A multiplexed striatal architecture for generalized spatial goal progress

This study demonstrates that the nucleus accumbens computes a generalized, scale-invariant distance-to-goal signal that supports flexible, memory-guided navigation by integrating dopaminergic inputs to bridge cognitive mapping and reinforcement learning.

The Gist

Imagine you are walking through a massive, shifting city. Sometimes you need to get to a coffee shop down the street; other times, you need to find a new restaurant that just opened. The city layout changes, the traffic jams vary, and sometimes you have to walk past your old favorite spots to get to the new one.

To navigate this chaos, your brain needs more than just a static map. It needs a dynamic "progress bar" that tells you how far you are from your goal, no matter where you started or how fast you are walking.

This paper reveals that a small, bean-shaped part of your brain called the Nucleus Accumbens (NAc) acts as the master engineer for this progress bar. Here is the story of what they found, broken down into simple concepts:

1. The "Progress Bar" That Doesn't Care About Time

Most of us think of distance in terms of time ("It's a 10-minute drive"). But the brain is smarter than that. The researchers found that the NAc calculates distance based on physical space, not time.

• The Analogy: Imagine you are walking up a hill. If you walk slowly, it takes longer, but you are still halfway up the hill. If you run, you get there faster, but you are still halfway up.
• The Discovery: The NAc neurons act like a ruler that stretches and shrinks. If you slow down, the neurons don't panic and say "We're running out of time!" Instead, they say, "We are still 50% of the physical path away from the goal." This allows the brain to generalize: whether you are walking a short path or a long one, the brain knows exactly how much "journey" is left.

2. The "Ghost" of Goals Past (Multiplexing)

Here is the really cool part. Usually, when you change your mind and pick a new destination, your brain forgets the old one. But the NAc is like a multitasking wizard.

• The Analogy: Imagine you are driving to a new restaurant (Goal A). On the way, you pass your old favorite pizza place (Goal B). Even though you aren't going to the pizza place anymore, your brain keeps a "ghost signal" for it.
• The Discovery: The NAc can hold two different "progress bars" at the same time without them mixing up. It uses a special trick called orthogonal coding. Think of it like a radio station: one signal is playing on the FM frequency (Current Goal), and another is playing on the AM frequency (Old Goal). They occupy the same space (the NAc) but don't interfere with each other. This lets the brain instantly remember, "Hey, if I wanted to go back to the pizza place, I'm actually very close to it," even while you are driving to the new restaurant.

3. The Fuel: Dopamine is the GPS, Not the Map

You might think the brain's famous "map" (the Hippocampus) does all this work. Surprisingly, the researchers found that the NAc does this job independently of the map.

• The Analogy: The Hippocampus is like a detailed paper map of the city. The NAc is the car's dashboard computer. You can have the map, but if the dashboard computer is broken, you don't know how far you've driven.
• The Discovery: When the researchers turned off the "map" (Hippocampus), the NAc's progress bar kept working perfectly. However, when they cut off the dopamine supply (the fuel from the Ventral Tegmental Area), the progress bar vanished.
• The Takeaway: Dopamine isn't just about "feeling good" or "wanting rewards." It is the essential fuel that powers the brain's ability to calculate how far you have to go. Without it, you lose your sense of progress.

4. The "What If?" Test

The researchers tested this by tricking the rats. They told them, "Go to the new restaurant!" The rats went. Then, they took the reward away at the new restaurant. The rats didn't just give up; they started sniffing around the old pizza place they had visited earlier.

• The Discovery: This "search behavior" was driven by that "ghost signal" in the NAc. But when they turned off the dopamine, the rats forgot the old pizza place entirely. They stopped searching for it. This proves that the NAc is the place where we keep a mental list of "what if I went back there?"

The Big Picture

This study changes how we see the brain's navigation system.

• Old View: The brain has a map (Hippocampus) and a reward system (Dopamine) that are separate.
• New View: The Nucleus Accumbens is the bridge. It takes the raw data of "where am I?" and turns it into a flexible, generalized sense of "how far am I from my goal?" It does this for the goal you have now, and it keeps a backup file for the goals you had before.

In short, your brain isn't just a GPS that tells you where you are; it's a smart assistant that constantly calculates your progress, remembers your past detours, and keeps you on track—all powered by a chemical spark called dopamine.

Technical Summary

1. Problem Statement

Flexible goal-directed behavior requires the brain to maintain a generalized internal metric that tracks progress toward goals across varying locations, routes, and contexts. While the hippocampus and entorhinal cortex are known to encode detailed spatial maps (cognitive maps), it remains unclear whether the brain constructs an abstract, generalized distance-to-goal state that can be applied to arbitrary start–goal combinations. Furthermore, it is unknown how the brain simultaneously maintains representations of current goals and latent (counterfactual) representations of previously visited goals to support flexible decision-making. Current reinforcement learning (RL) frameworks often rely on temporal-difference (TD) errors based on elapsed time, but navigation fundamentally requires inference over spatial state variables. This study seeks to identify the neural substrate for computing generalized spatial states and maintaining parallel representations of current and past goals.

2. Methodology

The researchers employed a multimodal approach combining large-scale electrophysiology, fiber photometry, and causal perturbations in rats navigating linear and 2D mazes.

