LIVE SINGLE CELL FLUORESCENCE MICROSCOPY; FROM

RESISTANCE DETECTION TO MITOCHONDRIAL DYSFUNCTION

A Dissertation

Presented to

The Graduate Faculty of The University of Akron

In Partial Fulfillment

of the Requirements for the Degree

Doctor of Philosophy

Lucille Ray

August 2020 LIVE SINGLE CELL FLUORESCENCE MICROSCOPY; FROM ANTIBIOTIC

RESISTANCE DETECTION TO MITOCHONDRIAL DYSFUNCTION

Lucille Ray

Dissertation

Approved: Accepted:

______Advisor Department Chair Dr. Michael Konopka Dr. Christopher Ziegler

______Committee Member Dean of the College Dr. Leah Shriver Dr. Linda Subich

______Committee Member Interim Director, Graduate School Dr. Adam Smith Dr. Marnie Saunders

______Committee Member Date Dr. Sailaja Paruchuri

______Committee Member Dr. Abraham Joy

ii ABSTRACT

Single-cell fluorescence microscopy is a powerful tool which can be used to investigate the nature of cellular responses to external stimuli. The application of single- cell fluorescence microscopy to the growing problem of susceptibility detection stands to improve and refine our ability to address bacterial resistance in human infections. In cases of severe bacterial infection, determining the correct antibiotic to use to combat an infection is a race against the patient’s dwindling lifespan. By combining the RedoxSensor™ Green fluorescent dye with live single-cell imaging, we have developed a method for antimicrobial susceptibility testing which can identify susceptibility to a given antibiotic within 100 minutes of treatment. We show that this method is reproducible and can identify susceptibility to several bacterial cell wall targets and bacterial targeted . Our methodology has considerable applicability within the sphere of rational antibiotic drug design as well in its ability to identify antibiotic efficacy as a function of time instead of antibiotic concentration. We use our method to compare the efficacy of 4 recently synthesized . Our work lays the framework for expansion upon our method into microfluidics systems and use in screening candidate antimicrobial drugs.

Mitochondrial morphological analysis within living eukaryotic cells represents another challenge which requires careful application of single-cell fluorescence microscopy. The cuprizone mouse treatment model for multiple sclerosis is known to

iii generate enlarged mitochondria within oligodendrocytes, but much remains unknown about the dynamics of their formation. Cultured MO3.13 oligodendrocyte cells treated with cuprizone are shown to undergo mitochondrial enlargement within 8 hours of direct cuprizone exposure via single-cell mitochondrial size determination with MitoTracker

Red fluorescent dye. Cuprizone treatment is also shown to increase oxygen consumption rapidly following oligodendrocyte exposure using porphyrin-based fluorescence lifetime oxygen consumption measurement.

iv ACKNOWLEDGEMENTS

First and foremost, I must thank my partner and fiancée Lilith Freed, who has kept me sane throughout graduate school. Without their love and patience, I surely would have faltered during the process. I acknowledge the help of my peers at the University of

Akron for support and collaboration throughout this work. Particularly, group members

Kyle Whiddon, Krishna Ojha, and Ravindra Gudneppanavar provided key insights during my work and needed criticisms. Many undergraduate students assisted in bringing this project to completion under my direction, and I would like to specifically thank Erin

Merico, Kailey Christman, and Garrett Parker for their contributions.

Obviously, I could not have completed any of this without my research advisor

Dr. Michael Konopka who has had more patience with me than perhaps at times I deserved. Additionally, I appreciate the time and effort of my committee members Dr.

Leah Shriver, Dr. Adam Smith, Dr. Sailaja Paruchuri, and Dr. Abraham Joy. Dr. Joy’s lab provided invaluable collaboration using their antimicrobials. Dr. Shriver’s lab provided a greatly appreciated opportunity to work with eukaryotic cells and step outside the world of .

v TABLE OF CONTENTS

LIST OF FIGURES……………………………………………………………………...ix

LIST OF TABLES……………………………………………………………………….xii

LIST OF ABBREVIATIONS…………………………………………………………..xiii

CHAPTER

I: INTRODUCTION………………………………………………………………………1

Description of the Work

Live Single Cell Imaging

Mechanisms of Antibiotic Action and Resistance

Antibiotic Susceptibility Detection

Goals for AST Development

Cuprizone and Mitochondrial Health

II: MATERIALS AND METHODS……………………………………………...... 19

Bacterial Strains and Growth Conditions

Bacterial Sample Preparation

Microscopy

Antibiotic Susceptibility Testing

Zone of Clearance and Broth Macrodilution

AST Data Analysis

vi Eukaryotic Cell Culture

Cuprizone Sample Preparation

Oxygen Consumption Measurement

Oxygen Consumption Measurement Analysis

Mitochondrial Size Measurement

III: DEVELOPMENT OF ANTIBACTERIAL SUSCEPTIBILITY METHOD………28

Introduction

Results

Discussion

IV: RSG AST VALIDATION…………………………………………………………55

Introduction

Results

Discussion

V: RSG AST USING ANTIMICROBIAL ……………………………..65

Introduction

Results

Discussion

VI: INVESTIGATING IMMEDIATE IMPACTS ON MITOCHONDRIAL HEALTH

FOLLOWING CUPRIZONE TREATMENT ON A SINGLE CELL LEVEL………75

Introduction

Results

vii Discussion

VII: CONCLUSION……………………………………………………………………..85

REFERENCES…………………………………………………………………………..92

viii LIST OF FIGURES

Figure Page

1.1 Cell wall and cell membrane structure of Gram-positive and Gram-negative bacteria…………………………………………………………………………….4

1.2 Indirect and direct antibiotic action are compared………………………………..5

1.3 Shows the life cycle of in planktonic solution as well as surface attached …………………………………………………………..8

1.4 Mother Machine fASTest method for antibiotic exposure is shown…………….12

2.1 pCR4-TOPO vector schematic showing the ampicillin and kanamycin resistance elements …………………………………………………………………………20

2.2 Attofluor cell chamber assembly showing coverslip insertion and rubber o-ring 21

2.3 Microscope setup displaying the Okolab temperature control cage surrounding the Nikon Ti-E microscope ………………………………………………………….22

2.4 Graph of oxygen consumption in terms of PPM oxygen consumed compared to starting level for a single trial …………………………………………………...26

2.5 All trials oxygen consumption over time normalized and combined for cuprizone treated and normal MO3.13 cells ……………………………………….………26

3.1 Elongation of E. coli strain MG1655 in response to thick unrinsed poly-L-lysine coating.…………………………………………………………………………..30

3.2 Left: E. coli 60 minutes after 100 µg/mL ampicillin addition in the presence of 1 µM RSG. Right: E. coli 60 minutes after 100 µg/mL ampicillin addition in the presence of 1 µM RSG and 20 µM PI.…………………………………….…….32

3.3 Upper left: E. coli highlighted in red are above the signal threshold. Upper right: Selections made by the “Analyze Particles” command on the same image. Lower: Signal intensity results reported for all ROI selections.…………………….…...34

3.4 Threshold determination for a single cell……………...………………………...35

ix 3.5 Characteristic bulging seen following treatment of MG1655 E. coli with 100 µg/mL carbenicillin.……………………………………………………………...36

3.6 Distribution of MG1655 E. coli reaching RSG signal increase threshold following treatment with 100 µg/mL carbenicillin ………………………..……………….36

3.7 Growth curve of MG1655 E. coli in MOPS minimal media for 3 different subculture dilutions in log scale …………………………………………………37

3.8 Distribution of MG1655 E. coli reaching RSG signal increase threshold following treatment with 100 µg/mL carbenicillin without the presence of poly-L-lysine...38

3.9 MG1655 E. coli grown in a B04F Microfluidic Bacteria Plate with MOPS minimal media flowing through containing 1µM RSG.……………………..…..40

3.10 MG1655 E. coli treated with 100 µg/mL kanamycin………….………………...42

3.11 MG1655 E. coli treated with 5 µg/mL .………………………….….43

3.12 MG1655 E. coli treated with 5 µg/mL PMB ………………………….………...44

3.13 Upper: Structure of PMB with a circle indicating the portion removed to generate PBMN. Lower: The structure of PMBN…………………………………..….….46

3.14 MG1655 E. coli treated with 5 µg/mL PMBN with 1 µM RSG and 20 µM PI…47

3.15 MG1655 E. coli treated with 5 µg/mL PMBN with 1 µM RSG and no PI……...47

3.16 MG1655 E. coli treated with 5 µg/mL PMBN with 0.1 µM RSG.……….……...47

3.17 P. aeruginosa treated with an antibacterial polymer coating in the presence of 1 µM RSG and 20 µM PI.………………………………………………………….49

3.18 stained with RSG (green) and PI (purple) and treated with 100 µg/mL vancomycin..………………………………………………………...50

4.1 Graph of RSG signal increase time distribution for 4 trials of MG1655 E. coli treated with 100 µg/mL carbenicillin in MOPS minimal media ………...……...56

4.2 Graph of RSG signal increase time distribution for stationary phase and logarithmic phase MG1655 E. coli treated with 100 µg/mL carbenicillin in MOPS minimal media.……………………………………………………………….…57

4.3 Time distribution of MG1655 E. coli treated with 100 µg/mL carbenicillin reaching RSG signal threshold in LB media.…………………………………...58

x 4.4 Time distribution of MG1655 E. coli treated with 100 µg/mL carbenicillin reaching RSG signal threshold, with MOPS minimal media being compared to LB rich media………………………………………………………………………...58

4.5 Time distribution of MG1655 E. coli treated with 5 µg/mL PMB reaching RSG signal threshold in MOPS media...………………………………………………59

4.6 Time distribution of MG1655 E. coli treated with 5 µg/mL PMB compared to those treated with 100 µg/mL carbenicillin reaching RSG signal threshold in MOPS media.………………………………………………………………….…60

4.7 Ampicillin and kanamycin resistant MG1655 E. coli RSG signal after treatment with ampicillin and PMB………………………………………………………...61

4.8 Left: MG1655 E. coli pre-incubated with 10 µM CCCP, then stained with RSG. Right: The same cells, 5 minutes following treatment with 5 µg/mL PMB…...... 62

5.1 Polyurethane antimicrobial backbone illustrating the varied pendant group functionalization.……………………………………………………………...... 67

5.2 Box and whisker plot of the RSG signal increase time distribution for the 4 polymer antimicrobials tested …………………………………………………...70

5.3 Distribution of RSG signal increase timing for polyurethane polymer antimicrobials (32 µg/mL) compared to PMB (5 µg/mL).……………...……….72

5.4 Composite image of E. coli treated with 32 µg/mL Cy5 labeled 75 A polymer (purple) antimicrobial and RSG (green)…………………………………………73

5.5 TD image of E. coli treated with 32 µg/mL Cy5 labeled 75 A polymer antimicrobial.……………………………………………………….……………74

6.1 Normal MO3.13 cell stained with MitoTracker Red FM, illustrating the threshold- based ROI analysis.………………………………………………………...... 78

6.2 MO3.13 cells treated with 1 mM cuprizone for 8 hours, then stained with MitoTracker Red FM dye.…………………………………………………….....79

6.3 Recorded cross-sectional area of mitochondria from CPZ treated versus non-CPZ treated MO3.13 cells..…………………………………………………………...80

6.4 MO3.13 cells treated with 2.5% ethanol vehicle, then stained with MitoTracker Red FM dye showing mitochondria structure …………………….…………….81

6.5 Oxygen consumption over time for control, CPZ treated, and rotenone treated MO3.13 cells.……………………………………………………………………82

xi 6.5 Shows normal and enlarged megamitochondria in IAR-20 cells………….…….84

xii LIST OF TABLES Table Page 5.1 Polyurethane polymers composition ratios and molecular weights…………….67 5.2 Single factor ANOVA analysis of the RSG signal increase distribution between 75 A, 81 C, and 68 B……………………………………………………………71 5.3 Single factor ANOVA analysis of the RSG signal increase distribution between 75 A, 81 C, 68 B, and 115 B……………………………………………………71 6.1 Single factor ANOVA analysis for control versus cuprizone treated MO3.13 cell respiration……………………………………………………………………….82

xiii LIST OF ABBREVIATIONS AMP Antimicrobial peptide ANOVA Analysis of variance AST Antimicrobial susceptibility test ATCC American Type Culture Collection CCCP [(3-chlorophenyl)hydrazono]malononitrile CCD Charge-coupled device DMEM Dulbecco’s modified Eagle medium FBS Fetal bovine serum LB Lysogeny broth LPS Lipopolysaccharide MHB Mueller-Hinton broth MIC Mean inhibitory concentration MOPS 3-(N-morpholino)propanesulfonic acid MRSA Methicillin resistant Staphylococcus aureus MS Multiple sclerosis OD Optical density PBP Penicillin binding protein PDMS Polydimethylsiloxane PI Propidium iodide PMB PMBN Polymyxin B nonapeptide PPM Parts per million ROI Region of interest RSG RedoxSensor™ Green SCMA Single cell morphological analysis

xiv CHAPTER I

INTRODUCTION

Description of the Work:

This work describes two disparate applications of fluorescence microscopy at the single cell level. Conceptually, it can be divided into two sections of unequal length. The first section, covering chapters III through V, concerns the development of a methodology for the detection of antibiotic susceptibility and applications of that method to traditional and polymer-based antibiotics. The second section, consisting of chapter VI, concerns the analysis of mitochondrial response to cuprizone treatment at the single cell level for an MO3.13 oligodendrocyte cell line. Fluorescence microscopy unites these two sections and combined they offer a robust representation of how powerful single cell analysis can be.

Live Single Cell Imaging:

Live cell imaging has been commonplace for nearly 80 years, dating back to early phase contrast studies of bulk cell culture to investigate cell division. Live single cell imaging is defined by the ability to track a single cell or multiple individual cells over time using microscopy, separating individual cell dynamics from the bulk. Advances in digital imaging software and region of interest (ROI) tracking have allowed automated image collection and analysis to simplify single cell analyses. Tracking individual cells using phase-contrast generates morphological data, but without visible markers for

1 internal cellular elements the technique is limited. Live single cell imaging techniques grew extensively with the ability to establish stable protein-based fluorescence within healthy cells. The expression of green fluorescent protein (GFP) in and

Caenorhabditis elegans opened the floodgates of genetic reporting using fluorescently tagged protein expression and allowed for localization of target proteins within individual cells (Chalfie, 1994). Fluorescent dyes have been used to add targeted labelling to live cell studies but can undergo significant photobleaching, have cell toxicity, and face limited specificity. However fluorescent dyes have advantages in price, availability, and ease of use that far outweigh these issues when appropriate controls are taken.

Live/dead fluorescent stains are commonly used in microscopy and flow- cytometry to identify and estimate the number of living and dead cells in a population.

Flow cytometry allows high-throughput detection of single cell characteristics. Single cells are passed through a microfluidic system which focuses the cells to a single cell stream which is interrogated by laser scanning. Flow cytometry represents a kind of single-cell imaging, but it only represents a snapshot in time for each cell. When considering live/dead stains this means that a cell can only be seen to be alive or dead at a specific time point. Live/dead assays also suffer from imperfect accuracy based upon strain identity and subpopulation dynamics where stain may be pumped out of the cell at different rates (Stiefel, 2015). Single-cell tracking combined with fluorescence microscopy allows individual cells to be monitored constantly to evaluate individual signal dynamics and combat these issues.

2 Direct Versus Indirect Mechanisms of Antibiotic Action:

Antibiotics have a multitude of mechanisms by which they accomplish bacterial cell death which cannot be fully explored in this work. To simplify and provide distinctions between antibiotic mechanisms, we will consider a separation of “direct” versus “indirect” bactericidal action. Indirect bactericidal action involves the inhibition of proteins or cell building blocks, such that cellular function can no longer proceed as normal and a cascade leading to cell death occurs from that point. This includes antibiotic classes such as β-lactams, fluoroquinolones, aminoglycosides, and .

