A proof of concept reinforcement learning based tool for non parametric population pharmacokinetics workflow optimization.

The building of population pharmacokinetic models can be described as an iterative process in which given a model and a dataset, the pharmacometrician introduces some changes to the model specification, then perform an evaluation and based on the predictions obtained performs further optimization. This process (perform an action, witness a result, optimize your knowledge) is a perfect scenario for the implementation of Reinforcement Learning algorithms. In this paper we present the conceptual background and a implementation of one of those algorithms aiming to show pharmacometricians how to automate (to a certain point) the iterative model building process.We present the selected discretization for the action and the state space. SARSA (State-Action-Reward-State-Action) was selected as the RL algorithm to use, configured with a window of 1000 episodes with and a limit of 30 actions per episode. SARSA was configured to control an interface to the Non-Parametric Optimal Design algorithm, that was actually performing the parameter optimization.The Reinforcement Learning (RL) based agent managed to obtain the same likelihood and number of support points, with a distribution similar to the reported in the original paper. The total amount of time used by the train the agent was 5.5 h although we think this time can be further improved. It is possible to automatically find the structural model that maximizes the final likelihood for an specific pharmacokinetic dataset by using RL algorithm. The framework provided could allow the integration of even more actions i.e: add/remove covariates, non-linear compartments or the execution of secondary analysis. Many limitations were found while performing this study but we hope to address them all in future studies.

Developing New Drugs for Mycobacterium tuberculosis Therapy: What Information Do We Get from Preclinical Animal Models?

Preclinical animal models of infection are employed to develop new agents but also to screen among molecules to rank them. There are often major differences between human pharmacokinetic (PK) profiles and those developed by animal models of infection, and these may lead to substantial differences in efficacy relative to that seen in humans. Linezolid is a repurposed agent employed to great effect for therapy of Mycobacterium tuberculosis In this study, we used the hollow-fiber infection model (HFIM) to evaluate the impact of different pharmacokinetic profiles of mice and nonhuman primates (NHP) versus humans on bacterial cell kill as well as resistance suppression. We examined both plasma and epithelial lining fluid (ELF) profiles. We examined simulated exposures equivalent to 600 mg and 900 mg daily of linezolid in humans. For both plasma and ELF exposures, the murine PK profile provided estimates of effect that were biased low relative to human and NHP PK profiles. Mathematical modeling identified a linkage between minimum concentrations (Cmin) and bacterial kill and peak concentrations (Cpeak) and resistance suppression, with the latter being supported by a prospective validation study. Finding new agents with novel mechanisms of action against M. tuberculosis is difficult. It would be a tragedy to discard a new agent because of a biased estimate of effect in a preclinical animal system. The HFIM provides a system to benchmark evaluation of new compounds in preclinical animal model systems against human PK effects (species scale-up estimates of PK), to safeguard against unwarranted rejection of promising new agents.

How to design a study to evaluate therapeutic drug monitoring in infectious diseases?

BACKGROUND: Therapeutic drug monitoring (TDM) is a tool to personalize and optimize dosing by measuring the drug concentration and subsequently adjusting the dose to reach a target concentration or exposure. The evidence to support TDM is however often ranked as expert opinion. Limitations in study design and sample size have hampered definitive conclusions of the potential added value of TDM. OBJECTIVES: We aim to give expert opinion and discuss the main points and limitations of available data from antibiotic TDM trials and emphasize key elements for consideration in design of future clinical studies to quantify the benefits of TDM. SOURCES: The sources were peer-reviewed publications, guidelines and expert opinions from the field of TDM. CONTENT: This review focuses on key aspects of antimicrobial TDM study design: describing the rationale for a TDM study, assessing the exposure of a drug, assessing susceptibility of pathogens and selecting appropriate clinical endpoints. Moreover we provide guidance on appropriate study design. IMPLICATIONS: This is an overview of different aspects relevant for the conduct of a TDM study. We believe that this paper will help researchers and clinicians to design and conduct high-quality TDM studies.

The Combination of Fosfomycin plus Meropenem Is Synergistic for Pseudomonas aeruginosa PAO1 in a Hollow-Fiber Infection Model.

