Skip to main content
NCBI home page
British Journal of Pharmacology logoLink to British Journal of Pharmacology
. 2015 Jun 29;172(15):3861–3874. doi: 10.1111/bph.13167

Model-based prediction of the acute and long-term safety profile of naproxen in rats

Tarjinder Sahota 1, Ian Sanderson 1, Meindert Danhof 1, Oscar Della Pasqua 1,2,3,✉
PMCID: PMC4523341  PMID: 25884765

Abstract

Background and Purpose

Despite the increasing importance of biomarkers as predictors of drug effects, toxicology protocols continue to rely on the experimental evidence of adverse events (AEs) as a basis for establishing the link between indicators of safety and drug exposure. Furthermore, biomarkers may facilitate the translation of findings from animals to humans. Combined with a model-based approach, biomarker data have the potential to predict long-term effects arising from prolonged drug exposure. Here, we used naproxen as a paradigm to explore the feasibility of a biomarker-guided approach for the prediction of long-term AEs in humans.

Experimental Approach

An experimental toxicology protocol was set up for evaluating the effects of naproxen in rats, in which four active doses were tested (7.5, 15, 40 and 80 mg·kg−1). In addition to AE monitoring and histology, a few blood samples were also collected for the assessment of drug exposure, TXB2 and PGE2 levels. Non-linear mixed effects modelling was used to analyse the data and identify covariate factors on the incidence and severity of AEs.

Key Results

Modelling results showed that besides drug exposure, maximum PGE2 inhibition and treatment duration were also predictors of gastrointestinal ulceration. Although PGE2 levels were clearly linked to the incidence rates, it appeared that ulceration severity is better predicted by measures of drug exposure.

Conclusions and Implications

These results show that the use of a model-based approach provides the opportunity to integrate pharmacokinetics, pharmacodynamics and toxicity data, enabling optimization of the design, analysis and interpretation of toxicology experiments.

Tables of Links

TARGETS
COX-1
COX-2
LIGANDS
Arachidonic acid PGE2
Naproxen TXB2
Open in a new tab

These Tables list key protein targets and ligands in this article which are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY (Pawson et al., 2014) and are permanently archived in the Concise Guide to PHARMACOLOGY 2013/14 (Alexander et al., 2013).

Introduction

A key purpose of preclinical general toxicity and safety pharmacology studies is to support the safe dose selection in humans. In particular, the need to understand the risks associated with long-term drug exposure falls within the remit of these two disciplines. However, despite the availability of a mixture of acute, midterm and chronic toxicity data prior to drug approval, the identification of long-term risks often happens in phase IV post marketing surveillance.

Ideally, earlier identification of potential risks would enable the use of evidence-based risk mitigation strategies. In order to achieve this goal, well-designed, integrated experimental protocols are needed in which time-dependent physiological changes arising from repeated exposure to a drug are characterized. Such an objective may be hampered by the use of traditional, empirical toxicology protocols, as they render the extrapolation of findings across species and across molecules rather difficult, preventing accurate translation of the pharmacological properties to man (Bai et al., 2013; Della Pasqua, 2013). Among other things, differences in sensitivity and target organ specificity continue to represent drawbacks for most clinical pathology parameters traditionally used for monitoring organ integrity both during preclinical toxicological assessment and clinical safety testing (Connelly and Bridges, 1991). Clearly, efforts are required to ensure the availability of tissue- and mechanism-specific data for accurate interpretation of acute and long-term safety findings.

In this context, biomarkers can be of great relevance, as they offer the possibility to discriminate between acute and chronic treatment effects. Over the last few years, several novel toxicity biomarkers have emerged as sensitive tools for detection, monitoring, quantification and prediction of safety and toxicity (O’Brien, 2008; Xie et al., 2013). Nevertheless, little attention has been given to the possibility of evaluating safety and toxicity using a mechanism-based approach whereby adverse events are assessed taking into account the underlying pharmacokinetic-pharmacodynamic (PKPD) properties of the molecule (McGonigle and Ruggeri, 2014). In the current investigation, we show therefore how PKPD modelling can be used to unravel the relationship between chronic drug exposure, pharmacodynamic effects, and overt symptoms and signs. The concept is illustrated by the correlation between naproxen concentrations, inhibition of PGE2 and TXB2 and gastric ulceration in rats.

Non-selective nonsteroidal anti-inflammatory drugs (NSAIDs), such as naproxen, act by blocking COX, which catalyses the rate-limiting step in the formation of prostanoids from arachidonic acid (Chakraborti et al., 2010). From a pharmacological perspective, various investigations have shown that both COX-1 and COX-2 are either constitutive or inducible in specific areas of the stomach of animals and humans (Morita, 2002; Coruzzi et al., 2007) (Figure 1). Hence, it can be anticipated that some balance between the activities of either isoform may be required for normal physiological function. Continuous COX-1 inhibition following prolonged administration of non-selective COX inhibitors is known to induce gastrointestinal (GI) adverse effects, especially ulceration and haemorrhagic bleeding (Loftin et al., 2002; Whittle, 2004; Wallace, 2008; Takeuchi, 2012). Unfortunately, at present, the dose selection of COX inhibitors disregards whether maximum, long-lasting blockade of either enzyme systems is strictly required for anti-inflammatory, analgesic response and how its pharmacology relates the observed adverse events (Huntjens et al., 2006). These considerations become essential when evaluating the side effects associated with long-term use of COX inhibitors, which include gastric and cardiac adverse events (Solomon et al., 2004; Schneeweiss et al., 2006).

Figure 1.

Figure 1

Open in a new tab

(Upper panel) Diagrammatic presentation of the processes underlying ulcer healing and factors affecting it. In the intact mucosa, COX-1 is the predominant COX isoform in the GI tract. In contrast, during wound healing, the expression of COX-2, rather than COX-1, is strongly increased in the repair zone. (Lower panel) Gastric effects of non-selective and COX-2-selective NSAIDs in normal or damaged gastric mucosa. The different effects of non-selective or selective COX-2 inhibition are explained by differences in COX-2 tissue expression (printed with permission from Halter et al., 2001; Coruzzi et al., 2007).

In spite of known interspecies differences that exist in GI-related morbidity (Kargman et al., 1996), we hypothesize that the characterization of the relationship between markers of COX inhibition and adverse events enables the prediction of safety windows for chronic treatment with selective and non-selective COX inhibitors. In fact, various studies provide further evidence of a multistage pathogenic mechanism for NSAID enteropathy by which the topical action of NSAIDs may initiate mucosal damage (Jackson et al., 2000; Halter et al., 2001), which is then converted to macroscopic damage by the concomitant inhibition of COX, with decreased mucosal PGs, presumably because of their effect on the microvasculature (Fornai et al., 2014).