Ampicillin, a β-lactamase, targets penicillin binding protein to inhibit the formation of the bacterial cell wall over time, tying its action to the speed of cell growth. Indirect antibiotic action efficiency is linked to cell respiration and the rate of bacterial growth in this way, with slowly growing and slowly respiring cells exhibiting greater resilience to antibiotic action (Toumanen et al., 1986; Lobritz et al., 2015). Direct antibiotic action involves targeting of the antibiotic molecule which is sufficient to produce bactericidal action without inhibiting cellular processes. Contact between the antibiotic and the bacterial cell is enough on its own to cause cell death. Direct action typically involves destroying or solubilizing components of the outer membrane of Gram-negative bacteria or the cell wall of Gram-positive bacteria (Fig. 1.1). A key class of antibiotics which use direct action are polymyxins such as polymyxin B (PMB) and colistin. Polymyxins first localize to outer membrane of Gram-negative bacteria by interactions with lipopolysaccharides (LPS), then destabilize both outer and inner membrane integrity, effectively lysing the cell (Velkov et al., 2010). Because the action of direct antibiotics

3 occurs irrespective of the cells physiological state, the only limiting factor is how quickly the drug can be delivered to the surface of the bacteria.

A graphical representation of indirect versus direct antibiotic action is given in

Fig 1.2. The delineation between direct and indirect antibiotic action tends to draw a line between peptide-based antimicrobials which structurally destabilize bacteria, and non- peptide antibiotics which target proteins and cellular building blocks. Delineating the two will be essential to discuss the varying speed at which antibiotic action can occur, and how we can develop a method which categorizes antibiotic efficiency with respect to time.

Figure 1.1: Cell wall and cell membrane structure of Gram-positive and Gram-negative bacteria (Figure reproduced from Silhavy et al., 2010. Copyright 2010 Cold Spring Harbor Laboratory Press.)

4

Figure 1.2: Indirect antibiotic action shown in the upper portion typically involves a target protein or building block, which causes damage to the bacterial cell over the course of normal growth. Direct antibiotic action shown in the lower portion involves a direct contact which immediately impacts bacterial survival. Typically a component of the outer membrane or cell wall is targeted and destabilized. Antibiotic Resistance, Tolerance and Persistence:

Antibiotic resistance is a growing threat to global healthcare outcomes due to the increase in multi-drug resistant and limited development rate of novel antibiotics (Aslam et al., 2018). Antibiotic resistance evolves from advantageous mutations that modify antibiotic binding targets, many of which occur naturally in bacterial populations. Some mutations that convey antibiotic resistance come at a fitness cost when compared to wild-type bacteria, and so are not selected for during normal growth conditions. Resistance causing mutations can undergo positive selection upon exposure to antibiotics when all bacteria are not eliminated during antibiotic challenge.

Multi-drug resistant pathogens commonly arise via positive selection due to excessive use of antibiotics in commercial food animals, the rate of which continues to rise each year (Graham et al., 2009; Van Boeckel et al., 2015). Mutations which do not come at a

5 fitness cost during normal growth conditions are even more dangerous, as they compete easily with non-resistant bacterial populations without antibiotic induced selection pressure (Melnyk et al., 2015). The mechanism by which the bacteria accomplishes resistance is tied to whether the mutation is likely to involve a fitness penalty. A middle ground between resistance and susceptibility exists known as antibiotic tolerance.

Tolerance is defined by the ability of a bacterial population or subpopulation to resist the bactericidal action of an antibiotic for longer than the wild type bacteria, while still experiencing a bacteriostatic effect (Tuomanen et al., 1986). The concentration of antibiotic required to kill the bacteria generally remains the same, while the rate of killing can change dramatically. Tolerance can be mediated by mutation but can be triggered by the active genetically regulated response of bacteria to starvation conditions (Nguyen et al., 2011).

Persistence is another form of antibiotic tolerance whereby a subpopulation of bacteria can survive antibiotic challenge indefinitely to repopulate later. Persistence is generated when a subset of cells enters a phenotypic state which is effectively one of hibernation, unable to be killed by indirectly acting antibiotics which rely upon growth to cause cell death (Balaban et al., 2004). The persister phenotype is made more successful in bacteria which form . Biofilms are defined as adherent, non-planktonic, bacteria which exist within a secreted protective extracellular matrix (Donlan, 2002).

Biofilms can severely limit the amount of antibiotic which reaches the surface of bacteria, increasing the concentration of antibiotic required. Biofilm production also contributes significantly to the issue of resistance by spawning multi-drug resistant cells which neither die nor grow in the presence of high concentrations of antibiotics (Lewis, 2005;

6 Vickery et al., 2013). The matrix is composed of extracellular polymeric substances

(EPS) which are high molecular weight polymers that impart structural stability to the film and prevent or slow the movement of antibiotics and other chemicals through the matrix to the contained cells. Bulk planktonic cells are in constant flux with biofilm aggregates in solution during normal growth, but during antibiotic challenge planktonic bacteria are eliminated while the mature surface-attached biofilms remain in hibernation to disperse planktonic cells following treatment (Fig 1.3). Biofilms allow bacteria to colonize surfaces more effectively and represent a major challenge for surface sterilization in hospitals where resistant microbes are more likely to accumulate, as cleaning agents can be ineffective against these dormant bacterial populations (Smith et al., 2008). Human gut microbe populations are defined by high-level biofilm organization which makes gastrointestinal infections such as Clostridium difficile difficult to treat (De

Vos, 2015). Currently available antibiotic treatments can wipe out beneficial bacterial populations and enrich resistant harmful strains within the gut microbiome, damaging host metabolism and immunity (Sun et al., 2019).

7 Figure 1.3: Reproduced from Rumbaugh et al., 2020. a) Shows the life cycle of Pseudomonas aeruginosa in planktonic solution as well as surface attached biofilm. Biofilm aggregates can exchange with planktonic cells in solution, while mature biofilms hold a bulk of hibernating cells which can release back into the bulk following antibiotic challenge. b) Shows a mature biofilm initiating a dispersion response during appropriate bulk solution conditions. To approach the issue of biofilm formation and the treatment of highly resilient bacterial populations, the application of antibiotics with high rates of biofilm penetration and low propensity for resistance are required. Naturally occurring nonribosomal peptides which have antimicrobial actions, typically referred to by the class name antimicrobial peptides (AMPs), the most successful class of antibiotics in terms of low resistance rates and potency (Joo et al., 2016; Spohn et al., 2019). PMB, which is a cyclic

8 AMP, penetrates P. aeruginosa biofilms and causes oxidative damage even at sublethal concentrations (Lima et al., 2019). The primary drawback of PMB and other polymyxins is inherent cytotoxicity due to lack of sufficient specificity in membrane targeting bacteria versus eukaryotic membranes (Duwe at al., 1986). Vancomycin, a glycopeptide

AMP, destroys biofilms in Staphylococcus aureus and remains the last line of defense against multidrug resistant strains such as methicillin resistant S. aureus (MRSA) (Post et al., 2016). The power of naturally occurring AMPs continues to foster development of novel synthetic AMPs and AMP mimetics which show promise despite poor drug trial outcomes to date (Cardoso et al., 2020). AMP development for potential drug use focuses on the ability of AMPs to directly target the bacterial membrane and design methods to increase specificity for bacterial rather than eukaryotic membranes, while still retaining broad spectrum applicability (Li et al., 2017). Detecting the susceptibility of bacteria to membrane targeted AMPs and methods to compare their relative potency therefore remains a high priority for both AMP drug development and hospital-side resistance detection.

Current State of Antibiotic Susceptibility Detection:

Antibiotic susceptibility detection is well-suited to take advantage of the low-cost generality of fluorescent dyes. Antibiotic resistance is typically generated through the exposure of bacteria to concentrations of antibiotic which are insufficient to kill them.

Through this exposure, strains which have advantageous mutations that inhibit the antibiotic’s function in some way evolve and no longer respond to treatment. An ideal test for antibiotic resistance is one which costs relatively little money compared to the generality and speed of its application. Traditional antibiotic susceptibility test (AST)

9 methods, while cheap and easy to use, are tied to cell culturing and can require upwards of 12 hours (Kirn, 2013; Boardman et al., 2015; Kostic et al., 2015). Broth microdilution remains the most commonly used of these methods and consists of inoculating one or multiple microplates with the bacteria to be tested and varying concentrations of antibiotic. Plates can be read in an automated fashion by scanning the optical density

(OD) in each well to determine if bacterial growth has occurred and occluded the well.

The lowest concentration of antibiotic which inhibits growth on the plate is considered the minimal inhibitory concentration (MIC) for the tested antibiotic.

New rapid AST methods using microfluidic devices are in development with an emphasis on monitoring growth rate and cell morphology on a single cell level. The fastest possible detection based on this concept is fASTest, which detects growth rate anomaly in E. coli after 1 to 2 cellular doubling times, around 20 minutes (Baltekin et al.,

2017). fASTest takes advantage of the Mother Machine, a microfluidic device which maintains constant growth over time by ejecting newly formed cells from individual mother cells of a single bacterial width (Fig 1.4) (Wang et al., 2010). fASTest does not provide universal support however, as the loading process of the Mother Machine is not adaptable to organisms which are highly motile or form biofilms such as P. aeruginosa and is completely incompatible with Gram-positive organisms. Single cell morphological analysis (SCMA) attempts to bridge this gap by monitoring more general changes in cell morphology using microscopy (Choi et al., 2014). In SCMA the shape, size and number of bacteria are monitored over time following treatment with antibiotic in microfluidic agarose channels. SCMA allows for automated characterization of bacterial morphology as resistant or susceptible and can yield results within 4 hours for clinical samples.

10 SCMA suffers from lowered accuracy for Gram-negative bacteria compared to Gram- positive bacteria due to filamentation or swelling which Gram-negative bacteria can undergo as a response to antibiotic challenge (Choi, 2014). Gram-negative bacteria treated with β-lactam antibiotics represent the greatest challenge to SCMA, as filamentation and swelling are defining features of β-lactam exposure (Yao, 2012).

Neither SCMA nor fASTest can differentiate tolerant bacterial populations from susceptible populations as tolerant bacterial populations will cease growth in the same manner as susceptible populations.

11 Figure 1.4: Mother Machine fASTest method for antibiotic exposure is shown. The schematic for the microfluidic device is show at top. Reference cells on the left grow normally, while cells exposed to antibiotic on the right grow more slowly. Resistant subpopulations, also known as heteroresistance, represent another hurdle for all AST methods. Resistant subpopulations allow bacteria to regrow following treatment with an antibiotic that the bulk of the bacterial population is susceptible to.

Subpopulation resistance is due to phenotypic heterogeneity which may produce small populations which are not dominant due to inferior growth characteristics but nonetheless survive within the bacterial bulk. Heteroresistance alone does not guarantee bacteria will

12 regrow in the case of human infection, as the immune system may eliminate the persistent bacteria left behind after infection. Heteroresistance is prevalent in bacterial infections and nearly impossible to detect using traditional MIC based methods (Nicoloff et al., 2019). By approaching susceptibility from a single-cell perspective a population can be sampled at the finest possible grain, allowing for the highest chance of detecting resistant cells.

Microscopy and AST:

Testing infectious bacteria using single-cell microscopy with an appropriate fluorescent probe is not hard to conceptualize but faces several hurdles to being medically useful. Ease of use is one area where all single-cell microscopy-based techniques have been criticized. Purchasing laser-based confocal microscopes for the express purpose of antibiotic resistance detection is a steep investment for many hospitals and time on existing units is valuable. LED fluorescence microscopes continue to rise in prevalence as a low-cost alternative to laser and mercury vapor lamp (MVP) microscopes and have proven sensitive enough to be on par with more expensive options for the detection of tuberculosis infection (Ojha, 2020). Throughput is another area where microscopy suffers compared to traditional AST methods. Microfluidics offers increased throughput to microscope-based methods by reducing the cross-sectional area required for imaging, in some cases down to the width of a single bacteria. Microfluidics also affords the opportunity to run tests in parallel with multiple conditions being monitored on the same microscope, saving microscope time. Even with these increases, the throughput does not come close to that of growth-based AST techniques which can test as many parallel conditions as there is sample for. Fluorescent probe AST methods must

13 therefore compete through speed of detection as their primary advantage and justification for use.

AST methods need to get faster due to rising antibiotic resistance around the world coupled with decreasing antibiotic drug discovery (Brown et al., 2016). Overuse of existing antibiotics in both livestock and humans has precipitated an impending antibiotic crisis where simple infections may be difficult or impossible to treat due to complete resistance (Rossolini et al., 2016). New antibiotic development has not been a profitable commercial venture for several decades due to the rate at which bacteria acquire resistance to newly deployed drugs. High-throughput screening continues to produce new classes of compounds, such as the recent 2-pyrazol-1-yl-thiazole derivatives (Ivanenkov et al., 2019), but bringing such compounds to market still requires significant commercial investment. New compound classes have relatively unknown pharmacokinetics which present difficulties in the drug development pipeline and can result in the abandonment of otherwise promising drugs. Widely available rapid AST could help to reduce the quantity of inappropriate antibiotic prescriptions which contribute to increased bacterial resistance

(Ventola, 2015). Rapid AST can also improve outcomes in cases of aggressive highly resistant bacterial infections where an appropriate antibiotic needs to be identified quickly. Resistant bacterial infections can cause sepsis with inappropriate treatment, where the immune system undergoes an extreme response as it becomes overwhelmed by the infection. Sepsis is estimated to result in 5 million deaths annually at a mortality rate of 26% for severe cases, with treatment typically hinging on finding an appropriate antibiotic within a matter of hours (Fleischmann et al., 2016).

Testing Novel Antibiotics:

14 Developing new antibiotics requires extensive testing for susceptibility using commercial AST methods to generate MICs. MIC is considered the gold standard for antibiotic effectiveness despite having shortcomings when considering antibiotic distribution within the body. Each drug has a rate of clearance which is different, meaning that the peak antibiotic concentration after treatment may be reached for hours, or merely minutes (Levison et al., 2009). In order to account for the clearance rate, antibiotics are dosed to provide a peak concentration which is significantly greater than the MIC, resulting in large doses multiple times per day for many antibiotics. Missed doses of antibiotics contribute to evolving bacterial resistance, as the antibiotic concentration falls below the MIC for an extended period. Bacteria which evolve an ability to withstand antibiotic exposure for longer periods before dying can develop a tolerance to antibiotics which is different and more difficult to detect than pure resistance, as it does not change the MIC (Brauner et al., 2017). Therefore, the most effective antibiotic is one which has a low MIC value compared to host toxicity, as well as being efficient with respect to the time required to kill the .

Live cell imaging AST would by its nature provide time resolved information about antibiotic treatment. Such a technique would be useful when testing newly developed antibiotics and inform synthetic chemistry approaches to generating more efficient forms. Polymer-based antimicrobials make use of targeted development extensively based upon modifying chain lengths and substituents to lower MIC and reduce host toxicity. A rising class of antibacterial polymers is naturally occurring host defense particles known as antimicrobial peptides (AMPs) and synthetically developed mimetics (Mahlapuu et al., 2016). AMPs and their mimetics largely target the membranes

15 of both Gram-positive and Gram-negative organisms directly, making them more difficult to develop resistance against as the bacteria have to drastically alter the composition of the membrane to acquire it (Epand et al., 2016; Zasloff, 2002). They are also promising in their ability to attack biofilms as a class through a variety of mechanism (Yasir et al., 2018). Because the targeting of AMP mimetics can be tuned to reduce toxicity to the host by increasing specificity for bacterial membranes, promising low toxicity and low MIC antimicrobial peptides have already been produced (Mankoci et al., 2019). Improving the efficiency of these polymers even further by decreasing the exposure time required to eliminate pathogens would help continue their development as replacements to traditional small molecule antibiotics.

