Evolving data analysis of an Oral Lipid Tolerance Test toward the standard for the Oral Glucose Tolerance Test: Cross species modeling effects of AZD7687 on plasma triacylglycerol

Abstract We have developed a novel mechanistic pharmacokinetic‐pharmacodynamic (PK/PD) model to describe the time course of plasma triglyceride (TAG) after Oral Lipid Tolerance Test (OLTT) and the effects of AZD7687, an inhibitor of diacylglycerol acyltransferase 1 (DGAT1), in humans, rats, and mice. Pharmacokinetic and plasma TAG data were obtained both in animals and in two phase I OLTT studies. In the PK/PD model, the introduction of exogenous TAG is represented by a first order process. The endogenous production and removal of TAG from plasma are described with a turnover model. AZD7687 inhibits the contribution of exogenous TAG into circulation. One or two compartment models with first order absorption was used to describe the PK of AZD7687 for the different species. Nonlinear mixed effect modeling was used to fit the model to the data. The effects of AZD7687 on the plasma TAG time course during an OLTT as well as interindividual variability were well described by the model in all three species. Meal fat content or data from single vs repeated dosing did not affect model parameter estimates. Body mass index was found to be a significant covariate on the plasma TAG baseline. The system parameters of the model will facilitate analysis for other compounds and provide tools to bring the standard of OLTT data analysis closer to the analyses of Oral Glucose Tolerance Test data maximizing knowledge gain.

the Oral Protein Tolerance Test (OPTT). Phenotypic flexibility has been postulated as a measure of health, suggesting that health can be evaluated by the ability to adapt to conditions of temporary stress and therefore the use of a standardized nutritional challenge test that combines properties of all three previously mentioned challenges into a combined oral protein-glucose-lipid tolerance test (OPGLTT) is expected to demonstrate subtle improvements of phenotypic flexibility, thereby enabling substantiation of nutritional health effects. 1 In such a test, multi-biomarker panels of interlinked measurements are generated, to fully evaluate and understand the results of such a test, a dynamic and quantitative understanding of the interrelation of the different components will be required. Before such a multi-biomarker mathematical model can be developed, understanding of the dynamics of the key biomarkers of each of the three individual tests is required. The behavior of plasma glucose and insulin after an OGTT has been widely studied and nowadays can be successfully described mathematically. [2][3][4][5][6] However, the same level of data analysis and modeling is lacking for plasma triglyceride (TAG) data after an OLTT.
Elevated levels of TAG are a major component of the metabolic syndrome and are important risk factors in the development of atherosclerotic cardiovascular disease. 7 Measuring plasma TAG by means of an OLTT can determine the efficiency with which the individual uses lipid components. 8 The analysis of the experimental data generated in an OLTT setting both in human and in animal studies vary substantially, from comparison of area under the curve (AUC) and/or the incremental area under the curve (iAUC), 9 use of a 3point test, 10 a scoring system, 7 time-matched comparison of single or multiple time-points, 11 comparison of the TAG clearance constants between groups 12 and comparison of mean values during a defined time-interval. 13 These type of analyses have varied statistical validity to compare between different groups of one study 9 and they also provide a pharmacodynamic measurement that sometimes has been used to establish PK/PD relationships when a xenobiotic have been used to modify the response compared to a control treated group. 14 However, this analysis of the OLTT data does not provide sufficient granularity to allow for an equivalent level of knowledge gain from an OLTT study as the modeling of blood glucose and insulin data from an OGTT study delivers. [2][3][4][5][6] A more mechanistic PK/PD modeling approach to mathematically describe the time course of the TAG excursion will provide the backbone for more meaningful analysis of OLTT data and in turn allowing for a deeper understanding of the potential action of different drugs on the postprandial lipid profile.

