Whole-organism phenotypic screening for anti-infectives promoting host health

Abstract

To date, antibiotics have been identified on the basis of their ability to kill bacteria or inhibit their growth rather than directly for their capacity to improve clinical outcomes of infected patients. Although historically successful, this approach has led to the development of an antibiotic armamentarium that suffers from a number of shortcomings, including the inevitable emergence of resistance and, in certain infections, suboptimal efficacy leading to long treatment durations, infection recurrence, or high mortality and morbidity rates despite apparent bacterial sterilization. Conventional antibiotics fail to address the complexities of in vivo bacterial physiology and virulence, as well as the role of the host underlying the complex, dynamic interactions that cause disease. New interventions are needed, aimed at host outcome rather than microbiological cure. Here we review the role of screening models for cellular and whole-organism infection, including worms, flies, zebrafish, and mice, to identify novel therapeutic strategies and discuss their future implications.

Since the discovery of Penicillin G by Alexander Fleming in 1928, the widespread use of antibiotics in clinical practice has arguably improved human health more than any other intervention in the history of medicine. The advent of antibiotics revolutionized modern medicine, rendering infections that were formerly life threatening as trivial and dramatically decreasing morbidity and mortality from post-operative and post-partum infections. Most conventional ‘Fleming-style’ antibiotics were identified by their ability to kill or inhibit bacteria growing in rich media in vitro. Despite their early success, a number of limitations have hindered their efficacy: (1) antibiotic resistance is rising rapidly, and efforts to identify novel classes of antibiotics have been largely unsuccessful; (2) treatment durations for some infections are long and are plagued by relapse after cessation of therapy; and (3) morbidity and mortality remain high in certain infections despite therapies that achieve bacterial sterilization. There is little doubt that alternative approaches for treating bacterial infections must be explored if we are to overcome these challenges. While many reviews discuss the challenges of conventional antibiotic discovery in the face of rising resistance^1,2, this Perspective will focus on the later challenges of suboptimal efficacies and persisting morbidity and mortality in some infections. Rather than selecting for therapeutic agents that simply kill bacteria, we propose a shift in strategy to discover candidate compounds that attain the desired phenotype, namely improving host health. We explore the strengths and weaknesses of host model systems amenable to chemical screening for small molecules that improve host health, including cells in culture, worms, flies, zebrafish, and mice. Finally, we discuss the feasibility of nonconventional, host-targeted therapies for bacterial infections.

Though antibiotics are currently very effective for a wide range of bacterial infections, there remain a large number of infections in which antibiotic therapy is suboptimal. For example, antibiotics have limited utility in necrotizing fasciitis, a rapidly progressive soft-tissue infection caused by monomicrobial or polymicrobial infection within fascial tissue planes³. Indeed, without early recognition and aggressive surgical management, mortality rates from necrotizing fasciitis can exceed 30%⁴. Similarly, antibiotics alone typically fail to clear infection of implanted hardware or devices without removal of the foreign body because of failure of conventional antibiotics to effectively address bacteria in biofilms^5,6. Furthermore, in other bacterial infections such as endocarditis and tuberculosis^7,8, prolonged durations of treatment with cocktails of antibiotics are required to achieve a durable cure. Finally, there is still a high degree of mortality and morbidity from common bacterial infections, such as pneumonia, despite appropriate antimicrobial treatment, resulting in sterilization^9,10. The need to consider the complex dynamic between host and pathogen is perhaps best illustrated in patients with sepsis, a life-threatening infection that is characterized by a dysregulated host response and organ dysfunction⁹. Severe forms of sepsis account for the most deaths in hospitalized patients⁹, with an estimated incidence of over one million cases per year in the United States alone with mortality rates of 15–20%¹¹, rates that are not tied to infections being untreatable with resistant organisms. Thus, it is clear that alternative interventions are needed that take into account the complex cellular biology of bacteria in vivo, host response, and dynamic host–pathogen interactions, in addition to agents that only achieve microbiological cure. Though efforts have been made to find such adjunctive therapies for sepsis, all have failed to clinically improve host outcome^12,13.

Nonconventional approaches

Though by far the most pervasive approach, ‘Fleming-style’ antibiotics, or small molecules that kill or inhibit the growth of bacteria, constitute only a subset of therapies that could, in principle, be deployed to treat bacterial infections (Fig. 1). Given the current crisis of antibiotic resistance and the pressing need to develop new therapeutic agents, recent attention has turned to nonconventional approaches¹⁴. These nonconventional approaches encompass strategies that are nonetheless ‘Fleming-like’, including phage lysins, antimicrobial peptides, antibodies, and wild-type or engineered bacteriophages that ultimately kill or neutralize bacteria in vitro or in vivo¹⁴ (Fig. 1). Other nonconventional options that target the bacterium but do not have classical in vitro activity include targeting of bacterial virulence mechanisms, biofilms, and/or bacterial proteins essential for survival only during infection (in vivo essential genes; Fig. 1), which has been reviewed elsewhere^14–16. Examples of these nonconventional approaches include the use of small molecules or antibodies that bind and neutralize bacterial toxins or other proteins¹⁴, a strategy that is currently being employed in the treatment of Clostridium difficile with the recent US Food and Drug Administration approval of bezlotoxumab, which binds to toxin B¹⁷.

Fig. 1 |. Conventional and nonconventional approaches to treating bacterial infection.

