The caudate nucleus contributes causally to decisions that balance reward and uncertain visual information

  1. Takahiro Doi
  2. Yunshu Fan
  3. Joshua I Gold
  4. Long Ding  Is a corresponding author
  1. Department of Neuroscience, University of Pennsylvania, United States
  2. Department of Psychology, University of Pennsylvania, United States
  3. Neuroscience Graduate Group, University of Pennsylvania, United States
8 figures, 1 table and 2 additional files

Figures

Monkeys biased toward choices associated with large reward.

(A) Task design and timeline. Monkeys reported the perceived motion direction with saccades to one of the two choice targets. The motion stimulus was turned off upon detection of saccade. Correct …

Figure 2 with 2 supplements
Caudate activity reflected motion strength, reward context, choice, and the expected reward size associated with the choice.

(A, B) Activity of two example neurons. Shades: coherence levels. Colors: reward context (A) and reward size (B). Firing rates were computed using a 200 ms running window (50 ms steps). Only correct …

Figure 2—figure supplement 1
Example neurons with different kinds of task-relevant modulations.

Same format as Figure 2A and B. Note that colors indicate reward contexts in (A-C) and reward size in (D-E). Bars above the curves indicate the epochs used for regression analysis. Regression …

Figure 2—figure supplement 2
Modulation patterns of “combination neurons” during motion viewing.

Color map showing the presence of significant non-zero regression coefficients for each combination neuron during Epoch 5 in Figure 1A (from 100 ms after motion onset until 100 ms before saccade …

Figure 3 with 2 supplements
Caudate microstimulation affected the monkeys’ decision behavior.

(A, B) Two example sessions from monkey C showing different patterns of microstimulation effects. Black: trials without microstimulation; red: trials with microstimulation. Open circles and dashed …

Figure 3—figure supplement 1
Effects of caudate microstimulation for all sessions.

(A) Scatter plots for microstimulation-induced effects for the Ipsi-LR blocks (x-axis) and Contra-LR blocks (y-axis). A1, choice bias (logistic shift in %coh, positive: biasing toward the …

Figure 3—source data 1

Fitting results for choice (logistic) and RT (linear) data.

The same data were used to plot Figure 3—figure supplement 1 and generate summary statistics in Supplementary file 1a.

https://cdn.elifesciences.org/articles/56694/elife-56694-fig1-v1.csv
Figure 3—figure supplement 2
The average RT difference between the two reward contexts alone could not account for the reward context-dependent microstimulation effects.

(A) The average RT was longer for Ipsi-LR and Contra-LR blocks for the two monkeys, respectively (Wilcoxon signed rank test, p=0.002 and <0.0001). Data were from trials without stimulation. Dashed …

Illustration of correlation patterns induced by different hypothesized actions of microstimulation.

(A) Drift-diffusion model. Motion evidence (E) is modeled as samples from a Gaussian distribution (mean = signed coherence, variance = 1). The decision variable is computed as the time integral of E

Figure 5 with 4 supplements
Microstimulation induced correlated changes in the reward modulation of drift and bound.

(A) Scatter plots of changes in drift and bound induced by electrical microstimulation (abscissa and top histograms) and by interactions between electrical microstimulation and reward condition …

Figure 5—source data 1

Fitting results for choice and RT data using the DDM for sessions with significant microstimulation effects.

https://cdn.elifesciences.org/articles/56694/elife-56694-fig5-data1-v1.csv
Figure 5—figure supplement 1
DDM fits to example sessions in Figure 3.

(A) DDM fit for the example session in Figure 3A. Lines are DDM fits to both choice and single-trial RT data (circles, correct trials; crosses, error trials). Same format as Figure 3A. (B) …

Figure 5—figure supplement 2
DDM fitting results.

(A) Histogram of the difference in AIC between the full model, in which all DDM parameters were allowed to vary by reward context and microstimulation status, and a reduced model, in which all DDM …

Figure 5—figure supplement 3
Biases in drift and bounds together accounted for biases measured in logistic fits.

