A White-Box Machine Learning Approach for Revealing Antibiotic Mechanisms of Action.

Current machine learning techniques enable robust association of biological signals with measured phenotypes, but these approaches are incapable of identifying causal relationships. Here, we develop an integrated "white-box" biochemical screening, network modeling, and machine learning approach for revealing causal mechanisms and apply this approach to understanding antibiotic efficacy. We counter-screen diverse metabolites against bactericidal antibiotics in Escherichia coli and simulate their corresponding metabolic states using a genome-scale metabolic network model. Regression of the measured screening data on model simulations reveals that purine biosynthesis participates in antibiotic lethality, which we validate experimentally. We show that antibiotic-induced adenine limitation increases ATP demand, which elevates central carbon metabolism activity and oxygen consumption, enhancing the killing effects of antibiotics. This work demonstrates how prospective network modeling can couple with machine learning to identify complex causal mechanisms underlying drug efficacy.

Synthetic biology looks good on paper.

Tremendous progress has been made in the design and implementation of synthetic gene circuits, but real-world applications of such circuits have been limited. Cell-free circuits embedded on paper developed by Pardee et al. promise to deliver specific and rapid diagnostics on a low-cost, highly scalable platform.

Horizontal gene transfer enables programmable gene stability in synthetic microbiota.

The functions of many microbial communities exhibit remarkable stability despite fluctuations in the compositions of these communities. To date, a mechanistic understanding of this function-composition decoupling is lacking. Statistical mechanisms have been commonly hypothesized to explain such decoupling. Here, we proposed that dynamic mechanisms, mediated by horizontal gene transfer (HGT), also enable the independence of functions from the compositions of microbial communities. We combined theoretical analysis with numerical simulations to illustrate that HGT rates can determine the stability of gene abundance in microbial communities. We further validated these predictions using engineered microbial consortia of different complexities transferring one or more than a dozen clinically isolated plasmids, as well as through the reanalysis of data from the literature. Our results demonstrate a generalizable strategy to program the gene stability of microbial communities.

Disturbing the Spatial Organization of Biofilm Communities Affects Expression of agr-Regulated Virulence Factors in Staphylococcus aureus.

Staphylococcus aureus uses quorum sensing and nutrient availability to control the expression of agr-regulated virulence factors. Quorum sensing is mediated by autoinducing peptide (AIP), which at a high concentration reduces expression of surface attachment proteins (coa, fnbpA) and increases expression of exotoxins (lukS) and proteases (splA). Nutrient availability can be sensed through the saeS/saeR system. Low nutrients increase expression of saeR, which augments expression of coa and fnbpA, distinct from the activity of AIP. The formation of spatial structure, such as biofilms, can alter quorum sensing and nutrient acquisition. In natural environments, biofilms encounter forces that may alter their spatial structure. These forces may impact quorum sensing and/or nutrient acquisition and thus affect the expression of agr-regulated virulence factors. However, this has not been studied. We show that periodically disturbing biofilms composed of S. aureus using a physical force affected the expression of agr-regulated virulence factors. In nutrient-poor environments, disturbance increased the expression of coa, fnbpA, lukS, and splA. Disturbance in a nutrient-rich environment at low or high disturbance amplitudes moderately reduced expression of coa and fnbpA but increased expression of lukS and splA. Interestingly, at an intermediate amplitude, the overall expression of agr-regulated virulence factors was the lowest; expression of lukS and splA remained unchanged relative to an undisturbed biofilm, while expression of coa and fnbpA significantly decreased. We hypothesize that these changes are a result of disturbance-driven changes in access to AIP and nutrients. Our results may allow the identification of environments where virulence is enhanced, or reduced, owing to a disturbance. IMPORTANCE Bacteria, such as Staphylococcus aureus, integrate signals from the environment to regulate genes encoding virulence factors. These signals include those produced by quorum-sensing systems and nutrient availability. We show that disturbing the spatial organization of S. aureus populations can lead to changes in the expression of virulence factors, likely by altering the ways in which S. aureus detects these signals. Our work may allow us to identify environments that increase or reduce the expression of virulence factors in S. aureus.

A multiplexable assay for screening antibiotic lethality against drug-tolerant bacteria.