Conventional, ‘Fleming-like’ approaches that also kill bacteria in vitro include (1) antibiotics, (2) phage lysins, (3) bacteriophages, and (4) antimicrobial peptides. Nonconventional approaches that could improve host health but do not kill or inhibit bacterial growth in vitro include (5) neutralizing antibodies, (6) inhibition of bacterial in vivo essential genes, (7) inhibition of virulence, (8) stimulating host immunity (for example, through immune response pathways like TLR signaling (I), autophagy (II), or induction of immune response genes (III)), and (9) dampening host immunity. Investigating host targets that improve host outcome during bacterial infection could occur at the cellular, organoid, or whole-organism level.

Alternatively, one could target the host itself using small molecules or biologics to either stimulate the immune response when it is inadequate or dampen it when it is excessive (Fig. 1). In fact, this latter approach is already adjunctive care for a number of bacterial infections because of its demonstrated clinical benefit. For example, the administration of corticosteroids as adjunctive therapy has been shown to have benefits for treating meningitis and Pneumocystis carinii pneumonia in HIV patients^18,19; more recent data suggests that this treatment may more generally benefit other pneumonias as well²⁰. In contrast, its value in septic shock has been considerably more controversial²¹, pointing to the challenge of finding such interventions and the likelihood that they may be beneficial in only certain subsets of patients and syndromes. In this Perspective, we focus on discussing host models that are amenable to high-throughput chemical screening for small molecules that improve host health as one approach to finding such interventions. While such screens have the potential to identify molecules that have ‘Fleming-like’ activity or inhibit bacterial virulence or bacterial in vivo essential genes, we will restrict our discussion to how these models might be used to discover chemical probes that can further our understanding of the host response to infection and how one might better intervene therapeutically by targeting the host.

Improving host health through pathogen and host parameters

The relationship between the pathogen and the host can be defined by a number of parameters including pathogen burden, bacterial pathogenicity, host immune response, host immune-cell function, host genotype, and host susceptibility to death, to name a few (Box 1). Interventions to favor host survival could potentially affect one or a few of these parameters. For example, interventions that augment the host’s ability to restrict bacterial expansion and control bacterial burden, such as conventional antibiotics or molecules that stimulate host immunity to clear bacteria (for example, P4 peptide¹⁴), can result in increased host survival or improved host health. Alternatively, interventions that increase the host’s ability to ‘tolerate’ the pathogen without changing the (quantitative) pathogen burden and instead modify the (qualitative) pathogenicity or toxicity of the pathogen, such as antivirulence approaches or host-targeting molecules that blunt the effects of toxins, or those that modulate responses downstream of host immunity, such as host metabolism, can also improve host health. One approach to analyzing the impact of these parameters on host health has been recently reported through disease-tolerance curves²², which describe the relationship between host health and pathogen load within a population (Box 1; Fig. 2a). By using Drosophila melanogaster as a model host for Listeria monocytogenes infection, the relationship between host health and bacterial load was determined to be sigmoidal and described by the parameters of vigor, slope, EC₅₀, and severity of disease²² (Box 1; Fig. 2a). In this model, interventions that decrease bacterial burden should move an individual leftward along the curve (Fig. 2b). Alternatively, mutations that decrease bacterial pathogenicity shift the entire curve to the right, thereby increasing the EC₅₀ such that for a certain pathogen burden²² there is improved host health (Fig. 2c). This work suggests that interventions such as antivirulence approaches or host-targeting molecules that blunt the effects of toxins should shift disease-tolerance curves to the right, increasing the ability of an individual host to ‘tolerate’ the pathogen. Thus, by applying disease-tolerance-curve analysis to novel interventions such as a new compound, one could potentially gain insight into the mechanisms by which the intervention provides benefit to the host.

Box 1 |. Parameters that affect host health.

A number of variables may affect host health over the course of an infection, including the initial inoculum size of the infecting organism, the burden of the pathogen in the host over time, the pathogenicity (i.e., the ability to cause host damage) of the infecting organism, the immune response of the host to infection—which may be determined by host genotype—and susceptibility of the host to death. Ecological immunologists have defined parameters that affect host health following infection in terms of resistance, vigor, resilience, and tolerance⁹⁹.

Resistance is defined as an organism’s ability to control bacterial burden^22,99. Thus, immune mechanisms that restrict bacterial burden can be considered to affect host resistance.

Vigor is defined as the health of an organism before infection²². One might imagine that elderly patient populations or patients undergoing chemotherapy for cancer may have different levels of vigor compared to young, healthy adult populations.

Resilience is defined as a host’s ability to return to its original state of health before infection²². It is understood to refer to an individual host organism, unlike disease-tolerance curves, which reflect disease tolerance within a population²².

A disease-tolerance curve is the dose–response curve of host health as a function of pathogen burden measured within a population^22,99.

In the context of D. melanogaster infection with L. monocytogenes, disease-tolerance curves have been determined to be sigmoidal in shape and governed by the parameters of vigor, slope, EC₅₀, and severity of disease²² (Fig. 2a). However, these curves may not always be sigmoidal in shape; a disease-tolerance curve to cancer modeled in D. melanogaster has been determined to be linear¹⁰⁰.

Fig. 2 |. Disease-tolerance curve analysis of parameters that affect host health.

a, Representation of a disease-tolerance curve of D. melanogaster infected with L. monocytogenes that has been determined to be sigmoidal and is defined by the parameters of vigor, slope, EC₅₀, and severity of disease²². b, Interventions that decrease bacterial burden (for example, ‘Fleming-like’ antibiotics) increase host resistance and should move an individual leftward along the curve. c, Interventions that decrease bacterial pathogenicity should shift disease-tolerance curves to the right (red curve), increasing the EC₅₀ and the ability of an individual to ‘tolerate’ the pathogen. d, Disease-tolerance curves can shift in more complicated ways if the intervention affects more than one parameter (for example, EC₅₀, slope, and the severity of disease). Images in a and c are adapted from ref.²² under a Creative Commons CC BY 4.0 license.

