Addressing Antibiotic Resistance with Bacterial Cytological Profiling: A High-Throughput Method for Antibiotic Discovery

1. Introduction

The World Health Organization (WHO) has declared antimicrobial resistance (AMR) as one of the most severe global health threats facing humanity. AMR is the ability of a microbe to survive and grow in the presence of a chemical thought to prevent this effectively. It has been estimated that in 2019 alone, antimicrobial resistance killed at least 1.27 million people globally, more deaths than HIV/AIDS or malaria, with 4.95 million deaths associated with AMR[1]. According to the Centers for Disease Control and Prevention’s Antibiotic Resistance Threats Report[2], in the United States, over 2.8 million antibiotic-resistant infections appear every year, leading to over 35,000 deaths. Furthermore, AMR has been predicted to lead to a total loss of up to $100 trillion for the global economy by 2050[3]. These alarming statistics underscore the urgent need to develop effective therapeutics to combat antimicrobial resistance.

The efforts undertaken in the field of AMR until now have not been enough despite the enormous research effort and inventive therapeutic approaches. Since 1940, antimicrobials have been used widely[4]^,[5] and beyond treating infections, antibiotics enabled many modern medical procedures, such as open-heart surgeries, organ transplants, and cancer therapies[6]. Even before 1940 and for about 60 years after, most antibiotics were discovered by culturing microbial samples from soil for compounds already expressed by microbes[7]^,[6]. However, over the last 20 years, the lipopeptides and the oxazolidinones have been the only two new antibiotic classes created and have been effective only against Gram-positive bacteria[8]. The last novel antibiotic class introduced to kill Gram-negative bacteria was the quinolones when nalidixic acid was synthesized in 1962[9]. Although recent developments have shown potential against Gram-negative bacteria, such as Zosurabalpin, a new antibiotic that disrupts bacterial lipopolysaccharide (LPS) transport from the inner membrane to the outer membrane[10], further advancements in antibiotic discovery are needed. To facilitate the discovery of novel druggable pathways, new high-throughput screens based on Bacterial Cytological Profiling have been developed.

This review emphasizes the use of bacterial cytological profiling (BCP) as a highly effective method for discovering novel antibiotics and rapidly identifying antibiotic targets in a cost-effective manner. BCP initially creates a library that captures the overall profile of bacterial morphological and physiological changes at a single-cell level induced by antibiotics with known mechanisms of action. This profile includes details on bacterial cell shapes and sizes, fluorescent intensities and spatial distribution of DNA, and fluorescent distribution of membrane dyes[11,12,13]. The library is then utilized to classify existing antibiotics based on the specific components of bacterial cells they target and to discover new antibiotics. In this review, we also highlight how BCP is used to expand our quantitative understanding of antibiotic pharmacodynamics and bacterial stress responses, as well as how BCP enhances the development of non-traditional antibacterial strategies such as phage therapies[14,15,16,17,18,19,20].

2. Antibiotic Mechanism of Action and Antibiotic Targets

AMR arises from either genetic alterations or phenotypic changes in pathogens[21,22]. To effectively tackle antibiotic-resistant bacteria, it is essential to understand how antibiotics work, which is known as their mechanism of action (MOA), see Table 1. Understanding MOA involves studying how antibiotics affect bacterial physiology and molecular interaction with bacterial targets (Figure 1). However, identifying the MOA presents a significant limitation in drug discovery.

Traditionally, the pathway inhibited by a compound has been identified mainly through macromolecular synthesis (MMS) assays. These assays use radioactively labelled precursors for peptidoglycan, lipid, protein, RNA, or DNA synthesis[23], therefore enabling the identification of whether one or more pathways are targeted. Despite being an important technique, MMS assays are limited by low accuracy, low resolution, low throughput and time-consuming[13].

To address the limitations associated with MMS assays, diverse alternative techniques for determining the MOA have been developed[40]. These include biochemical approaches, such as affinity chromatography, which identifies direct biophysical interactions between antimicrobials and their targets where the antibiotic interacts with protein from whole-cell extracts[41,42,43]. Genetic approaches, such as selection for resistance and resistance screening[41], focus on the genetic comparison between non-resistant strains and strains that have evolved and mutated to become resistant to a specific antibiotic.

While these additional methods offer several benefits, they also have limitations that influence their effectiveness. The primary challenge is the required time to conduct the assays. Additionally, these methods require large amounts of purified compounds, which can be difficult to obtain, especially since newly discovered compounds are usually available in limited amounts[13]. Apart from all limitations in determining MOA, discovering novel compounds that are active against Gram-positive and Gram-negative bacteria remains challenging.

