Alzheimer's disease (AD) poses a serious public health threat to the United States. Disease-modifying drugs slowing AD progression are in urgent need, but they are still unavailable. According to the amyloid cascade hypothesis, inhibition of β- or γ-secretase, key enzymes for the production of amyloid β (Aβ), may be viable mechanisms for the treatment of AD. For the discovery of γ-secretase inhibitors (GSIs), the APP-overexpressing Tg2576 mouse has been the preclinical model of choice, in part because of the ease of detection of Aβ species in its brain, plasma, and cerebrospinal fluid (CSF). Some biological observations and practical considerations, however, argue against the use of the Tg2576 mouse. We reasoned that an animal model would be suitable for GSI discovery if the pharmacokinetic (PK)/pharmacodynamic (PD) relationship of a compound for Aβ lowering in this model is predictive of that in human. In this study, we assessed whether the background 129/SVE strain is a suitable preclinical pharmacology model for identifying new GSIs by evaluating the translatability of the intrinsic PK/PD relationships for brain and CSF Aβ across the Tg2576 and 129/SVE mouse and human. Using semimechanistically based PK/PD modeling, our analyses indicated that the intrinsic PK/PD relationship for brain Aβx-42 and CSF Aβx-40 in the 129/SVE mouse is indicative of that for human CSF Aβ. This result, in conjunction with practical considerations, strongly suggests that the 129/SVE mouse is a suitable model for GSI discovery. Concurrently, the necessity and utilities of PK/PD modeling for rational interpretation of Aβ data are established.
Alzheimer's disease (AD) poses a serious public health threat to the United States. The patient population in the United States is estimated to be 5.4 million in 2011, with an annual socioeconomic cost of several hundred billion dollars (Alzheimer's Association et al., 2011). Disease-modifying drugs slowing AD progression are in urgent need, but they are still unavailable. One of the main hypotheses of AD etiology is the amyloid cascade hypothesis, which, based on multiple lines of strong evidence (Hardy and Higgins, 1992; Iversen et al., 1995; Shastry, 1998; Luo et al., 2001; Li et al., 2007; Karlnoski et al., 2009), postulates that AD is caused by abnormal accumulation and deposition of amyloid β (Aβ) in the brain. Aβ is generated through cleavage of the amyloid precursor protein (APP) first by β-secretase and then by γ-secretase. Inhibition of γ-secretase has been shown to decrease Aβ generation in vitro and in vivo; therefore, it has been proposed to be a potential disease-modifying mechanism for the treatment of AD. Programs seeking potent and selective γ-secretase inhibitors (GSIs) have been pursued in the pharmaceutical industry and reviewed periodically (Kreft et al., 2009; Imbimbo et al., 2011).
Discovery of GSIs entails the use of animal models, one of which is the transgenic Tg2576 mouse (Hsiao et al., 1996). Tg2576 mice overexpress the human APP695 isoform harboring the Swedish mutation (K670N/M671L) at the β-cleavage site, resulting in high susceptibility to β-secretase cleavage. Compared with their background 129/SVE strain, Tg2576 mice demonstrate drastically elevated levels of Aβ40 and Aβ42, accelerated Aβ plaque deposition in the brain, behavioral deficits, and impairments in learning and memory, resembling some pathological and symptomatic characteristics of AD (Hsiao et al., 1996; Chapman et al., 1999; Kawarabayashi et al., 2001; Sasaki et al., 2002; Middei et al., 2006). Tg2576 mice were thought to be an appropriate tool for researching AD and identifying therapeutic agents for AD. Tg2576 mice have been widely used to characterize GSI efficacy in Aβ lowering and behavioral, cognitive, and functional improvement (Lanz et al., 2003, 2004; Best et al., 2007; Prasad et al., 2007; Martone et al., 2009).