In reality, a given intervention may affect multiple parameters and thus have the potential to shift the pathogen–host relationship, and therefore disease-tolerance curves, in more complex ways²². For example, one could imagine that inhibition of virulence could lead to both decreased host damage and increased pathogen clearance, thus affecting host resistance, the EC₅₀ of the curve, and potentially the severity of disease itself (Fig. 2d). Understanding how small molecules that result in improved host health impact the various factors involved in determining the outcome of infection and the ensuing disease-tolerance curves will be important for building a more informed, systematic approach to alternative therapeutic discovery. Given our current limited understanding of all these factors, how they interact, and where one could intervene to favor host health, perhaps it is not surprising that rational, target-based approaches have fallen short, necessitating consideration of other approaches such as phenotypic screening.

Phenotypic versus target-based screening

Although target-based chemical screening has dominated drug-discovery efforts, it is increasingly recognized that many ‘first-in-class’ drug-discovery success stories have been disproportionately identified in phenotypic screens^23,24. Furthermore, a target-based approach is predicated on the basis of the reasonable knowledge of what is a good target, an assumption that grossly overestimates our understanding of disease processes, as illustrated by the high failure rate of candidates in clinical trials, including candidates targeting sepsis. Thus, host targets and small molecules that improve host health might be more effectively identified in phenotypic rather than target-based screening approaches. As there has been more success in identifying new antibiotics in phenotypic screens for bacterial cell killing or growth inhibition than in target-based, biochemical screens, we suggest that phenotypic screening, with a pivot toward selecting for improved host health rather than for a microbiological cure, may afford promising chemical probes to elucidate the biology of the host–pathogen balance and small-molecule leads with the potential for therapeutic benefit.

Cellular host models

On the path to performing phenotypic screens for host health, in vitro host models begin to integrate the role of the host, albeit limited to the cellular level. To date, however, such screens carried out in cellular host models have typically selected for inhibition of bacterial growth in host cells or inhibition of host cell death. Recently, the advent of new technologies such as high-content screening, including imaging-based analysis, has allowed more complex descriptions of a compound’s activity, even from the primary screen, that can provide clues to mechanisms of action²⁵. Indeed, such models are revealing examples of small molecules or peptides that can impact infection including the following: toll-like receptor (TLR) agonists^26–28, small molecule inhibitors of the protein kinases Akt and Abl that have also been shown to restrict proliferation of Salmonella typhimurium and Mycobacterium tuberculosis in cultured cells^29–31, or small molecules like rapamycin, carbamazepine, valproic acid, EGFR inhibitors, and fluoxetine (SSRI) that reduce M. tuberculosis bacterial burden in cultured cells, in part, by inducing autophagy^29,32. A number of these studies have shown that results from cellular host-based assays do extend to whole-organism infection, as Abl and EGFR inhibitors prevent M. tuberculosis replication in mice²⁹.

The disadvantage of cellular host models as surrogates for survival of the entire organism is that many viable targets and small molecules that can improve host outcome are likely to be missed in single host cell models. First, this reductionist model lacks the rich complexity of the immune response as a whole, which is a highly orchestrated response involving different immune cell types and various organ systems, as well as the complex proximal spatial organization of of pathogens into discrete lesions surrounded by a range of host cells that include both immune and supporting cell types, and more long-range effects whereby soluble signals such as cytokines can recruit additional host cells and activate more systemic responses, including interactions with other organ systems. In addition, the phenotypic assay used for high-throughput screening of cellular models often focuses simply on the ability to inhibit bacterial expansion, an oversimplified phenotype that does not necessarily map to improved host health (for example, the host has poor resilience and cannot readily return to its state of health before infection; Box 1), highlighting the disconnect between microbiological (sterilization) and patient cures. Finally, most cellular host models suffer from the limitations inherent to all two-dimensional cell-culture-based models in that immortalized host cell lines are often characterized by mutations and chromosomal abnormalities that can ultimately affect their function, and primary cells may be difficult to obtain (if human tissue is specifically needed), may survive poorly ex vivo following harvest for the period of time needed to adequately model infection, or may themselves rapidly loose functions specific to their cell type (for example, primary human hepatocytes); this is changing, however, with the advent of more sophisticated three-dimensional (3D) culture. Though the development of new in vitro models such as organoid 3D models^33,34 and models derived from human pluripotent stem cells^35,36 are beginning to make exciting strides to mitigate some of these limitations, with continued progress anticipated as the ability to incorporate numerous different cell types simultaneously into these systems arises, they will not completely recapitulate the whole organism.

Whole-organism screening

In an ideal world, one would screen for small molecules in exactly the biological setting in which one desires a therapeutic effect, namely in infected humans. However, appropriate ethical standards force us to consider alternatives that may optimize the successful translation from model to human disease. One obvious and attractive approach is screening in animal models that serve as a surrogate for human infection. Indeed, this approach has historically been effective, with perhaps the most successful whole-organism screens for anti-infectives resulting in the discoveries of the anti-syphilitic drug salvarsan³⁷ (Fig. 3a), the sulfa drug prontosil³⁸ (Fig. 3b), and the anti-parasitic drug avermectin^39,40. Salvarsan was discovered by Paul Ehrlich through screening in mice and rabbits³⁷. Described as the world’s first blockbuster drug, salvarsan was the most prescribed drug for syphilis from 1910 until the introduction of penicillin in 1942⁴¹. Similarly, prontosil was discovered by Gerhard Domagk in 1932 in a screen for industrial dyes that rescued mice from streptococcal sepsis³⁸. Strikingly, prontosil had no in vitro activity against streptococcal species because it requires metabolism in vivo to the active sulfanilamide. Domagk won the Nobel Prize for this work in 1939³⁸. Finally, avermectin, which was subsequently optimized to ivermectin (Fig. 3c), was discovered in mice by William Campbell at Merck by screening bacterial fermentation extracts for antiparasitic activity and is currently used in both human and veterinary medicine to treat parasitic worm infections including river blindness, strongyloidiasis, and lymphatic filariasis^39,40. William Campbell shared the 2015 Nobel Prize for Physiology and Medicine for the discovery of avermectin along with Satoshi Ōmura^39,40.

