Re-evaluation of FDA-approved antibiotics with increased diagnostic accuracy for assessment of antimicrobial resistance

Summary Accurate assessment of antibiotic susceptibility is critical for treatment of antimicrobial resistant (AMR) infections. Here, we examine whether antimicrobial susceptibility testing in media more physiologically representative of in vivo conditions improves prediction of clinical outcome relative to standard bacteriologic medium. This analysis reveals that ∼15% of minimum inhibitory concentration (MIC) values obtained in physiologic media predicted a change in susceptibility that crossed a clinical breakpoint used to categorize patient isolates as susceptible or resistant. The activities of antibiotics having discrepant results in different media were evaluated in murine sepsis models. Testing in cell culture medium improves the accuracy by which MIC assays predict in vivo efficacy. This analysis identifies several antibiotics for treatment of AMR infections that standard testing failed to identify and those that are ineffective despite indicated use by standard testing. Methods with increased diagnostic accuracy mitigate the AMR crisis via utilizing existing agents and optimizing drug discovery.

In brief Heithoff et al. observe that antibiotic testing in cell culture medium improves the accuracy by which laboratory testing predicts clinical outcomes in mice. Test methods with increased diagnostic accuracy will address the antimicrobial resistance crisis by improving the way antibiotics are developed, tested, and prescribed.

