MicroMPN: methods and software for high-throughput screening of microbe suppression in mixed populations

ABSTRACT

Screening assays are used to test if one or more microbes suppress a pathogen of interest. In the presence of more than one microbe, the screening method must be able to accurately distinguish viable pathogen cells from non-viable and non-target microbes in a sample. Current screening methods are time-consuming and require special reagents to detect viability in mixed microbial communities. Screening assays performed using soil or other complex matrices present additional challenges for screening. Here, we develop an experimental workflow based on the most probable number (MPN) assay for testing the ability of synthetic microbial communities to suppress a soil-borne pathogen. Our approach, fluorMPN, uses a fluorescently labeled pathogen and microplate format to enable high-throughput comparative screening. In parallel, we developed a command-line tool, MicroMPN, which significantly reduces the complexity of calculating MPN values from microplates. We compared the performance of the fluorMPN assay with spotting on agar and found that both methods produced strongly correlated counts of equal precision. The suppressive effect of synthetic communities on the pathogen was equally recoverable by both methods. The application of this workflow for discriminating which communities lead to pathogen reduction helps narrow down candidates for additional characterization. Together, the resources offered here are meant to facilitate and simplify the application of MPN-based assays for comparative screening projects.

IMPORTANCE

We created a unified set of software and laboratory protocols for screening microbe libraries to assess the suppression of a pathogen in a mixed microbial community. Existing methods of fluorescent labeling were combined with the most probable number (MPN) assay in a microplate format to enumerate the reduction of a pathogenic soil microbe from complex soil matrices. This work provides a fluorescent expression vector available from Addgene, step-by-step laboratory protocols hosted by protocols.io, and MicroMPN, a command-line software for processing plate reader outputs. MicroMPN simplifies MPN estimation from 96- and 384-well microplates. The microplate screening assay is amenable to robotic automation with standard liquid handling robots, further reducing the hands-on processing time. This tool was designed to evaluate synthetic microbial communities for use as microbial inoculates or probiotics. The fluorMPN method is also useful for screening chemical and antimicrobial libraries for pathogen suppression in complex bacterial communities like soil.

INTRODUCTION

Microbes are ubiquitous in nature, and they live in communities where they perform distinct functions within their ecosystems. Given this community-level functionality, there is an increasing interest in engineering synthetic multispecies communities whose interactions and emergent behaviors can be leveraged to perform specific tasks (1–4). One such area of interest is screening microbial communities for their ability to suppress pathogenic microorganisms (5–9). To evaluate the effect of a community for the suppression of a target pathogen, a screening method must be used to accurately distinguish viable pathogenic cells from non-viable ones and non-target microbes in a sample. This work presents a unified set of accessible, straightforward laboratory protocols and software amenable to the evaluation of the suppressive effects of engineered synthetic communities in complex environmental microcosms.

Suppression screening assays in complex microcosms are challenging to implement because they require a high level of accuracy and specificity for the enumeration of the pathogen of interest. Plating on selective media is the most commonly used method of enumeration, which relies primarily on the composition of the media to achieve specificity (10). The degree of selection, however, is also dependent on the microbe and environmental sample of interest (11). Though simple to implement, this method is space- and labor-intensive because it requires counting hundreds or thousands of bacterial colonies on multiple plates per experiment. Molecular methods like quantitative polymerase chain reaction (qPCR) allow for the specific detection of microbial DNA, circumventing the need for manual counting. However, qPCR does not differentiate cell viability unless dyes like propidium monoazide are added to inactivate free DNA (12, 13). Optical methods like flow cytometry and epifluorescence microscopy can distinguish intact cells of target microorganisms with the use of labeled antibodies and membrane-impermeable DNA dyes (14, 15), but sample processing is time-consuming (16, 17). These screening techniques also rely on expensive laboratory equipment and specialized training, limiting their accessibility and application. In our domain of interest (soil), the sample environment itself further complicates the use and optimization of these methods and can significantly increase sample processing requirements. For optical detection methods, such as flow cytometry and microscopy, it is necessary to separate soil particles from the microbial population (16). For PCR-based approaches, humic acids and organic matter present in a soil sample can hinder nucleic acid isolation, compromise extraction efficiency, and interfere with enzymatic activity (18).

The most probable number (MPN) assay provides an alternative approach, which has long been used for surveying the concentrations of microbes in complex environments (19–22). MPN enumeration is based on the use of replicate samples that are serially diluted. The number of tubes with growth at each dilution, as indicated by an optical signal such as turbidity (23) or color change (24, 25), is used to calculate the concentration of microbes in the undiluted sample (26). A standard MPN assay relies on the properties of the growth medium. Only the microbes capable of metabolizing the nutrients in the medium (e.g., anaerobes, methanogens, and iron reducers) will grow and produce a measurable signal (27–29). For increased specificity, variants of this approach have used microbial strains that were engineered to produce luminescence (30, 31) or express fluorescence (32). This allows for the differentiation of a targeted microbe from the rest of the sample using an intrinsically produced signal. Other implementations of the MPN assay have been adapted to microplates, increasing their throughput and reducing their runtime, labor, and space requirements (24, 33–36).