Antibiotic screens typically rely on growth inhibition to characterize compound bioactivity-an approach that cannot be used to assess the bactericidal activity of antibiotics against bacteria in drug-tolerant states. To address this limitation, we developed a multiplexed assay that uses metabolism-sensitive staining to report on the killing of antibiotic-tolerant bacteria. This method can be used with diverse bacterial species and applied to genome-scale investigations to identify therapeutic targets against tolerant pathogens.

Population structure of blaKPC-harbouring IncN plasmids at a New York City medical centre and evidence for multi-species horizontal transmission.

BACKGROUND: Carbapenem-resistant Enterobacterales (CRE) are highly concerning MDR pathogens. Horizontal transfer of broad-host-range IncN plasmids may contribute to the dissemination of the Klebsiella pneumoniae carbapenemase (KPC), spreading carbapenem resistance among unrelated bacteria. However, the population structure and genetic diversity of IncN plasmids has not been fully elucidated. OBJECTIVES: We reconstructed blaKPC-harbouring IncN plasmid genomes to characterize shared gene content, structural variability, and putative horizontal transfer within and across patients and diverse bacterial clones. METHODS: We performed short- and long-read sequencing and hybrid assembly on 45 CRE isolates with blaKPC-harbouring IncN plasmids. Eight serial isolates from two patients were included to assess intra-patient plasmid dynamics. Comparative genomic analysis was performed to assess structural and sequence similarity across plasmids. Within IncN sublineages defined by plasmid MLST and kmer-based clustering, phylogenetic analysis was used to identify closely related plasmids. RESULTS: Comparative analysis of IncN plasmid genomes revealed substantial heterogeneity including large rearrangements in serial patient plasmids and differences in structure and content across plasmid clusters. Within plasmid sublineages, core genome content and resistance gene regions were largely conserved. Closely related plasmids (≤1 SNP) were found in highly diverse isolates, including ten pST6 plasmids found in eight bacterial clones from three different species. CONCLUSIONS: Genomic analysis of blaKPC-harbouring IncN plasmids revealed the presence of several distinct sublineages as well as substantial host diversity within plasmid clusters suggestive of frequent mobilization. This study reveals complex plasmid dynamics within a single plasmid family, highlighting the challenge of tracking plasmid-mediated transmission of blaKPC in clinical settings.

Conjugation dynamics depend on both the plasmid acquisition cost and the fitness cost.

Plasmid conjugation is a major mechanism responsible for the spread of antibiotic resistance. Plasmid fitness costs are known to impact long-term growth dynamics of microbial populations by providing plasmid-carrying cells a relative (dis)advantage compared to plasmid-free counterparts. Separately, plasmid acquisition introduces an immediate, but transient, metabolic perturbation. However, the impact of these short-term effects on subsequent growth dynamics has not previously been established. Here, we observed that de novo transconjugants grew significantly slower and/or with overall prolonged lag times, compared to lineages that had been replicating for several generations, indicating the presence of a plasmid acquisition cost. These effects were general to diverse incompatibility groups, well-characterized and clinically captured plasmids, Gram-negative recipient strains and species, and experimental conditions. Modeling revealed that both fitness and acquisition costs modulate overall conjugation dynamics, validated with previously published data. These results suggest that the hours immediately following conjugation may play a critical role in both short- and long-term plasmid prevalence. This time frame is particularly relevant to microbiomes with high plasmid/strain diversity considered to be hot spots for conjugation.

Thermal stress induces tissue damage and a broad shift in regenerative signaling pathways in the honey bee digestive tract.

Honey bee colonies in the USA have suffered from increased die-off in the last few years with a complex set of interacting stresses playing a key role. With changing climate, an increase in the frequency of severe weather events, such as heat waves, is anticipated. Understanding how these changes may contribute to stress in honey bees is crucial. Individual honey bees appear to have a high capacity to endure thermal stress. One reason for this high-level endurance is likely their robust heat shock response (HSR), which contributes to thermotolerance at the cellular level. However, less is known about other mechanisms of thermotolerance, especially those operating at the tissue level. To elucidate other determinants of resilience in this species, we used thermal stress coupled with RNAseq and identified broad transcriptional remodeling of a number of key signaling pathways in the honey bee, including those pathways known to be involved in digestive tract regeneration in the fruit fly such as the Hippo and JAK/STAT pathways. We also observed cell death and shedding of epithelial cells, which likely leads to induction of this regenerative transcriptional program. We found that thermal stress affects many of these pathways in other tissues, suggesting a shared program of damage response. This study provides important foundational characterization of the tissue damage response program in this key pollinating species. In addition, our data suggest that a robust regeneration program may also be a critical contributor to thermotolerance at the tissue level, a possibility which warrants further exploration in this and other species.

