Real-time holographic monitoring of insect cell morphology during baculovirus coinfection for adeno-associated virus vector production using machine learning-based classification models.

The baculovirus expression vector system (BEVS) is a scalable platform used to produce recombinant adeno-associated virus vectors (rAAV) in insect cells. A major challenge in this system is reducing the formation of empty rAAV capsids, which do not contain vector DNA and lack therapeutic value. The proportion of empty capsids is influenced by the balance between the two baculovirus constructs that coinfect the producer cells: Bac-Rep-Cap, which supplies AAV replication and capsid proteins, and Bac-GOI-ITR, which delivers the therapeutic gene of interest. Digital holographic microscopy (DHM) is a label-free imaging technique that allows real-time monitoring of cell morphology in suspension cultures. Previous studies have used DHM to track cell density and baculovirus infection; however, its ability to evaluate different coinfections has not been explored. In this study, we combined DHM with machine learning to identify morphological patterns associated with various coinfections for rAAV production. Shaker-flask experiments with different Bac-Rep-Cap: Bac-GOI-ITR ratios created a dataset of cell morphologies to train a predictive classification model. When applied to real-time bioreactor measurements, the model revealed shiftsin the classification patterns related to the initial multiplicity of infection (MOI). The integration of DHM with the classification model has the potential to produce a qualitative "process fingerprint," where deviations in morphological patterns can serve as early indicators of suboptimal coinfection. Such early warning signs enable timely batch termination, reducing downstream processing of inconsistent material. Overall, DHM combined with machine learning offers a non-invasive, real-time tool for process benchmarking and quality assurance in rAAV manufacturing.