Treating high-density bacterial infections is a challenging clinical problem. We have a paucity of new agents that can address this problem. Pseudomonas aeruginosa is a particularly difficult pathogen to treat effectively because of the plethora of resistance mechanisms it carries. Fosfomycin is an agent discovered circa 40 years ago. Recently, it has been resurrected in the United States and studied for intravenous therapy. We hypothesized that, to maximize its utility, it would require combination chemotherapy when used in a clinical circumstance in high-bacterial-burden infections. We chose to examine the combination of meropenem plus fosfomycin. These agents were studied in the hollow-fiber infection model. We utilized a fully factorial study design, looking at 2 doses of meropenem alone (1 and 2 g 8-hourly) and two doses of fosfomycin alone (6 and 8 g 8-hourly), as well as all possible combinations plus a no-treatment control. We used a high-dimensional model of 5 inhomogeneous differential equations with 5 system outputs to analyze all data simultaneously. Combination therapy outperformed all monotherapy regimens, with all combinations driving >6 log10 CFU/ml of bacterial killing. Combination therapy was able to counterselect resistance emergence (meropenem mutants being killed by the combination, as well as fosfomycin mutants being killed by the combination) in all regimens studied. The analysis demonstrated that the combination was significantly synergistic for bacterial cell killing and resistance suppression. Meropenem plus fosfomycin is a promising combination for therapy of high-burden Pseudomonas aeruginosa infections and requires further study.

Pharmacodynamics of teicoplanin against MRSA.

OBJECTIVES: The overall study aim was to identify the relevant preclinical teicoplanin pharmacokinetic (PK)/pharmacodynamic (PD) indices to predict efficacy and suppression of resistance in MRSA infection. METHODS: A hollow-fibre infection model and a neutropenic murine thigh infection model were developed. The PK/PD data generated were modelled using a non-parametric population modelling approach with Pmetrics. The posterior Bayesian estimates derived were used to study the exposure-effect relationships. Monte Carlo simulations from previously developed population PK models in adults and children were conducted to explore the probability of target attainment (PTA) for teicoplanin dosage regimens against the current EUCAST WT susceptibility range. RESULTS: There was a concentration-dependent activity of teicoplanin in both the in vitro and in vivo models. A total in vivo AUC/MIC of 610.4 (total AUC of 305.2 mg·h/L) for an MRSA strain with an MIC of 0.5 mg/L was needed for efficacy (2 log10 cell kill) against a total bacterial population. A total AUC/MIC ratio of ∼1500 (total AUC of ∼750 mg·h/L) was needed to suppress the emergence of resistance. The PTA analyses showed that adult and paediatric patients receiving a standard regimen were only successfully treated for the in vivo bactericidal target if the MIC was ≤0.125 mg/L in adults and ≤0.064 mg/L in children. CONCLUSIONS: This study improves our understanding of teicoplanin PD against MRSA and defines an in vivo AUC/MIC target for efficacy and suppression of resistance. Additional studies are needed to further corroborate the PK/PD index in a variety of infection models and in patients.

Tools for the Individualized Therapy of Teicoplanin for Neonates and Children.

The aim of this study was to develop a population pharmacokinetic (PK) model for teicoplanin across childhood age ranges to be used as Bayesian prior information in the software constructed for individualized therapy. We developed a nonparametric population model fitted to PK data from neonates, infants, and older children. We then implemented this model in the BestDose multiple-model Bayesian adaptive control algorithm to show its clinical utility. It was used to predict the dosages required to achieve optimal teicoplanin predose targets (15 mg/liter) from day 3 of therapy. We performed individual simulations for an infant and a child from the original population, who provided early first dosing interval concentration-time data. An allometric model that used weight as a measure of size and that also incorporated renal function using the estimated glomerular filtration rate (eGFR), or the ratio of postnatal age (PNA) to serum creatinine concentration (SCr) for infants <3 months old, best described the data. The median population PK parameters were as follows: elimination rate constant (Ke) = 0.03 · (wt/70)-0.25 · Renal (h-1); V = 19.5 · (wt/70) (liters); Renal = eGFR0.07 (ml/min/1.73 m2), or Renal = PNA/SCr (µmol/liter). Increased teicoplanin dosages and alternative administration techniques (extended infusions and fractionated multiple dosing) were required in order to achieve the targets safely by day 3 in simulated cases. The software was able to predict individual measured concentrations and the dosages and administration techniques required to achieve the desired target concentrations early in therapy. Prospective evaluation is now needed in order to ensure that this individualized teicoplanin therapy approach is applicable in the clinical setting. (This study has been registered in the European Union Clinical Trials Register under EudraCT no. 2012-005738-12.).