Quantitative Microbial Risk Assessment of Antimicrobial Resistant and Susceptible Staphylococcus aureus in Reclaimed Wastewaters.

The annual risks of colonization, skin infection, bloodstream infection (BSI), and disease burden from exposures to antibiotic-resistant and susceptible Staphylococcus aureus (S. aureus) were estimated using quantitative microbial risk assessment (QMRA). We estimated the probability of nasal colonization after immersion in wastewater (WW) or greywater (GW) treated across a range of treatment alternatives and subsequent infection. Horizontal gene transfer was incorporated into the treatment model but had little effect on the predicted risk. The cumulative annual probability of infection (resulting from self-inoculation) was most sensitive to the treatment log10 reduction value (LRV), S. aureus concentration, and the newly calculated morbidity ratios and was below the health benchmark of 10-4 infections per person per year (ppy) given a treatment LRV of roughly 3.0. The predicted annual disability-adjusted life years (DALYs), which were dominated by BSI, were below the health benchmark of 10-6 DALYs ppy for resistant and susceptible S. aureus, given LRVs of 4.5 and 3.5, respectively. Thus, the estimated infection risks and disease burdens resulting from nasal colonization are below the relevant health benchmarks for risk-based, nonpotable, or potable reuse systems but possibly above for immersion in minimally treated GW or WW. Strain-specific data to characterize dose-response and concentration in WW are needed to substantiate the QMRA.

Drug detoxification dynamics explain the postantibiotic effect.

The postantibiotic effect (PAE) refers to the temporary suppression of bacterial growth following transient antibiotic treatment. This effect has been observed for decades for a wide variety of antibiotics and microbial species. However, despite empirical observations, a mechanistic understanding of this phenomenon is lacking. Using a combination of modeling and quantitative experiments, we show that the PAE can be explained by the temporal dynamics of drug detoxification in individual cells after an antibiotic is removed from the extracellular environment. These dynamics are dictated by both the export of the antibiotic and the intracellular titration of the antibiotic by its target. This mechanism is generally applicable for antibiotics with different modes of action. We further show that efflux inhibition is effective against certain antibiotic motifs, which may help explain mixed cotreatment success.

Dissecting the effects of antibiotics on horizontal gene transfer: Analysis suggests a critical role of selection dynamics.

Horizontal gene transfer (HGT) is a major mechanism responsible for the spread of antibiotic resistance. Conversely, it is often assumed that antibiotics promote HGT. Careful dissection of the literature, however, suggests a lack of conclusive evidence supporting this notion in general. This is largely due to the lack of well-defined quantitative experiments to address this question in an unambiguous manner. In this review, we discuss the extent to which HGT is responsible for the spread of antibiotic resistance and examine what is known about the effect of antibiotics on the HGT dynamics. We focus on conjugation, which is the dominant mode of HGT responsible for spreading antibiotic resistance on the global scale. Our analysis reveals a need to design experiments to quantify HGT in such a way to facilitate rigorous data interpretation. Such measurements are critical for developing novel strategies to combat resistance spread through HGT.

Collective antibiotic tolerance: mechanisms, dynamics and intervention.

Bacteria have developed resistance against every antibiotic at a rate that is alarming considering the timescale at which new antibiotics are developed. Thus, there is a critical need to use antibiotics more effectively, extend the shelf life of existing antibiotics and minimize their side effects. This requires understanding the mechanisms underlying bacterial drug responses. Past studies have focused on survival in the presence of antibiotics by individual cells, as genetic mutants or persisters. Also important, however, is the fact that a population of bacterial cells can collectively survive antibiotic treatments lethal to individual cells. This tolerance can arise by diverse mechanisms, including resistance-conferring enzyme production, titration-mediated bistable growth inhibition, swarming and interpopulation interactions. These strategies can enable rapid population recovery after antibiotic treatment and provide a time window during which otherwise susceptible bacteria can acquire inheritable genetic resistance. Here, we emphasize the potential for targeting collective antibiotic tolerance behaviors as an antibacterial treatment strategy.

