Monitoring Sugar Consumption and Ethanol Production During Yeast Fermentation Using a Raman Plate Reader and Multivariate Analysis

This column demonstrates the use of a Raman plate reader (RPR) combined with multivariate analysis (MVA) to monitor sugar consumption and ethanol production during yeast fermentation. Using partial least squares (PLS) regression models, glucose, fructose, and sucrose were simultaneously quantified in mixed solutions, and the effect of yeast-induced turbidity on model transferability was evaluated and corrected using turbidity-matched calibration standards. The workflow was then applied to time-course fermentation samples containing ternary and binary sugar mixtures, tracking substrate consumption—including rapid sucrose hydrolysis by yeast-secreted invertase and the preferential utilization of glucose over fructose—alongside ethanol production over the course of fermentation. These results demonstrate that combining automated, plate-based Raman acquisition with multivariate analysis provides a rapid, label-free, and high-throughput approach for monitoring multiple fermentation components simultaneously, with potential value for strain screening, media optimization, and process development.

Fermentation is a foundational bio-manufacturing technology used to produce a broad range of bio-derived products, including food ingredients, biopolymers, biofuels, specialty chemicals, pharmaceuticals, and other high-value metabolites. The performance of a fermentation process depends on multiple factors, including the microbe, the particular microbial strain, media composition, carbon source, nutrient availability, temperature, pH, aeration, and cultivation time. Therefore, process development commonly requires the evaluation of numerous combinations of biological and operating conditions to maximize product yield and productivity while minimizing unwanted byproducts and batch-to-batch variability. Raman spectroscopy can be used both for optimizing fermentation conditions and for monitoring in real time the progress of a reaction.

Background

Among the parameters used to evaluate fermentation performance, substrate consumption and product formation provide direct insight into microbial activity and process progression. Sugars frequently serve as the primary carbon and energy sources in microbial fermentation, and their rates of consumption can reflect differences in strain performance, substrate preference, metabolic state, and nutrient limitation. Tracking individual sugars is particularly important when complex media or mixed-carbon-source formulations are used, because microorganisms may consume available substrates at different rates or in a preferred sequence. In this column we will show how Raman spectroscopy can monitor ethanol production and identify suitable endpoints by Saccharomyces cerevisiae yeast, a process that has been known and used for thousands of years. Time-resolved measurement of changes in both substrate and product concentrations can support media optimization, strain screening, comparison of fermentation conditions, and early identification of stalled or atypical fermentation behavior.

Conventional analytical approaches, including high-performance liquid chromatography (HPLC), enzymatic assays, and spectrophotometric methods, remain essential for fermentation analysis. These techniques can provide sensitive and informative measurements, but each has practical limitations in high-throughput process development. Chromatographic methods, while providing comprehensive profiling, require intensive sample preparation, human labor, and separation time, resulting in a delay between sample collection and information availability. Enzymatic and colorimetric assays can provide faster screening but are often specific to a single analyte and may require multiple reagents, calibration procedures, and separate assays when several sugars or metabolites must be quantified. As the number of strains, formulations, and process conditions under investigation increases, the analytical workload can become a significant bottleneck in fermentation optimization.

Raman spectroscopy is an attractive complementary technique for fermentation analysis because it is label-free, requires minimal to no sample preparation, and can provide chemically specific information from complex liquid samples. A single Raman spectrum can contain contributions from multiple substrates, products, and metabolic byproducts, creating the potential to evaluate several components from one measurement. Because water produces a relatively weak Raman signal, the technique is particularly well suited to aqueous biological samples. Fermentation spectra are nevertheless complex and almost always contain overlapping bands, fluorescence backgrounds, high turbidity, and matrix-dependent variation that limit direct univariate analysis. When Raman spectroscopy is combined with multivariate analysis (MVA), spectral variation distributed across multiple Raman bands can be used to relate to reference concentration data. MVA methods such as partial least squares (PLS) regression can therefore support quantitative estimation of individual components even when their spectral features overlap.