In this work, we combine and augment several versions of the MPN technique, creating a unified set of tools for profiling synthetic microbial communities using fluorescence and MPN enumeration in laboratory microplates. The MicroMPN workflow includes (i) labeling the target pathogen with a publicly available broad-host-range expression vector containing the inducible mRFP1 fluorescent protein, (ii) step-by-step open access protocols for MPN dilution, fluorescence induction, data collection with a microplate reader, and data analysis, and (iii) new software for processing data from one or more microplates of any layout into MPN estimates. The protocols and software were validated through a series of experiments where we evaluated the ability of synthetic microbial communities to suppress the plant pathogen Ralstonia solanacearum in soil. Our unified methods were able to specifically quantify Ralstonia suppression in mixed soil communities with precision and accuracy equivalent to spot plating on selective solid media.

The resources presented here are intended to support and simplify the process of bacterial antagonist-based pathogen screening in complex microcosms. With the cost of sequencing technology decreasing, an ever-greater number of projects are focused on understanding microbial community dynamics and their encoded functionalities across environments (37). Moreover, the wider adoption of artificial intelligence in microbiology will accelerate our learning and predictive capabilities (38–41). Combining these two technologies will result in massive quantities of microbial data being generated for hypothesis testing. Ultimately, high-throughput workflows like ours will be essential for experimental testing and validation and will enable the rapid assessment of biologically relevant treatments for further characterization.

RESULTS

MicroMPN workflow

The workflow can be divided into two parts (Fig. 1).

Fig 1.

MicroMPN workflow. (1) Resuspend samples (e.g., soil) in an appropriate buffering agent and transfer sample aliquots to a microplate. (2) Proceed to serially dilute in selective or semi-selective media. Include inducer in the media (only if applicable). Record spectrofluorometric data prior to initializing incubation. (3) Incubate microplate to enable microbial growth and fluorescent protein expression. (4) Record spectrofluorometric data (e.g., time 48 hours). (5) Proceed to calculate MPN values using our software MicroMPN. (5A) Input file 1: the user specifies the layout of a microplate. This file contains the number of replicates, dilution series, and the number of samples per plate in “wellmap” format, a simple TOML-based file specification. MicroMPN parses the information of a microplate from the TOML file for the downstream calculation of an MPN value. (5B) Input file 2: user provides a CSV file (e.g., 48-hour time point data) with the plate name, the well location in a microplate, and the spectrofluorometric data. Users must specify a cutoff value, which the program uses to determine what wells are positive or negative for microbial growth. (5C) Output file: MicroMPN will calculate an MPN value, a bias-corrected MPN value, a 95% confidence interval, and a rarity index value.

Part 1: fluorMPN

The experimental portion of this workflow, which centers around the processing of environmental samples using the MPN fluorescence microplate assay, is referred to as the fluorMPN assay for brevity. This method determines unknown concentrations of a tractable microorganism in microbially heterogeneous samples. This version of the MPN assay requires the use of a fluorescently tagged microbe. The fluorescence signal serves as an indicator of target specificity and viability. Upon completion of an experiment, samples are transferred and processed on a microplate. The plate layout design used here is based on Rowe et al. (35) in a 96-well format: 12 dilutions by 8 replicates per sample. Next, samples are serially diluted in semi-selective media and, when necessary, in the presence of an inducer (i.e., homoserine lactone). This reduces the background noise found in the samples and induces the expression of fluorescence by the microbe. The plates are then incubated to enable growth and fluorescence expression of the target. The incubation time required for the assay will depend on the study system; we describe ours below. The relative fluorescence units (RFUs) are collected with a plate reader.

Part 2: MicroMPN

The computational portion of this workflow produces an estimate of the unknown concentration of a target, based on the output of the fluorMPN assay, using an integrated software package called MicroMPN. The software is available for download from GitHub (https://github.com/USDA-ARS-GBRU/micrompn) and the Python Package Index (https://pypi.org/project/micrompn/) (42). Input consists of a user-specified plate layout file and a CSV data file from the plate reader; output consists of the MPN estimate for a given sample, a 95% confidence interval for the MPN estimate, and a rarity index, which flags improbable values (e.g., when the number of positive wells increases with dilution) (43). These estimates are calculated based on the methods presented by Jarvis et al. (26), which are implemented in the “MPN” R package (44). Our software performs the statistical calculations using Scipy (45), resulting in fast MPN estimation (1.05 ms, S.D. 0.014 per MPN estimation, on a 2020 Apple laptop computer with a 2.0 GHz quad-core 10th-generation Intel Core i5 processor). MicroMPN generates estimates for an arbitrary number of samples and dilution factors, which can be spread across multiple plates, potentially with multiple samples per plate. To parse the data, users provide a Wellmap-formatted file specifying the arrangement of each plate, including the dilution sequence, replicate number, and sample locations; the Wellmap format is a well-documented specification protocol based on the widely used TOML configuration file format (46). Once the data is imported, a user-supplied cutoff value (see Processing of Microcosms), relevant to the data type collected on the plate reader, is used to determine which wells are positive or negative. The number of positive replicates per dilution factor is recorded for each sample on the plate. The number of positive replicates, total number of replicates, and value of the dilution factors are stored as arrays and used to directly estimate an MPN value for each sample on each plate (26). Text S1 further describes the theoretical principles and mathematical equations used to derive an MPN value.