The process of discovering and developing new classes of antibiotics is particularly challenging, as they must exhibit acceptable pharmacokinetic properties, demonstrate safety, and efficacy[9]. Moreover, producing antibiotics offers limited profit margins due to the high production costs and the extended process of research, testing, and approval[9]. Therefore, new, high-throughput screening platforms are needed for the fast and inexpensive screening of large libraries of synthetic and natural compounds that are highly effective against human pathogens[44,45]. The following section reviews the quick and scalable bacterial cytological profiling methods (BCPs) and discusses their availability for some of the most important human pathogens as outlined in the latest WHO 2024 report.

3. BCP to Identify the Mechanism of Action

In 2013, Poochit et al. designed Bacterial Cytological Profiling (BCP) analysis for E. coli cells using different classes of antibiotics[13] (Figure 2). BCP data are obtained using fluorescent microscopy of E. coli cells stained with fluorescent membrane and DNA dyes as well as fluorescent reporter for membrane permeability. Using image analysis software, various bacterial cell parameters, such as cell length, width, solidity, and DNA content, are extracted[13]. Subsequently, complex multidimensional data are analyzed using the Principal Component Analysis (PCA) technique, to cluster cells based on their cytological profile to identify the MOA of known and unknown antibacterial compounds (Figure 2B, Box 1). Since cytological profiling produces data at a single-cell level[12], it uses morphological data such as bacterial chromosomal condensation, or cell shape changes in response to antimicrobials to differentiate between different targeted metabolic pathways [13] (Figure 2B). Furthermore, this approach can lead to the identification of antibiotics that are effective against multidrug-resistant bacteria[11].

Box 1. Principle Component Analysis (PCA).

BCP takes advantage of the limited presence of cell-cycle checkpoints in bacteria[13]. When stressed by antibiotics, bacteria show phenotypical changes that are characteristic of the antibiotic target. For example, compounds that target the ribosome by stopping protein synthesis (e.g., tetracycline and chloramphenicol) produce circular chromosomes and wide cells[13,55] (Error! Reference source not found.).

During antibacterial treatment, rod-shaped bacteria (bacilli) can shrink and take on an oval form, known as ovoid cells[56,57]. There are no clearly defined names for these cells, however, as they have been referred to in literature as ‘round forms’[58,59]; ‘round cells’[60,61,62]; ‘spherical forms’; ‘spherical cells’[63,64,65]; or ‘coccoid forms’[66,67]. Filamentation, or cell elongation, occurs when rod-shaped bacteria (or sometimes cocci) synthesize peptidoglycan for their side walls but not for their division walls, leading to abnormally elongated cells[68]. This process results from the inhibition of septal peptidoglycan synthesis[56]. Filamentous cells can be also induced when DNA synthesis is inhibited[69,70] or DNA is damaged[71,72,73] by a process known as the SOS response that inhibits cell division [36](Fig. 2).

Antibiotic treatments can drastically alter bacterial cell size, induce localized swelling, bulge formation, blebbing, and thicken peptidoglycan[68]. Occasionally, antibiotic-treated cells can lose cell walls, turning bacterial cells into spheroplasts and protoplasts. Spheroplast are Gram-negative bacteria that lost their peptidoglycan layer, but kept their outer membrane, whereas protoplasts are formed from Gram-positive bacteria that lack the peptidoglycan layer[74]. Bacterial variants that completely lack a cell wall, encompassing both Gram-negative and Gram-positive bacteria, are also known as L-forms[75,76,77,78].

Phenotypical changes could confer an increase in fitness to bacteria in the presence of antibiotics[79]. Resistance to antibiotics usually takes the form of reducing the concentration of intracellular antibiotic or by reducing the binding affinities of the cellular targets to the antibiotic[80]. By using available BCP data, recent studies have shown that by reducing the surface-to-volume ratio (S/V), bacteria can effectively reduce the antibiotic concentration inside a cell, thereby promoting cell growth by decreasing antibiotic influx[80]. Similarly, an increase in S/V can benefit the cell in alternative ways such as increasing the antibiotic efflux rate or the rate of nutrient uptake[80,81,82]. These studies explain how cell shape transformations promote bacterial survival under antibiotic treatments – pointing towards potential new druggable targets that control cell shape and size under stress.