Bacterial temporal dynamics enable optimal design of antibiotic treatment.

There is a critical need to better use existing antibiotics due to the urgent threat of antibiotic resistant bacteria coupled with the reduced effort in developing new antibiotics. ß-lactam antibiotics represent one of the most commonly used classes of antibiotics to treat a broad spectrum of Gram-positive and -negative bacterial pathogens. However, the rise of extended spectrum ß-lactamase (ESBL) producing bacteria has limited the use of ß-lactams. Due to the concern of complex drug responses, many ß-lactams are typically ruled out if ESBL-producing pathogens are detected, even if these pathogens test as susceptible to some ß-lactams. Using quantitative modeling, we show that ß-lactams could still effectively treat pathogens producing low or moderate levels of ESBLs when administered properly. We further develop a metric to guide the design of a dosing protocol to optimize treatment efficiency for any antibiotic-pathogen combination. Ultimately, optimized dosing protocols could allow reintroduction of a repertoire of first-line antibiotics with improved treatment outcomes and preserve last-resort antibiotics.

Sub-inhibitory antibiotic treatment selects for enhanced metabolic efficiency.

Bacterial growth and metabolic rates are often closely related. However, under antibiotic selection, a paradox in this relationship arises: antibiotic efficacy decreases when bacteria are metabolically dormant, yet antibiotics select for resistant cells that grow fastest during treatment. That is, antibiotic selection counterintuitively favors bacteria with fast growth but slow metabolism. Despite this apparent contradiction, antibiotic resistant cells have historically been characterized primarily in the context of growth, whereas the extent of analogous changes in metabolism is comparatively unknown. Here, we observed that previously evolved antibiotic-resistant strains exhibited a unique relationship between growth and metabolism whereby nutrient utilization became more efficient, regardless of the growth rate. To better understand this unexpected phenomenon, we used a simplified model to simulate bacterial populations adapting to sub-inhibitory antibiotic selection through successive bottlenecking events. Simulations predicted that sub-inhibitory bactericidal antibiotic concentrations could select for enhanced metabolic efficiency, defined based on nutrient utilization: drug-adapted cells are able to achieve the same biomass while utilizing less substrate, even in the absence of treatment. Moreover, simulations predicted that restoring metabolic efficiency would re-sensitize resistant bacteria exhibiting metabolic-dependent resistance; we confirmed this result using adaptive laboratory evolutions of Escherichia coli under carbenicillin treatment. Overall, these results indicate that metabolic efficiency is under direct selective pressure during antibiotic treatment and that differences in evolutionary context may determine both the efficacy of different antibiotics and corresponding re-sensitization approaches.IMPORTANCEThe sustained emergence of antibiotic-resistant pathogens combined with the stalled drug discovery pipelines highlights the critical need to better understand the underlying evolution mechanisms of antibiotic resistance. To this end, bacterial growth and metabolic rates are often closely related, and resistant cells have historically been characterized exclusively in the context of growth. However, under antibiotic selection, antibiotics counterintuitively favor cells with fast growth, and slow metabolism. Through an integrated approach of mathematical modeling and experiments, this study thereby addresses the significant knowledge gap of whether antibiotic selection drives changes in metabolism that complement, and/or act independently, of antibiotic resistance phenotypes.

Selection and horizontal gene transfer underlie microdiversity-level heterogeneity in resistance gene fate during wastewater treatment.