The experiments described below evaluated the accuracy of the fluorMPN assay and MicroMPN software in enumerating the soil-borne pathogen R. solanacearum in soil microcosms unamended and amended with synthetic microbial communities. Their performance was compared to that of spotting on agar.

Evaluating vector retention by rfp-Rs5 in soil lacking selective antibiotics

To be effective, a fluorescent reporter must be retained in the bacterium without antibiotic selection since the antibiotics would alter the experiment. To evaluate retention, the soil was inoculated with Ralstonia, and the mean colony counts of Ralstonia on semi-selective South Africa (SMSA) gentamicin 0 mg/L (GEN-0) and GEN-20 agar were compared (Text S2). For unamended soil microcosms, there were no significant differences in mean counts at any of the time points collected (Fig. 2). By 24 hours of soil incubation, the concentration of Ralstonia in the soil had increased almost by a factor of 10 (log₁₀ = 1 CFU/g) and remained stable throughout the experiment. Plasmid stability was also tested in soil with a 10-member microbial community. Plasmid stability was evaluated 72 hours post-inoculation of soil (Text S2). In this experiment, the mean counts between SMSA GEN-0 and GEN-20 were significantly different (T₇₄ = 2.75, P = 0.037) with a small decrease in log abundance of 0.05 or a fold change of 1.1 on the plates with gentamycin.

Fig 2.

Assessing the stability of pFLxR5 in strain rfp-Rs5 in soil microcosms. Box plots of Ralstonia log₁₀-transformed CFU/g counts obtained from SMSA GEN-0 and GEN-20 at different time intervals. A box plot extends to the 25th and 75th percentiles, with the median, mean (x), upper and minimum values (whiskers), and outliers of the data set being displayed. Quartiles were calculated including the median. Sample size per time interval for unamended soil and soil with a synthetic community were 8 and 75, respectively. Differences between count means for SMSA GEN-0 and GEN-20 were assessed through a paired t-test at a significance level α < 0.05. P-values adjusted to account for multiple comparisons using the Bonferroni correction (^*P = 0.037, ns = not significant). For unamended samples, the table displays log₁₀ differences between 2 hours and the rest of the time points within an agar condition. For soil with a synthetic community, the log difference is also displayed.

Comparing the fluorMPN assay to spotting on agar

Bacterial counts from the fluorMPN assay and spotting on agar were compared in unamended soil and soil plus synthetic community samples. The results from both soil systems displayed linear relationships between methods of quantification. In unamended soil systems (Fig. 3A), the two methods were found to be significantly, positively correlated (r = 0.96, P = 1.3E−6) where the fitted regression model explained 91.23% of the shared variance. A similar correlation output was obtained in the presence of synthetic communities (Fig. 3B, r = 0.82, P = 7.7E−09). However, the fitted regression model explained 66.43% of the shared variance. To showcase the sensitivity of the software in recognizing plates with improbable results, Fig S1 displays the same data set as in Fig. 3B with the addition of improbable MPN data points.

Fig 3.

Assessing the strength and relationship of counts obtained by fluorMPN microplate assay and spotting on agar. Data plots display a correlation analysis between log₁₀-transformed CFU/g and MPN/g counts of rfp-Rs5 obtained from unamended soil (A) and soil and synthetic community (B) microcosms. A Pearson correlation test was implemented; correlation coefficient (r) and P-value are displayed. The regression equation and coefficient of determination (R²) are also included. The 95% confidence bands for the linear model are shown in gray. For both (A) and (B), data plotted correspond to counts of Ralstonia 3 days post-inoculation of soil. In (A), soil was inoculated with four concentrations (see legend) of strain rfp-Rs5 (n = 3 soil replicates per concentration). In (B), data plotted correspond to community treatments (C2–C5) and control (rfp-Rs5). For both methods being compared, no counts of Ralstonia were obtained for communities C1 and C6 due to background interference.