BCP has been successfully employed to study the MOA of various antibacterial agents, including azithromycin[83], diphenylureas[84] and thailandamide[85]. It has also been used to identify the cellular pathways targeted by anticancer metal complexes[86] and to study the response of bacteria to antibiotics in different growth conditions[87]. Additionally, BCP has been used to identify the cellular pathways targeted by antibacterial molecules affecting different cellular pathways[88]^,[89], making it a valuable tool not only for determining antibacterial targets but also to potentially identify novel MOA i.e., ones that target new proteins or new pathways (Figure 3).

BCP can also be used to determine the MOA of treatments beyond antibiotics, such as phage therapies, revealing how phages disrupt essential cellular pathways[14,15,16,17,18,19,20]. BCP allows the visualization of bacterial chromosomal condensation, cell shape and overall cellular morphology changes within bacterial cells during phage infection. These changes not only reveal the pathways and cellular targets phages use to propagate their lifecycle but also highlight the role of bacterial defense mechanisms in combating phage infection[15,17,20]. BCP has demonstrated how the overexpression of phage-related proteins can induce specific phenotypic changes as a result of the activation of a bacterial defense system to suppress phage propagation[14]. Additionally, BCP has been fundamental in assessing the impact of different antibiotics on phage replication, revealing that certain antibiotics can synergize with phages to enhance bacterial cell lysis. In contrast, others inhibit phage propagation by disrupting essential bacterial processes[15]. This dual capability of BCP to show both the direct effects of phage infection and the influence of external agents such as antibiotics, makes it a high-throughput tool in studying phage-bacteria dynamics.

4. BCP of Important Human Pathogens

Most importantly, BCP has been successfully used to study some of the most important human pathogens from the WHO Bacterial Priority Pathogens List (Table 2). In 2017, the first Bacterial Priority Pathogen list was created by the WHO in collaboration with researchers from the Division of Infectious Diseases at the University of Tübingen, Germany which used a multicriteria analysis technique to inform research and development (R&D) for future antibacterial compounds[90]. Now, seven years after the introduction of the list there have been novel antibiotics put onto the market either with effectiveness in vivo or in vitro against pathogens deemed critical priority, but unfortunately, resistance has been found in almost every one[91,92]. This year the WHO updated this list to tackle new developments in antimicrobial resistance to give an updated and directions for policy makers and insight on future developments. The new Bacterial Priority Pathogens List 2024 includes 15 resistant pathogens, ordered at various levels of priority from medium; high; to critical[93] (Table 2). Out of 15 pathogen groups, bacterial cytological profiling is not available for 30 % of them: Non-typhoidal Salmonella, Neisseria gonorrhoeae, Group A and B Streptococci, Haemophilus influenzae. Therefore, urgent BCPs regarding these severe pathogenic organisms are needed.

5. BCP to Identify New Druggable Cell Pathways

BCP is used to scan new antibacterial components to identify their specific targets (Fig. 3). As demonstrated, BCP effectively differentiates between various morphological changes induced by different antibiotics, thereby providing insights into the antibiotic’s MOA. If a novel antibiotic places bacteria in a distinct region of the PCA plot compared to known antibiotic targets, it could indicate a new pathway target or MOA previously uncharacterized (Fig. 3). For example, if the PCA analysis shows that the morphology of bacteria treated with a new antibiotic clusters in a region associated with membrane or RNA targets (Arrows 2 and 3 in Error! Reference source not found.), it directly indicates the antibiotic's mode of action. Conversely, if the antibiotic's effect causes a morphology change that places bacteria in a novel zone, as illustrated with Arrow 4, it may suggest the discovery of a new antibacterial pathway or target.

Figure 3. Representation of Principal Component Analysis (PCA) using bacterial morphologies to determine the MOA of a novel antibiotic. Arrow 1 indicates the antibiotic used against certain bacteria, which can change their shape depending on the antibiotic's MOA. If the bacteria exhibit a morphology as indicated by arrow 2, the antibiotic targets the membrane. Conversely, if the bacteria display a morphology as indicated by arrow 3, the antibiotic targets RNA. However, if the morphology is completely different from the known and clustered morphologies, as shown by arrow 4, it suggests that the antibiotic targets a novel pathway and if the bacteria do not show any change, it suggests that they are not susceptible to this antibiotic or, in the worst-case scenario, that they are resistant to the antibiotic.