Goals for AST Development:

Our goal for this project was to develop an AST methodology using fluorescence microscopy which would be cheap, straightforward in its application, and provide results faster than commercially available methods. Such a method would ideally use a commercially available dye which is readily available for replication and expansion of our results by other laboratories. We developed and describe here an AST method using

RedoxSensor™ Green (RSG) dye which generates susceptibility results within 2 hours at the upper limit, and 15 minutes at the lower limit. While our method is preliminary in nature and not suited for commercial deployment, we also considered ways in which it might be expanded upon and present them here. Using the RSG-based AST method to test newly synthesized antimicrobials would improve knowledge of their efficacy using time resolved treatment information. Our method proved to be sensitive to cell wall and cell membrane antibiotics, which makes it useful for testing membrane targeted AMP

16 mimetics. By comparing killing-time efficiency for several polymer antimicrobial variants, we hope to inform further synthetic design choices.

Cuprizone and Mitochondrial Health:

Cuprizone (bis-cyclohexanone oxaldihydrazone) is a chopper chelating compound which is used in modeling multiple sclerosis (MS) disease pathology. Cuprizone causes toxic demyelination in oligodendrocyte cells of mice, mimicking the demyelination seen in multiple sclerosis patients (Blakemore et al., 1972). MS disease etiology is currently unknown, so cuprizone model studies instead focus on investigating methods of demyelination reversal which might be applicable to MS. Cuprizone treatment causes oligodendrocyte mitochondria to undergo enlargement, forming megamitochondria similar to those seen in alcoholic liver disease (Wagner et al., 1975; Wakabayashi et al.,

1974). Mitochondrial dysfunction plays an important role in MS disease pathology, but mitochondrial dynamics and respiration in live cells have not been well characterized

(Barcelos et al., 2019). Cuprizone disrupts the energy metabolism of treated mice and cultured MO3.13 oligodendrocytes significantly, damaging NAD+ generation and potentially causing energy shortfall (Taraboletti et al., 2017). Mitochondrial superoxide dismutase activity is elevated in oligodendrocytes of the cuprizone mouse model, indicative of a possible increase in superoxide load within these mitochondria (De et al.,

1982). While long term treatment of cuprizone is known to precipitate mitochondrial stress, it is of great interest to investigate short term treatment to establish how quickly cuprizone can impact mitochondrial health.

By utilizing single cell imaging techniques and bulk oxygen consumption measurements, we investigated changes in mitochondrial health in MO3.13

17 oligodendrocytes following direct treatment with cuprizone. Live-cell imaging of mitochondria offers direct information about mitochondrial size, shape, and distribution within the cell (Mitra et al., 2010). Our work used the mitochondrial stain MitoTracker®

Red FM dye (ThermoFisher) to generate good signal resolution while staining only viable mitochondria. By comparing the size and shape of stained mitochondria within MO3.13 oligodendrocytes over time, we identified changes in mitochondria morphology following short term cuprizone treatment. We combined this study of mitochondrial morphology with a porphyrin-based bulk oxygen consumption methodology (Dmitriev et al., 2012). Cuprizone has been suggested to directly impair the function of complex IV in the by acting as a copper chelator (Faizi et al., 2016). Our measurements of oxygen consumption for cuprizone treated cells established the early impact of cuprizone treatment on respiration rate for MO3.13 oligodendrocytes.

18 CHAPTER II

MATERIALS AND METHODS

Bacterial Strains and Growth Conditions:

Escherichia coli MG1655 was grown in morpholinopropanesulfonate (MOPS)- buffered glucose minimal medium (Neidhart et al., 1974). Staphylococcus aureus (ATCC

#25923) was grown in LB (Luria-Bertani) Broth, Miller (BD Biosciences). These strains were grown in test tubes containing 5 mL of their respective media at 37°C with shaking

(200 rpm). To generate the β-lactam resistant strain for testing, the pCR4-TOPO empty vector (ThermoFisher) was inserted into E. coli MG1655 by transformation (Fig. 2.1).

19 Figure 2.1: pCR4-TOPO vector schematic showing the ampicillin and kanamycin resistance elements (ThermoFisher).

Bacterial Sample Preparation:

Cells for analysis were grown to between 0.2 and 0.6 OD600. Prior to imaging, the logarithmic phase cells were incubated with 0.1 or 1.0 µM RSG (0.1 or 1.0 µM) and PI

(20 µM) for 15 minutes with shaking in a 37° C incubator. If required, CCCP was added to a final concentration of 10 µM. RSG, CCCP, and PI were all components of the

20 BacLight RedoxSensor Green vitality kit (Thermo/Invitrogen). For each experiment, 20

µL of pre-incubated bacterial cell solution was added to the center of a coverglass in a stainless-steel observation chamber. Specifically, the Attofluor (ThermoFisher) chamber with round 25 mm diameter #1.5 thickness coverglass (Electron Microscopy Sciences) was used as a contained system for antibiotic exposure while imaging (Fig. 2.2). Each coverglass was coated with poly-L-lysine (Sigma) to assist cell adhesion by exposure to a

0.01% solution for 5 minutes.

Figure 2.2: Attofluor cell chamber assembly showing coverslip insertion and rubber o-ring (ThermoFisher). Sample and media are pipetted directly onto the inserted coverslip following chamber assembly. Microscopy:

All images were collected on a Nikon Ti-E inverted microscope with the Nikon

A1 confocal system using either a 60x Plan Apo λ (NA 1.4) oil objective or 100x Plan

Apo λ (NA 1.45) oil objective. RSG was excited with a solid state 488 nm laser and its fluorescence emission was collected from 500-550 nm. Excitation of PI was done with a

21 solid state 488 nm laser and emission filtered from 663 nm to 738 nm. NIS Elements was used for automated data collection and maintenance of detector settings. A constant temperature of 37 °C was maintained for all imaging experiments using a temperature control system and cage incubator (Okolab; Fig. 2.3). Full field images were taken every

15 seconds.

Figure 2.3: Microscope setup displaying the Okolab temperature control cage surrounding the Nikon Ti-E microscope. Temperature control is set using a digital controller and the entire stage is equilibrated based upon stage mounted temperature sensors. Antibiotic Susceptibility Testing:

22 The observation chamber was mounted on an adjustable microscope stage. After starting automated image collection, 1 mL of media containing the antibiotic condition to be tested, which had been pre-warmed to 37° C, was slowly added to the chamber by pipette mixing. The condition containing media was identical to the cell solution media, including the same concentration of RSG and PI, except for the addition of the antibiotics which were added at the following final concentrations: ampicillin (Sigma-Aldrich), 100

µg/mL; carbenicillin (Sigma-Aldrich), 100 µg/mL; polymyxin B (Sigma-Aldrich), 5

µg/mL; polymyxin B nonapeptide (Sigma-Aldrich), 5 µg/mL; vancomycin (Sigma-

Aldrich), 100 µg/mL.

Zone of Clearance and Broth Macrodilution:

Both traditional methods were completed using Mueller Hinton II cation adjusted media (BD). Zone of clearance agar plates were prepared with BBL™ Sensi-Disc™

Antimicrobial Susceptibility Test Discs (BD) containing either 100 µg ampicillin or 300

U polymyxin B. 200 µL of E. coli were plated at 0.5 McFarland standard and allowed to incubate overnight at 37° C. Measurement of clearance around antimicrobial disks was taken after 24 hours had elapsed. Broth macrodilution was performed using 4 mL of media for each culture using antibiotic dilutions of 2, 4, 8, 16 and 32 µg/mL for carbenicillin and 0.5, 1, 2, 4, and 8 µg/mL for polymyxin B. Tubes were inoculated with

4 µL of cells at 0.5 McFarland standard and allowed to grow overnight at 37° C in a non- shaking incubator. Visual growth after 24 hours was used to determine resistance or susceptibility. The carbenicillin broth macrodilution was repeated with E. coli in MOPS minimal media and required 48 hours of growth for positive identification.

AST Data Analysis:

23 NIS Elements processing software was used to add scale bars and time stamps to collected images. Image J distribution Fiji was used to analyze RSG signal and cell death counts on a frame by frame basis (Schindelin et al., 2012, 2015). All processing was completed manually using thresholding to identify cells whose RSG signal reached at least 5x initial average cell signal.

Eukaryotic Cell Culture:

Dulbecco’s Modified Eagle’s Medium (DMEM), Fetal Bovine Serum (FBS), and penicillin streptomycin (P/S) were purchased from Corning (Manassas, VA, USA).

HyClone DMEM/F-12 (1:1; w/o phenol red) was purchased from GE Healthcare Life

Sciences (Logan, UT, USA). Bis(cyclohexanone)oxaldihydrazone (for spectrometric det. of Cu; ≥ 99%) was purchased from Sigma Aldrich (St. Louis, MO, USA). Non-denatured ethanol (200 Proof) was purchased from Decon Laboratories Inc. (King of Prussia, PA,

USA). Rotenone (≥ 97%) was purchased from MP Biomedicals (Solon, OH, USA).

The immortal human MO3.13 oligodendrocyte cell line was cultured in DMEM supplemented with 10% FBS and 1% P/S and maintained in a humified environment at

37 °C with 5% CO2. MO3.13 cell solutions were prepared at a final density of 1.0 x 104 cells in DMEM/F12 containing 10% FBS, 1% P/S, and 2.5% EtOH.

Cuprizone Sample Preparation:

All 10 mM cuprizone stock solutions were prepared in DMEM/F12 containing

1% P/S and 25% EtOH. Serum (10% FBS) was added to the cuprizone solutions used in the mitochondrial swelling experiments. These stock solutions were sonicated in a heated

FS-14 Solid State/Ultrasonic (Fisher Scientific, USA) water bath until fully dissolved.

24 During this period, the solution was vortexed for 30 sec after 30 min increments. For the treatment of MO3.13, the 10 mM cuprizone stock solutions were diluted to a 1 mM final concentration.

Oxygen Consumption Measurements:

MO3.13 cell solution was incubated on a 3.2 µL volume glass well plate for 15 minutes in DMEM without FBS. Media was removed and replaced with DMEM containing 1 mM cuprizone and 2.5% ethanol. The glass well plate was sealed using a

#1.5 thickness 22x22mm coverslip, pre-treated with FluoSpheres carboxylate modified

.04 µm Pt luminescent beads on the well-facing side. Oxygen consumption was monitored via the fluorescent lifetime of the beads (ex 390/em 650), collected every 30 seconds. Detection was accomplished using an Andor iStar CCD detector DH743-18-F-

A3 with a BNC Model 557 Pulse Generator providing excitation pulses. After oxygen consumption measurement was completed, total cell count was taken via imaging the entire well.

Oxygen Consumption Measurement Analysis:

Oxygen consumption data was collected as a ratio of fluorescence lifetime, which was converted to PPM oxygen. Each dataset was truncated to its most linear portion, excluding the nonlinear portions at the beginning of the experiment and at the end when oxygen content became limited. The datasets were converted to terms of change in PPM oxygen by subtracting the initial PPM value of each truncated dataset from all points

(Fig. 2.4). All datasets for each condition were combined into unified datasets for visualization and slope comparison (Fig. 2.5).

25 Change in PPM Oxygen Over Time for Cuprizone Treated Cells 0.0002 0 -0.0002 0 5 10 15 20 25 30 -0.0004 y = -6.76E-05x + 3.99E-05 R² = 9.22E-01 -0.0006 -0.0008 -0.001

Delta PPM Delta -0.0012 -0.0014 -0.0016 -0.0018 -0.002 Time (Min)

Figure 2.4: Graph of oxygen consumption in terms of PPM oxygen consumed compared to starting level for a single trial. Best fit line applied to show the calculated slope value.

Change in PPM Oxygen Over Time for Cuprizone Treated vs Control Cells 0.001 y = -7.33E-05x - 2.31E-07 0 R² = 8.93E-01 0 10 20 30 40 50 -0.001 Control -0.002 Cuprizone Linear (Control) Delta PPM Delta -0.003 Linear (Cuprizone) -0.004 y = -6.08E-05x + 2.26E-07 R² = 6.68E-01 -0.005 Time (Min)

Figure 2.5: All trials oxygen consumption over time normalized and combined for cuprizone treated and normal MO3.13 cells. Best fit lines have been applied to both sets of data, with the equation for control cells appearing above the data points and cuprizone appearing below the data points. Mitochondrial Size Measurements:

26 MO3.13 cells were treated with either 1 mM cuprizone in DMEM containing

2.5% ethanol. Treated cells were stained with MitoTracker® Red FM dye

(ThermoFisher) 30 minutes prior to imaging. Individual cells were imaged from top to bottom via 0.5 µm step z-stack. Image data was analyzed to compare mitochondrial size between cells prior-to and after treatment with cuprizone over the course of 8 hours.

MO3.13 cells were imaged after exposure to identical media conditions without cuprizone for 24 hours as a control.

27 CHAPTER III

DEVELOPMENT OF ANTIBACTERIAL SUSCEPTIBILITY METHOD

Introduction:

RSG was considered for AST method development based upon its use as a live- dead stain and ties to bacterial reductase activity. RSG is supplied in an oxidized non- fluorescent form, but when reduced via bacterial reductases it fluoresces following excitation by 488 nm light. The intensity of the RSG signal can therefore be considered a marker for bacterial reductase activity in healthy cells. Multiple antibiotics with unrelated targets have been implicated in the production of reactive oxygen species or adversely impacting electron transport chain function (Kottur et al., 2016). If such changes could be detected via RSG we reasoned that it would provide a generalized mechanism for antibiotic resistance detection using a single dye. RSG is marketed as a flow cytometry dye but can dynamically monitor respiration response in live single cell imaging

(Konopka et al., 2011, Kalyuzhnaya et al., 2008).

Results:

A protocol for staining and imaging E. coli using RSG was established using previous studies as a basis (Konopka et al., 2011, Kalyuzhnaya et al., 2008). We first used Mattek 35 mm glass bottom dishes to image E. coli stained with RSG and maintained the microscope at 37° C using a temperature control cage. These open top dishes with poly-L-lysine pre-coated bottoms allowed for the addition and removal of media while cells were retained through interaction with the poly-L-lysine. When

28 subcultured in the presence of RSG at staining concentration in the dark, cells grew at a normal rate and showed typical RSG fluorescence levels when imaged on slides and in dishes. We established that RSG-stained E. coli were able to grow to confluence in dishes when additional fresh media containing RSG was added. Stainless steel Attofluor™ cell chambers were introduced to lower the cost per experiment with only the poly-L-lysine coated coverglass at the bottom of the chamber needing to be exchanged each time.

During imaging plastic lids were employed to prevent the loss of solution to evaporation.

ε poly-L-lysine, a poly lysine polymer with 25-35 lysine residues, has strong antimicrobial properties which target the outer membrane of E. coli (Hyldgaard et al.,

2014). Poly-l-lysine coatings used for cell culturing typically employ higher weight formulations between 5,000 to 300,000 Daltons to avoid toxicity. Coatings of high weight poly-l-lysine can still cause unwanted effects on the division and viability of E. coli depending primarily upon coating thickness (Colville et al., 2010). We used a 70,000 to 150,000 Dalton formulation provided as 0.01% sterile-filtered solution to soak the coverslips used for each experiment, then thoroughly rinsed the surfaces with sterile ultrapure water before air drying to thin the coating. When using a coating which was dried without rinsing, elongation of the E. coli occurred consistent with disruption of bacterial cell division (Fig. 3.1).

29 Figure 3.1: Elongation of E. coli strain MG1655 in response to thick unrinsed poly-L-lysine coating. To test the RSG signal changes in response to antibiotic treatment in real time the

RSG-stained cells had to be imaged prior to treatment, then at fixed periods following treatment with an antibiotic. K-12 MG1655 strain E. coli was used to as a starting point for bacterial testing as a commonly used minimally manipulated lab strain. Mueller-

Hinton Broth (MHB) is the gold-standard for AST and is endorsed by the Clinical

Laboratory Standards Institute (CLSI, 2020). To limit fluorescent background noise and cell movement during experimentation we used MOPS minimal media for strain growth and imaging. Minimal medias produce slower growing bacteria which are not as motile, making them easier to image and characterize. E. coli was grown to log phase at 37° C to ensure cells were in an active growth phase during testing. Stationary phase and starving cells may activate the stringent response which imparts additional resistance to antibiotics

(Gilbert et al., 1990). E. coli were stained with RSG with shaking at 37° C for 15 minutes prior to microscopy to allow signal to stabilize. All bacteria treated with antibiotics were

30 tested using broth macrodilution and zone of clearance analysis to independently verify susceptibility to the antibiotics used.

