Summary of known Ant⁻ strains and mutations found

In addition to the previously identified nonsense mutations in the antC biosynthetic operon, in silico analysis of all B. anthracis genomes in GenBank identified a novel nonsense mutation SNP present in the anthrose biosynthetic gene antB of strain Ba4599 [50]. This strain was associated with an anthrax outbreak among injection heroin users in Europe [55,56] (Fig 1B). Additionally, we identified 2 strains in our collection, one from Chile and one from Poland, having neither the wild-type nor mutant SNP. They also failed to produce a PCR product when the anthrose biosynthetic genes were targeted for standard PCR fragment analysis, indicating an absence of these genes. The phylogenetic relationships of these strains to the major clades of B. anthracis are depicted in Fig 1C, and their source countries are depicted in Fig 1D.

In silico simulation of anthrose operon inactivation

Computer simulation studies were performed to assess possibilities of anthrax mutations over time in the anthrose biosynthetic operon. The goals of the simulation studies were (1) to investigate probabilities of mutation within the coding region of the anthrose biosynthetic operon based on simulated mutation rates extrapolated from previously published molecular clock studies and observed wildlife outbreak frequencies; and (2) to investigate probabilities of whether conversion of Ant⁺ anthrax to Ant⁻ anthrax over a prolonged period of time is solely due to natural mutation or if there is a recent yet unknown selective pressure for Ant⁻ B. anthracis.

Since B. anthracis is a spore former, it has 2 very different states. As a spore, it does not multiply and remains dormant in the environment, which can last for a long period of time (years or decades). During vegetative growth, it actively replicates inside the host, which is a relatively short period of time. The number of generations per year is highly variable and is associated with the number of host infections. During enzootic years, host infections are sporadic, but during epizootic years, the number of host infections increases [57]. To capture this feature of anthrax disease in our simulation, 2 different mutation probabilities were incorporated into the simulations: one for enzootic years and one for epizootic years. Within the simulation, every seventh year was assumed to be a year of epizootic anthrax outbreak with much higher mutation, insertion, and deletion probabilities due to increased bacterial growth during years of epizootic outbreaks (Fig 1E).

Estimates of the proportion of wild-type B. anthracis with the 8-bp AAAAAAAG doublet to triplet nonsense mutation in antC by the end of the simulation period are shown in Fig 1F, and the estimates are summarized in S1 Table. Data interpretation for scenario 1 would be as follows: The median estimated proportion of Ant⁻ bacteria is 0.0031 with the 95% confidence interval equal to (0.0021; 0.0047). It follows from Fig 1F and S1 Table that when utilizing mutation probability values documented in the literature, only a small proportion of bacteria 0.0031 (95% CI (0.0021; 0.0047)) is expected to become Ant⁻ by simple sequence repeat (SSR) mutations in antC over the simulated time of 600 years. Even if mutation probability increased 100- or 1,000-fold, which is unlikely in the natural environment without the addition of a driving evolutionary pressure, then the proportion of Ant⁻ bacteria would only increase to 0.1394 (95% CI (0.1327; 0.1463)) and to 0.3193 (95% CI (0.3102; 0.3279)), respectively (Fig 1F). In conclusion, the naturally acquired mutation observed in WAG B. anthracis leading to Ant⁻ mutations are unlikely when using the documented probabilities even though the highest probability of SSR mutations was included in scenario 1. If there was a 100- (scenario 2) or 1,000-fold (scenario 10) increase in anthrax mutation probabilities due to some unexplained ecological selective pressures, beyond the already liberal probabilities utilized in the scenario 1 simulation, the naturally observed SSR becomes more likely (sc02 to sc10; Fig 1G).

Fig 1G and S2 Table provide summaries of the estimated mean population Hamming [58] distance across 10,000 population elements at the end of the simulation period. After being submitted to the triplet mutational round, the sequences were submitted to point mutation effects, and the Hamming distance was used to count the number of different characters between the 2 aligned DNA sequences (i.e., number of base pairs different after the 600-year simulation). The sequences at the end of the simulations were compared with the input sequences to compute the Hamming distances of interest. Estimates in Fig 1G and S2 Table refer to averages over the entire population for each of 1,000 runs and corresponding quantiles across 1,000 runs (each run contains 10,000 DNA sequences). For comparison, the original wild-type fragment length input for the simulation study consisted of 2,000 bp encompassing the region of the anthrose operon where natural mutations are observed. At the 50th percentile of scenario 01 (sc01), which utilized the most liberal baseline mutational probability approximations from the literature, only 0.0076 bp over the 2,000 bp input would be different over the 600-year simulation. The data suggest that, for probabilities documented in the literature and used in sc01 of the simulation study, overall mutation in the anthrose biosynthetic operon is not expected to be observed over the considered simulation length. If we add multipliers stepwise through the scenarios, we approach observations of the bp differences observed in Ant⁻ isolates from West Africa. An alternative, but unlikely, explanation is that pressure for mutations in the anthrose operon have been occurring for a much longer period of time than the 600-year simulation that was run. This then begs the question, why are we finding numerous anthrose-deficient strains now, and why at several locations around the world despite the slow evolutionary rate of B. anthracis?