Population pharmacokinetics and pharmacodynamics of teicoplanin in neonates: making better use of C-reactive protein to deliver individualized therapy.

OBJECTIVES: There is uncertainty about the optimal teicoplanin regimens for neonates. The study aim was to determine the population pharmacokinetics (PK) of teicoplanin in neonates, evaluate currently recommended regimens and explore the exposure-effect relationships. METHODS: An open-label PK study was conducted. Neonates from 26 to 44 weeks post-menstrual age were recruited (n = 18). The teicoplanin regimen was a 16 mg/kg loading dose, followed by 8 mg/kg once daily. Therapeutic drug monitoring and dose adjustment were not conducted. A standard two-compartment PK model was developed, followed by models that incorporated weight. A PK/pharmacodynamic (PD) model with C-reactive protein serial measurements as the PD input was fitted to the data. Monte Carlo simulations (n = 5000) were performed using Pmetrics. The AUCs at steady state and the proportion of patients achieving the recommended drug exposures (i.e. Cmin >15 mg/L) were determined. The study was registered in the European Clinical Trials Database Registry (EudraCT: 2012-005738-12). RESULTS: The PK allometric model best accounted for the observed data. The PK parameters medians were: clearance = 0.435 × (weight/70)0.75 (L/h); volume = 0.765 (L); Kcp = 1.3 (h-1); and Kpc = 0.629 (h-1). The individual time-course of C-reactive protein was well described using the Bayesian posterior estimates for each patient. The simulated median AUC96-120 was 302.3 mg·h/L and the median Cmin at 120 h was 12.9 mg/L; 38.8% of patients attained a Cmin >15 mg/L by 120 h. CONCLUSIONS: Teicoplanin population PK is highly variable in neonates, weight being the best descriptor of PK variability. A low percentage of neonates were able to achieve Cmin >15 mg/L. The routine use of therapeutic drug monitoring and improved knowledge on the PD of teicoplanin is required.

Individualization of piperacillin dosing for critically ill patients: dosing software to optimize antimicrobial therapy.

Piperacillin-tazobactam is frequently used for empirical and targeted therapy of infections in critically ill patients. Considerable pharmacokinetic (PK) variability is observed in critically ill patients. By estimating an individual's PK, dosage optimization Bayesian estimation techniques can be used to calculate the appropriate piperacillin regimen to achieve desired drug exposure targets. The aim of this study was to establish a population PK model for piperacillin in critically ill patients and then analyze the performance of the model in the dose optimization software program BestDose. Linear, with estimated creatinine clearance and weight as covariates, Michaelis-Menten (MM) and parallel linear/MM structural models were fitted to the data from 146 critically ill patients with nosocomial infection. Piperacillin concentrations measured in the first dosing interval, from each of 8 additional individuals, combined with the population model were embedded into the dose optimization software. The impact of the number of observations was assessed. Precision was assessed by (i) the predicted piperacillin dosage and by (ii) linear regression of the observed-versus-predicted piperacillin concentrations from the second 24 h of treatment. We found that a linear clearance model with creatinine clearance and weight as covariates for drug clearance and volume of distribution, respectively, best described the observed data. When there were at least two observed piperacillin concentrations, the dose optimization software predicted a mean piperacillin dosage of 4.02 g in the 8 patients administered piperacillin doses of 4.00 g. Linear regression of the observed-versus-predicted piperacillin concentrations for 8 individuals after 24 h of piperacillin dosing demonstrated an r(2) of >0.89. In conclusion, for most critically ill patients, individualized piperacillin regimens delivering a target serum piperacillin concentration is achievable. Further validation of the dosage optimization software in a clinical trial is required.

Impact of Bolus dosing versus continuous infusion of Piperacillin and Tazobactam on the development of antimicrobial resistance in Pseudomonas aeruginosa.

Management of nosocomial pneumonia is frequently complicated by bacterial resistance. Extended infusions of beta-lactams are increasingly being used to improve clinical outcomes. However, the impact of this strategy on the emergence of antimicrobial resistance is not known. A hollow-fiber infection model with Pseudomonas aeruginosa (PAO1) was used. Pharmacokinetic (PK) profiles of piperacillin-tazobactam similar to those in humans were simulated over 5 days. Three dosages of piperacillin-tazobactam were administered over 0.5 h or 4 h, with redosing every 8 h. Two initial bacterial densities were investigated (∼10(4) CFU/ml and ∼10(7) CFU/ml). The time courses of the total bacterial population and the resistant subpopulation were determined. All data were described using a mathematical model, which was then used to define the relationship between drug concentrations, bacterial killing, and emergence of piperacillin resistance. There was logarithmic growth in controls in the initial 24 h, reaching a plateau of ∼9 log10 CFU/ml. Bacterial killing following administration of piperacillin via bolus dosing and that after extended infusions were similar. For the lower initial bacterial density, trough total plasma piperacillin concentration/MIC ratios of 3.4 and 10.4 for bolus and extended-infusion regimens, respectively, were able to suppress the emergence of piperacillin resistance. For the higher initial bacterial density, all regimens were associated with progressive growth of a resistant subpopulation. A stratified approach, according to bacterial density, is required to treat patients with nosocomial pneumonia. Antimicrobial monotherapy may be sufficient for some patients. However, for patients with a high bacterial burden, alternative therapeutic strategies are required to maximize bacterial killing and prevent antimicrobial resistance.

Role for a phage promoter in Shiga toxin 2 expression from a pathogenic Escherichia coli strain.

Shiga toxins (Stxs), encoded by the stxA and stxB genes, are important contributors to the virulence of Escherichia coli O157:H7 and other Stx-producing E. coli (STEC) strains. The stxA and stxB genes in STEC strains are located on the genomes of resident prophages of the lambda family immediately downstream of the phage late promoters (p(R')). The phage-encoded Q proteins modify RNA polymerase initiating transcription at the cognate p(R') promoter which creates transcription complexes that transcend a transcription terminator immediately downstream of p(R') as well as terminator kilobases distal to p(R'). To test if this Q-directed processive transcription plays a role in stx(2)AB expression, we constructed a mutant prophage in an O157:H7 clinical isolate from which p(R') and part of Q were deleted but which has an intact pStx, the previously described stx(2)AB-associated promoter. We report that production of significant levels of Stx2 in this O157:H7 isolate depends on the p(R') promoter. Since transcription initiating at p(R') ultimately requires activation of the phage lytic cascade, expression of stx(2)AB in STEC depends primarily on prophage induction. By showing this central role for the prophage in stx(2)AB expression, our findings contradict the prevailing assumption that phages serve merely as agents for virulence gene transfer.

N-mediated transcription antitermination in lambdoid phage H-19B is characterized by alternative NUT RNA structures and a reduced requirement for host factors.

Gene expression in lambdoid phages in part is controlled by transcription antitermination. For most lambdoid phages, maximal expression of delayed early genes requires an RNA polymerase modified by the phage N and host Nus proteins at RNA NUT sites. The NUT sites (NUTL and NUTR) are made up of three elements: BOXA, BOXB and an intervening spacer sequence. We report on N antitermination in H-19B, a lambdoid phage carrying shiga toxin 1 genes. H-19B N requires NusA, but not two other host factors required by lambda N, NusB and ribosomal protein S10. The H-19B NUT site BOXA is not required, whereas the BOXB is required for N action. H-19B nut sites have dyad symmetries in the spacer regions that are not in other nut sites. Changes in one arm of the dyad symmetry inactivate the NUT RNA. Compensating changes increasing the number of mutant nucleotides but restoring dyad symmetry restore activity. Deletion of the sequences encoding the dyad symmetry has little effect. Thus, the specific nucleotides composing the dyad symmetry seem relatively unimportant. We propose that the RNA stem-loop structure, called the 'reducer', by sequestering nucleotides from the linear RNA brings into proximity sites on either side of the dyad symmetry that contribute to forming an active NUT site.

The exciting future of antifungal therapy.