Activated sludge is the centerpiece of biological wastewater treatment, as it facilitates removal of sewage-associated pollutants, fecal bacteria, and pathogens from wastewater through semi-controlled microbial ecology. It has been hypothesized that horizontal gene transfer facilitates the spread of antibiotic resistance genes within the wastewater treatment plant, in part because of the presence of residual antibiotics in sewage. However, there has been surprisingly little evidence to suggest that sewage-associated antibiotics select for resistance at wastewater treatment plants via horizontal gene transfer or otherwise. We addressed the role of sewage-associated antibiotics in promoting antibiotic resistance using lab-scale sequencing batch reactors fed field-collected wastewater, metagenomic sequencing, and our recently developed bioinformatic tool Kairos. Here, we found confirmatory evidence that fluctuating levels of antibiotics in sewage are associated with horizontal gene transfer of antibiotic resistance genes, microbial ecology, and microdiversity-level differences in resistance gene fate in activated sludge.

Purine and pyrimidine synthesis differently affect the strength of the inoculum effect for aminoglycoside and ß-lactam antibiotics.

The inoculum effect has been observed for nearly all antibiotics and bacterial species. However, explanations accounting for its occurrence and strength are lacking. We previously found that growth productivity, which captures the relationship between [ATP] and growth, can account for the strength of the inoculum effect for bactericidal antibiotics. However, the molecular pathway(s) underlying this relationship, and therefore determining the inoculum effect, remain undiscovered. We show that nucleotide synthesis can determine the relationship between [ATP] and growth, and thus the strength of inoculum effect in an antibiotic class-dependent manner. Specifically, and separate from activity through the tricarboxylic acid cycle, we find that transcriptional activity of genes involved in purine and pyrimidine synthesis can predict the strength of the inoculum effect for ß-lactam and aminoglycosides antibiotics, respectively. Our work highlights the antibiotic class-specific effect of purine and pyrimidine synthesis on the severity of the inoculum effect and paves the way for intervention strategies to reduce the inoculum effect in the clinic.

Metabolic genes on conjugative plasmids are highly prevalent in Escherichia coli and can protect against antibiotic treatment.

Conjugative plasmids often encode antibiotic resistance genes that provide selective advantages to their bacterial hosts during antibiotic treatment. Previous studies have predominantly considered these established genes as the primary benefit of antibiotic-mediated plasmid dissemination. However, many genes involved in cellular metabolic processes may also protect against antibiotic treatment and provide selective advantages. Despite the diversity of such metabolic genes and their potential ecological impact, their plasmid-borne prevalence, co-occurrence with canonical antibiotic resistance genes, and phenotypic effects remain widely understudied. To address this gap, we focused on Escherichia coli, which can often act as a pathogen, and is known to spread antibiotic resistance genes via conjugation. We characterized the presence of metabolic genes on 1,775 transferrable plasmids and compared their distribution to that of known antibiotic resistance genes. We found high abundance of genes involved in cellular metabolism and stress response. Several of these genes demonstrated statistically significant associations or disassociations with known antibiotic resistance genes at the strain level, indicating that each gene type may impact the spread of the other across hosts. Indeed, in vitro characterization of 13 statistically relevant metabolic genes confirmed that their phenotypic impact on antibiotic susceptibility was largely consistent with in situ relationships. These results emphasize the ecological importance of metabolic genes on conjugal plasmids, and that selection dynamics of E. coli pathogens arises as a complex consequence of both canonical mechanisms and their interactions with metabolic pathways.

Tradeoff between lag time and growth rate drives the plasmid acquisition cost.

Conjugative plasmids drive genetic diversity and evolution in microbial populations. Despite their prevalence, plasmids can impose long-term fitness costs on their hosts, altering population structure, growth dynamics, and evolutionary outcomes. In addition to long-term fitness costs, acquiring a new plasmid introduces an immediate, short-term perturbation to the cell. However, due to the transient nature of this plasmid acquisition cost, a quantitative understanding of its physiological manifestations, overall magnitudes, and population-level implications, remains unclear. To address this, here we track growth of single colonies immediately following plasmid acquisition. We find that plasmid acquisition costs are primarily driven by changes in lag time, rather than growth rate, for nearly 60 conditions covering diverse plasmids, selection environments, and clinical strains/species. Surprisingly, for a costly plasmid, clones exhibiting longer lag times also achieve faster recovery growth rates, suggesting an evolutionary tradeoff. Modeling and experiments demonstrate that this tradeoff leads to counterintuitive ecological dynamics, whereby intermediate-cost plasmids outcompete both their low and high-cost counterparts. These results suggest that, unlike fitness costs, plasmid acquisition dynamics are not uniformly driven by minimizing growth disadvantages. Moreover, a lag/growth tradeoff has clear implications in predicting the ecological outcomes and intervention strategies of bacteria undergoing conjugation.