However, conventional Raman measurements are generally performed on one sample at a time with manual sample loading, focusing, and acquisition. These manual workflows reduce their speed and practicality for studies involving numerous fermentation conditions or repeated sampling over time. A Raman plate reader (RPR) addresses this gap by combining automated Raman acquisition with a multi-well plate format. Samples collected throughout a fermentation experiment can be stored, transferred to a plate, and analyzed together in a standardized batch. This approach supports efficient retrospective analysis of fermentation trajectories while reducing manual handling and improving consistency across measurements.

In this study, we used the Raman plate reader combining with MVA to monitor sugar consumption and ethanol production during yeast fermentation. Aliquots collected at defined intervals were analyzed simultaneously following completion of the fermentation experiment. Our feasibility study evaluates the ability of a plate-based Raman workflow to quantify changes in multiple fermentation components and to track substrate utilization and product formation over time.

Sugar Identification and Quantification in Mixtures

Before jumping into fermentation samples, we first evaluated the feasibility of distinguishing and quantifying individual sugar molecules, specifically glucose, fructose, and sucrose, in mixed solutions. Stock solutions of each sugar were prepared at 200 mM in Milli-Q water. Individual calibration standards were prepared for each sugar at concentrations ranging from 20 to 160 mM. The spectra from all three sets of single-component standards were combined to develop a PLS2 calibration model with three response variables corresponding to the concentrations of glucose, fructose, and sucrose.

For independent validation, ternary mixtures containing 33–100 mM of each sugar were prepared in triplicate. The calibration standards and validation mixtures were prepared in a standard 96-well plate and analyzed using the Raman plate reader with 785-nm excitation. Spectra were acquired using a 2 s integration time and two accumulations per well. Under these conditions, a complete 96-well plate was analyzed automatically in less than 8 min, without manual intervention during spectral acquisition.

As shown in Figure 1a, 200 mM glucose, fructose, and sucrose solutions produced distinct Raman spectral profiles, although many of their bands overlapped across the fingerprint region because of their structural similarity. The spectral similarity makes quantification based on a single Raman band challenging and therefore benefits from MVA. Note that while sucrose is a dimer of glucose and fructose, its spectrum cannot be described as a sum of the spectra of the two components.

Before model development, the Raman spectra were preprocessed using standard normal variate (SNV) normalization, Savitzky-Golay 2^nd-derivative filtering, and mean centering. The concentration response variables were autoscaled to account for differences in their relative variation. A PLS model with 3 latent variables (LVs) was used to quantify all three sugars simultaneously.

The model successfully quantified each component in the ternary validation mixtures. The validation mixtures produced coefficients of determination of 0.998, 0.997, and 0.996 for glucose, fructose, and sucrose, respectively. Their corresponding root-mean-square errors of prediction (RMSEP) were 3.04, 4.70, and 2.28 mM. Prediction slopes ranged from 0.963 to 0.988, indicating close agreement between the predicted and prepared concentrations, with a small tendency toward underprediction. These results demonstrate that Raman can distinguish and quantify multiple structurally similar sugars from a single rapid measurement, even when their spectral features overlap.

Turbidity Affects Model Transferability

One of the major challenges in Raman analysis of fermentation samples is the effect of turbidity caused by suspended yeast cells. Turbidity can alter the excitation light penetration and Raman signal collection, leading to changes in signal intensity and background level. These optical effects can reduce the transferability of calibration model developed using clear aqueous standards.

To evaluate the turbidity effect, we added 30 µL of yeast suspension prepared at 2 mg/mL into each well containing the calibration and validation sugar solutions. The plate was immediately re-acquired upon yeast addition with the same acquisition settings. As shown in Figure 2a, the spectra changed noticeably after yeast addition. The aqueous sugar solutions have a consistent background, while the yeast-containing solutions exhibited much greater variations in their baselines.

