Logo: to the web site of Uppsala University

Antibiotic resistance is a severe danger to human health, and combination therapy with several antibiotics has emerged as a viable treatment option for multi-resistant strains. CombiANT is a recently developed agar plate-based assay where three reservoirs on the bottom of the plate create a diffusion landscape of three antibiotics that allows testing of the efficiency of antibiotic combinations. This test, however, requires manually assigning nine reference points to each plate, which can be prone to errors, especially when plates need to be graded in large batches and by different users. In this study, an automated deep learning-based image processing method is presented that can accurately segment bacterial growth and measure distances between key points on the CombiANT assay at sub-millimeter precision. The software was tested on 100 plates using photos captured by three different users with their mobile phone cameras, comparing the automated analysis with the human scoring. The result indicates significant agreement between the users and the software ([Formula: see text] mm mean absolute error) and remains consistent when applied to different photos of the same assay despite varying photo qualities and lighting conditions. The speed and robustness of the automated analysis could streamline clinical workflows and make it easier to tailor treatment to specific infections. It could also aid large-scale antibiotic research by quickly processing hundreds of experiments in batch, obtaining better data, and ultimately supporting the development of better treatment strategies. The software can easily be integrated into a potential smartphone application, making it accessible in resource-limited environments. Integrating deep learning-based smartphone image analysis with simple agar-based tests like CombiANT could unlock powerful tools for combating antibiotic resistance.

Population heterogeneity in bacterial phenotypes, particularly antibiotic resistance, is increasingly recognized as a major medical concern1. One specific form of phenotypic heterogeneity, known as heteroresistance (HR), refers to the presence of small subpopulations of resistant bacterial cells within a larger, otherwise susceptible population2. During antibiotic treatment, these rare resistant subpopulations can survive and proliferate, often leading to treatment failure and persistent infections. HR is clinically significant yet remains underdiagnosed because standard antibiotic susceptibility testing (AST) methods lack the sensitivity to detect resistant subpopulations that exist at very low frequencies, typically between 1 in 1,000,000 and 1 in 100,000 cells.

To address this diagnostic challenge, we have developed a 3D-printed droplet microfluidics-based platform combined with a texture-based image analysis tool to detect rare resistant subpopulations with frequencies as low as 10⁻⁶. The microfluidic chip consists of two main sections: (i) a droplet generation unit featuring a T-junction where an oil phase and an aqueous phase (containing bacteria in Mueller-Hinton broth mixed with antibiotics) to form discrete droplets (Fig. 1a), and (ii) an incubation chamber capable of housing approximately 2000–3500 droplets. In our multiplex design (Fig. 1b), multiple droplet generators and corresponding incubation zones are integrated to enable high-throughput screening. Each droplet typically encapsulates 1000–5000 bacterial cells, depending on parameters such as droplet volume, bacterial concentration (CFU/mL), the antibiotic used, and the expected frequency of resistant cells. If resistant cells are present, they can proliferate inside the droplets under antibiotic stress, altering the internal texture patterns. These changes can be quantified using image texture analysis, particularly through reduced homogeneity and correlation. Using this approach, we successfully detect HR in clinical isolates of Escherichia coli, Klebsiella pneumoniae, Pseudomonas aeruginosa, Acinetobacter baumannii, and Staphylococcus aureus against various antibiotics.

Population heterogeneity in bacterial phenotypes, such as antibiotic resistance, is increasingly recognized as a medical concern. Heteroresistance occurs when a predominantly susceptible bacterial population harbors a rare resistant subpopulation. During antibiotic exposure, these resistant bacteria can be selected and lead to treatment failure. Standard antibiotic susceptibility testing methods often fail to reliably detect these subpopulations due to their low frequency, highlighting the need for improved diagnostic approaches. Here, we present a droplet microfluidics method where bacteria are encapsulated in droplets containing growth medium and antibiotics. The growth of rare resistant cells is detected by observing droplet shrinkage under microscopy. We validated this method for three clinically important antibiotics in Escherichia coli isolates obtained from bloodstream infections and showed that it can detect resistant subpopulations as infrequent as 10^-6 using only 200 to 300 droplets. In addition, we designed a multiplex microfluidic chip to increase the throughput of the assay.

