Oxygen-based soft-sensing of glucose demand in Chinese hamster ovary fed-batch cultures across scales Authors: Valentin Jossen, Adrian Joseph, Gonzalo Mesado, Siewhoon Lim, Gilles Broccard Published: 2026 ABSTRACT AbstractAbstractGlucose is a critical nutrient in mammalian cell culture processes, influencing cellular metabolism, productivity, and product quality. In industrial Chinese hamster ovary (CHO) fed-batch processes, glucose concentrations are commonly controlled using manual offline measurements and operator-based feed adjustments. Although online Process Analytical Technology (PAT) solutions and oxygen uptake rate (OUR)-based feeding approaches have previously been investigated, their implementation in existing Good Manufacturing Practice (GMP) manufacturing environments remains challenging due to instrumentation, validation, and process integration requirements. In this study, the relationship between cumulative oxygen consumption and glucose utilization was investigated across 58 CHOK1SV GS-KO® fed-batch cultures performed in stirred glass and single-use bioreactors ranging from 10 L to 2 kL scale. Building on previously reported oxygen-based glucose estimation concepts, a lactate-adjusted framework was developed to account for both lactate production and re-consumption during the fed-batch process. Oxygen transfer rates were calculated from standard online process parameters and integrated over time to determine cumulative oxygen consumption, which was subsequently correlated with cumulative glucose utilization. The results demonstrated robust linear correlations between cumulative oxygen consumption and lactate-adjusted cumulative glucose utilization across different projects, cell lines, and process scales (R2 ≥ 0.984). Incorporation of lactate metabolism improved the robustness of the oxygen-glucose relationship, particularly during metabolic transitions associated with lactate re-consumption. The predictive capability of the framework was further evaluated in three 200 L fed-batch processes by prospectively estimating glucose consumption over sequential 24 h intervals using cumulative oxygen consumption and routine offline metabolite measurements. Overall, the study demonstrates that cumulative oxygen consumption provides a robust and scalable basis for oxygen-based soft-sensing of glucose demand in industrial CHO fed-batch cultures. The presented framework enables predictive glucose demand estimation using standard bioreactor process data without requiring additional glucose-specific PAT hardware and may support future implementation of automated feeding concepts in mammalian cell culture processes.Key points• Cumulative oxygen consumption correlates linearly with glucose utilization.• Lactate-adjusted O₂–glucose correlations are consistent across bioreactor scales.• Glucose control can be automated using existing bioreactor process data.Supplementary InformationThe online version contains supplementary material available at 10.1007/s00253-026-13882-2.Keywords: Oxygen transfer rate (OTR), O2 [cumulative], Glucose control, Gluc[cumulative], CHOK1SV GS-KO®, Fed-batch, Soft-sensing Introduction The production of biopharmaceuticals through mammalian cell culture processes has enabled major therapeutic advances over recent decades (Lu et al. 2020). In particular, Chinese hamster ovary (CHO) cells have become the predominant production platform for monoclonal antibodies and other recombinant proteins due to their robust growth characteristics, scalability, and ability to generate human-compatible post-translational modifications. Cell culture media plays a central role in these processes by supplying essential nutrients, growth factors, and metabolic precursors required for cell proliferation and recombinant protein production (Yao and Asayama 2017; Ritacco et al. 2018; Combe and Sokolenko 2021). In addition, media composition strongly influences the accumulation of inhibitory metabolites such as lactate and ammonia, which can negatively affect cell growth, productivity, and product quality (Pereira et al. 2018; Hartley et al. 2018; Luginsland et al. 2024). Continuous optimization of cell culture media and feeding strategies has therefore been essential for achieving higher viable cell densities and increasing monoclonal antibody titers in industrial fed-batch processes (Landauer 2014; Reinhart et al. 2015; Pan et al. 2017; Galbraith et al. 2018). Among the key nutrients used in mammalian cell culture, glucose plays a particularly important role as a primary carbon and energy source. Beyond its role in cellular metabolism, glucose availability also influences glycosylation patterns, protein expression, and overall product quality. Consequently, maintaining appropriate glucose concentrations throughout the fed-batch process is critical for robust and reproducible bioprocess performance. In industrial manufacturing environments, glucose concentrations are commonly controlled using routine offline measurements combined with manual adjustment of glucose feed rates. Although this approach is straightforward to implement and broadly compatible with existing Good Manufacturing Practice (GMP) infrastructure, it remains susceptible to operator-dependent variability and delayed response to metabolic changes occurring during cultivation. To improve process monitoring and feeding accuracy, various Process Analytical Technology (PAT) approaches have been