Some factors, however, argue against the utility of Tg2576 mice for identifying AD therapeutic agents. First, in Tg2576 mice and neuronal cultures containing the Swedish mutation, the β-cleavage kinetics and compartmentalization of APP processing are probably different from those in the background systems (Hook et al., 2008; Yamakawa et al., 2010). Potency shift of β-secretase inhibitors and GSIs in the Tg2576 and background systems have been observed (Burton et al., 2008; Yamakawa et al., 2010). These observations imply that the pharmacokinetic(PK)/pharmacodynamic (PD) relationship of a therapeutic agent obtained from experiments using Tg2576 mice may not translate to nontransgenic animals and humans.
Second, there is a tremendous cost incurred by using Tg2576 mice to screen and characterize Aβ-lowering compounds. Because of husbandry and genotyping considerations, it is much more expensive to use Tg2576 mice than nontransgenic mice.
The advent of a rodent Aβ assay with high sensitivity makes it feasible to detect inhibitory effects of GSIs in nontransgenic mice. Therefore, nontransgenic mice are currently in use for testing the GSI efficacy of Aβ lowering (Basi et al., 2010).
Throughout the progression of our GSI program, a question prevailed as to what preclinical model is suitable for compound screening and selection. Because efficacy of Aβ lowering in preclinical models is the predominant criterion for advancing a GSI to clinical development, we reasoned that an animal model would be suitable if the PK/PD relationship of a compound for Aβ lowering in this model is predictive of that in human, and that with everything else being equal an animal model requiring less capital commitment is more favorable. We propose here that 129/SVE mice can serve as a suitable model for screening and profiling GSIs. This is based on our quantitative analyses of the PK/PD relationship of several GSIs in Tg2576 and 129/SVE mice and human. We have found that: 1) 129/SVE mice are generally no less sensitive than Tg2576 mice to GSI treatment in terms of brain Aβ lowering; 2) in 129/SVE mice, the intrinsic PK/PD relationship of CSF Aβ40 tracks with that of brain Aβ42; and 3) in human, the intrinsic PK/PD relationship of CSF Aβ is comparable with that in the CSF of 129/SVE mice. As a result, 129/SVE mouse brain Aβ lowering is indicative of approximately the same potential effect in human CSF, suggesting the utility of 129/SVE mice for the discovery of GSIs.
Materials and Methods
Four structurally diverse GSIs, begacestat (GSI-953), semagacestat (LY450139), (2R)-2-[N-[(4-chlorophenyl)sulfonyl]-N-[2-fluoro-4-(1,2,4-oxadiazol-3-yl)benzyl]amino]-5,5,5-trifluoropentanamide (BMS-708163) (Gillman et al., 2010), and PF-6239574 [invented by Elan Pharmaceuticals, Dublin, Ireland, referred to as Elan-X hereafter (Truong et al., 2009)] (structures shown in Supplemental Fig. S1) used in this study were synthesized in-house. Their basic properties are summarized in Table 1.
Determination of Compounds' In Vitro Potency Against Aβ42.
Test compounds were solubilized in 100% dimethyl sulfoxide (DMSO) and serially diluted in 3.162-fold decrements, resulting in 11 concentrations ranging from 5 mM to 50 nM.
Chinese hamster ovary wild-type APP695 cells were seeded at 22,000 cells/100 μl well in 96-well tissue culture plates in cell growth media and incubated for 24 h at 37°C. Cells were washed once with cell growth media, and then 100 μl of fresh media was added before dosing. An equivalent volume (1 μl) of either DMSO or positive control compound was first added to the control wells manually to obtain minimum or maximum inhibition values, respectively, for the assay signal window. One microliter of each serially diluted test compound was added to the cell plate, resulting in a 100-fold dilution and a 1% final DMSO concentration. The plates were then incubated for approximately 24 h at 37°C. This procedure produced conditioned media in each well, which was tested for Aβ42 levels by ELISA.
Coating of ELISA plates was initiated by the addition of 50 μl/well of 10G3 Aβ42-specific antibody at 3 μg/ml in 0.1 M sodium bicarbonate, pH 9.0, into black 384-well Maxisorp plates (Thermo Fisher Scientific, Waltham, MA) with overnight incubation at 4°C. The 10G3 Aβ42-specific capture antibody was then flicked from the ELISA plates, and the plates were washed four times with 100 μl of ELISA wash buffer using the Multidrop Combi Reagent Dispenser (Thermo Fisher Scientific). Then, 90 μl/well of ELISA blocking buffer was added to the plates by using the multidrop dispenser. Ambient temperature incubation was allowed for a minimum of 1 h. Blocking buffer was then removed, and 20 μl of assay buffer was added to the wells. At that point, 40 μl (in duplicate) of experimental conditioned media (described above) was transferred into the wells of the blocked ELISA plates containing the 10G3 capture antibody, followed by overnight incubation at 4°C.
The ELISA plates were washed four times with 100 μl of ELISA wash buffer to remove unbound Aβ peptides. Europium (Eu)-labeled anti-Aβ1-16 (6E10) monoclonal antibody was diluted with ELISA buffer (50 μl/well of Eu-6E10 diluted by 1:10,000 and 20 μM EDTA). Incubation at ambient temperature for a minimum of 2 h was followed by four washes with 100 μl of ELISA wash buffer. DELFIA Enhancement Solution (30 μl/well; PerkinElmer Life and Analytical Sciences, Waltham, MA) was added, and plates were incubated at ambient temperature for 1 h. The plates were read on an EnVision plate reader (PerkinElmer Life and Analytical Sciences) using standard DELFIA time-resolved fluorescence settings.
Maximum percentages of inhibition values were determined with a control GSI compound, and minimum percentages of inhibition were determined in the absence of compound. IC50 was determined with nonlinear regression fit analysis by using in-house software.
PK/PD Data Collection.
All preclinical and clinical PK/PD data involved in this study are listed in Table 2. Preclinical PK/PD data were collected from Tg2576 and 129/SVE mice in-house as detailed below. Clinical PK/PD data were obtained from the public domain. The LY450139 CSF PK/PD data in healthy volunteers were digitized from figures 2A and 5A in Bateman et al. (2009); the placebo-normalized time courses of CSF Aβ1-40, Aβ1-42, and Aβ1-x are presented in Supplemental Fig. S2. The BMS-708163 plasma PK and CSF Aβ1-40 and Aβ1-42 time courses in healthy volunteers were captured from presentations at assorted conferences (Leil et al., 2010; Tong et al., 2010; Meredith et al., 2011).
129/SVE mice from Taconic Farms (Germantown, NY) and Tg2576 mice from Charles River Laboratories, Inc. (Wilmington, MA), housed four mice per cage, were acclimated for 1 week before dosing.
Animal Treatment and Sample Collection.
All animal studies followed the Institutional Animal Care and Use Committee-approved protocol that complies with the National Institutes of Health Guide for the Care and Use of Laboratory Animals (Institute of Laboratory Animal Resources, 1996). The studies and related designs are presented in Table 2. All compounds were formulated in a stock vehicle comprised of 98% (v/v) Phosal 50 PG and 2% Tween (Polysorbate 80); the final formulation for dosing was comprised of 12% (by volume) of this mixture in water (88% by volume). At the specified time points after dosing, the mice were euthanized by carbon dioxide asphyxiation. Whole blood samples (1 ml) were collected by cardiac puncture into EDTA-containing tubes, and plasma was separated by centrifugation at 1500g for 10 min at 4°C. The plasma was distributed into separate tubes and stored at −20°C for measurements of drug concentration and Aβ level. CSF samples (10 μl) were collected by cistern magna puncture using a sterile 25-gauge needle followed by suction with 10 μl of Rainin pipette (Rainin Instruments, Woburn, MA). CSF samples were distributed into separate tubes for drug exposure measurements (3 μl) and Aβ analysis (remainder) and frozen on dry ice. Whole brain was collected, the cerebellum was removed, and the remainder was bisected into left and right hemispheres (hemi brains). Brain pieces were weighed and frozen on dry ice, with cerebellum being used for subsequent exposure analysis. CSF and brain samples were stored at −80°C before analysis.
Plasma Aβx-40, CSF Aβx-40, and brain Aβx-40 and Aβx-42 were measured using ELISA. Frozen hemi brains were homogenized (10% w/v) in 50 mM Tris buffer, pH 8.0, containing 5 M guanidine HCl, by using a QIAGEN (Valencia, CA) tissue lyser. Each sample was homogenized with a 5-mm stainless-steel bead, four times, at a shaking rate of 24 times/s for 90 s, then incubated at room temperature for 3 h, and ultracentrifuged at 125,000g for 1 h at 4°C. The resulting supernatant was removed and stored in a 96-well polypropylene deep well plate at −80°C. Aβ peptides were further purified through solid-phase extraction using Waters (Milford, MA) Oasis reverse-phase HLB 96-well column plates (60 mg). Column eluates in ammonium hydroxide from 800 μl of brain supernatant were evaporated to complete dryness and stored at −80°C until assay. Plasma was diluted 1:1 in the guanidine homogenization buffer and incubated overnight at 4°C with constant rotation. The entire mixture was further purified through solid-phase extraction. Column eluates in ammonium hydroxide from 175 μl of plasma were evaporated to complete dryness and stored at −80°C until assay. CSF did not require any purification before assay.
For the Aβx-40 assay, a 384-well black Nunc Maxisorp plate (VWR, West Chester, PA) was coated with 15 μl/well (4 μg/ml) capture antibody (Rinat 1219; Pfizer). For the Aβx-42 assay, 15 μl/well (8 μg/ml) capture antibody (Rinat 10G3; Pfizer) was used. The plates were sealed and incubated at 4°C overnight, and then washed four times with phosphate-buffered saline containing 0.05% Tween 20 (PBS-T) and blocked with 75 μl of blocking buffer (1% bovine serum albumin in PBS-T) for 2 h at room temperature.
For 129/SVE mouse Aβ measurements, after washing the plates with PBS-T, the rodent Aβ40 (California Peptide, Napa, CA) or Aβ42 (California Peptide) standard was serially diluted in blocking buffer, and 15 μl was applied to the plate in quadruplicate. Dried brain samples were reconstituted in 120 μl of blocking buffer, which corresponds to a 6.67-fold concentration. Fifteen microliters of undiluted brain sample was added to the Aβx-42 assay plate in triplicate, and 15 μl of a 1:2 diluted brain sample was added to the Aβx-40 assay plate in triplicate. Dried plasma samples were reconstituted in 40 μl of blocking buffer, which corresponds to a 4.375-fold concentration, and 15 μl was applied in duplicate to an Aβx-40 assay plate. Frozen CSF samples were thawed on ice and diluted 1:8 in blocking buffer, and 15 μl was applied in duplicate to an Aβx-40 assay plate. All sample dilutions were determined previously to be within the linear range of detection of the assay.
For Tg2576 mouse Aβ measurements, after washing the plates with PBS-T, the human Aβ40 (California Peptide) or Aβ42 (California Peptide) standard was serially diluted in blocking buffer, and 15 μl was applied to the plate in quadruplicate. Dried brain samples were reconstituted in 120 μl of blocking buffer. The samples were then diluted 1:20 and 1:5 for Aβ40 and Aβ42 measurements, respectively. For Aβx-40 measurements in plasma and CSF, dried plasma samples were diluted 1:18 in blocking buffer, whereas CSF samples were diluted 1:175 in blocking buffer.
All plates were incubated with standards and samples for 2 h at room temperature, and then washed with PBS-T. Fifteen microliters of detection antibody (4G8-Biotin; Covance Research Products, Princeton, NJ; 200 ng/ml) was added to each well and incubated for 2 h at room temperature. The plates were then washed with PBS-T, and 15 μl of europium-labeled streptavidin (PerkinElmer Life and Analytical Sciences) with 50 ng/ml in blocking buffer was added for a 1-h incubation in the dark at room temperature. The plates were washed again with PBS-T, and 15 μl of PerkinElmer Enhancement solution (PerkinElmer Life and Analytical Sciences) was added to each well for a 20-min incubation at room temperature. The plates were read on an Envision plate reader using DELFIA time-resolved fluorimetry (excitation 340/emission 615).
Measurement of Compound Concentrations in Samples.
