Induced pluripotent stem cells are grown in flat monolayers for laboratory studies and in three-dimensional aggregates for manufacturing. The metabolic behavior differs dramatically between the two formats — aggregates develop spatial gradients in oxygen, nutrients, and redox state that monolayers don't have. Scaling up isn't just making the culture bigger; it's moving into a different metabolic regime.
Reader note (2026-11-17): Checked against the full text of arXiv:2603.17090 (Cheng, Jahromi, Wang, Caldwell, Cai, Choy, Auclair, Campbell, Zhao, Ring, Yoon, Harcum and Xie; March 2026), which I had earlier held as matching. The framework description matches: a monolayer kinetic network extended with a multiscale aggregate model, and the paper attributes the spatial and metabolic heterogeneity of aggregates to diffusion limits on oxygen and nutrients. The paper also reports limits on its validation that the essay leaves out: for the static pyruvate cultures the model showed slightly reduced predictive performance, particularly for alanine, which the authors suggest may come from unmodeled compartmentalization of pyruvate; the model substantially underestimated cell density in the LGLL condition; those datasets had two replicates against six for the historic static cultures; and the isotope-labeling results show low 13C enrichment in several TCA-cycle intermediates, which the authors say makes it difficult to distinguish compartmentalization effects from measurement constraints.
The authors (arXiv:2603.17090) build a multiscale mechanistic model linking molecular reactions, single-cell metabolism, and macroscopic culture dynamics. Monolayer kinetics provide the molecular foundation; a biological systems-of-systems framework models the aggregate, coupling intracellular fluxes with extracellular gradients. Isotopic tracers, metabolite profiling, and two-photon redox imaging validate the model across scales.
The through-claim: the gap between laboratory and manufacturing iPSC culture is metabolic heterogeneity — gradients that don't exist in the monolayer but dominate the aggregate. A monolayer model scaled up gives the wrong answers because it assumes uniformity that the aggregate doesn't have. The multiscale model bridges the formats by explicitly representing what changes between them: spatial structure produces metabolic states that a uniform model can't predict.