Genomic sequencing of Bacillus anthracis strains from Chile and Poland and verification of large genomic deletions

Illumina sequencing was performed, and reads were aligned to B. anthracis Ames. Single large gaps in the read alignment including the ant operon and surrounding areas were observed for UF00242 and UF00910 (Fig 2A and 2B). Following de novo assembly with SPAdes, contigs in both contained a spliced region coinciding with the boundaries of the deletion identified by read alignments, confirming the read alignment data. In silico MLVA typing of the Chilean B. anthracis strain verified presence in the A.Br003/004 lineage by MLVA-15, similar to other strains from Argentina. This strain contained a 19,108-bp chromosomal deletion when compared to B. anthracis Ames. The Polish strain was previously typed as being in group B [59], and in silico MLVA typing verified presence in the B.Br.CNEVA lineage. This strain contained a 59,157-bp deletion centered around the ant operon. With strain Ba4599 in the group A1.a lineage, the Chilean strains in the A.Br003/004 lineage, the Polish strain in the B.Br.CNEVA lineage, and the previously published WAG strains in the A_β/E clade [3,5,6], we have now identified several genetically and geographically distant Ant⁻ strains (Fig 1C and 1D). Primers were designed to amplify small products across the deleted regions present in Chile (UF00242) and Poland (UF00910) Ant⁻ strains (Fig 2C). These primer pairs would give prohibitively large products if the strains had the wild-type genomic material. Primers 2 and 3 were designed to produce an approximately 450-bp fragment present only in the Chilean Ant⁻ isolate. PCR of Ames, UF00242, and UF00910 using primers 2 and 3 resulted in positive PCR product only in the Chilean UF00242 strain, verifying the 19,108-bp deletion found in the de novo sequence and the alignments (Fig 2D). Primers 1 and 4 were designed to produce an approximately 900-bp fragment present only in the Polish strain UF00910. PCR of the strains Ames, UF00242, and UF00910 using primers 1 and 4 resulted in positive PCR product only in the Polish UF00910 strain, further verifying the 59,157-bp deletion found in the de novo sequence assembly and the alignments (Fig 2E).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Genome sequencing reveals 2 unique chromosomal deletions encompassing the anthrose operon.

(A) Illumina sequence alignment of strain UF00242 and (B) UF00910 to the Ba Ames antABCD region. The approximate size of the deletions is shown. (C) The exact deletion sizes shown in relation to the Ba chromosome with primer sites used to verify deletions indicated by arrows. (D) Agarose gel showing the approximately 450-bp fragment generated from the Chile strain (UF00242) and not in Ames or Polish strain UF00910. (E) Agarose gel showing the approximately 900-bp fragment generated from the Poland strain (UF00910) and not in Ames or Chilean strain UF00242.

https://doi.org/10.1371/journal.pbio.3001052.g002

Creation of anthrose mutant in Ba Sterne and analysis of anthrose content in Ba Sterne

Although we have Ant⁻ mutant wild strains, the heterologous strain backgrounds minimize their utility in disentangling the downstream results of anthrose loss, so we removed anthrose from the B. anthracis Sterne vaccine strain. Sterne lacks pXO2 and so is Cap⁻. The poly-D-glutamic acid capsule is produced by wild-type vegetative bacteria and is required for full virulence but is not involved in sporulation [60,61]. B. anthracis Sterne sporulates and is Cap⁻Tox⁺, causing a localized infection at the site of inoculation and toxemia-induced mortality in susceptible mice compared to disseminated bacterial growth following infection observed with Cap⁺Tox^- B. anthracis [62,63]. Cap⁺Tox⁺ B.anthracis generate mortality via toxemia and disseminated bacterial growth. Sterne can cause anthrax disease including lethal factor (LF) and edema factor (EF) toxemia in complement-deficient mouse strains, representing a useful model for understanding the role of anthrose in onset of toxemia-induced mortality [63]. We first created the ΔantC mutant by allelic replacement and verified the deletion by PCR (S1A Fig). The absence of anthrose in the spore was verified by gas chromatography–mass spectrometry (GC–MS) retention time comparison (S1B Fig). The unique mass spectrum from the compounds at the observed retention time peaks (S1D Fig) allowed us to propose the structure of the trimethylsilyl (TMS)-derivatized anthrose observed by GC–MS (S1C Fig). The retention time peak doublets at approximately 35 minutes were only present in the pure anthrose and Sterne spore preparations. These peaks corresponded to a unique mass spectrum not in the standard NIST databases. These spectra were not found in the ΔantC spore sample verifying functional removal of anthrose in the ΔantC spore.

Physiological and molecular pathogenic characterization of the Bacillus anthracis Sterne anthrose deletional mutant

Growth of the strains was carried out in 96 well plates. Sterne 34F2 and the ΔantC mutant grew at similar rates and to similar final densities. The complemented mutant exhibited a slight lag most likely due to expression from the plasmid based antC. Eventually, all strains reached the same final culture density. Normally, anthrose is produced during late logarithmic growth immediately prior to sporulation, during stage II of the sporulation associated expression wave [28,61]. Artificial expression of ant genes from the complementation plasmid throughout lag and log phase growth may dysregulate the natural growth pathways of B. anthracis.

Germination is an important physiological step for spore formers. Spore surface molecules, such as anthrose, may affect exosporium permeability to germinants or allow close interaction of germinant signals and exosporium. The germination rates of spores in response to heat and the potent germinant inosine were measured as a decrease in optical density over time. The Sterne vaccine isolate and complement strain had the same germination rates and final spore germination completion levels in response to the germinant inosine (Fig 3A). The ΔantC mutant had a slower germination rate and lower percentage of cells successfully germinating in the presence of both inosine and L-alanine, germinants targeting 2 different germination pathways. Sporulation of the ΔantC mutant was tested by comparing heat-treating or ethanol treating to untreated aliquots grown in culture. The ΔantC mutant produced more heat resistant cells, in terms of total number and relative percentage of total cells, at 48 hours with the wild-type and complement catching up by 72 hours (Fig 3B, top panel). The ΔantC mutant produced more ethanol resistant cells at 24 hours (total number and relative percentage of total cells) and 72 hours (relative percentage of total cells) than the wild-type (Fig 3B, bottom panel). Spore association percentages in non-phagocytic A549 human lung epithelial cells were not significantly different (Fig 3C). However, phagocytic RAW264.7 murine macrophages were associated with Sterne ΔantC spores at a significantly higher level than wild-type spores and ΔantC complement spores (Fig 3C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Anthrose deficiency affects B. anthracis physiology and the pathogenesis of anthrax.

