Visual Overview
Abstract
The abnormal accumulation of amyloid-β (Aβ) in the brain parenchyma has been posited as a central event in the pathophysiology of Alzheimer’s disease. Recently, we have proposed a systems pharmacology model of the amyloid precursor protein (APP) pathway, describing the Aβ APP metabolite responses (Aβ40, Aβ42, sAPPα, and sAPPβ) to β-secretase 1 (BACE1) inhibition. In this investigation this model was challenged to describe Aβ dynamics following γ-secretase (GS) inhibition. This led an extended systems pharmacology model, with separate descriptions to characterize the sequential cleavage steps of APP by BACE1 and GS, to describe the differences in Aβ response to their respective inhibition. Following GS inhibition, a lower Aβ40 formation rate constant was observed, compared with BACE1 inhibition. Both BACE1 and GS inhibition were predicted to lower Aβ oligomer levels. Further model refinement and new data may be helpful to fully understand the difference in Aβ dynamics following BACE1 versus GS inhibition.
Introduction
The amyloid cascade hypothesis posits that the pathologic cascade leading to Alzheimer’s disease (AD) is triggered by abnormal accumulation of amyloid-β protein (Aβ) in the brain parenchyma (Karran et al., 2011). Inhibition of Aβ production in the brain is therefore a therapeutic target for treating AD with a potentially disease-modifying effect (Cole and Vassar, 2007; Husain et al., 2008).
Aβ is generated through sequential proteolytic cleavage of β-amyloid precursor protein (APP) by β-secretase (BACE1) and γ-secretase (GS) (Esler and Wolfe, 2001), as schematically depicted in Fig. 1. In the first cleavage step, the N-terminal secreted fragment soluble APPβ (sAPPβ) and the C-terminal membrane-bound 99-amino acid fragment (C99) are formed by BACE1. Subsequently, C99 is cleaved by GS yielding Aβ species of different chain lengths of which Aβ38, Aβ40, and Aβ42 are the most common isoforms. In an alternative pathway, cleavage of APP by α-secretase leads to the formation of soluble APPα (sAPPα) and the C-terminal membrane-bound 83-amino acid fragment. α-Secretase cleavage precludes Aβ formation. Recently, a new APP processing pathway was reported by Willem et al. (2015), in which sequential cleavage of APP by η-secretase and BACE1 or ADAM10 produces Aη-β and Aη-α, respectively. There may be other alternate processing pathways of APP that are unidentified to date.
These observations show that the accumulation of Aβ species is governed by a biochemical network, in which there are multiple enzymes that may serve as a target to modify the exposure to distinct Aβ peptide species. The network structure complicates the prediction of the effect of inhibitors of the various enzymes on the exposure to the various Aβ species. This may explain the mixed observations in some of the early clinical trials with enzyme inhibitors. Against this background we have recently proposed a systems pharmacology model to describe the effect of BACE1 inhibition on multiple Aβ species (van Maanen et al., 2016). Throughout the article the term recent model is used to refer to this model.
The recent systems pharmacology model of the APP processing pathway was developed to characterize APP metabolite responses to BACE1 inhibition by MBi-5 (van Maanen et al., 2016), which is distinct from MSD’s BACE1 inhibitor MK-8931 (MSD, Kenilworth, NJ). The model took into account the kinetics and interrelationships of sAPPβ, sAPPα, Aβ40, and Aβ42. In the model, sAPPβ was used as a surrogate substrate for C99 in the GS cleavage step, modulating the responses of Aβ40 and Aβ42 in the presence of the BACE1 inhibitor. A precursor APP pool, shared by sAPPα and sAPPβ, was included to describe the effect on all four biomarkers with a single drug effect. The effect of BACE1 inhibition was included in the model as inhibition of the pathway mediated by BACE1. Using this model, it was demonstrated that BACE1 inhibition resulted in a larger absolute reduction of cerebrospinal fluid (CSF) levels of Aβ40 than of Aβ42, since the effect on Aβ42 was modulated by back-conversion from an Aβ oligomer (AβO) pool.
There is growing evidence that AβOs have a central role in the pathogenesis of AD (Klein, 2013). Toxic AβOs are considered to be the drivers of neurodegeneration. AβOs might exist in a complex equilibrium with Aβ monomers and fibrils (Benilova et al., 2012). Treatments that prevent Aβ production may reduce the concentration of AβOs and subsequently promote the release of soluble Aβ from fibrils to restore the equilibrium (Rosenblum, 2014). Inhibitors of the two secretases that generate Aβ from APP, BACE1, and GS inhibitors have been proposed as potential disease-modifying approaches in the treatment of AD (Husain et al., 2008); BACE1 acts earlier in the cascade, affecting sAPPβ, sAPPα, Aβ40, and Aβ42. Sequentially, GS inhibition interferes later in the amyloidogenic APP pathway and affects Aβ40 and Aβ42 only.
