A Pharmacometabonomic Approach To Predicting ... - ACS Publications

Page 1 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Proteome Research

A Pharmacometabonomic Approach to Predicting Metabolic Phenotypes and Pharmacokinetic Parameters of Atorvastatin in Healthy Volunteers Qing Huang1,2, Jiye Aa1, Huning Jia1,3, Xiaoqing Xin1,3, Chunlei Tao4, Linsheng Liu5, Bingjie Zou3, Qinxin Song1, Jian Shi6, Bei Cao6, Yonghong Yong7, Guangji Wang1, Guohua Zhou3* The first two authors contributed equally to this work 1. China Pharmaceutical University, Nanjing 210009, China; 2. Jiangsu Institute for Food and Drug Control, Nanjing 210008, China; 3. Department of Pharmacology, Jinling Hospital, Medical School of Nanjing University, Nanjing 210002, China; 4. Anhui University of Chinese Medicine, Hefei 230038, China; 5. Clinical Pharmacology Research Laboratory, The First Affiliated Hospital of Soochow University, Suzhou 215006, China; 6. Nanjing Drum Tower Hospital, the Affiliated Hospital of Nanjing University Medical School, Nanjing 210008, China; 7. The First Affiliated Hospital of Nanjing Medical University, Nanjing 210029, China Correspondence: Guohua Zhou

Guohua Zhou Professor, Dr., Corresponding author Department of Pharmacology, Jinling Hospital, Medical School of Nanjing University No. 305, Zhongshan East Road, Nanjing, Jiangsu 210002, PR China Tel: (+86) 84514223 Fax: (+86)84514223 E-mail: [email protected]

Check list

the number of References: 71 the number of Figures: 5 the number of Tables: 2 the number of Supplementary Figures: 5 the number of Supplementary Tables: 6 the number of Supplementary methods: 1

1

ACS Paragon Plus Environment


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 2 of 41

ABSTRACT Genetic polymorphism and environment each influence individual variability in drug metabolism and disposition. It is preferable to predict such variability, which may affect drug efficacy and toxicity, before drug administration. We examined individual differences

in

the

pharmacokinetics

of

atorvastatin

by

applying

gas

chromatography-mass spectroscopy (GC-MS)-based metabolic profiling to pre-dose plasma samples from 48 healthy volunteers. We determined the level of atorvastatin in plasma using LC/MSMS. With the endogenous molecules which showed a good correlation with pharmacokinetic parameters, a refined partial least squares model was calculated based on pre-dose data from a training set of 36 individuals, and exhibited good predictive capability for the other 12 individuals in the prediction set. In addition, the model was successfully used to predictively classify individual pharmacokinetic responses into subgroups. Metabolites such as tryptophan, alanine, arachidonic acid, 2-Hydroxybutyric acid, cholesterol and isoleucine were indicated as candidate markers for predicting, showing better predictive capability for explaining individual differences than conventional physiological index. These results suggest that a pharmacometabonomic approach offers the potential to predict individual differences in pharmacokinetics, and therefore to facilitate individualized drug therapy.

Key Words: Pharmacometabonomics, metabolomics, pharmacokinetics, personalized medicine, precision medicine, atorvastatin, prediction

2


Page 3 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


INTRODUCTION Personalized therapy is becoming increasingly important in medical treatment. It is expected to benefit patients, achieving better therapeutic effects with fewer side effects by tailoring stratified therapy to selected groups of patients based on some measure of how they will respond. Such personalization of drug treatments requires the ability to predict how different individuals will respond to a particular drug.

Thus far, most personalized approaches are based on investigations into genetic polymorphism and somatic mutation.1-5 Although a pharmacogenomic approach is useful in guiding drug treatment (especially single gene disorders), genomic information alone can only partly explain an individual’s response to drugs, since most diseases comprise a complex interplay between genetic and environmental influences.6, 7 In addition to genetic variation, personal physiopathological conditions, lifestyles, diet, environmental factors, and a combination/interaction of these factors also contribute to personalized medicine.