(A) Ba germination in response to 1 mM inosine as germinant (top) and 60 mM L-alanine (bottom). The experiment was carried out in triplicate for each strain. The OD₆₀₀ read every 2 minutes for 1 hour, and data are presented as percent of starting OD₆₀₀. The symbols represent the mean, and the vertical bars are the standard deviation of the mean at the time points. (B) Sporulation of WT Sterne compared to the ΔantC and ΔantC/COMP strain as indicated after growth in modified G sporulation media at 24, 48, or 72 hours. The standard error of the mean of 3 heat treatments (top panel) or 6 ethanol treatments per strain are shown (bottom panel). (C) Spore internalization assay using human lung epithelial cell line A549 and mouse macrophage cell line RAW264.7 after infection of spores at an MOI of 10:1. Numbers are internalized spores as a percentage of total spores determined by dilution plating of inoculum and cell lysates. Then the percentages were normalized to WT Ba Sterne to obtain relative internalization %. (D) Survival of A/J following intranasal instillation of A/J mice with 10⁶ spores of the indicated strain. (E) In a predetermined endpoint study, mice were infected by the subcutaneous route with 1,000 spores, and spleens were collected 48 hours postinfection. Each dot represents data from a single mouse with mean and standard deviation shown. p-Values were calculated by comparing Sterne (blue dots) to ΔantC (red dots) and ΔantC/COMP (green dots). Splenic bacterial burdens were low, but the levels of ΔantC mutant were significantly higher than wild-type and the complemented mutant as calculated by 1-way ANOVA. (F) Total bacterial organ burden of each strain from 8/10 mice challenged with 2 × 10⁷ total spores from the CI study. The means and SEM are shown with significance tested by the Student t test. (G) CI values were log transformed, and the standard deviation is shown. The Wilcoxon signed rank test was used to determine the significance of CI difference from 1. *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001; ****p ≤ 0.0001; ns = not significant. The data underlying Fig 3A–3G can be found in S1 Data. CI, competitive index; IN, intranasal challenge; MOI, multiplicity of infection; SC, subcutaneous challenge; WT, wild-type.

https://doi.org/10.1371/journal.pbio.3001052.g003

LD₅₀ survival, organ CFU, competitive dissemination index, and Biothrax vaccination studies in an anthrax mouse model

Purified spore colony-forming units (CFUs) were determined and then used to perform an LD₅₀ study, survival study, and splenic bacterial organ loading A/J mice challenged via the intranasal or subcutaneous route (S2 Fig). Preparations of spores were visualized under phase contrast microscopy and showed that there was no obvious clumping of spores (S3A Fig). Our results were similar to previously published LD₅₀ of Sterne in this model [60]: 60% survival after challenge with wild-type Sterne was observed, and only 40% after challenge with the ΔantC mutant (a 20% difference following subcutaneous or intranasal challenge Figs 3D and 4A). Probit regression analysis was used to calculate 7 and 14-day LD₅₀ doses (Table 1). At 7 days, the subcutaneous LD₅₀ was calculated as 1,810 spores for the wild-type Sterne and 971 spores for the ΔantC anthrose mutant, a 53.5% reduction in spore LD₅₀. The Sterne ΔantC complement had a 7-day LD₅₀ at 3,171 spores. The 14-day LD₅₀ remained the same for wild-type and ΔantC mutant but decreased to 986 spores for the ΔantC complement. Mice were more resistant to intranasal challenge but resulted in similar differences in LD₅₀. This indicates an increase in mortality and increase in pathogenicity during the early stages of infection with Ant⁻ Sterne isolate. The harmonic mean TTD of 1,000 ΔantC mutant spore-challenged A/J mice was 2.76 (66.24 hours) or 2.6 days (62.4 hours) sooner than either the Sterne wild-type or the ΔantC mutant complement, respectively. Following subcutaneous inoculation between the shoulders, symptoms such as severe draining lymph node and facial edema occurred markedly sooner in ΔantC challenged mice than in wild-type or complement-challenged mice. Upon dissection of mice euthanized 96 hours after subcutaneous challenge, subcutaneous hemorrhaging was more widespread following ΔantC spore challenge. Spleens from groups of 10 mice subcutaneously infected with 1,000 spores of the wild-type Sterne, ΔantC, or ΔantC/COMP were removed and homogenized, and total bacterial CFUs counts were determined (Fig 3E). Overall, the splenic burden was low, as described elsewhere [63]. However, there were significantly more ΔantC mutant bacteria present in the spleens of infected animals compared to wild type or complement (approximately 6 times more bacteria). Only 1/10 mice infected with wild-type Sterne had measurable bacterial burden while 6/10 infected with ΔantC Sterne did and 2/10 for the ΔantC/COMP. Even though mortality in the A/J mouse model occurs primarily by toxemia caused by a localized infection, we found decreased LD₅₀, decreased TTD, and earlier onset of symptoms was linked to increased bacterial dissemination of the ΔantC mutant. The faster onset of lymph node swelling and increased subcutaneous hemorrhaging support this notion.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Alternate animal models and pure death process simulations coalesce at increased sample sizes.

(A) Subcutaneous infection of A/J mice with 1,000 spores of the indicated strains shows the consistent faster TTD and increased lethality of the ΔantC mutant. Data are from 3 independent experiments and 15 challenge mice. (B) Survival of G. mellonella infected with 10⁵ spores of Ba ΔantC spores (red squares and line) was significantly different than the Ba Sterne wild-type challenged larvae (blue circles and line) as determined by the Mantel–Cox log-rank test. **p < 0.005; *p < 0.05. (C) Plot of the average estimated loss rates in the 1,000 CFU subcutaneous mouse challenge for each strain under the “Separate dynamics model.” (D) Plot of the average estimated loss rates in the 10⁵ CFU G. mellonela injection challenge for each strain under the “Separate dynamics model.” (E) Simulation of 1,000 CFU dynamics using small sample sizes over the course of a typical animal. (F) Simulation of G. mellonella death dynamics using small sample sizes. (G) Number of animals per replicate versus the proportion of correctly calling the “Separate dynamics model” using the mouse dynamics (black circles) and the larva dynamics (white circles). The data underlying Fig 4A–4G can be found in S1 Data. CFU, colony-forming unit; SC, subcutaneous challenge; TTD, time to death.