The level of agreement on count estimates made by each method of quantification varied depending on the soil system. For unamended soil microcosms (Fig. 4A), the mean difference (mean bias, $\overline{d}$) ± standard deviation (SD) between the fluorMPN assay and spotting on agar was −0.19 ± 0.29. While in soil microcosms with added consortia (Fig. 4B), the mean bias and standard deviation of the differences were larger, at −0.56 ± 0.33. Nonetheless, the negative sign associated with the mean bias values from the comparisons suggests that on average the fluorMPN assay overestimates the counts of Ralstonia by a factor of 1.5 ± 2.0 and 3.6 ± 2.1, respectively, as compared to spotting on agar.

Fig 4.

Determining the level of agreement between fluorMPN microplate assay and spotting on agar. Bland–Altman plots for rfp-Rs5 count data obtained from unamended soil (A) and soil and synthetic community (B) microcosms. The x-axis displays the mean Ralstonia count in soil measured by the two different methods of quantification. The y-axis displays the between-methods difference of measured counts. In the plot, the red lines mark the 95% limits of agreement between the two methods, while the black line represents their mean bias ± standard deviation. A blue line was also plotted at x = 0 as a reference for perfect agreement between the methods. Gray lines around the bias line correspond to the 95% confidence interval. Improbable MPN values were excluded from calculating the bias and lower and upper agreement values.

The precision of Ralstonia counts obtained from each method was also evaluated by comparing their variances. As shown in Table 1, no significant difference in count precision was obtained when comparing the outputs from the fluorMPN assay to spotting on agar for both soil microcosms. Similar conclusions were obtained with a variance component analysis (Text S3), which estimates the proportion of variance associated with a random effect; the variance component analysis (VCA) was run for each method individually to evaluate the between-group variation (i.e., between soil communities) and error containing within-group variance (i.e., between soil replicates for a community).

TABLE 1.

Comparing variances between methods^a

Soil microcosms	Two-sample F-test for variance
	F-statistic	P-value (α = 0.05)	df
Unamended soil	1.11	0.87	11, 11
Soil and synthetic communities	1.01	0.97	32, 32

^a For each soil microcosm, the F statistic was calculated by comparing the variances associated with counts of rfp-Rs5 obtained with the fluorMPN assay (numerator) and spotting on agar (denominator). Two-tailed p values are presented.

Evaluating community suppressive activity across enumeration methods

In screening experiments, the absolute change in abundance is often less important than the relative suppression of the microorganisms screened. Except for community C4, the mean difference between each community and the control was significant for both the fluorMPN assay and spotting on agar (Fig. 5). The log difference and percent decrease between each community and the control were similar between enumeration methods.

Fig 5.

Evaluating the inhibitory activity of communities against strain rfp-Rs5 for each method of quantification. Soil counts of Ralstonia as determined by fluorMPN microplate assay (A) and spotting on agar (B). A box plot extends to the 25th and 75th percentiles, with the median, mean (x), upper and minimum values (whiskers), and outliers of the data set being displayed. Quartiles were calculated excluding the median. Statistical differences being displayed correspond to multiple comparisons between groups performed with Dunnett’s test where α < 0.05 (adjusted P-values: ^***P < 0.001). Shown above each plot, log differences and percentage decrease (shown in parenthesis) relative to rfp-Rs5 control.

DISCUSSION

This work presents a unified set of validated laboratory protocols for screening fluorescently labeled pathogens in soil matrices and automating MPN estimations directly from microplate formats. As described in Materials and Methods, our experimental and computational protocols are available through the platform protocols.io. The first protocol, “Most Probable Number Fluorescence Microplate Assay,” explains how to run the assay with clay soil matrices inoculated with Ralstonia. For other bacterial systems, soil types, and environmental conditions, users will need to optimize and adapt the steps of the protocol to best fit their experimental needs. This may include, among other things, changing any of the following parameters: number of dilution series, dilution volumes, media type, selection of an appropriate fluorescent label, deciding on constitutive expression of fluorophore vs induction (used here), and microplate incubation time. In the second protocol, “MicroMPN: Software for Automating Most Probable Number Estimates from Laboratory Microplates,” we cover how to run the software using RFU data as an example. We also provide users with example data, Wellmap format files, and detailed instructions for use. The software can be used with other data types (e.g., optical density and luminescence) by simply specifying an appropriate positive/negative threshold value for that data type.

MicroMPN software is distinct from other MPN calculation tools including spreadsheets (36, 47, 48), computer programs (33, 49–51), command-line programs (44), and online interfaces (52, 53). To our knowledge, there are two other publications in which computer-based methods were developed to calculate MPN values from microplates (33, 36). Irwin et al. (36) created an Excel spreadsheet that estimates an MPN value from a single plate. Walser (33) wrote a Windows graphical program in Visual Basic 5 that allowed users to add multiple samples per plate. In contrast, MicroMPN is a Python-based command-line application that is compatible with any system that can run Python 3 and has a command-line interface. It follows the dataflow programming paradigm where data are processed in multiple steps represented by a directed acyclic graph of inputs and outputs. This makes it amenable to integration with workflow managers like Nextflow (54) and Apache Beam (55). The program also separates the physical layout information from the plate reader data using a standardized layout format called Wellmap. Statistical estimations are made using Scipy: a library of fast, optimized algorithms for optimization and statistics (45). Additionally, MicroMPN uses several software development practices that increase code reliability, including unit testing, continuous integration, and containerization.