In a previous investigation, we have shown how safety biomarkers can be used in conjunction with general toxicity protocols to predict the safety window in humans using an empirically derived safety threshold (Sahota et al., 2014a). However, safety thresholds such as the no-observed adverse-effect level (NOAEL) have many statistical and experimental limitations (Dorato and Engelhardt, 2005; Sahota et al., 2014b). Here we demonstrate that the use of a mechanism-based approach enables one to explore different parameterizations of biomarker response and drug exposure, facilitating the identification of potential (causal) factors associated with the overall treatment risk. In contrast to traditional extrapolation methods such as allometric scaling, which relies primarily on the point estimates of safe exposures, the use of hierarchical models can account for the variability in drug elimination or differences with respect to physiological, biochemical (e.g. expression of drug-metabolizing enzymes) and other time-variant factors (e.g. disease). These time-variant factors are increasingly more important as chronic interventions are common across most therapeutic areas.

Methods

The present investigation is based on a previously published general toxicity study in rats by Sahota et al. (2014a), with the non-selective COX inhibitor naproxen. A detailed description of the study design, strain of rats, sample collection, and analysis and PKPD modelling can be found in Sahota et al. (2014a).

Summary of study design

Three different treatment durations were investigated (1, 2 and 4 weeks). Rats were given daily doses of naproxen by p.o. gavage. The initial protocol was aimed at evaluating four cohorts per treatment duration at dose levels of 0, 15, 40 and 80 mg·kg−1·day−1 naproxen. However, the proposed dose levels had to be modified during the experimental phase as all animals receiving 80 mg·kg−1 in the 1 week cohort suffered from moderate weight loss. Subsequent experiments for the 2 week and 1 month cohorts were therefore performed with 7.5, 15 and 40 mg·kg−1 doses.

Satellite animals received identical doses to the toxicology groups and were used to measure plasma drug concentration (PK) and biomarker (PD) data, namely TXB2 and PGE2. Briefly, the biomarker data were obtained from whole blood ex vivo assays using LPS and aspirin for the analysis of PGE2, whereas as a stimulus clotted blood was used for the assessment of TXB2. These protocols provide evidence of the pharmacological effects of COX inhibitors on COX-1 and COX-2 activity respectively. An optimized composite sampling scheme was used and sampling took place on days 1, 7, 14 and 28. Details regarding sample analysis can be found in Sahota et al. (2014a). Endpoints in toxicology groups included adverse events, GI histology and terminal PK and PD measurements. Animals in the toxicology groups were killed at the end of their treatment period (1, 2 and 4 weeks). An overview of the study protocol is depicted in Figure 2. For each treatment duration, four groups were tested each of which received four different dose levels (n = 8 animals per dose level in the toxicology group and n = 24 animals per dose level in the satellite arm). This resulted in an experimental protocol with 96 animals in the toxicology group and 96 animals in the satellite group.

Figure 2.

Figure 2

Open in a new tab

Schematic representation of the design of the experimental protocol. The use of an integrated approach allowed the combination of additional data from (standard) pharmacokinetic experiments (DMPK experiments). Animals were assigned to different groups according to treatment duration and dose level. Treatment duration varied between 1, 2 and 4 weeks. For each treatment duration, four groups were tested each of which received four different dose levels (n = 8 animals per dose level in the toxicology group and n = 24 animals per dose level in the satellite arm, i.e. 3 animals per sampling time point). DMPK = drug metabolism and pharmacokinetic experiments including serial blood sampling.

The experimental procotol was approved by the DEC (Dierexperimentencommissie), i.e., the Animal Experimentation Committee of the University of Leiden, The Netherlands. The protocol procedures were in agreement with the ARRIVE guidelines for reporting experiments involving animals (Kilkenny et al., 2010; McGrath et al., 2010) as well as with the Dutch regulations regarding the use of animals for scientific research, as defined by the Legislation on Animal Experiments (Wet op Dierproeven), currently in force in the Netherlands.

Histology

Histological evaluation of the stomach was performed to establish a correlation between acute and long-term adverse events. Immediately after the rats had been killed, the stomachs were removed and cut open along the greater curvature and then washed with warm saline. The inner surface was photographed to allow the quantification of the area covered by haemorrhagic ulceration, which was determined under a dissecting microscope. Gastric ulceration was measured in terms of incidence and severity (percentage stomach surface area affected by ulceration). The software Image J version 1.43 (Abramoff et al., 2004) was used to calculate ulcer area and total stomach surface area. The person who performed the ulceration measurement was blinded as to animal identification and treatment group.

PKPD model

Naproxen concentrations and biomarker data were analysed by non-linear mixed effects modelling, as implemented in NONMEM version 7.2.0. Further details of the model building, diagnostics and results of the PK and PKPD analysis are reported elsewhere (Sahota et al., 2014a). Briefly, the PK of naproxen in plasma was best described by a one-compartment model with first-order absorption, first-order elimination and non-linear dose-dependent bioavailability. Weight was included as a significant covariate on clearance and volume of distribution. The PKPD models for both biomarkers, PGE2 and TXB2 were characterized by direct sigmoid IMAX models. Parameter values, precision estimates and goodness-of-fit diagnostics are shown in Supporting Information Table S1. Model-predicted biomarker levels were used as input (independent variables) in the analysis of adverse events.

Data analysis

Final model parameters describing the incidence of gastric ulceration and percentage gastric area affected were performed via the numerical integration routine ADVAN13 in NONMEM 7.2.0 using FOCE with Laplacian estimation. Convergence was determined by successful minimization and completion of the covariance step. All fitting procedures were performed on a computer (AMD-Athlon XP-M 3000+) running under Windows XP with a FORTRAN compiler (Compaq Visual Fortran, version 6.1). Data processing, management and graphical display were performed in R (R Development Core Team, 2014). Model diagnostics and validation were performed according to graphical and statistical criteria. Goodness-of-fit plots including observed (OBS) versus individual prediction (IPRED), OBS versus population prediction (PRED), conditional weighted residuals (CWRES) versus time, and CWRES versus OBS were used for diagnostic purposes (Hooker et al., 2007).