https://doi.org/10.1371/journal.pbio.3001052.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Probit LD₅₀ calculations and harmonic mean TTD.

https://doi.org/10.1371/journal.pbio.3001052.t001

A competitive index (CI) study was performed to ascertain if loss of anthrose led to enhanced dissemination of the ΔantC mutant. Twenty-four hours after subcutaneous injection of 2 × 10⁷ spores (a mixture of the ΔantC mutant and its kanamycin-resistant ΔantC/COMP complement) in 10 mice, the pooled organs were homogenized, and total bacterial organ burden and kanamycin-resistant bacterial organ burden were determined. There were significantly higher levels of ΔantC mutant bacteria in the organs compared to ΔantC/COMP (Fig 3F). In this CI study, a CI > 1 indicates that higher levels of ΔantC mutant are present, while a CI < 1 indicates that more ΔantC/COMP are present within the organs of each animal, removing confounding effects host variability may have on the bacteria spores with different genetics. The in vitro mean CI was 1.4 (1.22 to 1.58), while the in vivo mean CI was 35.20 (−0.18 to 70.57) (Fig 3G). The Wilcoxon Signed Rank Test found that the in vitro CI data were not significantly different than 1, while the in vivo CI data, though wider ranging, were considered significantly different than 1. This suggests that the loss of anthrose allows the dissemination of more ΔantC mutant bacteria in the same host animal. Two of the 10 mice had no detectable bacteria recovered at a detection limit of 100 CFU. One of 10 mice had a CI less than 1, while the other 6 had CIs ranging from 3.799 to 129.4.

Lastly, we wanted to see if loss of anthrose from the spore would result in a reduction of vaccine efficacy in the Sterne A/J mouse model. Groups of 5 unvaccinated challenged mice and vaccinated challenged mice were observed for survival after 20 LD₅₀, 50 LD₅₀, and 100 LD₅₀ subcutaneous challenges (S3B Fig). At all doses, survival of vaccinated animals was 100% with little to no symptoms observed at any point after challenge. Unvaccinated mice succumbed quickly to infection (S3B Fig). These challenge doses cause vaccine failure in mice when using pXO1⁺/pXO2⁺ B. anthracis [64], indicating the necessity of using fully virulent B. anthracis strains in future studies with immunocompetent mice.

Pure death process fitting and simulations

Just like the log-rank tests, the pure death process model fitting and testing only weakly supported the hypothesis that death rates in mice challenged with Ant⁻spores were different than in those mice challenged Ant⁺ spores with anthrose due to overlapping confidence intervals (Fig 4A). However, as opposed to the log-rank test, which is designed to detect differences among trends and is based on a chi-squared approximation to the distribution of the test statistic [65], the death process modeling framework was easily amenable to showing that the lack of strength of the results was a small sample size issue, rather than an intrinsic property of the mortalities due to both the wild type and the mutant strain. Indeed, as Dennis and colleagues recently show in their extensive review of classic error statistics in science in a journal issue dedicated to statistical science, classical error statistics do not really compare multiple models simultaneously [66]. With Fisherian significance tests, a single model is tested, and if rejected, one of many plausible alternatives is inferred. Neyman–Pearson tests like the log-rank test appear to compare 2 models but are really significance tests constructed along the axis most likely to reject the “Null” if the “Alternative” is true. Significance tests are calculated as if the null hypothesis were true. Philosophically, this is awkward because as George Box wrote, “All models are wrong.” If so, one may question the gains of doing a significance test. If the test rejects the Null hypothesis, the researcher has not learned anything since it is known a priori that this model is wrong. If the test fails to reject, it only tells that the researcher does not have sufficient sample size or specified inappropriate effect size. Model comparison via consistent information criteria like the Bayesian Information Criterion (BIC) used here, coupled with process simulations, can and do remedy these problems [66,67].

In particular, after first estimating the mouse mortality rate using death process modeling (Fig 4C) under the hypothesis that the mutant dynamics was indeed faster than the wild-type, we simulated 1,000 realizations of the death process using these parameter estimates starting at varying initial sample sizes (i.e., the number of initially challenged mice). Thus, each 1,000 group of simulations contained 5, 10, 20, 40, or 80 challenged mice. One such simulation of small sample sizes is shown in Fig 4E using the mouse dynamics and in Fig 4F using the G. mellonella dynamics modeled in Fig 4D. For each simulation at each sample size, we tested using the BIC whether the “separate” or the “joint” dynamics model was better and recorded the proportion of times the “Separate dynamics model” was best, given it was the model used to simulate the data. In other words, for each sample size, we estimated the proportion of time the BIC correctly chose the Separate dynamics model as the best model. By doing so, we were able to infer desirable sample sizes needed to detect a higher death rate in the mutant strains when it actually is of the size estimated for the 1,000 CFUs assays. The proportion of time the BIC test correctly chose the Separate dynamics model (akin to a power curve) is plotted in Fig 4G.

Out of each set of simulations, we computed the proportion of cases where a significant difference using BIC was detected when it truly was there. That proportion is an estimate of the detection power of the test. We found that the true difference between survival of ΔantC and wild-type challenged mouse groups was correctly detected 6%, 13%, 57%, 92%, and 100% of the time for the initial sample sizes of 5, 10, 20, 40, and 80, respectively. These simulations strongly suggest that larger sample sizes would lead to significantly higher deaths in anthrose-negative B. anthracis infected animals (Fig 4G). As single epizootic outbreaks in wildlife and livestock are generally larger than the number of animals used in the mouse infection study (n = 15) or those in the infection simulation study (3 groups of 20, n = 60) [68], outbreaks of anthrose-deficient B. anthracis in the wild would yield significantly higher mortality rates than anthrose-positive strains when using these animal numbers as a baseline in a predictive epizootic model.

