Real‐time monitoring of biomass during Escherichia coli high‐cell‐density cultivations by in‐line photon density wave spectroscopy

An efficient monitoring and control strategy is the basis for a reliable production process. Conventional optical density (OD) measurements involve superpositions of light absorption and scattering, and the results are only given in arbitrary units. In contrast, photon density wave (PDW) spectroscopy is a dilution‐free method that allows independent quantification of both effects with defined units. For the first time, PDW spectroscopy was evaluated as a novel optical process analytical technology tool for real‐time monitoring of biomass formation in Escherichia coli high‐cell‐density fed‐batch cultivations. Inline PDW measurements were compared to a commercially available inline turbidity probe and with offline measurements of OD and cell dry weight (CDW). An accurate correlation of the reduced PDW scattering coefficient µs′ with CDW was observed in the range of 5–69 g L−1 (R2 = 0.98). The growth rates calculated based on µs′ were comparable to the rates determined with all reference methods. Furthermore, quantification of the reduced PDW scattering coefficient µs′ as a function of the absorption coefficient µa allowed direct detection of unintended process trends caused by overfeeding and subsequent acetate accumulation. Inline PDW spectroscopy can contribute to more robust bioprocess monitoring and consequently improved process performance.


| INTRODUCTION
Biomass concentration is a parameter of high interest during biotechnological cultivation processes. Its growth rate µ strongly influences product quality and quantity (Schuler & Marison, 2012).
For example, µ is a critical indication for the formation of growth-inhibiting secondary metabolites such as acetate in Escherichia coli (Åkesson et al., 2001) or ethanol in Saccharomyces cerevisiae (Sonnleitner & Käppeli, 1985). To monitor and control growth rates there is a specific need for inline biomass sensors that feature high reliability and robustness (Mears et al., 2017) as well as the ability to be calibrated precisely over wide concentration ranges with linear responses (Olsson & Nielsen, 1997). Possible offline methods that can be used for calibration are cell dry weight (CDW) and optical density (OD) measurements. Mauerhofer et al. (2018) concluded that OD measurement values of the same cell suspension have a high dependency on the optical design of the employed spectrometer and can vary between different spectrometers.
The process application of real-time and in situ measurement techniques, in general, is a mandatory step toward quality by design (Rathore & Winkle, 2009;Yu et al., 2014), as well as process integration (Gomes et al., 2014) and is part of the process analytical technology (PAT) directive issued by the United States Department of Health and Human Services, Food and Drug Administration (2004).
The performance of commercially available inline biomass sensors such as dielectric spectroscopy (Kaiser et al., 2008), infrared spectroscopy, fluorescence spectroscopy, and OD measurements have been reviewed and compared in literature (Biechele et al., 2015;Kiviharju et al., 2008) stating the need for modern sensor systems that provide relevant information about all stages of the bioprocess. A combination of capacitance and turbidity data was proposed to enhance real-time control strategies (Habegger et al., 2018); however, OD sensors based on transmission, side-and backscattering show nonlinear concentration effects (Vojinović et al., 2006) and are prone to probe fouling in highly turbid systems (Münzberg et al., 2016). Marquard et al. (2017) reported similar saturation effects for in situ microscopy measurements at high cell densities of E. coli.
Due to advantages in volumetric productivity, continuous manufacturing has become a focus of recent research (Zinn et al., 2021) and there is a specific need for novel monitoring tools (Kopp et al., 2019). As such a new PAT tool, photon density wave (PDW) spectroscopy allows for the independent inline determination of scattering and absorption properties of highly turbid dispersions (Bressel et al., 2013) and was successfully used to monitor biotechnological processes like the enzymatic casein hydrolysis in milk (Hass et al., 2015). Gutschmann et al. recently applied this technology to stirred tank reactor (STR) cultivations of Ralstonia eutropha and yielded intracellular polyhydroxyalkanoate formation rates for the first time (Gutschmann et al., 2019;Gutschmann, Maldonado Simões, et al., 2023;. Sandmann et al. (2022) employed PDW spectroscopy to successfully measure the formation of algae biomass in a photobioreactor.
The working principle of PDW spectroscopy is based on intensity-modulating laser light of discrete wavelengths between approximately 400 and 1000 nm. Light is guided into the system under investigation by a multimode optical emission fiber. Due to multiple light scattering in turbid materials, a PDW is created and can be described based on radiation transport theory. Secondary optical fibers at different distances transport light back to an avalanche photodiode. Then, a vector network analyzer is utilized to measure changes in modulation amplitude and phase. This information is used in a weighted nonlinear global analysis to calculate the optical coefficients for absorption µ a and scattering µ s ′.
The aim of this study is to evaluate the performance of PDW spectroscopy as a PAT tool for precise inline biomass monitoring in highly turbid bacterial cultivations. E. coli serves as a model organism due to its broad employment in industrial processes. Three high-celldensity cultivations were conducted to determine the crossexperimental robustness and accuracy of PDW spectroscopy to establish OD measurements both inline (Dencytee sensor) and offline (Photometer and Cedex Bio HT).