investigated for mammalian cell culture applications (Nikita et al. 2023). These include enzymatic and electrochemical glucose sensors (Spichiger and Spichiger‐Keller 2010; Bauer et al. 2012; Lederle et al. 2021; Reyes et al. 2022), spectroscopic methods such as Raman spectroscopy (Webster et al. 2018, 2021, 2023; Olin et al. 2023), and soft-sensing concepts based on oxygen uptake rate (OUR) measurements. In particular, OUR- and oxygen transfer rate (OTR)-based approaches have been explored as indirect indicators of cellular metabolic activity and substrate demand (Winckler et al. 2014; Martínez-Monge et al. 2019; Ihling et al. 2021, 2022). Earlier OUR-based feeding concepts were already investigated in hybridoma and mammalian cell culture processes during the 1990s, where oxygen consumption was correlated with glucose utilization to support feeding strategies. More recently, Goldrick et al. demonstrated online glucose estimation in CHO cultures using OTR measurements (Goldrick et al. 2018). These studies demonstrated the general feasibility of oxygen-based glucose estimation approaches, but also highlighted practical implementation challenges associated with bioreactor-specific engineering characterization, determination of kLa correlations, dependence on gas composition and hydrodynamics, and the continued requirement for offline metabolite measurements. An additional complexity in mammalian cell culture processes arise from lactate metabolism. During the exponential growth phase, CHO cells frequently convert glucose to lactate despite sufficient oxygen availability, a phenomenon commonly referred to as the Warburg effect (Warburg 1956). Following metabolic adaptation, lactate may subsequently be re-consume by the cells, thereby altering the apparent relationship between glucose uptake and oxygen consumption (Hartley et al. 2018; Luginsland et al. 2024). Consequently, oxygen-based estimation of glucose demand may require consideration of both lactate production and lactate re-consumption in order to accurately describe metabolic substrate utilization throughout the fed-batch process. The present study does not introduce oxygen-based glucose estimation as a fundamentally new concept. Instead, it extends previously reported OUR/OTR-based approaches by incorporating lactate-adjusted glucose utilization and evaluating the resulting oxygen-glucose correlations across industrially relevant CHO fed-batch scales ranging from 10 L to 2 kL. Historical datasets from 58 fed-batch cultivations performed in stirred glass and single-use (SU) bioreactors were analyzed with respect to oxygen transfer, glucose utilization, and lactate metabolism. Based on these analyses, a lactate-adjusted oxygen-based soft-sensing framework for predictive glucose demand estimation was developed and subsequently evaluated in 200 L proof-of-concept fed-batch cultivations using standard online process data and routine offline metabolite measurements. The overall objective of the study was therefore to evaluate whether cumulative oxygen consumption can provide a robust and scalable basis for predictive glucose demand estimation in industrial CHO fed-batch processes without requiring additional glucose-specific PAT hardware or extensive retrofitting of existing bioreactor infrastructure. Material and methods Theoretical and modeling aspects The proposed framework was developed as an oxygen-based soft-sensing approach intended to estimate glucose demand from routinely available bioreactor process data. The methodology therefore combines standard online process parameters with routine offline metabolite measurements. Since oxygen-based estimation of metabolic activity strongly depends on bioreactor-specific engineering characteristics and process conditions, bioreactor characterization and kLa determination were required prior to model application. The general concept of the model and the influencing parameters utilized for the data analysis are illustrated schematically in Fig. 1. Online process parameters, including impeller speed, temperature, volumetric gas flow rates (Air, O2, CO2, N2), bioreactor weight, and dissolved oxygen (DO) concentration, were continuously recorded during the cultivation processes and used for online calculation of the OTR. Offline measurements of glucose and lactate concentrations as well as the amount of glucose added to the bioreactor were subsequently used to calculate cumulative glucose utilization. A simplified representation of central carbon metabolism was considered to establish the relationship between cumulative oxygen consumption and glucose utilization. In this context, lactate formation during glycolysis was considered as an intermediate metabolic state that does not directly consume oxygen. Furthermore, oxygen consumption was assumed to be predominantly associated with oxidative glucose metabolism, while the contribution of amino acid oxidation replenishing tricarboxylic acid (TCA) cycle intermediates was considered secondary for the purpose of the present analysis. The implemented lactate correction should therefore be interpreted as a pragmatic approximation intended to improve the correlation between oxygen consumption and apparent glucose utilization rather than as a mechanistic model of cellular metabolism or lactate regulation. A detailed mathematical description of the OTR calculations and lactate-adjusted glucose consumption calculations is provided in