Parietal neurons encode information sampling based on decision uncertainty

Abstract

During natural behavior, animals actively gather information that is relevant for learning or actions; however, the mechanisms of active sampling are rarely investigated. We tested parietal neurons involved in oculomotor control in a task in which monkeys made saccades to gather visual information relevant to a subsequent action. We show that the neurons encode, before the saccade, the information gain (reduction in decision uncertainty) that the saccade was expected to bring for the following action. Sensitivity to information gain correlates with the monkeys’ efficiency at processing the information in the post-saccadic fixation, but is independent of neuronal reward sensitivity. Reward sensitivity, in turn, is unreliable across task contexts, inconsistent with the view that the cells encode economic utility. The findings suggest that parietal cells involved in oculomotor decisions show uncertainty-dependent boosts of neural gain that facilitate the implementation of active sampling policies, including the selection of relevant cues and the efficient use of the information delivered by these cues.

Access optionsAccess options

Rent or Buy article

Get time limited or full article access on ReadCube.

from$8.99

All prices are NET prices.

Fig. 1: Task design.
Fig. 2: Behavior is sensitive to IG.
Fig. 3: Neural responses are stronger in INF blocks and small-reward trials.
Fig. 4: Distribution of IG and RS effects in individual cells.
Fig. 5: IG signals cannot be explained by the VOI or the EV.
Fig. 6: The strength of IG signals correlates with performance.
Fig. 7: LIP neurons are enhanced by reward in a traditional one-step task.
Fig. 8: Controls ruling out alternative explanations.

Data availability

The data generated and analyzed for this study are available from the corresponding author upon request.

Code availability

The code written to analyze the data and to produce the figures for this study are available from the corresponding author upon request.

References

  1. 1.

    Hanks, T. D. & Summerfield, C. Perceptual decision making in rodents, monkeys, and humans. Neuron 93, 15–31 (2017).

  2. 2.

    Gottlieb, J. & Oudeyer, P. Y. Toward a neuroscience of active sampling and curiosity. Nat. Rev. Neurosci. 19, 758–770 (2018).

  3. 3.

    Gottlieb, J. Understanding active sampling strategies: empirical approaches and implications for attention and decision reseeaerch. Cortex 17, 30276–30279 (2017).

  4. 4.

    Gottlieb, J., Hayhoe, M., Hikosaka, O. & Rangel, A. Attention, reward and information seeking. J. Neurosci. 34, 15497–154504 (2014).

  5. 5.

    Bisley, J. W. & Goldberg, M. E. Attention, intention, and priority in the parietal lobe. Annu. Rev. Neurosci. 33, 1–21 (2010).

  6. 6.

    Thompson, K. G. & Bichot, N. P. A visual salience map in the primate frontal eye field. Prog. Brain Res. 147, 251–262 (2005).

  7. 7.

    Maunsell, J. H. Neuronal representations of cognitive state: reward or attention? Trends Cogn. Sci. 8, 261–265 (2004).

  8. 8.

    Foley, N. C., Kelley, S. P., Mhatre, H., Lopes, M. & Gottlieb, J. Parietal neurons encode expected gains in instrumental information. Proc. Natl Acad. Sci. USA 114, E3315–E3323 (2017).

  9. 9.

    Shenhav, A., Botvinick, M. & Cohen, J. The expected value of control: an integrative theory of anterior cingulate cortex function. Neuron 79, 217–240 (2013).

  10. 10.

    Kable, J. W. & Glimcher, P. W. The neurobiology of decision: consensus and controversy. Neuron 63, 733–745 (2009).

  11. 11.

    Sugrue, L. P., Corrado, G. S. & Newsome, W. T. Choosing the greater of two goods: neural currencies for valuation and decision making. Nat. Rev. Neurosci. 6, 363–375 (2005).

  12. 12.

    Peck, C. J., Jangraw, D. C., Suzuki, M., Efem, R. & Gottlieb, J. Reward modulates attention independently of action value in posterior parietal cortex. J. Neurosci. 29, 11182–11191 (2009).

