The choice of Kluyveromyces marxianus as the primary enzyme producer in this system reflects a long-standing recognition of its unusual physiological profile among industrial yeasts. Unlike most Saccharomyces-related species, K. marxianus grows readily at temperatures approaching 45 °C, tolerates a broad range of pH values, and can assimilate non-conventional carbon sources such as lactose and inulin. These traits have made it attractive for processes that run at elevated temperatures, where contamination risk is lower and cooling costs are reduced. Its thermotolerance also provided a practical selective advantage in the present study: plating samples on YPD agar incubated at 45 °C allowed K. marxianus colonies to be counted while S. cerevisiae, which cannot form colonies at that temperature, was excluded. This simple phenotypic selection, paired with lysine agar selective for S. cerevisiae, gave the researchers a low-cost means of tracking population dynamics across the fermentation without genetic markers or molecular quantification.

The ecological logic underlying the co-culture is an example of metabolic cross-feeding, a phenomenon well documented in natural microbial communities where the waste product of one organism becomes the substrate of another. In aerobic yeast fermentations, dextrose is converted to ethanol even in the presence of oxygen when sugar concentrations are high, a behavior known as the Crabtree effect. K. marxianus is a Crabtree-positive yeast, so during the early phase of the batch it channels a substantial fraction of consumed dextrose into ethanol rather than fully oxidizing it to carbon dioxide. For an enzyme producer, this is problematic for two reasons: ethanol diverts carbon away from biomass and protein synthesis, and accumulated ethanol inhibits cellular metabolism, with K. marxianus being notably sensitive compared to its S. cerevisiae relative. By introducing S. cerevisiae, which oxidizes ethanol efficiently once glucose is depleted, the consortium effectively recycles a waste stream back into biomass and carbon dioxide, simultaneously removing the inhibitor from the environment.

The temporal structure observed in the fermentation — a dextrose-consumption phase followed by an ethanol-respiration phase — is a classic diauxic-like pattern familiar from decades of yeast physiology research. During the first phase, heat flow and carbon dioxide evolution measurements reflect fermentative growth on sugar. Once dextrose is exhausted, the off-gas signature changes as the culture shifts to oxidative growth on ethanol, which typically produces a distinct respiratory quotient and a different heat yield per mole of substrate. Biocalorimetry is particularly well suited to detecting such transitions because metabolic heat is generated by every catabolic reaction and can be measured continuously and non-invasively. The heat flow curve effectively integrates all metabolic activity in the vessel, so a shift in the dominant catabolic pathway appears as a change in both the magnitude and the shape of the thermal signal. Combining this with off-gas carbon dioxide analysis provides two independent, real-time windows into the same metabolism, which strengthens confidence in the phase assignments.

The reported enzyme titers deserve contextualization. Carboxypeptidase Y is a well-characterized serine carboxypeptidase, originally studied extensively in S. cerevisiae, where it serves as a model for vacuolar protein sorting and zymogen activation. Its industrial appeal stems from its exopeptidase activity: by sequentially removing amino acids from the carboxyl terminus of peptides, it can debitter protein hydrolysates, a persistent challenge in the production of protein-based food ingredients, since many bitter peptides terminate in hydrophobic residues that the enzyme preferentially cleaves. Beyond food processing, CPDY has found roles in peptide sequencing, in the modification of pharmaceutical proteins, and more recently in the emerging bioeconomy, where hydrolysis of proteins bound to lignocellulosic residues can release amino acids that feed subsequent fermentation steps for products such as bioethanol and polyhydroxyalkanoates. Doubling the achievable titer through a co-culture strategy, rather than through genetic engineering or expensive medium supplements, is significant because it addresses the economics of enzyme manufacturing, where downstream recovery costs scale inversely with fermentation titer.