Given the purpose of the study was to discriminate between acute and long-term effects of naproxen, different parameterizations were considered to describe the observed drug effects during the course of treatment. Predicted exposure and biomarker levels for each individual animal were calculated from the final PKPD model using post hoc empirical Bayes estimates (using MAXEVAL = 0). Details of the calculation methods are described in Table 2004.

Table 1.

Calculation of biomarker response and exposure variables

Parameter name Symbol Calculation
Area under drug concentration versus time profile AUC Inline graphic
Area above biomarker concentration versus time profile AOC Inline graphic
Time under threshold (80% inhibition) TUT Inline graphic
Cumulative area under drug concentration versus time profile CAUC Inline graphic
Cumulative area over biomarker concentration versus time profile CAOC Inline graphic
Maximum drug concentration over 24 h period CMAX max({Cp(s) : t − 24 < s < t})
Maximum biomarker inhibition over 24 h period CMIN min({BCp(s) : t − 24 < s < t})
Open in a new tab

Individual-predicted naproxen concentrations and biomarker levels are denoted by Cp(t) and BCp(t) respectively.

Ulceration model

Because each histological examination was performed once per animal, no between-subject variability could be estimated. All random effects are therefore accounted for with the residual variability structure. Nevertheless, both the incidence and the severity of ulceration were considered during modelling. Incidence was modelled as the probability of occurrence of ulceration, Ui,j at the time of killing, tj, and severity was modelled as PERpred,i,j, the % gastric surface area affected, when ulceration is observable at the time of assessment.

A logit transformation was used to describe the incidence of stomach ulcers. The general equation describing the incidence of ulcers is given by equation 2004:

graphic file with name bph0172-3861-m1.jpg 1

where P(Ui,j) represents the probability of the presence of ulceration in individual i at time tj. COVi,j,k is the Kth covariate value for individual i and time tj. Θ1 is a parameter governing the baseline logit probability and Θk is the coefficient of the Kth covariate relationship.

For technical reasons, the severity of ulceration (i.e. percentage gastric surface area affected) was log transformed. The base model did not include any covariates on response. As shown in equations 2013 and 2013, two fixed effect model parameters were used, θ1 and θ2, which represent the typical values of the logit of the probability of ulceration and the severity of ulceration respectively.

graphic file with name bph0172-3861-m2.jpg 2
graphic file with name bph0172-3861-m3.jpg 3
graphic file with name bph0172-3861-m4.jpg 4

where PERobs,i,j and PERpred,i,j represent observed and predicted percentage ulceration, respectively, in individual i at time tj. εi,j is the random effect describing residual variability with mean 0 and estimated standard deviation.

Covariates

To explore the relationship between drug exposure, biomarkers and adverse events over the course of treatment, different secondary pharmacokinetic and pharmacodynamic parameters expressing systemic exposure and pharmacological activity were explored as covariates on the logistic model parameters using the stepwise covariate method (SCM) in PsN (Lindbom et al., 2005). The variables defined in Table 2004, in addition to body weight and age, were tested as potential influential factors on the incidence and severity of ulceration. Time measured in days (DAY) was also tested as a covariate and as a surrogate for time-dependent effects such as healing, tolerance or other mechanisms influencing ulceration incidence and/or severity. For the percentage gastric area affected, PER, the specification of the covariate relationship was based on the diagnostic plots of the base model. Linear, exponential and hyperbolic (sigmoid Emax) functions were considered during covariate model building. A hockey stick function was also tested to describe toxicity only manifesting above a threshold exposure/biomarker level. The linear relationship is described by equation 1996:

graphic file with name bph0172-3861-m5.jpg 5

where θ1 is population prediction and θslpoe is the slope of relationship between parameter and (centred) covariate.

Likewise, an exponential relationship was also evaluated as described by equation 2001:

graphic file with name bph0172-3861-m6.jpg 6

Because data were sparse and maximum effect may not have been reached, the maximum effect was fixed to 100% during the evaluation of the sigmoid Emax function, as shown in equation 2007:

graphic file with name bph0172-3861-m7.jpg 7

where γ is an estimated Hill coefficient and COV50 is the covariate value required to achieve 50% of maximum effect.

Covariate relationships were centred by the median value of the covariate so that in this case PERpred,i,j = θ2, as in the base model. To facilitate covariate testing, COV50 was reparameterized to be a factor θ of the median covariate value (equation 2013).

graphic file with name bph0172-3861-m8.jpg 8

As shown in equation 2010, the hockey stick function was implemented according to the following function:

graphic file with name bph0172-3861-m9.jpg 9

where θ1 is population prediction and θslope is the slope of relationship between parameter and (centred) covariate, and θTHRESH is the threshold value of the covariate where toxicity begins.

Covariates were incorporated into the model by stepwise forward inclusion. A significance level of P < 0.01 was used for inclusion, which represented a drop of at least 6.63 units in the objective function for each additional parameter. A final evaluation of the statistical significance of all factors identified during the previous step was performed by subtracting each covariate individually (backward elimination). The final structural model (i.e., fixed effects model) included only those covariates whose subtraction resulted in a decrease of at least 6.63 units in the objective function (P < 0.01). Finally, to investigate model uncertainty, a bootstrap SCM was performed using the bootscm option in PsN to estimate covariate inclusion probabilities (i.e. to estimate model specification uncertainty).

Model validation

The performance of the ulceration models was assessed by numerical and visual predictive checks. To that purpose, 1000 data sets were simulated with the final model parameter estimates. The mean and the 95% confidence intervals were calculated and displayed for the incidence and percentage gastric area affected. Validation procedures also included normalized prediction distribution errors, which are based on the assumption that the normalized (de-correlated) prediction distribution errors (discrepancies) are normally distributed (Comets et al., 2008). One hundred data sets were simulated using the final model, which was then tested for the assumption of normality of the prediction distribution errors.

Results

In total, 80 histological examinations were performed on toxicology group animals. These revealed gastric ulceration at all dose levels except for placebo, where no adverse events were observed. As a consequence, no NOAEL could be obtained for any of the treatment durations (Figure 3).

Figure 3.

Figure 3

Open in a new tab

Plots of the observed and predicted ulcer incidence (top) and severity (bottom). Dots in the upper panel show observed percentage of total animals in each cohort manifesting GI toxicity. Ulcer severity was measured as % of stomach area affected by ulceration. The shaded area depicts the 95% uncertainty in population prediction of the model (dashed lines depict the 50th percentile). The model was unable to describe the apparent negative dose–response trend observed after short treatment durations (i.e. 1 and 2 weeks), indicating that time-independent, long-lasting or irreversible processes may appear only after long-term treatment.