  13. 13.

    Barash, S., Bracewell, R. M., Fogassi, L., Gnadt, J. W. & Andersen, R. A. Saccade-related activity in the lateral intraparietal area. I. Temporal properties. J. Neurophysiol. 66, 1095–1108 (1991).

  14. 14.

    Sugrue, L. P., Corrado, G. S. & Newsome, W. T. Matching behavior and the representation of value in the parietal cortex. Science 304, 1782–1787 (2004).

  15. 15.

    Aston-Jones, G. & Cohen, J. D. An integrative theory of locus coeruleus-norepinephrine function: adaptive gain and optimal performance. Annu. Rev. Neurosci. 28, 403–450 (2005).

  16. 16.

    Reynolds, J. H. & Heeger, D. J. The normalization model of attention. Neuron 61, 168–185 (2009).

  17. 17.

    Balan, P. F., Oristaglio, J., Schneider, D. M. & Gottlieb, J. Neuronal correlates of the set-size effect in monkey lateral intraparietal area. PLoS Biol. 6, e158 (2008).

  18. 18.

    O’Neill, M. & Schultz, W. Coding of reward risk by orbitofrontal neurons is mostly distinct from coding of reward value. Neuron 68, 789–800 (2010).

  19. 19.

    Tobler, P. N., Christopoulos, G. I., O’Doherty, J. P., Dolan, R. J. & Schultz, W. Risk-dependent reward value signal in human prefrontal cortex. Proc. Natl Acad. Sci. USA 106, 7185–7190 (2009).

  20. 20.

    Monosov, I. E., Leopold, D. A. & Hikosaka, O. Neurons in the primate medial basal forebrain signal combined information about reward uncertainty, value, and punishment anticipation. J. Neurosci. 35, 7443–7459 (2015).

  21. 21.

    Monosov, I. E. & Hikosaka, O. Selective and graded coding of reward uncertainty by neurons in the primate anterodorsal septal region. Nat. Neurosci. 16, 756–762 (2013).

  22. 22.

    Schultz, W. et al. Explicit neural signals reflecting reward uncertainty. Philos. Trans. R. Soc. Lond. B Biol. Sci. 363, 3801–3811 (2008).

  23. 23.

    Pouget, A., Drugowitsch, J. & A., K. Confidence and certainty: distinct probabilistic quantities for different goals. Nat. Neurosci. 19, 366–374 (2016).

  24. 24.

    Fan, J. An information theory account of cognitive control. Front. Hum. Neurosci. 8, 680 (2014).

  25. 25.

    Vossel, S. et al. Spatial attention, precision, and bayesian inference: a study of saccadic response speed. Cereb. Cortex 24, 1436–1450 (2014).

  26. 26.

    Dayan, P., Kakade, S. & Montague, P. R. Learning and selective attention. Nat. Neurosci. 3 (Suppl.), 1218–1223 (2000).

  27. 27.

    Krebs, R. M., Boehler, C. N., Roberts, K. C., Song, A. W. & Woldorff, M. G. The involvement of the dopaminergic midbrain and cortico–striatal–thalamic circuits in the integration of reward prospect and attentional task demands. Cereb. Cortex 22, 607–615 (2012).

  28. 28.

    Chong, T. T. et al. Neurocomputational mechanisms underlying subjective valuation of effort costs. PLoS Biol. 15, e1002598 (2017).

  29. 29.

    Silvetti, M., Vassena, E., Abrahamse, E. & Verguts, T. Dorsal anterior cingulate–brainstem ensemble as a reinforcement meta-learner. PLoS Comput. Biol. 14, e1006370 (2018).

  30. 30.

    Bisley, J. W. & Goldberg, M. E. Neuronal activity in the lateral intraparietal area and spatial attention. Science 299, 81–86 (2003).

  31. 31.

    Louie, K., Grattan, L. E. & Glimcher, P. W. Reward value-based gain control: divisive normalization in parietal cortex. J. Neurosci. 31, 10627–10639 (2011).

  32. 32.