INTRODUCTION
The World Health Organization (WHO) identified antimicrobial resistance as a major threat to global health, food security, and economic stability. 1,2 Despite the scale and urgency, few promising drug candidates are currently in the clinical pipeline due to the high costs of drug development and risk that a newly approved antibiotic becomes ineffective due to bacterial resistance or is earmarked for use as a drug of last resort. [3][4][5] Additional factors include reduced incentives for pharmaceutical research and development for diseases that require relatively short courses of treatment (infectious diseases) relative to blockbuster drugs for pervasive diseases (cancer, cardiovascular diseases, hyperlipidemia, and immune disorders). 6,7 The healthcare industry paradigm for the evaluation of antibiotic efficacy is based on in vitro assays that do not consider host-pathogen interactions that can have a marked impact on drug potency. 8 The principal in vitro assay for antibiotic assessment, developed in the 1940s, uses a nutrient-rich bacteriologic medium, Mueller-Hinton broth (MHB). 9 This assay has been used globally for antimicrobial susceptibility testing (AST) to determine the minimum inhibitory concentration (MIC), the standard measurement of antibiotic activity. MICs determine the clinical breakpoint, the concentration of antibiotic used to indicate whether an infection with a given clinical isolate is likely to be treatable in a patient. [10][11][12] Clinical breakpoints are used by clinical microbiological laboratories to define patient isolates as susceptible (S) or resistant (R) to a panel of antibiotics. Thus, the in vitro MHB bioassay has been the criterion standard for guiding physician treatment practices, tracking outbreaks and epidemics, and assessing chemical structures in the development of novel therapeutics for more than half a century.
Despite these successes, in vitro bioassays are fundamentally flawed because antibiotic potency is highly context dependent, influenced by media composition (pH, buffers, osmolarity, nutrients); pathogen factors (load, virulence, resistance genes); host factors that can act synergistically with antimicrobials (antimicrobial peptides, complement, neutrophils); and the generation of reactive metabolic byproducts after antibiotic exposure. 13,14 Thus, AMR therapy is often reliant on clinical reasoning by physicians on a case-by-case basis with support from agencies that provide up-to-date guidance on clinical management. 15 Significant advances have been made to increase the accuracy by which in vitro assays predict clinical outcome. This is evidenced by (1) evaluation of antibacterial activity using patient serum, 16 (2) utilization of host-mimicking media to increase predictive accuracy, [17][18][19][20][21] and (3) antibiotic synergy with cationic antimicrobial peptides [22][23][24] and reactive metabolic byproducts, 25 with resultant translation to front-line therapies. 14,22,[26][27][28][29][30] However, significant hurdles remain, as many of these approaches require either patient specimens, simulation of host compartments, addition of purified biologicals, or exploitation of bacterial metabolic networks.
Here, we report the development of an alternative AST protocol for widespread clinical utility based on media that are more physiologically representative of in vivo infection conditions (mammalian cell culture medium, pooled human donor serum, or urine) vs. standard bacteriologic MHB medium. MHB supports the growth of bacteria and is not intended to mimic any aspect of the host environment. In contrast, cell culture medium supports the growth of mammalian cells, reflecting physiological conditions more consistent with in vivo sites of microbial infection, and human sera or urine are often the site/route of bacterial dissemination. MICs of clinically relevant antibiotics were evaluated against ESKAPE (Enterococcus faecium, S. aureus, K. pneumoniae, A. baumannii, P. aeruginosa, and Enterobacter spp.) pathogens (that escape the biocidal action of antibiotics) 31 in physiologic media, and diagnostic accuracy was assessed in murine models of sepsis. Using FDA-approved antibiotics, we find that AST in mammalian cell culture medium increased diagnostic accuracy, thereby providing justification for clinical utilization of existing antibiotics for the potential treatment of AMR infections.
Phase 2 of the study evaluated whether bacterial testing in physiologic media improved the MIC predictive accuracy of clinical outcome vs. that seen in standard MHB medium. This in vivo analysis examined a total of 26 antibiotic/pathogen combinations assayed individually in murine sepsis models (11 antibiotics; 11 clinical isolates; 7 bacterial spp.). A comparative statistical analysis of the predicted number of survivors (phase 1) vs. the actual number of survivors (phase 2) was performed to determine the accuracy by which MIC testing predicted in vivo efficacy.
Development of a standardized AST protocol for testing in human serum and urine Environmental sensitization to physiologic conditions during bacterial culture and drug testing can have up to a 1,000-fold effect on antibiotic susceptibility. 20 Thus, consideration of physiologic conditions should be implemented in a standardized AST protocol for widespread clinical utility. However, this presents a formidable challenge for test media consisting of human sera or urine that can be inhibitory to the bacterial culture of some pathogens. Although most bacterial pathogens tested exhibited robust growth after overnight culture in serum or urine pooled from human donors, several pathogens formed aggregates that impaired enumeration and subculture and/or did not support growth to adequate bacterial cell densities in microtiter plates required for reliable MIC determination. 36 We thus established a media supplementation/aggregate disruption protocol to enable comparative AST analyses in human serum or urine ( Figure S1; see STAR Methods). Briefly, bacterial isolates were sensitized to 100% pooled human donor serum or urine by overnight culture, agitated to separate bacterial cell aggregates, diluted into human fluids supplemented with 30% (v/v) Luria-Bertani broth (LB) to supply limiting nutrients, 37 and subjected to MIC testing performed in the supplemented human fluids using microtiter plates. This procedure allowed the sensitization of bacteria in human fluids and adequate bacterial cell densities for reliable MIC determination for all pathogens tested.
Comparative AST analysis in physiologic media vs. bacteriologic medium In head-to-head comparative analyses, antibiotics were evaluated for antibacterial activity against clinical isolates assayed in standard bacteriologic MHB medium, in mammalian cell culture medium (DMEM), and in pooled human donor sera or urine. Testing in physiologic media revealed that 14.7% (74/504) of the MIC values that were obtained predicted a change in susceptibility designation that crossed a clinical breakpoint (S to R; R to S) (Tables S1 and S2; Figures 1 and 2). Such altered susceptibility designations could potentially change physician decision-making provided they were supported by favorable clinical outcomes. Notably, susceptibility designations for several antibiotic/pathogen combinations derived from testing in DMEM frequently differed from those derived in MHB and in pooled human donor serum or urine. This is evidenced by the DMEMpredicted susceptibility (R to S) of (1) MRSA (USA300, MT3302) and E. cloacae to ceftriaxone, (2) MRSA (USA300, MT3302) to piperacillin/tazobactam, and (3) S. Typhimurium to streptomycin (vs. all other media predicting resistance) (Table 1), as well as DMEM-predicted resistance (S to R) of A. baumannii, K. pneumoniae and P. aeruginosa to colistin (vs. all other media predicting susceptibility).