the following sections. Schematic representation of the different components of the framework used for the data analysis. A Bioreactor illustrating the key process parameters used for calculation of cumulative oxygen consumption O2[cumulative] and cumulative glucose utilization Gluc[cumulative]. B Simplified representation of oxygen transfer in stirred bioreactor and central carbon metabolism in CHO cells, highlighting the relationship between glucose (Gluc), lactate (Lac), and oxygen (O2) Oxygen transfer rate and cumulative oxygen consumption The temporal change in dissolved oxygen concentration can generally be described by the difference between oxygen transfer into the liquid phase and oxygen uptake by the cells according to Eq. 1. The OTR describes the oxygen transfer from the gas phase into the liquid phase and depends on the volumetric mass transfer coefficient (kLa) and the concentration difference between the oxygen saturation concentration (cO2∗) and the actual dissolved oxygen concentration (cO2) according to Eq. 2. The dissolved oxygen probes provided relative oxygen saturation values (% DO), which were converted into absolute dissolved oxygen concentrations using the corresponding oxygen saturation concentration under the prevailing process conditions. The equilibrium oxygen concentration was dynamically calculated considering process temperature, pressure, and gas composition throughout the fed-batch process. The applied gas composition included air, oxygen enrichment, nitrogen, and carbon dioxide additions used for dissolved oxygen and pH control. The OUR was expressed as the product of the cell-specific oxygen consumption rate (qO2) and the viable cell concentration (Xv). Under controlled dissolved oxygen conditions, dO2dt was assumed to be approximately zero for instantaneous OUR estimation, resulting in OTR ≈ OUR according to Eq. 3. This quasi-steady-state assumption was applied only for instantaneous oxygen transfer calculations under DO-controlled conditions. In contrast, cumulative oxygen consumption was subsequently determined by temporal integration of the calculated OTR values. The kLa was estimated using bioreactor-specific Van’t Riet correlations according to Eq. 4. Bioreactor-specific Van’t Riet correlations were established from experimentally determined kLa datasets obtained under representative process conditions. Based on the calculated OTR values, cumulative oxygen consumption was determined by temporal integration over fed-batch duration according to Eq. 5. The cumulative oxygen consumption (O2[cumulative]) was subsequently obtained by summation over all process intervals according to Eq. 6. Cumulative glucose concentration consumed To estimate glucose utilization during the fed-batch processes, a simplified lactate-adjusted glucose consumption model was implemented. Since lactate production and re-consumption alter the apparent stoichiometric relationship between glucose uptake and oxygen consumption, incorporation of lactate metabolism was expected to improve the robustness of oxygen-based glucose estimation during metabolic transitions. For this purpose, the lactate/glucose yield (Y[Lac/Gluc]) was calculated for each time interval according to Eq. 7. The amount of glucose consumed (Gluc[con,t]) during each process interval was subsequently calculated according to Eq. 8. Each process interval was classified either as a lactate production phase (Lac[t-1] < Lac[t]) or a lactate consumption phase (Lac[t-1] > Lac[t]) depending on the temporal lactate concentration profile. During lactate production phase, glucose utilization was corrected according to the calculated lactate/glucose yield coefficient. During lactate re-consumption phases, lactate utilization was incorporated using a simplified stoichiometric factor. The cumulative glucose consumption (Gluc[cumulative]) was subsequently calculated according to Eq. 9. The implemented lactate correction should be interpreted as a metabolic adjustment intended to improve oxygen-glucose correlations rather than as a comprehensive mechanistic representation of central carbon metabolism. Since lactate metabolism in CHO cells is additionally influenced by metabolic adaption, mitochondrial activity, process temperature, pH, and amino acid metabolism, the presented model was intentionally simplified to support robust process-scale evaluation across multiple industrial datasets. Cell lines and fed-batch cultures In this study, data from different CHOK1SV GS-KO® fed-batch cultures producing various monoclonal antibodies were analyzed with respect to their oxygen, glucose consumption and lactate metabolism (Table 1). Fed-batch processes were evaluated across multiple bioreactor scales ranging from 10 L stirred glass bioreactors to 2 kL single-use stirred bioreactors. All cell lines were cultured using Lonza`s proprietary chemically defined and animal component-free platform media and feeds. Cells were thawed from research or working cell banks according to standard procedures and expanded through conventional seed train operations prior to inoculation of the production bioreactors. Following inoculation, the cells were cultivated for 15 days using Lonza`s fed-batch platform process. During the fed-batch process, glucose feed solutions were continuously added following feed initiation using a manually adjusted feeding strategy intended to maintain residual glucose concentrations within the