    Blanchard, T. C., Hayden, B. Y. & Bromberg-Martin, E. S. Orbitofrontal cortex uses distinct codes for different choice attributes in decisions motivated by curiosity. Neuron 85, 602–614 (2015).

  33. 33.

    Bromberg-Martin, E. S. & Hikosaka, O. Midbrain dopamine neurons signal preference for advance information about upcoming rewards. Neuron 63, 119–126 (2009).

  34. 34.

    Park, I. M., Meister, M. L., Huk, A. C. & Pillow, J. W. Encoding and decoding in parietal cortex during sensorimotor decision making. Nat. Neurosci. 10, 1395–1403 (2014).

  35. 35.

    Hikosaka, O. & Isoda, M. Switching from automatic to controlled behavior: cortico-basal ganglia mechanisms. Trends Cogn. Sci. 14, 154–161 (2010).

  36. 36.

    Isoda, M. & Hikosaka, O. A neural correlate of motivational conflict in the superior colliculus of the macaque. J. Neurophysiol. 100, 1332–1342 (2008).

  37. 37.

    Leathers, M. L. & Olson, C. R. In monkeys making value-based decisions, LIP neurons encode cue salience and not action value. Science 338, 132–135 (2012).

  38. 38.

    Gottlieb, J. & Goldberg, M. E. Activity of neurons in the lateral intraparietal area of the monkey during an antisaccade task. Nat. Neurosci. 2, 906–912 (1999).

  39. 39.

    Snyder, L. H., Batista, A. P. & Andersen, R. A. Change in motor plan, without a change in the spatial locus of attention, modulates activity in posterior parietal cortex. J. Neurophysiol. 79, 2814–2819 (1998).

  40. 40.

    Hosokawa, T., Kennerley, S. W., Sloan, J. & Wallis, J. D. Single-neuron mechanisms underlying cost–benefit analysis in frontal cortex. J. Neurosci. 33, 17385–17397 (2013).

  41. 41.

    Oristaglio, J., Schneider, D. M., Balan, P. F. & Gottlieb, J. Integration of visuospatial and effector information during symbolically cued limb movements in monkey lateral intraparietal area. J. Neurosci. 26, 8310–8319 (2006).

  42. 42.

    Asaad, W. F. & Eskandar, E. N. A flexible software tool for temporally-precise behavioral control in Matlab. J. Neurosci. Methods 174, 245–258 (2008).

  43. 43.

    Platt, M. L. & Glimcher, P. W. Neural correlates of decision variables in parietal cortex. Nature 400, 233–238 (1999).

  44. 44.

    Nystrom, M. & Holmqvist, K. An adaptive algorithm for fixation, saccade, and glissade detection in eyetracking data. Behav. Res. Methods 42, 188–204 (2010).

Download references

Acknowledgements

The work was supported by NIH grants R24 EY015634 and RO1EY25965 to J.G., and by fellowships from the Danish Council For Independent Research, the Reinholdt W. Jorck og Hustrus Fond and the Marie og MB Richters Fond to M.H.

Author information

J.G., N.D. and M.H. designed and implemented the experiment. M.H. and N.D. collected the data. M.H. analyzed the data. J.G. and M.H. wrote the paper.

Correspondence to Jacqueline Gottlieb.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Peer review information: Nature Neuroscience thanks Jochen Ditterich and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Integrated supplementary information

Supplementary Figure 1 The metrics of the first saccade were not influenced by IG and RS.

a. Metrics of the first saccade to the cue, including reaction time (RT), velocity, and accuracy. Each point is the mean and SEM across all sessions (n=87). The p-values are from a 2-way ANOVA. b. The correlation between the first saccade latency and the postsaccadic VT (Pearson correlation coefficient). Each point is one session (n=87). There was a weak tendency for the two values to be positively correlated in INF (red) and unINF blocks (blue). However, there was no inverse correlation that might indicate a tradeoff between the first saccade latency and the postsaccadic viewing duration.

Supplementary Figure 2 Task geometries used for testing the representation of different task components.