The objective of the current investigation was to elucidate the APP processing pathway further, by challenging the recently developed systems pharmacology model of the APP pathway to describe Aβ dynamics following GS inhibition. The aims were 1) to separate BACE1 and GS sequential cleavage steps, 2) to investigate possible differences in Aβ response following GS versus BACE1 inhibition, and 3) to evaluate if moderation of Aβ42 by back-conversion from an AβO pool could also be identified after inhibiting GS. To this end, CSF Aβ40 and Aβ42 response data from two studies of the GS inhibitor 23-((1r,4s)-4-(4-chlorophenylsulfonyl)-4-(2,5-difluorophenyl)cyclohexyl)propanoic acid (MK-0752) in cisterna magna ported rhesus monkeys (Cook et al., 2010) were analyzed simultaneously with data from a BACE1 inhibitor (MBi-5) study using the APP systems model.
Materials and Methods
Animals.
All animal studies were reviewed and approved by the MSD Institutional Animal Care and Use Committee (https://www.msdresponsibility.com/access-to-health/research-development/animal-research/). The National Institutes of Health Guide to the Care and Use of Laboratory Animals (https://grants.nih.gov/grants/olaw/Guide-for-the-Care-and-use-of-laboratory-animals.pdf) and the Animal Welfare Act (https://www.aphis.usda.gov/animal_welfare/downloads/awa/awa.pdf) were followed in the conduct of the animal studies. The cisterna magna ported rhesus monkey model was reported by Gilberto et al. (2003). The rhesus monkeys were chronically implanted with catheters in the cisterna magna, facilitating repeated sampling of CSF and plasma (through a jugular vascular access point). These rhesus monkeys were individually housed and captive bred in a closed colony.
In the first GS inhibitor study (study 1), six male animals weighing between 6.9 and 9.6 kg (mean, 8.2 kg) were included, in which five animals were 5–8 years of age (mean, 6 years) and one animal was 17 years of age at the time of the study. The second GS inhibitor study (study 2) included six male animals weighing between 6.1 and 12.3 kg (mean, 9.1 kg), all of which were 7–10 years of age (mean, 8 years). In the BACE1 inhibitor study (study 3), six male animals weighing between 5.2 and 11.7 kg (mean, 8.7 kg), all of which were 2–10 years of age (mean, 8 years) were included. One-half of the animals in the BACE1 inhibitor study also participated in the GS inhibitor studies.
Drug Administration and Sampling.
The effects of secretase inhibition were obtained in the three studies. In study 1, information about the effect of GS inhibition on Aβ40 and Aβ42 was obtained following a single oral administration of MK-0752 at 60 and 240 mg/kg (5 ml/kg) in a vehicle-controlled (sterile water) three-period crossover study. In study 2, the effect of GS inhibition on Aβ40 and Aβ42 was measured during a follow-up collection period, following a single oral dose of MK-0752 at 240 mg/kg (5 ml/kg) in a vehicle-controlled (sterile water) study. In study 3, the effect of BACE1 inhibition on sAPPα, sAPPβ, Aβ40, and Aβ42 were measured, following single oral administration of MBi-5 at 10, 30, and 125 mg/kg (5 ml/kg), or vehicle (0.4% methylcellulose) in a four-way full, crossover study. The study protocols of studies 1 and 2 and the pharmacological profile of MK-0752 were previously described by Cook et al. (2010). The detailed study protocol of study 3 and the pharmacological profile of MBi-5 were described by Dobrowolska et al. (2014). The study protocols are summarized subsequently.
In study 1, plasma and CSF drug concentrations were collected at 0 hours predose and 3, 5, 7, 9, 13, 16, 19, 22, 25, 28, 31, 34, 37, 40, and 49 hours postdose, resulting in 16 plasma and CSF pharmacokinetics (PK) samples for each monkey per treatment. In study 2, plasma and CSF samples were collected as described for study 1 and additional samples were taken at 73, 145, 169, 217, and 241 hours postdose, resulting in 21 plasma and CSF PK samples for each monkey. Then, 2 ml of blood and 1 ml of CSF were collected at each time point. The concentration of MK-0752 in the plasma and CSF samples was determined using liquid chromatography–tandem mass spectrometry. The concentrations of Aβ40 and Aβ42 were determined from CSF samples collected at the same time points as the PK samples, giving 16 measurements of each biomarker for each monkey per treatment in study 1 and 21 measurements of each biomarker for each monkey in study 2. The Aβ1-40 and Aβ1-42 assays used for the concentration measurements were described previously by Cook et al. (2010).
