Local striatal reward signals can be predicted from corticostriatal connectivity

Peter Smittenaar, Zeb Kurth-Nelson, Siawoosh Mohammadi, Nikolaus Weiskopf, Raymond J. DolanNeuroImage 2017 · DOI

TL;DR

This study demonstrates that the intricate wiring of the brain's "reward circuit" (corticostriatal connections) can predict local functional activity within the striatum, even within established subregions. This suggests that the brain's physical structure plays a crucial role in shaping its moment-to-moment operations, offering a more detailed understanding of how the striatum processes reward and action selection.

What Problem Does This Paper Address?

The striatum, a key component of the basal ganglia, is known to be involved in reward-guided action selection. Its structure is characterized by multiple "loops" connecting it to various cortical areas. A central question in neuroscience is how this complex anatomical organization translates into specific functional roles. Specifically, while we know that different cortical areas project to different parts of the striatum, it's unclear how precisely these "connectivity fingerprints" dictate the function of individual voxels (small 3D units) within the striatum, especially within classical subregions like the caudate nucleus and putamen. This paper aims to bridge this gap by testing whether the anatomical connections of a specific voxel in the striatum can predict its functional activity during a reward-based learning task.

Methods

The researchers employed a sophisticated approach combining functional Magnetic Resonance Imaging (fMRI) and diffusion-weighted MRI (dMRI).

  • fMRI: Participants performed a reinforcement learning task where they had to learn to associate specific cues with actions (hand or foot) that yielded rewards. fMRI data was used to measure brain activity, specifically looking at signals related to reward, expected value, and motor responses.
  • dMRI and Probabilistic Tractography: dMRI was used to map the white matter pathways of the brain. Probabilistic tractography was then applied to estimate the connectivity profile of each individual voxel in the striatum to 148 cortical regions. This essentially created a "connectivity fingerprint" for each voxel.
  • Leave-One-Out Cross-Validation (LOOCV): To predict functional activity from connectivity, a LOOCV approach was used. For each participant, the relationship between connectivity and function was learned from all other participants. This learned relationship was then used to predict the functional activity of the current participant's voxels based on their connectivity fingerprints. This method helps ensure the findings are generalizable.
  • Model Comparison: A reinforcement learning model was used to estimate participants' behavior and derive key variables like expected values and reward prediction errors, which were then used as regressors in the fMRI analysis.

Datasets

The study used data from 24 healthy adult participants. No new benchmarks were introduced; rather, existing fMRI and dMRI acquisition and analysis techniques were applied.

Key Insights and Ideas

  • Structure Dictates Function, Even Locally: The most striking finding is that the specific anatomical connections of a striatal voxel (its "connectivity fingerprint") can predict its functional activity. This holds true even within established striatal subregions, suggesting a finer-grained functional organization than previously appreciated based solely on anatomical boundaries.
  • Reward Signals are Highly Predictable: Local reward signals in both the caudate nucleus and putamen were strongly predictable from corticostriatal connectivity. This highlights the critical role of specific cortical inputs in shaping how the striatum processes rewarding outcomes.
  • Expected Value and Motor Control Show Differential Predictability: While expected value signals were predictable from connectivity in the caudate nucleus, this was not the case for the putamen. Conversely, effector-specific activity (hand vs. foot) was predictable in the putamen but not the caudate. This suggests that different striatal subregions might be specialized for processing different aspects of reward-guided behavior, with their functional differentiation being driven by distinct cortical inputs.
  • Connectivity Captures Unique Variance: Even when accounting for the general pattern of activity across all participants (the "functional group-average"), the corticostriatal connectivity still explained unique variance in local striatal function. This indicates that anatomical wiring provides information about function that goes beyond simply replicating average brain activity patterns.

Key Results

  • Corticostriatal connectivity profiles successfully predicted local variations in:
  • Reward signals in bilateral caudate nucleus and putamen.
  • Expected value signals in bilateral caudate nucleus.
  • Response effector activity (hand vs. foot) in bilateral putamen.
  • The predictive accuracy, measured by Pearson's correlation, was significantly greater than chance for these specific contrasts and regions.
  • While the functional group-average also showed predictive power, the study demonstrated that corticostriatal connectivity explained unique variance not captured by this average functional pattern.

Limitations and Future Work

  • Statistical Thresholding: The authors acknowledge that the statistical tests were not corrected for multiple comparisons, meaning some findings might be false positives. However, they argue that the replication of results across hemispheres strengthens their confidence.
  • Sample Size: The study involved 24 participants. The authors suggest that larger datasets, such as those from the Human Connectome Project, would be beneficial for exploring individual differences in structure-function relationships and their link to reinforcement learning parameters.
  • Generalizability of Models: The study used a specific reinforcement learning model. Future work could explore how different models might influence the observed structure-function relationships.
  • Future Directions: The authors propose that their findings have implications for how we understand striatal computations, suggesting a potential division between function driven by local connectivity and more diffuse, non-specific signals. Further research could explore this dichotomy and its impact on cognitive processes. They also emphasize the potential of combining diffusion imaging with other techniques like resting-state fMRI to gain a more comprehensive understanding of corticostriatal systems.

Summary updated

BibTeX
@article{smittenaar2017local,
  author = {Peter Smittenaar and Zeb Kurth-Nelson and Siawoosh Mohammadi and Nikolaus Weiskopf and Raymond J. Dolan},
  title = {{Local striatal reward signals can be predicted from corticostriatal connectivity}},
  journal = {NeuroImage},
  year = {2017},
  doi = {10.1016/j.neuroimage.2017.07.042},
}
Figure