process-specific target range (in accordance with the specification used by Webster et al. Webster et al. 2018, 2021, 2023). The existing feeding strategy already provided a robust glucose management across scales and therefore served as the reference process condition for evaluation of the oxygen-based predictive framework developed in the present study. The temperature was initially maintained at 36.5 °C and then adjusted to a lower temperature for the remainder of the 15-day culture period. The dissolved oxygen concentration and pH levels were monitored and regulated using in-line probes from Mettler Toledo. The DO value was maintained at a defined setpoint using combinations of air, O2, and N2 sparging. The pH was initially maintained at a setpoint and subsequently shifted to a more alkaline value for the remainder of the 15-day culture period. The pH was maintained using CO2 sparging and the addition of an alkaline solution. Table 1Overview of analyzed projects (n = 58 fed-batch processes)Number of bioreactors per scale and projectProject10L STR50L STR200L STR1kL STR2kL STRLP-001-221-LP-002-111-LP-003-111-LP-004-121-LP-005-111-LP-006--11-LP-007-22--LP-008-221-LP-009-211-LP-01021--4LP-0114-1-2LP-012--1-10 Overview of analyzed projects (n = 58 fed-batch processes) Offline analytics During each fed-batch process, samples were collected from the bioreactor on a daily basis for offline analysis of substrates, metabolites and cell growth. Glucose, lactate, and other metabolites were analyzed offline using the NOVA BioProfile FLEX 2 (Nova Biomedical), while viable (VCD) and total (TCD) cell densities were quantified with the Vi-Cell BLU (Beckmann Coulter). Bioreactor systems Process data were evaluated for benchtop (10 L glass STR), pilot-scale (50 L and 200 L single-use STR), and production-scale (1 kL and 2 kL single-use STR) bioreactors. Comprehensive bioprocess engineering data were available for all systems, owing to complete characterization of mixing and mass-transfer properties. Data analytics Process data analysis was performed using the open-source software R (version 4.2.1). Online process data streams obtained from MFCS or DeltaV systems and offline analytical measurements were consolidated into structured datasets comprising 12 projects and 58 fed-batch cultivations. Offline measurements were aligned with the online process data based on batch identifiers and electronic sampling timestamps. The majority of the analyses were performed retrospectively using historical process datasets. In addition, prospective proof-of-concept evaluations were performed at 200 L scale by predicting glucose consumption over sequential 24 h intervals based on cumulative oxygen consumption and routine offline metabolite measurements without active close-loop process control implementation. Statistical analyses were performed using the R packages ggpmisc (version 0.5.5) and metrics (version 0.1.4). Results Characterization of the volumetric mass transfer coefficient To evaluate oxygen mass transfer characteristics across the investigated bioreactor scales, experimentally determined kLa datasets (data not shown) obtained under representative process conditions were used to establish Van’t Riet correlations according to Eq. 4. The characterization studies were performed in alignment with Bandyopadhyay et al. 1967 and Bauer et al. 2020 for SU bioreactor characterization. The evaluated process conditions covered P/V values ranging from 14 to 61 W/m3 and ug values ranging from 7.6∙10–5 to 1.1∙10–3 m/s. The calculated kLa values showed good agreement with experimentally determined values and with literature data reported for comparable SU bioreactor systems (Noack et al. 2011; Löffelholz et al. 2013; De Wilde et al. 2014). Figure 2 compares normalized measured and calculated kLa values obtained using the established Van’t Riet correlations. The resulting correlation demonstrated strong predictive agreement (R2 = 0.997, n = 47), with deviations remaining withing a range of −5.7% to + 10.1%. The observed agreement indicated that the established correlations were sufficiently accurate for subsequent OTR estimation under the process conditions investigated. At the same time, the bioreactor-specific characterization effort required for determination of kLa correlations and P/V relationships illustrate one of the practical implementation challenges associated with oxygen-based soft-sensing approaches in industrial bioprocessing environments. Comparison of normalized kLa values (measured versus calculated) obtained using the established Van’t Riet correlation. The dashed lines indicate the deviation band (−5.7% to + 10.1%) encompassing the scatter between the experimentally determined and calculated normalized kLa values Calculation of OTR and O2 [cumulative] during the fed-batch processes Based on the online process parameters and the Van’t Riet correlations, OTR values were calculated throughout the fed-batch processes for all investigated cultivations (Fig. 3). In all cases, oxygen demand increased substantially during the initial 120–150 h of cultivation, corresponding to rapid cell proliferation and increasing metabolic activity. After approximately 160 h, OTR values approached a plateau phase and remained relatively stable until the end of the fed-batch process. Maximum calculated OTR values reached up to 175 mg O2/L/h (equivalent to 5.4 mmol O2/L/h), depending on the respective