Upstream process development and robust analytical characterization of bioreactor cell culture processes are key to the successful production of live virus vaccines (LVVs). Viable cell density (VCD) and cell viability (CV) are among the most important parameters governing the outcome of microbial and mammalian cell culture production of biopharmaceuticals. While a number of offline techniques exist for measuring VCD and CV, they are time consuming and prone to high variability, limiting their utility for process characterization and ultimately process control. In this work, we explore dielectric spectroscopy as an in situ process analytical technology (PAT) for real time characterization of VCD and CV for an LVV process, with the goal of enabling enhanced bioreactor upstream process development and commercialization. To date, LVV processes have not routinely utilized dielectric spectroscopy for accurate quantification of VCD and CV, largely due to challenges caused by dynamic changes associated with cell growth and death specific to LVV production. Thus, the capability to monitor VCD and CV in situ in real time for LVV processes remains a pressing need that will be highly enabling to LVV process development & manufacturing.

In LVV production, the expected VCD and CV trends undergo significant changes over the course of the process, as outlined in Fig. 1 (Liu, 2017, Tapia et al., 2016). At the process outset, VCD is initially low and begins to grow in an approximately sigmoidal fashion, eventually plateauing. At this point, a media exchange may be performed, and viral infection is initiated. This is followed by a stationary phase, where VCD remains approximately constant and viral infection begins to progress. The initial lag phase for cell growth persists for roughly 36–48 h. This is followed by the exponential growth phase, which lasts for roughly 48–60 h, after which cell growth plateaus into the deceleration phase, during which viral infection is carried out. In the present LVV process, the stationary phase lasts for ∼ 4 days, during which viral replication takes place and accelerates, leading into the final death phase, which is allowed to proceed for ∼ 4 days until the culture is harvested. As viral replication accelerates, VCD begins to drop off, often sharply, during the viral replication and cell death phase. In general, cells may undergo varying degrees of rupture/lysis during which the virus is released into the cell culture media. In the process reported, not all cells are expected to fully rupture prior to harvest of the culture (Robinson et al., 2014, Feng et al., 2013, Jones et al., 2021, Bird et al., 2014). CV tends to follow similar trends to VCD, but may differ due to CV reflecting a ratio of viable to total cells. CV may thus vary more widely from process to process, particularly during the growth and death phases. Either during or at the end of the cell death phase, the culture is harvested to obtain the targeted viral product. Establishing control and consistency over VCD and/or CV throughout a process is key to process robustness and achieving a desirable viral titer yield at the time of harvest.

A variety of offline assays for monitoring VCD and CV have been developed. Common examples include manual hemocytometry and dye staining or automated cell counting instrumentation (Pappenheimer, 1917, Louis et al., 2011, Piccinini et al., 2014, Chan et al., 2015, Pamphilon et al., 2013). Differentiation of total, dead, and/or viable cells using these offline analytical strategies is performed typically by visualization (e.g. manual observation or automated imaging). Each of these methods typically involve a similar approach in that cell culture aliquots are collected, sometimes followed by adjustments of total concentration, and cells are stained to facilitate visualization. These approaches are time consuming, introduce the potential for user error, create potential safety hazards (particularly during a viral infection), and the results can be highly variable (Piccinini et al., 2017, Felice et al., 2009, Simon et al., 2016).