We first used ampicillin to test the RSG signal response of cells treated with antibiotics. Ampicillin is a β-lactam antibiotic which targets penicillin binding proteins to inhibit transpeptidation during synthesis, thereby preventing cell wall formation. 20 µL of cells stained with 1 µM RSG were added to an Attofluor™ chamber directly via pipette and allowed to adhere to the poly-L-lysine coating for 5 minutes before imaging began. 1 mL of MOPS minimal media containing 1 µM RSG and 100

µg/mL ampicillin, pre-warmed to 37° C, was added directly via pipette with gentle mixing. A high concentration of ampicillin was used compared to the MIC of 16 µg/mL in order to encourage the strongest possible response from the cell. Cells were manually imaged for 120 minutes at intervals of 15 minutes. A notable increase in RSG signal intensity was observed for some cells, with the signal saturating the detector (Fig. 3.2).

This experiment was repeated with the addition of PI to serve as a dead stain to provide context (Fig. 3.2). We hypothesized that the RSG signal increase was occurring prior to the appearance of PI signal and was a direct result of antibiotic exposure.

31 Figure 3.2: Left: E. coli 60 minutes after 100 µg/mL ampicillin addition in the presence of 1 µM RSG. Yellow circles indicate cells which have increased in RSG signal (green) to saturation and appear to have swollen or burst. Right: E. coli 60 minutes after 100 µg/mL ampicillin addition in the presence of 1 µM RSG and 20 µM PI. Increased RSG signal is seen in many cells and is not generally overlapping with PI signal (purple). Yellow circles indicate cells with significant RSG and PI signal overlap. To confirm the nature of the RSG signal increase and evaluate it for use as a marker of antibiotic susceptibility, single cell tracking was employed. For these and future experiments, carbenicillin was used for its superior stability in solution compared to ampicillin. Physical chamber setup remained the same while data acquisition was automated. Cells were focused using the Nikon Perfect Focus System (PFS) to maintain focus during the experiment without the need for manual adjustment. The PFS uses an

870-nm LED and CCD system to track the focal plane and the user-set sample position.

We used a scan interval of 15 seconds for collecting images with a collection time per frame of roughly 2 seconds. This allowed for the possibility of collecting data from up to

2 separate locations in each chamber imaged, with travel time and automatic refocusing between locations being the major limiting factor. After at least 2 frames were collected antibiotic treatment was added to the Attofluor chamber as before. Care was taken to

32 dislodge as few cells as possible during additional media addition. Cells were tracked manually using the FIJI ImageJ software analysis suite (Schindelin et al., 2012; 2015).

Thresholding was used on the first frame of data collection to establish a baseline level of cell signal to compare against. Signal threshold was lowered until all visible cells were highlighted. These were analyzed automatically using the “Analyze Particles” command, with a selection criterion of >5 pixels area to generate region of interest (ROI) selections (Fig. 3.3). The mean signal intensity was taken for all cells and compared to the signal intensity of cells undergoing RSG signal increase following ampicillin treatment. Mean initial signal multiplied by a factor of 5 was established as the signal threshold for antibiotic action (Fig. 3.4). Complete data analysis automation was explored but ultimately abandoned. Small particle size, overlapping cell bodies, and adjacent cells frustrate approaches which attempt to track unique cells. High levels of cellular movement across the frame in both the z and x-y axis cause cells to be measured multiple times in the absence of scrutiny.

33 Figure 3.3: Upper left: E. coli highlighted in red are above the signal threshold of 152 out of a maximum of 4095 (arbitrary units). Upper right: Selections made by the “Analyze Particles” command on the same image. Lower: Signal intensity results reported for all ROI selections with mean signal of 365 listed across all 320 cells.

34 Figure 3.4: Threshold determination for a single cell. Yellow circle on first frame indicates the cell of interest. Some signal is seen above the threshold in red on the second frame but does not reach the threshold of 5-pixel area for cell tracking. Third frame shows an ROI selection around the cell indicating signal above the threshold and the cell being counted as having increased RSG signal. Single cells tracked over time for carbenicillin treatment showed increases in

RSG signal preceding PI signal. Some cells which increased in RSG signal also showed characteristic bulging and swelling consistent with β-lactam antibiotic action (Fig 3.5)

(Yao et al., 2012). Bulging cells followed a pattern which consisted of stabilization of bulge size, followed by membrane bursting which resulted in immediate loss of elevated

RSG signal and appearance of PI signal. For this initial trial 269 of an estimated 565 identified cells reached RSG signal levels above the threshold. These cells decreased in

RSG signal below the threshold level following the appearance of PI signal. The remainder of the cells did not increase in RSG signal and instead decreased in RSG signal below initial levels. Mean time to RSG signal increase was found to be 54 minutes (Fig.

3.6). This trial represented a high cell density with over 500 cells within a 210 µm x 210

µm space from a culture at 0.5 OD. Based upon MG1655 growth curve measurements, we determined that 0.5 OD to 0.6 OD is near the very end of the logarithmic phase in

MOPS (Fig. 3.7). We sought to test cultures in future experiments between 0.1 and 0.3

OD to ensure testing logarithmic phase cells.

35 Figure 3.5: Characteristic bulging seen following treatment of MG1655 E. coli with 100 µg/mL carbenicillin. Scale bar represents 5 µm.

Time Distribution of E. coli Increasing in RSG Signal Intensity Following Carbenicillin Treatment 60 52 50 47 38 40 30 30 27 22 20 14 14 11 Number of Cells of Number 10 7 7

0 0 10 20 30 40 50 60 70 80 90 100 10 Minute Time Bins

Figure 3.6: Distribution of MG1655 E. coli reaching RSG signal increase threshold following treatment with 100 µg/mL carbenicillin. Each bar represents a 10- minute time binning starting from 0 – 10 minutes and ending at 100 – 110 minutes.

36 Growth Curve Log Scale MG1655 E. coli in MOPS 10

4.12 4.51 3.07 3.22 2.57 2.712.88 2.95 2.1172.37 2.486 1.447 1.216 1.277 1 0 100 200 0.725 300 400 500 600 0.537 0.343 0.243 0.145 0.113 0.1 0.094 0.061 0.052

OD (log scale) (logOD 0.037 0.028 0.015 0.015 0.017 0.01 0.012 0.007 0.005 0.005 0.003 0.003 0.002 0.001 Time Post Subculture (Minutes)

250 1000 10000

Figure 3.7: Growth curve of MG1655 E. coli in MOPS minimal media for 3 different subculture dilutions in log scale. The dilutions are, left to right, 1:250, 1:1000, and 1:10000. Data points of OD are labeled. Straight portions of the curves indicate logarithmic phase growth. To ensure that the applied poly-L-lysine coating was not necessary to observe changes in RSG signal, the experiment was replicated on bare glass without the poly-L- lysine coating. We did not observe a difference in the characteristics of the antibiotic induced RSG signal changes (Fig. 3.8). 238 of an estimated 476 total observed cells underwent RSG signal increase above the threshold value. Increased movement of cells in solution made multiple replications of this condition difficult to achieve.

37 Time Distribution of Cells Increasing in RSG Signal Intensity Following Carbenicillin Treatment Without Poly-L-lysine 80 73 75 70 60 50 40 35 30 21 20

Number of Cells of Number 12 6 9 10 1 3 2 1 0 0 10 20 30 40 50 60 70 80 90 100 10 Minute Time Bins

Figure 3.8: Distribution of MG1655 E. coli reaching RSG signal increase threshold following treatment with 100 µg/mL carbenicillin without the presence of poly- L-lysine. Each bar represents a 10-minute time binning starting from 0 – 10 minutes and ending at 100 – 110 minutes. Microfluidics was approached to improve method reliability. Microfluidic devices have the advantage of controlling cell movement while providing precise antibiotic delivery. We used the CellASIC™ ONIX Microfluidic Platform for live cell imaging combined with a B04F Microfluidic Bacteria Plate. The ONIX system provided constant flow of media into the plate and switch between multiple media conditions as needed.

The microfluidic plate trapped logarithmic phase E. coli grown in MOPS minimal media within a chamber segregated by differing height PDMS silicone pillars. Once immobilized, fresh MOPS media containing RSG was flowed through the chamber to stain the cells prior to antibiotic administration. Cells stained with RSG in the microfluidic system did not achieve appreciable levels of RSG signal above background fluorescence. We attempted to address this by pre-staining cells with RSG. Pre-stained cells did not maintain their initial level of fluorescence and decreased in fluorescence despite RSG being present in the media flowing through the chamber. RSG signal

38 increase was observed during long term trials of cell growth without antibiotic due to unrelated cell death by chamber overcrowding but did not appear until several hours of growth (Fig. 3.9). Carbenicillin treatment did not produce increases in RSG signal but did result in cell death as indicated by PI signal. High levels of background fluorescence we observed upon addition of RSG to the microfluidic system for any trials. RSG is likely absorbed and sequestered by the PDMS silicone contained within the plate, reducing the amount of RSG in cells within the system.

39 Figure 3.9: MG1655 E. coli grown in a B04F Microfluidic Bacteria Plate with MOPS minimal media flowing through containing 1µM RSG. RSG signal can be seen to be low for the upper two panels through 2 hours of growth. By 8 hours increased RSG signal can be seen in the lower left panel for select cells which are dying as the plate becomes overcrowded. The lower right panel shows the crowding of cells as seen via brightfield. Different antibiotics were evaluated for use with RSG using the Attofluor chamber method. Kanamycin was tested at 100 µg/mL using an identical setup to carbenicillin. Kanamycin binds to the 30S subunit of the E. coli ribosome and causes the production of non-functional protein in the cell. Kanamycin did not produce a similar increase in RSG signal but instead saw a decrease in RSG signal over the course of 2

40 hours (Fig. 3.10). Death in kanamycin treated cells also proved difficult to detect with PI signal, with the majority of the cells not showing PI signal after 2 hours. While a consistent decrease in RSG signal could be used as an antibiotic susceptibility marker we considered it too ambiguous to use for AST. Norfloxacin was similarly evaluated at 5

µg/mL and found to decrease RSG signal but attain PI signal (Fig. 3.11). Norfloxacin inhibits DNA gyrase to prevent E. coli from dividing correctly. These results led to a consideration of the differences in antibiotic action that would lead to an RSG signal increase for carbenicillin, but not for norfloxacin or kanamycin. All three antibiotics have been implicated in production of ROS which could theoretically impact RSG signal

(Lobritz et al., 2015). Carbenicillin differs from the other antibiotics by virtue of targeting cell wall machinery. Because the previously observed increases in RSG signal during carbenicillin treatment occurred in time with cell bulging and membrane stress, we suspected that the RSG signal increase could be related to cell membrane and cell wall instability.

41 Figure 3.10: MG1655 E. coli treated with 100 µg/mL kanamycin. Left panel shows RSG signal (green) pre-addition. Right panel shows RSG signal 2 hours post- addition.

42 Figure 3.11: MG1655 E. coli treated with 5 µg/mL norfloxacin. Upper left panel shows RSG signal (green) pre-addition. Upper right panel shows RSG signal 2 hours post-addition. Lower panel shows PI signal (purple) 2 hours post-addition. To test the role of cell wall and cell membrane stability in antibiotic induced RSG signal increase we employed polymyxin-B (PMB). PMB is a powerful antibacterial which first binds to the outer membrane of Gram-negative organisms, then causes a disruption of both the outer and inner membranes, resulting in cell death (Derris et al.,

2014). The action of PMB is direct, physical, and does not rely upon enzymatic action to

43 complete its action unlike the previously tested antibiotics. MG1655 E. coli treated with 5

µg/mL PMB underwent RSG signal increase rapidly following treatment with all cells increasing in signal uniformly within 5 minutes (Fig. 3.12). 227 of 227 tracked cells underwent RSG signal increase. The uniform RSG signal increase is likely driven by the of PMB which is not sensitive to or effected by the physiological state of the cells which it attacks. While β-lactams and fluoroquinolones are less effective against slowly growing genetically susceptible cells such as persister cells, PMB should be equally effective against all cells which are genetically susceptible (Lewis, 2005).

Figure 3.12: MG1655 E. coli treated with 5 µg/mL PMB. Increase in RSG signal (green) is seen at 5 minutes in the center panel. Increased RSG disappears by 15 minutes in the right panel. Polymyxin-B can be broken down from its base form to only the cyclic cationic portion which is known as polymyxin-B nonapeptide (PMBN) (Fig. 3.13). PMBN is extremely specific to the outer membrane of Gram-negative organisms and inhibits their growth in a bacteriostatic manner at high concentrations, with notable activity against

Pseudomonas aeruginosa (Tsubery et al., 2000). The MIC for E. coli for PMBN is in most cases >500 µg/mL (Tsubery et al., 2001). We tested PMBN to further investigate if outer membrane destabilization alone was enough to cause an RSG signal increase. We immediately encountered a contradiction compared to the literature when exposing

44 MG1655 E. coli to 5 µg/mL PMBN, 1 µM RSG, and 20 µM PI. Cells experienced both an RSG increase identical in nature to that seen with PMB, followed by a loss of RSG signal and the appearance of PI signal consistent with cell death (Fig. 3.14). As a bacteriostatic antibiotic specific to the outer membrane PMBN should not destabilize the inner membrane enough to allow PI to enter the cell. We ran an identical experiment without the addition of PI and observed the same increase in RSG signal, but never observed a decrease in RSG signal. The increased RSG signal instead persisted for more than 8 hours without fading (Fig. 3.15). We considered that RSG could be hampering cell growth and contributing to bacterial stasis in combination with the PMBN. We grew

MG1655 E. coli in the absence of RSG with 5 µg/mL of PMBN and confirmed that it did not cause growth inhibition alone. We attempted the trial again while reducing the RSG concentration to 0.1 µM to determine if the action was concentration dependent. We observed cell growth in combination with increased RSG signal across 2 hours of growth

(Fig. 3.16). This should not be considered a false positive regarding using RSG signal increase for AST, as the increase correctly indicates a susceptibility to the permeabilization activity of PMBN which can be used to increase the potency of non- effective antibiotics.

45 Figure 3.13: Upper: Structure of PMB with a circle indicating the portion removed to generate PBMN. Lower: The structure of PMBN (Sigma-Aldrich).

46 Figure 3.14: MG1655 E. coli treated with 5 µg/mL PMBN with 1 µM RSG and 20 µM PI. Left panel shows RSG signal (green) pre-treatment. Middle panel shows RSG signal 5 minutes post-treatment. Right panel shows PI signal (purple) 1 hour post- treatment.

Figure 3.15: MG1655 E. coli treated with 5 µg/mL PMBN with 1 µM RSG. Left panel shows RSG signal (green) 1 minute post-treatment. Middle panel shows RSG signal 10 minutes post-treatment. Right panel shows RSG signal 3 hours post-treatment.

Figure 3.16: MG1655 E. coli treated with 5 µg/mL PMBN with 0.1 µM RSG. Left panel shows RSG signal (green) pre-treatment. Middle panel shows RSG signal 1 hour post-treatment. Right panel shows RSG signal 2.5 hours post-treatment. Include a section here about experiment 09/08/2017, which showed than an active proton gradient is not required in order to see the RSG signal increase because it occurred

47 in the presence of CCCP. We also tested with Sytox Blue in the 08/23/2017 dataset which gives secondary confirmation that we are seeing increase in signal occurring at approximately the same time as RSG. We used Sytox Blue because it prevents any signal leaching into the channel from RSG without needing to collect data in channel series which slows down data collection. 10/4/7 experiment more importantly used CSF instead of growth media to show that it can be done in a human fluid analogue.