In study 3, plasma and CSF drug concentrations were collected at 0 hours predose and 3, 5, 7, 9, 13, 16, 19, 22, 25, 28, 31, 49, 55, 58, 73, and 145 hours postdose, resulting in 17 plasma and CSF PK samples for each monkey per treatment group. Then, 2 ml of blood and 1 ml of CSF were collected at each time point. The concentration of MBi-5 in the plasma and CSF samples was determined using liquid chromatography–tandem mass spectrometry. The concentrations of Aβ40, Aβ42, sAPPα, and sAPPβ were determined from CSF samples collected at 22, 20, and 1 hours predose and 2, 4, 6, 8, 12, 15, 18, 21, 24, 27, 30, 48, 54, 57, 72, and 144 hours postdose, giving 19 measurements of each biomarker for each monkey per treatment group. Then, 1 ml of CSF was collected at each time point. The specific enzyme-linked immunosorbent assays used for the concentration measurements were described previously (Sankaranarayanan et al., 2009; Wu et al., 2011).
PK-Pharmacodynamics (PD) Analysis.
PK-PD modeling analysis was performed by means of nonlinear mixed effects modeling using the software package NONMEM, version 7.2.0 (Bauer, 2011). In this approach, structural (fixed) effects and both intra- and interindividual variability are taken into account. Typical values of the structural model parameters [population parameters, which define the average value for a parameter in a population (θ)], the variance and covariance of the interindividual variability (ω2), and the variance of the residual error (σ2) are estimated. The best models were chosen based on the minimum value of the objective function, the precision of the parameter estimates, and visual inspection of goodness-of-fit plots. A more detailed description of the modeling procedure was described in van Maanen et al. (2016). To evaluate the performance of the model, a visual predictive check was performed in which the median and the 90% interquantile range of the data simulated with the final parameter estimates were overlaid with the observations. The predictive capacity is considered sufficient when the median and 90% of the predictions line up within the fifth, 50th, and 95th percentiles of the observations. The NONMEM software package was implemented on an Intel QuadCore (Intel Core i7 CPU860, 2.80 GHz, 3.24 GB RAM) and Compaq Visual Fortran, version 6.6 (Compaq Computer Corporation, Houston, TX) was used as compiler. Data management and model assessment was done using the statistical software package S-PLUS for Windows, version 8.0 (Professional, Insightful Corp., Seattle, WA).
Model Description.
The systems model of the APP processing pathway was developed by sequential analysis of PK and PD data following administration of MBi-5 and MK-0752. The PK models of MBi-5 and MK-0752 were based on simultaneous analysis of plasma and CSF PK data of each compound. The results of the PK data analysis of MBi-5 have been reported elsewhere by van Maanen et al. (2016). The results of the PK data analysis of MK-0752 are reported in the Supplemental Material. The PK models adequately described the plasma and CSF concentration time profiles of MBi-5 and MK-0752, respectively, thus the models could serve as input for PD model analysis.
The interrelationships of APP metabolite responses to BACE1 inhibition were described recently using a comprehensive systems model of the APP processing pathway (van Maanen et al., 2016). To describe the effect of the GS inhibitor, the model had to be extended. The extended systems model of the APP processing pathway included a compartment for C99. The relation between Aβ and C99 was included in the model, representing the GS cleavage step, on which the drug effect of MK-0752 was implemented. In addition, sAPPβ was no longer used as the driver of Aβ response and a sAPPβ elimination path was incorporated into the model. The formation rate constant of C99 was set to be equal to the formation rate constant of sAPPβ (Rinβ), since sAPPβ and C99 are both products of the same BACE1 cleavage step.
The biomarker response profiles of MBi-5 and MK-0752 measured in CSF were adequately described by a model containing compartments for seven moieties: APP, sAPPβ, sAPPα, C99, Aβ40, Aβ42, and AβO (Fig. 2). The production of APP was assumed to be constant and described by a zero-order input rate constant, RinAPP. The production of the APP metabolites was assumed to be first order, i.e., dependent on its precursor concentration. The relationship between APP and its metabolites (sAPPβ, sAPPα, C99, Aβ40, and Aβ42) and AβO is described by eqs. 1–7:(1)(2)(3)(4)(5)(6)(7)where Routa is the sAPPα degradation rate and Routb is the sAPPβ degradation rate. The initial conditions for these differential equations are the baseline levels: APPbase, sAPPαbase, sAPPβbase, C99base, Aβ40base, Aβ42base, and AβObase, respectively.
The rate of change of APP with respect to time in the presence of the BACE1 inhibitor is expressed by eq. 1, in which the BACE1 cleavage inhibition is incorporated by the factor BACE1 cleavage inhibition factor (EFFB). The rate of change of C99 with respect to time in the presence of the GS inhibitor is described by eq. 4, in which the GS cleavage inhibition is incorporated by the factor GS cleavage inhibition factor (EFFG). EFFB and EFFG are the degrees of inhibition caused by MBi-5 and MK-0752, respectively. Generally, the degree of inhibition is described by a sigmoidal Imax function, as given by eq. 8:(8)where EFF is APP cleavage inhibition; Ctarget is the drug target site concentration of MBi-5 or MK-0752, respectively; IC50 is the Ctarget that results in 50% inhibition of BACE1 or GS; Imax is the maximum response; and the generalized additive model (GAM) is the Hill coefficient. The Ctarget value is derived from the respective PK models as follows:(9)where AUCCSF and AUCplasma are the areas under the CSF and plasma concentration-time curves, respectively; and Ctarget is assumed to be in steady state (SS) with Cplasma.
