Quality control of the culture media during early stages of NK-cell expansion process

Introduction: Natural killer (NK) cells have emerged as a promising platform for cancer immunotherapy due to their ability to recognize and eliminate malignant cells without prior sensitization or antigen presentation. Monitoring the composition of culture media throughout the expansion process may provide valuable information for process control within a quality-by-design (QbD) framework.

Objectives: This work applies Fourier-transform infrared (FTIR) spectroscopy combined with machine learning to characterize the global molecular signature of culture media used during NK-cell expansion and to discriminate between different culture media formulations, including compositionally similar media.

Methods: NK cells were isolated from umbilical cord blood samples under legal procedures. Cells were cultured in eight-commercially media: M1-M8. Culture media samples were analysed using a Vertex-70 FTIR-spectrometer with HTS-XT high-throughput accessory. Supervised-classification models were developed based on the best 578-features from normalized first-derivative spectra using the Information Gain algorithm.

Results: A k-nearest neighbours (kNN) model achieved an F1-score of 0.725 for classifying the eight media across days 0-27. To evaluate early process monitoring, models were trained using samples from the first nine culture days only. A Naïve Bayes (NB) classifier reached an F1-score of 0.669, indicating that medium identity can be inferred from FTIR spectra at early culture stages. For the closely related media M1, M2 and M3, NB classifiers achieved F1-scores of 0.728 (days 9-20) and 0.731 (days 0-9).

Conclusions:The results indicate that FTIR-spectroscopy combined with machine learning can successfully discriminate between different culture media used in NK-cell manufacturing, including formulations with similar compositions. Moreover, medium identification is feasible from samples collected during early culture stages, highlighting the potential of this rapid, non-destructive, and cost-effective approach for process monitoring and quality control in NK-cell expansion workflows.