Assessment of MIC predictive accuracy of clinical outcome in murine models of Gram-positive and Gramnegative sepsis
The activities of antibiotics that had discrepant results in physiologic media were evaluated for MIC predictive accuracy of clinical outcome vs. that seen in standard MHB medium. This in vivo analysis constituted the individual assay of 26 antibiotic/pathogen combinations in murine models of sepsis (11 antibiotics; 11 clinical isolates; 7 bacterial spp.) ( Figure 3; Table 1). The dose/route of infection and sepsis disease progression was based on established animal models of sepsis (see STAR Methods). 38,39 All mice in the mock-treated groups died, providing an indication of expected mortality in the absence of effective treatment. Briefly, pairwise comparisons of test accuracy were performed between the media across all pathogen and antimicrobial combinations and between MHB and DMEM for each antibiotic and pathogen (see STAR Methods). Diag-nostic accuracy was calculated as the number of animals that were predicted to survive and did survive combined with the number of animals that were predicted to succumb and did succumb divided by the total numbers of animals. Statistical analyses returning a p value of <0.05 were considered significant. Diagnostic accuracy of discordant MICs that crossed a clinical breakpoint increased from 54% in MHB to 77% in DMEM (p = 0.014), but accuracy decreased to 34% in pooled human donor sera or urine (p = 0.006). Increased diagnostic accuracy in cell culture medium was a reflection of improved prediction of antibiotic treatment success from 61% in MHB to 87.7% in DMEM (p = 0.026) and a trend for improved prediction of treatment failure from 37% in MHB to 50.7% in DMEM (p = 0.37).
Increased diagnostic accuracy in DMEM was demonstrated in several animal models of infection vs. that seen in MHB and in pooled human donor serum or urine. This is evidenced by the DMEM-predicted treatment success (R to S) of (1) MRSA (USA300, MT3302) and E. cloacae with ceftriaxone ( Figures 3A,  3B, and 3G), (2) MRSA (USA300, MT3302) with piperacillin/tazobactam ( Figures 3A and 3B), and (3) S. Typhimurium with streptomycin ( Figure 3K) (vs. all other media predicting resistance), as well as DMEM-predicted treatment failures (S to R) of A. baumannii, K. pneumoniae (13883), and P. aeruginosa with colistin ( Figures 3F, 3H, and 3J) (vs. all other media predicting susceptibility). An exception is the treatment success (but DMEM-predicted failure) of K. pneumoniae (13883) with tetracycline ( Figure 3H).  Figures 3G, 3H, and 3K), as well as treatment failure (R and R) of K. pneumoniae (MT3325) with azithromycin ( Figure 3I). Taken together, these findings suggest that testing in DMEM cell culture medium, either alone or in combination with standard bacteriologic MHB medium, provides an approach to identify presently available antibiotics for the potential treatment of AMR infections.

