Modelling complex growth profiles of Bacteroides fragilis and Escherichia coli on various carbohydrates in an anaerobic environment

Zachary McGuire, Sainitya Revuru, Sheng Zhang, Amanda Blankenberger, Moiz Rasheed, Jacob D. Hosen, Guang Lin, Mohit S. VermabioRxiv 2023 · DOI

TL;DR

Researchers have developed a unified mathematical model capable of describing the entire lifecycle of a bacterial culture—including growth, stationary, and death phases—within a single equation. By introducing a time-dependent carrying capacity to the classic logistic growth model, the team successfully mapped the complex growth profiles of Bacteroides fragilis and Escherichia coli across 13 different carbohydrates.

What Problem Does This Paper Address?

In predictive microbiology, most researchers rely on models like the Logistic or Gompertz equations. While these are excellent for "sigmoidal" (S-shaped) growth, they have a major blind spot: they typically ignore the death phase. In a closed system, bacteria eventually exhaust resources or accumulate toxic byproducts, leading to a population decline.

Currently, scientists often have to "segment" their data—using one model for growth and a completely different one for death—or use complex "cybernetic" models that require deep knowledge of a cell's internal metabolism. This paper addresses the need for a versatile, "black-box" model that can fit various curve shapes (growth, death, or both) without needing prior metabolic data.

Methods

The authors modified the standard logistic growth equation. In the classic version, the "carrying capacity" ($K$) is a constant representing the maximum population the environment can support.

The innovation here is treating $K$ as a time-dependent variable ($K(t)$). They assumed that once the population reaches its peak ($Y_{max}$), the carrying capacity begins to decline at a linear rate ($d$).

  • The Model: $\frac{dY(t)}{dt} = r \cdot Y(t) \cdot (1 - \frac{Y(t)}{Y(t') - d \cdot (t - t')})$
  • Parameters: The model relies on two primary values: $r$ (the per capita growth rate) and $d$ (the rate of carrying capacity decline).
  • Workflow: They used the Digital Environment to Enable Data-driven Science (DEEDS) platform and custom Python scripts (parsing.py and modelFit.py) to automate data processing and parameter extraction from optical density ($OD_{600}$) measurements.

Datasets

The researchers generated an experimental dataset to test the model's versatility:

  • Bacteria: Bacteroides fragilis (an anaerobic gut commensal) and Escherichia coli LF82 (an adherent-invasive strain).
  • Substrates: 13 different carbohydrates, including monosaccharides (like D-glucose and L-fucose) and disaccharides (like Sucrose and Maltose).
  • Conditions: Each sugar was tested at three concentrations (100 mM, 10 mM, and 1 mM) in an anaerobic environment over 72 hours.
  • Replicates: Eight technical replicates were used for each condition to ensure statistical significance.

Key Insights and Ideas

  • The "Boom and Bust" of B. fragilis: Unlike E. coli, which reached a stable stationary phase on most sugars, B. fragilis exhibited significant death phases, particularly on maltose and lactulose. This decline was concentration-dependent, suggesting that the "crash" is tied to how the bacteria process specific sugars.
  • Base Media Differences: E. coli was able to grow on the chemically defined base media (ZMB1) alone by utilizing included amino acids. B. fragilis, however, was strictly dependent on the added carbohydrates to initiate growth.
  • The Phage Mystery: The authors investigated whether the sudden death in B. fragilis was caused by "prophages" (viruses hidden in the bacterial genome) waking up and killing their hosts. Their experiments ruled this out, implying the death is likely due to chemical changes in the environment (like pH drops or toxic metabolite buildup).

Key Results

  • Model Versatility: The proposed model successfully fit a wide variety of shapes, from standard S-curves to profiles showing rapid population crashes, outperforming traditional models that can only handle one or the other.
  • Growth Drivers: For E. coli, growth was generally optimized at 10 mM concentrations for most sugars. For B. fragilis, growth was highly substrate-specific; for example, it utilized N-acetyl glucosamine more efficiently than glucose.
  • Statistical Robustness: By using the product of $r$ and $Y_{max}$, the authors created a metric to rank the "quality" of growth on different carbon sources, providing a clear hierarchy of bacterial preference for different sugars.

Limitations and Future Work

The authors identify three main areas for improvement:

  1. Lag Phase: The current model does not include a specific parameter for the "lag phase" (the delay before growth starts), which is a common feature in bacterial cultures.
  2. Averaged Rates: The growth rate ($r$) is calculated as an average over the entire period rather than an instantaneous value, which can mask subtle changes in growth velocity.
  3. Biological Causality: While the model describes how the population changes, it doesn't explain why (e.g., it can't distinguish between death caused by acid accumulation vs. nutrient exhaustion).

Future work will focus on adapting the model to handle multiple growth rates (biphasic growth) and investigating the specific chemical changes that trigger the death phases observed in gut microbes like B. fragilis.

Summary updated

Abstract

Abstract Previously published models for microbial growth focus only on either death or growth and are unable to account for differently shaped growth curves. Currently, there is no model capable of incorporating combinations of microbial growth trends. This study creates a bacterial growth model that incorporates growth, death, lag, and tail phases as well as applies this model to the growth trends of Bacteroides fragilis and Escherichia coli on 13 different carbohydrate substances. Growth trends were collected by measuring the optical densities over 72 hours for either B. fragilis or E. coli in a chemically defined media supplemented by a mono- or disaccharide. The Digital Environment to Enable Data-driven Science (DEEDS) platform was utilized to parse data and apply the developed model to obtain parameter values. E. coli was found to grow on the chemically defined media alone while B. fragilis was unable to grow on it alone. E. coli growth was led by 10 mM concentration of substrates while B. fragilis growth was substrate dependent. Bacterial death only occurred for B. fragilis but was found to be dependent on concentration for the two most significant substrates. A singular model was developed that does not require prior knowledge of metabolomics and is capable of incorporating a combination of growth and death trends.
BibTeX
@misc{mcguire2023modelling,
  author = {Zachary McGuire and Sainitya Revuru and Sheng Zhang and Amanda Blankenberger and Moiz Rasheed and Jacob D. Hosen and Guang Lin and Mohit S. Verma},
  title = {{Modelling complex growth profiles of <i>Bacteroides fragilis</i> and <i>Escherichia coli</i> on various carbohydrates in an anaerobic environment}},
  year = {2023},
  doi = {10.1101/2023.05.01.538938},
  howpublished = {bioRxiv},
}
Figure