Validation of the fluorMPN assay for suppression screening

Our results confirmed previous observations that MPN and spotting (i.e., CFU) methods correlate positively and strongly but differ in their absolute counts (23, 33, 34, 56). For both unamended soil and soil with synthetic communities, we observed that MPN count estimates on average were consistently greater than CFU counts. A previous study sought to identify intrinsic sources of variability (e.g., human error or laboratory procedure) within MPN and CFU methods that could drive the differences in count estimates obtained for a single source (similar experimental design to ours) (57). Through probabilistic modeling of each method, based on their conditional probability distributions, the authors concluded that the process of dilution and enumeration is zero-censored resulting in a small, known, positive bias in MPN estimates, which we also observed.

The positive bias in MPN relative to true concentration has been documented previously, and Salama et al. (58) and Haas (59) have proposed methods to adjust for MPN bias that bring the uninoculated samples in line with the CFU estimates. Surprisingly, the difference between CFU and MPN estimates is higher in the synthetic community than in the pathogen-only soil experiments (1.5× vs 3.6×). Haas (59) provides a possible explanation for this. MPN is based on a maximum likelihood estimate assuming a Poisson distribution of cells, where each cell is disaggregated and uniformly diluted at each stage in the dilution series. This assumption is imperfect. The Poisson is a special case of the negative binomial distribution where there is no overdispersion. However, overdispersion can occur if bacteria form clumps with different numbers of bacteria and if those clumps disaggregate during handling. It is reasonable that the aggregation characteristics of the mixed sample are different from the single inoculation samples. In fact, enumerating bacteria from soil slurries violates the assumptions behind both the CFU and MPN methods and affects the accuracy of CFU counts as well (60). For screening, this is of less importance since it is the relative reduction not the absolute quantification that is most important.

We increased the complexity of the unamended soil microcosm by adding synthetic bacterial communities and then evaluated the performance of the fluorMPN assay. For four different communities, we estimated the concentration of Ralstonia across multiple wells of soil, which served as technical replicates (see clustering of data points in Fig. 3B). For each well, a single count was generated using both methods, such that each well served as a replicate of the count method. A treatment-agnostic F-test of the pooled data indicated that the overall variance (across all the sampled wells) did not differ based on the method of enumeration. In other words, regardless of the method used to generate the count data, the amount of biological variability recovered across wells for all communities was the same. Based on the equal variances of the count data, we infer that the precision of these methods is the same, even though the absolute mean counts are slightly different for each method.

To evaluate whether the fluorMPN assay recovered the effects of a given treatment on a population of interest, we compared the suppressive action of a community to data obtained by spotting on agar (Fig. 5). Out of the four communities tested, three displayed inhibitory action against Ralstonia, while one had no effect. The suppressive trends of a community and their relative values of inhibition were conserved between the fluorMPN assay and spotting on agar. This suggests that the fluorMPN microplate assay can be applied to projects profiling treatments in soil environments. Moreover, because the level of biological variation recovered by both methods was the same, we conclude that the fluorMPN microplate assay is generally suitable for hypothesis testing in the same contexts as spotting on agar.

Validation of plasmid stability under experimental conditions

Antibiotic selection is often used for the long-term maintenance of plasmids, but it is undesirable in experimental evaluations. Therefore, plasmid retention in R. solanacearum was tested in the absence of gentamicin in both soil systems. Results for unamended soil microcosms demonstrated that plasmid pFLxR5 was retained by the bacterial population for 3 days in soil without the addition of antibiotics. In the presence of a synthetic community, however, we noticed a minor decrease in the mean counts of colonies that grew on agar plates with gentamicin versus no gentamicin control plates. Although this change was statistically significant, biologically, a 1.1-fold change is a negligible difference in cell counts, suggesting that Ralstonia does not lose plasmid pFLxR5 under the evaluated conditions. Similarly, other studies have shown plasmid retention and reporter expression within bacterial systems in the absence of selective pressure (61–65). Specifically, a study with Ralstonia showed that under replication-favorable conditions or in the presence of environmental challenges (e.g., broth, in-planta, and soil), a plasmid was retained with a detectable expression of the fluorescent reporter (62). Recognizing that we cannot rule out the possible loss of plasmid pFLxR5 beyond 72 hours of soil incubation, these experiments could be performed with other vectors for which their mechanism of stability has been profiled in longer timeframes under different environmental conditions (62, 63, 66, 67). Alternatively, promoter-reporter sequences can be inserted into the genome of the bacterium. However, chromosomal insertions can result in a lower fluorescent signal due to a single copy of the inserted gene as compared to a multicopy plasmid, unless specific promoters are used to enhance expression (68).