We first directly applied the PLS model built from the clear aqueous standards reported in the previous section to the yeast-containing validation mixtures. As demonstrated in Figure 2b, the predicted concentrations remained correlated with the prepared concentrations, but there were systematic slope differences between the calibration and validation data for all sugars. This result indicates that the spectral effects introduced by yeast turbidity were not fully corrected by preprocessing alone.

A second PLS2 model was therefore developed using calibration standards containing the same yeast concentration as the validation samples. The same preprocessing and modeling approach was applied. Matching the turbidity of the calibration and validation samples substantially improved prediction performance, as shown in Figure 2c. The root-mean-square errors of prediction (RMSEP) improved from a range of 9.91 – 13.16 mM to a range of 4.99 – 6.54 mM. This confirms that calibration samples should reproduce the optical and matrix characteristics of the samples being analyzed. For fermentation process monitoring, including representative yeast concentrations or a range of turbidity levels in the calibration design could improve model robustness across different stages of the process.

Sugar Consumption and Ethanol Production Trajectories During Yeast Fermentation

After evaluating sugar quantification in prepared mixtures, our next step was to determine whether the sugar PLS model could be used to track substrate consumption during yeast fermentation. Two sugar solution formulations were prepared for the fermentation study. The first contained a ternary sugar mixture composed of equal parts of glucose, fructose, and sucrose at 67 mM each. The second contained a binary sugar mixture composed of only glucose and fructose, also at 67 mM each.

As summarized in Figure 3, budding yeast, Saccharomyces cerevisiae, was activated in the sugar-containing solutions at 42 °C and then allowed to ferment at room temperature. Aliquots were withdrawn from the fermentation broth every 30-60 min to generate a time-course sample set. The aliquots were immediately stored at 4 °C upon collection to slow further fermentation until analysis. After completion of the fermentation experiment, all collected aliquots were centrifuged. The clarified supernatants were transferred into a 96-well plate in triplicate and analyzed together using the Raman plate reader with 785 nm excitation.

Sugar concentrations over time were quantified using a PLS model developed from both aqueous sugar standards and yeast turbidity-matched sugar standards. The calibration data contains single-sugar dilutions, sugar mixtures, and corresponding yeast-containing sets to better represent the complex optical conditions encountered in fermentation samples.

The glucose, fructose, and sucrose concentrations predicted by the sugar PLS model was plotted as a function of fermentation time, as shown in Figure 4. Interestingly, in the ternary sugar mixture, sucrose concentration decreased rapidly and reached approximately zero within the first 30 min of fermentation. This behavior is consistent with the known ability of S. cerevisiae to secrete invertase, an enzyme that hydrolyzes sucrose into glucose and fructose before cellular uptake.¹ The rapid sucrose breakdown explains the initial increase in the predicted concentrations of both monosaccharides. After this initial increase, glucose and fructose were consumed at different rates, with glucose decreasing faster than fructose. This trend confirms that glucose is the preferred carbon source for S. cerevisiae compared to fructose.²

The binary sugar mixture provided a simpler system to evaluate monosaccharide consumption without additional contribution from sucrose hydrolysis. In this case, the predicted glucose and fructose concentrations decrease over time without an initial increase. Glucose was consumed more rapidly than fructose by approximately two-fold, further supporting the preferential utilization of glucose over fructose under these experimental conditions.

In addition to sugar consumption, Raman spectra were used to evaluate ethanol production during fermentation. As shown in Figure 5a-b, the time color-coded Raman spectra changed systematically over the fermentation time course for both sugar formulations. The zoomed-in spectra highlight the growth of the ethanol peak near 884 cm^-1. For visualization, the spectra in the inset were baseline-corrected to more clearly show the increase of this peak over time.