| Bacterial strain
All cultivations were performed using the E. coli wild-type strain W3110 (DSM-5911; Leibniz Institute DSMZ-German Collection of Microorganisms and Cell Cultures).

| Growth media and preculture conditions
The W3110 strain was dilution smeared on lysogeny broth (LB) (LB-Medium, Carl Roth GmbH + Co. KG) plates from glycerol cryostocks and incubated overnight at 37°C. First, liquid precultures were picked from single colonies and grown for 7 h in 20 mL standard LB medium (10 g L −1 peptone, 5 g L −1 yeast extract, 5 g L −1 NaCl).
The device was built at the University of Potsdam and supplied to the TU Berlin. The 25-mm-diameter stainless-steel probe was inserted into the reactor through the lid via a DN25 Ingold port and autoclaved in situ. The optical coefficients, the reduced scattering coefficient µ s ′ and the absorption coefficient µ a , were determined at a wavelength of λ = 638 nm with a moving set of 10 repeating measurements for one set of optical coefficients. This resulted in a temporal measurement resolution of 0.8 min −1 . For raw data analysis, a refractive index of 1.388 for the suspension was used. The refractive index was determined by a laser refractometer (DSR-λ; Schmidt + Haensch GmbH & Co.).

| Total cell density (TCD)
A Dencytee sensor (unit DN12-120; Hamilton Bonaduz AG) was used to monitor the biomass concentration. The sensor utilizes transmission-based absorbance measurements at 880 nm with an optical path length of 5 mm to automatically correlate the results to the TCD.

| CDW
Each CDW value was determined by centrifugation of 2 × 2 mL sample aliquots (4°C, 6500g, 10 min) and drying at 80°C until constant weights were achieved. The average of the measured values was taken as a result.

| OD
OD 583 values were measured by two different systems at the wavelength λ = 583 nm each. For measurements using a UV/VIS photometer (Ultrospec 2100 Pro UV/Visible Spectrophotometer; Biochrom Ltd.) 0.9% (w v −1 ) aqueous NaCl solution (saline) was both used as a blank reference and to dilute the samples to achieve OD 583 values ≤ 0.5. Measurements were conducted in 1.6 mL semi-micro polystyrene cuvettes (Carl Roth GmbH + Co. KG) with an optical pathlength d = 10 mm. For analysis of OD 583 via a pipetting robot (Cedex Bio HT Analyzer; Roche Diagnostics AG), the test kit OD Bio HT (PMMA acryl cuvettes, d = 5 mm) was used.

| Media components
Glucose and acetate concentrations were determined with the Cedex Bio HT Analyzer using the test kits Glucose Bio HT and Acetate V2 Bio HT. cients µ s ′ and µ a measured inline by PDW spectroscopy were obtained before the end of the batch phases from 6 h (C1) and 7 h (C2 and C3) onward which corresponded to a lower limit of detection of 5 g L −1 CDW. Before reaching the limit of detection, biomass could already be tracked by the raw light intensity detected by PDW spectroscopy without processing the signal into the optical coefficients (Supporting Information: Figure S4). In C1 and C2, the reduced scattering coefficient µ s ′ increased in correlation with the rising CDW up to the final concentrations. All correlation values are stated in Table 1. During C3, the correlation between CDW and µ s ′ ended at 19.6 h and 68.9 g L −1 , although the measured CDW increased to a final concentration of 86.1 g L −1 . A general increase of the absorption coefficient µ a was observed over the course of all experiments, but no direct correlation of the signal with CDW was found. As an example, during C1 µ a increased until 17.   showed good correlations with both OD 583 -based results (µ Photometer , µ Cedex ). At the beginning of the fed-batch phases, µ Dencytee displayed a systematic underprediction of the growth rate in all cultivations.

