Machine Learning-Led Optimization of Combination Therapy: Confronting the Public Health Threat of Extensively Drug Resistant Gram-Negative Bacteria.

Developing optimized regimens for combination antibiotic therapy is challenging and often performed empirically over many clinical studies. Novel implementation of a hybrid machine-learning pharmacokinetic/pharmacodynamic/toxicodynamic (ML-PK/PD/TD) approach optimizes combination therapy using human PK/TD data along with in vitro PD data. This study utilized human population PK (PopPK) of aztreonam, ceftazidime/avibactam, and polymyxin B along with in vitro PDs from the Hollow Fiber Infection Model (HFIM) to derive optimal multi-drug regimens de novo through implementation of a genetic algorithm (GA). The mechanism-based PD model was constructed based on 7-day HFIM experiments across 4 clinical, extensively drug resistant Klebsiella pneumoniae isolates. GA-led optimization was performed using 13 different fitness functions to compare the effects of different efficacy (60%, 70%, 80%, or 90% of simulated subjects achieving bacterial counts of 102 CFU/mL) and toxicity (66% of simulated subjects having a target polymyxin B area under the concentration-time curve [AUC] of 100 mg·h/L and aztreonam AUC of 1,332 mg·h/L) on the optimized regimen. All regimens, except those most heavily weighted for toxicity prevention, were able to achieve the target efficacy threshold (102 CFU/mL). Overall, GA-based regimen optimization using preclinical data from animal-sparing in vitro studies and human PopPK produced clinically relevant dosage regimens similar to those developed empirically over many years for all three antibiotics. Taken together, these data provide significant insight into new therapeutic approaches incorporating ML to regimen design and treatment of resistant bacterial infections.

A mechanism-based pathway toward administering highly active N-phage cocktails.

Bacteriophage (phage) therapy is being explored as a possible response to the antimicrobial resistance public health emergency. Administering a mixture of different phage types as a cocktail is one proposed strategy for therapeutic applications, but the optimal method for formulating phage cocktails remains a major challenge. Each phage strain has complex pharmacokinetic/pharmacodynamic (PK/PD) properties which depend on the nano-scale size, target-mediated, self-dosing nature of each phage strain, and rapid selection of resistant subpopulations. The objective of this study was to explore the pharmacodynamics (PD) of three unique and clinically relevant anti-Pseudomonas phages after simulation of dynamic dosing strategies. The Hollow Fiber Infection Model (HFIM) is an in vitro system that mimics in vivo pharmacokinetics (PK) with high fidelity, providing an opportunity to quantify phage and bacteria concentration profiles over clinical time scales with rich sampling. Exogenous monotherapy-bolus (producing max concentrations of Cmax = 7 log10 PFU/mL) regimens of phages LUZ19, PYO2, and E215 produced Pseudomonas aeruginosa nadirs of 0, 2.14, or 2.99 log10 CFU/mL after 6 h of treatment, respectively. Exogenous combination therapy bolus regimens (LUZ19 + PYO2 or LUZ19 + E215) resulted in bacterial reduction to <2 log10 CFU/mL. In contrast, monotherapy as a continuous infusion (producing a steady-state concentration of Css,avg = 2 log10PFU/mL) was less effective at reducing bacterial densities. Specifically, PYO2 failed to reduce bacterial density. Next, a mechanism-based mathematical model was developed to describe phage pharmacodynamics, phage-phage competition, and phage-dependent adaptive phage resistance. Monte Carlo simulations supported bolus dose regimens, predicting lower bacterial counts with bolus dosing as compared to prolonged phage infusions. Together, in vitro and in silico evaluation of the time course of phage pharmacodynamics will better guide optimal patterns of administration of individual phages as a cocktail.

Bayesian optimization of tacrolimus exposure in stable kidney transplant patients.

STUDY OBJECTIVE: The objective was to compare tacrolimus AUC0-12 determined by Non-Compartmental Analysis (NCA) using intensive sampling to Maximum a Posteriori-Bayesian (MAP-Bayesian) estimates from robust (n = 9 samples/subject) and sparse (n = 2 samples/subject) sampling in 67 stable KTRs and a validation group of similar patients. DESIGN: This open-label, prospective, single center 12-h PK study included nine serial samples collected in KTRs to determine steady-state NCA tacrolimus AUC0-12 . SETTING: This study was conducted at a single site within a large, urban hospital in the western New York area. PATIENTS: This study described tacrolimus pharmacokinetics in stable kidney transplant recipients on maintenance tacrolimus therapy. INTERVENTION: Robust and sparse AUC0-12 estimates by a MAP-Bayesian approach were obtained using the Advanced Dosing Solutions (AdDS) and ADAPT5 freeware. Limited sampling strategies were evaluated using the original population PK model (n = 67), which was also assessed using a validation group (n = 15). AUC0-12 agreement was tested by paired t-tests with intraclass correlation coefficient (ICC) and Bland Altman analysis. MEASUREMENTS AND MAIN RESULTS: A total of 35 Black and 32 White stable KTRs (estimated glomerular filtration rate [eGFR] = 55.2 ± 15.7 mL/min/1.73m2 ) received the tacrolimus dose of 3.4 ± 1.7 mg/study with troughs of 6.8 ± 1.8 ng/mL. The NCA-AUC0-12 was 123.8 ± 33.6 µg·h/L compared to MAP-Bayesian estimates for Robust-AUC0-12 of 124.7 ± 33.3 µg·h/L and optimal 2-specimen Sparse-AUC0-12 of 119.7 ± 32.7 µg·h/L for the training group. Comparison of Robust-AUC0-12 to NCA-AUC0-12 had an ICC of 0.96 (p = 0.99) while comparison of Robust-AUC0-12 to Sparse-AUC0-12 using Pre-dose trough [C(t0h )] and 1 h [C(t1h )] resulted in an ICC of 0.93 (p = 0.014). In the validation group, 5 Black and 10 White KTRs (eGFR = 56.4 ± 16.8 mL/min/1.73m2 ) received a mean tacrolimus dose of 1.9 ± 1.2 mg/study with a trough of 6.0 ± 1.7 ng/mL. The validation group's NCA-AUC0-12 (88.4 ± 33.1 µg·h/L) was comparable to Robust-AUC0-12 (85.1 ± 33.8 µg·h/L, ICC = 0.93; p = 0.12) and Sparse-AUC0-12 determined from C(t0h ) and C(t4h ) (86.7 ± 33.9 µg·h/L, ICC = 0.91; p = 0.61). CONCLUSION: MAP-Bayesian estimation for patient-specific AUC0-12 using sparse, two-specimen sampling is comparable to NCA and may enhance tacrolimus TDM in stable KTRs.