It should be noted that despite an initial protocol design with dose levels up to 80 mg·kg−1, dose levels of naproxen had to be reduced for the 2 week and 1 month cohorts for ethical reasons. This was due to moderate weight loss observed in all animals receiving 80 mg·kg−1 in the 1 week cohort. The animals in this cohort were immediately killed. The final protocol was therefore modified for the 2 week and 1 month cohorts, which received lower doses (7.5, 15 and 40 mg·kg−1). Histological examinations were performed on these animals and terminal blood samples taken. No other adverse events were observed.

Logistic models for gastric ulcerations

Empirical analysis of these data revealed some peculiarities in the data, in that there was no significant dose-response until week 4. In fact, due to the significantly negative parameter Θ2 (in Table 2013), the data revealed a possible negative dose–response relationship before week 4. Moreover, the incidence of ulcers was much lower in the week 4 cohort than in shorter treatment durations. Exploratory evaluation of the relationship between naproxen exposure and biomarker levels instead of dose did not provide further evidence of an apparent relationship.

Table 2.

Logistic model estimates describing ulcer incidence and severity

Parameter Value SEM
logit(P(Ui,j = 1)) = θ1 + θ2*IMAXPGE,i,j + θ3*DAYi,J
θ1 −0.226 0.305
θ2 0.042 64.8%
θ3 −0.066 0.022
Inline graphic
θ4 0.21% 17.2%
θ5 149.9 74%
θ6 94.3% 13%
Residual variability (proportional error) 78% 16%
Open in a new tab

After an initial attempt to describe the data without the use of covariates, a clear model misspecification was observed. As shown in Figure 3, the apparent negative dose–toxicity relationship for week 1 and week 2 was not replicated by the final model predictions. An overview of the parameter estimates for the final model is summarized in Table 2013, whereas Figure 4 shows the model specification uncertainty, as determined by the results of the bootstrap procedure. The wide, flat distribution reflects high model specification uncertainty (i.e. two or more different models may be indistinguishable). In addition, stepwise model building combined with bootstrap methods showed a possible negative correlation between the incidence of ulceration and treatment duration, indicating an acute effect dissipating over time (Figure 5). By contrast, upon incorporation of this time-dependent effect into the final model, the overall fit improved (Figure 6). CMAX and AUC were shown not be the primary drivers of toxicity, although these parameters may be considered indirectly correlated with risk.

Figure 4.

Figure 4

Open in a new tab

Model specification uncertainty. Bars indicate model selection probability as determined by the bootstrap SCM ordered from most to least probable. Only the top five most probable models are displayed. A wider, flatter distribution reflects high model specification uncertainty (i.e. two or more different models may be indistinguishable). The overlaid table in the inset shows numerical details of model selection probability.

Figure 5.

Figure 5

Open in a new tab

Visual predictive checks. The upper panel depicts the prediction distribution for the incidence of ulcers by dose level. Observed incidence is highlighted by the red lines whereas dashed lines represent the 90% prediction intervals (PIs). Of the eight groups, one would expect on average one group to fall beyond the 90% PIs, which is in line with the observed results. Ulcer incidence is determined as the observed percentage of total animals manifesting GI toxicity. The lower panel shows ulceration severity against predicted normalized exposure for different variables of interest (AOC = area above biomarker concentration vs. time profile; CAUC = cumulative area under the concentration vs. time profile). The shaded area represents the 95% PI; dots depict the actual data. PK = naproxen concentrations.

Figure 6.

Figure 6

Open in a new tab

Differences in the sensitivity of explanatory variables describing the relationship between drug exposure/biomarker levels and adverse events as determined by the incidence of ulcers and ulceration severity. Dashed lines represent the median profile of simulated values using the final model, whereas shaded area represent the 95% prediction intervals. See text for details on the units of the independent variables (x-axis). Based on statistical criteria, it appears that maximum inhibition of PGE2 was the best predictor of adverse event incidence. However, cumulative TXB2 inhibition was found to be the best explanatory variable for the severity of ulceration. Given the degree of model uncertainty, a note of caution is required before assigning clinical meaning to these covariates.

Discussion and conclusions

Model-based evaluation of safety pharmacology and toxicology data

Current practices in toxicology and safety pharmacology rely on the concept of thresholds of drug exposure (e.g. NOAEL) as a proxy for the risk of adverse events, which are treated in a mechanism- and time-independent manner. The disadvantage of such an approach is that long-term toxicity can become conflated with acute toxicity, which in turn could be mitigated or related to entirely different physiological mechanisms (Blantz, 1996; Dom et al., 2012). Another hurdle to overcome in the assessment of risk is that general toxicity studies are not designed to characterize the relationship between drug exposure and toxicity, but rather to explore the boundary between therapeutic and toxic exposures (Josa et al., 2001). As such, data can be uninformative with respect to understanding the causal factors and underlying mechanisms associated with unwanted pharmacological effects. Clearly, these inefficiencies in experimental protocol design also violate the principle of the 3 Rs (reduction, refinement and replacement) and ultimately contribute to biased conclusions about the long-term benefit-risk ratio of an intervention (Balls, 1994). By contrast, the use of a model-based approach provides the opportunity to integrate safety and toxicity data and assess in a strictly quantitative manner the contribution of influential factors, namely drug exposure and biomarkers of pharmacological activity to potential adverse events (Danhof et al., 2005; 2008,; Bai et al., 2013).

From a methodological perspective, general toxicology studies represent a challenge for model-based analysis because sparse pharmacokinetic data derived from satellite animals need to be linked to adverse event data, which are also typically sparse. In addition, lack of individual exposure profiles often prevents further evaluation of the role of relevant physiological or pathophysiological measures, such as biochemistry, haematology or biomarker data as influential covariates on treatment outcome. Typical experimental protocols in toxicology research yield therefore less informative data sets, as compared with studies aimed at the characterization of PKPD relationships, which are now commonly used in early drug development (Knight, 2007).

An important aspect of our analysis was therefore to show that without major modification to existing general toxicity protocols, it is possible to explore and eventually elucidate the causal relationship between drug administration, exposure, and the incidence and severity of adverse events associated with chronic therapy. In addition to the integrated analysis of pharmacokinetic, pharmacodynamic and toxicity data, here we have highlighted that the lack of NOAEL (due to the presence of adverse events at all tested dose levels) has not prevented us from further characterizing the exposure–adverse event relationships. The main modifications consisted of the additional collection of biomarker data from animals and the choice to treat histological observations as a continuous data type. The incorporation of biomarkers into the assessment of long-term toxicity enabled us to further understand time dependencies and nonlinearities in downstream effects related to the primary pharmacological target (Huntjens et al., 2010).