| Scattering properties as a function of absorption properties
The simultaneous and independent determination of scattering and absorption properties by PDW spectroscopy provides additional process information by discussing µ s ′ as a function of µ a , as shown in The first decrease in slope was observed after 10.6 h, while the first increase in acetate concentration was detected between 10 and 11.8 h (Figure 1). The second change of slope in µ s ′/µ a for C1 was F I G U R E 2 Biomass growth rates µ of cultivation 1-3 based on all applied measurement systems. Growth rates calculated from inline measurements: reduced scattering coefficient µ s ′ (µ PDW , red circle), total cell density TCD 880 (µ Dencytee , gray square). Growth rates calculated from offline measurements: cell dry weight (µ CDW , green circle), optical density OD 583 (µ Cedex , orange upward triangle and µ photometer , blue downward triangle). PDW, photon density wave.  Table 1.

| DISCUSSION
The aim of the conducted experiments was to evaluate the potential of PDW spectroscopy as an inline sensor for biomass measurements during E. coli high-cell-density cultivations and to achieve a thorough comparison to established optical in-and offline methods. The results obtained during three different 3.7 L STR fed-batch cultivations ( Figure 1) showed that the reduced scattering coefficient µ s ′ measured by PDW spectroscopy correlated with the CDW in a concentration range of 5-69 g L −1 (R 2 = 0.98, PBE ± 3.2 g L −1 ) F I G U R E 3 Reduced scattering coefficient µ s ′ by PDW spectroscopy measured during cultivation 1-3 as a function of respective absorption coefficient µ a , both measured at 638 nm. Values greater µ s ′= 1.1 mm −1 and µ a = 0.03 mm −1 are displayed in Supporting Information: Figure S5. Cultivation 1 (C1, red square), cultivation 2 (blue circle), and cultivation 3 (green triangle). Regarding the offline methods, the Cedex Bio HT Analyzer achieved the highest overall accuracy (R 2 = 0.99, PBE ± 2.2 g L −1 ), compared to R 2 = 0.97 and PBE ± 4.1 g L −1 achieved by using a benchtop UV/VIS photometer. The higher accuracy of OD 583 measurements conducted via the Cedex Bio HT Analyzer compared to UV/VIS photometer-based measurements can be explained by the pipette-based dilution of samples for the UV/VIS photometer measurements, necessary due to the deviation from the linear Lambert-Beer behavior above OD = 0.8 for E. coli (Myers et al., 2013).
The correlation between µ s ′ and CDW values below 5 g L −1 was not considered, since light scattering in that regime is too low (µ s ′ < 0.1 mm −1 at 638 nm) for quantitative data analysis in PDW spectroscopy. This effect is wavelength dependent and material specific, thus here related to E. coli cultivations. This limitation can be explained by the measurement principle that requires a sufficient amount of multiple light scattering and more pronounced light scattering than light absorption (µ s ′ >> µ a ) (Bressel et al., 2013;Fishkin et al., 1996). Given the low cell density during the early batch phase, absorption effects are predominant. Thus, quantitative PDW spectroscopy theory is not applicable in that regime. However, also in the early stage of cultivation, changes in the raw data of PDW spectroscopy are related to biomass growth. Here, a cell growthrelated process trend can be generated in real time from the changes in the raw light intensity values measured by the PDW spectrometer (Supporting Information: Figure S4). It can be used for relative process monitoring before a sufficient concentration of cells to cause multiple light scattering is reached for quantitative PDW spectroscopy data analysis.
The third cultivation resulted in a final yield of 86.1 g L −1 CDW, but none of the investigated optical measurement technologies showed a correlation with the CDW exceeding 76 g L −1 (Figure 1;   Figure 4). This can possibly be explained by cell lysis or significant morphological changes of the cells during the terminal phase of the cultivation that influence the optical properties, as has been shown to influence light scattering in the case of lysis of erythrocytes (Valenzeno & Trank, 1985). This hypothesis is supported by the decrease of the reduced scattering coefficient after 23 h while the absorption coefficient kept increasing until the end of the process.
