A better method to analyze bacterial mutation
Having a better understanding of bacterial mutation rates can play a vital role in public health applications such as food safety and infectious disease prevention. For nearly 75 years the Luria-Delbrück fluctuation assay has been a valuable tool for determining microbial mutation rates in the laboratory, though its use has been limited to mathematically inclined scientists and labs with sufficient computing resources. Improvements in technology and computational methods have made using the fluctuation assay feasible for a wider range of researchers, and a newly updated software package from the Texas A&M School of Public Health has further expanded the protocol’s reach.
In a new paper published in the journal G3: Genes, Genomes, Genetics and chosen as a 2018 Spotlight article, Qi Zheng, PhD, associate professor in the School of Public Health, discusses the latest version of his fluctuation assay software, rSalvador. He outlines the history of software designed for use with the fluctuation assay, examines the use of various computational methods involved in studying bacterial mutation and discusses important new features of the rSalvador software package, which Zheng has developed over the past 15 years.
In the early 2000s, Zheng developed software known as SALVADOR, making a new application of the fluctuation assay available to a wider range of researchers. However, SALVADOR ran in Mathematica, a proprietary commercial application whose cost presented an additional roadblock to some researchers. Several years later a web-based tool known as FALCOR was developed to give researchers a more affordable way to do basic fluctuation analysis. The new rSalvador package is more powerful, expanding beyond the capabilities of FALCOR and SALVADOR, and runs in the free and open-source statistical programming language R, making it available for many more researchers.
“Analyzing fluctuation assay data can be tricky for the uninitiated,” Zheng said. “Fluctuation assay analyses rely on certain mathematical assumptions. Having a thorough understanding of these assumptions and a versatile software package can help researchers avoid many potential pitfalls.”
The article focuses on a few of these pitfalls and provides examples to help researchers avoid making common mistakes.
For example, one typical overlooked problem is caused by a common laboratory practice called partial plating. When cultures contain too many cells for researchers to realistically count, they will often plate a small portion of the cell culture for analysis. However, researchers are often not aware of the need for adjustment for the additional random variation introduced by partial plating.
Zheng uses a raisin bread analogy to explain why partial plating should be accounted for when analyzing fluctuation assay data. “Like the raisins on a piece of raisin bread, mutants are randomly, but not necessarily evenly, distributed through the culture,” Zheng said. “If you cut a ten percent portion out of a piece of raisin bread, you often may not get precisely a ten percent share of the raisins.”
Adjustment for partial plating requires an algorithm that Zheng devised in 2008, and in the article, he uses published data to illustrate rSalvador’s capability to account for partial plating.
Another strength of rSalvador is its new computational methods for comparing fluctuation assay data from different experiments. Researchers are typically interested in more than a simple mutation rate estimate; they often need to compare mutation rates between different bacteria strains or across many different conditions. Previous research has relied on statistical tests like the Student’s t-test and the Mann-Whitney U test for this, but these tried-and-true statistical tests are often inappropriate for fluctuation assay data. One reason for the inapplicability of the Mann-Whitney test is that cultures in two experiments conducted under different conditions often reach noticeably different average cell population size.
“Directly comparing the numbers of mutants between two such experiments is like comparing apples and oranges,” Zheng said. “Applying the U test blindly to fluctuation assay data accomplishes just that. Continued use of inappropriate statistical tests to fluctuation assay data could dampen the credibility of a costly research project.”
rSalvador improves on this by giving two mutation rate comparison techniques: a likelihood ratio test and an empirical method to check whether confidence intervals overlap for two mutation rates. Zheng published the likelihood ratio test for fluctuation assay data in 2016, and he adapted the confidence interval-based method a year earlier.
“These techniques have been shown by simulation to give better results and they can also accommodate experiments involving partial plating,” Zheng said.
The software also provides methods to mitigate several other shortcomings found in recent analyses of fluctuation assay data.
In the article, extensive detail on rSalvador is provided as well as several examples to guide researchers on which of the software package’s methods are best for different experiments. The paper catalogs key mathematical assumptions underlying current methods for fluctuation assay data, which may serve as a road map for future research.
“This package provides a new tool for scientists who employ the fluctuation assay protocol to further understand the mechanisms and roles of bacterial mutation,” Zheng said. “It is remarkable that this 75-year-old experimental protocol today finds an increasingly wide acceptance in the mutation research community. I am elated to be able to help scientists make better use of this time-honored protocol.”