Decoupled choice-driven and stimulus-related activity in parietal neurons may be misrepresented by choice probabilities

Abstract

Trial-by-trial correlations between neural responses and choices (choice probabilities) are often interpreted to reflect a causal contribution of neurons to task performance. However, choice probabilities may arise from top-down, rather than bottom-up, signals. We isolated distinct sensory and decision contributions to single-unit activity recorded from the dorsal medial superior temporal (MSTd) and ventral intraparietal (VIP) areas of monkeys during perception of self-motion. Superficially, neurons in both areas show similar tuning curves during task performance. However, tuning in MSTd neurons primarily reflects sensory inputs, whereas choice-related signals dominate tuning in VIP neurons. Importantly, the choice-related activity of VIP neurons is not predictable from their stimulus tuning, and these factors are often confounded in choice probability measurements. This finding was confirmed in a subset of neurons for which stimulus tuning was measured during passive fixation. Our findings reveal decoupled stimulus and choice signals in the VIP area, and challenge our understanding of choice signals in the brain.Choice-related signals in neuronal activity may reflect bottom-up sensory processes, top-down decision-related influences, or a combination of the two. Here the authors report that choice-related activity in VIP neurons is not predictable from their stimulus tuning, and that dominant choice signals can bias the standard metric of choice preference (choice probability).

Fig. 1

Example single-unit recordings from awake, behaving monkeys during heading discrimination. a, b The behavioral data from two example sessions, along with simultaneous single-unit recordings from VIP. Top row: psychometric curves represent the ratio of rightward choices as a function of heading, based on visual (red) or vestibular (blue) cues. The data (circles) were fitted with cumulative Gaussian functions (solid curves). Second row: average firing rates (FRs) are presented as a function of heading. Leftward and rightward pointing gray triangles mark the average FRs for leftward and rightward choices, respectively. For each unit, one hundred (randomly selected) overlaid spikes are presented (insets). Third row: raster plots depict the spikes for each trial, as a function of time. Stimulus motion, having a Gaussian velocity profile, spanned the epoch from 0 to 1 s. Horizontal black lines in the raster plots separate different headings. c–f Average FRs as a function of heading are presented for four additional VIP units (3 units tested with the visual condition in red, and 1 unit tested with the vestibular condition in blue) from four separate sessions. All error bars denote SEM

Fig. 2

Heading and choice partial correlations. Partial correlations were calculated between neuronal firing rates and stimulus headings (given the actual choices made by the monkey) or choices (given the stimulus headings). Only cells that showed a significant main effect of heading or choice were included (p < 0.05, two-way ANOVA). a Partial correlations are presented as scatterplots for VIP (left) and MSTd (right). Red and blue symbols represent the data from the visual and vestibular conditions, respectively, and circle and triangle symbols represent the new and old data sets, respectively. Arrows indicate the data points corresponding to the VIP tuning curves presented in Fig. 1 (letters A–F for the respective Fig. 1 subplots). b The proportion of neurons (same data as in a) with significant partial correlations for choice-only, heading-only or both heading and choice. In VIP, neurons with significant choice partial correlations are prevalent; while MSTd shows a greater proportion of significant heading partial correlations. In b, the new and old data were pooled for each brain area (results were similar for each data set individually). For VIP new (old) data, N = 140 (50) and 95 (58) for visual and vestibular conditions, respectively. For the MSTd new (old) data, N = 80 (217) and 25 (97) for visual and vestibular conditions, respectively. See also Supplementary Fig. 1 for a simulation and Supplementary Figs. 2–4 for further analyses of these data