Impact of soil particulates

The particulate nature of soil poses challenges for the culture-based enumeration of microorganisms. Both fluorMPN and spotting rely on serial dilution to ensure that microorganisms are uniformly suspended in liquid. However, the adhesion of microbes to particles—and the particles themselves—may interfere with this. Additionally, soil particles scatter the excitation light in undiluted samples, making the first several dilutions unreadable, but this is generally not a problem if the microbe of interest has a concentration of more than about 10² MPN/g. To circumvent issues associated with soil particles, samples could be filtered before serially diluting. However, this option seems unlikely for projects that require considerable sample screening. Alternatively, after mixing a sample, centrifugation of the deep well plate at low speed and for a short period would bring large soil particles to the bottom.

Beyond soils, further automation, and concluding remarks

The applications of the fluorMPN assay and MicroMPN software extend beyond our soil system. It can be used for high-throughput screening of chemical libraries, metabolites, probiotics, water and food microbial contaminants, in vitro mammalian cell infections, microbial stress responses, and many other conditions. In our experiments, each 96-well microplate contained only a single sample, with 12 dilution factors and 8 replicates; however, users can customize the number of samples per 96- and 384-well plate. To improve the automation of fluorMPN, several of its steps can be scaled up using liquid handling robotics. Our work does not include specific protocols for liquid handling of the dilutions required to fully automate screening. Our experience working with several liquid handlers is that these are highly platform-dependent. Fortunately, one of the most common uses of a liquid handler is a dilution series. Thus, this basic operation is usually well documented in the literature for each specific platform.

In conclusion, this work provides detailed protocols for screening microbial suppression in complex microcosms software to automate experimental analysis and empirical validation of the method with a soil pathogen. These resources should simplify the process of bacterial antagonist-based pathogen screening in complex environments and provide treatment leads for further evaluation and optimization.

MATERIALS AND METHODS

Growing conditions of Ralstonia solanacearum

Ralstonia solanacearum strain Rs5 (alternate strain name: UW576) phylotype IIA-sequevar 7 from the southeastern United States was used as our model agricultural pathogen (69, 70). To grow wild-type Rs5, a casamino acid–peptone–glucose (CPG) (71) medium was used. Cultures of fluorescently tagged strain Rs5 (transformation detailed below) were grown in CPG supplemented with 10 mg/L of gentamicin (CPG GEN-10). Ralstonia strains were grown at 30°C.

Construction of inducible RFP-expressing Ralstonia solanacearum

The incorporation of fluorescence in combination with MPN was adapted from reference (32). In our experimental design, fluorescence expression was regulated by an inducer. To make R. solanacearum tractable in soil, strain Rs5 was transformed with an inducible vector expressing a red fluorescent protein, following a previously described protocol (72), albeit with a few modifications (Text S4). Ralstonia was transformed with vector pFLxR5 (Plasmid ID 149482, Addgene, USA), which has the antibiotic marker GEN and a LuxR/P_LuxB promoter-regulator pair regulating an mRFP cassette [a detailed description of vector design and assembly can be found elsewhere (73)]. Throughout the manuscript, the terms strain rfp-Rs5 or Ralstonia are used interchangeably when referring to the transformed strain.

During MPN quantification, the expression of RFP was induced in strain rfp-Rs5 by the addition of N-(3-oxohexanoyl) homoserine lactone (OC6) (MilliporeSigma, USA) to media at a working concentration of 10.2 µM; the inducer was resuspended in the solvent dimethylformamide. To prevent loss of plasmid throughout incubation in CPG broth and agar, 10 mg/L of gentamicin (GEN-10) was always included in the medium preparation.

Assembling synthetic bacterial communities

Six communities (C1–C6), each composed of 10 strains, were evaluated for their biocontrol activity against R. solanacearum in soil microcosms. Strains used to assemble bacterial communities were randomly selected from a pool of 42 isolates. Lennox broth, supplemented with 20 mM glucose, was used to grow 18 strains isolated from clay soil. The rest of the strains were obtained from the NRRL Agricultural Research Service Collection (USDA, Peoria, IL) and were grown in a tryptone–yeast extract–glucose medium (5 g tryptone, 5 g yeast extract, 1 g K₂HPO₄, and 1 g of glucose per liter of deionized water). All strains were grown at 30°C.

To assemble a community, strains were grown individually in a 2 mL deep well plate under shaking conditions for 16 hours. Grown cultures were centrifuged at 4,000 rpm for 30 min at room temperature (RT). Next, supernatants were aspirated, and cell pellets were resuspended in 600 µL of 10% glycerol diluted in 0.5% NaCl. After resuspension, OD₆₂₀ measurements for each strain were recorded using a SpectraMax M3 multimode microplate reader (Molecular Devices LLC, USA) and used as a proxy of their concentrations before assembly (Table S1). Each community was assembled by pooling 500 µL of each strain.