Because each LIP neuron has a limited retinotopic RF that moves with the eyes, we could not observe the responses to all the task stages responses in a single geometry. Therefore, in addition to the standard geometry that revealed how the neurons encoded the cue, we used two control geometries designed to capture the responses to the saccade targets during different task epochs. Here we describe the logic of these geometries in detail, taking as example a neuron that has an RF in the left hemifield. a. Standard geometry for testing responses to the cue, during the delay period before the first saccade. This is the main geometry discussed in the text. As shown in Fig. 1a, during the delay period the stationary dots were at the center of the RF of the cell (left panel, dark cone) and the targets were at a 90-degree angular separations, which typically fell outside the RF. After the first saccade the RF moved to a different location. In this example, the RF would fall outside the monitor viewing area and is omitted for simplicity. By focusing on the delay and presaccadic periods (thick border) we observed the responses to the cue and the first saccade to it, but we could not observe the responses to the second saccade. b. Geometry 1: responses to the second saccade targets during the delay period before the first saccade. The display was rotated so that a target was in the RF during the delay period, while the cue was outside the RF (left panel). This allowed us to observe how the neuron responded to the targets during the delay period (when the monkeys prepared their saccade to the cue). c. Geometry 2: responses to the second saccade during the interval between the first and second saccade. The monkeys began by fixating an eccentric location and made their first saccade to the cue that was positioned at the center of gaze. This saccade brought the RF onto one of the targets. Therefore, this allowed us to examine the responses when the second saccade was directed toward the RF (last column) or in the opposite direction (not shown).

Supplementary Figure 3 Controls for saccade parameters and consistency.

a. The neurons were not strongly modulated by saccade parameters. Left: The bars show the regression coefficients (mean and SEM across all the cells, n=87) estimated with the stepwise model (Methods, eq. 1). Points indicate the regression coefficient for each individual cell. The numbers in parentheses indicate the % of cells contributing to each value (the cells for which the corresponding factor was accepted in the stepwise regression). The coefficients were not significantly different from 0 for any descriptor (N.S., p > 0.05) Right: The result was confirmed by an alternative analysis using a fixed model in which all saccade parameters were included for all cells. Same conventions as in left panel. b. Distribution of cell-by-cell area under the curve (AUC) for IG and RS. We computed a receiver operating characteristic for each cell (n=87) and computed the area under the curve (AUC) as a measure of the cell’s discrimination ability. In this analysis a curve that bows away from the diagonal (away from AUC= 0.5) indicates discrimination between two conditions. We calculated the AUC twice, first by comparing the firing rate distributions on INF (signal) relative to unINF (noise) trials across both RS, and second by comparing the distributions for small reward (signal) relative to large reward (noise) trials across INF and unINF blocks. (We chose this specification of signal and noise to maintain the convention that AUC > 0.5 indicates higher discrimination. This explains the sign reversal for RS, whereby regression coefficients are negative, indicating enhancement by smaller rewards, but AUC indices are above 0.5, indicating significant discrimination). 44% of neurons showed significant AUC for IG (p< 0.05, bootstrap; orange) and 38% showed significant AUC for RS (teal). The average indices were significantly greater than 0.5 across the population (colored triangles; IG: mean 0.55 (0.012), Wilcoxon signed rank test p < 10-4; RS: mean 0.55 (0.011), p < 10-4). Arrowheads show marginal means. c. AUC and regression coefficients are highly correlated throughout the delay period. Left panels show a comparison of the regression coefficients (abscissa) and AUC (ordinate) across individual cells (n=87). Black points indicate p<0.05 along the abscissa, colored circles indicate p<0.05 along the ordinate. Gray vertical lines indicate null effects (β = 0 and AUC = 0.5). r and p values refer to the Pearson correlation coefficient. The two indices are highly correlated for both IG (top) and RS (bottom). Right panels show time-resolved analysis using regression (gray) and AUC (colored). Each index was computed in a sliding window (50 ms width, 20 ms steps) aligned to cue onset and saccade onset. Data points show the mean and SEM across cells, bold colors indicate p< 0.05 in the individual bin. d. Peristimulus time variance (PSTV). PSTVs were constructed by calculating the variance of spike counts in bins of 50 ms, stepped by 10 ms. The panel shows the mean variance across the population of cells (n=87) for each of the four contexts, aligned to cue onset and 1st saccade onset. Variances show the same pattern as mean firing rates (cf Fig. 3b). e. Distribution of cell-by-cell firing rate variances. The variance of the mean firing rate in the delay period were calculated for both INF (red, upward-facing) and unINF (blue, downward-facing) conditions and tested for bimodality using Hartigan’s Dip Test. The vast majority are unimodal (colored if p>0.05, otherwise gray if p<0.05). Population means indicated by arrowheads at the top of the panel. At odds with a mixture-of-states hypothesis, in unINF blocks the vast majority of cells have unimodal firing rate distributions and firing rate variance is lower relative to INF blocks.