It is assumed that the system is in SS when no treatment is given (EFFB = 1, EFFG = 1). These SS conditions were used to derive part of the system parameters. From SS and eq. 1 it follows that the zero-order input rate constant of APP (RinAPP) is:(10)where APPbase is the baseline level of APP, assumed to be equal to the sum of the baseline levels of sAPPα and sAPPβ. All alternate pathways are represented by the terms for α-secretase.
Using SS conditions and eq. 2 the sAPPα formation rate constant (Rinα), equivalent to the α-secretase cleavage step, can be derived:(11)where sAPPαbase is the baseline level of sAPPα. The sAPPβ and C99 formation rate constant (Rinβ), equivalent to the BACE1 cleavage step, follows from SS conditions and eq. 3:(12)where sAPPβbase is the baseline level of sAPPβ.
From eq. 5 and SS, the Aβ degradation rate constant (Kout), is deduced:(13)where C99base is the baseline level of C99. It should be noted not that C99 is not an observed measure. From eq. 4 and SS the baseline level of C99 can be calculated:(14)Combining eqs. 13 and 14, the Aβ42 formation rate constant (Kin42), equivalent to a GS cleavage step, can be written as:(15)where Kpl and Krev are the Aβ42 oligomerization and dissociation rate constants, respectively, which are dependent on the baseline values of Aβ42 and the AβO pool (Aβ42base and AβObase, respectively) according to eq. 16:(16)The model structure includes four transit compartments (Fig. 2), one for each biomarker measured in CSF (sAPPα, sAPPβ, Aβ40, and Aβ42), to account for transport from the target site in the brain to CSF. These transit processes are described, in general, by eq. 17:(17)where Ktr is the transit rate constant for the particular APP metabolite xAx (sAPPα, sAPPβ, Aβ40, or Aβ42).
Results
Aβ Response to GS and BACE1 Inhibition Was Described by Separate Descriptive Models.
Initially, empirical PK-PD models were developed to quantify the exposure-response relationships for each CSF APP metabolite of the BACE1 inhibitor MBi-5 (Aβ40, Aβ42, sAPPα, and sAPPβ) and GS inhibitor MK-0752 (Aβ40 and Aβ42) in rhesus monkeys. For the BACE1 inhibitor MBi-5, the empirical PK-PD models for Aβ40, Aβ42, sAPPα, and sAPPβ were discussed recently in van Maanen et al. (2016). For MBi-5 and MK-0752, we now present the empirical PK-PD models for Aβ40 and Aβ42. The exposure-response relationship for each Aβ-inhibitor combination was described by a transit model with one or two compartments, with the drug effect modeled relative to or subtracted from baseline using an Imax function. Table 1 presents a summary overview of the results of these models. For each inhibitor, the empirical models identified similar drug effects for Aβ40 and Aβ42: for MBi-5 the identified potencies were Aβ40: 0.0254 μM [95% confidence interval (CI), 0.0246–0.0262] and Aβ42: 0.0455 μM (95% CI, 0.0351–0.0559) and for MK-0752 the identified potencies were Aβ40: 0.432 μM (95% CI, 0.300–0.564) and Aβ42: 0.567 μM (95% CI, 0.402–0.732).
The separate empirical models revealed potential challenges for the combined analysis. First, there were study differences in Aβ baselines: 1.5-fold higher for the Aβ40 baseline and 3.4-fold higher for the Aβ42 baseline in the GS inhibitor studies (study 1 and 2) compared with the BACE1 inhibitor study (study 3). Consequently, the ratio of Aβ42:Aβ40 was higher in the GS inhibitor studies: 0.078 for the GS inhibitor studies and 0.034 for the BACE1 inhibitor study, respectively. Second, the mean transit time through the compartments of the models was lower for Aβ42 after BACE1 inhibition than Aβ42 after GS inhibition. This indicated that the response of Aβ42 to BACE1 inhibitor will appear earlier in CSF than with GS inhibition. Sequentially, BACE1 inhibition interferes earlier in the amyloidogenic APP pathway. This suggested a temporal difference in relative response progression of Aβ42 following BACE1 versus GS inhibition. For Aβ40, the mean transit time was higher after BACE1 inhibition than after GS inhibition; however, overlapping confidence intervals suggest insignificant differences.