Infection of Galleria mellonella with anthrax spores

The death process analysis approach involved estimation from experimental values followed by power estimation via simulations. Using the estimates from the simulations, the study was repeated by experimentally infecting G. mellonella larvae with B. anthracis spores. We hypothesized that using the numbers generated by the power test using the mouse infection variables, anthrose would significantly impact infection in the G. mellonella model. Using the G. mellonella infection model, we found that a dose of 10⁵ ΔantC spores killed the larvae faster, producing significantly different survival curves (Fig 4B). The ΔantC mutant killed 100% of larvae by 56 hours postinfection, while approximately 22% of wild-type Sterne infected larvae survived to the 72-hour study endpoint. The dynamics generated in the G. mellonella challenge model were utilized in a down-sampling death process simulation to demonstrate the effects low sample size would have had on mortality observations without adequate sample size (Fig 4F).

From a biological standpoint, we consider the G. mellonella infection model suitable for our inferences because it has been used to characterize major B. anthracis Sterne virulence factors following vegetative cell challenge [69] and for down-selecting antibacterial peptides active against Ba for future use in an animal model [70]. It also allows challenge of desired animal numbers inexpensively and humanely while serving as an accurate innate immune animal model. G. mellonella possess several components mimicking the mammalian innate immune response. For example, phagocytic hemocytes and production of antimicrobial peptides and reactive oxygen species are triggered within the hemocoel in response to bacterial infections.

Bacterial strains, cell lines, and growth conditions

Escherichia coli DH5α was used as a cloning strain and grown in LB broth or agar at 37°C, and RHO3 was used as a mobilization strain and grown in the presence of 200 μg/ml of DAP (Sigma-Aldrich, USA; [78]). Kanamycin was used at 35 μg/ml in E. coli and 20 μg/ml in B. anthracis Sterne. Spectinomycin was used at 50 μg/ml in E. coli and 100 μg/ml in B. aanthracis. Conjugations were carried out as previously described [78,79]. B. anthracis Sterne 34F2 was obtained from Colorado Serum (USA) and grown with brain heart infusion (BHI) broth or agar at 37°C. Wild-type B. anthracis are from the Martin E. Hugh-Jones Bacillus anthracis Collection housed at the Emerging Pathogens Institute at the University of Florida. Bacteria were manipulated using BSL3 practices and procedures according to the Biosafety in Microbiological and Biomedical Laboratories (BMBL) in a CDC-/USDA-inspected and registered entity. A549 human lung epithelial cells and murine RAW264.7 macrophages were acquired from ATCC and maintained in DMEM+10% heat-inactivated FBS with antibiotic-antimycotic at 37°C and 5% CO₂.

Whole genome sequencing and bioinformatic analysis

gDNA was isolated using the QIAGEN DNeasy UltraClean Microbial Kit (Hilden, Germany), filter sterilized, and sterility verified according to internal procedures and CDC SAP guidance. gDNA was sheared, and libraries were produced with the NEBNext Ultra II Library Prep kit (NEB, Ipswich, Massachusetts, USA). The libraries were sequenced with the miSeq next-generation sequencing system (Illumina, San Diego, USA) at the UF EPI using v3 chemistry. Reads were quality controlled using standard methods, and SPAdes [80] was used for de novo genome assembly, while BWA-MEM [81] was used to align reads to B. anthracis Ames. Assembled genomes were analyzed using CLC Sequence Viewer 7.5.0 (QIAGEN), and alignments were viewed with Integrative Genomics Viewer version 2.5 (IGV; Broad Institute [82]). References sequences were obtained from GenBank. Whole genome SNP alignments were created with the PhAME software package developed at Los Alamos National Labs [83], and phylogenetic trees were generated with RAxML [84] within PhaME and visualized with iTOL [85]. BioSample accession numbers of the Illumina read files for UF00242 and UF00910 are SAMN15046868 and SAMN15046933 and have been deposited in BioProject PRJNA635616.

PCR and molecular methods

PCR was carried out using standard methods [86] and Q5 Pfu polymerase (NEB). Splicing by overlap extension (SOE) PCR was carried out by first amplifying 1 kb of each of the regions flanking the antC gene. These fragments were combined in a 15-cycle PCR without primers. A total of 5 μl of the initial SOE reaction was used as template with end primers in a 30-cycle PCR. The resulting approximately 1.9-kbp fragment was digested with BamHI and HindIII for further cloning as described below. PCR primers for genomic deletion verification were designed from the de novo assembled sequences of B. anthracis Sterne, UF0242, and UF0910.

Simulated mutational acquisition studies

The simulations were performed using the University of Florida Research Computing Environment and the HiPerGator 2.0 supercomputer cluster [87]. Simulations were performed in annual discrete time slots over the period of 600 years. Annual time slots were selected due to low mutation activity in shorter time slots (days, weeks, or months) and the predominantly dormant lifestyle of B. anthracis in the spore state. For this work, we focused on a 2,000-bp region of B. anthracis gDNA containing a segment of the anthrose biosynthetic operon known to accumulate nonsense mutations that will inactivate one of the biosynthetic genes and prevent deposition of anthrose onto the surface of the spore. Within the computer simulation, this piece of DNA was identically replicated 10,000 times in computer simulation to create a synthetic population. This synthetic population was investigated for mutations, insertions, and deletions for the period of 600 years. We initially observed for the AAAAAAAGAAAAAAAG SSR triplication event (AAAAAAAGAAAAAAAGAAAAAAAG creating a stop) in antC of WAG B. anthracis [4,6]. The anthrax mutation probability per generation for SSR mutations is much higher than individual base pair mutation probabilities due to the mechanisms of polymerase slippage during genome replication and has been calculated previously (between 10⁻⁴ and 10⁻⁵ for SSR compared to 5.2 × 10⁻¹⁰ for point mutations) [88,89]. Therefore, the SSR mutation probability per generation was always assumed to be 10⁴ higher than the individual base pair mutation probability used in the simulation, and those mutations were considered first at every time step of the simulation process before the individual base pair mutations. The individual base pair mutation probability value, denoted as λ, for B. anthracis was derived from the literature [59,88], and individual base pair mutations were considered second after SSR mutation within every time step of the simulation process. This mutation rate per generation was adjusted by the multiplier g resulting in mutation probability λ × g. This multiplier g is determined by the number of generations per unit of time within the simulation (i.e., year). Within the simulation, the mutations in each time slot are assumed to be independent from other time slots. The logic behind this probability comes from the assumption of the Poisson process for the mutation counts with independent time slots. This results in the expected number of mutations equal to λ × g which serves as a binary probability of a single mutation in the given time slot. Within the simulation study, every seventh year was approximated to be epizootic [57], and during epizootic years, 43 generations of B. anthracis were assumed to occur [59] with just a single generation per year for sporadic years. The resulted multiplier values were g = 43 and g = 1, respectively.