DISCUSSION
Our results indicate that re-evaluation of existing FDA-approved antibiotics may be an important augmentation to the development of new drugs to combat antimicrobial resistance. In headto-head comparisons of physiologic media (DMEM, sera, urine) vs. standard bacteriologic MHB medium, 15% of the MIC values obtained in physiologic media predicted a change in susceptibility that crossed a clinical breakpoint, the concentration of antibiotic used to define whether an infection with a given clinical isolate is likely to be treatable in a patient. Diagnostic accuracy of these discordant MICs increased when the testing was carried out in DMEM cell culture medium and assayed in murine models of Gram-positive and Gram-negative sepsis. The test advancement was a reflection of both improved prediction of antibiotic treatment success and a trend for improved prediction of antibiotic treatment failure. This led to the identification of potentially effective FDA-approved antibiotics for the treatment of AMR infections that standard testing failed to identify and also excluded those that were ineffective despite indicated use by standard testing. Additionally, diagnostic accuracy of bacteriologic medium increased when test results were in agreement with cell culture medium results but dropped when in disagreement. Thus, test agreement between the two culture test conditions may increase confidence for clinical decision-making, while contrary predictions may favor an adjunctive therapy to primary treatment. Clinical implementation of test methods with improved diagnostic accuracy provides a platform to expand the therapeutic armamentarium, improve clinical management and antibiotic stewardship, and facilitate the discovery of novel compounds with improved pharmacological properties.
This study provides a potential solution for addressing discrepant results between antibiotics indicated by standard AST and actual clinical outcomes. Indeed, the limitations of standard AST methods for predicting clinical efficacy are being increasingly recognized, as certain antibiotics dismissed by standard testing are effective at treating AMR infections. This is evidenced by b-lactams as adjunctive therapy for refractory bacteremia caused by MRSA and vancomycin-resistant Enterococcus, 22,26 azithromycin monotherapy for multidrug-resistant P. aeruginosa, 28 and azithromycin/piperacillin-tazobactam combination therapy for CRE Achromobacter xylosoxidans. 40 Further, despite limited azithromycin breakpoint designations for Enterobacterales (S. Typhi and Shigella spp.), 41,42 azithromycin has been used clinically for diarrheagenic E. coli, Shigella spp., Salmonella spp., and Campylobacter spp. 43,44 This has led to proposed azithromycin breakpoints for diarrheagenic E. coli, 45 and azithromycin is being considered as a standard therapy for specific Enterobacterales infections. [45][46][47] The mechanism by which DMEM cell culture medium improved the accuracy by which MIC testing predicted in vivo efficacy appears to rely on the presence of physiologic levels of sodium bicarbonate, an anionic buffer that plays a role in the maintenance of blood and tissue pH. 48 However, the role of bicarbonate in antibiotic susceptibility was not simply a reflection of pH stabilization 49 because removal of NaHCO 3 from exogenously buffered DMEM cell culture medium resulted in MICs similar to MHB in many species, 19 and, reciprocally, the addition of NaHCO 3 to exogenously buffered MHB resulted in MICs similar to DMEM cell culture medium. Rather, bicarbonate is a pleiotropic ionic factor that (1) stimulates global changes in bacterial gene expression with resultant changes in bacterial membrane permeability that impact susceptibility to cationic peptides, 49 (2) affects bacterial virulence gene expression and resultant susceptibility to b-lactam antibiotics, 50 and (3) contributes to the dissipation of the bacterial proton motive force (PMF) required for activity or import/export of various classes of antibiotics and several immune components (defensins, cathelicidins, bile salts). 51 Additionally, such changes in antibiotic susceptibility might also have an indirect effect on bactericidal action via stimulating host cytokine responses important for bacterial clearance. For example, bicarbonate-mediated changes in bacterial membrane permeability can increase b-lactam levels. b-Lactams can increase the expression of a-toxin in S. aureus, 52 which in turn can prompt an immunostimulatory interleukin-1-b (IL-1-b) response 53 with resultant enhanced host recognition/bacterial clearance for the successful treatment of S. aureus bacteremia. 54,55 Testing in each of the three physiologic media examined (DMEM, human sera, or urine) resulted in a similar fraction of MICs (15%) that predicted a change in clinical breakpoint Colored regions depict the fraction of pathogen-antibiotic combinations tested that exhibited a change in MIC (increased susceptibility or resistance) when derived in either mammalian cell culture medium (DMEM), pooled human donor sera, or urine relative to standard bacteriologic MHB medium; %2-fold (green), 4-fold (yellow), R8-fold (red). Percentages of pathogen-antibiotic combinations (test/standard condition) resulting in MICs that resulted in altered susceptibility designations are depicted. S, susceptible; I, intermediate; R, resistant. MICs were determined by broth microdilution. [10][11][12]  MICs and susceptibility designations were determined by broth microdilution [10][11][12] as detailed in Tables S1 and S2. MIC values were derived from the consensus of R6 independent determinations. MHB and DMEM assays: unless otherwise specified, bacterial culture and testing in MHB or DMEM were performed in unsupplemented medium. Sera and urine assays: bacteria were cultured overnight in 100% pooled human donor sera or urine, agitated to separate bacterial cell aggregates, diluted into human fluids supplemented with 30% LB, and subjected to MIC testing performed in supplemented human fluids in microtiter plates (see STAR Methods) (n R 6). Virulence assays: discordant MICs derived from antibiotic susceptibility testing in MHB, DMEM, human sera, and urine were examined for diagnostic accuracy following individual assay in murine sepsis models (n = 10) (see Figure 3; STAR Methods Report ll OPEN ACCESS classification. The increased diagnostic accuracy of DMEM relative to the other physiologic media tested is not driven solely by the presence of bicarbonate, as it is also present in human sera and urine (DMEM, 44 mM; sera, 25 mM; urine, 2.5 mM). 33,48,56,57 It may, however, reflect that DMEM supports the growth of mammalian cells, emulating physiological conditions more consistent with in vivo sites of microbial infection.
An alternate possibility is that results concerning the predictive power of human serum or urine (vs. DMEM or MHB) do not apply to mice but may apply to the human condition because of inherent milieu differences between the two species that impact drug potency. Notably, results of animal models of systemic infection may not be readily translated to other modes of infection, including respiratory, skin, urinary tract infections (UTIs), or even some cases of bacteremia (given the route of infection of the models used), and thus individual physiologic media might be more predictive for their corresponding site of infection. Therefore, conclusions concerning the predictive power of serum or urine (vs. DMEM or MHB) require further investigation using additional models of infection (e.g., respiratory, skin, UTIs).
AST in physiologic media may impact the means by which antibiotics are tested, developed, and prescribed and offers a number of advantages over conventional methods. Foremost is improved diagnostic accuracy. Additional advantages include growth support of most pathogens observed in clinical practice and ease of adoption to existing protocols/instrumentationmaking the methodological transition of culture conditions simple, scalable, and affordable. Refined AST methods have potential benefit to both empiric antimicrobial therapy (prior to the receipt of blood culture and AST results) and definitive antimicrobial therapy (subsequent to blood culture and AST results), 58 which may ultimately improve clinical management and patient outcome. Testing in mammalian cell physiologic media exemplified by DMEM provides a platform for evaluation both of FDAapproved antibiotics and other compounds under development, potentially leading to significant cost and life savings.

