Ordered developmental emergence of number-selective neurons in days-old zebrafish

Using whole-brain two-photon imaging in larval zebrafish, researchers revealed that number-selective neurons emerge in a specific developmental sequence, with cells tuned to single quantities appearing first and followed by those for higher numbers, ultimately forming a structured neuronal code for numerical cognition.

The Gist

The Big Idea: Counting Before You Can Talk

Imagine a baby who hasn't learned to speak yet, but can already tell the difference between one cookie and two cookies. Scientists have long wondered: How does the brain learn to "count" before it even knows what numbers are?

This study used tiny, transparent baby fish (zebrafish) to peek inside a living brain and watch how the "number sense" develops from scratch. They discovered that the brain doesn't just suddenly "get" math; it builds a number system in a very specific, ordered sequence, starting with the concept of "one."

---

The Experiment: A Movie Theater for Fish Brains

To see what was happening inside the fish's head, the scientists used a high-tech camera called a Two-Photon Light Sheet Microscope.

• The Analogy: Imagine the fish's brain is a dark city at night. Usually, you can only see a few streetlights at a time. But this microscope is like a magical drone that can fly over the entire city at once, turning on every single streetlight (neuron) simultaneously so you can see the whole map.
• The Setup: They glued the baby fish (at 3, 5, and 7 days old) in a gel and showed them dots on a screen.
• The Trick: They were very careful. If you show a fish one big dot and two tiny dots, the fish might just be reacting to the size or brightness, not the number. So, the scientists used a "mathematical recipe" to ensure the dots were different in number, but identical in total size, shape, and brightness. This forced the fish's brain to focus strictly on the count.

The Discovery: The "One" First, Then "Many"

The researchers watched the brain light up as the fish saw 1, 2, 3, 4, or 5 dots. Here is what they found about how the brain builds its number system:

1. The "One" Neurons Arrive Early (Day 3)
At 3 days old, the fish's brain is brand new. The scientists found that the very first neurons to specialize in counting were the "One-Tuned" neurons.

• The Analogy: Think of the brain as a construction site. Before you can build a skyscraper (complex math), you have to lay the foundation. The "One" neurons are the foundation. They are the only specialized workers on the site at the start.

2. The "Many" Neurons Join Later (Days 5 & 7)
As the fish got older (5 and 7 days), new types of workers arrived. Neurons that could distinguish "Two," "Three," and "Four" started popping up.

• The Analogy: It's like a classroom. First, the teacher teaches the concept of "One." Once the students master that, the teacher introduces "Two," then "Three." The brain doesn't learn them all at once; it adds them one by one.

3. The Brain Gets More Efficient (The Pruning)
Here is the surprising part: As the fish got older, the total percentage of neurons dedicated to counting actually went down.

• The Analogy: Imagine a chaotic party where everyone is shouting. At first, almost everyone is shouting about numbers. But as the party matures, the crowd gets quieter and more organized. The brain "prunes" (cuts away) the excess noise. It stops needing thousands of neurons to guess the number; it just needs a few highly skilled, precise neurons to do the job perfectly.

Where is the "Math Center"?

In humans, we often think math happens in the front of the brain (the prefrontal cortex). But in these fish, the number neurons were found mostly in the midbrain and forebrain.

• The Analogy: In a human, the math department is in the "Executive Office" at the top of the building. In a fish, the math department is right in the "Lobby" (the midbrain), which handles immediate sensory information. This suggests that counting might be a very ancient, basic survival skill that evolved before complex thinking.

The "Decoder" Test: Can a Computer Read the Fish's Mind?

To prove these neurons were actually doing the math, the scientists built a computer program (an AI decoder) that listened to the brain activity.

• The Result: The computer could look at the pattern of flashing neurons and guess, "Ah, the fish is looking at 3 dots!" with high accuracy.
• The Takeaway: Even though the fish can't speak, its brain is sending a clear, coded message about numbers. The code was so strong that the computer could read it better than random guessing.

Why Does This Matter?

This study tells us that number sense is a fundamental building block of life, not something we learn only after school.

• Survival: A baby fish needs to know if there is one predator or two. It needs to know if there is one food source or a school of fish.
• Evolution: The fact that fish, birds, and humans all seem to have these "number neurons" suggests that the ability to estimate quantity is an ancient tool that nature gave us to survive.

Summary in One Sentence

The brain builds its ability to count like a ladder: it starts with the rung for "One," adds the rungs for "Two" and "Three" as it grows, and eventually trims the fat to become a highly efficient, precise counting machine.

Technical Summary

1. Problem Statement

The ability to discriminate discrete quantities, known as the Approximate Number System (ANS), is a fundamental cognitive trait found across many species. While previous studies have identified number-selective neurons in specific brain regions of primates (parietal/prefrontal cortices) and birds, a comprehensive understanding of how these neural circuits emerge and organize across the entire brain during early development has remained elusive. Existing methods (e.g., electrodes) are often invasive, limited in spatial coverage, or biased toward specific regions, making it difficult to map the unbiased, whole-brain emergence of numerical coding. The authors aimed to identify when and where number-selective neurons appear in the developing zebrafish brain and to characterize the developmental trajectory of these circuits before numerically-driven behaviors (like shoaling or hunting) emerge.