While an exact biological twin was not achieved by conducting two identical cultivations the reproducibility of PDW results has been proven by previous studies (Gutschmann et al., 2019;Werner et al., 2017).
The growth rates calculated based on inline PDW spectroscopy were in excellent agreement with the growth rates calculated based on the different offline methods (Figure 2). In contrast, the inline TCD values led to a good agreement with the reference methods during the batch phase but displayed a systematic underestimation of µ during the first 4-8 h of the feeding phase in all experiments. The ability to measure and therefore control growth rates in real-time is a valuable tool in process control as it not only determines cells' lipid composition and protein leakage (Arneborg et al., 1993;Shokri et al., 2002) but in addition has a direct influence on product yield because various recombinant protein products have a maximum production rate at specific growth rates (Peebo & Neubauer, 2018).
Assessing the changes in absorption and scattering properties as a function of each other allows for a time-independent discussion of process trends (Figure 3). This holds great value for comparing different experiments and can potentially be used to detect process disturbances such as the observed acetate inhibition during C1 in this study. It may function as a time-independent approach for ensuring batch quality, providing a tool to define a "golden batch" trajectory.
While all three cultivations had a similar relation in the correlation of µ s ′/µ a at first, there was a clear change of slope for the trajectory of C1 at µ s ′ = 0.32 mm −1 and µ a = 0.006 mm −1 compared to the other cultivations. This can be attributed to an overcalculation of the exponential feed rate during C1 and the following overflow-related production of acetate, reaching growthinhibiting concentrations of 3.6 g L −1 between 10 and 11.8 h ( Figure 1). The change of slope relates to a processing time from 10.6 h onward and is likely caused by reduced growth rates. The corresponding growth rates calculated from CDW samples support this theory by decreasing from µ = 0.39 h −1 at 10 h to µ = 0.26 h −1 at 11.8 h and reaching 0.11 h −1 at 13 h (Figure 2), directly reducing the increase of µ s ′. The reduced scattering coefficient might further be influenced by changes in the cell shape or surface, induced by lysis (Haidinger et al., 2003) or aging. The possible effect of these processes on the observed increase of µ a need further investigation as it might generate a better quantitative understanding of the process. This approach and the additional insight it generates might be used as a control and monitoring parameter in turbidostat (Gresham & Dunham, 2014) and chemostat (Kittler et al., 2020) cultivations for continuous manufacturing.
PDW spectroscopy holds the distinct advantage that it is a direct method. Therefore, it does not rely on chemometric modeling like performed for fluorescence spectroscopy (Ödman et al., 2009).
Neither is there a need for advanced calibration accounting for suspended solids, ethanol concentration, or the conductivity of the medium that dielectric spectroscopy requires (Lopez et al., 2019).
To summarize, PDW spectroscopy allows for accurate predictions of high concentrations of biomass and its growth rate, based on the reduced scattering coefficient µ s ′, obtained inline and under process conditions. Due to the high accuracy across different experiments indicated by the low predicted biomass error and its ability to detect process disturbances, PDW spectroscopy should be further investigated as an analytical technology for process monitoring, development, and control.

| CONCLUSION
The conducted study finds that PDW spectroscopy can be applied as μ is the specific growth rate (h −1 ), μ max is the maximal growth rate during batch phase (h −1 ), μ set is the specific growth rate during exponential fed-batch phase (h −1 ), n is the factor for specific growth rate: 0.7 (cultivation 1) or 0.6 (cultivations 2 and 3), F 0 is the initial feed rate (L h −1 ),Y x/s is the biomass/glucose yield (g g −1 ), S i is the glucose concentration in feed (g L −1 ), X 0 is the biomass concentration at the end of the batch phase (g L −1 ), and V 0 is the volume at the end of the batch phase (L).