Scatter plots of microstimulation effects on choice bias, measured with logistic fit (abscissa) and DDM fits (ordinate). Bias values for the DDM were simulated using fitted parameters. Lines: linear …

Figure 5—figure supplement 4
Both monkeys showed similar correlation patterns between ∆drift and ∆bound.

(A-C) Results from monkey C. Same format as Figure 5B and D. Data points were color-coded based on ∆bound (rew) no estim values. Red dashed lines replotted the linear regression in A, with x-values …

Figure 6 with 3 supplements
Control analysis results.

(A) The average microstimulation effects on ∆drift and ∆bound that were independent of reward context were not correlated. Same format as Figure 5D. Linear regression, t-test, p=0.60. (B-D) The …

Figure 6—figure supplement 1
∆drift and ∆bound values obtained by fitting the simulated data to the “NoCollapse” DDM model did not show the same correlation patterns as the original data.

Linear regression, t-test, p=0.25, 0.65, and 0.66 for the three panels, respectively.

Figure 6—figure supplement 2
Both monkeys showed significant negative correlation between ∆drift (rew) and ∆bound (rew) on trials before microstimulation began.

Each data point represents one session. n = 18 and 21 for monkeys C and F, respectively. Linear regression, t-test, p<0.0001 for both monkeys. Red dashed lines re-plot the regression for “(rew) no …

Figure 6—figure supplement 3
The correlated ∆drift (rew × estim) and ∆bound (rew × estim) effects represented the dominant features of the data.

Scatter plot of the projection values for the first principal component (PC) based on eight fitted parameters (x-axis, [me or z] × [contra-LR or ipsi-LR] × [with estim or no estim]) and for the …

Figure 7 with 1 supplement
Microstimulation effects on coordination depended on neural selectivity and baseline adjustments of drift and bound.

(A and B) Principal components (PCs) were estimated for reward context modulation of Δdrift and Δbound without microstimulation (left panels; mean subtracted) for the two monkeys separately. The …

Figure 7—source data 1

Neural selectivity for different task factors at the microstimulation sites.

Column “indRewcontCoh” was used to divide the two subsets of stimulation sites in C and D. PCA was performed on DDM fitting results in Figure 5—source data 1.

https://cdn.elifesciences.org/articles/56694/elife-56694-fig7-data1-v1.csv
Figure 7—figure supplement 1
MRI reconstruction of recording and microstimulation sites.

(A) Recording and microstimulation sites projected on coronal slices. AC: anterior commissure. Positive numbers: anterior to AC. Negative numbers: posterior to AC. Scale bars: 5 mm. (B) Summary of …

Author response image 1
Monkeys did not show trends in their drift/bound biases across sessions.

Data were from trials without microstimulation in microstimulation sessions.

Tables

Key resources table
Reagent type
(species) or resource
DesignationSource or referenceIdentifiersAdditional
information
Software and AlgorithmsPython 3.5Python Software Foundationhttps://www.python.org/
Software and AlgorithmsMATLABMathworkshttps://www.mathworks.com
Software and AlgorithmsPsychophysics ToolboxKleiner et al., 2007http://psychtoolbox.org/
Software and AlgorithmsPandas v0.19.2Python Data Analysis Libraryhttps://pandas.pydata.org/
Software and AlgorithmsScikit-learn v0.18.1Pedregosa et al., 2011https://scikit-learn.org/stable/
Software and AlgorithmsScipy v0.18.1SciPy.orghttps://docs.scipy.org/doc/scipy/reference/stats.html

Additional files

Supplementary file 1

a, Median and p values for microstimulation-induced effects for all 55 sites, as measured by logistic fits to the choice data and linear fits to the RT data.P values were from Wilcoxon signed rank test.

Bold: p<0.05. b, Median and p values for microstimulation-induced effects in 39 effective sites, as measured by best DDM fits to the choice and RT data. P values were from Wilcoxon signed rank test. Bold: p<0.05.

https://cdn.elifesciences.org/articles/56694/elife-56694-supp1-v1.docx
Transparent reporting form
https://cdn.elifesciences.org/articles/56694/elife-56694-transrepform-v1.docx

Download links