Figure 3. Representation of Principal Component Analysis (PCA) using bacterial morphologies to determine the MOA of a novel antibiotic. Arrow 1 indicates the antibiotic used against certain bacteria, which can change their shape depending on the antibiotic's MOA. If the bacteria exhibit a morphology as indicated by arrow 2, the antibiotic targets the membrane. Conversely, if the bacteria display a morphology as indicated by arrow 3, the antibiotic targets RNA. However, if the morphology is completely different from the known and clustered morphologies, as shown by arrow 4, it suggests that the antibiotic targets a novel pathway and if the bacteria do not show any change, it suggests that they are not susceptible to this antibiotic or, in the worst-case scenario, that they are resistant to the antibiotic.

Together, BCP significantly enhances drug development by offering a precise, fast and systematic method for characterizing the effects of new antibacterial agents. Its ability to identify target-specific morphological changes provides a comprehensive tool for uncovering novel antibiotic targets and advancing our understanding of bacterial physiology.

6. BCP Limitations

Even with all the advantages we mentioned about BCP, it has certain limitations. BCP can identify the general target of an antibiotic, but it cannot provide precise information about the exact site within the target that is affected. For instance, while BCP can indicate that an antibiotic targets the ribosome, it cannot specify which part of the ribosome is involved.

BCP requires staining dyes to evaluate DNA content and cell size and shape, with fluorescent dye intensity being essential for determining the antibiotic MOA. A wide variety of dyes, protein fusions, and reporter strains have been used in BCP, facilitating both fast MOA detection, and discovery of new MOAs. However, despite the abundance of possible dyes, strains, and assays selecting the most appropriate ones for specific phenotypic experiments remains challenging, as more information is needed to understand cellular functions (Error! Reference source not found.). This is exacerbated by the complexity of bacterial physiology with many processes being overlapped by mechanisms such as metabolic flux or co-dependent regulation [107]^,[108].

Table 3. Studies with bacterial profiles following the original BCP method. This table includes the organisms studied, the dyes used to visualize cellular components, data availability, and image analysis methods used to extract data for profiling.

*Pipelines, scripts, or instructions are detailed and/or included in the paper. Programs are also widely accessible.

Table 3. Studies with bacterial profiles following the original BCP method. This table includes the organisms studied, the dyes used to visualize cellular components, data availability, and image analysis methods used to extract data for profiling.

*Pipelines, scripts, or instructions are detailed and/or included in the paper. Programs are also widely accessible.

Table 4. Studies using protocols similar to the original BCP method. This table includes the organisms studied, the dyes used to visualize cellular components, whether the data is available online, and image analysis methods used to extract data for profiling.

*Pipelines, scripts, or instructions are detailed and/or included in the paper. Programs are also widely accessible.

Table 4. Studies using protocols similar to the original BCP method. This table includes the organisms studied, the dyes used to visualize cellular components, whether the data is available online, and image analysis methods used to extract data for profiling.

*Pipelines, scripts, or instructions are detailed and/or included in the paper. Programs are also widely accessible.

7. BCP Potential Improvements

As shown in Error! Reference source not found. and 4, a variety of fluorescent dyes have been used to investigate the cytological profiles of different bacterial organisms. However, newly developed dyes have the potential to provide more detailed information that could help in building a more comprehensive response profile. Despite the prevalence of cell wall targeting antibiotics in BCP experiments, direct methods for visualizing cell wall synthesis and remodeling during antibiotic exposure have been lacking. In 2012, Kuru et al. discovered a groundbreaking method for bacterial cell wall staining using fluorescent amino acids[133]. The cell wall provides the shape and structural integrity of the cell. It is made of peptidoglycan (PG), which consists of glycan strands cross-linked by D-amino acid (DAA)[134]. The team introduced HADA and NADA, two fluorescent D-amino acids (FDAAs) attached to a D-amino acid backbone (3-amino-d-alanine). This chemical biology approach aims to detect and visualize the exact location and amount of new peptidoglycan layer synthesis in bacteria. By using HADA or NADA as a fluorescent peptidoglycan label during cell wall synthesis, the technique also allows researchers to observe morphological changes in bacteria over time. This is relevant as HADA and NADA can be implemented in the methodology of BCP to investigate the growth modes of bacteria under antibiotic exposure as they exhibit a diverse growth pattern – that could confer selective advantages in their environments[135]^,[80].

Other fluorescent dyes are available to quantitatively probe bacterial physiological states: ThT and DiBAC₄ for bacterial membrane potential[136,137,138,139], carboxy-H₂DCFDA for reactive oxygen species (ROS)[139], and DAF-FM for reactive nitrogen species (RNS)[139]. By integrating membrane potential, ROS, and RNS into cytological profiles could provide additional information regarding bacterial physiology and bacterial stress response during antibiotic treatment.