The rise in antifungal resistance has limited treatment options for serious fungal infections, emphasizing the need for effective combination therapies. However, low-cost and rapid systems to evaluate synergy and antagonism in antifungal combinations are lacking. Here, we introduce a novel in vitro testing method for assessing antifungal interactions in C. albicans, enabling the simultaneous testing of three antifungal agents in a single agar plate with overnight results. This method, validated against the checkerboard assay, provides consistent fractional inhibitory concentration (FICi) measurements with reduced variability and workload. We applied this method in a comprehensive screen of 92 clinical C. albicans isolates for three antifungals-amphotericin B, fluconazole, and anidulafungin-yielding assessments of a total of 276 distinct combinations of antifungals and isolates. Results revealed isolate-specific interaction patterns, with amphotericin B and fluconazole showing synergy in 1% of isolates, anidulafungin and fluconazole in 19.5%, and amphotericin B and anidulafungin in 23.9%. These findings underscore the need for isolate-specific testing in clinical settings. This proposed assay aims to present a solution to that as a scalable high throughput approach to this clinical problem.

Background

Antibiotic heteroresistance is a common bacterial phenotype characterised by the presence of small resistant subpopulations within a susceptible population. During antibiotic exposure, these resistant subpopulations can be enriched and potentially lead to treatment failure. In this study, we examined the prevalence, misclassification, and clinical effect of heteroresistance in Escherichia coli bloodstream infections for the clinically important antibiotics cefotaxime, gentamicin, and piperacillin–tazobactam.

Methods

We conducted a retrospective cohort analysis of patients (n=255) admitted to in-patient care and treated for E coli bloodstream infections within the Uppsala region in Sweden between Jan 1, 2014, and Dec 31, 2015. Patient inclusion criteria were admission to hospital on suspicion of infection, starting systemic antibiotics at the time of admission, positive blood cultures for the growth of E coli upon admission, and residency in the Uppsala health-care region at the time of admission. Exclusion criteria were growth of an additional pathogen than E coli in blood cultures taken at admission or previous inclusion of the patients in the study for another bloodstream infection. Antibiotic susceptibility of preserved blood culture isolates of E coli was assessed for cefotaxime, gentamicin, and piperacillin–tazobactam by disk diffusion and breakpoint crossing heteroresistance (BCHR) was identified using population analysis profiling. The clinical outcome parameters were obtained from patient records. The primary outcome variable was length of hospital stay due to the E coli bloodstream infection, defined as the time between admission and discharge from inpatient care as noted on the physician’s notes. Secondary outcomes were time to fever resolution, admission to intermediary care unit or intensive care unit during time in hospital, switching or adding another intravenous antibiotic treatment, re-admission to hospital within 30 days of original admission, recurrent E coli infection within 30 days of admission to hospital, and all-cause mortality within 90 days of admission.

Findings

A total of 255 participants with a corresponding E coli isolate (out of 500 screened for eligibility) met the inclusion criteria, with 135 female patients and 120 male patients. One (<1%) of 255 strains was BCHR for cefotaxime, 109 (43%) of 255 strains were BCHR for gentamicin, and 22 (9%) of 255 strains were BCHR for piperacillin–tazobactam. Clinical susceptibility testing misclassified 120 (96%) of 125 heteroresistant bacterial strains as susceptible. The BCHR phenotypes had no correlation to length of hospital stay due to the E coli bloodstream infection. However, patients with piperacillin–tazobactam BCHR strains who received piperacillin–tazobactam had 3·1 times higher odds for admittance to the intermediate care unit (95% CI 1·1–9·6, p=0·041) than the remainder of the cohort, excluding those treated with gentamicin. Similarly, those infected with gentamicin BCHR who received gentamicin showed higher odds for admittance to the intensive care unit (5·6 [1·1–42·0, p=0·043]) and mortality (7·1 [1·2–49·2, p=0·030]) than patients treated with gentamicin who were infected with non-gentamicin BCHR E coli.

Interpretation

In a cohort of patients with E coli bloodstream infections, heteroresistance is common and frequently misidentified in routine clinical testing. Several negative effects on patient outcomes are associated with heteroresistant strains.

Antibiotic resistance is a global concern with significant implications for healthcare, food production, and the environment. Therapies involving antibiotic combinations are frequently employed as a strategy to overcome antibiotic resistance. When present in combination, the efficacy of antibiotics may be enhanced or weakened, and as antibiotic interactions are a priori generally unpredictable, they need to be experimentally determined. Though antibiotics are regularly present at sublethal concentrations (e.g., in patients with suboptimal dosing regimens, late after the last dosage, difficult-to-penetrate tissues, and also in the natural environment), effects of antibiotic combinations are generally studied at lethal dosages. To address this, we developed the sublethal interaction factor (SIF) assay, based on the Bliss independence model, to quantify antibiotic combination effects at sublethal concentrations. SIF assay uses the whole growth curve, instead of only the growth rate, and determines reliably the outcome of the interaction between two antibiotics at sublethal concentrations. The SIF method was validated against the CombiANT assay and showed high sensitivity and specificity, attesting to its usability.