A Systems Model to Describe APP Metabolite Responses to GS and BACE1 Inhibition.
Recently, we reported a systems pharmacology model, incorporating the pharmacokinetics of MBi-5 and APP metabolite (Aβ40, Aβ42, sAPPβ, and sAPPα) concentrations (van Maanen et al., 2016). In the current analysis, the model was extended to describe the dynamics of Aβ responses after exposure to a GS inhibitor. To this end, the APP metabolite responses of Aβ40, Aβ42, sAPPβ, and sAPPα following BACE1 inhibition and Aβ40 and Aβ42 response following GS inhibition were analyzed simultaneously.
To closer match the APP processing pathway, a C99 compartment was added to the model structure. Since sAPPβ and C99 are both products of the same BACE1 cleavage step, the formation rate constant of C99 was set to be equal to the formation rate constant of sAPPβ (Rinβ). The effect of BACE1 inhibition was incorporated in the model as inhibition of Rinβ. The effect of GS inhibition was modeled as inhibition of the Aβ40 and Aβ42 formation rate out of the C99 compartment, consistent with the GS cleavage step. The elimination of sAPPβ (Routb) could now be described as a separate parameter.
Inclusion of a C99 compartment and the estimated GS inhibition rates implied rebound of Aβ40 and Aβ42 response after GS inhibition; simulations indicated an excessive response above baseline upon cessation of GS inhibition (data not shown). However, the data did not suggest any significant rebound. Therefore, the model was refined by adding an alternative elimination pathway of C99 (Kout99) and hence preventing rebound. Kout99 reflects the production of other Aβ species than Aβ40 and Aβ42 through the GS cleavage of C99, as well as an alternative pathway of clearance independent of GS. The resulting model structure is presented in Fig. 2.
Interstudy baseline differences were evaluated by adding baseline data from two other studies (studies A and B) (see the Supplemental Material). From this, it became apparent that a correction for Aβ baseline differences between studies needed to be included. The underlying biologic system was assumed to be the same during all studies. Therefore, scaling factors were included in the model predictions (individual predicted values, or IPRED) for Aβ40 and Aβ42 outside of the system. To improve the model description further, differences in parameter values following BACE1 or GS inhibition were investigated. The formation rate constant of Aβ40 (Kin40) was fixed to the value identified recently (0.574 hour−1) following BACE1 inhibition (van Maanen et al., 2016) and a significantly lower Kin40 value was identified after GS inhibition (0.349 hour−1 (95% CI: 0.296–0.402); also, a substantial reduction of the Aβ42 oligomerization rate constant (Kpl) after GS inhibition was found (95% reduction). Including these differences improved the description of all of the biomarkers. Overall, the data were adequately captured across studies (Figs. 3–6). Only a slight underprediction was observed for the baseline level of Aβ40 in study 1 (Fig. 3A) and the maximal Aβ42 response to 240 mg/kg MK-0752 (Fig. 5, B and C) and 125 mg/kg MBi-5 (Fig. 5F). Additional diagnostic plots are provided in the Supplemental Material.
The Model Separated Drug-Specific and System-Specific Parameters.
The population parameters and intra- and interanimal variability were optimized for all study populations simultaneously and are reported in Table 2. Interanimal variability was included as exponential in nature, reflecting lognormal distributions of the individual model parameters, for the baseline of sAPPβ and the IC50 values of MBi-5 and MK-0752. Since the baselines of other APP metabolites were modeled as a function of the baseline of sAPPβ, the interanimal variability of sAPPβ was propagated in these biomarkers. Residual variability was included for each APP metabolite (sAPPβ, sAPPα, Aβ40, and Aβ42), as proportional error models, assuming a normal distribution. Drug-specific parameters (IC50, GAM, and Imax) could be distinguished from system-specific parameters (Kin40, Kout99, Routa Routb, KtrAP, Kpl, and Krev). The correlations between parameters were all below 0.95.
The transit rate constant from brain-to-CSF for Aβ40 and Aβ42 was assumed to be equal and fast. Since the transit rate for sAPPβ and sAPPα can only be estimated relative to the transit rate of Aβ, the latter was fixed to an arbitrary high value (10 hour−1). The transit rate constant for sAPPβ and sAPPα was estimated to be 0.0847 hour−1, which should be interpreted relative to the Aβ transit rate constant. Correction factors on Aβ42 and Aβ40 for study differences in levels compared with study 3 were 3.7 (95% CI, 3.40–4.00) and 1.37 (95% CI, 1.26–1.48), which were imputed as multipliers on the respective IPREDs of Aβ42 and Aβ40. The IC50 value of MBi-5 was estimated to be 0.0185 μM (95% CI, 0.0149–0.0221); the IC50 value of MK-0752 was 0.445 μM (95% CI, 0.337–0.553). The Hill coefficients for both compounds slightly deviated from unity: MBi-5, 1.49 (95% CI, 1.35–1.63), and MK-0752, 1.73 (95% CI, 1.54–1.92).
