The cell growth quantifier (CGQ)
The CGQ has been developed to allow accurate and parallelisable monitoring of biomass in shake flasks in a fully automated and non-invasive manner. Special focus during the development was put on the measurement range with the goal of building a system able to monitor the complete bioprocess, from inoculation (OD600 < 0.5) to harvesting (OD600 > 50). Currently there are three other devices available, which can serve as a technical benchmark for the CGQ system, namely the SFR Vario (Presens GmbH, Germany), the OD-Monitor (TAITEC, Japan) and the OD-Scanner (BugLab LLC, USA).
The CGQ implements the well-known technique of light scattering with light source and sensor mounted directly into the spring clamp under the shake flask (Fig. 1a). Depending on the current biomass concentration inside the flask, different amounts of light are scattered towards the sensor (Fig. 1b). Backscattered light measurements are known to exhibit a wide dynamic range and are also the basis for SFR Vario and OD-Scanner. In contrast, the OD-Monitor performs transmission measurements, which strongly reduces the upper detection limit to OD600 < 10 and which requires relatively high and thus in case of oxygen limited processes unfavourable filling volumes of about 20 % [43].
In addition to the dynamic range requirements of typical bioprocesses, fully automated measurements under continuous shaking are favourable to avoid effects like sedimentation, poor aeration and mixing, anaerobic metabolic stress and to reduce the scientist’s manual workload. Similar to the CGQ, the SFR Vario and the OD-Monitor allow for such measurements, while the OD-Scanner requires the flask to be taken off the shaker and to be hold still for the measurement period by the operator.
Automated biomass monitoring in shake flasks requires some kind of handling the highly dynamic liquid distribution in a continuously shaken environment. Current approaches turn the dynamic nature of shaken bioreactors into a static measurement setup. The simplest method is to stop the shaking movement as required for the OD-Scanner operation. Other methods include accelerometric synchronisation in combination with a median filter (SFR Vario) [8] and a light path choice for minimal fluid fluctuations in combination with averaging (OD-Monitor) [44].
In contrast to the aforementioned static measurement methods, the CGQ for the first time implements a dynamic approach of biomass monitoring in a continuously shaken system. For this purpose it collects millions of single backscatter readings at a dynamic measurement frequency >500 kHz, which together form periodic raw signal series (simplified shown in Fig. 1c). As the actual biomass concentration can be assumed to be constant during the few seconds of raw data acquisition, the periodic raw data series are essentially a function of the liquid height above the sensor. The CGQ collects a complete “image” of the dynamic liquid distribution over several subsequent shaking movement periods instead of simplifying the fluid dynamics to one single static measurement. From the resulting huge amount of data at a macroscopic fermentation time point, robust backscattering readings are then calculated for almost each kind of liquid distribution observable in typical shake flask experiments, even at highest shaking speeds and extremely low filling levels (i.e. 350 rpm at <5 % filling volume for optimal oxygen transfer rates). Furthermore, the data-driven CGQ technology provides real-time biomass data at high resolution and smoothness, without the need for smoothing and filtering (as described in [8]), where subsequent measurement delays of 5–10 min are introduced, which can cause significant systematic biomass underestimation errors.
In order to avoid the interference of external light sources with the highly sensitive CGQ measurements and to allow for accurate biomass monitoring at typical low inoculation cell densities, each monitored flask is darkened with a light shielding cover as shown in Fig. 1d.
One of the most favourable shake flask properties is their parallelisability. The currently available devices do not support such experimental setups and can be operated in parallel with only four flasks (SFR Vario, OD-Monitor) or even only one flasks (OD-Scanner) at the same time. The CGQ allows for fully parallelised biomass measurements in up to 8 or 16 flasks, all connected to one base station (see Fig. 1d), so that each flask on a typical shaker tray is monitored. Additionally the CGQ requires not much more space on the tray than the standard spring clamp, which furthermore contributes to its applicability in highly parallelised shake flask experiments.
Out of all currently existing solutions for non-invasive biomass monitoring in shake flasks, the CGQ system was identified as the optimal choice for an efficient triplicate based parallel evaluation of four different S. cerevisiae strains grown on various carbon sources due to its comparably small size, high parallelisability and dynamic range, real-time data availability and smoothness and its applicability under almost any cultivation condition. With the bundled CGQuant software (see screenshot in Fig. 1d), all data analysis tasks, including the calibration, the comparison, annotation and documentation of data sets and the calculation of growth rates (see “Methods” section) could be accomplished intuitively and satisfactorily.