Response Surface Methodology, the optimization framework used here, is a statistical design approach that models the response of a process — in this case enzyme activity — as a mathematical function of several input variables, allowing interactions between factors to be captured rather than examined one at a time. For a medium containing yeast nitrogen base, ammonium chloride, and dextrose, the interactions are non-trivial: the carbon-to-nitrogen ratio influences both the rate of ethanol formation and the balance between biomass and product formation, while the trace elements and vitamins supplied by yeast nitrogen base support enzyme folding and secretion. By fitting a quadratic response surface to a modest number of experiments, RSM identifies near-optimal compositions efficiently, which matters in bioprocess development where each shake-flask experiment consumes days of labor and materials.

The application of a Long Short-Term Memory network to this fermentation reflects a broader shift in bioprocess engineering toward data-driven modeling. Traditional mechanistic models of microbial growth, such as Monod-type kinetics extended with product inhibition terms, require assumptions about which variables matter and how they interact, and they struggle when consortia introduce species interactions that are hard to parameterize. LSTM networks, by contrast, learn temporal dependencies directly from time-series data, using gated memory cells that can retain information over long sequences — a property well matched to bioprocesses where the state of the culture at one hour depends on events many hours earlier. The reported error metrics, a mean squared error of 0.0060 and a mean absolute error of 0.071, indicate that the model reproduced the measured trajectories of metabolic transitions and enzyme yields with small residuals, though such metrics should always be interpreted relative to the scale of the measured variables and validated on data not seen during training. If robust, a model of this kind could support soft sensing — inferring hard-to-measure quantities like enzyme activity from continuously available signals such as heat flow and off-gas composition — and could feed model-predictive control strategies that adjust feeding or aeration in real time.

The division of labor observed in the consortium also raises questions about population stability, which the CFU plating addressed directly. Synthetic and natural co-cultures are vulnerable to takeover by the faster-growing species, and the balance between K. marxianus and S. cerevisiae here depends on the sequential availability of their preferred substrates. The finding that S. cerevisiae remained a minority during the dextrose phase and expanded during the ethanol phase suggests that the interaction is self-regulating to some degree: the ethanol produced by one partner creates the niche occupied by the other. Such substrate-mediated mutualism is generally more stable than interactions requiring constant inoculum tuning, which bodes well for scale-up, although maintaining the 1:1 inoculation ratio at larger volumes and in repeated batches would still require monitoring.

From a sustainability standpoint, the strategy aligns with several trends in industrial biotechnology. Avoiding ethanol accumulation means less carbon is lost as an unusable byproduct, improving the effective carbon yield of the process. The use of GRAS-status organisms means the enzyme product can be manufactured without containment concerns that accompany recombinant or pathogenic hosts, simplifying regulatory pathways for food and pharmaceutical applications. And the integration of calorimetry and respirometry as process analytical tools fits within the quality-by-design paradigm increasingly expected in biomanufacturing, where real-time measurements replace end-point testing. The combination of a biological design solution — cross-feeding co-cultivation — with an information technology solution — deep learning-based prediction — illustrates how modern enzyme production processes are becoming simultaneously more ecological in their logic and more digital in their control, a pairing that may prove applicable to other fermentation products whose yields are limited by self-produced inhibitory metabolites.

Subject of Research: …

Article Title: Co-cultivation of yeast to enhance carboxypeptidase enzyme production: biocalorimetric, respirometric, and deep learning investigation

Article References: Ganesan, S. K., V, A. J., P, V. M., Dinakarkumar, Y., NE, S., & A, S. R. (2026). Co-cultivation of yeast to enhance carboxypeptidase enzyme production: biocalorimetric, respirometric, and deep learning investigation. Frontiers in Chemical Engineering, 8, Article 1894850. https://doi.org/10.3389/fceng.2026.1894850

Image Credits: AI Generated

DOI: 10.3389/fceng.2026.1894850

Keywords: Co-cultivation, yeast, enhance, carboxypeptidase, enzyme, production, biocalorimetric, respirometric, deep, learning, investigation, scientific research