Differences in APP Metabolite Interrelationships Following BACE1 and GS Inhibition.
The formation rate constant of Aβ42 (Kin42) was calculated according to eq. 15: 0.0186 and 0.0113 hour−1 in the BACE1 and GS inhibition studies, respectively. The formation rate constant of Aβ40 Kin40 was higher than Kin42 (0.574 and 0.349 hour−1, in the BACE1 and GS inhibition studies, respectively). This is in agreement with the previously reported ratio of Aβ42 and Aβ40 of about 1:10 in non-Alzheimer brain (Iwatsubo et al., 1994).
The resulting model was used to visualize the interrelationships of the biomarkers following BACE1 and GS inhibition, respectively. Also, the behavior of APP, C99, and AβO was predicted. The relationships of the biomarker responses to BACE1 inhibition were recently discussed in van Maanen et al. (2016). The differentiation in biomarker response to inhibition of BACE1 and GS was as follows: APP increases after BACE1 but not after GS inhibition (Fig. 7, A and B, respectively); C99 decreases following BACE1 inhibition and slightly increases following GS inhibition; and both BACE1 and GS inhibition are predicted to decrease AβO levels, implying that the formation of AβO is reduced by decreased levels of monomeric Aβ42 and that AβO is in dynamic equilibrium with monomeric Aβ42.
The simulated concentration of AβO should be interpreted as the level if Aβ42 monomers are tied in the “oligomer soup” in the brain. The AβO pool was modeled as a pool in equilibrium with monomeric Aβ42 without adaptation for the number of subunits in multimeric species contained in the AβO pool. Therefore, the simulated difference in AβO concentration shown in Fig. 7, A and B reflects a difference in the number of monomers incorporated in the AβO pool.
Discussion
In this analysis, a recently reported systems model of the APP processing pathway was extended to describe the interrelationships of Aβ40, Aβ42, sAPPβ, and sAPPα upon inhibition of BACE1 with MBi-5 and Aβ40 and Aβ42 upon inhibition of GS with MK-0752 simultaneously. BACE1 acts earlier in the cascade, affecting all four biomarkers, and sequentially GS inhibition interferes later in the amyloidogenic APP pathway and affects Aβ40 and Aβ42 only. Here, the combined model-based analysis of BACE1 and GS inhibitor response data facilitated the separation of the GS cleavage step from other processes involved.
To that end, the extended systems model of the APP processing pathway included a compartment for C99, wherein the relationship between C99 and Aβ represents the GS cleavage step. The MK-0752 concentration-dependent decrease of Aβ40 and Aβ42 could be described by incorporating a single drug effect in the model: inhibition of the formation rates of Aβ40 and Aβ42 out of the C99 compartment, equivalent to the GS cleavage step. The effect of MBi-5 was incorporated in the model as inhibition of the formation of sAPPβ and C99 (Rinβ), corresponding to the BACE1 cleavage step.
The identified IC50 values of MK-0752 in the empirical models were consistent with the single potency identified using the systems model. Aβ is the product of GS cleavage of C99, and therefore Aβ response measurement following GS inhibition provides a direct reflection of the drug action. Hence, the IC50 values based on empirical versus the mechanistic systems model were similar for MK-0752. The estimated IC50 value of MK-0752 (0.445 μM) was also similar to the brain IC50 value of 440 nM in guinea pigs (Cook et al., 2010).
For MBi-5, the identified IC50 values in the empirical models were significantly higher than the single potency of 0.0185 μM identified using the systems model. Since MBi-5 interferes earlier in the amyloidogenic APP pathway, inhibiting the formation of the Aβ precursor C99, the Aβ response does not directly reflect the drug action. A systems model that includes key processes such as the production, elimination, and brain-to-CSF transport for the APP metabolites can more accurately describe the IC50 value than an empirical model.
The identified IC50 value of MBi-5 in the current analysis was lower (0.0185 μM; 95% CI, 0.0149–0.0221) than recently identified (0.0269 μM; 95% CI, 0.0154–0.0384) (van Maanen et al., 2016); however, confidence intervals overlap. The more complete systems model presented here explains more fully the processes occurring in the APP pathway and therefore provides a more accurate characterization of the IC50 value. By analyzing the effects of two compounds with differing methods of action—i.e., GS and BACE1 inhibitors acting on different sequences in the APP processing pathway—the APP processing pathway could be further elucidated.
First, it was possible to identify the brain turnover of sAPPβ as a separate parameter. Thus, the brain turnover of sAPPβ [ln(2)/Routb = 0.39 hours] could be distinguished from the half-life of the brain-to-CSF transfer [ln(2)/KtrAP = 8.2 hours]. It should be noted that with only information on the concentrations in CSF, it is not possible to separate any delays related to transport of drug to the brain from delays due to transport of protein from brain to the cisterna magna (the site of measurement). The transport compartments and value of the transit rates reflect both processes.