Evaluation of online biomass data in comparison to offline data for CEN.PK2-1C on SCD medium
In order to evaluate the general validity of CGQ-derived online biomass data in the context of typically collected offline parameters, CEN.PK2-1C was grown on synthetic complete medium with d-glucose and samples were taken at various time points throughout the fermentation process. CGQ backscattered light intensities were converted into OD600 values by CGQuant using two different calibration files for CEN.PK2-1C as depicted in Fig. 5b. The samples were analysed by offline OD600 measurements, HPLC and microscopy. All these results are depicted in Fig. 2. Offline and online biomass data are in good agreement, thus demonstrating the general applicability of backscattered light measurements as an alternative for offline biomass determination. The well-known Crabtree-effect of S. cerevisiae being grown on glucose as carbon source can be clearly identified solely on the basis of the online biomass data. Here the initial rapid growth phase on glucose accompanied by ethanol formation took about 16 h and was followed by the typical metabolic shift to a much slower growth on the basis of respiratory ethanol metabolization. These observations are strongly supported by the HPLC-derived ethanol and glucose concentrations and are in good agreement with the data published by Anderlei et al. [45] for a comparable experimental setup.
Growth measurements of various yeast strains on different carbon sources
To demonstrate the properties of the CGQ technology, four different yeast strains, BY4741, W303, CEN.PK2-1C and Ethanol Red, were grown in synthetic complete (SC) medium with different carbon sources: d-glucose, d-galactose, maltose or ethanol. Therefore, pre-cultures were inoculated in 500 mL flasks, containing YEPD2% medium, until reaching OD600 of 1–2. Subsequently, cultures were washed and inoculated with a starting OD600 of 0.2 in 300 mL flasks, containing SC medium and the particular carbon source (“Cultivation conditions” section). Growth and apparent growth rate were followed by using the CGQ (Figs. 3, 4).
On glucose medium the fastest increase of scattered light intensity, the highest intensities as well as the highest growth rate could be observed for Ethanol Red (Fig. 3). Metabolization of galactose provided the highest growth rate for strain BY4741, while the highest intensities were reached by Ethanol Red. Due to the cultivation of the pre-cultures on glucose-containing medium, long lag-phases could be shown for all strains on galactose. Interestingly, W303-1A revealed the shortest lag-phase. On maltose medium (Fig. 4), Ethanol Red reached the highest intensities and growth rate, CEN.PK2-1C was slightly slower, but no growth could be observed for BY4741 as well as W303-1A. Whether the non-existing growth of both derivatives of S288C on maltose could be due to the progenitor’s inability for maltose utilization [16, 46] or due to other reasons cannot finally be concluded by these exemplarily intended experiments.
Another interesting observation which could be nicely demonstrated with the CGQ technology was the formation of double peaks within the apparent growth rates, as shown here for the growth rates of W303-1A on glucose and galactose (Fig. 3). Whether the monitored observations are primary derived from physiological (for instance alterations in glucose repression) or morphological alterations could be a subject of further studies.
On ethanol, only a very slow growth can be seen for Ethanol Red which probably is due to its selection for high ethanol production rates and titers [38] (Fig. 4). The highest intensities and growth rate were reached by CEN.PK2-1C. No growth could be detected for all four strains on synthetic glycerol medium (data not shown). Inconsistencies with the work of Swinnen et al. [47], where prototrophic versions of the laboratory strains CEN.PK, W303, and S288c exhibited growth in a different synthetic glycerol medium (according to [48]) after supplementation of nucleobases and amino acids, could be due to the different handling of pre-cultures, different pH-values, different composition of both media or different genotypic characters.
Additional observations
Ethanol or also small parts of fermentative carbon sources are metabolized by S. cerevisiae using respiratory pathways. Under the chosen shaking and aeration conditions, most aerobic cultivations can be expected to become oxygen limited as soon as a certain biomass concentration is reached. This has been shown within a comparable experimental setup [45] for S. cerevisiae DSM70449 on YEPD2% by online monitoring of oxygen transfer rates in conjunction with offline glucose and ethanol concentrations. Indeed, those oxygen limitations are observable in real-time biomass data (as shown in [49]). With growth rates being limited by the maximal oxygen transfer rate, the biomass concentration becomes a linear function of the fermentation time. As clearly depicted in Figs. 2 and 3 for growth on glucose as primary carbon source, a strictly linear biomass increase can be observed for the respiratory ethanol metabolization phase after the diauxic shift that follows the consumption of glucose.
Another interesting observation is the strain specific decrease in the scattered light intensity at the transition from sugar fermentation to respiratory ethanol consumption within the diauxic shift (Figs. 2, 3). This shift seems to cause morphological or cell size related changes that influence the scattering intensity to various degrees depending on the investigated strain (see also Fig. 5). While the growth curve of Ethanol Red exhibits a relatively fast and small signal decrease, scattered light intensities of BY4741 are decreasing much stronger and over a longer period of time. CEN.PK2-1C and W303-1A exhibit only very slight backscattering changes with minimal signal decreases over longer time periods (>6 h).