Through the global metabolic profiling of metabolites, which are the end products of systemic function and various metabolic pathways, metabolomics provides a unique method to characterize individual genetic variation and its net interactions among these contributing factors to drug responses. Clayton et al.(2006)8 first demonstrated that the pre-dose metabolic profile of urine samples could predict paracetamol toxicity and metabolism in rats, even without any previous genotypic knowledge, and 3



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 4 of 41

proposed the term pharmacometabonomic, which was defined as 'the prediction of the outcome of a drug or xenobiotic intervention in an individual based on a mathematical model of pre-intervention metabolite signatures. Various groups working with different biological models, including cell culture,9 rodents,10-13 and ultimately humans,14-19 eventually proved this strategy. Thus, pharmacometabonomics has become a potentially effective tool in the evaluation of individual differences for personalized medicine.7, 20-23

Thus far, most research has used pharmacometabonomics to predict individual drug responses and pharmacodynamic variations. We24 have used baseline metabolomic data to predict the pharmacokinetic parameters of triptolide in Sprague Dawley rats, and observed that some endogenous molecules in the baseline serum of Sprague Dawley rats correlated well with their maximum concentration in blood (Cmax) and area under the curve (AUC) values. Yoon and Hwang25 then measured the levels of metabolites in pre-dose urine samples from 29 healthy volunteers by LC-MS, and the result showed that pre-dose metabolite levels could predict individual variations in the pharmacokinetics of tacrolimus. To our knowledge, this is the only published work that uses pre-dose metabolomic data from urine samples to predict pharmacokinetic parameters in humans.

Atorvastatin is a new HMG-CoA reductase inhibitor that is widely used as an oral lipid-lowering drug to reduce plasma levels of LDL cholesterol (LDL-C) and the risk 4


Page 5 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


for coronary artery disease (CAD).26-30 There is considerable variability among individuals in response to atorvastatin therapy with respect to drug metabolism,31-37 treatment efficacy,38-41 and toxicity.42-45 The pharmacokinetic response to atorvastatin varies greatly among individuals; one study observed 45-fold or greater variability in plasma concentrations between patients taking the same daily atorvastatin dose.46 Such variability in atorvastatin pharmacokinetics is often associated with many individuals not reaching target therapy goals or suffering from severe adverse drug reactions, owing to the therapeutic effects as well as the ADRs are all proved to be related to atorvastatin concentrations in the plasma directly47-51. Apart from genetic polymorphisms (e.g., ABCB1,31, 32 OATP1B1,32, 33 ABCG2,34 CYPs,35, 36 and UGT37), life style, dietary preferences, concomitant medications, and environment also contribute to the variety of pharmacokinetic responses to atorvastatin among individuals.46, 52, 53 So far, it is difficult to predict an individual's PK response to atorvastatin. Therefore, the identification of predictive pre-treatment metabolic signatures (markers) would be of enormous clinical utility in order to assess individual variation, maximizing LDL-C lowering while minimizing the risk of adverse drug reactions.

In this study, we employed a pharmacometabonomic approach to predict the pharmacokinetic outcomes of atorvastatin dosing in humans. To our knowledge, this is one of the first studies in humans showing the ability of pre-drug plasma metabolic profiles to predict pharmacokinetics parameters on GC/MS platform. Pre-dose 5



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 6 of 41

baseline metabolites in plasma samples from 48 healthy volunteers were assessed using a combined gas chromatography and mass spectrometry (GC/MS)-based non-targeted, broad spectrum pathway agnostic metabolomics platform. Multivariate statistical techniques were used to screen potential markers of individual diversity by correlating endogenous molecules in baseline plasma with pharmacokinetic parameters. The data revealed that this pharmacometabonomic approach is able to effectively predict individual pharmacokinetic outcomes of atorvastatin dosing.

MATERIAL AND METHODS Healthy human volunteers. Forty-eight healthy male Chinese volunteers took part in this study. Primary diagnosis of the study participants consisted of hematology, urinalysis, biochemistry, serology, and a physical examination. Participants were excluded if they had any history or indication of renal, gastrointestinal, or hepatic abnormality. In addition, individuals with acute or chronic disease were excluded from the study. All volunteers were hospitalized at clinical research units for 72 h (24 h before and 48 h after drug administration), given the same regular diet with strict control over any other food intake, and kept under the same environmental and lifestyle conditions in order to avoid outside influences on pharmacokinetic and metabolomic data. Written informed consent was issued by all volunteers prior to study participation. The Research and Ethics committees of Jinling Hospital (Nanjing, China) approved the experimental protocol.