Invasive fungal infections are becoming more common. Current therapy is generally limited to amphotericin B in its parent and lipid formulations, 5-fluctyosine, fluconazole, and itraconazole. Toxicity, drug-drug interactions, and increasing fungal resistance reduce the usefulness of these drugs, and the need for new therapies is pressing. This article briefly discusses the limitations of antifungal minimum inhibitory testing, and then summarizes new antifungal drugs in development that have been tested in humans. It also addresses novel treatment strategies such as drug combination therapy, pharmacological reformulations to improve the efficacy or reduce the toxicity of current antifungal drugs, immune function augmentation, and vaccine development. All of these strategies, although in their infancy, will enhance the clinician's ability to care for patients with invasive fungal infections.

Arrangement and functional identification of genes in the regulatory region of lambdoid phage H-19B, a carrier of a Shiga-like toxin.

H-19B is a lambdoid phage that carries the genes (stx-I) encoding the two toxin subunits of a Shiga-like toxin; Escherichia coli lysogens of H-19B are converted to toxin producers. Based on the determination of a 17-kb region of the H-19B genome and functional studies, we have identified the early regulatory region and associated genes of H-19B, as well as the location of the late regulatory region and the toxin and lysis genes. A comparative analysis of the sequence of the H-19B genome reveals the presence of ORFs and genes found in analogous positions on the genomes of a number of other lambdoid phages. A cloned genomic fragment that confers immunity to an infecting H-19B phage contains an ORF of an analogous size and genomic location for a repressor gene, adjacent to a putative operator region. The lambda replication genes, O and P, are conserved in H-19B except for a 39-bp insert in the O gene creating two new O protein-binding sites in the origin of replication (ori), giving H-19B six binding sites as opposed to the four sites found in lambda. We identify ORFs and sequences involved in transcriptional regulation encoding N-like antitermination systems like those found in other lambdoid phages and nearly identical to sequences found in phage HK97. Our functional studies show that these sequences support antitermination even though they contain significant differences from those of other lambdoid phages. We also identify ORFs and sequences analogous to the Q-p'R late antiterminators-promoters found in other lambdoid phages. The Shiga-like stx-I genes are located directly downstream of the promoter, p'R, for the late genes, and upstream of the lysis genes.

Functional and genetic analysis of regulatory regions of coliphage H-19B: location of shiga-like toxin and lysis genes suggest a role for phage functions in toxin release.

Analysis of the DNA sequence of a 17 kb region of the coli lambdoid phage H-19B genome located the genes encoding shiga-like toxin I (Stx-I) downstream of the gene encoding the analogue of the phage lambda Q transcription activator with its site of action, qut at the associated pR' late promoter, and upstream of the analogues of lambda genes encoding lysis functions. Functional studies, including measurement of the effect of H-19B Q action on levels of Stx expressed from an H-19B prophage, show that the H-19B Q acts as a transcription activator with its associated pR'(qut) by promoting readthrough of transcription terminators. Another toxin-producing phage, 933W, has the identical Q gene and pR'(qut) upstream of the stx-II genes. The H-19B Q also activates Stx-II expression from a 933W prophage. An ORF in H-19B corresponding to the holin lysis genes of other lambdoid phages differs by having only one instead of the usual two closely spaced translation initiation signals that are thought to contribute to the time of lysis. These observations suggest that stx-I expression can be enhanced by transcription from pR' as well as a model for toxin release through cell lysis mediated by action of phage-encoded lysis functions. Functional studies show that open reading frames (ORFs) and sites in H-19B that resemble components of the N transcription antitermination systems controlling early operons of other lambdoid phages similarly promote antitermination. However, this N-like system differs significantly from those of other lambdoid phages.

A branch in the ToxR regulatory cascade of Vibrio cholerae revealed by characterization of toxT mutant strains.

Co-ordinate expression of genes associated with pathogenicity in Vibrio cholerae requires two transcription activators, ToxR and ToxT. Work carried out to date suggests that ToxR activates transcription of the toxT gene and that ToxT directly activates transcription of several genes whose products play a role in colonization or CT production by V. cholerae. Previous work also suggests that ToxR can directly activate transcription of the CT operon (ctxAB) independently of ToxT, thereby implying a degree of complexity in control of the cixAB operon not found with other genes of the ToxR regulon. We tested the regulatory cascade model of virulence gene expression by constructing strains of classical and El Tor V. cholerae deleted for the coding sequence for the putative DNA-binding domain of toxT. Phenotypic analysis of these strains suggests that V. cholerae has ToxT-dependent and ToxT-independent branches of its virulence regulon. The results also raise questions about the precise role for ToxR in activation of ctxAB transcription.

