Jonah Larkins-Ford

Tuberculosis requires lengthy multi-drug therapy. Mycobacterium tuberculosis occupies different tissue compartments during infection, making drug access and susceptibility patterns variable. Antibi...

Mycobacterium abscessus infections are difficult to treat and are often considered untreatable without tissue resection. Due to the intrinsic drug-resistant nature of the bacteria, combination therapy of three or more antibiotics is recommended. A major challenge in treating M. abscessus infections is the absence of a universal combination therapy with satisfying clinical success rates, leaving clinicians to treat infections using antibiotic lacking efficacy data. We systematically measured drug combinations in M. abscessus to establish a resource of drug interaction data and identify patterns of synergy to help design optimized combination therapies. We measured approximately 230 pairwise drug interactions among 22 antibiotics and identified 71 synergistic pairs, 54 antagonistic pairs, and four potentiator-antibiotics not previously reported. We found that commonly used drug combinations in the clinic, such as azithromycin and amikacin, are antagonistic in lab reference strain ATCC19977, whereas novel combinations, such as azithromycin and rifampicin, are synergistic. Another challenge in developing universally effective multidrug therapies for M. abscessus is the significant variation in drug response between isolates. We measured drug interactions in a focused set of 36 drug pairs across a small panel of clinical isolates with rough and smooth morphotypes. We observed highly strain-dependent drug interactions that cannot be predicted from single-drug susceptibility profiles or known drug mechanisms of action. Our study demonstrates the immense potential to identify synergistic drug combinations in the vast drug combination space and emphasizes the importance of strain-specific combination measurements for designing improved therapeutic interventions.

A challenge in designing treatment regimens for tuberculosis is the necessity to use three or more antibiotics in combination. The combination space is too large to be comprehensively assayed; therefore, only a small number of possible combinations are tested. We narrowed the prohibitively large search space of combination drug responses by breaking down high-order combinations into units of drug pairs. Using pairwise drug potency and drug interaction metrics from in vitro experiments across multiple growth environments, we trained machine learning models to predict outcomes associated with higher-order combinations in the BALB/c relapsing mouse model, an important preclinical model for drug development. We systematically predicted treatment outcomes of >500 combinations among twelve antibiotics. Our classifiers performed well on test data and predicted many novel combinations to be improved over bedaquiline + pretomanid + linezolid, an effective regimen for multidrug-resistant tuberculosis that also shortens treatment in BALB/c mice compared to the standard of care. To understand the design features of effective drug combinations, we reformulated classifiers as simple rulesets to reveal guiding principles of constructing combination therapies for both preclinical and clinical outcomes. One example ruleset is to include a drug pair that is synergistic in dormancy and another pair that is potent in a cholesterol-rich growth environment. These rulesets are predictive, intuitive, and practical, thus enabling rational construction of effective drug combinations based on in vitro pairwise drug synergies and potencies. As more preclinical and clinical drug combination data become available, we expect to improve predictions and combination design rules.

A lengthy multidrug chemotherapy is required to achieve a durable cure in tuberculosis. Variation in Mycobacterium tuberculosis drug response is created by the differing microenvironments in lesions, which generate different bacterial drug susceptibilities. To better realize the potential of combination therapy to shorten treatment duration, multidrug therapy design should deliberately explore the vast combination space. We face a significant scaling challenge in making systematic drug combination measurements because it is not practical to use animal models for comprehensive drug combination studies, nor are there well-validated high-throughput in vitro models that predict animal outcomes. We hypothesized that we could both prioritize combination therapies and quantify the predictive power of various in vitro models for drug development using a dataset of drug combination dose responses measured in multiple in vitro models. We systematically measured M. tuberculosis response to all 2- and 3-drug combinations among ten antibiotics in eight conditions that reproduce lesion microenvironments. Applying machine learning to this comprehensive dataset, we developed classifiers predictive of multidrug treatment outcome in a mouse model of disease relapse. We trained classifiers on multiple mouse models and identified ensembles of in vitro models that best describe in vivo treatment outcomes. Furthermore, we found that combination synergies are less important for predicting outcome than metrics of potency. Here, we map a path forward to rationally prioritize combinations for animal and clinical studies using systematic drug combination measurements with validated in vitro models. Our pipeline is generalizable to other difficult-to-treat diseases requiring combination therapies. One Sentence Summary: Signatures of in vitro potency and drug interaction measurements predict combination therapy outcomes in mouse models of tuberculosis.