Using RSG signal increase to develop an AST methodology requires the signal increase to be applicable to more than one strain of E. coli. Ideally the methodology developed herein would be applied to the full variety of infectious Gram-positive and

Gram-negative organisms. We first explored the applicability of the RSG signal increase to another strain of E. coli. E. coli ATCC® 25922 differs from MG1655 in containing the

O-antigen, significantly changing its outer membrane structure. The absence of the O- antigen makes the LPS layer rough and more hydrophobic compared to a smooth LPS layer which is more difficult for hydrophobic antibiotics to penetrate (Tesujimoto et al.,

1999). 25922 E. coli were treated with 100 µg/mL carbenicillin, 1 µM RSG, and 20 µM

PI in the same manner as MG1655 E. coli. RSG signal increase was observed to follow the same pattern as previously discussed.

Pseudomonas aeruginosa is a notable Gram-negative organism which creates difficult to eliminate biofilms containing persister cells. Biofilms created by P. aeruginosa colonize hospital equipment such as endotracheal tubes used to mechanically ventilate patients (Guillon, 2018). In collaboration with Dr. Abraham Joy’s lab in the

Department of Polymer Science we tested a polymer antibiotic fabricated by Dr. Elaheh

Chamsaz which prevents P. aeruginosa from colonizing surfaces. (Chamsaz et al., 2017).

48 We found that P. aeruginosa underwent RSG signal increase upon contact with the antibiotically active cationic coumarin coating (Fig. 3.17). RSG is also seen to penetrate tight groupings of these cells which constitute small pockets of biofilm.

Figure 3.17: P. aeruginosa treated with an antibacterial polymer coating in the presence of 1 µM RSG and 20 µM PI. Upper left shows initial RSG signal (green) following deposition of the cells onto the antibacterial coating with a yellow circle indicating the presence of a cluster of cells in a biofilm. Upper right shows RSG signal after 5 minutes in contact with the polymer surface. Lower shows RSG signal after 30 minutes in contact with the polymer surface. Increased RSG signal upon antibacterial exposure was also observed in Gram- positive Staphylococcus aureus. Vancomycin (100 µg/mL) induced an increase in RSG fluorescence signal in the S. aureus preceding cell death (Fig. 3.18). Single cell tracking

49 proved infeasible for S. aureus due to high cell motion in the z-axis and the bundling nature of growth.

Figure 3.18: Staphylococcus aureus stained with RSG (green) and PI (purple) and treated with 100 µg/mL vancomycin. Time course shows progression from pre-addition of vancomycin to 8 hours post-addition. Discussion:

RSG offers an advantage over PI and other dyes in several areas. PI signal only appears post cell death and may be difficult to interpret and track across the experiment.

PI tends to dilute out in dead cells, creating a smeared signal which is difficult to interpret. PI does not offer any information on the health of the cell prior to death. With

RSG, cells which are experiencing high levels of signal prior to antibiotic treatment can be identified and excluded.

Other dyes were considered for use in developing a single-cell fluorescence-based

AST method. PI was used throughout the development of RSG as an AST methodology as an independent confirmation of cell death. PI did not prove suitable for use in AST due primarily to its ability to actively impact the lethality of antibiotics. When tested on E. coli in combination with PMBN, PI was found to cause cell death despite the natural resistance of E. coli to PMBN. With RSG it was found that increased signal effects could

50 be effectively modulated by concentration. PI also proved more difficult to track on a single cell basis compared to the RSG signal increase. PI signal was often diluted upon β- lactam induced cell death by spilling out of cells with poor inner membrane integrity. For cells which remained intact enough to sequester PI, the long retention time for the signal posed an additional problem when attempting to differentiate dead cells which would float into the signal space of other cells. Because RSG signal is reduced after cell death, dead cell movement did not present the same issue compared to PI. Issues with PI signal consistency and reliability are well known and can be especially unreliable with respect to exponentially growing cells such as in Sphingomonas and Mycobacterium frederiksbergense where up to 40% of normally growing cells can be stained in a healthy culture (Shi et al., 2007).

The limitations of the current methodology are numerous with respect to employment in a clinical setting. The current percentage of RSG signal increase for susceptible bacteria is unacceptably low for enzymatically targeted antibiotics.

Carbenicillin saw increased RSG signal in between 50-70% of susceptible E. coli populations across all the experiments collected. PMB in contrast saw signal increase in every cell that we were able to track across all the experiments collected. It is tempting to conclude that increases in signal for E. coli treated with carbenicillin are simply being missed in the 15 seconds between every frame collected. There is evidence to suggest that some signals are missed in the form of cells we have detected which experience RSG signal increase above the threshold for a single frame before decreasing in signal precipitously in the next. The more likely culprit is that of a sub-population of cells which simply do not meet the criteria to undergo RSG signal increase prior to death. The nature

51 and cause of RSG signal increase has not been entirely elucidated by our study and can only be guessed at. Our data suggests a dependence on the destabilization of the outer membrane to drive the RSG signal increase in E. coli, independent of ETC function. A possible interpretation of non-increasing cells would therefore be a sub-population which experiences inner membrane destabilization prior to outer membrane destabilization, resulting in cell death as indicated with PI without RSG signal increase. Experiments completed in the absence of PI show that such an effect is not due to PI presence, as some cells were observed to decrease in RSG signal relative to initial levels without signs of growth or life.

Single cell tracking in the currently prescribed format is far too cumbersome to be employed on a clinical level. Cells were tracked primarily on a manual basis, with each frame of each experiment reviewed to account for cell movement. Improvement in cell tracking is difficult to approach in open cell chambers with small and motile bacteria.

Automated particle tracking for each cell would have gone a long way towards automating the process, but multiple hurdles stood in the way of employment for this work. Particle tracking would have to be based on both the transmitted light (TD) image and the RSG channel image due to having low signal to noise in the RSG channel by design to allow for a high signal ceiling. Tracking at the low 512 x 512 resolution of the images where individual cells can occupy as few as 10 pixels total presents an additional challenge. Tracking would have had to account additionally for cell movement, preserving the identity of each cell over time regardless of motility, and discard cells which “fell” into frame from a higher out of focus z-axis position. Cells which were undergoing division or changing shape such as bulging cells present an additional

52 challenge due to appearing segmented but only representing one cellular body. An increased image resolution of 1024 x 1024 was approached as a solution, but generated file sizes of 1 gigabyte per 30 minutes of collected data without the possibility of collecting data from multiple x-y positions in parallel.

The most direct solution to simplify cell tracking would be a move to a more chemically inert microfluidic flow system. RSG is marketed for flow cytometry and should maintain stable initial signal under flow, unlike what was observed using the

ONIX system. A PDMS-free microfluidics system which immobilizes bacteria for microscopy would suffice to observe RSG signal dynamics for laboratory isolates. If combined with a cell sorting system, clinical isolate samples could be tested directly by capturing bacteria within the microfluidics system and subsequently treating them with candidate antibiotics in the presence of RSG. Clinical isolates to be tested for susceptibility are primarily urine and blood samples which contain a mix of bacteria and host cells. Blood represents the most difficult medium to separate due to the incredible disparity between blood cell concentration and bacterial concentration, with only 1 to 100 bacteria per mL of whole blood (Reimer et al., 1997). While commercially available solutions for direct isolation of bacteria from clinical isolates do not currently exist, they may become available in the near future. Centrifugation can rapidly separate bacteria by sedimentation under the correct conditions, but recovery rates around 69% mean the possibility of missing resistant subpopulations is too great (Buchanan et al., 2017).

Promising methods have been suggested using microfluidics and electrophoretic force, selectively targeting the charged bacterial membrane. Bacteria have been isolated at up to

53 91.3% yield with controlled dielectrophoretic force, using low energy pulses to avoid impacting cell viability as determined by colony streaking tests (Yoon et al., 2019).

With further development RSG AST could provide results in a fast, efficient, and cheap manner, rivaling currently available methods. Single cell antibiotic susceptibility detection methods provide more accuracy than bulk measurements because of the ability to identify subpopulations. With appropriate microfluidic design parallelization is limited only by sample quantity for this method, allowing multiple candidate antibiotics to be tested simultaneously. RSG is a commercially available dye with a low cost of 0.46 USD per test at a concentration of 0.1 µM based upon ThermoFisher pricing at the time of writing. Our method represents a proof-of-concept for antibiotic susceptibility testing using single-cell tracking, fluorescence microscopy, and fluorescent reporter dyes.

54 CHAPTER IV

RSG AST VALIDATION

Introduction:

Given the method of RSG AST developed in the previous chapter, we investigated the reproducibility and validity of the method as an alternative to extant methods. Reproducibility is key in tests for antibiotic susceptibility to eliminate ambiguity in results that may affect treatment outcomes. We define reproducibility within the scope of our method to be tied to the ability to observe RSG signal changes consistently whenever bacteria are treated with an antibiotic that they are susceptible to.

One limitation on reproducibility already mentioned is that of cells which do not increase in RSG signal following treatment with enzymatically targeted antibiotics such as carbenicillin.

RSG signal increase in treated bacteria has been characterized thus far by the time past treatment when the signal intensity reaches a set threshold. For the time distribution of increased RSG signal to be meaningful, it must be reproducible and relevant to antibiotic efficacy. Direct and enzymatically targeted antibiotic action should consistently be able to be differentiated from each other based upon signal timing. This establishes the ability of the method to differentiate the speed of action of a given antibiotic on a given bacteria, thereby delineating efficacy in terms of exposure time needed to affect bacterial integrity. Novel antibiotics are currently evaluated by MIC when considering effectiveness. RSG AST could evaluate the efficacy of novel antibiotics in the time

55 regime as well, highlighting drugs which would need the least circulation time to eliminate infection.

Results:

Testing MG1655 E. coli for susceptibility was explored through multiple trials to confirm the reproducibility of the method. The time distribution of RSG signal increase in MOPS minimal media and carbenicillin treatment across 4 trials and 749 cells remained below 100 minutes for 99.2% of cells which saw signal increase (Fig. 4.1). The average time to reach RSG signal threshold following treatment was 45.1 minutes. Time distribution of the signal increase was significantly different across the 4 trials according to single-factor ANOVA analysis. Averages for each individual trial were 30.8, 42.7,

43.9, and 54.3 minutes. Variations in the speed of antibiotic action are expected for carbenicillin. All 4 trials were conducted using logarithmic phase cells, but no two sets of cells grown at different times will be physiologically identical.

Time Distribution of Cells Meeting RSG Signal Threshold for Multiple MOPS Carbenicillin Treatment Trials 40 35 30 25 20 15 10 Percentage of Cells of Percentage 5 0 0 10 20 30 40 50 60 70 80 90 >100 10 Minute Time Bin

Figure 4.1: Graph of RSG signal increase time distribution for 4 trials of MG1655 E. coli treated with 100 µg/mL carbenicillin in MOPS minimal media.

56 We tested MG1655 E. coli approaching stationary phase at OD 1.1 to compare the time distribution to logarithmic phase cells. Cells which took over 100 minutes to reach

RSG signal threshold increased dramatically from 0.8% of total logarithmic phase cells to

13.8% of cells approaching stationary phase (Fig. x). These cells likely represent a subpopulation within the group of cells tested which have a different physiological state compared to the bulk. Because carbenicillin action is tied to cell growth, stationary phase cells which grow at a slower rate take longer to be affected.

Time Distribution of Cells Meeting RSG Signal Threshold for Stationary vs Logarithmic Carbenicillin Treatment 30.0 25.0 20.0 15.0 10.0 5.0 Percentage of Cells of Percentage 0.0 0 10 20 30 40 50 60 70 80 90 >100 10 Minute Time Bin

Stationary Logarithmic

Figure 4.2: Graph of RSG signal increase time distribution for stationary phase and logarithmic phase MG1655 E. coli treated with 100 µg/mL carbenicillin in MOPS minimal media. Rich media was tested to determine if RSG signal increase time distribution would change with increased nutrient availability and logarithmic phase cell growth rate.

E. coli underwent greater movement within the Attofluor™ chambers when grown and tested in rich LB media compared to MOPS, which resulted in increased difficulty with single cell tracking. 2 suitable trials with 426 cells total were able to be tracked and evaluated (Fig. 4.3). The average time to reach RSG signal threshold following treatment

57 was 27.8 minutes. Comparing LB media to MOPS minimal media averages highlights the decrease in time needed to reach the RSG signal threshold (Fig. 4.4).

Time Distribution of Cells Meeting RSG Signal Threshold for Carbenicillin Treatment Trials in LB Media 60

50

40

30

20

Percentage of Cells of Percentage 10

0 0 10 20 30 40 50 60 70 80 90 100 10 Minute Time Bins

Figure 4.3: Time distribution of MG1655 E. coli treated with 100 µg/mL carbenicillin reaching RSG signal threshold in LB media. Two separate trials are shown.

Time Distribution of Cells Meeting RSG Signal Threshold for MOPS vs LB Carbenicillin Treatment 50.0

40.0

30.0

20.0

10.0 Percentage of Cells of Percentage 0.0 0 10 20 30 40 50 60 70 80 90 100 10 Minute Time Bins

MOPS LB

Figure 4.4: Time distribution of MG1655 E. coli treated with 100 µg/mL carbenicillin reaching RSG signal threshold, with MOPS minimal media being compared to LB rich media. MOPS minimal media is shown in blue, LB rich media is shown in orange.

58 PMB trials were compared to test the variation in RSG signal increase timing during treatment. RSG signal increase in both trials was seen within 10 minutes of treatment with PMB (Fig. 4.5). In these experiments it appears that the action of PMB against E. coli may be diffusion limited as the RSG signal increase occurs rapidly following addition. Comparing PMB to carbenicillin treatment shows clear delineation between a directly acting antibiotic and enzymatically acting antibiotic (Fig. 4.6).

Time Distribution of Cells Meeting RSG Signal Threshold for Polymyxin B Treatment 60

50

40

30 Trial 1 20 Trial 2

Percentage of Cells of Percentage 10

0 0 1 2 3 4 5 6 7 8 9 10 Minutes Post Antibiotic Addition

Figure 4.5: Time distribution of MG1655 E. coli treated with 5 µg/mL PMB reaching RSG signal threshold in MOPS media. Two separate trials are shown.

59 Figure 4.6: Time distribution of MG1655 E. coli treated with 5 µg/mL PMB compared to those treated with 100 µg/mL carbenicillin reaching RSG signal threshold in MOPS media. Carbenicillin is shown in blue, PMB in orange. The inset graph shows a further breakdown of the distribution of RSG signal timing within the 0 – 10 minute time binning. RSG AST was tested using bacteria with resistance to the antibiotics which they were treated with to evaluate the potential for false positive detection. MG1655 E. coli containing a PCR4-Topo vector conferring ampicillin and kanamycin resistance were prepared via transformation. To prevent pre-exposure to ampicillin, the cells were grown in kanamycin to ensure plasmid retention. When exposed to ampicillin for 30 minutes, cells did not undergo characteristic RSG signal increase. After 30 minutes the cells were exposed to PMB to provide a positive control and the RSG signal for all cells increased as expected (Fig. 4.7).

60 Figure 4.7: Top left: Ampicillin and kanamycin resistant MG1655 E. coli RSG signal prior to addition of 100 µg/mL ampicillin. Top right: RSG signal after 30 minutes of exposure to ampicillin, immediately prior to PMB addition. Bottom: RSG signal 3 minutes after PMB was added. Persister cells represent a troubling subset of cells which are generally only susceptible to directly acting antibiotics due to their slow cell growth, metabolism, and respiration rate. To investigate whether RSG AST could detect susceptibility in these types of cells we utilized carbonyl cyanide m-chlorophenyl hydrazone (CCCP) to uncouple the electron transport chain in E. coli. CCCP chemically uncouples the electron transport chain by acting as a protonophore and disrupts ATP synthase by dissipating the proton motive force. CCCP treated bacteria are functionally similar to the persister cell phenotype (Grassi et al., 2017). MG1655 E. coli treated with 10 µM CCCP for 30

61 minutes were still found to undergo RSG signal increase upon exposure to 5 µg/mL PMB

(Fig. 4.8).

