← The Frontier
Field Notes6 min read

Field Notes: Scaling Islet Differentiation Past the 1L Barrier

Notes from the bench on what actually breaks when you move iPSC-to-islet differentiation from 125mL spinner flasks to stirred-tank bioreactors — and what we did about it.

Scale-up in cell manufacturing is never linear. Everyone knows this abstractly. Fewer people have lived through the specific failure modes of moving an islet differentiation protocol from a 125mL spinner flask to a 1L stirred-tank bioreactor.

These are notes from that process.

What Breaks First: Shear Stress

The first casualty of scale-up is almost always aggregate integrity. At small scale, gentle agitation in a spinner flask creates a predictable, low-shear environment. In a stirred-tank system, even at equivalent impeller speeds, the shear profile is fundamentally different.

Our first 1L runs produced islet-like clusters with acceptable endocrine marker expression but dramatically reduced insulin secretion in glucose-stimulated insulin secretion (GSIS) assays. The clusters looked right. They didn't function right.

The culprit: mechanical disruption of the tight cell-cell contacts that beta cell maturation requires.

The Fix: Impeller Geometry and Stage-Specific Agitation

The solution wasn't a single change — it was a redesign of the agitation profile at each differentiation stage.

Early stages (definitive endoderm, primitive gut tube) tolerate higher shear. Later stages — particularly the islet maturation phase — require dramatically lower agitation. We now run a staged agitation protocol that drops impeller speed by ~40% during the final maturation window.

Combined with a switch to a pitched-blade impeller configuration, GSIS recovery improved substantially compared to early scale-up attempts.

What We Still Don't Understand

Oxygen gradients in larger vessels remain an open problem. Dissolved oxygen sensors give you a single-point reading; the actual DO profile across a 1L vessel during dense aggregate culture is heterogeneous in ways that are hard to characterize without expensive inline imaging.

This is where the digital twin work becomes important. We're building computational models of the bioreactor environment specifically to address this blind spot.

More in a future dispatch when we have cleaner data.

Views expressed in this post are solely those of Dhruv Sareen in his personal and academic capacity and do not reflect the positions of any affiliated institution or organization. Full disclaimer

← Back to The Frontier