Second, a higher elimination rate for C99 than for sAPPβ was identified. The elimination rate of C99 reflects the production of other Aβ species as well as an alternative pathway of clearance that is independent of GS. The elimination rates of sAPPβ and sAPPα had similar values. Both are soluble fragments of APP, with overlapping sequence only differentiating in the 16 C-terminal amino acids. The functions that are related to the shared domains of sAPPα and sAPPβ are identical. Therefore, it is not unlikely that sAPPα and sAPPβ also have similar elimination pathways.
Third, a divergence in Aβ dynamics following BACE1 versus GS inhibition was found: a significant difference in Aβ40 formation rate constant was identified. This may reflect that the implementation of the GS cleavage step in the model is a simplification of the underlying system. Matsumura et al. (2014) reported multiple interactive pathways for stepwise successive processing of C99 by GS, which were hypothesized to define the Aβ isoforms and quantity of each Aβ. If Aβ40 and Aβ42 are indeed the products of consecutive GS cleavage, this may be reflected in the identified divergence in Aβ dynamics following GS inhibition versus BACE1 inhibition. An alternative explanation of the differentiation in Aβ dynamics following BACE1 and GS inhibition may be that a feedback mechanism (unknown) was activated.
Fourth, the systems analysis suggests a difference in oligomerization of Aβ42 after GS and BACE1. However, the maximal Aβ42 responses to the higher-dose groups of MBi-5 and MK-0752 were not adequately captured. Therefore, this should be interpreted with caution. The interpretation of the AβO pool was recently discussed in van Maanen et al. (2016). In the recent model, inclusion of an AβO pool in the model could account for the differential effect of MBi-5 on Aβ40 and Aβ42 response observed in the data. In the Supplemental Material, the observed change in the ratio of Aβ42:Aβ40 after GS and BACE1 inhibition is presented. After GS inhibition, there is less difference in response of Aβ40 and Aβ42 observed, and thus less change in the ratio Aβ42:Aβ40. This may be caused by differential activation of some feedback mechanism on Aβ production or oligomerization, or model simplification of successive GS cleavage, as discussed previously. Once quantitative data of AβO response following BACE1 and GS are added, a difference in oligomerization may be confirmed.
There were large variations in biomarker measurements between studies. The observed study variations in Aβ42 and Aβ40 levels could result from factors related to analytical procedures, such as differences in laboratory procedures among centers and technicians, sample handling, or sample storage. To account for this, correction factors on Aβ42 and Aβ40 for study differences in studies 1 and 2 compared with study 3 were implemented in the model on IPRED. The biologic system was assumed to be the same in all studies. When incorporating correction factors on IPRED, these are outside the system; therefore, these are assumed to not affect the system parameters and model-derived differences following GS versus BACE1 inhibition.
Therapeutic strategies aimed at reducing the Aβ burden have the potential for eliciting a disease-modifying effect. Lower Aβ levels can be achieved by decreasing Aβ production. Inhibitors of the two secretases that generate Aβ from APP, BACE1, and GS inhibitors are and have been pursued in the pharmaceutical industry. The BACE1 inhibitor E2609 (Eisai, Tokyo, Japan) is currently in phase III clinical development and AZD 3293 (Eli Lilly [Indianapolis, IN] and AstraZenica [Cambridge, UK]) recently progressed to phase III. Recently, a phase II/III trial of MK-8931 (MSD) was halted after an interim analysis showed a lack of efficacy in patients with mild-to-moderate AD, but the drug has advanced to phase III with a prodromal AD study population. The GS inhibitor MK-0752 was progressed to phase I, but then discontinued due to tolerability issues. The GS inhibitor avagecestat (BMS-708163) was discontinued after phase II because of lack of efficacy and adverse effects of the gastrointestinal and dermatological system (Crump et al., 2012). The development of the GS inhibitors begacestat (GSI953) (Martone et al., 2009) and semagacestat (LY450139) (Doody et al., 2013) was also stopped. To date, no GS inhibitor demonstrated therapeutic success in AD patients.
A better understanding of the drug-induced modulation of the APP system after GS inhibition may be obtained through a quantitative comprehension of its concentration-response relationships. Several studies on the PK and PD of GS inhibitors have been reported. Das et al. (2011) reported a two-compartment model describing Aβ response to GS inhibition as observed in plasma and CSF in rhesus monkeys. Their model postulates an inhibitory mechanism of Aβ clearance by GS inhibition. However, in their model aspects of the Aβ production, transport, and clearance processes were simplified.