Fig. 3

Relationship between heading and choice partial correlations and choice probabilities. Average choice probability (CP; green-orange heat map) is presented in the plane of partial correlations for the visual (a) and vestibular (b) data from VIP (left column) and MSTd (right column). The background gray color represents CP = 0.5; shades of green and orange represent CP > 0.5 and CP < 0.5, respectively. White dashed lines (drawn by eye, identical for all four subplots) outline the regions characterized predominantly by high CPs (CP > 0.5). Marginal plots show CP values plotted against heading and choice partial correlations. Cell samples and N values are the same as in Fig. 2. Circle and triangle symbols mark the new and the old data sets, respectively. See also Supplementary Fig. 1D for CP simulation

Fig. 4

Partial correlations as a function of time. Partial correlation data are presented for areas VIP (left two columns) and MSTd (right two columns). Heading partial correlation data are presented in red for the visual condition and in blue for the vestibular condition. The respective choice partial correlations are presented in black. a The average partial correlations were aligned to each cell’s heading preference by inverting the signs of both the heading and choice partial correlations when the overall heading partial correlation was negative (done separately for visual and vestibular conditions). On average, the choice partial correlations across cells cancel out, indicating no consistent alignment to heading preferences. b Squared partial correlations (R ²) are presented, such that positive and negative contributions from different neurons did not cancel. The gray lines below a depict the stimulus velocity profile. Cell selection and N values are the same as in Figs. 2 and 3. The old and new data were pooled in this plot, and the horizontal axis shows time normalized to stimulus duration, which was different for the new and old data sets. See also Supplementary Fig. 5 for a more direct comparison of the choice partial correlations across areas and stimulus modalities (R ²)

Fig. 5

Partial correlation results for congruent and opposite cells. a Visual and vestibular heading partial correlations are presented for neurons classified as ‘congruent’ or ‘opposite’ based on whether the visual and vestibular partial correlations for heading had the same or opposite signs, respectively. Significantly congruent or opposite cells (p < 0.05 for both the visual and vestibular heading partial correlations) are marked by filled data points. Circle and triangle symbols mark the new and the old data sets, respectively. N values represent the number of cells (the number of significantly congruent or opposite cells in parentheses). Only cells which showed a significant main effect of heading or choice (p < 0.05, two-way ANOVA) for both visual and vestibular conditions were included. b Heading and choice partial correlations are presented for the same cells. Red and blue symbols represent the data for the visual and vestibular conditions, respectively. c Choice partial correlations measured during the visual and vestibular conditions are compared for the same cells. Solid lines and the shaded regions represent type-II regressions with their 95% confidence intervals. r and p-values for the regressions in panel b are presented on the respective plots. For all the regressions in black (a, c) p < 0.001 (except for the second plot in a, for which p = 0.02)

Fig. 6

Partial correlation time-course for congruent and opposite cells. Average partial correlation (R) values are presented for congruent and opposite cells in VIP (a) and MSTd (b). Here, the data were aligned to each cell’s visual heading preference by inverting the partial correlation signs such that the visual heading partial correlation was always positive. This was done together for visual and vestibular heading and choice partial correlations, such that positive choice (or positive vestibular heading) R-values indicate responses with the same sign as the visual heading partial correlation. Cell selection and N values are the same as in Fig. 5. The old and new data were pooled in this plot, and the times along the horizontal axis are normalized to stimulus duration. All error bars denote SEM

Fig. 7

Heading, but not choice, partial correlations are predicted by heading preferences measured during fixation. (a) Heading and b choice partial correlation histograms are presented for the data with rightward and leftward global heading tuning preferences (lighter-colored upward histograms, and darker-colored downward histograms, respectively). The data are presented separately for VIP and MSTd (left and right columns, respectively) and visual and vestibular conditions (red and blue shades, respectively). p-values (Wilcoxon rank sum tests) are presented on the plots. N = 93 for VIP visual (49 classified ‘Left’ and 44 classified ‘Right’), N = 92 for VIP vestibular (45 classified ‘Left’ and 47 classified ‘Right’), N = 265 for MSTd visual (141 classified ‘Left’ and 124 classified ‘Right’), and N = 265 for MSTd vestibular (124 classified ‘Left’ and 141 classified ‘Right’)