project and cell line. Comparable OTR profiles and maximum oxygen transfer rates were observed for the same cell lines across different process scales, indicating robust scalability of the metabolic oxygen demand. Comparable OTR values in the range of 38–118 mg O2/L/h (= 1.2–3.9 mmol O2/L/h) have also been described in literature by various authors (Huang et al. 2010; Goldrick et al. 2018; Pappenreiter et al. 2019; Ihling et al. 2021, 2022) for viable CHO cell concentrations in the range of 9–20 × 106 cells/mL in shake flasks and stirred bioreactors. The correlation between OTR and VCD also allows the estimation of qO2 during the fed-batch processes. It is noteworthy that the calculated values for qO2 agreed well with values reported by Seidel et al. (Seidel et al. 2021) and Maschke et al. (Maschke et al. 2022), who reviewed and summarized qO2 literature values for CHO cells. To reduce variability associated with instantaneous OTR fluctuations and to establish a more robust descriptor of overall metabolic activity, cumulative oxygen consumption O2[cumulative] was determined by temporal integration of the calculated OTR values over the fed-batch duration. The final summation of the integrated OTR values enhanced the precision of the calculations by reducing the variability of O2 during the processes. The resulting O2[cumulative] values were subsequently correlated with Gluc[cumulative]. Time-dependent profiles of the OTR in the investigated bioreactors during the fed-batch processes. The data are grouped according to the respective project and batch number. Cell growth, glucose consumption, and the consideration of lactate metabolism Figure 4 shows the time-dependent VCD profiles for the investigated fed-batch cultivations across all process scales. While differences in maximum VCD were observed between projects due to cell line-specific and process-specific characteristics, comparable growth trends were obtained for the same cell line across different bioreactor scales. Peak viable cell densities were reached on average after approximately 237 h of cultivation. The relatively small scale-dependent differences in VCD indicated robust process scalability and consistent metabolic behavior across the investigated bioreactor systems. Similar time-dependent VCD profiles and peak cell concentrations (15–32 × 106 cells/mL) have also been described in literature for similar CHO fed-batch processes and stirred bioreactors (Xu et al. 2017; Webster et al. 2018, 2021). Cultures with slower cell growth and lower maximum VCD exhibited a lower oxygen demand, as evidenced by the OTR profiles during the fed-batch processes (e.g., LP-003 and LP-004). Therefore, calculated oxygen consumption was a reliable indicator of the cells’ metabolic activity. Time-dependent profiles of viable cell density (VCD) during the investigated fed-batch processes. The data are grouped according to the respective project number As a consequence of the rising number of viable cells and the resulting metabolic activity, a decrease in glucose concentration was observed over the initial 48 to 119 h (dependent on the specific process version) of the fed-batch processes (Fig. 5A). After initiation of the glucose feed, the glucose concentration was maintained within the desired operational range for the residual glucose, and near the target concentration. Glucose trends comparable to those shown in Fig. 5A were also described by Webster et al. for Lonza’s fed-batch platform processes in stirred bioreactors with different CHOK1SV GS-KO® clones (Webster et al. 2018, 2021). The secretion of lactate in rapidly proliferating mammalian cells, despite an abundance of oxygen, is referred to as the Warburg effect (Warburg 1956). Lactate is produced as a product of glycolysis by the hydrogenation of pyruvate through lactate dehydrogenase (LDH) and is subsequently co-secreted together with a proton via the monocarboxylate transporter (MCT) (Hartley et al. 2018; Luginsland et al. 2024). Consequently, lactate accumulates during the exponential growth phase of a standard fed-batch process, preceding a switch in cell metabolism. Following this transition, cells begin to consume extracellular lactate. Accordingly, lactate production and consumption must be considered when calculating the glucose consumed to ensure a further correlation with oxygen. Figure 5B illustrates the time-dependent profiles of lactate concentrations across the various fed-batch processes. As expected, extracellular lactate was produced and accumulated during the exponential growth phase. With the exception of a few cases, the maximum lactate concentrations (1.0–4.2 g/L) were reached after 78–140 h (approximately 3–5 days). These observations were consistent with the findings of previous studies on CHO fed-batch processes, as referenced in the literature (Dean and Reddy 2013; Webster et al. 2018; Luginsland et al. 2024). In consideration of the glucose and lactate concentrations, the lactate-corrected Gluc[cumulative] concentrations during the fed-batch processes were calculated in accordance with the methodology outlined before. The calculated Gluc[cumulative] concentrations for the various fed-batch processes and bioreactors ranged from 36.0 to 61.1 g/L, and the Gluc[cumulative] concentrations for the same cell lines exhibited substantial consistency across the different process scales. Time-dependent profiles of glucose (A) and lactate (B) concentrations during the investigated fed-batch processes. The data are grouped according to the respective project number. For glucose profiles, the dashed lines (–) indicate the lower and upper residual glucose limits, while the dash-dotted line (-.