Mechanism-based analysis of long-term safety and toxicology data

The present analysis suggested that the ulcerative effect is acute and diminishes with sustained long-term exposure. In addition, CMAX and AUC were not selected features of the final model, although these parameters may be indirectly correlated with risk. Our attempt to establish a relationship between drug exposure/biomarker levels and adverse events revealed clear differences in the sensitivity of explanatory variables used to describe the incidence of ulcers and ulceration severity. Out of tested relationships, the maximum inhibition of PGE2 (IMAXPGE) was the best predictor of adverse event incidence, with the bootstrap SCM showing low model uncertainty. On the other hand, cumulative TXB2 inhibition (CAOCTXB) was found to be the best explanatory variable for the severity of ulceration (Figure 6). Other physiologically plausible explanatory factors such as maximum PGE2 inhibition or DAY (treatment duration) were found to be fraught with significant model uncertainty. Physiologically, it is plausible that the ulcerative effect of naproxen primarily reflects the blockade of tissue damage repair by COX-2 (Brzozowski et al., 2001), whereas the severity of ulcerations may potentially be linked to the antiplatelet (i.e. COX-1 dependent) prostanoid inhibition in the GI tract (Capone et al., 2007).

PKPD relationships as a translational factor for the evaluation of risk in humans

It is important to emphasize that for the GI toxicity, animal data are not easily translated to humans. In humans, it is still unclear what clinical end point is best related to the inhibition of prostanoid biosynthesis in the GI tract and which isozyme is involved. The results of randomized clinical trials together with the knowledge of the impact of the different drugs on COX-1 and COX-2 activity at therapeutic doses (using human whole blood assays ex vivo) suggest therefore that gastroduodenal lesions develop as a consequence of persistent, moderate inhibition of mucosal COX-1 activity in a large proportion of exposed patients, while bleeding complications occur as a result of transient, high-grade inhibition of platelet COX-1 in a very small percentage of exposed patients.

Randomized clinical trials with coxibs versus traditional NSAIDs have focused mainly on three classes of clinical end points (Rostom et al., 2007; Higuchi et al., 2009). First, GI symptoms such as dyspepsia, which have a poor signal-to-noise ratio and for which the COX-1 dependence of the signal is uncertain. In addition, these symptoms also fail to correlate with the presence of detectable GI lesions. Second, endoscopically detectable lesions, which are largely COX-1 dependent and have a favourable signal-to-noise ratio. However, it remains uncertain whether they are reliable predictors of serious GI complications. Third, serious GI complications (i.e. perforation, ulcer and bleeding; perforation, obstruction and bleeding). These measures have an uncertain signal-to-noise ratio because of lack of adequately sized placebo-controlled studies. Moreover, the haemorrhagic nature of the most prevalent component of these combined endpoints makes it likely that they reflect primarily GI complications related to inhibition of platelet by COX-1.

Taking into account the points mentioned above, it becomes clear that the relationship between chronic exposure and the incidence and severity of adverse events should be considered a critical but not sufficient requirement to predict safety and toxicity in humans. Therefore, PKPD models need to be parameterized in such a way that it is possible to discriminate between drug-specific and system-specific parameters. Understanding of pharmacokinetic differences in conjunction with detailed information on potential system-specific differences, such as varying metabolic capacity, is sine qua non conditions to translate and accurately interpret safety findings (Zuideveld et al., 2007; Chain et al., 2013). Even when estimation of such parameters may be impractical, inferences can be made about their magnitude. Undoubtedly, a mechanism-based approach is likely to yield more reliable predictions than the currently accepted use of empirical cover or safety margin which disregard any possible pharmacological basis for both observed and unobserved adverse events. In fact, the role of COX inhibition in the characterization of NSAID-induced adverse events was recently investigated by Massó González et al. (2010), who describe the relationship between the degree of inhibition of whole blood COX-1 and COX-2, the maximum plasma concentration (Cmax) of individual NSAIDs, and relative risk of upper GI bleeding/perforation in humans. Drugs that have a long half-life or slow-release formulation and/or correlate with profound and coincident inhibition of both COX isozymes were found to be associated with a greater risk of upper GI bleeding/perforation.

Whereas the experimental results for naproxen seem to concur with the conclusions drawn by Massó González et al. (2010), formal extrapolation of our findings requires further information on system-specific properties, including potential differences in gastric mucosa susceptibility to ulceration and expression and activity of isozymes during maintenance and repair processes. Rats appear to be more susceptible to GI toxicity than humans and show gender-specific differences in ulceration, so any prediction without correcting for such differences is therefore likely to overestimate risk (Urushidani et al., 1978; Lanza et al., 1979). In addition, we have shown that multiple models with different explanatory variables meet the statistical criteria used for fitting procedures. These apparently conflicting findings can be interpreted as model uncertainty due to design or even imprecision in parameter estimation. On the other hand, these same results can also be considered hypotheses generating (i.e., they shed light into the possible or even plausible combination of mechanisms underpinning the causal path(s) between drug exposure and toxicity). This latter aspect is essential for extrapolating data from animals to humans. In fact, in a study performed by Huntjens et al. (2006), the authors conclude that the main determinant of the primary anti-inflammatory, analgesic effect is the degree of target engagement at the tested dose ranges, as defined by the inhibition of PGE2 and TXB2.

Potential limitations

Unfortunately, our investigation is just a first step into the use of a pharmacologically based approach to the evaluation of adverse events. This work does not address differences between the occurrence of adverse events in animals versus humans. This is a weakness that applies to any analysis of these data regardless of whether it is based on PKPD modelling or traditional safety thresholds (e.g. NOAEL). We anticipate that to address this limitation, systematic, integrated analysis of drug effects in vitro, in animals and humans is required. Such an analysis implies access to data from different doses and compounds, so that one can discriminate between drug- and system-specific properties.

