Towards real-time monitoring, control and real-time release

Real-time monitoring provides information if the process runs within the specified operating range. This is useful information and can be linked to release criteria, but it could be further exploited for process control. Then raw material variations, fluctuations in process parameters could be controlled and a much more robust process can be obtained. This is then also the step towards automation, because human intervention could be minimized. In bioreactors real-time monitoring is used to control the process to deliver optimal nutrients to reach a high titer or product quality. A typical example is the addition to manganese ions to trigger the glycosylation of antibodies (Tharmalingam et al., 2015) addition of amino acids or control the level of glucose at a minimum level. Bioreactor control is much more advanced compared to downstream processing. Besides pooling there are not a lot of opportunities to control batch chromatography. An interesting approach is to control the gradient shape in batch chromatography to optimize resolution and binding capacity (Sellberg et al., 2017). The authors named it salt trajectory. An optimizer adapts the shape of the gradient according to the retention behavior of product and impurities and the salt gradient is modulated. The wider application of such a controls system would require a fast-real-time monitoring of the effluent stream of the chromatography column, because they still used off-line measurements of the column effluent. Such a system maximizes productivity while resolution is not affected. Usually increasing productivity will be obtained at the expense of reduced resolution. Implementation of soft sensors such as the (Walch et al., 2019) into a column chromatography with a gradient optimizer would lead to an optimal system where product quality and quantity is monitored, and the information is further utilized to control the system and operate under high resolution.
Another dilemma is the optimization of dynamic binding capacity. When chromatography is operated with constant velocity an optimum of productivity at a certain residence time is achieved. At this residence time the column utilization is rather low, but by increasing the column utilization the productivity is reduced. At high velocity/low residence time the breakthrough curve is shallow and at low velocity steep, therefore the dynamic binding capacity and column utilization is higher at low velocity (Carta et al., 2010; Eslami et al., 2022a; Eslami et al., 2022b). One solution is countercurrent chromatography, which allows maximizing column utilization and productivity, but on expense equipment/instrument complexity(Heeter et al., 1996). Countercurrent loading or periodic counter current chromatography is a way to render batch chromatography into continuous chromatography and in addition column utilization is improved. In its simplest form a column is overloaded, and the breakthrough loaded on the second column. Concurrently the third column is washed, eluted, and regenerated. This concept has been extensively used for purification of recombinant antibodies with protein A affinity chromatography(Badr et al., 2021; Davis et al., 2021; Farid et al., 2015; Gerstweiler et al., 2021; Gomis-Fons et al., 2020; Rathore et al., 2022b; Scheffel et al., 2022; Sun et al., 2022; Vetter et al., 2021; Vogg et al., 2018). In order to control the overloading of the first column a control algorithm using absorbance at 280 nm has been implemented. First a breakthrough of the impurities is observed and when the column is close saturation a second breakthrough is caused by the product. This signal can be used as a control algorithm for switching columns (Chmielowski et al., 2017; Gerstweiler et al., 2022; Godawat et al., 2012). This will only work when rather pure feedstock such as an antibody produced in chemical defined media is purified. Crude feedstocks would fully saturate the UV sensor and the difference between the signals of the impurities and product might be too low in order to get useful information to control the loading. In this case a soft sensor such as the Walch et. al. approach (Walch et al., 2019) will be a general solution to control loading in PCC. Near infrared spectroscopy (NIRS) has been also placed at the column entrance and exit and calibrated with a chemometrics approach (Thakur et al., 2020). It is not yet clear if this sensor alone would work for crude solutions or if only the concentration range is wider compared to a UV-cell. If only a wider concentration range must be covered flow VP is a good solution.
Another way to solve the problem of productivity and column utilization is to change the flow rate during loading. A dual flowrate during loading has been proposed by Lacki and coworkers (Ghose et al., 2004a; Ghose et al., 2004b) and then extend to either multiple steps (Ramakrishna et al., 2022) or a linear flow-rate gradient (Chen et al., 2021)or a model predicted gradient (Eslami et al., 2022a; Eslami et al., 2022b; Gomis-Fons et al., 2021; Sellberg et al., 2018)). The loading is started with a high velocity and then subsequently reduced. In order to establish an optimized system an optimizer and controller are necessary. Such a model predictive control results in high productivity and high column utilization but with less complex equipment. In order to run such a controlled process a soft sensor is required at the column in-let and outlet when crude feedstocks are purified and UV280 is not sufficient to monitor product concentration (Eslami et al., 2022a; Eslami et al., 2022b; Sellberg et al., 2018). Such a controller has been implemented in the XAMIris (Evon, Austria) (Figure 8).