Boltzmann Model Outperforms Classic Kinetics in Microbial Polymer Fermentation

A twist of mathematics is reshaping how scientists understand one of biotechnology’s most versatile microbial products. In a study published in the journal 3 Biotech, researchers led by Shilin Hu and Feilong Sun of Xi’an Polytechnic University in China have shown that a four-parameter Boltzmann empirical model can describe the batch fermentation of poly-γ-glutamic acid, or γ-PGA, by the bacterium Bacillus subtilis natto more faithfully than the classical Logistic-Luedeking-Piret framework that has dominated fermentation kinetics for decades. The work matters because γ-PGA, a sticky, water-absorbing biopolymer produced naturally by certain Bacillus strains, is finding uses everywhere from wastewater remediation and agriculture to cosmetics and drug delivery, and better models of its production translate directly into better industrial processes.

Poly-γ-glutamic acid is an unusual biopolymer. Unlike proteins, whose amino acids are linked by peptide bonds in a single direction, γ-PGA is built from repeating glutamate units connected through amide bonds between the amino group of one unit and the gamma-carboxyl group of the next. That architecture gives the molecule remarkable properties: it is biodegradable, edible, non-toxic, and capable of holding large amounts of water. It is also produced by the same bacterium that ferments soybeans into the Japanese food natto, which is why Bacillus subtilis natto is a workhorse strain for making it industrially. As demand for sustainable, bio-based materials grows, engineers are under pressure to squeeze every gram of product from every liter of fermentation broth, and that is where kinetic models come in.

For generations, fermentation engineers have leaned on a family of equations known as unstructured kinetic models, which describe how biomass, substrate, and product change over time without tracking every internal metabolic reaction. The most widely used combination pairs the Logistic equation, which captures the sigmoidal, S-shaped growth curve of bacteria, with the Luedeking-Piret equation, which links product formation to both the growth rate and the size of the existing cell population. This Logistic-Luedeking-Piret, or L-LP, framework works well for many fermentations, but it carries a hidden assumption: once the product reaches its peak, it stays there. The equations simply cannot describe a decline in product concentration after the maximum has passed.

That assumption broke down in the Xi’an experiments. Over a 36-hour batch fermentation, the researchers tracked three variables: biomass, the concentration of γ-PGA, and the concentration of glucose, the carbon source feeding the whole process. Bacterial growth followed the familiar sigmoidal pattern, with a maximum specific growth rate of 0.7362 per hour and a peak biomass concentration of 1.82 grams per liter in dry cell weight. γ-PGA production climbed steadily to a maximum of 18.76 grams per liter. But then something the classical model could not handle occurred: as glucose became depleted and the culture entered carbon-starved conditions, the γ-PGA concentration declined modestly. The product was not stable; the fermentation was asymmetric, with accumulation on the way up and losses on the way down.

To understand exactly how the classical model failed, the team did something clever. They built a diagnostic extension of the L-LP framework by adding a first-order product degradation term, essentially allowing the model to subtract a quantity of γ-PGA proportional to how much was already present. This modification was not meant to be the final answer; it was a probe, designed to quantify how much of the mismatch between the standard model and the real data could be attributed to the post-peak decline. By comparing the extended and unextended L-LP models, the researchers could confirm that product degradation under carbon-depleted conditions was the key phenomenon the classical equations missed, rather than some other feature of the fermentation’s behavior.

The alternative the researchers evaluated came from a different branch of mathematics. The four-parameter Boltzmann model is an empirical sigmoidal function, borrowed conceptually from the curves that describe phase transitions and other smooth but sharply changing natural processes. Its four adjustable parameters let it control where the transition begins, how steeply it rises, and, crucially, what happens at the end of the curve. Where the Logistic-Luedeking-Piret system forces the product concentration to level off at its peak value forever, the Boltzmann formulation has the flexibility to represent a near-plateau phase that does not behave like a perfect plateau, including the subtle late-stage sag in γ-PGA concentration as glucose ran out.

The quantitative comparison between the two frameworks was striking. For biomass accumulation, the Boltzmann model matched the Logistic model almost exactly, with both achieving a coefficient of determination of roughly 0.99, meaning they explained 99 percent of the variance in the growth data. Biomass, in other words, is the easy part; the classic sigmoidal description remains fully adequate. But for the product and the substrate, the picture changed dramatically. The Boltzmann model reduced the root-mean-square error, a standard measure of average prediction error, by approximately 45 percent for γ-PGA and 47 percent for glucose relative to the classical model, while also lowering the error for biomass by about 10 percent.

Information-theoretic criteria reinforced the verdict. The researchers used Akaike’s Information Criterion, or AIC, a statistical tool that balances how well a model fits the data against how many parameters it consumes, penalizing unnecessary complexity. Even though the Boltzmann model uses four parameters per curve, it won decisively: the AIC difference in favor of the Boltzmann description reached 11.72 for γ-PGA and 12.70 for glucose. In the conventions of model selection, differences of this size constitute strong evidence that the better-fitting model is genuinely superior, not merely a fluke of extra flexibility. For process engineers, that means the Boltzmann approach is not just fitting noise; it is capturing real structure in the fermentation dynamics that the classical framework is blind to.

The study is honest about the limits of the new approach. The Boltzmann model reproduced the near-plateau phase of γ-PGA accumulation more accurately than the classical equations, but it did not fully capture the slight decrease in product concentration at the very end of the run. The diagnostic L-LP extension with a degradation term still played a valuable role in explaining why that decline happens, pointing to enzymatic or metabolic consumption of the polymer once the carbon source is gone. In other words, the Boltzmann model is the better empirical tool for prediction, while the degradation-extended L-LP framework serves as a mechanistic hypothesis about what the cells and their enzymes are doing when glucose disappears.

The practical implications extend well beyond the statistics. Because γ-PGA production is highly sensitive to fermentation conditions, and because the polymer’s high viscosity complicates mixing and oxygen transfer in industrial reactors, having a kinetic model that accurately reflects the entire batch, including the carbon-starved tail, gives process designers a quantitative foundation for optimization. The authors position their Boltzmann framework as a practical basis for subsequent process optimization and fed-batch scale-up studies, where feeding glucose at the right moments could prevent the product degradation phase entirely and push final titers higher. The work, funded by the Shaanxi Provincial Science and Technology Key Project and Xi’an Municipal research programs, is a reminder that even in a field as established as fermentation engineering, the choice of an equation can be the difference between a process that plateaus and one that reaches its true potential.

Subject of Research: Non-linear Boltzmann kinetic modeling of poly-γ-glutamic acid batch fermentation by Bacillus subtilis natto

Article Title: Modeling the asymmetric metabolic shifts in poly-γ-glutamic acid batch fermentation using a non-linear Boltzmann approach

Article References: Hu, S., Li, H., Wang, Y., Li, W., Shang, M., & Sun, F. (2026). Modeling the asymmetric metabolic shifts in poly-γ-glutamic acid batch fermentation using a non-linear Boltzmann approach. 3 Biotech, 16(10), Article 447. https://doi.org/10.1007/s13205-026-05082-6

Image Credits: AI Generated

DOI: 10.1007/s13205-026-05082-6

Keywords: poly-γ-glutamic acid, Bacillus subtilis natto, batch fermentation, fermentation kinetics, Boltzmann model, Logistic-Luedeking-Piret model, biopolymer, bioprocess engineering, kinetic modeling, AIC model selection, product degradation, fed-batch scale-up