregation. Journal of cognitive

Appendix A. Supplementary data

Figure S1. Percentage correct per PAS level for each TBS condition. A main effect of experience and an interaction effect of experience x TBS condition were found, p < 0.001.

Figure S2. Response bias for first order performance per TBS condition for medium trials with 95% confidence interval. No significant effect of TBS was found, p > 0.05.

Figure S3. Response bias for metacognitive performance for easy (left) and medium (right) trials with 95% confidence interval. Significant main effects of difficulty level and criterion were found. An interaction effect revealed a significant difference between the second and third criterion for easy and medium trials, p < 0.05.

0 0,5 1 1,5 2 2,5 3 aPFC MT sham Re sp o n se b ias Medium -2 -1,5 -1 -0,5 0 C1 C2 C3 Re sp o n se b ias

Easy

Medium