Growing a Root from Its Genome : Emergent Tissue Layers and Zones in a GPU-Native Simulation of the Arabidopsis Root Tip

Models of the Arabidopsis root tip usually take its cell identities as given. Here we grow a root tip from a single genome. In a two-dimensional median section, every cell reads the same network of 62 genes (26 core genes from an earlier, unpublished model of callus and 36 added for the root) and starts with the same expression; the only spatial cues are the polarity of auxin transport, auxin from the shoot and the free surface. Identities and zones are never given to the simulation; they are read from gene expression. Cells are polygons of a shared mesh, relaxed towards mechanical equilibrium under turgor at every step. In 48 simulated hours, a tip-shaped template grows into a root of 3 000-4 500 cells with epidermis, cortex, endodermis, pericycle and stele in order along most of its length, SHORT-ROOT and SCARECROW in the endodermis, an auxin maximum, a quiescent centre and a cap at its tip, and meristem, elongation and mature zones. A WOX5 gate built from published interactions and one hypothesis keeps a quiescent centre to 48 h in 17 of 20 simulated roots that were not used for tuning (85%; Wilson 95% interval 64-95%; criteria fixed before the run), against 9 of 20 roots used for tuning, before the gate. The niche is not maintained beyond 48 h: grown on to 73 h, only 6 of these 20 roots still have a quiescent centre (30%; 15-52%). All pre-registered criteria hold in only 1 of 20 roots, because the quiescent centre is too wide at 0 h; without that count, 12 of 20. The meristem is too long, and cortical cells made after the start of growth elongate only three- to four-fold instead of eightto thirty-fold, a limit set in large part by the numerical growth scheme. Mutants are a major weakness: several look like their published phenotype, but an exploratory re-run of a mutant benchmark fixed in advance for our first model passes one of nine predictions. Nor does the root regenerate: cut near its tip, it re-forms an auxin maximum and a columella but no quiescent centre. On a first, simpler model and without any polarity cue, a mass of identical cells still sorts into concentric layers and, when PIN carriers follow the auxin flux, concentrates auxin into a few foci, but forms no stem-cell niche: the genome lacks a lateral inhibition between foci. We state the limitations and the next stages.