Heteroresistance to vancomycin among methicillin-resistant Staphylococcus aureus (MRSA) remains a diagnostic and therapeutic problem in clinical microbiology. In this prospective cohort study of 842 adult patients with MRSA bacteremia in S. Korea, we investigate the prevalence, risk factors, and clinical implications of the heteroresistant vancomycin-intermediate S. aureus (hVISA) phenotype. The hVISA phenotype is detected in 22% of cases. Multivariable regression analysis reveals strong positive associations between hVISA and hospital-acquired infection, prior anti-MRSA therapy, vancomycin exposure, and particularly vancomycin MIC (odds ratio 15.2 per 1 mg/L increase, p < 0.001). Strikingly, patients infected with hVISA strains have a lower 90-day mortality compared to those with fully susceptible strains (hazard ratio 0.66, p = 0.019), suggesting a possible trade-off between resistance and virulence. However, in hVISA strains treated with vancomycin, outcomes reverse: mortality more than doubled (HR 2.5, p < 0.001), bacteremia persisted longer, and relapse rates increased fivefold. Using maximally selected rank statistics, we identify a PAP–AUC threshold of 0.65 as the first clinically derived breakpoint predictive of mortality risk, providing an actionable definition of vancomycin heteroresistance. These findings underscore the clinical relevance of hVISA, and support routine testing for heteroresistance to inform treatment decisions.

Treatments with antibiotic combinations are becoming increasingly important even though the supposed clinical benefits of combinations are, in many cases, unclear. Here, we systematically examined how several clinically used antibiotics interact and affect the antimicrobial efficacy against five especially problematic Gram-negative pathogens. A total of 232 bacterial isolates were tested against different pairwise antibiotic combinations spanning five classes, and the ability of all combinations in inhibiting growth was quantified. Descriptive statistics, principal component analysis (PCA), and Spearman's rank correlation matrix were used to determine the correlations between the different combinations on interaction outcome. Several important conclusions can be drawn from the 696 examined interactions. Firstly, within a species, the interactions are in general conserved but can be isolate-specific for a given antibiotic combination and can range from antagonistic to synergistic. Secondly, additive and antagonistic interactions are the most common observed across species and antibiotics, with 87.1% of isolate-antibiotic combinations being additive, 11.6% antagonistic, and only 0.3% showing synergy. These findings suggest that to achieve the highest precision and efficacy of combination therapy, not only isolate-specific interaction profiling ought to be routinely performed, in particular to avoid using drug combinations that show antagonistic interaction and an expected associated reduction in efficacy, but also discovering rare and potentially valuable synergistic interactions.

IMPORTANCE

Antibiotic combinations are often used to treat bacterial infections, which aim to increase treatment efficacy and reduce resistance evolution. Typically, it is assumed that one specific antibiotic combination has the same effect on different isolates of the same species, i.e., the interaction is conserved. Here, we tested this idea by examining how several clinically used antibiotics interact and affect the antimicrobial efficacy against several bacterial pathogens. Our results show that, even though within a species the interactions are often conserved, there are also isolate-specific differences for a given antibiotic combination that can range from antagonistic to synergistic. These findings suggest that isolate-specific interaction profiling ought to be performed in clinical microbiology routine to avoid using antagonistic drug combinations that might reduce treatment efficacy. Antibiotic combinations are often used to treat bacterial infections, which aim to increase treatment efficacy and reduce resistance evolution. Typically, it is assumed that one specific antibiotic combination has the same effect on different isolates of the same species, i.e., the interaction is conserved. Here, we tested this idea by examining how several clinically used antibiotics interact and affect the antimicrobial efficacy against several bacterial pathogens. Our results show that, even though within a species the interactions are often conserved, there are also isolate-specific differences for a given antibiotic combination that can range from antagonistic to synergistic. These findings suggest that isolate-specific interaction profiling ought to be performed in clinical microbiology routine to avoid using antagonistic drug combinations that might reduce treatment efficacy.