We also understand that establishing correlations between clinical and preclinical data would be useful to support claims about the predictive value of the approach. However, this would also imply establishing the sensitivity and specificity of the endpoints, including a range of compounds, rather than a single drug. Our initial intent was therefore to show that any attempt to translate or predict drug effects in humans requires careful consideration of the PKPD relationships of the adverse events of interest. In addition, we realize that the logistic models proposed here are empirical in nature, but we believe that such a parameterization may not necessarily undermine its use prospectively. Fully mechanistic models (i.e. systems biology type of models) will be possible only for known targets and molecules which have already been widely tested. For novel chemical and biological entities, understanding of the mechanisms underlying adverse events is very limited, and in most cases unknown at the time safety pharmacology and toxicology protocols are implemented. This issue will remain even when modelling techniques are applied (Nyman et al., 2012). The difference, however, is the possibility to identify drug- (CL, Vd) or system-specific properties (e.g. IC50, IMAX) that can be generalized and/or scaled across species. The merit of this work is therefore to show that markers of pharmacological activity can be treated as explanatory variables for adverse events. These findings can be used in conjunction with allometrically scaled pharmacokinetic data to explore the implications of such effects in humans.

The current manuscript is a first step towards a more pharmacologically based approach for the evaluation of toxicity. The limitations of the modelling exercise presented here are mostly due to the nature of the data package available during (early) preclinical evaluation (i.e. the types of experimental protocols currently used in safety pharmacology and general toxicology). We acknowledge that the absence of ulcerations in vehicle-treated animals and the lack of additional cohorts with lower exposure levels may represent a weakness in our investigation. True baseline rates for ulceration could not be factored into the analysis, nor was it possible to accurately establish the adverse event rates at lower doses. We also recognize that adopting a different modelling approach to link the probability of adverse effect with the time course of the selected biomarkers and unobserved tissue repair mechanisms may provide further insight into the drivers of toxicity, more specifically ulceration and bleeding (Motsko et al., 2006). However, this approach would require a much more labour-intensive protocol design, including destructive sampling at different intervals to ensure sufficient data are available on the time course of biomarkers and tissue repair status.

In summary, the findings of this study demonstrate the feasibility and potential benefits of a model-based approach for the evaluation of chronic safety pharmacology and toxicity. The present work has shown that even for relatively frequent adverse events, model uncertainty can be significant and therefore one should quantify it. This probably arises from the fact that toxicity studies are generally designed to find safety windows and not to explore the entire exposure-adverse event profile. Moreover, the availability of PKPD relationships may allow us to make inferences about untested doses and dosing regimens, providing an opportunity for risk mitigation, even before obtaining experimental clinical data.

Author contributions

O. D. P. and T. S. defined the research questions, planned the experimental protocol and selected the analysis methods. T. S. and I. S. performed the in vivo experiments and data analysis. T. S. wrote the first draft of the paper that has been finalized with the contributions of M. D. and O. D. P.

Conflict of interest

The authors declare that there are no conflicts of interest.

Glossary

AUC

area under the concentration versus time curve

CAOC

cumulative area over biomarker concentration versus time profile

CMAX

maximum drug concentration over the period of 24 h

CMIN

maximum biomarker inhibition over the period of 24 h

CWRES

conditional weighted residuals

IMAX

maximum biomarker inhibition

IPRED

individual prediction

NOAEL

no-observed adverse-effect level

NSAID

nonsteroidal anti-inflammatory drug

OBS

observed concentration or levels

PD

pharmacodynamics

PK

pharmacokinetics

PKPD

pharmacokinetic-pharmacodynamic

PRED

population prediction

SCM

stepwise covariate method

Supporting Information

Additional Supporting Information may be found in the online version of this article at the publisher's web-site:

Table S1 Summary of naproxen population PK and PKPD model parameters (Sahota et al., 2014).

bph0172-3861-sd1.docx (13.3KB, docx)