Open-source maximum a posteriori-bayesian dosing AdDS to current therapeutic drug monitoring: Adapting to the era of individualized therapy.

Recent updates in the therapeutic drug monitoring (TDM) guidelines for vancomycin have rekindled interest in maximum a posteriori-Bayesian (MAP-Bayesian) estimation of patient-specific pharmacokinetic parameters. To create a versatile infrastructure for MAP-Bayesian dosing of vancomycin or other drugs, a freely available, R-based software package, Advanced Dosing Solutions (AdDS), was created to facilitate clinical implementation of these improved TDM methods. The objective of this study was to utilize AdDS for pre- and post-processing of data in order to streamline the therapeutic management of vancomycin in healthy and obese veterans. Patients from a local Veteran Affairs hospital were utilized to compare the process of full re-estimation versus Bayesian updating of priors on healthy adult and obese patient populations for use with AdDS. Twenty-four healthy veterans were utilized to train (14/24) and test (10/24) the base pharmacokinetic model of vancomycin while comparing the effects of updated and fully re-estimated priors. This process was repeated with a total of 18 obese veterans for both training (11/18) and testing (7/18). Comparison of MAP objective function between the original and re-estimated models for healthy adults indicated that 78.6% of the subjects in the training and 70.0% of the subjects in the testing datasets had similar or improved predictions by the re-estimated model. For obese veterans, 81.8% of subjects in the training dataset and 85.7% of subjects in the testing dataset had similar or improved predictions. Re-estimation of model parameters provided more significant improvements in objective function compared with Bayesian updating, which may be a useful strategy in cases where sufficient samples and subjects are available. The generation of bespoke regimens based on patient-specific clearance and minimal sampling may improve patient care by addressing fundamental pharmacokinetic differences in healthy and obese veteran populations.

Rewinding Extinction in the Northern White Rhinoceros: Genetically Diverse Induced Pluripotent Stem Cell Bank for Genetic Rescue.

Extinction rates are rising, and current conservation technologies may not be adequate for reducing species losses. Future conservation efforts may be aided by the generation of induced pluripotent stem cells (iPSCs) from highly endangered species. Generation of a set of iPSCs from multiple members of a species can capture some of the dwindling genetic diversity of a disappearing species. We generated iPSCs from fibroblasts cryopreserved in the Frozen Zoo®: nine genetically diverse individuals of the functionally extinct northern white rhinoceros (Ceratotherium simum cottoni) and two from the closely related southern white rhinoceros (Ceratotherium simum simum). We used a nonintegrating Sendai virus reprogramming method and developed analyses to confirm the cells' pluripotency and differentiation potential. This work is the first step of a long-term interdisciplinary plan to apply assisted reproduction techniques to the conservation of this highly endangered species. Advances in iPSC differentiation may enable generation of gametes in vitro from deceased and nonreproductive individuals that could be used to repopulate the species.

Bacterial Mixology: Combining Pharmacodynamic Models to Predict In Vitro Competition of MCR-1-Harboring E. coli.

The emergence of mobile colistin resistance (mcr)-mediated polymyxin resistance has resulted in a significant detriment to the utility of the polymyxins in the clinical setting. Though the risk for horizontal transfer of an mcr-containing plasmid is a major component of the transmissibility, selection of polymyxin resistant subpopulations is still a major risk factor for developing polymyxin-resistant infections. Using static time-kills over 24 h (h), we performed competition studies by mixing known inocula of isogenic Escherichia coli strains (wildtype [WT] and mcr-1-harboring) and treating with a concentration array of polymyxin B. These results were then compared to a priori predictions of bacterial-killing effects by polymyxin B on a mixed population of E. coli cells using a previously published mechanism-based model. The data showed that both selective pressure between WT and mcr-1-harboring strains as well as underlying polymyxin B heteroresistance within each of the two strains contributed to bacterial regrowth despite treatment with high concentration polymyxin B. Moreover, the simulations showed that when mcr-1-harboring cells were 1% or 10% of the total population, regrowth by 24 h was still observed in ≥50% of the simulated subjects for both a 106 and 108 inoculum. These results indicate that at lower inoculums with a low proportion of mcr-1-harboring cells, selective pressure from a pharmacokinetic-optimized regimen of polymyxin B still results in regrowth and selection of polymyxin-resistant cells.