To quantify ethanol production, we developed a separate PLS model with ethanol-water mixtures prepared at concentrations ranging from 0 to 23% v/v. We note that nominal ethanol concentrations were calculated from the prepared ethanol-water volume ratios. Although ethanol-water mixtures exhibit slight volume contraction, this effect was not corrected because the fermentation samples fall in the low concentration range where the impact on predicted ethanol percentage is negligible. To minimize spectral interference from sugar bands, only a narrow spectral window from 877 to 895 cm^-1 was used for ethanol prediction. The spectra were preprocessed using Savitzky-Golay 2^nd-derivative filtering and mean centering. Two latent variables were selected. As shown in Figure 5c, the calibration model showed good performance with a fit slope of 0.9999.

We then applied the ethanol prediction model to the fermentation samples. As demonstrated in Figure 5d, the predicted ethanol percentage increased over time for both the ternary and binary sugar mixtures overall, confirming ethanol production during yeast fermentation. In the binary sugar mixture, the ethanol signal began to plateau around 6 h and decreased slightly at the final time point. This trend may reflect ethanol evaporation combined with depletion of fermentable sugars near the end of the experiment, as the predicted glucose and fructose concentrations were close to zero (Figure 4b). Additional experimental replicates and orthogonal ethanol measurements would be useful to confirm whether the final decrease reflects true process behavior or prediction variability.

Summary and Discussion

In this work, we explored the feasibility of using a Raman plate reader combined with multivariate analysis for time-resolved analysis of yeast fermentation samples. Glucose, fructose, and sucrose were first evaluated in prepared aqueous mixtures to determine whether individual sugars could be distinguished and quantified from overlapping Raman spectra. A PLS2 model successfully predicted the concentration of each sugar in ternary mixtures, demonstrating that multivariate Raman analysis can extract component-specific concentration information from chemically similar analytes.

The effect of yeast-induced turbidity was then evaluated. Applying a model developed from clear aqueous standards directly to yeast-containing samples resulted in systematic prediction differences, while a turbidity-matched calibration improved model performance. This result highlights an important consideration for fermentation analysis: calibration samples should reflect the optical and matrix conditions of the samples being measured.

When applied to yeast fermentation samples, the Raman plate reader workflow captured interesting substrate consumption trends. In the ternary sugar mixture, we detected rapid sucrose hydrolysis by yeast-secreted invertase, resulting in monosaccharide concentration increase during the beginning of fermentation. In both ternary and binary sugar mixtures, glucose was consumed faster than fructose, indicating a consistent carbon source preference. Raman spectra also showed an increase in ethanol concentration, and a separate PLS model was used to estimate ethanol production during fermentation. The primary value of this approach is its ability to analyze many fermentation aliquots efficiently in a standardized plate-based format with high-throughput and minimal manual intervention during acquisition. This makes the method well-suited for retrospective comparison of fermentation trajectories across strains, media formulations, carbon sources, and process conditions. Because Raman spectra can contain information from multiple components in a single measurement, the same dataset can support simultaneous monitoring of substrate depletion, product formation, and other chemical changes.

Future studies could further expand the utility of this workflow by incorporating more comprehensive calibration sets, additional biological replicates, and orthogonal validation measurements such as HPLC or enzymatic assays. Calibration models that include varying experimental conditions and complex fermentation media could improve model transferability to more complex fermentation settings. With further validation, automated Raman plate reader analysis could provide a rapid, reagent-free, and information-rich complement to conventional fermentation analysis, particularly for high-throughput process development and screening studies.

References

(1) Koschwanez, J. H.; Foster, K. R.; Murray, A. W. Sucrose Utilization in Budding Yeast as a Model for the Origin of Undifferentiated Multicellularity. PLoS Biol. 2011, 9 (8), e1001122. DOI: 10.1371/journal.pbio.1001122

(2) D’Amore, T.; Russell, I.; Stewart, G. G. Sugar Utilization by Yeast During Fermentation. J. Ind. Microbiol. 1989, 4 (4), 315–323. DOI: 10.1007/BF01577355