The dominant method for generating Chinese hamster ovary (CHO) cell lines that produce high titers of biotherapeutic proteins utilizes selectable markers such as dihydrofolate reductase (Dhfr) or glutamine synthetase (Gs), alongside inhibitory compounds like methotrexate (MTX) or methionine sulfoximine (MSX), respectively. Recent work has shown the importance of asparaginase (Aspg) for growth in media lacking glutamine–the selection medium for Gs-based selection systems. We generated a Gs/Aspg double knockout CHO cell line and evaluated its utility as a novel dual selectable system via co-transfection of Gs-Enbrel and Aspg-Enbrel plasmids. Using the same selection conditions as the standard Gs system, the resulting cells from the Gs/Aspg dual selection showed substantially improved specific productivity and titer compared to the standard Gs selection method, however, with reduced growth rate and viability. Following adaptation in selection medium, the cells improved viability and growth while still achieving ~5-fold higher specific productivity and ~3-fold higher titer than Gs selection alone. We anticipate that with further optimization of culture medium and selection conditions this approach would serve as an effective addition to workflows for the industrial production of recombinant biotherapeutics.

The control of nutrient availability is critical to large-scale manufacturing of biotherapeutics. However, the quantification of proteinogenic amino acids is time-consuming and thus is difficult to implement for real-time in situ bioprocess control. Genome-scale metabolic models describe the metabolic conversion from media nutrients to proliferation and recombinant protein production, and therefore are a promising platform for in silico monitoring and prediction of amino acid concentrations. This potential has not been realized due to unresolved challenges: (1) the models assume an optimal and highly efficient metabolism, and therefore tend to underestimate amino acid consumption, and (2) the models assume a steady state, and therefore have a short forecast range. We address these challenges by integrating machine learning with the metabolic models. Through this we demonstrate accurate and time-course dependent prediction of individual amino acid concentration in culture medium throughout the production process. Thus, these models can be deployed to control nutrient feeding to avoid premature nutrient depletion or provide early predictions of failed bioreactor runs.

Despite their therapeutic potential, many protein drugs remain inaccessible to patients since they are difficult to secrete. Each recombinant protein has unique physicochemical properties and requires different machinery for proper folding, assembly, and post-translational modifications (PTMs). Here we aimed to identify the machinery supporting recombinant protein secretion by measuring the protein-protein interaction (PPI) networks of four different recombinant proteins (SERPINA1, SERPINC1, SERPING1 and SeAP) with various PTMs and structural motifs using the proximity-dependent biotin identification (BioID) method. We identified PPIs associated with specific features of the secreted proteins using a Bayesian statistical model, and found proteins involved in protein folding, disulfide bond formation and N-glycosylation were positively correlated with the corresponding features of the four model proteins. Among others, oxidative folding enzymes showed the strongest association with disulfide bond formation, supporting their critical roles in proper folding and maintaining the ER stability. Knock down of ERP44, a measured interactor with the highest fold change, led to the decreased secretion of SERPINC1, which relies on its extensive disulfide bonds. Proximity-dependent labeling successfully identified the transient interactions supporting synthesis of secreted recombinant proteins and refined our understanding of key molecular mechanisms of the secretory pathway during recombinant protein production.