8. Image Analysis Tools for BCP and Data Availability

Fluorescent microscopy has allowed us to visualize many cell components in great clarity, however, quantitative image analysis of both cellular and sub-cellular structures has been a continual challenge[125,140]. The requirement for accurate tools is highlighted by the scale of the bacterial objects, with typically 100-300 pixels per typical E. coli cell[124], while antibiotic-treated cells could have an order of magnitude larger sizes[13]. In addition to variations in size, bacteria exhibit a diverse array of shapes[135]. While most studied bacteria in BCP (see Table 3 and 4) are rod-shaped or spheroid, there is growing interest in bacteria with more complex shapes that emerge after exposure to antibiotics, such as seen in Caulobacter crescentus[79].

To precisely define cell boundaries and to segment cellular components, sub-pixel segmentation methods are required[124],[141]. Many segmentation solutions are currently available as user-friendly plugins such as MicrobeJ (using classical segmentation), within easily accessible platforms such as ImageJ[140,142,143]. Stand-alone image analysis programs are also available but these can at times be less supported and less accessible[118]. Classical image segmentation techniques have been used since the 1960s[144,145], laying the groundwork for the more advanced artificial intelligence methods used nowadays.

In 2016 SuperSegger was created to improve upon flaws in segmentation through thresholding in bacterial phase-contrast images and it combines classical segmentation with Deep learning[146]. To correct common errors in segmentation from both the thresholding and watershed, SuperSegger uses a shallow neural network trained from the segmentation data[147]. Recently, Deep neural networks (DNNs) which have now become the backbone of most Deep learning segmentation methods are now widely recognized as superior tools for cell segmentation[148]. As showcased in Table 3, Deep learning is significantly underutilized in original BCP studies. However, some recent BCP profiles have been created using Deep learning for segmentation and object detection (Table 4).

Among recent and easily available Deep learning segmentation algorithms, Pachitariu et al. demonstrated that Cellpose outperformed the popular programs Mask R-CNN and StarDist when applied to a varied dataset of different cell types and cell-like-objects, showcasing it as a powerful general solution for cell segmentation[142]. Cutler et al. assessed the performance of (at the time) state-of-the-art cell segmentation algorithms on a wide array of bioimages of bacteria with different morphologies. This led to the design of the algorithm, Omnipose, which outperforms all segmentation algorithms tested across a varied dataset of bacterial cell sizes, shapes, and optical characteristics and as such has been used extensively in research[124]. Beyond segmentation, Deep learning has been used to predict morphologies of unlabeled cells, allowing researchers to simulate more tags than are currently possible to image at once and potentially extending the power of cytological profiling[149,150].

Segmented data availability (Table 3 and 4) is invaluable for scientific communities and accelerates new findings. By using published BCP data and mathematical modelling, the researchers uncovered the robustness of scaling behavior between cell surface area and volume in E. coli[81] and B. subtilis[82], inferred cell physiological alterations upon antibiotic treatments[151], and proposed a new antibiotic resistance pathway mediated by cell surface-to-volume ratio (S/V) transformations[80]. Therefore, the availability of BCP data is good practice and should be considered as a benchmark for all future BCP platforms, especially for pathogenic bacteria (Table 2).

9. Conclusions

Despite significant advances in research and the development of new tools, combating antimicrobial resistance (AMR) requires a multifaceted approach. Continued investment in research and development, global collaboration, and the effective implementation of surveillance and prevention strategies are crucial. Bacterial Cytological Profiling (BCP) stands out as a rapid and cost-effective technique that facilitates drug discovery by revealing the mechanism of action of novel antibacterial agents through detailed physiological and morphological analysis. Furthermore, BCP could be used to identify phenotypic changes when multiple antibiotics are used, revealing unique or overlapping cell morphologies induced by these combinations[12]. However, systematic explorations of cytological profiles for drug combinations are still missing.

Apart from bacteria, cytological profiling methods are also widely used for other organisms such as yeast[152,153], fungi[154], and human cells[155,156,157]. Deep learning techniques employed for yeast and human cells have been used without feature extraction, however, this method has not yet been applied to bacteria. Therefore, wider availability, applications and integration of machine learning tools across different scientific fields are needed.

Besides BCP being used to discover new antibiotics, BCP has been used to investigate complex interactions between bacteria and their predators – bacteriophages[14,15,16]. BCP enables the identification of metabolic pathways and cellular processes targeted by phages and antibiotics, both individually and in combination. Therefore, BCP reveals molecular mechanisms governing the phage-bacteria interaction, ultimately paving the way for more effective phage-based antibacterial therapies.