Soil microcosms

The clay soil used in this study to assemble soil microcosms—unamended soils and soils inoculated with synthetic bacterial communities—was retrieved in 2021 from the University of Florida, North Florida Research and Education Center (Quincy, FL, USA). The soil was collected from four sections of a plot (30.551302°N, −84.599687°W) at a depth of 15 cm and combined as one sample. The soil was stored in sealed containers at 4°C for 1 year. For the creation of the soil microcosms, the soil was mixed with a trowel, then transferred to an autoclaved container, and air-dried at RT for 6 days. Finalizing day 6 of drying, we macerated soil with a mortar and pestle, followed by loading a 2 mL Eppendorf 96-deep-well plate (Thermo Fisher Scientific, USA). Early morning the proceeding day, plates were inoculated with the treatments detailed below. Inoculated systems were kept in a Percival incubator in the dark at 30°C ± 2°C and 80% ± 5% relative humidity for a limited time.

Quantifying strain rfp-Rs5 in soil microcosms following the MicroMPN workflow

Unamended soil microcosms

In previous studies, MPN fluorescence microplate assays have been used for surveying the concentration of microbes in environmental or food samples (33, 74). Here, we validate this approach by inoculating soil microcosms with four known concentrations of fluorescently labeled Ralstonia. Three colonies of the bacterium were inoculated in 6 mL of CPG GEN-10 and grown overnight for 16 hours at 250 rpm at 30°C. The overnight culture was centrifuged for 30 min at 4,000 rpm, then resuspended in 6 mL of 0.5% NaCl. After resuspension, three aliquots were further 10-fold diluted in 0.5% NaCl, creating four different concentrations of Ralstonia: undiluted (i.e., 10⁰), 10⁻¹, 10⁻², and 10⁻³. Ralstonia treatments were applied to soil in triplicate using 150 µL of inoculum. To obtain the concentration of Ralstonia from the microcosms, soil samples were processed 72 hours post-inoculation.

Synthetic community microcosms

We extended the fluorMPN assay beyond environmental surveying to evaluate whether engineered bacterial communities significantly reduced the concentration of Ralstonia in soil. Pre-assembled bacterial communities were thawed at 30°C, and 150 µL from each community was used to inoculate six independent wells of soil. An additional six wells reserved for rfp-Rs5 controls were pre-treated with 150 µL of 10% glycerol diluted in 0.5% NaCl. Community-inoculated and control wells were incubated for 120 hours. An overnight culture of Ralstonia was prepared as described in the previous section, except the culture was not further diluted before adding to the soil. As before, an inoculum volume of 150 µL was also applied to each soil well. The samples were further incubated for 72 hours before terminating the experiments. Data for C1–C3 and C4–C6 and their corresponding rfp-Rs5 control were collected on separate days.

Processing of microcosms

To obtain the concentration of Ralstonia from the microcosms, 1 mL of autoclaved de-ionized water was added to each well of a deep well plate. The plates were then sealed with an Eppendorf silicone mat (VWR International LLC, USA) and manually shaken for 2 min. From each soil well, samples were split as shown in Fig S2.

For the fluorMPN component, from every soil well, eight aliquots of 100 µL were transferred to the first column of a 96 clear-bottom black-well plate (Thermo Fisher Scientific, USA). In our plate layout, each row (A–H) was assigned a replicate value (1–8), while each column (1–12) represented a particular dilution factor (1e00–1e−11). Sequential 10-fold serial dilutions were performed in CPG broth supplemented with the following components: 10 mg GEN, 50 mg 2,3,5-triphenyl tetrazolium chloride (TZC), 25 mg bacitracin (BAC), 0.5 mg penicillin (PEN), and 100 mg cycloheximide (CHX), per liter of CPG broth; mass quantities were derived from the modified semi-selective SMSA medium recipe (75). Inducer OC6 was added to the media at a final concentration of 10.2 µM. Diluted samples were incubated at 30°C under shaking conditions. To prevent the drying of wells, additional media lacking OC6 were supplemented at the 24-hour mark of incubation. Fluorescence data were collected at T0 and T48. Measurements of RFU were used as an indicator of the presence of Ralstonia in a well. The SpectraMax M3 microplate reader was set to an excitation wavelength of 571 nm and emission at 620 nm, with a 5-second shaking step before measurement. RFU readings were taken from the top of a black-well plate using the plate adapter provided by the machine. For each microplate, wells were considered RFP-positive if RFU (T48) was greater than RFU (T0) at the five-sigma level (µ_T0 + 5σ_T0). This value was used as our user-supplied cutoff value to estimate MPN values using our command-line tool MicroMPN. We have included a Jupyter Notebook “Determining Threshold” in our GitHub repository (https://github.com/USDA-ARS-GBRU/MicroMPN_data-files) that illustrates the determination of a five-sigma threshold from the background (42).