References

  1. Abramoff MD, Magalhaes PJ, Ram SJ. Image processing with ImageJ. Biophotonics Int. 2004;11:36–42. [Google Scholar]
  2. Alexander SPH, Benson HE, Faccenda E, Pawson AJ, Sharman JL, Spedding M, et al. The Concise Guide to PHARMACOLOGY 2013/14: enzymes. Br J Pharmacol. 2013;170:1797–1867. doi: 10.1111/bph.12451. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Bai JP, Fontana RJ, Price ND, Sangar V. Systems pharmacology modelling: an approach to improving drug safety. Biopharm Drug Dispos. 2013 doi: 10.1002/bdd.1871. doi: 10.1002/bdd.1871. [DOI] [PubMed] [Google Scholar]
  4. Balls M. Replacement of animal procedures: alternatives in research, education and testing. Lab Anim. 1994;28:193–211. doi: 10.1258/002367794780681714. [DOI] [PubMed] [Google Scholar]
  5. Blantz RC. Acetaminophen: acute and chronic effects on renal function. Am J Kidney Dis. 1996;28(1 Suppl. 1):S3–S6. doi: 10.1016/s0272-6386(96)90561-2. [DOI] [PubMed] [Google Scholar]
  6. Brzozowski T, Konturek PC, Konturek SJ, Sliwowski Z, Pajdo R, Drozdowicz D, et al. Classic NSAID and selective cyclooxygenase (COX)-1 and COX-2 inhibitors in healing of chronic gastric ulcers. Microsc Res Tech. 2001;53:343–353. doi: 10.1002/jemt.1102. [DOI] [PubMed] [Google Scholar]
  7. Capone ML, Tacconelli S, Sciulli MG, Anzellotti P, Di Francesco L, Merciaro G, et al. Human pharmacology of naproxen sodium. J Pharmacol Exp Ther. 2007;322:453–460. doi: 10.1124/jpet.107.122283. [DOI] [PubMed] [Google Scholar]
  8. Chain A, Dubois V, Danhof M, Sturkenboom M, Della Pasqua O. Identifying the translational gap in the evaluation of drug-induced QTc-interval prolongation. Br J Clin Pharmacol. 2013;76:708–724. doi: 10.1111/bcp.12082. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Chakraborti AK, Garg SK, Kumar R, Motiwala HF, Jadhavar PS. Progress in COX-2 inhibitors: a journey so far. Curr Med Chem. 2010;17:1563–1593. doi: 10.2174/092986710790979980. [DOI] [PubMed] [Google Scholar]
  10. Comets E, Brendel K, Mentre F. Computing normalised prediction distribution errors to evaluate nonlinear mixed-effect models: the npde add-on package for R. Comput Methods Programs Biomed. 2008;90:154–166. doi: 10.1016/j.cmpb.2007.12.002. [DOI] [PubMed] [Google Scholar]
  11. Connelly JC, Bridges JW. Species variation in target organ toxicity. In: Cohen GM, editor. Target Organ Toxicity. Boca Raton, FL: CRC Press; 1991. pp. 89–115. . In: (ed.). [Google Scholar]
  12. Coruzzi G, Venturi N, Spaggiari S. Gastrointestinal safety of novel nonsteroidal anti-inflammatory drugs: selective COX-2 inhibitors and beyond. Acta Biomed. 2007;78:96–110. [PubMed] [Google Scholar]
  13. Danhof M, Alvan G, Dahl SG, Kuhlmann J, Paintaud G. Mechanism-based pharmacokinetic-pharmacodynamic modeling-a new classification of biomarkers. Pharm Res. 2005;22:1432–1437. doi: 10.1007/s11095-005-5882-3. [DOI] [PubMed] [Google Scholar]
  14. Danhof M, de Lange EC, Della Pasqua OE, Ploeger BA, Voskuyl RA. Mechanism-based pharmacokinetic-pharmacodynamic (PK-PD) modeling in translational drug research. Trends Pharmacol Sci. 2008;29:186–191. doi: 10.1016/j.tips.2008.01.007. [DOI] [PubMed] [Google Scholar]
  15. Della Pasqua O. Translational pharmacology: from animal to man and back. Drug Discov Today Technol. 2013;10:e315–e317. doi: 10.1016/j.ddtec.2013.03.001. [DOI] [PubMed] [Google Scholar]
  16. Dom N, Knapen D, Blust R. Assessment of aquatic experimental versus predicted and extrapolated chronic toxicity data of four structural analogues. Chemosphere. 2012;86:56–64. doi: 10.1016/j.chemosphere.2011.08.050. [DOI] [PubMed] [Google Scholar]
  17. Dorato MA, Engelhardt JA. The no-observed-adverse-effect-level in drug safety evaluations: use, issues, and definition(s) Regul Toxicol Pharmacol. 2005;42:265–274. doi: 10.1016/j.yrtph.2005.05.004. [DOI] [PubMed] [Google Scholar]
  18. Fornai M, Antonioli L, Colucci R, Pellegrini C, Giustarini G, Testai L, et al. NSAID-induced enteropathy: are the currently available selective COX-2 inhibitors all the Same? J Pharmacol Exp Ther. 2014;348:86–95. doi: 10.1124/jpet.113.207118. [DOI] [PubMed] [Google Scholar]
  19. Halter F, Tarnawski AS, Schmassmann A, Peskar BM. Cyclooxygenase 2 – implications on maintenance of gastric mucosal integrity and ulcer healing: controversial issues and perspectives. Gut. 2001;49:443–453. doi: 10.1136/gut.49.3.443. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Higuchi K, Umegaki E, Watanabe T, Yoda Y, Morita E, Murano M, et al. Present status and strategy of NSAIDs-induced small bowel injury. J Gastroenterol. 2009;44:879–888. doi: 10.1007/s00535-009-0102-2. [DOI] [PubMed] [Google Scholar]
  21. Hooker AC, Staatz CE, Karlsson MO. Conditional weighted residuals (CWRES): a model diagnostic for the FOCE method. Pharm Res. 2007;24:2187–2197. doi: 10.1007/s11095-007-9361-x. [DOI] [PubMed] [Google Scholar]
  22. Huntjens DR, Spalding DJ, Danhof M, Della Pasqua OE. Correlation between in vitro and in vivo concentration-effect relationships of naproxen in rats and healthy volunteers. Br J Pharmacol. 2006;148:396–404. doi: 10.1038/sj.bjp.0706737. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Huntjens DR, Spalding DJ, Danhof M, Della Pasqua OE. Impact of chronic inflammation on the pharmacokinetic-pharmacodynamic relationship of naproxen. Eur J Pain. 2010;14:227. doi: 10.1016/j.ejpain.2009.05.017. , e1–10. [DOI] [PubMed] [Google Scholar]
  24. Jackson LM, Wu KC, Mahida YR, Jenkins D, Hawkey CJ. Cyclooxygenase (COX) 1 and 2 in normal, inflamed, and ulcerated human gastric mucosa. Gut. 2000;47:762–770. doi: 10.1136/gut.47.6.762. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Josa M, Urizar JP, Rapado J, Dios-Viéitez C, Castañeda-Hernández G, Flores-Murrieta F, et al. Pharmacokinetic/pharmacodynamic modeling of antipyretic and anti-inflammatory effects of naproxen in the rat. J Pharmacol Exp Ther. 2001;297:198–205. [PubMed] [Google Scholar]
  26. Kargman S, Charleson S, Cartwright M, Frank M, Riendeau M, Mancini J, et al. Characterization of prostaglandin G/H synthase 1 and 2 in rat, dog, monkey, and human gastrointestinal tracts. Gastroenterology. 1996;111:445–454. doi: 10.1053/gast.1996.v111.pm8690211. [DOI] [PubMed] [Google Scholar]
  27. Kilkenny C, Browne W, Cuthill IC, Emerson M, Altman DG. Animal research: Reporting in vivo experiments: the ARRIVE guidelines. Br J Pharmacol. 2010;160:1577–1579. doi: 10.1111/j.1476-5381.2010.00872.x. : . [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Knight A. Animal experiments scrutinised: systematic reviews demonstrate poor human clinical and toxicological utility. ALTEX. 2007;24:320–325. doi: 10.14573/altex.2007.4.320. [DOI] [PubMed] [Google Scholar]