Figure 4.8: Left: MG1655 E. coli pre-incubated with 10 µM CCCP, then stained with RSG. Right: The same cells, 5 minutes following treatment with 5 µg/mL PMB. Discussion:

RSG AST provides a method by which to determine the efficacy of antibiotic action in terms of time. The timing of RSG signal increase for carbenicillin was shown to depend upon the growth rate of the E. coli tested. When cells approached stationary phase, the slower rate of bacterial cell division led to a larger population of cells requiring over 100 minutes to undergo RSG signal increase. When cells were grown in rich media instead of minimal media the faster rate of bacterial cell division led to faster RSG signal increase. We expect that other indirect action antibiotics which target the bacterial cell membrane and cell wall would behave similarly, such as vancomycin in S. aureus. Direct and indirect mechanisms of antibiotic action were seen to have easily differentiable distributions of RSG signal increase when comparing PMB to carbenicillin. The speed of

62 PMB action was replicable with a remarkably narrow window for the time of increase for all cells.

Persister cells which have low or completely dormant rates of cell growth would likely fail to undergo RSG signal increase at all in the presence of an indirectly acting antibiotic. Persisters would additionally have an extremely low level of cellular respiration, exacerbating any staining with RSG at all. E. coli treated with CCCP to uncouple the electron transport chain did show RSG signal increase when exposed to

PMB, indicating that directly acting antibiotics might be able to elicit an RSG signal response from persisters that are susceptible to them. This also indicates that the RSG signal increase mechanism seen during antibiotic exposure may not be driven by electron transport chain activity at all. RSG may undergo active transport out of healthy cells in the outer membrane, such as by drug efflux pumps, which is disrupted when the outer membrane is destabilized. The resulting influx of RSG into the cell may cause an increase in signal due to some portion of the RSG being reduced immediately upon contact with many reducing agents present in the cell.

It remains to be seen how effective the method is at detecting small subpopulations of resistant cells. In future experimentation, doping known ratios of resistant cells into cultures of non-resistant cells would provide a baseline for the sensitivity of detection for RSG AST. Sensitivity for indirect antibiotics would be expected to be poor due to susceptible cells do not always increase in RSG signal.

Comparative analysis would be required by testing multiple antibiotics to compare the percentage of cells which produced RSG signal increase for each antibiotic to determine the most appropriate antibiotic. For directly acting PMB however, the sensitivity could be

63 tied linearly with the sample size. As all observed susceptible cells appear to undergo signal increase, as long as sample size is sufficient to include at least one resistant cell then resistance should be detected.

64 CHAPTER V

RSG AST USING ANTIMICROBIAL POLYMERS

Introduction:

Antimicrobial peptides (AMPs) are found naturally in the immune responses of prokaryotic and eukaryotic organisms and disrupt hostile microbial growth. Structurally they vary considerably in α-helical and β-sheet content, with some peptides having no secondary structure in solution. AMPs have activity which is potent at low concentrations, driven by targeting the cell membrane and cell wall either as part of their anti-bacterial action or to allow translocation to the cytoplasm. To achieve their specificity AMPs use cationic amino acids to target negatively charged bacterial membranes and an amphiphilic structure to allow for membrane insertion. Nisin, originally isolated from the bacterium Lactococcus lactis, has been used as a food preservative for decades thanks to its broad-spectrum activity against Gram-positives into the parts per billion range. Nisin, like most AMPs, is difficult for susceptible organisms to acquire resistance to which accounts for its longevity of use. One exception is

Staphylococcus aureus which can become resistant by upregulating an existing ABC transporter which recognizes the molecule. In the absence of an existing transporter which can recognize the AMP, reorganization of the fundamental cellular membrane structure may be necessary to generate resistance. Naturally derived and synthetic AMP molecules suffer from in vivo toxicity when given orally and are primarily used for topical application i.e. Neosporin®.

65 Recently a set of antimicrobial peptide-mimetic were developed and characterized with strong activity against Gram-negative organisms and low hemotoxicity (Mankoci et al., 2017; 2019). These polyurethanes are direct membrane acting antibiotics which interact with both the outer and inner membrane of Gram- negative bacteria. Their structure consists of a repeating polyurethane backbone with varied functionalization using amino acid pendant groups (Fig. 5.1). 50%, 80%, or 100% of the polyurethane backbone was composed of subunits carrying a lysine-like monomer which provides specificity and action against negatively charged Gram-negative membranes, designated as the “m” subunit. The remainder of the subunits carried pendant groups which changed the overall charge, polarity, and hydrophobicity of the polyurethane molecules and thereby also altered specificity and efficacy, designated the

“n” subunits. Using RSG AST we were able to investigate the effect of differing pendant groups on antibiotic efficacy. We tested a selection of the polyurethanes whose structures have been previously published, named for convenience as 75 A, 81 C, 68 B, and 115 B

(Table 5.1). 75 A and 68 B were synthesized at a high and low molecular weight using exclusively lysine-like pendant monomers, with no “n” subunit. 81 C was 50% functionalized using valine-like pendant monomers, while 115 B was 20% functionalized using arginine-like pendant monomers. Valine, as a hydrophobic pendant group, should cause greater affinity for eukaryotic membranes and therefore causes increased toxicity to mammalian cells. Arginine, as a cationic pendant group, should increase the affinity for the anionic outer membrane of E. coli.

66 Figure 5.1: Polyurethane antimicrobial backbone illustrating the varied pendant group functionalization. Above indicates the polymer backbone structure with n and m subunits where m is always the mLysine pendant group. Below indicates the monomers being combined to yield the final polymer backbone structure for a containing the mValine and mLysine pendant groups, where OCN(CH2)6NCO is serving as the polyurethane backbone linker between them.

Polyurethane Table n monomer pendant group %n:%m ratio Molecular weight 75 A None, all mLysine 0% : 100% 30 kDa 68 B None, all mLysine 0% : 100% 6 kDa 81 C mValine 50% : 50% 8 kDa 115 B mArginine 20% : 80% 35 kDa

Table 5.1: Polyurethane polymers composition ratios and molecular weights.

67 These polymers were characterized for membrane penetrating potential based upon either β-galactosidase release or DiSC3(5) fluorescence (Mankoci et al., 2017;

2019). β-galactosidase activity assay provides a measure of membrane permeability based upon the leakage of cytoplasmic β-galactosidase into the bulk solution converting

ONPG into lactose following antibiotic challenge (Lehrer et al., 1988). This methodology does not provide single cell information and is not time resolved with respect to the moment membrane damage first occurs during antibiotic challenge. DiSC3(5) membrane staining offers membrane polarization information based upon the self-quenching nature of the dye when aggregated inside cells with polarized membranes (te Winkel et al.,

2016). The bacterial membrane becomes depolarized following antibiotic challenge, causing the signal of DiSC3(5) to increase in bulk solution when monitored by fluorimetry. This method also does not provide single-cell information, and DiSC3(5) is not amenable for antimicrobial susceptibility testing over time due to the toxic nature of the dye itself. A similarly voltage sensitive dye DiBAC4(3) is non-toxic and amenable to measurement of membrane potential over time, with fluorescence in cells increasing when membrane potential drops (Deere et al., 1995). Increase in membrane potential is not specific to depolarization caused by bacterial cell death however, and can mistakenly identify cells which have entered into stationary phase and are no longer maintaining a membrane potential strong enough to exclude the dye (te Winkel et al., 2016). The lack of a specific assay for antimicrobial membrane penetration highlights a need in this area of antibacterial development for an easy-to-use and specific technique.

Results:

68 Polyurethane polymers 75 A, 81 C, 68 B, and 115 B were tested in Mueller-

Hinton broth (MHB), in order to replicate previous conditions used to establish MIC for those polymers (Mankoci et al., 2017; 2019). We completed 2 RSG AST trials for each polyurethane antimicrobial tested for a total 748 cells (Fig. 5.2). The distribution of RSG signal increase timing for each polymer compared to each other was not significantly different for 3 of the polyurethanes, 75 A, 81 C and 68B (P-value 0.057 single factor

ANOVA; Table 5.2). Polyurethane 115 B (mean of 2.65 minutes required for RSG signal increase) was found to be significantly different in its timing distribution compared to 75

A, 81 C, and 68 B (P-value 0.0001 single factor ANOVA; Table 5.3). The increase in

RSG signal for cells treated with 115 B was slightly more rapid than that of the other 3 polyurethanes, consistent with its functionalization making it more attracted to the E. coli outer membrane. The slight increase in speed is not likely due to any increase in diffusion speed for 115 B, as it was the heaviest polyurethane tested.

69 Figure 5.2: Box and whisker plot of the RSG signal increase time distribution for the 4 polymer antimicrobials tested.

70 SUMMARY Groups Count Sum Average Variance 75 A 169 518.75 3.069527 1.630926 81 C 194 675.75 3.483247 3.753928 68 B 263 862.5 3.279468 2.594787

ANOVA Source of Variation SS df MS F P-value F crit Between Groups 15.47855 2 7.739276 2.872824 0.057288 3.010184 Within Groups 1678.338 623 2.693961

Total 1693.816 625

Table 5.2: Single factor ANOVA analysis of the RSG signal increase distribution between 75 A, 81 C, and 68 B.

SUMMARY Groups Count Sum Average Variance 75 A 169 518.75 3.069527 1.630926 81 C 194 675.75 3.483247 3.753928 68 B 263 862.5 3.279468 2.594787 115 B 122 323.5 2.651639 2.43446

ANOVA Source of Variation SS df MS F P-value F crit Between Groups 56.55816 3 18.85272 7.10952 0.000103 2.616873 Within Groups 1972.907 744 2.651757

Total 2029.466 747

Table 5.3: Single factor ANOVA analysis of the RSG signal increase distribution between 75 A, 81 C, 68 B, and 115 B. We compared the RSG signal increase distribution to that of PMB to highlight the slight difference between the two sets of directly acting antibiotics (Fig. 5.3).

Remarkably, all the polyurethanes acted faster to produce an RSG signal increase compared to PMB. Given that the polyurethanes have a much larger size, mixing

71 throughout the media upon antibiotic addition would be slower compared to PMB. The trials for PMB were conducted in MOPS minimal media, compared to the rich MHB media for the polyurethanes, which is a possible cause of the difference in signal timing.

RSG Signal Distribution of Polymer Antimicrobials Versus Polymyxin B 60.0 50.0 40.0 30.0 20.0 10.0 Percentage of Cells of Percentage 0.0 0 1 2 3 4 5 6 7 8 9 10 Minutes Post Antibiotic Addition

75 A 81 C 68 B 115 B PMB

Figure 5.3: Distribution of RSG signal increase timing for polyurethane polymer antimicrobials (32 µg/mL) compared to PMB (5 µg/mL). A Cy5 labeled version of polyurethane 75 A was prepared to investigate the mechanism of action of the antimicrobial polyurethanes in vivo. Cy5 is a far-red fluorescent dye, commonly conjugated to proteins and nucleic acids, which is far enough from RSG signal to avoid signal bleed when used together. MG1655 E. coli were treated with Cy5 75 A both in combination with RSG to observe the overlap signal development.

Cy5 75 A was seen to surround the cells concurrently with increasing RSG signal, outlining the cell completely (Fig. 5.4). The RSG signal was seen to shrink within the cell leaving some space between it and the Cy5 signal, suggesting the presence of a possible periplasmic space. The existence of a periplasmic space was also suggested by TD images of the same cells (Fig. 5.5). Apparent invaginations can be seen in the cells where

72 more light was able to shine through the cell to the detector. The appearance of this phenomenon was confirmed in cells treated with non-labeled 75 A in the absence of RSG as well.

Figure 5.4: Composite image of E. coli treated with 32 µg/mL Cy5 labeled 75 A polymer (purple) antimicrobial and RSG (green). Circled cell has been surrounded by 75 A, but Cy5 label signal does not yet appear within the cell where RSG signal is seen. Scale bar represents 2 µm.

73 Figure 5.5: TD image of E. coli treated with 32 µg/mL Cy5 labeled 75 A polymer antimicrobial. Areas which show apparent invaginations are circled. Scale bar indicates 2 µm. Discussion:

The antimicrobial polyurethanes tested showed similar RSG signal increase profiles, consistent with their shared mechanism of action. The baseline antimicrobial activity of the polyurethanes is driven by the “n” monomer containing a lysine-like functionalization, with the “m” monomers serving to enhance that activity or to increase mammalian compatibility. Polyurethane 115 B had the fastest RSG signal increase timing in the group, which may be attributable to the positively charged arginine pendant group functionalization. Increasing the overall positive charge of the polyurethane may make the polymer more efficient when targeting bacteria, but also increases the toxicity to mammalian cells compared to the other 3 polyurethanes (Mankoci et al., 2019). The trade

74 off in polyurethane efficiency for the more hemocompatible 81 C can be considered negligible and the previously reported MIC values were equivalent for both 81 C and 115

B at 16 µg/mL. RSG AST might be more useful in screening a larger library of modified polyurethanes which are geared towards producing even greater mammalian compatibility to identify which pendant groups specifically, and at what ratios, impede membrane targeting.

Cy5 labeled polyurethane 75 A provided an opportunity to investigate the antimicrobial action on the single cell level. The Cy5 75A surrounded E. coli during treatment, occurring concurrently with increased RSG signal, indicating that increased

RSG signal occurred prior to movement of 75 A into the interior of the cell. Some cells spent some time in the surrounded state, and in that time typically developed what appeared to be invaginations or inclusions. The shrinking in area of RSG signal was indicative of interior cytoplasmic volume shrinking, suggestive of a plasmolysis event.

After infiltration of Cy5 75A into the interior of the cell, Cy5 signal was seen throughout the cell without apparent regard for the invaginations of inclusions. Our data suggests that the antimicrobial mechanism of these polyurethanes may occur in part through plasmolysis events which would greatly destabilize in interior cytoplasmic membrane of the cells and lead to cell death. As the polymer surrounds the cells as part of its action, it could effectively create a local hypertonic environment which would drive plasmolysis.

75 CHAPTER VI

INVESTIGATING IMMEDIATE IMPACTS ON MITOCHONDRIAL HEALTH

FOLLOWING CUPRIZONE TREATMENT ON A SINGLE CELL LEVEL

Introduction:

Mitochondrial size and shape have a critical role in the control of respiration rate and metabolic pathway function (Campello et al., 2010). Impaired mitochondrial fusion and fission has been implicated in the general progression of neurodegenerative diseases

(Knott, 2008). In multiple sclerosis disease pathology disease progression is tied to reducing mitochondrial function (Regenold et al., 2008). Irregular mitochondrial fission and fusion occurs in cuprizone mouse models in tandem with demyelination in the corpus collosum (Kashani, 2014). Greatly enlarged megamitochondria which form in alcoholic liver disease and in cuprizone treatment animal models as a result of these irregularities have impaired respiratory capability (Wagner, 1975; Wakabayashi, 1974). The formation of megamitochondria occurs over a period of several weeks to months in animal-model oligodendrocytes and is well characterized, however there is no consensus on how cuprizone induces demyelination and mitochondrial dysfunction (Praet, 2014).

In the MO3.13 oligodendrocyte precursor cell line exposure to cuprizone in culture resulted in metabolic dysregulation which damaged NAD+ formation, disrupting energy generation (Tarabolett et al., 2017). We investigated mitochondrial swelling in living MO3.13 oligodendrocyte precursor cells treated with cuprizone over a period of 8 hours. Mitochondrial shape and size were measured using 3-dimensional imaging of live

76 cells stained with MitoTracker™ FM dye to preserve their integrity. Our goal was to identify whether the previously observed metabolic dysfunction was accompanied by irregular mitochondrial morphology. By treating mitochondria within the MO3.13 as being similar to individual “cells” we were able to directly translate the ROI style of tracking used in RSG AST. We combined the measurement of mitochondrial morphology with measurements of cellular respiration upon initial cuprizone treatment as a measure of the ability of cuprizone to directly block respiration. It has been suggested that cuprizone acting as a copper chelator can directly impair the electron transport chain through complex IV (Faizi et al., 2016). Changes in respiration rate for MO3.13 cells treated with cuprizone was measured using a fluorescence lifetime porphyrin sensor

(Dmitriev et al., 2012). Our method for respiration detection does not allow individual cell respiration to be measure due to sensitivity limitations but does allow for the per cell oxygen consumption rate to be computed.