-) indicates the target residual glucose concentration The relationship between O2[cumulative] and Gluc[cumulative] Linear regression analyses were performed for each fed-batch cultivation based on cumulative oxygen consumption and lactate-adjusted cumulative glucose utilization. Figure 6 illustrates the resulting correlations for the projects investigated and fed-batch runs. In all cases, strong linear relationships were obtained between cumulative oxygen consumption and cumulative glucose utilization (R2 ≥ 0.984). Comparable ratios of ΔGluc[cumulative]/ΔO2[cumulative] were obtained for the same cell lines (d¯ΔGluc[cumulative]/ΔO2[cumulative] ~ 0.5–7.6%). The evaluation performed on a project-by-project and cell line basis also demonstrated a robust linearity (R2 ≥ 0.985) across the different process scales (data not shown). The ΔGluc[cumulative]/ΔO2[cumulative] ratios for the various fed-batch processes were found to fall within the range of 1.19 to 1.87 × 10–3. Comparable ratios were also determined by analyzing the data listed in Ihling et al. 2022 and Goldrick et al. 2018 for experiments conducted in shake flasks and a bench-scale bioreactor (Goldrick et al. 2018; Ihling et al. 2022). Goldrick et al. applied a comparable approach to that presented in this publication, but they used the cumulative OTR instead of the cumulative O2 and did not specifically consider the lactate metabolism. The exclusion of lactate metabolism from the calculation resulted in a substantial overestimation of glucose utilization for the datasets analyzed in this study (Supplementary Material, Figures S1–S2), thereby introducing significant deviations between the calculated and experimentally observed glucose consumption. Therefore, the use of cumulative O2 instead of cumulative OTR, coupled with the incorporation of lactate metabolism, was essential to achieve robust correlations. Thus, the present study demonstrated that incorporation of lactate-adjusted glucose utilization substantially improved the robustness of the oxygen-glucose correlation during metabolic transition phases characterized by lactate production and subsequent lactate re-consumption. In contrast to previously reported OUR/OTR-based approaches primarily evaluated at laboratory scale, the present analysis demonstrates robust oxygen-glucose correlations across multiple industrially relevant CHO fed-batch scales ranging from 10 L to 2 kL. To evaluate predictive capability and precision of the various correlations, the Gluc[cumulative] values derived from O2[cumulative] were calculated and compared with the Gluc[cumulative] values obtained from the experimental data (Fig. 7). It is evident that in all cases, the cumulative glucose concentration could be accurately predicted on the basis of the cumulative oxygen concentration (R2 = 0.987–0.998). The mean difference across all data points was found to be ± 0.6%, with root mean squared errors ranging between 0.91 and 3.76 g/L. Correlation between cumulative oxygen consumption (O2[cumulative]) and lactate-adjusted cumulative glucose utilization (Gluc[cumulative]) for the investigated fed-batch processes. The data are grouped according to the respective project and batch number. (•) calculated Gluc[cumulative] values. (–) linear regression line Comparison between lactate-adjusted cumulative glucose utilization determined from experimental process data and cumulative glucose utilization predicted from cumulative oxygen consumption (O2[cumulative]). The data are grouped according to the respective project and batch number Evaluation of predictive glucose demand estimation in 200 L proof-of-concept cultivations The predictive capability of the proposed framework was further evaluated in three 200 L fed-batch cultivations. Importantly, these cultivations were not operated using active closed-loop APC control. Instead, the framework was evaluated prospectively by predicting glucose consumption over sequential 24 h intervals using cumulative oxygen consumption and previously established oxygen-glucose correlations. As illustrated in Fig. 8A, the predicted glucose utilization values showed close agreement with experimentally determined glucose consumption throughout the fed-batch duration. Only minimal, and in most cases no, readjustment of the oxygen-glucose correlation was necessary following incorporation of the daily offline glucose and lactate measurements, as evidenced by the strong linearity maintained throughout the entire fed-batch processes. Stable predictive performance was maintained throughout the cultivation, including during the transition from lactate production to lactate re-consumption. The proof-of-concept evaluation therefore demonstrated that cumulative oxygen consumption can provide a robust descriptor of overall metabolic activity and glucose demand throughout industrial CHO fed-batch cultivation. Building on the obtained correlations, a flexible future concept for automated glucose feeding was developed with the intent of supporting multiple projects and bioreactor scales while remaining feasible for GMP applications. The proposed approach uses daily offline glucose and lactate data (which are routinely collected