Figure 1.
Experimental stimuli and procedure.
(A) Experimental procedure. Part I: subjects placed self-paced bids for the opportunity to eat or avoid eating 60 different food items. Part II: subjects performed a liking-rating task for the same food items via a 4AFC button press (Strong Dislike, Weak Dislike, Weak Like, Strong Like) while their EEG responses were recorded. Part III: a randomly selected bidding trial was implemented that determined which food, if any, subjects had to eat. (B) Sample stimuli, bidding task screen. (C) Sample stimuli and liking-rating task screen (Part II).
Figure 2.
(A) Comparison of bids and liking ratings. Although individual subjects (markers) varied in their utilization of the bidding range, there was a strong logistic relationship between bid values and preference ratings (thick line, average of individual fits). (B) Aggregate response histogram. (C) Average median RTs by liking rating.
Figure 3.
Artifact removal via independent component analysis (ICA).
(A) Trial-by-trial activity from a frontal sensor (E17, left) in a subject (PXM) with excessive blinking, sorted by reaction time. This subject's pattern of blinking after the response creates large positive deflections at this sensor following key press (black line). (B) Topographic scalp representations at 1042 ms (left) show a stereotyped frontal pattern consistent with eye blink, visible as a large deflection in the ERP (right). (C) ICA analysis extracts a component corresponding to the eye blinks, as indicated here by the scalp topography. Such components are also visible in the data of other subjects with less obvious artifacts in their EEG data (7 subjects shown here). The polarity of the independent component is arbitrary. (D) PXM's data following ICA cleaning. In contrast to the raw data, the major peaks now reflect brain-related activity. (E) Corresponding brain-related ICA components in PXM and the other depicted subjects.
Figure 4.
Value signals in the stimulus-locked data.
(A) Heat map summarizing the results of the linear regression analysis in the 128 sensors and 18 time windows (each 50 ms long). Each box indicates the p-value for a mixed-effects group estimate of the effect of liking rating on the activity of a specific sensor during a specific time window. P-values are corrected for multiple comparisons using a permutation test. (B) ERP responses in parietal sensors during the 150 to 250 ms time window. (C) ERP responses in central and frontotemporal sensors during the 400 to 550 ms time window. (D) ERP responses in frontal sensors during the 700 to 800 ms time window. In each case, the left maps show the scalp distributions of t-values for the liking rating variable in the linear regression in two- (top) and three-dimensional (bottom) projections. The sensors of interest (SOIs) from which the ERP is extracted are shown in the white boxes. ERPs were computed by averaging waveforms first within and then across subjects. In each case the ERP shows a parametric response that correlates with the liking ratings during the time window highlighted in gray. In the 400–550 ms window, the SOIs also included bilateral frontal and lateral sensors with high negative t values.
Figure 5.
Validation of the distributed source reconstruction technique.
Note that the spatial resolution of reconstruction is limited due to inherent constraints and the realignment of subjects into a common neuroanatomical space. Depicted spatial precision reflects the high statistical threshold used: F≥30, corresponding to p<0.00005 (uncorrected). (A) Visual evoked potential (VEP). Top: Topography and waveform of the VEP recorded from occipital sensors. The time window used for source reconstruction (100–150 ms) is shown in gray. Bottom: Sources of the VEP included visual areas such as primary visual cortex (V1), middle and inferior occipital gyrus, and precuneus (PCun). Activity from one maximal intensity projection (MIP) (circled) is highly similar to the original EEG data. (B) Movement-related cortical motor potential. Top: Topography and waveform of the motor potential recorded from parietal sensors, for the comparison of contralateral versus ipsilateral motor output. Bottom: Time courses of sources localized to M1 (circled) show a similar contra-ipsi pattern.
Figure 6.
Exploratory distributed source reconstruction, stimulus-locked data.
Depicted spatial precision reflects the high statistical threshold used: p<0.05, FWE-corrected. (A) Glass brain displays showing the maximum intensity projections (MIP) of all sources for each time window. (B) Representative sources of interest for each time window superimposed on the MNI brain. Top: 150 to 250 ms. Prominent sources include: fusiform and lingual gyrus (FG, LG), middle and superior temporal gyrus (MTG, STG), and parahippocampal gyrus; insula (Ins), associated with gustatory processing; and inferior frontal gyrus (IFG). Middle: 400 to 450 ms. While posterior sources are still visible, sources associated with value computation emerge within this time window, including ventral striatum (VStr), Brodmann area 46 (BA46), and ventromedial prefrontal cortex (vmPFC). Bottom: 700 to 800 ms. In this time window, overlapping with average median RT (710 ms), there is continued activity in subgenual cingulate (Cg25) and intraparietal sulcus (IPS), as well as emerging activity in middle and superior frontal gyrus (MFG, SFG). See also Tables S1, S2, S3.
Figure 7.
Preliminary Granger causal connectivity analysis.
Note that due to the coarse spatial resolution of the source reconstruction technique, the selected regions of interest (ROIs) are not precise anatomical regions but rather represent broad divisions of cortex. (A) Model and causal connectivity across 3 time windows of interest (orange: 250–500 ms, cyan dotted: 550–700 ms, magenta dashed: 800–950 ms). All displayed connections are significant for a threshold of p = 0.01, Bonferroni-corrected. (B) Causal flow measures of the difference between the number of outgoing and incoming causal connections. Large positive values indicate “causal sources,” nodes with strong causal influence on the system, while negative values signify “causal sinks.” (C) Unit causal density measures of the summed causal interactions for a given node. High unit causal density indicates the presence of a hub in the causal network.
Figure 8.
Value signals in the response-locked data.
(A) Heat map summarizing the results of the linear regression analysis in the 128 sensors and 14 40-ms time windows. Each box indicates the p-value for a mixed-effects group estimate of the effect of liking rating on the activity of a specific sensor during a specific time window. P-values are corrected for multiple comparisons using a permutation test. (B) Top: Scalp distribution of t-values for the liking rating variable over time. Bottom: ERP grand average waveforms for preference, −400 to −320 ms pre-response. (Because significant effects of preference were largely restricted to the same electrodes over time, only this time window is displayed as an example.) The ERP shows a parametric response that correlates with the liking ratings during the time window highlighted in gray. (C) Exploratory distributed source reconstruction for the response-locked data. Identified sources were similar to those in the stimulus-locked analysis. Activity was also localized to anterior and dorsal cingulate cortex (ACC, dACC) and premotor cortex (PM), regions known to be engaged in response selection and planning.
Figure 9.
Stimulus-locked source reconstruction maps (FWE-corrected p<0.01) versus mOFC activations in 3 published fMRI studies.
The source reconstruction of the EEG data from 400–550 ms (red) and 700–800 ms (yellow) shows vmPFC sources consistent with previous results from fMRI (green, cyan, and blue). Notably, the closest match with the EEG data comes from Litt et al. (2010), who used a similar evaluation task with appetitive and aversive stimuli.