6


Page 7 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Study design. This was a single-center, randomized, open-label clinical trial. Sequential blood samples (6 ml) were collected in sodium heparinized tubes immediately before (0 h, after fasted overnight) and at 0.25, 0.5, 0.75, 1, 1.5, 2, 3, 4, 6, 8, 12, 24, 36, and 48 h after oral drug administration (atorvastatin, 20 mg). After centrifugation at 3000×g for 10 min, the resultant plasma was separated and stored at −80°C until analysis.

Analysis of atorvastatin concentrations. Atorvastatin was extracted from sodium heparin

anti-coagulated

plasma

samples

by

liquid–liquid

extraction.

The

concentrations of atorvastatin in plasma samples were determined using an AB API 3000 liquid chromatographic-tandem mass spectrometry system (LC-MS/MS, USA) with ESI interface. The SRM transitions of precursor ions to product ions were 559.2 →440.4 (m/z) for atorvastatin, and 285.2→154.2 (m/z) for the IS (diazepam) (see Supplementary Materials and Methods online for details).

Pharmacokinetic analysis. Pharmacokinetic analysis was performed using the DAS software version 2.1.1. The concentration-time curves were graphed, and the AUC and Cmax were calculated.

GC-MS–based global metabolic profiling. GC-MS–based global metabolic profiling was adapted from previous reports54,

55

with some modifications. The

analysis process involved plasma sample preparation, derivatization, GC-MS analysis, 7



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 8 of 41

data preprocessing, and the identification of metabolites (see Supplementary Materials and Methods online for details).

Statistical analysis. Statistical analysis was performed using a one-way ANOVA embedded in SPSS (version 16.0) with a significance level of 0.05 or 0.01. All significance

tests

were

two-sided.

Correlations

among

metabolites

and

pharmacokinetic parameters were obtained by deriving Spearman's correlation coefficient between each pair using SPSS (version 16.0). Multivariate statistical analysis (MVSA) was performed based on the dataset using SIMCA-P 13 software (Umetrics, Umea, Sweden), as published.56

Data were modeled using principal

component analysis (PCA) and partial least-squares (PLS) regression. To test the validity of all multivariate statistical models, 7-fold cross-validation was used, whereby a seventh of the data were left out of the model and then predicted back in, repeating the process until all of the data had been excluded at least once to calculate the Q2Y value (a measure of the cross-validated predictive ability). In addition, the permutation test was performed with an iteration of 100 to test the validity of models. The goodness-of-fit for a model was evaluated using three quantitative parameters: R2X, the explained variation in X; R2Y, the explained variation in Y; and Q2Y, the predicted variation in Y. The model’s parameters were checked carefully to avoid over-fitting of the model (the methods of modeling and validation are shown in Supplementary Materials and Methods online).

8


Page 9 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


The Pathway Analysis. Pathway enrichment and topology analyses were performed utilizing Metaboanalyst (http://www.metaboanalyst.ca). The Pathway Analysis module combines results from powerful pathway enrichment analysis with the pathway topology analysis to help researchers identify the most relevant pathways involved in the conditions under study. By uploading the discriminatory compounds that were significantly correlated with pharmacokinetic parameters, the built-in Homo sapiens (human) pathway library for pathway analysis and hypergeometric test for over-representation analysis was employed. A results report was then presented graphically as well as in a detailed table. Potential biomarkers were identified based on the identified metabolic pathways and the statistics. Databases such as KEGG, HMDB, Lipid Maps and MetaCyc were used to deduce some mechanisms, as well as by the literature reports.

RESULTS Healthy human volunteers Forty-eight healthy male Chinese volunteers were selected on the basis of their medical history and routine clinical laboratory tests, including hematology, urinalysis, biochemistry, serology, and physical examination (see Methods). The volunteers’ mean age was 25.75 ± 2.65 years, mean body weight was 67.13 ± 6.96 kg, and mean body mass index was 21.87 ± 1.58 kg/m2. Additional baseline characteristics of the study participants are described in Table S-1.

9



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 10 of 41

Pharmacokinetic study To evaluate the pharmacokinetic response, we measured the plasma concentration of atorvastatin at various time points after oral administration (20mg) in the 48 healthy volunteers. Graphing the plasma concentration of atorvastatin versus time revealed a high degree of individual variation regarding pharmacokinetic responses (Figure 1).