Correlation of OD600 and scattering signal in shake flasks
In order to determine absolute biomass concentrations from scattering signals it is required to correlate offline biomass concentration measurements and the obtained scattering intensities. This correlation was investigated with different S. cerevisiae strains for exponentially growing cells as well as diluting stationary cell suspensions to prove the general applicability of CGQ’s scattering based biomass estimation in shake flasks and to identify its limitations.
Data collected via the dilution series approach (see “Correlation experiments” section) are shown in Fig. 5a for the three S. cerevisiae strains CEN.PK2-1C, W303-1A and Ethanol Red. The curves exhibit a strictly monotonic nonlinear correlation, which can be accurately described by the mixed exponential-sigmoid-linear model function (“Calculation basics—correlation backscatter vs. OD600”, “Calculation basics—growth experiments” sections). Even though all investigated strains belong to the same species, considerable strain-to-strain differences were observed for the correlation data. This might be attributed to differences in cell wall and cell interior composition as well as to cell size and morphology. Interestingly, W303-1A and Ethanol Red curves were highly similar up to an OD600 of 7 compared to the CEN.PK2-1C curve. W303-1A exhibited the lowest scattering. For W303, Kokina et al. [23] recently described changes in the correlation of cell count and OD600 due to increased cell swelling of aged cultures. As the cell from the presented correlation experiments originate from >72 h old cultures, the lower scattering of W303 cells could be explained by increased cell size and thus reduced backscattering [50].
Significant correlation differences were obtained between the dilution series approach and the growth approach (Fig. 5b). Within the range of OD600 = 1.5–3.5 the backscattered light intensity is much weaker for the exponentially growing cells than for the stationary cells. At an OD600 of about four the two correlation curves intersect. This behaviour can be explained by growth phase related changes of S. cerevisiae’s cell size. From Mie-theory of light scattering by homogeneous spheres it is known that backscattering intensities increase with decreasing particle volume, as shown by Latimer and Pyle [50] for various cell volumes and scattering angles. Hence it must be concluded from the correlation data sets that, within the range of OD600 = 1.5–3.5, the scattering particles from the growth approach are considerably larger than those from the dilution series approach, which is indeed caused by the mother-daughter-cell-aggregates from budding during the exponential growth, while the stationary phase as well as the slow growth on ethanol is dominated by single, non-aggregated and thus stronger backscattering cells, as clearly shown by the microscopic images in Fig. 2. Remaining differences at an OD600 > 4 could be due to fermentatively formed ethanol, which has been shown to induce significant changes to S. cerevisiae’s cell volume (size reductions >20 %) [51], as well as cell wall thickness, roughness and composition [52].
While morphologic, cell size, growth phase and even measurement technique related phenomena may introduce considerable variance into the direct correlation of backscattered light intensities and OD600 or other biomass estimates, these correlation data are still useful to calculate desired biomass values from backscatter readings as measured by the CGQ system. This is exemplarily shown in Fig. 2, where exponential growth phase (dark blue) and stationary growth phase (light blue) correlations were used to calculate OD600 from the CGQ’s backscatter readings and where offline data (yellow) were collected on a completely different UV–VIS-spectrometer. The good agreement between offline and online data could be reached by correlation data refinement within the CGQuant software. For this purpose, two additional offline OD600 (yellow squares with purple edges) values were entered into the software, the cell density after inoculation and after 31.5 h of fermentation.
Resolution and reproducibility of light scattering based biomass estimation in shake flasks
Under optimal conditions, the CGQ’s backscatter readings provided sufficient resolution for biomass estimation even below an OD600 of 0.2, as exemplarily depicted in Fig. 6a. Positive correlations of scattering signal and OD600 could be observed for all other data sets at the latest from OD600 = 0.4 onwards. Throughout all measurements during this study, a high reproducibility (Fig. 6b) was observed for both, correlation and growth experiments. The baseline signal (see Fig. 4 BY4741, W303 1A) was highly stable with a background noise of 1 % and a signal drift smaller than 0.02 % per hour with respect to the average baseline signal.
The limits of resolution and technical reproducibility during backscatter monitoring are mainly introduced by the comparability of real shake flasks and their positioning in the spring clamps. Shake flask bottoms may exhibit different thicknesses, inhomogeneities, scratches and unevenness, all influencing the optical path. Spring clamps may show various degrees of stability and deformation, which can influence the positioning of the flask and thus the liquid distribution above the sensor. Many of these effectors are only recognisable at very low biomass concentrations, but under rarely observed worst-case conditions lens-like scaling effects can be observed as exemplarily shown in Fig. 6b for CEN.PK2-1C(5). However, algorithms have been implemented into the CGQuant software allowing for the elimination or reduction of such external factors.