| Rat
Full description of the OLTT in the rat has been published previously. 17 In brief, 53 male Han-Wistar rats (∼230 g), previously main-

| Mouse
Oral Lipid Tolerance Test in the mouse was carried out in a similar way to rats. In brief, 41 Male ICR mice (∼25 g) were dosed by oral gavage with either CMC vehicle (n = 8) or AZD7687 suspended in CMC at 0.1, 1, or 3 mg/kg (n = 11) per group and 0.5 hour later dosed with an oral gavage of fat emulsion containing 20% soybean oil (Intralipos 20%) at 10 mL/kg. N = 3 of the AZD7687-treated animals (per group) were serially bled for PK determination at 1, 2, 3, All studies involving animals are reported in accordance with the ARRIVE guidelines for reporting experiments involving animals. 18,19 2.2 | The PK/PD model Biological plausibility combined with stable parameter estimation was used in the evaluation of different potential PK/PD models to help select the optimal one. The key components of the final PK/PD TAG model are described in Figure 1. The introduction of exogenous TAG into the system is represented by a first order process from the lipid depot compartment (assumed to be the gut) to the central compartment (plasma). A lag time to accommodate the time delay from ingestion of the fatty meal to the resulting increase in plasma TAG was needed. In addition, there is an endogenous production and removal of TAG from plasma described with a turnover model controlled by the turnover rate (kin) (production) and the fractional turnover rate (kout) (loss). Turnover model was reparameterized using the relationship of kin to baseline plasma TAG (R0) (R0 = kin/kout) to increase numerical stability.
Ka TAG is the absorption rate of exogenous TAG into plasma, V TAG is the volume of distribution of exogenous TAG, and Tlag TAG is the time delay for the absorption of TAG from the SMM. The contribution of meal intake to the dynamics of TAG was incorporated into the model by considering the macronutrient composition of the meals. TAG from meals (clinical studies) in units of kcal was converted to grams of TAG input (9 kcal of fat is equivalent to 1 gr of TAG). The total energy content of the meal was 1100 kcal. The contribution of corn oil intake (rat studies) to the dynamics of TAG was incorporated into the model by assuming 100% of the corn oil was fat. Volume of oil administered was converted to grams applying a density value of 0.9 gr/mL. The contribution of soybean oil intake (mouse studies) to the dynamics of TAG was incorporated into the model by assuming 100% of the soybean oil was fat. Volume of soybean oil on the fat emulsion administered was calculated and converted to grams applying a density value of 0.917 gr/mL. It was assumed that all TAG administered was absorbed in the absence of AZD7687.
Based on the role of DGAT-1, it is postulated that AZD7687 inhibits the contribution of exogenous TAG to circulation. A nonlinear relationship (equation 1) provided a stable parameterization.
where C is the AZD7687 concentration in plasma and IC 50 is the concentration of compound that results in 50% of max inhibition.
The fixed structure of the final PK/PD model is shown in Figure 1 and its mathematical description on equations 2, 3, and 4.
Residual error was estimated for plasma TAG using a proportional error model.
C TAG represents the plasma TAG concentration. Additionally, a lag time (Tlag TAG ) to accommodate the time delay from ingestion of the fatty meal to the resulting increase in plasma TAG was applied to the A TAG equation.

| Pharmacokinetic models
Human PK was described with a 2-compartment model with first order absorption, body weight was found to be a significant covariate for individual estimates of clearance (Cl) and volume of distribution (V). In the rat and mouse, as PK data were only collected for 4-11 hours post dose, the plasma AZD7687 concentration data during the OLTT were characterized by 1-compartment PK model. Full description of all the PK models is provided in supplementary section S1.

| Modeling approach, implementation, and evaluation
A sequential approach of fitting the AZD7687 plasma concentrations first and then the plasma TAG secondly was used to select the structure of the models and get good initial estimates of the parameters. Subsequently, both AZD7687 plasma concentrations and plasma TAG levels were simultaneously fitted on the final analysis.
Models were implemented using nonlinear mixed effects modeling in Phoenix NLME, version 1.3 (Certara). The analysis was conducted using the first order conditional estimation (FOCE-ELS) during the sequential and simultaneous analysis (the expectation maximization method (QRPEM) was used for the simultaneous fitting of the human data due to speed and ability to converge).
Interindividual variability was evaluated on all PK/PD parameters with an assumed log-normal distribution of individual parameters.
In addition, the potential effects on AZD7687 PK and plasma TAG of subject characteristics, such as age, body weight, height, age, Phoenix NLME %CV reported was calculated using the Hessian method.
For the evaluation of the final models, 1000 data sets were simulated in Phoenix NLME. The median and the 95% prediction intervals of the individual concentration-time profiles of AZD7687 and TAG were superimposed on the respective observed data.

