ECONTI ran a continuous E. coli process end to end with model-based control

ECONTI was a research project funded by the Austrian Research Promotion Agency (FFG) to build a continuous manufacturing platform for biotechnological products with a low ecological footprint. enGenes Biotech led the consortium and the upstream process, Qubicon built the automation, BOKU University developed the chromatography on the multi-column system and resin of Tosoh Bioscience, SimVantage contributed the fluid-dynamics models for scale-up, and Novasign the hybrid models and the digital twin of the process. The project ran from 2023 to 2025, and its results are published in Microbial Cell Factories.

The process is one continuous chain. A seed reactor grows the cells and feeds them into a production reactor, whose outflow runs through a filtration that removes the cells. The clarified supernatant goes on to multi-column chromatography, which captures the Protein A. To control a chain like this, the behaviour of every unit operation has to be known, so each one was described by its own model and the models were connected into one digital twin of the process.

The upstream runs in two stages. The seed reactor is held at constant conditions and supplies a steady stream of fresh cells. In the production reactor the strain stops growing and releases Protein A into the culture supernatant, and production stayed stable for more than 500 hours.

To find its best conditions, the production reactor ran an intensified DoE. Within one continuous run its temperature (25, 30 and 35 °C) and pH (6.5, 7.0 and 7.5) were shifted every 36 hours, with the centre point repeated midway and at the end. The titer peaked at 4 g/L at 35 °C and pH 6.5, 60% higher than the about 2.5 g/L at the centre point, and these conditions ran the integrated process.

The outflow of the production reactor ran continuously into a microfiltration with vibrating membranes. The membrane holds the cells back and passes the supernatant with the Protein A on to chromatography, fully automated and without an operator.

Novasign built a hybrid model of the filtration from the process data Qubicon recorded. A neural network inside the mass balances of Novasign’s TFF framework predicts the permeate flow from the speeds of the feed and retentate pumps, and with it how much supernatant reaches the columns.

The loop was built in the FFG research consortium: Qubicon’s automation platform called Novasign’s switching model at each cycle.

Capture ran on multi-column chromatography. While some columns load, the others wash, elute, clean and re-equilibrate, and the columns switch in turn, five of them in the integrated run. The right moment to switch depends on how much Protein A a column can still bind, and that changes with the titer and conductivity of the feed and with the age of the column.

The model describes binding with a Langmuir isotherm whose maximum capacity falls exponentially as the conductivity rises, and corrects it for column aging, about 12% less over the first 100 cycles. A breakthrough model on top of it gives the switch time, the moment just before the product would break through the loaded columns.

During operation, inline HPLC measured the titer and a sensor the conductivity. Novasign’s model predicted the switch time from both, and Qubicon chose the matching chromatography method and sent it to the Tosoh Octave BIO, so the loading times were recalculated automatically. The control ran for 10 days in a test with feed titer and conductivity changed every day, and kept working in the integrated process when an unknown molecule from the culture lowered the binding capacity.

In Novasign Studio the models of the four unit operations were connected into one simulation of the chain. Change the conditions of the reactors, the speeds of the filtration pumps or the conductivity, and the predicted switch time follows.

The process was operated continuously for up to 30 days. Fully integrated, from the bioreactors to the columns, the automated chain ran for 72 hours without manual intervention.

A techno-economic model at 1,200 kg of Protein A a year puts the continuous process at €8,465 per kg, against €12,382 per kg for a fed-batch process, 32% lower.

The same model estimates a 39% smaller facility and 45% less energy for heating, ventilation and air conditioning than the fed-batch process.