• Behavioral Tasks:
  • Linear Maze: Rats navigated a 2m track with 10 wells, alternating between goal pairs. Goal locations changed dynamically across blocks, requiring rapid updating of navigation strategies.
  • 2D Maze: A 1.5m × 1.5m arena with a 5×5 grid of wells. Rats performed memory-guided navigation to a fixed "home" well from random starting points, contrasting with cue-guided navigation to random targets.
  • Reward Omission: A paradigm where expected rewards were withheld to induce search behavior toward previously rewarded locations.
• Neural Recordings:
  • Neuropixels Probes: Used for large-scale population recordings (45–149 neurons) in the Nucleus Accumbens (NAc), Ventral Tegmental Area (VTA), and Hippocampal CA1.
  • Fiber Photometry: Used to measure dopamine concentration in the NAc using the sensor dLight1.2.
  • Tetrode Recordings: Used for single-unit recordings in VTA and CA1.
• Causal Perturbations:
  • Pharmacogenetics (DREADDs): Silencing of Hippocampus (HPC) and Medial Entorhinal Cortex (MEC) using hM4Di and the agonist DCZ.
  • Optogenetics:
    • Inhibition: Silencing VTA dopamine neurons (using stGtACR2) or VTA-to-NAc axon terminals (using eOPN3) to test necessity for spatial coding.
    • Activation: Used to identify dopaminergic neurons.
• Analysis Techniques:
  • Dimensionality Reduction: Uniform Manifold Approximation and Projection (UMAP) and Persistent Homology to analyze population trajectory topology.
  • Decoding: Linear Support Vector Regression (SVR) to decode behavioral variables (distance, time, proportional distance) from neural activity.
  • Orthogonality Analysis: Calculating the angle between decoding subspaces for current vs. previous goals.

3. Key Contributions & Results

A. Generalized, Scale-Invariant Distance-to-Goal Coding

• Finding: NAc population activity encodes a proportional distance-to-goal signal (distance normalized by total journey length).
• Evidence:
  • Decoders trained on NAc activity accurately predicted proportional distance across different goal locations and running directions.
  • The signal tracks physical path length, not elapsed time. When animals slowed down mid-journey, the distance decoder remained accurate, while time-based decoding failed.
  • This coding is generalized: It transfers to novel journeys and 2D environments, tracking path distance rather than Euclidean distance, suggesting reliance on path integration/self-motion.
  • Specificity: This signal is present during memory-guided navigation but absent during cue-guided navigation, indicating it is not a simple motivational ramp but a computed spatial state.

B. Independence from Hippocampal Maps

• Finding: NAc distance coding does not rely on online hippocampal or medial entorhinal cortical activity.
• Evidence:
  • Pharmacogenetic silencing of HPC and MEC (verified by object location memory deficits) did not impair the accuracy of NAc distance-to-goal decoding.
  • In contrast, hippocampal (CA1) neurons showed poor generalization of distance coding across start-goal combinations compared to NAc.
  • This suggests the NAc computes spatial progress independently of canonical allocentric map updates.

C. Critical Dependence on VTA Dopamine

• Finding: Dopaminergic input from the VTA is essential for NAc distance coding and accurate goal targeting.
• Evidence:
  • Optogenetic silencing of VTA dopamine terminals in the NAc abolished the ramp-up/ramp-down firing patterns of NAc neurons and severely impaired goal-directed navigation (increased error rates).
  • Silencing did not affect basic locomotion or well recognition but disrupted the specific computation of spatial progress.

D. Multiplexed Coding of Current and Previous Goals

• Finding: The NAc simultaneously encodes distances to the current goal and previously rewarded goals within orthogonal subspaces.
• Evidence:
  • In a task where rats navigated to a new goal after passing a previous one, a subset of NAc neurons maintained activity representing the distance to the previous goal even after the animal switched targets.
  • Orthogonality: Decoders for current and previous goals were trained on the same population. The regression vectors for these two goals were nearly orthogonal (~80.77°), allowing parallel evaluation of counterfactual states without interference.
  • Memory Persistence: This latent representation of previous goals persisted during continuous task engagement but decayed rapidly (within ~1 min) during rest periods.
  • Dissociation from Dopamine: While VTA dopamine signals encoded the current goal distance, they did not encode the previous goal distance. The multiplexed memory emerges within NAc circuit dynamics, not via direct inheritance from dopamine inputs.

E. Behavioral Relevance: Memory-Guided Search

• Finding: NAc dopamine is required to reinstate search behavior toward previously rewarded locations when current goals fail.
• Evidence:
  • When reward was omitted at the current goal, rats naturally searched for the previously rewarded well.
  • Optogenetic suppression of VTA-to-NAc dopamine during this omission period abolished the bias toward the previous goal, while leaving the preference for the current goal pair intact. This confirms the NAc's role in flexibly switching between current and latent goal representations.

4. Significance

This study fundamentally shifts the understanding of the striatum's role in navigation and reinforcement learning:

1. Bridging Cognitive Maps and RL: It identifies the NAc as a substrate that bridges detailed spatial mapping (cognitive maps) and reinforcement learning. It computes a normalized spatial state variable (proportional distance) that allows RL algorithms to operate over physical space rather than just time.
2. Generalization: The NAc provides a coordinate-invariant metric for progress that generalizes across arbitrary start-goal pairs, solving a key limitation of hippocampal remapping.
3. Multiplexing Counterfactuals: The discovery of orthogonal coding for current and previous goals reveals a mechanism for maintaining "what-if" scenarios (counterfactuals) in parallel. This allows animals to rapidly re-evaluate alternative strategies without erasing the current plan.
4. Dopamine's Role: It refines the role of dopamine from a simple reward prediction error signal to a critical modulator of spatial state abstraction, necessary for maintaining the internal metric of progress and the flexibility to switch between latent goals.

In summary, the NAc acts as a dynamic, multiplexed processor that computes a generalized metric of spatial progress, enabling flexible, long-range goal-directed behavior in changing environments.