PAT offers a means to circumvent these challenges presented by traditional offline cell counting assays. PAT enables in situ process monitoring through real time measurements of process data, such as temperature, pH, particle size, or dissolved metabolites, offering the opportunity for accelerated process development and scale up (Schaefer et al., 2014, Chanda et al., 2015, Wasalathanthri et al., 2020, Rathore et al., 2010, Chen et al., 2011, Scott and Wilcock, 2006, Smith et al., 2021, Ralbovsky and Smith, 2021, James et al., 2006, Lomont et al., 2022, Lomont and Smith, 2022, Ralbovsky et al., 2022, Ralbovsky et al., 2022, Smith et al., 2019, Smith et al., 2020, Wei and Smith, 2023). For the purposed of tracking VCD and CV, a probe-based dielectric spectroscopy PAT measurement (also known as capacitance) can enable real time, in situ monitoring directly in the bioreactor (Asami et al., 1976, Irimajiri et al., 1975, Yardley et al., 2000, Harris and Kell, 1983, Markx and Davey, 1999, Carvell and Dowd, 2006, Metze et al., 2020). A dielectric spectroscopy probe is minimally perturbative to the process and can be implemented across a wide range of scales from milliliters to thousands of liters, and both re-usable and single-use configurations are available. The operating principle behind the measurement relies of viable, intact cells polarizing across their membranes in response to an applied field (Fig. S1); this polarization is detected by the probe as a change in the measured dielectric (Asami et al., 1976, Irimajiri et al., 1975, Yardley et al., 2000, Harris and Kell, 1983, Markx and Davey, 1999, Carvell and Dowd, 2006, Metze et al., 2020). Dead cells, with presumably damaged or ruptured membranes, will not exhibit this same polarization of ions across their membranes and thus should not contribute to the response measured, allowing selective detection of viable cells (Kell et al., 1998, Harris et al., 1987, Stoicheva et al., 1989). Thus, there is a fundamental relation to expect that dielectric spectroscopy can directly correlate with VCD. Given that CV is reported as a ratio of viable to total cells, it may also correlate well with dielectric spectroscopy. However, as compared to VCD, CV may be expected to show greater deviation in some cases (i.e. total cell counts can vary for a given value of CV). Dielectric spectroscopy can be measured at a wide range of frequencies, and establishing a high correlation between this in-line PAT technology and the existing “gold-standard” offline methods for measuring VCD and CV represents an important step toward the widespread adoption of dielectric spectroscopy as a measure of these key parameters in biopharmaceutical production processes, including LVV production processes. While capacitance measurements have been explored as a means of tracking VCD (Asami et al., 1976, Irimajiri et al., 1975, Yardley et al., 2000, Harris and Kell, 1983, Markx and Davey, 1999, Carvell and Dowd, 2006, Metze et al., 2020), widespread implementation for direct process monitoring and control in LVV processes remains limited. The vast majority of work to date in this field has focused on recombinant protein expression cell culture (non-LVV) applications and have not routinely employed or explored multivariate data analysis approaches. Establishing a general approach for generating strong correlations with traditional offline cell viability metrics throughout the entire course of a cell cultivation process, particularly during the viral replication and cell death phase of LVV processes, which present unique challenges in LVV applications, remains difficult (Metze et al., 2020, Cannizzaro et al., 2003, Opel et al., 2010, Metze et al., 2020).

In this work, we explored dielectric spectroscopy collected across 25 frequencies, spanning 50 to 20,000 kHz, in a biopharmaceutical LVV upstream bioreactor cell culture process using adherent Vero cell culture on microcarriers. Vero cells have received significant attention for their capabilities in producing a wide variety of LVVs, and they are currently the most used continuous cell line for viral vaccine manufacturing, making them a highly relevant target for the present study (Kiesslich and Kamen, 2020, Kiesslich et al., 2020, Yang et al., 2019, Genzel et al., 2010, Mangion et al., 2020, Chen et al., 2011, Sène et al., 2022). Dielectric spectroscopy as a PAT measurement offers an especially attractive alternative to traditional offline cell counting assays for microcarrier growth processes, a historically difficult sample matrix. We analyzed the measured dielectric spectra across all frequencies for correlation to offline measurements of VCD and CV obtained using a traditional offline cell counting assay. Both univariate and multivariate analysis, in the form of partial least squares (PLS) regression, were explored. The optimal univariate measurement frequencies were successfully determined. Multivariate PLS-based approaches were observed to offer significantly improved correlations of in situ dielectric spectroscopy data to offline VCD and CV as compared to univariate approaches.

To the best of our knowledge, this is the first report of an inline, real time analytical methodology to successfully predict VCD and CV during live virus-based vaccine production. Given that LVVs represent one of the oldest and historically successful approaches to vaccine development, the potential impact of this new methodology can be highly enabling for rapid, innovative vaccine production. We note that the process conditions associated with LVV cell culture are significantly different from those associated with perfusion cell culture and monoclonal antibody production, where dielectric spectroscopy has been much more widely applied. Specifically, application of dielectric spectroscopy to real time monitoring of LVV processes has historically been challenging due to the difficulty of accurately tracking viable cell density viability during the cell death phase as viral titer increases, cell morphology changes, and cells may clump/aggregate, among other changes that take place. Thus, development of an analytical methodology capable of successfully predicting VCD and CV for LVV processes serves to fill an existing gap and pressing need for LVV production.

Here we demonstrate that a single fit can be achieved throughout all phases of the process via multivariate analysis. VCD and CV are crucial for evaluating the harvest timing for the culture and ensuring the robustness of the LVV process being carried out. As such, our new proposed analytical methodology utilizing in situ dielectric spectroscopy with multivariate analysis offers a new, significant ability to progress vaccine production within this live virus modality. Overall, this work demonstrates the power of dielectric spectroscopy for real time, in situ monitoring of VCD in an LVV process. The fundamental approach to data collection and data analysis demonstrated in this work are broadly applicable and readily transferable to other biopharmaceutical viral vaccine production processes, including within diverse cell lines, viral vectors, and scales.