MPN estimations were then generated with MicroMPN. For MPN data, a single count datum corresponds to RFU data collected from a single microplate. The final calculated MPN value is based on the volume of a sample used, not solely on the dilution step. Therefore, we adjusted the MPN estimate by a factor of 10 to account for the difference in volume and normalized it by the averaged mass of soil per well of a 96-well deep well plate. We further log₁₀-transformed the MPN estimate before any statistical analysis.

For the spotting component, a single aliquot of 100 µL of resuspended soil was 10-fold serially diluted in 1× phosphate-buffered saline (Thermo Fisher Scientific, USA). Diluted samples were spotted in triplicate on supplemented (GEN, TZC, BAC, PEN, or CHX) CPG agar without an inducer. CFU data were collected by 48 hours post-spotting. For CFU data, a single count datum corresponds to averaged replicate counts originating from a single well of soil.

For step-by-step protocols on the methods described in this article, visit protocols.io. The two protocols we created were “Most Probable Number Fluorescence Microplate Assay” (dx.doi.org/10.17504/protocols.io.q26g7yqk1gwz/v1) and “MicroMPN: Software for Automating Most Probable Number Estimates from Laboratory Microplates” (dx.doi.org/10.17504/protocols.io.81wgbymenvpk/v1).

Statistical analysis

All statistical analyses were performed in R v4.3.0 (76). All plots were graphed using Microsoft Excel (v2304). The strength and relationship between the fluorMPN assay and spotting on agar were evaluated with a Pearson correlation and linear regression analysis. For the former, we used the “cor.test” function, while for the latter, the “lm” function was used; both functions are from the “stats” package (76). The assumptions of skewness, kurtosis, and heteroscedasticity for the linear regression model were evaluated with the “gvlma” function of the “gvlma” package (77).

We used the Bland–Altman (BA) plot and analysis to display and assess the degree of agreement between count measurements obtained using the two methods of quantification (78–80). The BA plot is a scatter plot that displays the differences (CFU – MPN) and means ((CFU + MPN)/2) for every paired count. The level of agreement between the two methods is then evaluated by calculating the mean of the differences [i.e., mean bias ($\overline{d}$)] and SD associated with the differences. Additionally, 95% limits of agreement are calculated around the bias using the equation $\overline{d}$ ± 1.96 × SD. The bias and limits of agreement were calculated with the function “agree_test” from the “SimplyAgree” package (81). The assumption of normality was evaluated with a Shapiro–Wilk test based on the pairwise differences between the methods using the R function “shapiro.test.” The assumption of homoscedasticity was evaluated with a Breusch–Pagan test around the bias. The confidence intervals (95%) for the bias and limits of agreement were calculated as described in Haghayegh et al. (80). The limits of agreement contain 95% of the variation between the two methods (78, 80). If no bias exists between the methods, then the average of the differences should equal zero. However, if bias does exist, then the sign of the bias value indicates whether a method is over- or underestimating the counts relative to their average. To compare the precision of the fluorMPN assay to spotting on agar, we evaluated the differences in variances between methods by running Fisher’s two-sided F-test with the “var.test” function from the “stats” package (76).

Finally, to assess if the community suppression was independent of the enumeration method used, we compared the mean of every community to the mean of the control (rfp-Rs5). Differences between group means were compared by a one-way ANOVA [function “aov” from “stats” package (76)] followed by post hoc Dunnett’s test [function “DunnettTest” from “DescTools” package (82)]. The assumption of normality was assessed with a Shapiro–Wilk test [function “shapiro.test” from “stats” package (76)], while the assumption of homoscedasticity was evaluated with Levene’s test [function “leveneTest” from “car” package (83)]. Since the assumption of normality was violated for spotting on agar, we ran an additional non-parametric Kruskal–Wallis test [function “kruskal.test from “stats” package (76)], followed by post hoc Dunn’s test [function “dunnTest” from “FSA” package (84)] for multiple comparisons (Text S5). The results from the parametric, non-parametric, and post hoc tests for spotting on agar were the same independently of the statistical analysis performed.

DATA AVAILABILITY

The raw and normalized MPN and CFU values, R scripts, and Excel files used here are available via GitHub (https://github.com/USDA-ARS-GBRU/MicroMPN_data-files). The software is available for download from GitHub (https://github.com/USDA-ARS-GBRU/micrompn) and the Python Package Index (https://pypi.org/project/micrompn/) (42). The software is registered with bio.tools under the identifier (biotools:micrompn) (85).

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/spectrum.03578-23.

Supplemental material. spectrum.03578-23-s0001.docx.

Text S1 to S5, Fig. S1 and S2, and Table S1.

OPEN PEER REVIEW. reviewer-comments.pdf.

An accounting of the reviewer comments and feedback.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.