Supplementary Figure 4 Model comparisons.

Population activity is best fit by a model containing only IC and RS terms. The ordinate shows the relative Bayesian Information Criterion (BIC) scores for the full set of models containing all the possible combinations of 7 regressors: IG (i, referred to as “ic” in the legend), RS (r), IG*RS (x), expected value (e), value of information (v), decision accuracy (d), and trial completion rate (c). Note that these terms describe average performance in a recording session, so that the coefficients capture variance across the population of cells rather than within the dataset of each cell. The models are ordered from left to right according to the number of parameters. Excess BIC is defined as the difference in BIC score relative to the best model. The best model (excess BIC = 0) was the one including only two parameters (I and R). The BOLD font indicates the models with excess BIC < 10.

Supplementary Figure 5 Analysis of speed accuracy tradeoffs.

Each point shows the average decision accuracy and VT for one session (n=87). The lines show the best fit regression line across all INF data points. The large dots show the average values for INF trials with large and small RS and the x’s are the projections of these values on the regression line. For monkey S, the x’s are obscured by the large dots. In unINF blocks, decision accuracy was close to ceiling and largely independent of VT, confirming that the monkeys decided based on their prior. The only relationship we found was a negative correlation between accuracy and VT that was specific to monkey S (r = -0.45, p < 10-4) but not monkey M (r = -0.11, p = 0.27), suggesting that monkey S had more attentional lapses on sessions with longer VT. In INF blocks, in contrast, decision accuracy was strongly positively correlated with VT in both monkeys (red points; overall r = 0.23, p= 0.0023; monkey M, r = 0.3, p = 0.0024; monkey S, r = 0.48, p < 10-4), confirming the results of the within-session analysis (Fig. 6a,b). We further noted that speed-accuracy tradeoffs in INF blocks were sensitive to reward size, and this sensitivity differed by monkey. Monkey M tended to slow down and gain accuracy on large relative to small reward trials. In the left panel, the dark red dots lie above and to the right of the pale red dots, as shown by their mean values (large symbols). Monkey S showed the opposite pattern, making faster and less accurate decisions when high rewards were at stake. To measure these behavioral effects, we computed the regression of the speed-accuracy function across all the sessions (black line) and the projections of each point onto this line. The difference between these projections on large reward versus small reward trials indicate the direction and extent to which the monkeys adjusted the speed accuracy regimes according to RS in each session and is plotted on the abscissa in Fig. 6c.

Supplementary Figure 6 Responses on the standard memory guided saccade task.

Average PSTHs showing population firing rates of the 87 cells in our sample during a standard memory-guided saccade task (MGS), aligned on cue and saccade onset. The black trace shows the PSTH for the RF center location, and the gray trace shows the average for 3 remaining locations. Shading indicates SEM. The thick bar on the x-axis shows the interval during which the target was present. During the memory period (300 to 800 ms after target onset) activity was significantly higher for the RF relative to the other locations, across the population (two-sided Wilcoxon signed rank, p < 10-14 across the population, p < 10-8 in monkey M, p < 10-6 in monkey S) and in the vast majority of individual cells (93%).

Supplementary information

Supplementary Figures 1–6.

Reporting Summary

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark