🧠 neuroscience

Head direction cells use a head-referenced dual-axis updating rule in 3D space

By recording head direction cell activity in rats exploring a hemispherical surface, this study demonstrates that these cells update their firing in 3D space using a head-referenced dual-axis rule that integrates head-plane and earth-horizontal plane rotations relative to gravity.

Original authors: Williams, M., Street, J., Burgess, N., Jeffery, K.

Published 2026-02-20

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Williams, M., Street, J., Burgess, N., Jeffery, K.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine your brain has a built-in GPS. Inside this GPS, there's a tiny, super-specialized team of neurons called Head Direction (HD) cells. Think of these cells as the brain's internal compass needle. In a flat world (like a rat running on a flat floor), this compass works perfectly: it just points "North," "South," "East," or "West" based on which way the animal's nose is pointing.

But what happens when the animal climbs a sphere, a tree, or a wall? The world becomes 3D. The animal isn't just turning left or right anymore; it's tilting up, down, and rolling sideways.

This is where the paper gets exciting. The researchers wanted to solve a mystery: How does the brain's compass keep working when you are walking on a giant ball?

The Problem: The "Berry-Hannay" Glitch

Imagine you are holding a map while walking on a sphere. If you just turn your head left and right (like you do on flat ground), your compass gets confused because the ground itself is curving beneath you. In physics, this is called a "Berry-Hannay error." It's like trying to drive a car on a globe using only a flat map; eventually, your directions will be totally wrong.

The researchers suspected the brain doesn't just use a simple "left-right" turn signal. They thought the brain uses a Dual-Axis Rule.

The Solution: The "Two-Handed" Compass

The paper proposes that the brain's compass uses two hands to keep its bearings, rather than just one.

Hand 1 (The Local Spin): This hand tracks how much the animal turns its head left or right relative to its own body. It's like spinning a top on a table.
Hand 2 (The Gravity Tilt): This hand tracks how the animal's body is tilting relative to gravity (the pull of the Earth). It's like a carpenter's level that always knows which way is "down."

The brain adds these two movements together. By combining "how much I turned my head" with "how much the ground tilted under me," the brain can calculate exactly where "North" is in the real world, even if the animal is upside down or on a steep slope.

The Experiment: Rats on a Giant Ball

To test this, the scientists put rats on a giant, 1-meter-wide sphere (like a beach ball) and recorded the electrical signals from their brains while they scurried around.

They tested four different theories (or "reference frames") to see which one made the most sense of the data:

The Flat Earth Theory: The brain only cares about the horizontal plane (ignoring the curve).
The Surface Theory: The brain cares about the angle of the ball's surface under the rat's feet.
The Head Theory: The brain cares about the angle of the rat's own head.
The "Two-Handed" Theory (Dual Axis): The brain combines the head's turn with the gravity tilt.

The Big Discovery

The results were clear: The "Two-Handed" theory using the Head as the reference won.

When the scientists plotted the data using the Head-Referenced Dual-Axis (DAH) rule, the brain's compass signals became incredibly sharp and precise. It was like taking a blurry photo and suddenly bringing it into perfect focus.

The Analogy: Imagine trying to describe the direction of a lighthouse beam.
- If you only say "It's pointing East," that's fine on flat ground.
- But if you are on a ship rolling in a storm, "East" changes depending on how the ship tilts.
- The rat's brain realized: "I need to know how my head is turning, AND how my body is tilting against gravity, to know where the lighthouse really is."

Why Does This Matter?

Efficiency: The brain is lazy (in a good way). It doesn't need to build a complex 3D map of every possible angle. Instead, it uses a clever 2D trick (the ring attractor) and just adds a "tilt correction" factor. It's like using a flat map but adding a "tilt sensor" to your phone to make it work on a hill.
The "Head" vs. The "Floor": The study proved that the brain cares more about the rat's head than the floor it's walking on. Even if the rat is walking on a curved wall, the compass is anchored to the head's orientation relative to gravity, not the wall's surface.
Navigation in the Dark: This system allows animals to navigate in 3D spaces (like caves or trees) even if they can't see landmarks, because they are constantly updating their internal compass based on how they move and how gravity pulls on them.

The Bottom Line

This paper shows that our brains are brilliant engineers. When faced with the complexity of moving in 3D space, the brain doesn't try to build a massive, complicated 3D model. Instead, it uses a smart, two-part formula: "Turn of the Head" + "Tilt against Gravity."

This allows the brain's internal compass to stay accurate, sharp, and reliable, whether the animal is running on a flat floor, climbing a tree, or scurrying across a giant sphere. It's a perfect example of how nature finds the simplest, most efficient solution to a complex problem.

1. Problem Statement

Head direction (HD) cells function as a neural compass, signaling the azimuth (horizontal facing direction) of the head. In 2D environments, these cells update via a "ring attractor" network driven by left-right (yaw) head turns. However, in 3D space (e.g., climbing a sphere), a purely yaw-based updating rule accumulates Berry-Hannay errors due to non-horizontal rotations (pitch and roll), causing the internal compass to drift from the true global direction.

While previous studies suggested a "Dual Axis" (DA) rule might correct this—summing head rotation around a local axis and the rotation of that local axis around the gravity axis—two critical questions remained unanswered:

Does the DA rule operate during unrestricted, free movement over a continuously curved surface (as opposed to restricted planes or passive rotation)?
Is the "local axis" in this dual-axis calculation referenced to the environment's surface (allocentric) or the animal's head (egocentric)?

2. Methodology

The researchers conducted an electrophysiological study using rats to test these hypotheses.

Subjects & Recording: 10 adult male rats were implanted with tetrodes in the anterodorsal thalamic nucleus (ADN) and postsubiculum (PoS), key regions for HD cell processing.
Experimental Apparatus: Rats foraged freely on the upper surface of a 1-meter diameter hemisphere (a sphere with the bottom cut off). This allowed for continuous 3D movement and varying head orientations relative to gravity.
Data Acquisition:
- Position: Tracked via 5 cameras and a head-mounted LED (50 Hz).
- Orientation: Tracked via an Inertial Measurement Unit (IMU) calibrated against visual ground truth to capture 3D head pose (pitch, roll, yaw).
Analysis Pipeline:
- Cell Selection: 436 HD cells were identified from 2,270 single units based on tuning curve sharpness ( $\kappa > 0.5$ ).
- Modeling: The researchers fitted firing rates to a modified von Mises function under four distinct reference frame models:
  1. GA (Gravity Axis): Standard horizontal azimuth only (null hypothesis for 3D).
  2. DAS (Dual Axis - Surface): Local axis = Surface normal; Global axis = Gravity.
  3. DAH (Dual Axis - Head): Local axis = Head's dorso-ventral axis; Global axis = Gravity.
  4. HA (Head Axis only): Purely egocentric yaw, ignoring gravity/surface tilt (null control).
- Metric: Tuning curve quality was assessed via peak firing rate, Rayleigh vector length, sharpness ( $\kappa$ ), and Pearson's correlation ( $R$ ) between the model and actual data.

3. Key Contributions

First Demonstration of DA Rule in Free 3D Navigation: This is the first study to confirm that the dual-axis updating mechanism operates during active, unrestricted locomotion over a continuously curved surface, rather than just in constrained or passive conditions.
Resolution of Reference Frame Ambiguity: The study definitively identifies that the "local axis" for the dual-axis calculation is head-referenced (egocentric), not surface-referenced.
Mathematical Formulation: The paper provides a specific mathematical formulation (Equation 2) for the DAH rule, describing how the firing rate depends on the sum of head rotation in the head-plane and head-axis rotation in the horizontal plane.

4. Results

Superiority of the DAH Model: When tuning curves were plotted using the DAH (Head-Referenced Dual Axis) model, they were significantly taller, sharper, and better fitted by the von Mises function compared to all other models.
- Statistical analysis (Linear Mixed-Effects Models) confirmed that DAH outperformed DAS, GA, and HA across all metrics (peak rate, Rayleigh vector, $\kappa$ , and model fit) with $p < 0.001$ .
Surface vs. Head: The DAS model (surface-referenced) performed better than the standard GA model but was significantly inferior to the DAH model. This indicates the brain does not use the surface normal to update direction but rather the head's own orientation.
No Elevation Tuning: The study found no evidence of cells specifically tuned to elevation (pitch angle) or pure head-axis rotation (HA model), suggesting the system relies on the integrated dual-axis signal rather than separate elevation detectors.
Error Correction: The DAH model successfully corrected for the geometric distortions inherent in 3D movement, maintaining a stable preferred firing direction (PFD) relative to the global environment despite the rat's changing pitch and roll.

5. Significance and Implications

Dimensionality Reduction: The findings suggest the brain uses a dimensionality reduction strategy. Instead of constructing a complex 3D volumetric compass (which would require encoding roll and pitch), the brain maintains a planar (2D) compass that is continuously updated via a dual-axis rule. This allows for efficient navigation using a 1D ring attractor network even in 3D space.
Neural Mechanism: The results support a model where the HD system integrates:
1. Dynamic signals: Vestibular inputs (semicircular canals) detecting head turns.
2. Static signals: Otolith inputs detecting head tilt relative to gravity.
  These are combined to compute a "tilt-compensated compass."
Limitations and Future Directions: The paper discusses the "Hairy Ball Theorem," noting that a planar vector field on a sphere must have singularities (points where direction is undefined). While the DA rule solves the error accumulation problem, the system may still struggle with "inversion" (upside-down movement), a phenomenon observed in bats but not rats. This implies that while the DA rule is robust for most 3D navigation, the full integration of spatial signals in extreme orientations remains an open question.
Broader Cognitive Impact: This mechanism likely influences how other spatial cells (grid cells, place cells) encode 3D space, suggesting that the "cognitive map" may be fundamentally planar rather than volumetric, with implications for understanding non-spatial cognition that relies on spatial frameworks.