Results:

MO3.13 cells were stained with MitoTracker™ Red FM dye and imaged to establish a baseline identification of the mitochondrial network (Fig. 6.1). Single cells were imaged across the Z-axis in 0.5 µm steps, meaning that each cell had 30 to 40 images collected from the top to bottom of the cell, allowing greater delineation of the mitochondrial network. Measuring the size of mitochondrion within the network was not straightforward as healthy cells have a relatively interconnected mitochondrial network.

The size of individually bright points within the mitochondrial network was used as an estimation of overall mitochondrial size within the network. MO3.13 cells from the same

77 culture were treated with 1 mM cuprizone for 8 hours, then were stained and imaged using MitoTracker™ Red FM dye in the same manner (Fig. 6.2).

Figure 6.1: Normal MO3.13 cell stained with MitoTracker Red FM, illustrating the threshold-based ROI analysis. Signal thresholding was used to identify discrete junctions in the mitochondrial network. The size of these junctions was computed as area values in µm2.

78 Figure 6.2: MO3.13 cells treated with 1 mM cuprizone for 8 hours, then stained with MitoTracker Red FM dye. Inset shows a zoomed in section on a single enlarged mitochondrion. Mitochondria where enlarged throughout the cell and contained visible internal space in some cases. MitoTracker™ Red FM staining is dependent upon the mitochondrial membrane potential, indicating that these mitochondria are viable and non- depolarized despite swelling. Mitochondrial swelling has been seen in long-term cuprizone treatment animal models and short-term treated oligodendrocyte cell cultures

(Blakemore et al., 1972; Cammer et al., 1999). No live cell imaging of cuprizone- enlarged mitochondria to show polarization and viability has been previously reported.

Mitochondrial cross-sectional area was compared for pre-treatment cells and the 8-hour 1 mM cuprizone treatment cells (Fig. 6.3). Cuprizone treated cross-sectional mean area was significantly (ANOVA p = 6.7 x 10-6) larger than pre-treatment, with 0.30 µm2 and 0.21

79 µm2 respectively. This represents a progression towards the megamitochondria found in longer treatment regimes. MO3.13 cells were treated with 2.5% ethanol to compare against cuprizone treated cells, as 2.5% ethanol was used in the media to keep the cuprizone dissolved. No sign of enlarged mitochondria formation was seen (Fig. 6.4).

Figure 6.3: Recorded cross-sectional area of mitochondria from cuprizone treated versus non-cuprizone treated MO3.13 cells.

80 Figure 6.4: MO3.13 cells treated with 2.5% ethanol vehicle, then stained with MitoTracker Red FM dye showing mitochondria structure. Scale bars represent 5 µm. Long-term treatment with cuprizone is known to cause electron transport chain failure, but initial cellular respiration response to cuprizone treatment has not been characterized. MO3.13 respiration rate was measured and compared for 30-minute treatments of 1 mM cuprizone, 10 µM rotenone, and ethanol vehicle alone. Cells treated with 1 mM cuprizone consumed oxygen at a rate of 118 attomoles/cell·sec compared to

91.3 attomoles/cell·sec for vehicle cells (Fig. 6.5). Cells treated with 10 µM rotenone served as a control, as rotenone rapidly and effectively blocks electron transport chain function, consistent with a lower oxygen consumption rate of 66.5 attomoles/cell·sec.

Significantly increased oxygen consumption for cuprizone compared to control indicates cellular distress occurring rapidly following cuprizone treatment (single factor ANOVA p

= .03; Table 6.1).

81 Figure 6.5: Oxygen consumption over time for control, cuprizone treated, and rotenone treated MO3.13 cells. A more negative slope indicates a greater rate of oxygen consumption.

SUMMARY Groups Count Sum Average Variance Control 3 2.67E-16 8.89E-17 6.61E-35 Cuprizone 3 3.5E-16 1.17E-16 1.65E-34

ANOVA Source of Variation SS df MS F P-value F crit Between Groups 1.17E-33 1 1.17E-33 10.10867 0.033554028 7.708647 Within Groups 4.61E-34 4 1.15E-34

Total 1.63E-33 5

Table 6.1: Single factor ANOVA analysis for control versus cuprizone treated MO3.13 cell respiration.

82 Discussion:

Mitochondrial size following cuprizone treatment was found to be increased compared to control cells as determined by cross-sectional area (0.30 µm2 versus 0.21

µm2). The percent change in mitochondrial area that we observed is consistent with the formation of megamitochondria in vivo (Das et al., 2012). Mitochondria in cuprizone treated cells also appeared noticeably more punctate in their distribution within the cell and less interconnected. While this relationship was not characterized systematically it would indicate dysfunction in mitochondrial network formation which is driven primarily by Mitofusin-2 (Filadi et al., 2018). Megamitochondria formation has not previously been observed after short exposure to cuprizone, indicating that changes in mitochondrial morphology can occur more quickly than previously understood. Formation of megamitochondria by free-radical inducing chemicals such as hydrazine occurred within

22 hours of exposure in RL-34 and IAR-20 kidney cell lines and appeared visually similar to the enlarged mitochondria we observed (Karbowski et al., 1998) (Figure 6.6).

The impact of cuprizone on mitochondrial dynamics could be mediated by free-radical production within the cell given the increased activity of superoxide dismutase and high levels of oxidative stress in the cuprizone mouse model (De et al., 1982; Praet et al.,

2014).

83 Figure 6.6: Reproduced from Karbowski et al. Left shows normal mitochondria in IAR-20 cells. Right shows enlarged megamitochondria in the IAR-20 cells following treatment with chloramphenicol. Both are stained with Mito Tracker Red CMXRos. MO3.13 cells showed increase in respiration which occurs immediately following cuprizone treatment. This response represents an initial response which contrasts with previously observed reductions in respiration after long term exposure (Wagner et al.,

1975; Wakabayashi et al., 1974). Rotenone, by comparison, caused an immediate decrease in respiration as it directly impairs electron transfer from complex I to ubiquinone. This suggests that even at higher than physiological concentrations, cuprizone does not directly block the electron transport chain as a copper chelator.

Instead, cuprizone likely disrupts enzymatic function via induced genetic dysregulation.

Reduced complex IV activity in the cuprizone mouse model has been attributed primarily to targeted copper as its primary mechanism of action (Acs et al., 2013).

However, mitochondria have reduced complex IV expression after a week in the cuprizone mouse model (Varhaug et al., 2020). Erythropoientin also shows protection against the effects of cuprizone through upregulation of complex IV and other mitochondrial respiration genes (Kashani et al., 2017). The effect of cuprizone on

84 oligodendrocyte gene regulation likely overshadows its ability to directly chelate copper in cells.

85 CHAPTER VII

CONCLUSION

Summary of Work:

This work has established a single-cell method for detecting antibiotic susceptibility which was found applicable to both Gram-positive and Gram-negative bacteria. Our method is based upon monitoring RSG signal dynamics over time to observe increases in single bacterial cells. Time required to detect antibiotic susceptibility via this method was found to be vastly improved compared to commercially available technologies. For indirectly acting β-lactams susceptibility detection was achieved within

100 minutes, while directly acting PMB susceptibility could be detected within just 10 minutes. Despite the speed of detection, there are several notable drawbacks to our AST method. RSG signal was not seen to increase in all cells treated with β-lactam antibiotics which were known to be susceptible, indicating the possibility that for any given indirectly acting antibiotic there will always be cells which do no undergo RSG signal increase despite susceptibility. Single cell RSG signal tracking is time consuming with the current system, requiring semi-manual cell identification and tracking. Our attempt at adaption of the method to an ONIX microfluidic system to improve the tracking efficiency was unsuccessful due to absorption of the dye into the PDMS-based microfluidic chamber. However, there is no indication that a glass based microfluidic system would behave differently than our current setup in terms of RSG signal dynamics.

86 The RSG AST method was found to be reproducible, and RSG signal increase timing was found to be dependent upon cell physiology for indirectly acting β-lactams.

The increase in RSG signal was seen to occur earlier in rich media compared to minimal media, and later in stationary compared to logarithmic cells. E. coli which were resistant to β-lactams did not undergo RSG signal increase when exposed to carbenicillin. We did not find any examples of false positive susceptibility detection in bacteria which should have been resistant to antibiotics that they were tested with. In the case of treatment with bacteriostatic PMBN, we found that RSG signal increase was based upon the concentration of RSG used in solution and that PI caused false positive bactericidal identification. Differentiating between bacteriostatic and bactericidal antibiotic action could be an important point of use for RSG AST in investigating novel antibiotic compounds.

Based upon the body of experiments undertaken during development and validation of the RSG AST method, the method by RSG signal increase occurs following antibiotic action can be suggested. The signal increase appears to be tied to outer membrane stability in E. coli, as increase in signal can occur with exposure to PMBN, an antibiotic which only causes outer membrane disruption and does not penetrate the inner cytoplasmic membrane. In addition, the signal increase does not appear to require electron transport chain activity since the protonophore CCCP did not abolish RSG signal increase from PMB exposure. In E. coli treated with Cy5 labeled 75 A antimicrobial polyurethane RSG signal increase was seen to occur concurrently with polyurethane buildup outside of the cell and did not require localization of 75 A into the cytoplasm.

We theorize that RSG undergoes active transport out of healthy E. coli via efflux pumps

87 in the outer membrane, and disruption of that transport during outer membrane destabilization causes an increase in RSG signal. The reducing cytoplasmic environment would be sufficient to reduce a large quantity of RSG upon contact, instantly elevating the level of signal.

The antimicrobial polyurethanes tested by RSG AST did not offer large differences in antibiotic efficacy based upon pendant group functionalization. 115 B was found to be slightly more efficient due to its arginine functionalization which is better able to target the negatively charged bacterial outer membrane. RSG AST may be better suited for use alongside a high-throughput screening for antibiotic drug molecules where there are a variety of drug mechanisms expected. RSG AST would excel at identifying high efficiency directly targeted antimicrobials which could be further refined and tested.

An initial high-throughput screen could be performed using natural product discovery methods capable of screening millions of clones per day (Scanlon et al., 2014). The mechanism of action of these candidate natural products would be unknown so it would be ideal to identify products which are high efficiency such as directly acting antibiotic molecules. Because directly targeted antimicrobials have such a rapid RSG signal increase timing, each screening candidate could be tested within ten minutes. Combined with a microfluidic approach, such a screening technique could easily screen hundreds of drug candidates per day.

A single cell imaging approach was also applied to identify morphological changes in mitochondria within MO3.13 oligodendrocyte cells treated with cuprizone.

Cuprizone treatment generated enlarged mitochondria consistent with those previously described in the mouse animal model. Cuprizone treatment was able to induce changes in

88 mitochondrial morphology within 8 hours of direct treatment, much faster than had previously been reported. Per cell respiration was also measured using a platinum- porphyrin based oxygen consumption measurement. Within the first hour of cuprizone treatment, oxygen consumption of treated cells rose compared to vehicle and rotenone treated cells. The increase in oxygen consumption seen in these cells may indicate an immediate cellular stress brought on by cuprizone exposure and indicates that cuprizone copper chelation does not reduce respiration rate, even at higher than physiological concentrations of cuprizone.

Future Work and Applications:

The RSG AST method described herein provides improvement over existing methods for the detection of antibiotic resistance but does not yet establish a widely distributable methodology and protocol appropriate for clinical use. Despite the failure of the method to adapt to the ONIX microfluidic system that we tested; we believe that the transition into an appropriate microfluidic system will make this method clinically viable.

Our method parallels other microscopy AST methods in development but is inherently more universally applicable than growth-based or morphology-based methods such as fASTest and SCMA. The antibiotic treatment induced RSG signal phenomenon appeared in all tested Gram-positive and Gram-negative organisms and is fundamentally uncomplicated to analyze when analysis can be fully automated. Our primary goal of future work in this area would be to develop a PDMS-free microfluidic chip system which would provide straightforward loading and treatment of bacteria using RSG AST.

The most straightforward version of this system would provide for loading via syringe for clinical samples whereby bacteria would be automatically concentrated in a cell trap

89 within the chip. Within the chip eukaryotic cells would be excluded by size using narrow capture channels to trap and concentrate bacteria within an imaging area. Following syringe loading, a wash solution would be run through via fresh syringe followed by RSG staining and antibiotic challenge combined with fluorescence imaging. By concentrating the bacteria into a size-constricted space bacterial movement would be greatly reduced and Gram-positive bacteria could be reduced to a monolayer for simplified automated analysis. The most challenging aspect of this future work would be fabrication and design of a microfluidic system which does not absorb RSG while maintaining a manufacturing cost which is attractive.

Our methodology has possible future applications in antibiotic discovery and synthetic antimicrobial design. MIC remains the gold standard by which antibiotic efficacy is measured despite the lack of time-to-kill information given by that measure.

By including RSG AST into antimicrobial design as a measure of contact time required to generate antimicrobial action, antimicrobials can be designed to not simply kill at lower concentrations but to kill more quickly. This can be extended to not just more efficient killing of planktonic bacteria but more efficient elimination of biofilms as well due to the demonstrated ability of RSG to penetrate a bacterial biofilm. A goal for RSG AST for antimicrobial design would be to test antimicrobial polymer variants based upon rate of

RSG signal increase within biofilm populations. The development of new antimicrobials which can quickly infiltrate biofilms represents a major challenge of current antimicrobial design which RSG could contribute to. Antibiotic drug discovery can benefit from RSG AST in much the same ways that antimicrobial design can. The major hurdle with respect to drug discovery that would need to be addressed is throughput. Our

90 primary goal for drug discovery applications would be establishing a high throughput methodology using RSG to test antimicrobial containing natural isolates. A multi-well imaging plate would be well suited for initial screens. Multiple wells with adhered bacteria would be monitored simultaneously at low magnification with separate antimicrobial isolates to identify the presence of RSG signal increase. The strength of the

RSG signal response would allow for identification of RSG increase occurring at low magnifications as point signals without being able to identify single cell dynamics.

Candidate isolates which induced RSG signal increases at low magnification would then be imaged further at higher magnification to establish the RSG signal dynamics and compare relative antimicrobial strength. RSG AST represents a straightforward fluorescence technique with bountiful future applications.

The cuprizone model of oligodendrocyte demyelination still has much to be understood with respect to mitochondrial physiology. Cuprizone induces mitochondrial dysfunction rapidly in MO3.13 oligodenrocytes, with enlarged mitochondria forming within 8 hours of treatment. Changes in mitochondrial size of this nature had not been previously identified under such short timespans of exposure. Ongoing work includes tying the mitochondrial morphology changes that we have observed to transcriptional changes within the cell. Examining transcriptional dysregulation for superoxide dismustase, mitofusin and other mitochondria specific genetic markers will provide additional context to our work. Respiration rate should be characterized for a greater range of time points in order to understand the progression from initially increased levels of respiration to the point of mitochondrial dysfunction ultimately impairing that

91 respiration. Our current respiration data only rules out the ability of cuprizone to immediately block electron transport chain function via rapid chopper chelation activity.

92 REFERENCES

1. Aslam, B. et al. Antibiotic resistance: a rundown of a global crisis. Infection and Drug Resistance vol. 11 1645–1658 (2018).

2. Balaban, N. Q., Merrin, J., Chait, R., Kowalik, L. & Leibler, S. Bacterial persistence as a phenotypic switch. Science (80-. ). 305, 1622–1625 (2004).

3. Baltekin, Ö., Boucharin, A., Tano, E., Andersson, D. I. & Elf, J. Antibiotic susceptibility testing in less than 30 min using direct single-cell imaging. Proc. Natl. Acad. Sci. U. S. A. 114, 9170–9175 (2017).