Fig. 3 |. Small molecules with host-targeting activity and/or identified through whole-organism screening.

a, Salvarsan, reported to be a mixture of cyclic As–As bonded species⁹⁸, identified by screening mice and rabbits infected with Treponema pallidum³⁷. b, Prontosil, the first sulfa drug, identified in a screen of industrial dyes that rescued mice from streptococcal sepsis³⁸. c, Ivermectin, derivatized from avermectin, which was identified by screening bacterial fermentation extracts in mice for antiparasitic properties^39,40 and targets invertebrate glutamate-gated chloride channels. d, RPW-24, identified in a screen for small molecules that rescue C. elegans from E. fecalis infection⁴⁸; thought to stimulate C. elegans host response⁴⁹. e, 2-Amino acetophenone, enhances the survival of D. melanogaster and M. musculus infected with P. aeruginosa^62,63. f, 16,16-dimethyl-prostaglandin E2 (a stabilized derivative of prostaglandin E2 currently being evaluated as a therapeutic for graft-versus-host disease), identified in a D. rerio screen for compounds that perturbed hematopoietic stem cell numbers^23,76. g, PROTO-1, protects zebrafish hair cells from aminoglycoside-induced death^23,77. h, Dorsomorphin⁷⁸, inhibits bone morphogenetic protein (BMP) to treat fibrodysplasia ossificans progressiva (FOP) and anemia of inflammation²³. i, anthracyclines (for example, epirubicin, shown), chemotherapeutic agents found to have a profound protective effect in a murine polymicrobial sepsis model⁸⁶.

Although the examples of salvarsan, metabolized protonsil, and ivermectin still work in the ‘Fleming style’ of targeting pathogen viability, they nevertheless demonstrate the potential of whole-organism screening to find effective therapeutics. In fact, whole-organism screening may be needed in cases such as salvarsan, in which adequate in vitro culture systems do not exist for some pathogens, and protonsil, wherein activation of a prodrug by the host is required. Admittedly, in this era, their choice of animal models is generally considered unfeasible, which has necessitated the consideration of other host–pathogen models that are more amenable to screening for anti-infective candidates, including model organisms like the nematode worm Caenorhabditis elegans, the fruit fly D. melanogaster and the zebrafish Danio rerio. Powerful genetic tools and the ability to carry out chemical screens in these organisms are paving the way toward a more detailed mechanistic understanding of how different cell types and organ systems work together to impact host health. Notably, studies in these organisms do often translate to mice, which are the current gold standard for modeling mammalian infection. It is through these host models that we are beginning to ask the question, how do we improve host health and not just attain microbiological cure?

C. elegans

For over 15 years, the nematode worm C. elegans has been used to model the pathogenesis of a number of clinically relevant human pathogens, including Pseudomonas aeruginosa, S. typhimurium, Staphylococcus aureus, and Enterococcus fecalis⁴². In many cases, the bacterial virulence factors required for infection in mammals and worms are the same, thus validating the use of worms as a model host⁴³. Importantly, studies have also revealed conservation of innate immune pathways between worms and mammals⁴⁴ (Table 1). For example, defense against bacterial infection in worms relies on activation of the PMK-1 p38 mitogen-activated protein kinase (MAPK) pathway, which is related to the p38 MAPK cascade that is activated downstream of TLR signaling in insects and mammals⁴⁴. The activation of the PMK-1 p38 MAPK cassette during infection plays a role in inducing both antimicrobial peptide synthesis and the unfolded protein response (UPR)⁴⁵.

Table 1 |.

Comparison of model hosts amenable to chemical screening for anti-infectives

Organism	C. elegans	D. melanogaster	D. rerio	M. musculus
Size	1 mm	2 mm	0.9-4.5 mma	10 cm
Chemical screening throughput (molecules screened/studyb)	High (37,20048)	Medium (2,000–6,10055–59, c)	Medium (≤1,000d)	Low 54/ avermectin 606/salvarsane
Mode of inoculation	Feeding	Needle pricking Injector pumping Feeding	Automated (or manual) microinjection Immersion	Orogastric, injection, inhalation
Comparison of host immunity
Toll/TLR	+	+	+	+
MyD88	−	+	+	+
p38 MAPK pathway	+	+	+	+
NF-κB	−	+	+	+
Immune cells	−	+	+	+
Cytokines	−	+	+	+
Immune organs	−	−	+	+

^aZebrafish embryo and larvae lengths range from 0.9 mm 16 h post fertilization to 4.5 mm by 7 d post fertilization, which is the relative span during which many zebrafish infection models are conducted.

^bThroughput values represent the number of small molecules screened in a given study.

^cThe published number of compounds screened in flies for other phenotypes with the caveat that this this level of throughput can be impacted by the mode of inoculation.

^dThroughput for D. rerio is based on the authors’ personal experience infecting zebrafish embryos manually by microinjection, which may be increased if the inoculum is given through automated microinjection⁸¹ or immersion⁷⁰.