Limitations of the study
In vitro assays are subject to inherent limitations since they fail to recapitulate the full spectrum of interactions of antibiotics between Report ll OPEN ACCESS the intact animal host and pathogen that are highly heterogeneous in time and space. Antibiotic concentrations can be modulated by absorption, distribution, metabolism, and excretion and further influenced by the dynamic nature of the infective process (nutrient availability, innate immune synergy, reactive metabolic product synergy). 13,14,59 Although improved diagnostic accuracy in DMEM was observed across a diversity of bacterial species and antimicrobials, these findings cannot be conclusive or generalized for MIC determination of individual bacterial species without increasing the number of clinical isolates tested to ensure sufficient clinical representation. Additionally, clinical outcomes derived from systemic infection may not apply to localized infections (respiratory, skin, UTIs), and thus testing in physiologic media more representative of the corresponding site of infection might increase the accuracy by which MIC assays predict in vivo efficacy. Further, human clinical efficacy and toxicity studies will need to be conducted to assure that these findings are applicable to patients with various infections and sepsis.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

DECLARATION OF INTERESTS
Antibiotic treatment Infected mice were treated (or mock-treated) with the following dosing regimens beginning 2 h post-infection: azithromycin (100 mg/kg/day), 23 72 or tetracycline (100 mg/kg/day). 76 All drug doses were delivered by the i.p. route once every 12 h except ertapenem, which was delivered once every 8 h. Mouse survival was assessed for 10 days post-infection.

Ethics statement
Human subjects approval was obtained from the Institutional Human Subjects Use Committee of the University of California, Santa Barbara and the Institutional Review Board of Santa Barbara Cottage Hospital. All animal experimentation was conducted following the National Institutes of Health guidelines for housing and care of laboratory animals and performed in accordance with Institutional regulations after pertinent review and approval by the Institutional Animal Care and Use Committee at the University of California, Santa Barbara.

QUANTIFICATION AND STATISTICAL ANALYSES
Statistical analysis of mouse survival Log-rank (Mantel-Cox) test was used to compare differences in survival between groups for Kaplan-Meier survival curves; significance was determined using GraphPad Prism version 9.2.0. P values of less than 0.05 were considered significant (n =10/cohort).

Statistical analysis of predicted & actual outcome
Antimicrobial susceptibility in murine sepsis models was evaluated using R Statistical Software (v4.2.0). 77 Statistical analyses returning a p value of <0.05 were considered significant. Experimental challenge experiments evaluated outcomes for multiple antimicrobials and bacterial species focusing on scenarios where there were discrepant classifications of antimicrobial susceptibility between tests performed utilizing different antimicrobial susceptibility testing media. All mice in the mock treated groups died providing an indication of expected mortality in the absence of effective treatment. Correct classification of susceptibility to an antimicrobial was anticipated to be associated with an increased proportion of mice surviving challenge. Conversely, an antimicrobial susceptibility classification of resistance was anticipated to be associated with an increased proportion of mice succumbing to challenge. The relationship between the susceptibility classification provided by each susceptibility testing method was compared by determining the proportion of animals that survived and died following challenge. Susceptibility testing method accuracy reflected the proportion of individual mouse outcomes that were consistent with the susceptibility predictions of the testing method where antimicrobial susceptibility was assigned a prediction of survival (the desired outcome should the antimicrobial be utilized in clinical practice) and antimicrobial resistance was assigned mortality (the outcome observed with mock treatment). Pairwise comparisons of test accuracy were performed between the media across all pathogen and antimicrobial combinations; further pairwise comparisons of test accuracy were performed between MHB and DMEM for each antibiotic and pathogen. EUCAST recommendations call for dose adjustments for ''intermediate'' susceptibility (I) 78 and, thus for statistical analysis, test results were dichotomized to either susceptible or not (intermediate or resistant) consistent with the EUCAST recommendations for standard antimicrobial dosing. Diagnostic accuracy was calculated as the number of animals that were predicted to survive and did survive, combined with the number of animals that were predicted to succumb and did succumb, divided by the total numbers of animals. For this statistical analysis, the predicted number of survivors for susceptibility (10/10 animals) and intermediate/resistance (0/10 animals) was compared to the actual number of survivors observed. The accuracy of antimicrobial susceptibility testing methods was calculated using the epiR R package (v2.0.52). 79 Fisher's exact test with false discovery rate (FDR) adjustment for multiple comparisons was used to compare the accuracy of susceptibility test methods using the RVAideMemoire R package (v0.9-81-2).