  29. Lanza FL, Royer GL, Jr, Nelson RS, Chen TT, Seckman CE, Rack MF. The effects of ibuprofen, indomethacin, aspirin, naproxen, and placebo on the gastric mucosa of normal volunteers. Dig Dis Sci. 1979;24:823–828. doi: 10.1007/BF01324896. [DOI] [PubMed] [Google Scholar]
  30. Lindbom L, Pihlgren P, Jonsson EN. PsN-Toolkit–a collection of computer intensive statistical methods for non-linear mixed effect modeling using NONMEM. Comput Methods Programs Biomed. 2005;79:241–257. doi: 10.1016/j.cmpb.2005.04.005. [DOI] [PubMed] [Google Scholar]
  31. Loftin CD, Tiano HF, Langenbach R. Phenotypes of the COX-deficient mice indicate physiological and pathophysiological roles for COX-1 and COX-2. Prostaglandins Other Lipid Mediat. 2002;68–69:177–185. doi: 10.1016/s0090-6980(02)00028-x. [DOI] [PubMed] [Google Scholar]
  32. Massó González EL, Patrignani P, Tacconelli S, García Rodríguez LA. Variability among nonsteroidal anti-inflammatory drugs in risk of upper gastrointestinal bleeding. Arthritis Rheum. 2010;62:1592–1601. doi: 10.1002/art.27412. [DOI] [PubMed] [Google Scholar]
  33. McGonigle P, Ruggeri B. Animal models of human disease: challenges in enabling translation. Biochem Pharmacol. 2014;87:162–171. doi: 10.1016/j.bcp.2013.08.006. [DOI] [PubMed] [Google Scholar]
  34. McGrath J, Drummond G, McLachlan E, Kilkenny C, Wainwright C. Guidelines for reporting experiments involving animals: the ARRIVE guidelines. Br J Pharmacol. 2010;160:1573–1576. doi: 10.1111/j.1476-5381.2010.00873.x. : . [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Morita I. Distinct functions of COX-1 and COX-2. Prostaglandins Other Lipid Mediat. 2002;68–69:165–175. doi: 10.1016/s0090-6980(02)00029-1. [DOI] [PubMed] [Google Scholar]
  36. Motsko S, Rascati K, Busti A, Wilson J, Barner J, Lawson K. Temporal relationship between use of NSAIDs, including selective COX-2 inhibitors, and cardiovascular risk. Drug Saf. 2006;29:621–632. doi: 10.2165/00002018-200629070-00007. [DOI] [PubMed] [Google Scholar]
  37. Nyman AM, Schirmer K, Ashauer R. Toxicokinetic-toxicodynamic modelling of survival of Gammarus pulex in multiple pulse exposures to propiconazole: model assumptions, calibration data requirements and predictive power. Ecotoxicology. 2012;21:1828–1840. doi: 10.1007/s10646-012-0917-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. O’Brien PJ. Cardiac troponin is the most effective translational safety biomarker for myocardial injury in cardiotoxicity. Toxicology. 2008;245:206–218. doi: 10.1016/j.tox.2007.12.006. [DOI] [PubMed] [Google Scholar]
  39. Pawson AJ, Sharman JL, Benson HE, Faccenda E, Alexander SP, Buneman OP, et al. NC-IUPHAR. The IUPHAR/BPS Guide to PHARMACOLOGY: an expert-driven knowledgebase of drug targets and their ligands. Nucl. Acids Res. 2014;42:D1098–1106. doi: 10.1093/nar/gkt1143. (Database Issue): [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. R Development Core Team. Vienna, Austria: 2014. R: a language and environment for statistical computing. R Foundation for Statistical Computing. Available at http://www.R-project.org/ (accessed 15/10/2014) [Google Scholar]
  41. Rostom A, Muir K, Dubé C, Jolicoeur E, Boucher M, Joyce J, et al. Gastrointestinal safety of cyclooxygenase-2 inhibitors: a Cochrane collaboration systematic review. Clin Gastroenterol Hepatol. 2007;5:818–828. doi: 10.1016/j.cgh.2007.03.011. [DOI] [PubMed] [Google Scholar]
  42. Sahota T, Sanderson I, Danhof M, Della Pasqua O. Model-based analysis of thromboxane B2 and prostaglandin E2 as biomarkers in the safety evaluation of naproxen. Toxicol Appl Pharmacol. 2014a;278:209–219. doi: 10.1016/j.taap.2014.03.010. [DOI] [PubMed] [Google Scholar]
  43. Sahota T, Danhof M, Della Pasqua O. 2014b. Utility of model based approaches to predict the risk of adverse events from preclinical toxicology protocols. In: Pharmacology-based toxicity assessment – towards quantitative risk prediction in humans. PhD Thesis, University of Leiden, 2014.
  44. Schneeweiss S, Solomon DH, Wang PS, Rassen J, Brookhart MA. Simultaneous assessment of short-term gastrointestinal benefits and cardiovascular risks of selective cyclooxygenase 2 inhibitors and nonselective nonsteroidal antiinflammatory drugs: an instrumental variable analysis. Arthrit Rheumat. 2006;54:3390–3398. doi: 10.1002/art.22219. [DOI] [PubMed] [Google Scholar]
  45. Solomon DH, Schneeweiss S, Glynn RJ, Kiyota Y, Levin R, Mogun H, et al. Relationship between selective cyclooxygenase-2 inhibitors and acute myocardial infarction in older adults. Circulation. 2004;109:2068–2073. doi: 10.1161/01.CIR.0000127578.21885.3E. [DOI] [PubMed] [Google Scholar]
  46. Takeuchi K. Pathogenesis of NSAID-induced gastric damage: importance of cyclooxygenase inhibition and gastric hypermotility. World J Gastroenterol. 2012;18:2147–2160. doi: 10.3748/wjg.v18.i18.2147. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Urushidani T, Okabe S, Takeuchi K, Takagi K. Strain differences in aspirin-induced gastric ulceration in rats. Jpn J Pharmacol. 1978;28:569–578. doi: 10.1254/jjp.28.569. [DOI] [PubMed] [Google Scholar]
  48. Wallace JL. Prostaglandins, NSAIDs, and gastric mucosal protection: why doesn’t the stomach digest itself? Physiol Rev. 2008;88:1547–1565. doi: 10.1152/physrev.00004.2008. [DOI] [PubMed] [Google Scholar]
  49. Whittle BJ. Mechanisms underlying intestinal injury induced by anti-inflammatory COX inhibitors. Eur J Pharmacol. 2004;500:427–439. doi: 10.1016/j.ejphar.2004.07.042. [DOI] [PubMed] [Google Scholar]
  50. Xie HG, Wang SK, Cao CC, Harpur E. Qualified kidney biomarkers and their potential significance in drug safety evaluation and prediction. Pharmacol Ther. 2013;137:100–107. doi: 10.1016/j.pharmthera.2012.09.004. [DOI] [PubMed] [Google Scholar]
  51. Zuideveld KP, Van der Graaf PH, Peletier LA, Danhof M. Allometric scaling of pharmacodynamic responses: application to 5-HT1A receptor mediated responses from rat to man. Pharm Res. 2007;24:2031–2039. doi: 10.1007/s11095-007-9336-y. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Table S1 Summary of naproxen population PK and PKPD model parameters (Sahota et al., 2014).

bph0172-3861-sd1.docx (13.3KB, docx)

Articles from British Journal of Pharmacology are provided here courtesy of The British Pharmacological Society

RESOURCES