^eScreening was conducted in both mice and rabbits in order to identify salvarsan (compound 606)³⁷.

From a technical standpoint, the C. elegans infection model is extremely attractive for screening for chemical modulators of infection. Large numbers (~300) of genetically identical progeny can be produced from a single hermaphroditic parent worm⁴³. Worms are transparent, allowing infection with fluorescently tagged bacteria to be visualized in real time. Their size (~1 mm in length) permits arraying in a 96- or 384-well format for chemical screening⁴⁶. For these assays, worms are infected by feeding on a lawn of the pathogen of interest and then arrayed into screening plates with a large particle sorter^47,48. Worm viability following infection is then quantified with automated microscopy and image analysis^47,48. Automation has made this the highest throughput whole organism amenable to chemical screening, with a throughput on the order of 10,000s⁴⁸.

A number of small molecules have been identified that cure the worms of infection but do not themselves have bactericidal or static activity against the infecting bacterium in vitro^46,48. One of these small molecules, RPW-24 (Fig. 3d), was identified in a screen for compounds that prolonged the survival of worms infected with the Gram-positive organism E. fecalis⁴⁸ but also extended the survival of worms infected with the Gram-negative bacterium P. aeruginosa⁴⁹. Because the virulence factors of these two pathogens are quite disparate, the authors of the study speculated that this small molecule might target the host. Using a combination of whole-organism expression profiling, classical epistasis analysis, and RNA interference (RNAi) in C. elegans, they demonstrated that RPW-24 stimulates the induction of host immune-response genes and promotes nematode survival through the p38 MAPK cascade and the transcription factor ATF-7⁴⁹. Though the precise target of RPW-24 is not known, the enormous genetic tractability of the organism and readily available loss-of-function mutants were used to narrow down possible relevant pathways and suggest candidate targets⁴⁹. This study demonstrates the feasibility of identifying host-targeted small molecules that improve host outcome through whole-organism screening.

Although worms are clearly a powerful host model for studying bacterial infections, some aspects of host immunity are not conserved with vertebrates (Table 1). For example, worms do not have dedicated immune cells and therefore do not have adaptive immunity or express cytokines, similarly to mammals. Moreover, mammalian genes that are important for PAMP recognition and innate immune signaling such as MyD88 and NFκB do not have obvious orthologs in worms⁴⁴. Finally, the single TLR homolog identified in worms does not play a central role in innate immunity to a number of pathogens, though it does appear to be important for worm avoidance of Serratia marcescens and defense against infection with Salmonella enterica⁴⁴.

D. melanogaster

Flies (D. melanogaster) have long been used as a model host for bacterial pathogenesis studies due to their genetic tractability and the greater similarity of host immunity to mammals compared to worms, including the presence of phagocytic cells and more similar innate immune signaling cascades⁵⁰ (Table 1). For example, the first Toll receptor was originally identified in D. melanogaster and was found to sense many pathogens, including bacteria and fungi⁵⁰. Needless to say, since the original discovery of the first TLR, the voluminous body of literature in the past few decades illuminating the key role of TLRs in sensing pathogens to activate the innate immune response, including the identification of downstream signaling pathways and pathogen ligands that activate these pathways, attests to the power of flies as a host infection model and the way that understanding mechanisms of innate immune defense in flies translates to human infection.

Flies have been used to model infection with a wide array of Gram-negative, Gram-positive, and fungal pathogens, including P. aeruginosa, S. typhimurium, Mycobacterium marinum, S. aureus, Streptococcus sp., Candida albicans, and L. monocytogenes⁵¹. Like worms, large numbers of organisms can be evaluated, making flies a relatively high-throughput model compared to vertebrate infection models. Infection can be elicited through needle pricking or by more high-throughput methods of inoculation such as feeding or injector pumping, which enables more precise control of the inoculum⁵¹. Flies can be treated with small molecules through feeding or direct injection⁵², and assessment of bacterial burden is relatively facile in this host through homogenization and plating for colony forming units⁵¹.

Several studies in flies have played important roles in highlighting the potential disconnect between pathogen burden and host health, laying the foundation for the concept of disease tolerance²². A forward genetic screen for D. melanogaster mutants with altered sensitivity to infection from L. monocytogenes demonstrated that a number of mutant flies had a decreased mean time to death following infection but no difference in bacterial burden over time compared to wild-type flies⁵³. Strikingly, there was no overlap between D. melanogaster genes that impacted time to death identified from a whole-organism screen and host genes identified as being important in Drosophila–L. monocytogenes cell culture infection models⁵³, underscoring important differences between studying infection in a dish as opposed to a whole organism. Similarly, a study found that it is possible to have little correlation between bacterial burden and host survival in flies across 11 different genotypes infected with P. aeruginosa; flies can have low bacterial burden and poor survival or high bacterial burden and still remain relatively healthy⁵⁴. These studies suggest that there are host factors that determine the relative health of the organism irrespective of bacterial burden, and thus interventions that focus solely on control or elimination of pathogen burden, characteristic of conventional antibiotics aiming for microbiological cure, may not be optimal for organism health.

