FPL Confirmed in Mouse Cortex -- But the Insect Scaling Trend Breaks at the Mammal Boundary
FPL holds in 79,554 mouse V1 cortex neurons (r=+0.455, p=0) -- first mammalian confirmation. But the insect scaling trend (0.317 to 0.819 across 428x) does not extend to mammals. Mouse cortex with 79k neurons gives weaker FPL than larval Drosophila with 2,952. Brain architecture matters more than neuron count.
Four Species. One Law. Different Strengths.
The MICrONS Cubic Millimeter dataset (The MICrONS Consortium et al. 2025, Nature) contains 303 million synaptic connections between 82k neurons in mouse primary visual cortex. We pre-registered four hypotheses (DISC_F15_MICRONS_v1, hash de0983ea, commit ead83b5, 2026-06-15) and ran the analysis on the public S3 data.
H1 CONFIRMED: r(d1, d2) = +0.455 at 79,554 connected neurons (p = 0).
FPL holds in mammalian cortex. This is the first confirmation across the insect-mammal boundary and the fourth independent connectome confirmation.
The Scaling Trend Breaks at the Phylum Boundary
- Three insect connectome experiments had suggested a monotonic scaling law -- r(d1, d2) increases with brain size:
- C. elegans (302 neurons, F12): r = +0.317
- Larval Drosophila (2,952 neurons, F13): r = +0.663
- Adult Drosophila (129,117 neurons, F14): r = +0.819
H2 predicted r > 0.70 for mouse cortex based on this trend. The result: r = +0.455. Despite 79k neurons (27x more than larval Drosophila), mouse cortex gives a weaker FPL signal than a 2,952-neuron insect larva.
The insect scaling tendency is insect-specific. It reflects brain architecture, not neuron count.
Why Cortex is Weaker: Laminar Specialization
The adult Drosophila top hub is the APL neuron: d1 = 14,000, d2 = 14,786. It contacts every Kenyon cell and receives from every Kenyon cell. Same neuron, both directions, simultaneously. This generalist architecture creates near-perfect d1/d2 co-persistence.
The mouse cortex top d2 hub has d2 = 19,718 but d1 = 601. It receives from 19,718 distinct neurons but projects to only 601. This is a structural specialist -- the opposite of APL.
- This split is driven by cortical layers:
- Layer 4 neurons: dense thalamic and recurrent input (high d2), short local axons (low d1)
- Layer 5/6 pyramidal neurons: project to many brain areas (high d1), more selective input (lower d2)
- Layer 2/3 interneurons: variable
The layer-specific d1/d2 role separation weakens hub co-persistence. Cortex sits in an intermediate regime between full generalist (insect) and full polarity (npm software).
Hub Regime Spectrum: Five Points Now Characterized
FPL is confirmed across the full spectrum. Its strength encodes the brain's structural organization class.
Binary > Weighted (Third Replication)
H4: r(d1_binary, d2_binary) = 0.455 > r(d1_weighted, d2_weighted) = 0.445. CONFIRMED.
Three consecutive pre-registered connectome experiments (F14, F15 explicit; F13 partial) confirm that binary contact diversity is a more fundamental FPL signal than synapse-count weighting. The number of distinct partners matters more than the total synapse volume.
Pre-Registered Before Analysis
All four hypotheses were hashed (de0983ea...) and committed to the public preregistrations repository before any FPL correlation was computed. The artifact filter (remove neurons with d1 or d2 > 81,995 -- impossible for real cells) was specified in the pre-registration, not discovered post-hoc.
2,441 over-merged segmentation fragments were excluded; 79,554 nucleus-verified neurons remained.
IRDME pre-registration-first structural analysis. All experiments committed before analysis.
Reproducibility
This result was pre-registered before analysis. SHA-256 hash: de0983ea651a5e6b6ac978fef0aee7aa3a02be51
Verify at github.com/vladi160/preregistrations