In the simulation study, 10 different scenarios were considered, and the probability parameter λ × g was used in baseline simulation scenario 01 (sc01) for mutations, insertions, and deletions of individual base pairs. For each subsequent scenario, the parameter λ × g was increased by multipliers 100 (sc02), 200 (sc03), 300 (sc04), …, 800 (sc09) and by multiplier 1,000 for the last 10th scenario (sc10). To produce stable estimates 1,000 simulation replications were generated for each of 10 scenarios. The mutation probabilities were assumed to be the same for wild-type and mutant B. anthracis. The simulations for each scenario were performed independently for each element of the synthetic population. The uncertainty estimates were summarized across 1,000 replications for each scenario, and summary statistics were computed for each simulation replication within each simulation scenario, and uncertainty estimates were presented in S1 and S2 Tables in the form of quantiles across these 1,000 replications.

In S1 and S2 Tables, the column with 0% corresponds to the smallest statistic value across 1,000 replications for the given scenario, while 100% corresponds to the largest. The column with 50% corresponds to the median value of the statistic of interest, while 2.4% and 97.60% percentiles correspond to the 95% confidence interval for the given statistic. Input sequences, documentation, code, analysis, and data for these simulations are available at https://github.com/akirpich-ap/anthrax-simulations.

Mutant generation and complementation

The anthrose-deficient Sterne strain was created using previously published methods [90]. Briefly, SOE PCR generated the ΔantC fragment for cloning into pRP1028 by BamHI/HindIII digest. The temperature-sensitive plasmid was transformed into RHO3 then mated into Sterne overnight at 30°C on BHI+DAP. Colonies containing the allelic replacement plasmid were streaked on BHI-Sp100 and incubated at 37°C to select for chromosomal integration. Red fluorescence was verified by viewing under fluorescent excitation, and colonies were restreaked. Mutants were resolved by introducing I-SceI-expressing plasmid pRP1099 into merodiploids and plating on BHI-Km20 as above using RHO3. Colonies no longer expressing RFP were PCR screened for mutation of antC then streaked on BHI to cure pRP1099. Deletion mutant colonies were resuspended in 95% ethanol and incubated for 1 hour to ensure that saved isolates retained the ability to sporulate after several rounds of culturing.

For complementation of the ΔantC mutant, the antC gene was amplified by PCR to add NheI and PstI sites, restriction digested with NheI and PstI and cloned into pRP1099 cut with the same enzymes to produce pRepU-Kan-antC, and plasmid confirmed by NheI/PstI digest. The complementation vector was transferred into Ba ΔantC by conjugation using RHO3 as previously described [78], followed by selection on kanamycin and PCR verification of plasmid presence.

Spore generation and purification

Starter Sterne cultures were inoculated into 3 ml of BHI with shaking at 37°C overnight. The next day, 200-μl aliquots were plated on Difco sporulation media (DSM) agar. The cultures were incubated at 30°C for 5 days. The sporulation efficiency of each plate was checked by microscopy and was always ≥99 complete prior to harvesting in 2 ml of ice-cold sterile Milli-Q water. Spore preparations from multiple plates were combined, washed 2 times in ice-cold water, then purified through a diatrizoic acid gradient. The spore suspensions were resuspended in 95% ethanol to kill any residual vegetative cells, incubated for 1 hour with occasional vortexing, then washed 3 times in ice-cold sterile Milli-Q water. Aliquots were frozen at −80°C and periodically thawed, diluted, plated, and colonies enumerated to determine viable spores. Multiple spore preparations were used for the work in this manuscript. Spore were visualized under phase contrast microscopy using an EVOS XL Core Imaging System (Thermo Fisher Scientific, Waltham, Massachusetts, USA). Spores appear as phase bright points.

Methanolysis, monosaccharide derivatization, and GC–MS of spore preparations

The 1 × 10⁸ spores of the indicated strains or pure anthrose (Sigma-Aldrich, St. Louis, Missouri, USA) were resuspended with 1 M methanolic HCl in glass bottles with TFPE lids and incubated for 16 hours at 80°C [91]. The samples were dried in a speed vac. This was repeated twice with 100% methanol. TMS derivatization was carried out using TMSI (Sigma-Aldrich; St. Louis, Missouri, USA) according to the manufacturer’s instructions. GC–MS analysis was carried out at the Mass Spectrometry and Education Center in the Department of Chemistry at the University of Florida.

Growth curve analysis

Growth curves were carried out by growing starter cultures overnight in BHI at 37°C with shaking. The cultures were diluted to an OD₆₀₀ of 1 then inoculated into modified G media or BHI at a 1:100 dilution. Growth curves were carried out in a Biotek Synergy H1 96-well plate reader in triplicate with orbital shaking at 425 cpm, and the OD₆₀₀ was read every 5 minutes.

Germination and sporulation assays

Germination assays were carried out by thawing frozen aliquots of spores, prepared as above, on ice then diluting to 1 × 10⁹ spores/ml in ice cold 1X PBS. The spore suspensions were heat-activated for 30 minutes at 70°C, then inosine or L-alanine was added to a final concentration of 1 mM or 60 mM, respectively. The OD₆₀₀ was measured every 2 minutes at 37°C preceded by shaking for 5 seconds. The assay was carried out in biological triplicate and technical duplicate. Germination of each strain was normalized to the starting OD₆₀₀ of each strain and is expressed as percentage of starting OD₆₀₀.