A model-based meta-analysis of published and in-house (pre-)clinical GS inhibitors data was performed by Niva et al. (2013). The production and clearance of Aβ was described with a turnover model, with a drug effect on the production rate. Tai et al. (2012) also used turnover models to describe Aβ levels following GS inhibition in brain, CSF, and plasma in the wild-type rat. They proposed a quasi-static Aβ pool in the brain, which does not change after short drug exposure. The aforementioned approaches look solely at the behavior of Aβ and not at the behavior of the APP system as a whole. The understanding of the APP system is imperative to improve the prediction of drug effects on Aβ levels.
System biology–based approaches have been used to try to explain the APP pathway. Mizuno et al. (2012) reported the AlzPathway, a comprehensive map of intra-, inter-, and extracellular signaling pathways in AD. In this AD-signaling network, one of the central molecules was Aβ. These types of models are used to explore and identify drug targets, but are not concerned with pharmacology and the general principles of PK-PD modeling.
Systems pharmacology models integrate the best available understanding of the biology and pharmacology of the system responses. The model facilitates the integration of prior knowledge of biologic systems and assumptions about the pathology and pharmacology with emerging data. Subsequent model refinement can then be helpful in elucidating system behavior and identifying knowledge gaps and further experiments. By challenging the model, we can learn something: If the existing model does not capture the data, we need to ask why and try to understand what is going on, i.e., were hypotheses underlying the model incorrect or incomplete.
When planning a new study, a crossover study design, in which each rhesus monkey receives MBi-5 and MK-0752, should be considered. This design will facilitate an adequate separation of study differences and differences in system responses following BACE1 or GS inhibition since the first is canceled out. Also, if sAPPα and sAPPβ response to GS inhibition would additionally be measured, it is anticipated that a possible feedback mechanism in the APP pathway can be evaluated.
Conclusions
The development of a systems pharmacology model is an evolutionary process, integrating knowledge of the biologic system with emerging data. Using the sAPPβ pool as the moderator of Aβ in the recent model was a simplification of the underlying biologic system. Here, the systems model structure more closely resembles the underlying APP pathway, and the incorporation of the data following GS inhibition was essential for this.
The model characterized the response and interrelationships of the APP metabolites and gave insight into the biologic mechanisms of the system. The application of such a mechanistic approach that separates drug-specific and systems-specific parameters provides a robust characterization of the inhibitors. A differentiation in Aβ dynamics after BACE1 versus GS inhibition was found, which was reflected in a difference in Aβ40 formation. As such, the systems pharmacology analysis also points to parts of the APP system that require further investigation to fully understand the interference of secretase inhibitors on the system.
Acknowledgments
We thank Randall J. Bateman and Justyna A. Dobrowolska (Department of Neurology, Washington University, St. Louis, MO) for contributions to the study designs.
Authorship Contributions
Participated in research design: Savage, Michener, Kennedy, Stone, Kleijn.
Conducted experiments: Michener.
Performed data analysis: van Maanen, Ahsman, van Steeg, Stone.
Wrote or contributed to the writing of the manuscript: van Maanen, van Steeg, Ahsman, Danhof, Stone.
Footnotes
- Received August 19, 2017.
- Accepted February 5, 2018.
↵This article has supplemental material available at jpet.aspetjournals.org.
Abbreviations
- Aβ
- amyloid-β
- AβO
- amyloid-β oligomer
- AD
- Alzheimer’s disease
- APP
- amyloid precursor protein
- AUC
- area under the curve
- BACE1
- β-secretase
- CI
- confidence interval
- CSF
- cerebrospinal fluid
- Ctarget
- drug concentration target site
- C99
- C-terminal membrane-bound 99-amino acid fragment
- EFFB
- inhibition of BACE-1 cleavage
- EFFG
- inhibition of GS cleavage
- GAM
- generalized additive model
- GS
- γ-secretase
- IPRED
- individual predicted value
- Kin40
- amyloid-β40 formation rate constant
- Kin42
- amyloid-β42 formation rate constant
- Kout
- amyloid-β degradation rate constant
- Kpl
- oligomerization rate constant
- Krev
- amyloid-β oligomer dissociation rate constant
- MK-0752
- 23-((1r,4s)-4-(4-chlorophenylsulfonyl)-4-(2,5-difluorophenyl)cyclohexyl)propanoic acid
- PD
- pharmacodynamics
- PK
- pharmacokinetic
- RinAPP
- zero-order input constant for amyloid precursor protein
- Rinα
- soluble amyloid precursor protein α formation rate constant
- Rinβ
- soluble amyloid precursor protein β formation rate constant
- Routa
- soluble amyloid precursor protein α degradation
- Routb
- soluble amyloid precursor protein β degradation rate constant
- sAPPα
- soluble amyloid precursor protein α
- sAPPβ
- soluble amyloid precursor protein β
- SS
- steady state
- Copyright © 2018 by The American Society for Pharmacology and Experimental Therapeutics