Morphological profiling is a method to classify target pathways of antibacterials based on how bacteria respond to treatment through changes to cellular shape and spatial organization. Here we utilized the cell-to-cell variation in morphological features of Mycobacterium tuberculosis bacilli to develop a rapid profiling platform called Morphological Evaluation and Understanding of Stress (MorphEUS). MorphEUS classified 94% of tested drugs correctly into broad categories according to modes of action previously identified in the literature. In the other 6%, MorphEUS pointed to key off-target activities. We observed cell wall damage induced by bedaquiline and moxifloxacin through secondary effects downstream from their main target pathways. We implemented MorphEUS to correctly classify three compounds in a blinded study and identified an off-target effect for one compound that was not readily apparent in previous studies. We anticipate that the ability of MorphEUS to rapidly identify pathways of drug action and the proximal cause of cellular damage in tubercle bacilli will make it applicable to other pathogens and cell types where morphological responses are subtle and heterogeneous. Significance Tuberculosis is a leading cause of death in the world and requires treatment with an arduous multidrug regimen. Many new tuberculosis drugs are in development, and the drug development pipeline would benefit from more rapid methods to learn drug mechanism of action and off-target effects. Here we describe a high-throughput imaging method for rapidly classifying drugs into categories based on the primary and secondary mechanisms of cellular damage caused by different antibacterials called Morphological Evaluation and Understanding of Stress (MorphEUS). We anticipate that MorphEUS will assist in rapidly pinpointing pathway of action of antibacterials for tuberculosis and other bacterial infections.

Multidrug combination therapy is an important strategy for treating tuberculosis, the world’s deadliest bacterial infection. Long treatment durations and growing rates of drug resistance have created an urgent need for new approaches to prioritize effective drug regimens. Hence, we developed a computational model called INDIGO-MTB that identifies synergistic drug regimens from an immense set of possible drug combinations using the pathogen response transcriptome elicited by individual drugs. Although the underlying input data for INDIGO-MTB was generated under in vitro broth culture conditions, the predictions from INDIGO-MTB correlated significantly with in vivo drug regimen efficacy from clinical trials. INDIGO-MTB also identified the transcription factor Rv1353c as a regulator of multiple drug interaction outcomes, which could be targeted for rationally enhancing drug synergy.

Combinations of three or more drugs are used to treat many diseases, including tuberculosis. Thus, it is important to understand how synergistic or antagonistic drug interactions affect the efficacy of combination therapies. However, our understanding of high-order drug interactions is limited because of the lack of both efficient measurement methods and theoretical framework for analysis and interpretation. We developed an efficient experimental sampling and scoring method [diagonal measurement of n-way drug interactions (DiaMOND)] to measure drug interactions for combinations of any number of drugs. DiaMOND provides an efficient alternative to checkerboard assays, which are commonly used to measure drug interactions. We established a geometric framework to factorize high-order drug interactions into lower-order components, thereby establishing a road map of how to use lower-order measurements to predict high-order interactions. Our framework is a generalized Loewe additivity model for high-order drug interactions. Using DiaMOND, we identified and analyzed synergistic and antagonistic antibiotic combinations against Mycobacterium tuberculosis. Efficient measurement and factorization of high-order drug interactions by DiaMOND are broadly applicable to other cell types and disease models.