Though chemical screening for anti-infective molecules has, to date, not been performed in flies, chemical screens have been performed for a variety of other phenotypes, including phenotypes associated with fragile X syndrome^55,56, enhancers of radiation treatment for cancer⁵⁷, and inhibitors of tumor cell growth^58,59. Although it is not clear the degree to which compounds identified as having activity in flies will translate to mammals, the inverse often appears to be true: compounds with known mechanisms of action in mammals, like inhibition of microtubule and actin polymerization or inhibition of kinases or ion channels, do translate to flies⁶⁰. Encouragingly, known antibiotics do reduce bacterial load and enhance survival in flies following infection²², suggesting that this model has potential. For example, recent work has described the ability of 2-amino acetophenone (2-AA; Fig. 3e), produced by P. aeruginosa, to enhance the survival of infected flies by inhibiting its acute virulence⁶¹; this survival benefit also translates to infected mice⁶¹. Interestingly, 2-AA appears to target the host by epigenetic reprogramming of the pro-inflammatory cytokine loci through HDAC1, ultimately resulting in downregulation of pro-inflammatory cytokine expression both in vitro and in vivo^62,63. Strikingly, mice treated with 2-AA 4 days before P. aeruginosa infection had significantly improved survival compared to untreated controls, which could be reversed with an HDAC inhibitor, yet had a higher bacterial burden compared to untreated controls^62,63.

Like worms, flies also have drawbacks as a host model for human infection. Although they have phagocytic cells, they lack the multilineage immune-cell complexity that is characteristic of vertebrates⁵⁰. At the anatomical level, they lack vasculature and much of the organ system characteristic of mammals⁵⁰, including the lymph organs that are clearly important features of mammalian immunity. Thus, despite the power of these invertebrate models, investigators have searched for additional host model organisms that better mirror mammalian—specifically human–—immunity while preserving many of the advantages of these invertebrate systems.

D. rerio

Zebrafish (D. rerio) embryos have served as a model host for a variety of pathogens including M. marinum⁶⁴, S. aureus⁶⁵, S. typhimurium⁶⁶, and P. aeruginosa^67,68, as they combine a number of advantages of both worms and flies with vertebrate immunity. Similar to worms, the dynamics of infection can be visualized microscopically in real time in living zebrafish embryos because of the availability of genetically engineered or chemically induced transparent zebrafish. Importantly, zebrafish are jawed vertebrates and have an immune system that is very similar to that of mammals (Table 1), including multilineage immune cell complexity (macrophages, neutrophils, B and T cells) and expression of pro-inflammatory cytokines and complement proteins⁶⁹. Consistent and tunable inoculation of embryos and larvae is accomplished through microinjection, but is a lower throughput method than the feeding used to initiate C. elegans and some D. melanogaster infections. However, infection with M. marinum, which is a natural fish pathogen, can also be accomplished by embryo immersion in high concentrations of bacteria^64,70, resulting in colonization of the developing gills and transient colonization of the gut⁷⁰. Finally, bacterial expansion over time in zebrafish embryos is easily monitored either by plating for bacteria directly from homogenized fish or by using microscopy or luminometry to quantitate fluorescently or luminescently labeled bacteria, respectively, from infected embryos, which facilitates a more high-throughput way of determining bacterial expansion^70,71.

In conjunction with zebrafish physiology being highly related to that of higher order vertebrates such as mammals, the technical ease of working with them makes zebrafish embryos well suited for chemical screening, as evidenced by the numerous screens in fish that have been performed for a variety of phenotypes, including but not limited to cardiotoxicity, haematopoiesis, leukocyte migration, embryogenesis, and suppressors of leukemia²³. Compound administration is technically easy, as the fish are typically immersed in compound for the length of the assay, but most often requires absorption through the skin for efficacy²³, which can provide some indication of tissue penetration. Though confirming the target of any compound found to have activity in a whole-organism screen is challenging, the genetic tools necessary to approach this problem are growing in fish, including CRISPR–Cas9, TALENs and morpholinos^23,72–75. Some hits from these screens have translated to mice and are moving forward toward human studies including, for example, 16,16-dimethyl prostaglandin E2⁷⁶ (16,16-dimethyl PGE2; Fig. 3f), a stabilized derivative of prostaglandin E2 used to prevent graft-versus-host disease (GvHD); PROTO-1 (Fig. 3g), which protects zebrafish hair cells from aminoglycoside-induced death^23,77; and dorsomorphin⁷⁸ (Fig. 3h), which inhibits bone morphogenetic protein (BMP) to treat fibrodysplasia ossificans progressiva (FOP) and anemia of inflammation²³.

The feasibility of chemically screening zebrafish embryos for anti-infectives has also been demonstrated through studies showing the efficacy of conventional antibiotics. After manual microinjection of bacteria, exposure to known antibiotics in the surrounding water results in lower bacterial loads and increased survival of infected fish in multiple infection models^68,79. Manual microinjection is highly laborious, which limits throughput; however, recent efforts to develop an automated microinjection pipeline to infect, array, and image infected embryos^80–82 has been limited to infection of pathogens into the yolk of embryos at the early developmental stage (16–128 cells)⁸¹. Although this is advantageous because of the increased throughput (~2,500 embryos per hour)^81–83, it remains to be seen whether infection of the yolk in early developmental stage embryos accurately replicates features of infection in mice or humans, because the bacteria are directly introduced into an extremely rich growth media (the yolk) at a developmental stage without immune cells. Some pathogens such as M. marinum and Staphylococcus epidermidis injected into the yolk in this fashion have been shown to replicate slowly and disseminate as the embryo develops^80–82.