in a manufacturing environment) together with online process parameters from the previous day to continuously update the relationship between O2[cumulative] and glucose utilization. As shown in Fig. 8B, this correlation is then used to calculate the amount of glucose required over a defined interval (e.g., Δ1 h). The calculated requirement can be applied repeatedly until new offline measurements become available, at which point the correlation can be readjusted if needed. This structure allows the model to reflect the current metabolic state of the culture while relying exclusively on information that is already routinely available in SU bioreactor systems. A simplified representation of how this model could interface with an automated feed pump is provided in Fig. 8C. In a manufacturing environment, the frequency of updating the OTR calculation and applying new feed commands would need to be selected carefully. Very frequent updates may lead to unnecessary adjustments driven by short-term fluctuations, whereas longer update intervals may delay the response to metabolic changes such as shifts in lactate metabolism. Nevertheless, the underlaying model structure is compatible with a range of feedback frequencies and can be adapted to different operational and scale-dependent requirements. Results of the PoC study evaluating the predictive capability of the proposed oxygen-based soft-sensing framework and its potential applicability for future automated glucose feeding concepts. A Predictive estimation of cumulative glucose utilization during the 200 L PoC cultivations based on the Gluc[cumulative]/O2[cumulative] correlation. Glucose consumption was prospectively estimated over sequential 24 h intervals using cumulative oxygen consumption. Following daily offline sampling, the Gluc[cumulative]/O2[cumulative] correlation was updated if necessary and subsequently applied for prediction of the next 24 h interval. B Schematic representation of the proposed oxygen-based soft-sensing framework for predictive glucose demand estimation. C Simplified conceptual automation scheme illustrating potential future integration of the framework into an automated glucose feeding strategy Discussion The results of this study demonstrate that cumulative oxygen consumption provides a robust and scalable basis for describing glucose utilization in industrial CHO fed-batch cultures when lactate metabolism is incorporated into the analysis. Across multiple cell lines, projects, and reactor scales ranging from 10 L to 2 kL, strong linear correlations were obtained between cumulative oxygen consumption and lactate-adjusted cumulative glucose utilization. These findings indicate that cumulative oxygen consumption captures key aspects of the overall metabolic activity governing substrate utilization during fed-batch cultivation. Oxygen-based estimation of metabolic substrate demand is not a fundamentally new concept. OUR-based feeding strategies were already investigated for hybridoma and mammalian cell culture processes during the 1990 s, and more recently Goldrick et al. demonstrated online glucose estimation using OTR measurements in CHO cultures (Goldrick et al. 2018). However, practical implementation of such approaches has remained limited, partly due to the engineering complexity associated with bioreactor-specific characterization, determination of kLa correlations, dynamic gas composition effects, and process-specific calibration requirements. The present study therefore does not propose a fundamentally new APC strategy, but rather extends previously reported oxygen-based estimation approaches by incorporating lactate-adjusted glucose utilization and evaluating the resulting correlations across industrially relevant CHO fed-batch scales. One of the central findings of the present work was the importance of incorporating lactate metabolism into oxygen-based glucose estimation. During phases of lactate production and subsequent lactate re-consumption, omission of lactate metabolism resulted in systematic deviations between glucose utilization and oxygen consumption. Incorporation of lactate-adjusted glucose calculations substantially improved the robustness and linearity of the oxygen-glucose correlations, particularly during metabolic transition phases. At the same time, the implemented lactate correction should be interpreted as a pragmatic process-level approximation rather than a mechanistic representation of central carbon metabolism. Lactate metabolism in CHO cells is influenced by multiple factors including metabolic adaptation, mitochondrial activity, amino acid metabolism, pH, and temperature shifts (Hartley et al. 2018; Luginsland et al. 2024), which were not explicitly modeled in the present study. The proof-of-concept evaluation at 200 L scale further demonstrated that cumulative oxygen consumption can support predictive estimation of glucose demand over sequential cultivation intervals when combined with routine offline glucose and lactate measurements. Importantly, the presented framework has not yet been implemented as a fully autonomous closed-loop glucose control strategy. Instead, the study evaluated the predictive capability of oxygen-based glucose estimation using standard online process data and routine offline metabolite measurements. Consequently, the proposed framework should currently be considered a hybrid soft-sensing approach rather than a fully implemented