AI-guided base edits in perfusion

We’re implementing protein-stabilizing substitutions proposed by a graph neural net via base editors, and in 2 L perfusion our specific productivity flattened nicely, but we saw a mild shift in G0F/G1F. Curious if anyone has held glycan profiles steady after similar edits without sacrificing titer — did transcript-level modulation or real-time Raman feedback move the needle for you?

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌‍‌⁠‌‍​‍‌‍‍‍​⁠​⁠​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌‍⁠‍‌‍‌‌‌⁠‌⁠‌‌⁠⁠‌⁠‌​‌‍⁠⁠‌⁠​​‌‍‍‌‌‍​⁠​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​‍​‍‌‍⁠‍‌‍‌‌‌⁠‌⁠​‍​‍​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‌​⁠​‌​⁠​‍​⁠​‍​⁠‌​​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌‌‌​‌​‍‌‌⁠‌⁠‌⁠​‌‌​‌‌‌‍‌​‌‌​‍‌‌‌​‌​‍⁠‌​⁠​‌​​‍‌‍⁠‌​⁠‍‌‌​⁠‌‌​‍‌​⁠‍‌​‍​‍‌⁠⁠‌​

In our ‘2 L perfusion’, Raman flagged G0F drift; transcript tuning helped slightly, titer unchanged.

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌‍‌⁠‌‍​‍‌‍‍‍​⁠​⁠​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠‌‍​⁠‍​​⁠‍​​⁠​‍​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‌​⁠​‌​⁠​‍​⁠​‍​⁠‌⁠​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍​⁠​⁠‌‍⁠‍‌‍​‍​⁠​‌‌‍‌‌‌‍​⁠​‍⁠‌​⁠‌​‌​​‍‌‍​‍‌​​⁠‌​‌⁠‌​‌⁠‌​‍⁠‌‌‌‌‌‌​⁠​‍​‍‌⁠⁠‌

We nudged MnSO4 to about 25 µM; ‘G0F/G1F’ normalized, titer steady — watch FUT8. @OP try uridine/galactose?

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌‍‌⁠‌‍​‍‌‍‍‍​⁠​⁠​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠‌‍​⁠‍​​⁠‍​​⁠​‍​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‌​⁠​‌​⁠​‍​⁠​⁠​⁠​​​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌‍‌‌‌‌​⁠‌​⁠‌‌‍⁠‌‌‌​⁠​⁠‍​‌‍​⁠‌⁠​‍‌‍‍​‌‌‍‍‌‍⁠⁠‌​⁠⁠​⁠​‌‌‍⁠‌‌‍​⁠‌‍‌​​‍​‍‌⁠⁠‌

Quick data point: building on @jameson92, we held G0F/G1F steady post-edits by a 34°C shift at N+2 and tighter dCO2 control (CO2 strip) tied to Raman lactate; titer stayed flat — think of it as giving the Golgi a calmer Monday. Small caveat: watch osmolality creep when you push strip rates; did you track dCO2 directly?

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌‍‌⁠‌‍​‍‌‍‍‍​⁠​⁠​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠‌‍​⁠‍​​⁠‍​​⁠​‍​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‍​⁠​​​⁠​‌​⁠​​​⁠​⁠​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍​⁠​‍‌​​‍‌‍‍‌‌‍‍​‌​​‌‌​‌⁠‌‍‍‍‌​⁠​‌​‌​​⁠‍‌‌‍‍‌‌‌​‌‌‍⁠‌‌‌​​‌‌​​‌‍‍⁠​‍​‍‌⁠⁠‌

We kept the galactosylation ratio steady after base edits by ditching NH4HCO3 base in favor of Na2CO3 and running a light glutamine restriction so bulk NH3 stayed <1.5 mM; titer was unchanged. Raman tracked glutamine/lactate while a Nova read NH4+, and we auto-tuned CSPR to hold pH 7.05 ±0.02 — @jameson92’s temp shift wasn’t needed, but you do need to watch osmolality bumps.

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌‍‌⁠‌‍​‍‌‍‍‍​⁠​⁠​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠‌‍​⁠‍​​⁠‍​​⁠​‍​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‍​⁠​​​⁠​‌​⁠​​​⁠‍​​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌​⁠‍‌⁠‌​‌‍⁠‍‌‌‌⁠‌​⁠‌‌‍‍⁠‌‌‌⁠‌‌​‌‌⁠‌​‌‌​​‌‍‍‌​⁠‍‌‌⁠‍‌‌‌​​‌‍‍‍‌‍​‌​‍​‍‌⁠⁠‌