4. Barcelos, I. P. de, Troxell, R. M. & Graves, J. S. Mitochondrial Dysfunction and Multiple Sclerosis. Biology (Basel). 8, 37 (2019).

5. Blakemore, W. F. Observations on oligodendrocyte degeneration, the resolution of status spongiosus and remyelination in cuprizone intoxication in mice. J. Neurocytol. 1, 413–426 (1972).

6. Boardman, A. K., Campbell, J., Wirz, H., Sharon, A. & Sauer-Budge, A. F. Rapid microbial sample preparation from blood using a novel concentration device. PLoS One 10, (2015).

7. Brauner, A., Shoresh, N., Fridman, O. & Balaban, N. Q. An Experimental Framework for Quantifying Bacterial Tolerance. Biophys. J. 112, 2664–2671 (2017).

8. Buchanan, C. M. et al. Rapid separation of very low concentrations of bacteria from blood. J. Microbiol. Methods 139, 48–53 (2017).

9. Cammer, W. The neurotoxicant, cuprizone, retards the differentiation of oligodendrocytes in vitro. J. Neurol. Sci. 168, 116–120 (1999).

10. Campello, S. & Scorrano, L. Mitochondrial shape changes: Orchestrating cell pathophysiology. EMBO Reports vol. 11 678–684 (2010).

11. Cardoso, M. H. et al. Computer-Aided Design of Antimicrobial Peptides: Are We Generating Effective Drug Candidates? Frontiers in Microbiology vol. 10 (2020).

12. Chalfie, M., Tu, Y., Euskirchen, G., Ward, W. W. & Prasher, D. C. Green fluorescent protein as a marker gene expression. Science vol. 263 (1994).

93 13. Chamsaz, E. A., Mankoci, S., Barton, H. A. & Joy, A. Nontoxic Cationic Coumarin Polyester Coatings Prevent Pseudomonas aeruginosa Biofilm Formation. ACS Appl. Mater. Interfaces 9, 6704–6711 (2017).

14. Choi, J. et al. A rapid antimicrobial susceptibility test based on single-cell morphological analysis. Science Translational Medicine vol. 6 www.ScienceTranslationalMedicine.org (2014).

15. Das, S. et al. Mitochondrial morphology and dynamics in hepatocytes from normal and ethanol-fed rats. Pflugers Arch. Eur. J. Physiol. 464, 101–109 (2012).

16. De Vos, W. M. Microbial biofilms and the human intestinal microbiome. npj Biofilms and Microbiomes vol. 1 (2015).

17. De, A. K. & Subramanian, M. Effect of cuprizone feeding on hepatic superoxide dismutase and cytochrome oxidase activities in mice. Experientia 38, 784–785 (1982).

18. Deere, D., Porter, J., Edwards, C. & Pickup, R. Evaluation of the suitability of bis - (1,3-dibutylbarbituric acid) trimethine oxonol, (diBA-C 4 (3)-), for the flow cytometric assessment of bacterial viability. FEMS Microbiol. Lett. 130, 165–169 (1995).

19. Dmitriev, R. I. & Papkovsky, D. B. Optical probes and techniques for O 2 measurement in live cells and tissue. Cellular and Molecular Life Sciences vol. 69 2025–2039 (2012).

20. Donlan, R. M. Biofilms: Microbial Life on Surfaces. Emerging Infectious Diseases • vol. 8 http://www.microbelibrary.org/ (2002).

21. Duwe, A. K., Rupar, C. A., Horsman, G. B. & Vas, S. I. In Vitro Cytotoxicity and Antibiotic Activity of Polymyxin B Nonapeptide. ANTIMICROBIAL AGENTS AND CHEMOTHERAPY (1986).

22. Epand, R. M., Walker, C., Epand, R. F. & Magarvey, N. A. Molecular mechanisms of membrane targeting antibiotics. Biochim. Biophys. Acta - Biomembr. 1858, 980–987 (2016).

23. Filadi, R., Pendin, Di. & Pizzo, P. Mitofusin 2: From functions to disease. Cell Death and Disease vol. 9 (2018).

24. Fleischmann, C. et al. Assessment of global incidence and mortality of hospital- treated sepsis current estimates and limitations. Am. J. Respir. Crit. Care Med. 193, 259–272 (2016).

94 25. Gilbert, P., Collier, P. J. & Brown, M. R. W. Influence of Growth Rate on Susceptibility to Antimicrobial Agents: Biofilms, Cell Cycle, Dormancy, and Stringent Response. Antimicrob. Agents Chemother. 34, 1865–1868 (1990).

26. Graham, J. P., Evans, S. L., Price, L. B. & Silbergeld, E. K. Fate of antimicrobial- resistant enterococci and staphylococci and resistance determinants in stored poultry litter. Environ. Res. 109, 682–689 (2009).

27. Grassi, L. et al. Generation of persister cells of Pseudomonas aeruginosa and Staphylococcus aureus by chemical treatment and evaluation of their susceptibility to membrane-targeting agents. Front. Microbiol. 8, (2017).

28. Guillon, A. et al. Treatment of Pseudomonas aeruginosa Biofilm Present in Endotracheal Tubes by Poly-L-Lysine. Antimicrob. Agents Chemother. 62, (2018).

29. Hyldgaard, M. et al. The antimicrobial mechanism of action of epsilon-poly-L- lysine. Appl. Environ. Microbiol. 80, 7758–7770 (2014).

30. Ivanenkov, Y. A. et al. 2-Pyrazol-1-yl-thiazole derivatives as novel highly potent antibacterials. J. Antibiot. (Tokyo). 72, 827–833 (2019).

31. Joo, H. S., Fu, C. I. & Otto, M. Bacterial strategies of resistance to antimicrobial peptides. Philos. Trans. R. Soc. B Biol. Sci. 371, (2016).

32. Kashani, I. R. et al. Protective effects of melatonin against mitochondrial injury in a mouse model of multiple sclerosis. Exp. Brain Res. 232, 2835–2846 (2014).

33. Kirn, T. J. & Weinstein, M. P. Update on blood cultures: How to obtain, process, report, and interpret. Clinical Microbiology and Infection vol. 19 513–520 (2013).

34. Knott, A. B. & Bossy-Wetzel, E. Impairing the mitochondrial fission and fusion balance: A new mechanism of neurodegeneration. Ann. N. Y. Acad. Sci. 1147, 283–292 (2008).

35. Kostić, T. et al. Thirty-minute screening of antibiotic resistance genes in bacterial isolates with minimal sample preparation in static self-dispensing 64 and 384 assay cards. Appl. Microbiol. Biotechnol. 99, 7711–7722 (2015).

36. Lehrer, R. I., Barton, A. & Ganz, T. Concurrent assessment of inner and outer membrane permeabilization and bacteriolysis in E. coli by multiple-wavelength spectrophotometry. J. Immunol. Methods 108, 153–158 (1988).

37. Levison, M. E. & Levison, J. H. Pharmacokinetics and Pharmacodynamics of Antibacterial Agents. Infect. Dis. Clin. North Am. 23, 791–815 (2009).

95 38. Lewis, K. Persister cells and the riddle of biofilm survival. Biokhimiya 70, 327– 336 (2005).

39. Li, J. et al. Membrane active antimicrobial peptides: Translating mechanistic insights to design. Frontiers in Neuroscience vol. 11 (2017).

40. Lima, M. R. et al. Evaluation of the interaction between polymyxin B and Pseudomonas aeruginosa biofilm and planktonic cells: Reactive oxygen species induction and zeta potential. BMC Microbiol. 19, (2019).

41. Lobritz, M. A. et al. Antibiotic efficacy is linked to bacterial cellular respiration. Proc. Natl. Acad. Sci. U. S. A. 112, 8173–8180 (2015).

42. Mahlapuu, M., Håkansson, J., Ringstad, L. & Björn, C. Antimicrobial peptides: An emerging category of therapeutic agents. Frontiers in Cellular and Infection Microbiology vol. 6 (2016).

43. Mankoci, S. et al. Bacterial Membrane Selective Antimicrobial Peptide-Mimetic Polyurethanes: Structure-Property Correlations and Mechanisms of Action. Biomacromolecules 20, 4096–4106 (2019).

44. Mankoci, S., Kaiser, R. L., Sahai, N., Barton, H. A. & Joy, A. Bactericidal Peptidomimetic Polyurethanes with Remarkable Selectivity against Escherichia coli. ACS Biomater. Sci. Eng. 3, 2588–2597 (2017).

45. Melnyk, A. H., Wong, A. & Kassen, R. The fitness costs of antibiotic resistance mutations. Evol. Appl. 8, 273–283 (2015).

46. Mitra, K. & Lippincott-Schwartz, J. Analysis of mitochondrial dynamics and functions using imaging approaches. Curr. Protoc. Cell Biol. 4251–42521 (2010) doi:10.1002/0471143030.cb0425s46.

47. Nagarajan, D. et al. Computational antimicrobial peptide design and evaluation against multidrug-resistant clinical isolates of bacteria. J. Biol. Chem. 293, 3492– 3509 (2018).

48. Nguyen, D. et al. Active starvation responses mediate antibiotic tolerance in biofilms and nutrient-limited bacteria. Science (80-. ). 334, 982–986 (2011).

49. Nicoloff, H., Hjort, K., Levin, B. R. & Andersson, D. I. The high prevalence of antibiotic heteroresistance in pathogenic bacteria is mainly caused by gene amplification. Nat. Microbiol. 4, 504–514 (2019).

50. Ojha, A., Banik, S., Melanthota, S. K. & Mazumder, N. Light emitting diode (LED) based fluorescence microscopy for tuberculosis detection: a review. Lasers in Medical Science (2020) doi:10.1007/s10103-019-02947-6.

96 51. Post, V., Wahl, P., Richards, R. G. & Moriarty, T. F. Vancomycin displays time- dependent eradication of mature Staphylococcus aureus biofilms. J. Orthop. Res. 35, 381–388 (2017).

52. Praet, J., Guglielmetti, C., Berneman, Z., Van der Linden, A. & Ponsaerts, P. Cellular and molecular neuropathology of the cuprizone mouse model: Clinical relevance for multiple sclerosis. Neurosci. Biobehav. Rev. 47, 485–505 (2014).

53. Regenold, W. T., Phatak, P., Makley, M. J., Stone, R. D. & Kling, M. A. Cerebrospinal fluid evidence of increased extra-mitochondrial glucose metabolism implicates mitochondrial dysfunction in multiple sclerosis disease progression. J. Neurol. Sci. 275, 106–112 (2008).

54. Reimer, L. G., Wilson, M. L. & Weinstein, M. P. Update on Detection of Bacteremia and Fungemia. vol. 10 (1997).

55. Rossolini, G. M., Arena, F., Pecile, P. & Pollini, S. Update on the antibiotic resistance crisis. Curr. Opin. Pharmacol. 18, 56–60 (2014).

56. Rumbaugh, K. P. & Sauer, K. Biofilm dispersion. Nat. Rev. Microbiol. (2020) doi:10.1038/s41579-020-0385-0.

57. Scanlon, T. C., Dostal, S. M. & Griswold, K. E. A high-throughput screen for antibiotic drug discovery. Biotechnol. Bioeng. 111, 232–243 (2014).

58. Shi, L. et al. Limits of propidium iodide as a cell viability indicator for environmental bacteria. Cytom. Part A 71, 592–598 (2007).

59. Silhavy, T. J., Kahne, D. & Walker, S. The bacterial cell envelope. Cold Spring Harbor perspectives in biology vol. 2 (2010).

60. Smith, K. & Hunter, I. S. Efficacy of common hospital with biofilms of multi-drug resistant clinical isolates. J. Med. Microbiol. 57, 966–973 (2008).

61. Spohn, R. et al. Integrated evolutionary analysis reveals antimicrobial peptides with limited resistance. Nat. Commun. 10, (2019).

62. Stiefel, P., Schmidt-Emrich, S., Maniura-Weber, K. & Ren, Q. Critical aspects of using bacterial cell viability assays with the fluorophores SYTO9 and propidium iodide. BMC Microbiol. 15, (2015).

63. Sun, L. et al. Antibiotic-induced disruption of gut microbiota alters local metabolomes and immune responses. Front. Cell. Infect. Microbiol. 9, (2019).

97 64. Taraboletti, A. et al. Cuprizone Intoxication Induces Cell Intrinsic Alterations in Oligodendrocyte Metabolism Independent of Copper Chelation. Biochemistry 56, 1518–1528 (2017).

65. te Winkel, J. D., Gray, D. A., Seistrup, K. H., Hamoen, L. W. & Strahl, H. Analysis of antimicrobial-triggered membrane depolarization using voltage sensitive dyes. Front. Cell Dev. Biol. 4, (2016).

66. Tsubery, H., Ofek, I., Cohen, S. & Fridkin, M. Structure - Function studies of Polymyxin B nonapeptide: Implications to sensitization of Gram-negative bacteria. J. Med. Chem. 43, 3085–3092 (2000).

67. Tsubery, H., Ofek, I., Cohen, S. & Fridkin, M. N-terminal modifications of Polymyxin B nonapeptide and their effect on antibacterial activity. Peptides 22, 1675–1681 (2001).

68. Tsujimoto, H., Gotoh, N. & Nishino, T. Diffusion of macrolide antibiotics through the outer membrane of Moraxella catarrhalis. J. Infect. Chemother. 5, 196–200 (1999).

69. Tuomanen, E., Cozens, R., Tosch, W., Zak, O. & Tomasz, A. The rate of killing of Escherichia coli by β-lactam antibiotics is strictly proportional to the rate of bacterial growth. J. Gen. Microbiol. 132, 1297–1304 (1986).

70. Tuomanen, E., Durack, D. T. & Tomasz1, A. Antibiotic Tolerance among Clinical Isolates of Bacteria. ANTIMICROBIAL AGENTS AND CHEMOTHERAPY vol. 30 (1986).

71. Van Boeckel, T. P. et al. Global trends in antimicrobial use in food animals. Proc. Natl. Acad. Sci. U. S. A. 112, 5649–5654 (2015).

72. Velkov, T., Thompson, P. E., Nation, R. L. & Li, J. Structure-activity relationships of polymyxin antibiotics. Journal of Medicinal Chemistry vol. 53 1898–1916 (2010).

73. Ventola, C. L. Antibiotic Resistance Crisis Part 1: Causes and Threats. Pharmacy and Therapeutics vol. 40 (2015).

74. Vickery, K., Hu, H., Jacombs, A. S., Bradshaw, D. A. & Deva, A. K. A review of bacterial biofilms and their role in device-associated infection. Healthcare Infection vol. 18 61–66 (2013).

75. Wagner, T. & Rafael, J. ATPase complex and oxidative phosphorylation in chloramphenicol-induced megamitochondria from mouse liver. BBA - Bioenerg. 408, 284–296 (1975).

98 76. Wakabayashi, T. & Green, D. E. On the mechanism of cuprizone-induced formation of megamitochondria in mouse liver. Journal of Bioenergetics vol. 6 (1974).

77. Wang, P. et al. Robust growth of escherichia coli. Curr. Biol. 20, 1099–1103 (2010).

78. Yao, Z., Kahne, D. & Kishony, R. Distinct Single-Cell Morphological Dynamics under Beta-Lactam Antibiotics. Mol. Cell 48, 705–712 (2012).

79. Yasir, M., Willcox, M. D. P. & Dutta, D. Action of antimicrobial peptides against bacterial biofilms. Materials vol. 11 (2018).

80. Yoon, T., Moon, H. S., Song, J. W., Hyun, K. A. & Jung, H. Il. Automatically Controlled Microfluidic System for Continuous Separation of Rare Bacteria from Blood. Cytom. Part A 95, 1135–1144 (2019).

81. Yu, Z., Qin, W., Lin, J., Fang, S. & Qiu, J. Antibacterial mechanisms of polymyxin and bacterial resistance. BioMed Research International (2015) doi:10.1155/2015/679109.

82. Zasloff, M. Antimicrobial peptides of multicellular organisms. Nature vol. 415 www.nature.com (2002).

99