| AZD7687 plasma concentration
For all species, free plasma concentrations of AZD7687 were calculated based on measured concentrations of AZD7687 in plasma and corrected using a constant free fraction (fu) for each of the species.
To be able to do cross species comparisons, modeling was carried out using free concentrations of AZD7687, therefore, in this report the term concentration of AZD767 refers to free concentration and resulting PK and PK/PD parameters obtained to describe the PK and TAG profiles were obtained using free concentrations of AZD7687 in plasma.

| RESULTS
Plasma concentrations of AZD7687 in man were adequately described by a 2-compartment model with first order absorption while in the rat and mice were described by 1-compartment model with first order absorption. The pharmacokinetic parameters are shown in Table 1. The diagnostic plots and goodness of fit plots for the PK models are shown in supplementary section S1.
Effects of AZD7687 on the plasma TAG time course after SMM were well described by the PK/PD model proposed in all three T A B L E 1 Summary of the key PK and PK/PD parameters for AZD7687 in human, mouse, and rat (Full list of PK and PD parameters provided in Supplementary sections S1 and S2)

| Human data
The %fat of the different SMM tested did not significantly affect the model parameter estimates. No differences in the PK/PD relationship were observed between the single dose and multiple dose experiments. The resulting free in vivo IC 50 was estimated to be 0.008 μmol/L. Interindividual variability in the IC 50 was high (223%), additionally, interindividual variability was also estimated on ka TAG , V TAG , R0, and kout parameters according to a log-normal distribution of individual parameters (

| Rat data
The resulting free in vivo IC 50

| Mouse data
The resulting free in vivo IC 50 (0.081 μmol/L) showed a 62% interindividual variability which is substantially lower than for human.
Additionally, interindividual variability could also be estimated on Tlag TAG , ka TAG , and R0 parameters according to a log-normal distribution of individual parameters ( Table 1) These types of limited PK/PD analysis do not maximize the potential learnings that can be extracted from the studies. Application of more mechanistic PK/PD modeling approaches to data generated on the analogue experiment OGTT has enhanced the level of understanding of the relationship between disease and response in such a challenge-based study. 3,5,6 Additionally, the application of more mechanistic modeling approaches has also provided important insights into the mechanism of action of drugs. 2,4 Similar kind of gains ought to be achieved as well from more mechanistic modeling of data generated in OLTT.
There are models published that describe drug-induced changes in some plasma lipids in the absence of a lipid load [39][40][41][42][43][44] but to our knowledge none describe the plasma TAG excursion after an OLTT. In an OLTT setting, the "challenger" (fat load) can be considered the same as the pharmacodynamic response biomarker (plasma TAG) and therefore any successful mechanistic model will need to take into account the extra input of TAG simultaneously with the baseline data and the data during intervention of a TAG-lowering compound. Such approach has previously successfully been used to describe plasma glucose concentrations over time after an OGTT. 2-6 However, the above models had not been modified or applied to describe the plasma TAG time course following an OLTT challenge and therefore a gap still remained in the literature.
Clinical development of AZD7687, 15 a novel DGAT1 inhibitor has recently been stopped due to intolerable side effects. 16

DGAT1
is involved in the dietary absorption of TAG. [45][46][47] Denison et al 14,16 have published the results from an OLTT carried out with AZD7687 in humans after single and multiple doses of AZD7687. However, the PK/PD analysis of that data was based on iAUC of the TAG excursion and the compound exposure at the time of fat loading. In addition, it has been previously published that AZD7687 reduces the TAG excursion after an OLTT in rats in an exposure dependant manner. 15 Once again, the PK/PD analysis of that data was carried out  adipokine production, and lipoprotein production that are also triggered by an OLTT should be attempted.
In conclusion, this model provides not only a description of the effects of AZD7687 on plasma TAG in an OLTT setting across all MORENTIN GUTIERREZ ET AL.
| 9 of 11 three species but also the tools to start bringing the standard of OLTT data analysis closer to how OGTT data are analyzed maximizing the knowledge gain and the possibility to compare results between the tests. In the future, this model could contribute to the difficult task of building a cross-biomarker model that can help explain the multi-variable results obtained in OPGLTT.