Brain Networks Sensitive to Object Novelty, Value, and Their Combination

Abstract

Novel and valuable objects are motivationally attractive for animals including primates. However, little is known about how novelty and value processing is organized across the brain. We used fMRI in macaques to map brain responses to visual fractal patterns varying in either novelty or value dimensions and compared the results with the structure of functionally connected brain networks determined at rest. The results show that different brain networks possess unique combinations of novelty and value coding. One network identified in the ventral temporal cortex preferentially encoded object novelty, whereas another in the parietal cortex encoded the learned value. A third network, broadly composed of temporal and prefrontal areas (TP network), along with functionally connected portions of the striatum, amygdala, and claustrum, encoded both dimensions with similar activation dynamics. Our results support the emergence of a common currency signal in the TP network that may underlie the common attitudes toward novel and valuable objects.

Keywords: dynamic causal modeling; fMRI; macaque; novelty; resting state connectivity; value.

Figure 1

Stimuli and experimental paradigms. (A) Value training sessions included repeated association of abstract fractal objects with low or high rewards (good and bad fractals, respectively) for >10 days. (B) Perceptual familiarization sessions included repeated exposure of abstract fractal objects in passive viewing and in the absence of reward. (C) Example fractals used as good/bad and novel/familiar categories. (D) Schematic of independent novelty and value dimensions. (E) Test of GB and NF coding in fMRI in a passive viewing task using a block design. In all blocks, subject kept central fixation. In the base blocks, no object was shown. In the probe blocks, objects from one category (good/bad in GB scans and novel/familiar in NF scans, pseudorandomly ordered through sessions) were shown on the left or right hemifield at 6° eccentricity.

Figure 2

Cortical regions with significant novelty and value coding. (A) Cortical regions’ beta coefficients in significantly responsive voxels for GB contrast (left: value coding) and NF contrast (right: novelty coding) in monkey D shown in standard space (Reveley et al. 2017) (P < 0.001, α < 0.01 cluster corrected). (B) Same format as A but for Monkey U (P < 0.001, α < 0.01 cluster corrected). (C) Average GB (horizontal axis) and NF (vertical axis) beta coefficients for top 10 visually activated voxels across the entirety of a cortical area with at least one cluster of activated voxels in either GB or NF scans in monkey D. Each point represents one cortical region. A number of key areas are annotated. The vertical and horizontal gray dashed lines specify mean values of NF coefficients and GB coefficients across all activated cortical regions, respectively. The error bars indicate the standard error of the mean (s.e.m) across novelty and value dimensions. (D) Same format as C but for monkey U.

Figure 3

Resting state cortical networks account for novelty and value coding for member areas. (A) Schematic of the functional connectivity analysis steps based on resting correlation. The first base blocks from all of the runs were selected and concatenated after regression of nuisance parameters and band-pass filtering. The representative graph is extracted via PC-stable algorithm (see Methods). Having estimated the weighted undirected graph, spectral clustering was used to partition the graph into 4 networks. (B) Functional connectivity graph segmented into 4 networks using unsupervised spectral clustering in monkey D. Occipital: yellow, TP: green, ventral-IT: red, and parietal: blue. (C) The 4 networks for monkey D color-coded and shown on standard brain surface (Reveley et al. 2017). (D) The NF/GB distribution of beta coefficients in all subcortical and cortical areas (whether or not significant) was considered jointly as the null distribution and iso-probability contours marking the 68% (dashed line) and 95% (dotted line) interval of the joint distribution were drawn. Cortical voxels falling outside these confidence intervals were plotted and colored according to network membership (95% used for ventral-IT and TP networks and 68% used for parietal network). The marginal distributions of beta coefficients for the null distribution and for the 3 networks are also depicted across the GB and NF axes. (E–G) same format for monkey U.

Figure 4

Similarity of time-course and dynamics of NF/GB activations in the TP network. (A) Average MION signal correlation coefficient between NF and GB time-courses in voxels in each network. Error bars indicate s.e.m across voxels. (B) DCM model of the prefrontal-temporal interaction during good and novel object presentation. Bayesian model selection was used to arrive at the best DCM model for each monkey. For both monkeys the best model had contralateral visual input to the temporal node. The self-connections were modulated by input type (G,B,N,F) in monkey D but not in monkey U (see Supplementary Fig. 7 for details). The edge weights for presentation of good (left) and novel (right) were combined (averaged) across right and left blocks and are presented in contra versus ipsi format (T_c,T_i: contra and ipsilateral temporal nodes, P_c,P_i: contra and ipsilateral prefrontal nodes). The line widths are scaled by the absolute value of the edges. Edges with positive weights are red and those with negative weights are blue. For Monkey U self-connections in P nodes were inconsistent between the 2 hemispheres and are grayed out. Monkey D: T node included TEa, TEm, TEO, FST, IPa and P node included 8Av, 45b, 44, F5, F4, 12l, 45a, 46v. Monkey U: T node included TEa, TEm, TPO, FST, IPa and P node included 13m, 45b, 44, F5, F4, 45a, 46v.

Figure 5

Subcortical novelty and value coding in striatum, amygdala, claustrum, and hippocampus. (A) GB (left) and NF (right) significantly active voxels in coronal view (P < 0.001,  < 0.01 cluster-corrected). (B) Same as A in sagittal view. (C) Voxels with significant resting correlation with ventral-IT, TP, and parietal networks in coronal (top) and sagittal (bottom) views (P < 0.001,  < 0.01 cluster-corrected). (D) Overlay of TP-connected voxels (green transparent squares) and voxels with significant GB (left) and NF (right) coding (beta coefficients of GB and NF shown). Data in this figure are from monkey U. Amyg: amygdala, CD: caudate, CDt: caudate tail, Claus: claustrum, Hipp: hippocampus, Put: putamen.