Importantly, it is clear that targeting host mechanisms in combination with antibiotics to improve the outcome of infection in a zebrafish model can indeed translate to biology relevant to human infection. In a landmark study, the lta4h gene was identified in a zebrafish forward genetic screen as a gene required for defense against M. mariunum infection⁸⁴. Lta4h catalyzes production of the pro-inflammatory eicosanoid LTB4⁸⁴. Interestingly, two SNPs identified in the human homolog of lta4H were associated with pulmonary and meningeal tuberculosis⁸⁴. Though homozygosity of either allele was associated with increased disease severity in humans, heterozygosity was shown to be protective, resulting in a less severe disease⁸⁴. Studies of Lta4H in fish revealed that its deficiency results in greater production of anti-inflammatory lipoxins (LXA4) and decreased levels of the pro-inflammatory cytokine TNFα, together contributing ultimately to poor control of M. marinum replication in vivo⁸⁵. Conversely, overexpression of Lta4H results in excessive eicosanoid LTB4 production and excessive TNFα expression, contributing again to poor control of M. marinum replication in vivo, now due to an excessive inflammatory response⁸⁵. Importantly, bacterial load could be controlled in fish by tuning the inflammatory response with small molecules that antagonized either LTB4 or TNFα in fish with increased levels of Lta4H or in Lta4H-deficient fish to improve host survival⁸⁵. These observations in fish could be extended to human patient populations. The study showed a profound effect of glucocorticoid treatment on tubercular meningitis patient survival that was dependent on a patient’s genotype at a polymorphism in the LTA4H gene⁸⁵. Although dexamethasone treatment had only a very modest effect on patient survival as a whole in the patients studied, independent of host genotype, patients homozygous for the LTA4H allele predicted to have increased LTA4H expression and excessive pro-inflammatory cytokine responses showed greatly increased survival following dexamethasone treatment⁸⁵. This study is particularly important, as it demonstrates how studies in fish can be extrapolated to those in humans in which adjunctive glucocorticoid treatment is the standard of care for tubercular meningitis yet clearly has differential effects on host outcome based on lta4H genotype. Furthermore, it is illustrative of the fact that interventions targeting the host may be sensitive to host genotype, necessitating the identification of patient subpopulations who will benefit most from an intervention.

Mice

The story of the discoveries of salvarsan, prontosil, and ivermectin demonstrate the power of whole-organism phenotypic screens in mice. Though use of this model for screening is less feasible in the current era, except perhaps for small chemical collections such as repurposing libraries, several recent studies demonstrate the critical role of whole-organism models in the goal of targeting host health and not simply bacterial sterilization.

To date, many pathogenesis studies define whether a given bacterial gene is important for infection or whether a given small molecule has potential therapeutic efficacy by considering their effect solely on bacterial burden, often irrespective of the outcome on host survival. Yet, recent work in mice highlights the dramatic effects that interventions can have on host survival even without a substantial impact on bacterial burden. This point was best exemplified in a study exploring the effects of the anthracycline class of chemotherapeutic agents on host survival in a murine polymicrobial sepsis model (Fig. 3i)⁸⁶. The authors found that low-dose administration of anthracyclines (epirubicin, doxorubicin, and daunorubicin) provided a profound protective effect on host survival yet had little to no effect on bacterial burden in multiple organs in a cecal ligation and puncture (CLP) model⁸⁶. The protective effect of anthracyclines on host survival was found to be dependent on the DNA damage response mediated through the ATM serine/threonine kinase, (as Atm^−/− mice were not protected with anthracycline treatment, and the autophagy pathway; ATM is known to negatively regulate mTOR, an inhibitor of autophagy)⁸⁶. Perhaps the most striking finding of the paper was that the protective effect of anthracyclines on host survival was mediated through the activity of ATM and the autophagy pathway in the lung rather than in myeloid cell populations following CLP, as tissue-specific deletion of the Atm gene and the autophagy gene Atg7 in the lung, but not in myeloid cells, abrogated the protective effect of anthracyclines during sepsis⁸⁶. This study not only underscores the disconnect between bacterial burden and host health, but also emphasizes how little we understand about which host tissue types might even be optimally targeted to promote survival during infection, an aspect that cannot be discovered or explored in reductionist models.

The elegant and complex interplay between various host cell types and organ systems that support organism health during infection is only beginning to be understood and clearly differs between different pathogens. A recent study on the host requirements for Bacillus anthracis toxin–induced lethality in mice also yielded unexpected findings on the tissue requirements for host death. B. anthracis causes disease primarily through the effects of two toxins: anthrax lethal toxin (LT), a metalloproteinase that cleaves MAPK kinases and the inflammasome component NLRP1⁸⁷, and edema toxin⁸⁷ (ET), an adenylate cyclase. While the precise mechanism of toxin entry into host cells and the subsequent molecular effects of each toxin have been extensively studied, the mechanism by which these toxins ultimately lead to host death was not known until a number of tissue-specific knockout mice of the anthrax toxin receptor, which can bind both LT and ET and is required for their transport into host cells, were evaluated for their susceptibility to LT and ET⁸⁸. The authors surprisingly found that mortality is dependent on LT targeting of cardiomyocytes and vascular smooth muscle cells, whereas ET-dependent mortality is dependent on targeting hepatocytes⁸⁸. The examples of anthracycline and anthrax toxin both demonstrate that it is not necessarily obvious which tissue types and which organ systems would have the greatest impact on host survival during infection and that, indeed, this may differ between different infecting bacteria. Importantly, addressing these critical questions is now possible using whole-organism models of infection and the latest genetic tools such as tissue-specific siRNA or CRISPR–Cas9, which may ultimately guide the discovery of small molecules that specifically target a tissue that will have the greatest impact on host health improvement.