2. Methodology

The study employed a combination of advanced imaging, controlled stimulus design, and rigorous computational analysis:

• Model System: Larval zebrafish (Danio rerio) at three developmental stages: 3, 5, and 7 days post-fertilization (dpf). The fish expressed a pan-neuronal, nuclear-localized calcium indicator (Tg(elavl3:H2B::jGCaMP7f)).
• Imaging Platform: A custom-built two-photon light-sheet microscope was used to record whole-brain activity at single-cell resolution.
  • Resolution: ~17,500 active neurons per larva.
  • Speed: 1 Hz volumetric imaging rate (1 volume per second).
  • Duration: 90-minute recording sessions to ensure statistical power while minimizing photobleaching.
• Stimulus Design: Visual stimuli consisted of black dots on a red background (to avoid fluorescence interference).
  • Numerosities: 1, 2, 3, 4, and 5 dots.
  • Control of Confounds: To ensure neurons responded to number rather than continuous variables (size, density), the authors controlled for five geometrical covariates: Radius (R), Total Area (TA), Total Perimeter (TP), Convex Hull (CH), and Inter-Distance (ID). These were combined into six feasible pairings to systematically vary numerosity while holding non-numerical features constant.
• Data Processing Pipeline:
  • Preprocessing: Motion correction using ANTs; registration to age-specific brain templates using ITK-SNAP.
  • Segmentation: CaImAn (Python toolbox) was used for constrained non-negative matrix factorization to segment individual neurons and extract fluorescence traces ($\Delta F/F$).
  • Neuron Identification: A two-way permutation ANOVA was applied to identify "number-selective neurons." A neuron was classified as number-selective if it showed a significant main effect for numerosity ($p < 0.01$) but no significant effect for the geometric control conditions or their interaction.
  • Validation: A Leave-One-Out (LOO) cross-validation framework was used to verify the stability of neuron identity and tuning curves across independent data subsets.
• Decoding: A Support Vector Machine (SVM) with a linear kernel was trained on the calcium activity of identified number-selective neurons to predict the presented numerosity (1–5 dots) from neural activity.

3. Key Contributions

• First Whole-Brain Map: This is the first study to map the emergence of number-selective neurons across the entire brain of a vertebrate at single-cell resolution during early development.
• Developmental Timeline: The study establishes a precise timeline for the emergence of numerical coding, showing that number sense exists at 3 dpf, well before the onset of complex numerically-driven behaviors.
• Ordered Emergence: The discovery of a specific developmental sequence where neurons tuned to "1" appear first, followed by neurons tuned to higher numbers.
• Pruning Mechanism: Evidence of a developmental pruning process where the relative proportion of number-selective neurons decreases as the brain matures, suggesting a shift from a broad, noisy population code to a refined, efficient one.

4. Key Results

• Ordered Emergence of Number-Selectivity:

  • 3 dpf: The brain is dominated by neurons tuned specifically to numerosity 1 (approx. 96% of number-selective cells).
  • 5–7 dpf: Neurons tuned to 2, 3, 4, and 5 appear in increasing proportions. Concurrently, the proportion of "1-tuned" neurons decreases significantly (to ~86% at 7 dpf).
  • Tuning Properties: The tuning curves exhibit broad profiles consistent with Weber's Law (response decreases as numerical distance from the preferred number increases). While single-neuron tuning can be noisy or asymmetric, population-averaged curves become less noisy with age.

• Spatial Distribution:

  • Number-selective neurons are primarily localized in the forebrain and midbrain (optic tectum), with very few in the hindbrain.
  • In 3 dpf larvae, the midbrain contains significantly more number-selective neurons than the forebrain. By 5 and 7 dpf, the distribution becomes more balanced, likely due to forebrain maturation.
  • No specific topographic map (e.g., a "number column") was found; neurons for different numerosities are intermixed.

• Relative Proportion Decline:

  • While the absolute number of identified neurons increases with age, the relative proportion of number-selective neurons (relative to all active neurons) decreases significantly:
    • 3 dpf: ~10%
    • 5 dpf: ~4.8%
    • 7 dpf: ~3.2%
  • This suggests a developmental refinement where the brain relies on fewer, but more precisely tuned, neurons for numerical processing.

• Decoding Performance:

  • An SVM decoder trained on the activity of number-selective neurons could predict the presented numerosity with >2x chance accuracy (chance = 20%) across all age groups (3, 5, and 7 dpf).
  • Despite the "noisier" tuning curves in younger fish, the higher absolute abundance of number neurons in 3 dpf larvae compensated for this, allowing for robust decoding performance comparable to older fish.

5. Significance

• Foundational Cognitive Primitive: The study demonstrates that the neural hardware for number sense is present and functional at 3 dpf, immediately after the visual system becomes operational. This precedes behavioral evidence of numerical discrimination (which starts around 5–24 dpf), suggesting that numerical coding is an intrinsic feature of early sensory processing rather than a product of extensive learning.
• Evolutionary Conservation: The findings support the hypothesis that the neural basis for the Approximate Number System (ANS) is evolutionarily conserved, with similar broad tuning properties observed in mammals, birds, and now fish.
• Developmental Mechanism: The results propose a model of cognitive development where early numerical representations are "broad" and abundant (high redundancy), followed by a pruning and refinement phase that increases coding efficiency.
• Methodological Advance: The successful application of two-photon light-sheet microscopy for whole-brain functional mapping in developing vertebrates provides a new framework for studying the emergence of other cognitive primitives (e.g., fear, social behavior) at cellular resolution.

In summary, this paper provides a high-resolution, whole-brain view of how the brain constructs a "number sense," revealing a dynamic developmental process where specific neural circuits emerge in a strict temporal order and are subsequently refined through pruning to support complex cognitive functions.