Influenza A viruses (IAVs) inhibit host gene expression by a process known as host shutoff. Host shutoff limits host innate immune responses and may also redirect the translation apparatus to the production of viral proteins. Multiple IAV proteins regulate host shutoff, including PA-X, a ribonuclease that remains incompletely characterized. We report that PA-X selectively targets host RNA polymerase II (Pol II) transcribed mRNAs, while sparing products of Pol I and Pol III. Interestingly, we show that PA-X can also target Pol II-transcribed RNAs in the nucleus, including non-coding RNAs that are not destined to be translated, and reporter transcripts with RNA hairpin structures that block ribosome loading. Transcript degradation likely occurs in the nucleus, as PA-X is enriched in the nucleus and its nuclear localization correlates with reduction in target RNA levels. Complete degradation of host mRNAs following PA-X-mediated endonucleolytic cleavage is dependent on the host 5’->3’-exonuclease Xrn1. IAV mRNAs are structurally similar to host mRNAs, but are synthesized and modified at the 3’ end by the action of the viral RNA-dependent RNA polymerase complex. Infection of cells with wild-type IAV or a recombinant PA-X-deficient virus revealed that IAV mRNAs resist PA-X-mediated degradation during infection. At the same time, loss of PA-X resulted in changes in the synthesis of select viral mRNAs and a decrease in viral protein accumulation. Collectively, these results significantly advance our understanding of IAV host shutoff, and suggest that the PA-X causes selective degradation of host mRNAs by discriminating some aspect of Pol II-dependent RNA biogenesis in the nucleus. All viruses depend on host components to convert viral mRNAs into proteins. Several viruses, including influenza A virus, encode factors that trigger RNA destruction. The influenza A virus factor that serves in this capacity is known as PA-X. PA-X limits accumulation of host mRNAs and proteins in infected cells and suppresses host responses to infection, but to date its precise mechanism of action remains obscure. Here we report that PA-X selectively targets cellular mRNAs, while sparing viral mRNAs, thereby compromising host gene expression and ensuring priority access of viral mRNAs to the protein synthesis machinery. We demonstrate that complete degradation of mRNAs cut by PA-X is dependent on the host factor Xrn1 and that PA-X likely works in the cell’s nuclei. Interestingly, PA-X targeting appears to be selective for products of host RNA polymerase II, and canonical mRNA processing is required for cleavage. Even though viral mRNAs are spared from PA-X-mediated degradation, PA-X-deficient viruses displayed defects in the synthesis of certain viral mRNAs and decreased viral protein accumulation. Thus, PA-X-mediated host shutoff influences the efficiency of viral gene expression. These studies significantly advance our understanding of this important viral host shutoff protein and may provide future opportunities to limit the pathogenesis of influenza A virus infection.

Translation in eukaryotes is followed to detect toxins and virulence factors and coupled to the induction of defence pathways. Caenorhabditis elegans germline-specific mutations in translation components are detected by this system to induce detoxification and immune responses in distinct somatic cells. An RNA interference screen revealed gene inactivations that act at multiple steps in lipid biosynthetic and kinase pathways upstream of MAP kinase to mediate the systemic communication of translation defects to induce detoxification genes. Mammalian bile acids can rescue the defect in detoxification gene induction caused by C. elegans lipid biosynthetic gene inactivations. Extracts prepared from C. elegans with translation deficits but not from the wild type can also rescue detoxification gene induction in lipid-biosynthesis-defective strains. These eukaryotic antibacterial countermeasures are not ignored by bacteria: particular bacterial species suppress normal C. elegans detoxification responses to mutations in translation factors.

The rise of multidrug-resistant Acinetobacter baumannii and a concomitant decrease in antibiotic treatment options warrants a search for new classes of antibacterial agents. We have found that A. baumannii is pathogenic and lethal to the model host organism Caenorhabditis elegans and have exploited this phenomenon to develop an automated, high-throughput, high-content screening assay in liquid culture that can be used to identify novel antibiotics effective against A. baumannii . The screening assay involves coincubating C. elegans with A. baumannii in 384-well plates containing potential antibacterial compounds. At the end of the incubation period, worms are stained with a dye that stains only dead animals, and images are acquired using automated microscopy and then analyzed using an automated image analysis program. This robust assay yields a Z′ factor consistently greater than 0.7. In a pilot experiment to test the efficacy of the assay, we screened a small custom library of synthetic antimicrobial peptides (AMPs) that were synthesized using publicly available sequence data and/or transcriptomic data from immune-challenged insects. We identified cecropin A and 14 other cecropin or cecropin-like peptides that were able to enhance C. elegans survival in the presence of A. baumannii . Interestingly, one particular hit, BR003-cecropin A, a cationic peptide synthesized by the mosquito Aedes aegypti , showed antibiotic activity against a panel of Gram-negative bacteria and exhibited a low MIC (5 μg/ml) against A. baumannii . BR003-cecropin A causes membrane permeability in A. baumannii , which could be the underlying mechanism of its lethality.