Kinetics of expression of the Escherichia coli cad operon as a function of pH and lysine.

The Escherichia coli cadBA genes are regulated at the transcriptional level by external pH and lysine. The membrane-localized CadC protein is required for activation of this operon under inducing conditions, which include acidic external pH, lysine, and oxygen limitation. To better understand the mechanism by which CadC functions, the kinetics of cadBA expression as a function of pH and lysine were examined. By primer extension assays, cadBA expression was detected within 4 min following exposure of cells to one of the inducing stimuli (low pH or lysine), provided that the cells had first been grown to steady state in the presence of the other inducing stimulus. The induction time was three to four times longer when both inducing stimuli were added simultaneously. cadBA expression was shut off within 4 min following a shift from acidic to neutral pH. Treatment of cells with chloramphenicol prevented induction by acidic pH and lysine. Transcription of lysP (encodes a lysine transporter) was also examined, since it is a negative regulator of cadBA expression in the absence of lysine. lysP expression was repressed by lysine but not influenced by pH. Putative transcription start sites for lysP and cadC were determined. Together, these data suggest that CadC senses the lysine- and pH-induced signals separately and that one of the roles of lysine in inducing cadBA may be to repress expression of lysP, thus eliminating the repressing effects of LysP.

Altered pH and lysine signalling mutants of cadC, a gene encoding a membrane-bound transcriptional activator of the Escherichia coli cadBA operon.

The Escherichia coli CadC protein is required for activation of cadBA transcription under conditions of low external pH and exogenous lysine. cadBA encodes proteins involved in the decarboxylation of lysine to cadaverine, and cadaverine excretion. Sequence analysis suggested that CadC contains a single transmembrane segment separating a DNA-binding domain in the amino terminus from a periplasmic domain. Western analysis of subcellular fractions demonstrated that CadC is expressed and concentrated in the cytoplasmic membrane in cells grown either at an inducing pH (pH 5.8) or at a non-inducing pH (pH 7.6). Eight cadC mutants were isolated based on their ability to confer expression of a cadA-lacZ fusion independent of low external pH or exogenous lysine. Five of these mutants expressed the cadA-lacZ fusion at both pH 5.8 and pH 7.6, but retained the requirement for the lysine signal while the other three mutants displayed pH independence at pH 5.8 but not at pH 7.6. These results support a model in which CadC is a membrane-bound transcriptional activator of the cadBA operon capable of sensing (directly or indirectly) signals generated outside the cytoplasmic membrane as a consequence of acidic pH and lysine.

Roles of LysP and CadC in mediating the lysine requirement for acid induction of the Escherichia coli cad operon.

Expression of the Escherichia coli cadBA operon, encoding functions required for the conversion of lysine to cadaverine and for cadaverine excretion, requires at least two extracellular signals: low pH and a high concentration of lysine. To better understand the nature of the lysine-dependent signal, mutants were isolated which expressed a cadA-lacZ transcription fusion in the absence of lysine while retaining pH regulation. The responsible mutation in one of these isolates (EP310) was in cadC, a gene encoding a function necessary for transcriptional activation of cadBA. This mutation (cadC310) is in a part of the gene encoding the periplasmic domain of CadC and results in an Arg-to-Cys change at position 265, indicating that this part of the protein is involved in responding to the presence of lysine. Three other mutants had mutations mapping in or near lysP (cadR), a gene encoding a lysine transport protein that has previously been shown to regulate cadA expression. One of these mutations is an insertion in the lysP coding region. Thus, in the absence of exogenous lysine, LysP is a negative regulator of cadBA expression. Negative regulation by LysP was further demonstrated by showing that lysP expression from a high-copy-number plasmid rendered cadA-lacZ uninducible. Expression of cadA-lacZ in a strain carrying the cadC310 allele, however, was not affected by the plasmid-expressed lysP. Cadaverine was shown to inhibit expression of the cadA-lacZ fusion in cadC+ cells but not in a cadC310 background.