Four pharmacokinetic parameters for atorvastatin were estimated using DAS software (version 2.1.1): the maximum concentration in blood (Cmax), the time to reach Cmax (tmax), half-life (t1/2), and the area under curves (AUC0-t). The study participants’ maximum and minimum Cmax and AUC values differed by approximately 10-fold, despite controlling for physiological and environmental conditions. Cmax proved to be considerably important for a drug with a narrow therapeutic index and high individual variation to avoid adverse drug effects. AUC0-t represents total drug exposure and provides the best assessment of a drug’s variation among individuals. Therefore, we used Cmax and AUC0-t as response variables for subsequent pharmacometabonomic analysis.

We categorized the 48 study participants into three subgroups regarding respective Cmax and AUC values: low, medium, and high. Mean Cmax and AUC values are indicated in Table S-2. Next, we divided the participants into a training set (n=36) and a prediction set (n=12) for the subsequent pharmacometabonomic analysis. The prediction set was randomly selected to include 25% of individuals in the high 10


Page 11 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


(n=3/12), medium (n=6/24), and low (n=3/12) subgroups of the Cmax group and the AUC group; the remaining participants comprised the training set to construct a partial least squares (PLS) model, as described below. The pharmacokinetic parameters of Cmax and AUC differed significantly between the low and high subgroups (p1.0) were highly relevant to Cmax, and 57 variables (VIP >1.0) were highly relevant to AUC (enclosed by red boxes in Figure 2c, d).

Based on the above VIP metabolites, each of them also correlated with the corresponding pharmacokinetic parameter. The results of linear correlation regression analysis revealed that alanine, 2-hydroxybutyric acid, tocopherol, lysine, tryptophan, and some unidentified variables correlated well with Cmax (p1.0) were highly relevant to Cmax (c) or AUC (d) (as Y variables). 141x114mm (300 x 300 DPI)



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 3. Metaboanalyst (http://www.metaboanalyst.ca) generated topology map described the impact of baseline metabolites with high VIP values (VIP>1) on metabolic pathways. (a, b) The result of metabolic topology pathway analysis based on the baseline metabolites with high VIP values (VIP>1) in the initial Cmax(a) and AUC (b) model. (c, d) The result of over-representation pathway enrichment analysis based on the baseline metabolites with high VIP values (VIP>1) in the initial Cmax(c) and AUC (d) model. 169x164mm (300 x 300 DPI)


Page 38 of 41

Page 39 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 4. Refined PLS modeling to predict the Cmax and AUC of atorvastatin. (a, d) Measured versus predicted Cmax (a) and AUC (d)values for the two-component PLS model, in which all predictions related to model-building data. Colors from blue to red indicate increasing Cmax and AUC values. (b, e) Internal validation of the PLS model with 20 permutation tests to confirm predictability and data overfitting to predict Cmax (b) and the AUC (e). The result shows that all R2 (goodness-of-fit) and Q2 (predictive ability) values from the permuted models (left) were smaller than those of the original model (far right), demonstrating the validity of the PLS model. (c, f) PLS model external validity for predicting Cmax (c) and the AUC (f) of atorvastatin using the prediction set (n=12) samples: 3 in the low group (green), 6 in the medium group (blue), and 3 in the high group (red)]. Predicted values from the PLS model; all predictions of Cmax values exhibited a linear relationship with actual Cmax values (r=0.839) and all predictions of AUC values exhibited a linear relationship with actual AUC values (r=0.872). 99x56mm (300 x 300 DPI)



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 5. OPLS modeling to characterize pharmacokinetic response. (a, b) Score plot obtained using model-building data [low group (n=9, green), medium group (n=18, blue), and high group (n=9, red)] indicates relative clustering of three groups and distinct deviation of low-Cmax vs. high-Cmax subgroups (a) and low-AUC vs. high-AUC subgroups (b). (c, d) OPLS model validity for characterizing Cmax (c) and AUC (d) subgroups using 6 samples in the prediction set: low group (n=3, orange), high group (n=3, black). Using this prediction model, individuals can be categorized into the correct region. 119x81mm (300 x 300 DPI)


Page 40 of 41

Page 41 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Abstract graph 49x35mm (300 x 300 DPI)


A Pharmacometabonomic Approach To Predicting ... - ACS Publications

Recommend Documents