Motivational signals disrupt metacognitive signals in the human ventromedial prefrontal cortex

Abstract

A growing body of evidence suggests that, during decision-making, BOLD signal in the ventromedial prefrontal cortex (VMPFC) correlates both with motivational variables - such as incentives and expected values - and metacognitive variables - such as confidence judgments - which reflect the subjective probability of being correct. At the behavioral level, we recently demonstrated that the value of monetary stakes bias confidence judgments, with gain (respectively loss) prospects increasing (respectively decreasing) confidence judgments, even for similar levels of difficulty and performance. If and how this value-confidence interaction is reflected in the VMPFC remains unknown. Here, we used an incentivized perceptual decision-making fMRI task that dissociates key decision-making variables, thereby allowing to test several hypotheses about the role of the VMPFC in the value-confidence interaction. While our initial analyses seemingly indicate that the VMPFC combines incentives and confidence to form an expected value signal, we falsified this conclusion with a meticulous dissection of qualitative activation patterns. Rather, our results show that strong VMPFC confidence signals observed in trials with gain prospects are disrupted in trials with no - or negative (loss) - monetary prospects. Deciphering how decision variables are represented and interact at finer scales seems necessary to better understand biased (meta)cognition.

Fig. 1. Experimental design and behavioral results.

a Experimental paradigm. Participants viewed two Gabor patches on both sides of the screen (150 ms) and then chose which had the highest contrast (left/right, self-paced). After a jitter of a random interval between 4500 and 6000 ms, the incentive condition was shown (900 ms; green frame for win trials, gray frame for neutral trials, red frame for loss trials). Afterwards, participants were asked to report their confidence in the earlier made choice on a scale ranging from 50% to 100% with steps of 5%. The initial position of the cursor was randomized between 65% and 85%. Finally, subjects received feedback. The inter-trial interval (ITI) had a random duration between 4500 and 6000 ms. The calibration session only consisted of Gabor discrimination, without confidence rating, incentives, or feedback, and was used to adjust the difficulty so that every individual reached a performance of 70%. b Behavioral results. Individual-averaged accuracy (left), reaction times (middle) and confidence (right) as a function of incentive condition (−100/red, 0/gray, +100/green). Colored dots represent individuals (N = 32), gray lines highlight within-subject variation across conditions. Error bars represent sample mean ± standard error of the mean. Note that for confidence and accuracy, we computed the average per incentive level per individual, but that for reaction times, we computed the median for each incentive condition rather than the mean due to their skewed distribution. c Linear mixed-effect model (LMEM) results. The graph depicts fixed-effect regression coefficients (β) for incentive condition (Inc.) and absolute incentive condition (|Inc.|) predicting performance (top), reaction times (middle), and confidence (bottom). Error bars represent standard errors of fixed effects. *p < 0.05.

Fig. 2. Overview of general linear models for fMRI analyses.

a–b Events of interest. The timeline depicts the succession of events within a trial. a Yellow boxes highlight the two events/timing of interest (stimulus/choice and incentive/confidence), that are modeled as stick function for the functional magnetic resonance imaging (fMRI) analysis. We also modeled the feedback event as a stick function. c General linear models (GLMs) parametric regressors specification. The graph displays the different combinations of parametric modulators of each event of interest for all GLMs used to analyze the fMRI data.

Fig. 3. Whole-Brain fMRI Results.

a–c Whole-brain statistical blood-oxygen-level-dependent (BOLD) activity correlating with general linear model 1 (GLM1) “early certainty” (a), incentives (b), and confidence (c). d Whole-brain statistical maps of BOLD activity correlating with GLM3 “expected value”. N = 30. Unless otherwise specified, all displayed clusters survived p < 0.05 family-wise error (FWE) cluster correction. Voxel-wise cluster-defining threshold was set at p < 0.001, uncorrected. Red/yellow clusters: positive activations. Blue clusters: negative activations. For whole-brain activation tables see Supplementary Data 1.

Fig. 4. Activation in ventromedial prefrontal cortex across models.

a Anatomical ventromedial prefrontal cortex (VMPFC) region of interest (ROI). b–d Comparison of VMPFC activations to different specifications of early certainty during the choice moment (b), incentives during incentive/rating moment (c), and confidence during incentive/rating moment (d), as implemented in the different GLMs. Dots represent individual activations; bar and error bars indicate sample mean ± standard error of the mean. Gray lines highlight within-subject variation across the different specifications. N = 30. Cert early certainty; Inc. incentives; conf. confidence; EV expected value. Diamond-ended horizontal bars indicate the results of repeated-measure ANOVAs. Dash-ended horizontal bars indicate the result of post hoc paired t tests. ~p < 0.10; *p < 0.05; **p < 0.01; ***p < 0.001.

Fig. 5. Activation in ventromedial prefrontal cortex across incentives and time points.

a Qualitative ventromedial prefrontal cortex (VMPFC) activation patterns predicted under different models. The different boxes present how blood-oxygen-level-dependent (BOLD) signal should vary with increasing confidence in the three incentive conditions (green: +100; gray: 0; red: −100), under different hypotheses (i.e., encoding different variables), at different time points. Bar graphs in insets summarize these relationships as expected intercepts (or baseline—top) and slope (bottom). b–c VMPFC region of interest (ROI) analysis (N = 30). T values corresponding to baseline and regression slope were extracted in the three incentive conditions, and at the two time points of interest (b: stimulus/choice; c: incentive/rating). Dots represent individual activations; bar and error bars indicate sample mean ± standard error of the mean. Gray lines highlight within-subject variation across the different incentive conditions. Diamond-ended horizontal bars indicate the results of repeated-measure ANOVAs. Dash-ended horizontal bars indicate the result of post hoc paired t tests. ns: P > 0.05; *P < 0.05; **P < 0.01; ***P < 0.001.

Fig. 6. Activation in ventromedial prefrontal cortex across Y and Z dimensions.

a Large anatomical medial prefrontal cortex region of interest (ROI). The Y (blue) and Z (yellow) arrows indicate the dimensions over which the signal is extracted and marginalized—respectively, corresponding to the postero-anterior axis and ventro-dorsal axis. b–c MPFC region of interest (ROI) analysis of confidence activations, at the voxel-level, marginalized over the Y (b) and Z (c) dimensions. Voxel-wise T values corresponding to regression slope were extracted in the three incentive conditions (green: +100; gray: 0; red: −100), and at the two time points of interest (left: stimulus/choice; right: incentive/rating), averaged over two dimensions and plotted as a function of the third dimension. Dots and error bars indicate sample mean ± standard error of the mean (N = 30).