For sporulation assays, strains were grown to stationary phase in BHI at 37°C and shaken at 225 rpm. OD₆₀₀ was adjusted to 1 in BHI and used at 1:100 to inoculate flasks of modified G medium in triplicate for each strain [92] followed by incubation at 30°C and shaking at 225 rpm. For ethanol treatment spore counting at 24 and 72 hours, 50-μl aliquots were resuspended in 1 mL of ice cold 90% ethanol and kept on ice for 1 hour, being vortexed every 15 minutes for 15 seconds. Samples were centrifuged for 30 minutes at 4°C at 25,000 × g. Pellets were suspended in PBS + 0.05% Tween-20 (PBS/T), serially diluted, and plated on Bacto BHI agar. For colony counting, untreated aliquots were suspended in PBS/T, serially diluted, and plated on Bacto BHI agar. Plates were incubated at 37°C overnight, after which colonies were counted. The data shown are the average of 2 independent experiments carried out in triplicate. For heat-inactivation spore enumeration, cultures were grown as above and two 1-ml aliquots were removed every 24 hours for 3 days and dilution plated on BHI or heated at 95°C for 10 minutes then dilution plated, as described previously [93], to determine the number of spores and total bacteria present in the cultures. The experiment was carried out in triplicate.

Cellular spore association assay

A549 lung epithelial cells or RAW264.7 macrophages were seeded at 1.5 × 10⁵ cells/well in 24-well Corning CellBIND plates and allowed to attach overnight. The next day, monolayers were washed 3 times with 1X PBS. Spores were thawed on ice and quickly diluted to a multiplicity of infection (MOI) of 10:1 in 200 μl DMEM+10% heat-inactivated FBS, added to the monolayers, and incubated at 37°C in 5% CO₂ for 1 hour. After 1 hour, the supernatants were removed, monolayers were washed 3 times with 1X PBS, then media with 10 μg/ml of gentamicin was added for 1 hour to kill extracellular bacteria. Monolayers were washed 3 more times with 1X PBS to remove antibiotics and any uninternalized spores, then cold water was added to lyse cells, and the suspensions were incubated for 15 minutes then dilution plated on BHI to count the number of internalized bacteria. The initial inoculums were dilution plated to count the number of spores used to infect the monolayers and to calculate internalization efficiency per strain which is presented in relation to Sterne. This experiment was carried out in triplicate in 2 independent experiments. The data were averaged from the 6 data points, and the standard deviation was calculated.

Galleria mellonella infection study

G. mellonella were purchased from Carolina Biological Supply (Burlington, North Carolina, USA). In total, 100 to 250 mg worms were used in groups of approximately 25 Sterne, and ΔantC mutant spores were generated and purified as described above. Purified spores were stored at −80°C, thawed, and enumerated by dilution plating before diluting down to the desired inoculum of 10⁷ spores/ml in PBS. Spore dilutions were kept on ice, and a 27-gauge needle attached to a Hamilton 50-μl microsyringe was used for injections. Larvae were briefly placed on ice then held with forceps to inject 10 μl (10⁵ spores) into the fifth larval abdominal section containing the third distal set of prolegs. Worms were observed at room temperature for 30 minutes before being placed at 37°C. Worms were checked for survival every 8 hours over the 72-hour study by gentle prodding for movement.

Animal studies

Four to 6-week-old female A/J mice were purchased from The Jackson Laboratory (Bar Harbor, Maine, USA). Animals were housed in microisolator cages under pathogen-free conditions with food and water ad libitum. Spores were produced, purified as above, and frozen in 20% (v/v) glycerol aliquots at −80°C. An aliquot of each was thawed and spores enumerated by dilution plating on BHI. Dilution values were determined, and aliquots were thawed on ice and diluted with cold 1X PBS for target inoculation in 100 μl of PBS for subcutaneous injections or 20 μl of PBS for intranasal instillation. Groups of 5 mice (n = 5) were challenged by subcutaneous injection with 100-μl inoculum containing 10⁰, 10¹, 10³, 10⁴, or 10⁶ spores. The same was done for intranasal instillation of 20 μl, but mice were first anesthetized with intraperitoneal ketamine/xylazine injection. In 2 independent studiess, 10 mice in 2 additional experiments were infected subcutaneously with 10³ spores of each strain and were used to confirm our observations of decreased TTD and lower survival percentages near the LD₅₀ dose. The numbers from all experimental challenges were combined for total of 15 mice in Fig 4A. For the predetermined endpoint study, groups of 5 mice were humanely euthanized at 48 hours following 10³ spores subcutaneous challenge. Spleens were removed and processed in 5 ml of 1X PBS using a stomacher (Interscience, Saint-Nom-la-Breteche, France, EU). Undiluted and diluted aliquots were plated on BHI for CFU determination. Mice were observed thrice daily for the first 4 days then once daily until the end of the 14- or 21-day studies. Mice were euthanized when moribund or at the end of the study. LD₅₀ values were calculated by Probit regression using the Stata software package (StataCorp, College Station, Texas, USA).

Prime boost vaccination studies were carried out similarly as the LD₅₀ studies, but mice were primed by intramuscular injection of 50 μl (25 μl in each caudal thigh muscle) with the FDA-approved BioThrax anthrax vaccine adsorbed (AVA) vaccine (BEI Resources) diluted 1:4 in PBS and then boosted 2 weeks later. Four weeks post-boost, groups of 5 mice were subcutaneously challenged with 20, 50, or 100 times the LD₅₀ of each strain calculated for B. anthracis Sterne in this study, and survival was observed out to 21 days. These challenge doses were associated with vaccine failure in mice when using wild-type B. anthracis.

