Dopamine-Modulated Working Memory in a Ring Attractor Network

A spiking neural network simulation exploring how dopamine concentration affects the stability of working memory under distractor interference.

Built with Brian2.

---

Background

Ring attractor networks are a standard computational model for spatial working memory - the kind used when you hold a location in mind during a delay period. Neurons are arranged on a ring, connected by a Mexican-hat profile: strong local excitation, weak global inhibition. A localised "bump" of activity can persist without external input, encoding the remembered location.

The question here is: what happens to that bump when a distractor appears, and does dopamine level change the answer?

D1 receptor activation in prefrontal cortex is known to scale recurrent excitation. In this model, a single parameter da (∈ [0,1]) multiplies the excitatory synaptic weights, mimicking D1-mediated gain modulation. This lets us ask whether there is a critical DA threshold below which the memory trace becomes vulnerable to interference.

---

Experimental Protocol

0 ms: baseline (50 ms, spontaneous) -> cue ON (100 ms at neuron 50) -> cue OFF (t = 150 ms) -> distractor (30 ms, half-amplitude at neuron 0) -> silence -> end (t = 260 ms)

Three DA conditions are run with identical noise seeds so the only variable is dopamine level.

---

Results

DA level	Label	Late spikes (250–300 ms)	Final centroid offset
1.00	healthy	352	+3.7 neurons
0.88	fatigued	174	+3.7 neurons
0.78	depleted	48	+22.0 neurons

At DA = 0.78, the late-period bump centroid (265-295 ms) is shifted by +22 neurons relative to the cued location, indicating a large working-memory error after the distractor period. At DA ≥ 0.88, the final centroid remains within ~4 neurons of the cue, consistent with only noise-level perturbation.

These results indicate a sharp transition in attractor stability as a function of dopamine level, consistent with a bifurcation between stable and distractor-sensitive regimes - a critical DA value below which the memory trace becomes vulnerable to interference.

---

Key figure

Top row: population activity heatmaps across the three conditions. The bump narrows and weakens with decreasing DA; at low DA, the distractor visibly pulls activity away from the cued location during/after the distractor period (230–260 ms).

Bottom row: bump centroid tracked via circular mean of active neurons. High/moderate DA centroids stay near neuron 50 throughout. Low DA centroid drifts to ~72 after the distractor - a measurable working memory error.

---

Possible extensions

• DA-dependent synaptic plasticity - instead of static weight scaling, let dopamine modulate STDP learning rates and test whether the network can self-organize a stable bump.
• Multi-item working memory- maintain two competing bumps with asymmetric DA; test whether the stronger attractor suppresses the weaker one or whether they can coexist.
• Dopamine as an RL signal - couple the ring attractor to a reward prediction error signal and test whether the network can learn which locations are worth remembering.

---

Running it

Install dependencies:

pip install -r requirements.txt

Run the simulation:

python ring_attractor_dopamine.py

Output: ring_attractor_dopamine.png + per-condition centroid summary printed to stdout.

Main parameters are at the top of the file:

J_e = 7.0      # peak excitatory weight (mV)
J_i = 0.2      # flat inhibitory offset (mV)
sig_e = 0.05     # Gaussian half-width (fraction of ring)
da_levels = [1.0, 0.88, 0.78]   # DA conditions to compare

---

References

• Durstewitz, D., & Seamans, J. K. (2008). The dual-state theory of prefrontal cortex dopamine function with relevance to catechol-o-methyltransferase genotypes and schizophrenia. Biological Psychiatry, 64(9), 739–749. https://doi.org/10.1016/j.biopsych.2008.05.015
• Wang, X.-J. (2001). Synaptic reverberation underlying mnemonic persistent activity. Trends in Neurosciences, 24(8), 455–463. https://doi.org/10.1016/S0166-2236(00)01868-3
• Compte, A., Brunel, N., Goldman-Rakic, P. S., & Wang, X.-J. (2000). Synaptic mechanisms and network dynamics underlying spatial working memory in a cortical network model. Cerebral Cortex, 10(9), 910–923. https://doi.org/10.1093/cercor/10.9.910