Interactions between metabolism and growth can determine the co-existence of Staphylococcus aureus and Pseudomonas aeruginosa.

Most bacteria exist and interact within polymicrobial communities. These interactions produce unique compounds, increase virulence and augment antibiotic resistance. One community associated with negative healthcare outcomes consists of Pseudomonas aeruginosa and Staphylococcus aureus. When co-cultured, virulence factors secreted by P. aeruginosa reduce metabolism and growth in S. aureus. When grown in vitro, this allows P. aeruginosa to drive S. aureus toward extinction. However, when found in vivo, both species can co-exist. Previous work has noted that this may be due to altered gene expression or mutations. However, little is known about how the growth environment could influence the co-existence of both species. Using a combination of mathematical modeling and experimentation, we show that changes to bacterial growth and metabolism caused by differences in the growth environment can determine the final population composition. We found that changing the carbon source in growth media affects the ratio of ATP to growth rate for both species, a metric we call absolute growth. We found that as a growth environment increases the absolute growth for one species, that species will increasingly dominate the co-culture. This is due to interactions between growth, metabolism, and metabolism-altering virulence factors produced by P. aeruginosa. Finally, we show that the relationship between absolute growth and the final population composition can be perturbed by altering the spatial structure in the community. Our results demonstrate that differences in growth environment can account for conflicting observations regarding the co-existence of these bacterial species in the literature, provides support for the intermediate disturbance hypothesis, and may offer a novel mechanism to manipulate polymicrobial populations.

Infections caused by multiple types of bacteria are tough to treat. For example, co-infections with Staphylococcus aureus and Pseudomonas aeruginosa are so difficult to cure they may persist for years in humans and cause serious illness. But when these two types of bacteria are grown together in the laboratory, P. aeruginosa kills off all the S. aureus. Learning why these two types of bacteria can coexist in people but not in the laboratory may lead to new treatments to clear infections. It may also help scientists grow beneficial bacteria mixes that break down pollution or produce biofuels. Pajon and Fortoul et al. show that interactions between bacterial metabolism and growth rate determine whether S. aureus and P. aeruginosa coexist. In the experiments, they grew both types of bacteria in different environments with different food sources. They measured their growth and metabolism and how many bacteria of each species survived over time. Then, they used their data to develop a mathematical model and tested its predictions in the laboratory again. The type of bacteria that had more energy also grew faster and outcompeted the other species. Measuring the growth rate of the two species allowed the scientists to predict which one would win out and what the tipping point would be. Physically disrupting the mix of bacteria disrupted this relationship. These results may help explain what allows these bacteria to coexist in some settings but not others. It may enable scientists to develop new ways to treat infections with P. aeruginosa and S. aureus that work by manipulating growth in the two species. Bacterial growth and metabolism are known to drive antibacterial resistance. Studies in mice using drugs or other therapies to manipulate growth and metabolism may help scientists thwart these resistance mechanisms. The results may also help scientists design and grow beneficial multispecies bacteria communities.

Growth productivity as a determinant of the inoculum effect for bactericidal antibiotics.

Understanding the mechanisms by which populations of bacteria resist antibiotics has implications in evolution, microbial ecology, and public health. The inoculum effect (IE), where antibiotic efficacy declines as the density of a bacterial population increases, has been observed for multiple bacterial species and antibiotics. Several mechanisms to account for IE have been proposed, but most lack experimental evidence or cannot explain IE for multiple antibiotics. We show that growth productivity, the combined effect of growth and metabolism, can account for IE for multiple bactericidal antibiotics and bacterial species. Guided by flux balance analysis and whole-genome modeling, we show that the carbon source supplied in the growth medium determines growth productivity. If growth productivity is sufficiently high, IE is eliminated. Our results may lead to approaches to reduce IE in the clinic, help standardize the analysis of antibiotics, and further our understanding of how bacteria evolve resistance.