Competitive index study

To ensure pRepU-kan-antC was maintained by complemented spores, 1 × 10⁷ CFU ΔantC/COMP spores were inoculated in 3 ml of BHI in triplicate in the absence of kanamycin and grown with shaking at 37°C for 24 hours. After 24 hours, the cultures were dilution plated on BHI and BHI+Km20. CFU counts indicated no significant difference between kanamycin-resistant colonies on BHI+Km20 and total colonies on BHI. This means that the plasmid was stable after 24-hour growth at 37°C in the complement strain, facilitating the in vivo CI study. Here, the in vitro CI was determined by inoculating the ΔantC and ΔantC/COMP mixed in a total of 2 × 10⁷ spores and inoculated into 3 ml of BHI in triplicate. The cultures were incubated at 37°C with shaking for 24 hours, then diluted in PBS/T and plated for CFU counts on BHI and BHI+Km20. The calculated CFU/ml on BHI+Km20 (the ΔantC/COMP concentration) were subtracted from the calculated CFU/ml on BHI to obtain the ΔantC concentration. The in vitro CI was calculated by the following equation: . For the in vivo CI, ten 4- to 6-week-old female A/J mice were injected subcutaneously with 2 × 10⁷ spores (a mixture of ΔantC and ΔantC/COMP) in 100 μl of 1XPBS. Twenty-four hours later, mice were humanely killed, and the kidneys, liver, and spleen of each mouse were pooled in 5 ml of PBS/T. The organs were homogenized with a handheld homogenizer, diluted in PBS/T, with each dilution plated on BHI and BHI+Km20. The in vivo CI was calculated as described above for the in vitro CI.

Pure death process simulation studies

We used a log-rank test to test for significant differences in TTD between wild-type and anthrose-negative populations. These tests detect differences in patterns regardless of the nature of the process. To test biological hypotheses on virulence and pathogenicity of our mutant and wild-type strains, we modeled the subcutaneous infection assays with a pure death process [53], whose mechanistic approach is ideal for testing such hypotheses. Fitting this model via maximum likelihood is the best approach in our case because the exact time of death of every mouse is never recorded. Additionally, the model considers any possible dependencies in the infection process for mice sharing a cage.

Fitting the death process model to data.

The basic data unit consists of the observed pairs number of surviving mice or larvae and time: where q+1 is the total number of observed time points. Regarding each observation as the initial condition for the next one, from Eq 2 it follows that

Now, using the Markov property and letting τ_i = t_i−t_i−1 to simplify notation, the joint probability of the observations or likelihood function for the observations for 1 experimental batch can be written as

This formulation, however, assumes that the per-mouse death rate μ is a constant independent of dose. Hence, to incorporate our different biological hypotheses regarding the factors influencing the death rate, we wrote μ as a log-linear model of the categorical and continuous variables of interest. Specifically, to test the hypothesis that the death rate differed between the mutant (M) strain and the wild type (W) strain, we fitted a likelihood function with a different set of parameters governing the death rate of each strain, i.e., where the death rate of the mutants was and that of the wild type was and compared that fitting via the BIC with the model that stated that only 1 set of parameters was necessary to explain, both, the mutant strain trials and the wild-type trials with the Joint log-linear model

The model with the lowest BIC was deemed as the best model [102]. From heretofore, we will denote the hypothesis that states that each strain imposes a different death rate as the “Separate dynamics model” and its null counterpart as the “Joint dynamics model”. Finally, we compared the fit of these 2 models by pooling all data across all doses as well as dose per dose. The results of every fit, the BIC scores for every model, and the conclusions of every test are shown in Table 2. The code, instructions, and data output for these analyses are freely available in https://github.com/jmponciano/mice.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Results for the model selection test.

https://doi.org/10.1371/journal.pbio.3001052.t002

Determination of whether the mutant strains resulted in the same death rate than the wild-type strain (Null model or “Joint dynamics model”). If the difference in information criteria is greater than 3 points, the evidence supporting the conclusion (equal or different death rates) is strong and roughly corresponds to a significant hypothesis test at the 5% alpha level.

As shown in Table 2, none of the comparisons pointed toward a significant difference between the death rate of the mutant strain vis-à-vis the death rate of the wild-type strain. Using the data for the 1,000 CFUs dose, we plotted the average estimated loss rates for each strain under the “Separate dynamics model” (Fig 4C). This plot shows that although the mutants death rate is on average predicted to be faster than the death of the wild type, the confidence intervals of both average death trends overlap substantially.

This overlap of confidence intervals led us to investigate via simulations whether infection assays started with larger sample sizes and assumed to evolve according to the same estimated dynamics difference than the 1,000 CFU assays plotted in Fig 4C would yield significant differences. To do that, we simulated experimental assays in batches of 3 replicates, each started with initial sample sizes of 5, 10, 20, 40, and 80 mice using the Gillespie algorithm for stochastic processes[95], parameterized with the maximum likelihood estimates of the “separate model dynamics” for the 1,000 CFU data shown in Table 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. MLEs of the “separate model dynamics” for the 1,000 CFU mouse data in mice and larvae.

https://doi.org/10.1371/journal.pbio.3001052.t003

In summary, to determine the size of the population of mice or larvae to detect a difference of size equal to the one observed for subcutaneous challenge, we:

1. Simulated 1,000 data sets of the same dimension as the ones observed (3 batches of 5 mice followed for 15 days for each of the strains: wild type and Mutant) using a stochastic death process parameterized with the original data;
2. Estimated the death rate as a function of the dose for every simulated data set and for every strain;
3. Tested whether the estimated difference was significantly large using the BIC;
4. Computed the proportion of times the “true” difference was actually detected; and
5. Repeated steps 1 to 5 for data sets with 3 batches of 10, 20, 40, and 80 mice, respectively.

Results from the simulations were used to estimate the power of the infection assays in mice for each population size.

Ethics statement

Animal studies were carried out at the USDA accredited University of Florida animal care facility under approved UF IACUC protocol number 201810488. Animals were housed in microisolator cages under pathogen-free conditions with food and water ad libitum. Mice were euthanized by CO2 asphyxiation followed by cervical dislocation using AVMA-approved methods at humane endpoints or at the end of the study. Ketamine and xylazine cocktail were injected intraperitoneally prior to intranasal spore instillation.