A human brain begins as a single cell. Nine months later, it contains roughly 170 billion cells, neurons and glia, each in the correct place, each performing the right function. How that staggering increase in scale does not lead to chaos is one of developmental biology’s oldest questions.
The classical answer, proposed by Lewis Wolpert in 1969, relies on morphogen gradients: diffusible chemical signals that spread across a developing tissue, telling cells where they are by their concentration. But as Stan Kerstjens of Cold Spring Harbor Laboratory and ETH Zurich points out, morphogens have a fundamental limitation. “A morphogen gradient works beautifully over a millimeter,” he said, “but not over the orders-of-magnitude size increase during development.”
Writing in Neuron (published online March 2, 2026), Kerstjens and an international team propose a complementary mechanism: cells inherit their positional identity through their cellular family tree, a lineage-based information system that scales naturally as the brain grows.
The researchers built a mathematical model starting from a simple observation: cells descended from the same progenitor stay near each other during development. If each progenitor’s descendants maintain spatial coherence, the brain can encode positional information through lineage relationships alone, without requiring long-range chemical signals.
“Cells don’t have a GPS,” Kerstjens told reporters. “The only thing a cell ‘sees’ is itself and its neighbors. But its fate depends on where it sits. A cell in the wrong place becomes the wrong thing, and the brain doesn’t develop right.”
To test the model in real biological data, the team analyzed gene expression patterns from the Allen Developing Mouse Brain Atlas, tracking how tens of thousands of genes are expressed across space and time as the mouse brain develops. They applied principal component analysis, reducing the dimensionality of the data to reveal co-expression patterns they called “principal eigengenes.”
The patterns were striking: different eigengenes encode positional information at different spatial scales. Low-order components capture broad anterior-posterior gradients, while high-order components encode fine-grained regional patterns. Remarkably, as few as 10 to 20 genes can accurately predict eigengene values at any location, suggesting that positional information is compressed into a remarkably efficient genomic code.
Cross-Species Conservation
The team validated the model in two vertebrate species. In the mouse, they tracked brain-wide expression across multiple developmental stages using the Allen atlas. In larval zebrafish, they confirmed the same eigengene patterns using single-cell resolution data from a published atlas by Shainer et al. (2023, Science Advances).
The eigengene patterns were conserved between mouse, a mammal, and zebrafish, a teleost fish, indicating that the lineage-based positional mechanism is evolutionarily ancient, predating the divergence of these lineages roughly 450 million years ago.
The patterns also emerged early and persisted across developmental stages, suggesting that positional information established during early neurogenesis remains stable as the brain grows, which is exactly what a lineage-based system would predict.
Beyond Morphogens
Kerstjens emphasizes that the lineage-based model does not replace morphogen gradients but complements them. Early development likely relies on chemical gradients to establish broad axes, while the lineage-based mechanism takes over as the brain expands beyond the range of effective diffusion. The two systems together provide robust positional information at every scale.
The co-authors represent an unusually distributed collaboration. Rodney J. Douglas of the Institute of Neuroinformatics at UZH and ETH Zurich, where Kerstjens began the work during his PhD, is a senior author alongside Florian Engert of Harvard University and Anthony M. Zador of CSHL. Kerstjens joined Zador’s lab at CSHL in 2023.
The implications extend beyond developmental biology. The researchers suggest the same principles may apply to other developing tissues, including tumors, where abnormal lineage relationships could contribute to disordered organization. The work could also inform self-replicating AI systems that need to pass spatial information across artificial generations.
Source: Kerstjens S, Engert F, Douglas RJ, Zador AM. “A lineage-based model of scalable positional information in vertebrate brain development.” Neuron 2026; 114(9):1623-1634.e2. DOI: 10.1016/j.neuron.2025.12.043