APC strategy. Compared to dedicated glucose PAT solutions such as Raman spectroscopy or enzymatic glucose sensors, the presented framework offers several practical advantages. Since the approach relies primarily on routinely available process parameters and reactor characterization data, it can potentially be implemented without additional glucose-specific PAT hardware or extensive retrofitting of existing GMP manufacturing infrastructure. Furthermore, the observed robustness of the oxygen-glucose correlations across multiple reactor scales suggests that cumulative oxygen consumption may provide a scalable descriptor of metabolic substrate demand in industrial CHO fed-batch processes. At the same time, the present study also highlights several practical limitations associated with oxygen-based glucose estimation approaches. Accurate implementation requires prior bioreactor-specific engineering characterization, including determination of kLa correlations and estimation of P/V under representative process conditions. Dynamic gas composition effects and process-specific hydrodynamics additionally contribute to implementation complexity. Furthermore, the current framework still depends on routine offline glucose and lactate measurements for recalibration of the oxygen-glucose relationship. Consequently, the approach does not eliminate the need for offline analytics, even though stable linear correlations were already observed from day 1 onwards. However, the approach should currently be viewed as a predictive support framework rather than a replacement for established glucose monitoring strategies. Overall, the results demonstrate that cumulative oxygen consumption combined with lactate-adjusted glucose utilization provides a coherent and reproducible basis for predictive glucose demand estimation in industrial CHO fed-batch cultures. While additional validation under different process conditions, feeding strategies, and cell culture platforms would further strengthen general applicability, the presented framework already demonstrates the potential of oxygen-based soft-sensing concepts to support more standardized and scalable glucose management strategies in industrial mammalian cell culture processes. Supplementary Information Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 819 kb) Acknowledgements We would like to express our gratitude to Stefan Tauber for his efforts in collecting the process data from the 1kL and 2kL single-use production bioreactors. Furthermore, we would like to express our gratitude to the entire USP Pilot team in Visp for their invaluable contribution in evaluating the predictability of the developed approach during three fed-batch processes at 200L scale. Latin symbols [g/L], Glucose feed concentration [mg/L], Oxygen concentration [mg/L], Oxygen concentration at maximum saturation [-], Empirical constant C of Van’t Riet correlation [m], Impeller diameter [106 cells/mL], Mean difference in viable cell density [mg/L/h], Time derivative of oxygen concentration [m], Bioreactor vessel diameter [-], Geometrical ratio between impeller diameter and bioreactor vessel diameter [g/mg O2], Ratio between Gluc[cumulative] and O2[cumulative] [L/h], Volumetric glucose feed rate [g/L], Glucose concentration [g], Consumed amount of glucose for process time [t] [g], Cumulative amount of glucose consumed [m], Bioreactor vessel height [-], Geometrical ratio between bioreactor vessel height and diameter [m], Maximum liquid height [-], Geometrical ratio between maximum liquid height and bioreactor vessel diameter [1/h], Volumetric mass transfer coefficient [g/L], Lactate concentration [kg], Bioreactor weight [-], Newton number (= Power number) [mg/L/h], Oxygen transfer rate [mg/L/h], Oxygen uptake rate [W/m3], Volume specific power input [mol/cell/h], Cell specific oxygen consumption rate [h], Process time [m/s], Superficial gas velocity [m3], Total bioreactor vessel volume [m3], Minimum bioreactor working volume [m3], Maximum bioreactor working volume [-], Lactate yield per glucose equivalent Greek symbols [-], Exponent of Van’t Riet correlation [-], Exponent of Van’t Riet correlation [g/L], Differential concentration of glucose (ΔGluc) or lactate (ΔLac) Indices [-], Time at process time [t] [-], Time at process time [t-1] Author contributions Conceptualization, V.J., A.J. and G.M.; methodology, V.J., A.J. and G.M.; data analysis, V.J. and A.J.; validation, V.J.; formal analysis, S.L. and G.B.; writing – original draft preparation, V.J. and A.J.; writing – review and editing, G.M., S.L. and G.B.; visualization, V.J.; project administration, V.J. and A.J.; All authors have read and agreed to the published version of the manuscript. Funding This research did not receive any specific grant from funding agencies in the public or not-for-profit sectors. Data availability The data that support the findings of this study are not openly available due to reasons of sensitivity and are available from the corresponding author upon reasonable request. Declarations Ethical approval This article does not contain any studies with human participants or animals performed by any of the authors. Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary file1 (DOCX 819 kb) Data Availability Statement The data that support the findings of this study are not openly available due to reasons of sensitivity and are available from the corresponding author upon reasonable request.