Though mice have historically been and continue to be the workhorse model for studying human disease, many questions have been raised regarding whether mice are indeed a good model for human infection. While mice are often currently used as the gold standard for evaluating whether a given intervention might translate to humans, it cannot be ignored that the translation rate of host-targeted therapeutic candidates for sepsis from mice to humans is extremely poor⁸⁹. For conventional antibiotics, the high correlation between success based on in vitro efficacy and pharmacodynamics models and success in clinical translation in comparison to that of small-molecule candidates targeting other disease areas is likely tied to the fact that antibiotics are targeting the bacterium rather than the human. Thus, targeting the host in infection will likely be vulnerable to the lower translation rates due to the increased complexity of human disease and physiology compared to bacteria and inherent differences between mice and humans, such as the fact that mice are relatively refractory to inflammatory stimuli, such as endotoxin, in comparison to humans where the lethal dose in humans can be at least five orders of magnitude lower than that of mice⁹⁰. The recent literature examining the effect of inflammatory stimuli on gene-expression responses in whole blood cell populations isolated from human patients and mice suggests that although there are statistically significant similarities in gene-expression changes between mice and humans^91,92, the overlap in similarity of genes that were either up- or downregulated between the two species ranged from only 5.9%–15% in the samples analyzed⁹². Certainly, this somewhat low level of concordance could potentially be attributed to inherent differences in the proportions of immune-cell populations in whole blood between mice and humans, differences in the time frame of blood harvest between species, and large genetic differences expected in samples derived from human patients compared to the smaller genetic variations in inbred mouse lines⁹³, making it quite difficult to make an even-handed comparison between the gene-expression responses of mice and humans. This point highlights one important caveat to finding candidates based on screening in alternative host models: the efficacy of the candidate and the relevance of its underlying biology must be validated in humans rather than extrapolated from surrogate models. Thus, while mice may seemingly be a better model for human infection than other models including worms, flies, and zebrafish, all are imperfect.

Implications and future directions

The value of whole-organism models for elucidating the biology of infection is abundantly clear given the impact that that these models have had on our understanding of mammalian and even human immunity. At the same time, the limitations of such models in completely reflecting human infection biology must also be recognized, as evidenced by the limited translation of therapeutic candidates from animal models to human studies^94–96. In light of this, it is worth contemplating what fair expectations are for whole-organism screens in worms, flies, fish, and mice. We propose that one valuable role that these models may play is in the identification of novel chemical probes to further elucidate the biology of infection and potentially identify new targets. Of course, this value might be dependent on the ability to elucidate the mechanisms of action of novel small-molecule probes of interest. The availability of rapidly improving, comprehensive genetic, genomic, and proteomic tools among these organisms is making this challenge increasingly tractable^72–75. As worms, flies, and fish become progressively high-throughput, cost-effective models, they offer the tremendous benefit of increasing the feasibility to discover small-molecule probes in models that better recapitulate the complexity of the host response in the organism as a whole, wherein extremely diverse sets of cell-types organized in complex anatomical structures must communicate information regarding the health of the organism over both short and long distances.

Optimistically, one can hope that these studies will extend beyond the discovery of chemical probes and eventually translate to therapeutic intervention of human infection. Is this a reasonable expectation? Aside from the challenge of discovering new antibiotics with new mechanisms, for conventional antibiotic candidates targeting the bacterium in which preclinical in vivo pharmacodynamics models are highly predictive of clinical efficacy⁹⁷, this is a reasonable expectation. For host-targeting small-molecule candidates, the jury is still out. However, the fact that some, though not all, important immune mechanisms are evolutionarily conserved across the different host model organisms and the fact that many current drugs used in clinical medicine have been shown to also be efficacious in the different model organisms together suggest that indeed some candidates identified in these model organisms are expected to translate. Humanized versions of these model hosts might also improve translation potential. Furthermore, it is reasonable to speculate that the candidates most likely to translate to human infection may be ones that target host functions that are evolutionarily conserved across a large range of models, including worms, flies, zebrafish, and mice, as their conservation may be indicative of the importance of such functions.

Finally, the current lack of therapeutic approaches targeting the host in bacterial infection raises the more general question of the feasibility of this approach in the treatment of infection. However, current practice argues that this is indeed possible, with several examples of bacterial infections in which adjunctive therapy in combination with a conventional antibiotic results in superior survival compared to that with treatment with an antibiotic alone^18–20. Nevertheless, it is a valid concern that manipulating the host response to infection could be potentially detrimental, with augmentation of immunity resulting in uncontrolled inflammation and dampening of immunity causing immunocompromise and uncontrolled pathogen proliferation. Indeed, the study of Lta4H in zebrafish and its extrapolation to patients with tubercular meningitis attests to the fact that too much as well as too little inflammation both lead to a worse outcome for the host⁸⁵. Thus, one of the great challenges of finding effective host-targeting therapeutics may be the need to tune host immunity rather than completely inhibiting or augmenting it. To increase the complexity of the problem, the degree of tuning required to maximize host health may vary as a function of time throughout the infection course and as a function of underlying host genotype as it relates to the magnitude of host response. The desire for candidates that are able to tune immunity in subtle ways mirrors the quest for inhibitors of kinases in cancer that modulate function to dampen excessive activity without completely eliminating the kinase activity important for homeostasis of normal cells. The ability to obtain this level of control may well determine whether a particular host-targeting therapy will be effective^14,85. In this regard, the development of host-targeted therapies for infection may mirror innovative approaches now being taken in other disease areas, such as cancer, that are pushing toward personalized medicine that is tailored to the timing of intervention based on disease biomarkers and to the unique genotype of the patient.

Infection is still one of the leading causes of death worldwide. Given the growing problem of antibiotic resistance, the current poor rate of return in the development of new ‘Fleming-style’ antibiotics, and the ongoing problem that antibiotics alone are insufficient for improving host health in a variety of settings, there is a clear need to consider nonconventional approaches, such as targeting the host, to treat bacterial infections¹⁴. Multiple approaches will need to be explored to meet these challenges, and whole-organism screening is one valuable tool through which we can begin to explore how we might improve host health following infection instead of simply settling for microbiological cure.