All samples were frozen at −20°C until analyses. Sample preparation and liquid chromatography-tandem mass spectrometry (LC-MS/MS) methodology for all four compounds was developed internally. Standard curves were prepared in respective matrix via serial dilution in the concentration range of 0.488 to 1000 ng/ml (plasma and CSF) or 0.488 to 1000 ng/g (brain). For plasma, a 50-μl aliquot of sample was precipitated with 300 μl of acetonitrile containing internal standard. Samples were vortexed for 1 min, then centrifuged at 3000 rpm for 10 min. The supernatant (250 μl) was transferred to a 96-well plate, evaporated under nitrogen, and reconstituted with 100 μl of 75:25 water/acetonitrile. Frozen brain tissue was weighed, and an isopropanol/water (60:40) volume equivalent to four times the mass was added before homogenization in a bead beater (BioSpec Products Inc., Bartlesville, OK). For brain and CSF, a mixed matrix approach was used. To generate the brain standard curve, 50 μl of blank brain homogenate matrix was added to a 50-μl aliquot of each point in the plasma curve. Likewise, 50 μl of blank plasma was added to 50 μl of each brain homogenate sample. For CSF, 5 μl of blank artificial CSF matrix was added to a 5-μl aliquot of each point in the plasma curve to generate the CSF standard curve. Likewise, 5 μl of blank plasma was added to 5 μl of each CSF sample. These samples were then processed as described above for plasma.
LC-MS/MS analysis was carried out using a high-performance liquid chromatography system consisting of tertiary Shimadzu LC20AD pumps (Shimadzu Scientific Instruments, Columbia, MD) with a CTC PAL autosampler (Leap Technologies, Carrboro, NC) interfaced to an API 4000 LC-MS/MS quadruple tandem mass spectrometer (AB Sciex Inc., Ontario, Canada). LY450139, BMS-708163, GSI-953, Elan-X, and a structurally similar internal standard were separated on a Synergi Max RP, 2 × 30 mm, 4-μ, 80-A column (Phenomenex, Torrance, CA) by gradient elution using a flow rate of 0.5 ml/min. A 20-μl sample was injected into the column. The mobile phase consisted of solvent A (10 mM ammonium formate in 0.1% formic acid) and solvent B (acetonitrile). The gradient was as follows: solvent B was held at 5% for 0.5 min, linearly ramped from 5 to 90% in 1.5 min, held at 90% for 0.5 min, and then ramped to 5% over 0.5 min. The mass spectrometer was operated using negative electrospray ionization in multiple reaction monitoring mode. The ion pairs monitored were 360.1/203.1 for LY450139, 519.1/175.1 for BMS-708163, 389.7/237.7 for GSI-953, and 439/188 for Elan-X. All raw data were processed using Analyst Software v1.4.2 (AB Sciex Inc.).
Statistical Analysis and Data Presentation.
A two-way analysis of variance was used to compare the vehicle-normalized Aβ time courses of the compound-treated groups with concurrent vehicle controls in the plasma, CSF, and brain compartments. After a significant main effect by analysis of variance, a Bonferroni post test was applied to evaluate the significance of effect at each time point. The statistical analyses were conducted using Prism v5.01 (GraphPad Software Inc., San Diego, CA). All data are presented as mean percentage of vehicle control ± S.E.M.
We have observed hysteresis, where an Aβ time course lags behind an exposure time course in the CSF and brain, after treatment of various GSIs in multiple species. Therefore, a semimechanistically based indirect response model (Jusko and Ko, 1994) was used to analyze the observed exposure and Aβ data. Before modeling, the Aβ time courses in the treated groups were normalized by the concurrent vehicle time course to remove potential fluctuations in the Aβ levels caused by nonspecific effects. The indirect response model (Fig. 1) assumes that the control-normalized Aβ level in a given compartment (brain or CSF) under physiological conditions is governed by a zero-order generation process (with a rate of Kin) and a first-order clearance process (with a rate constant of kout), which can be described mathematically as
Per the pharmacological mechanism of inhibiting γ-secretase, the Aβ generation rate, Kin, is modified by an inhibitory sigmoidal effect on a GSI treatment, as expressed in eq. 2:
The CSF and brain compartments were modeled separately. Given the hypothesis that the lowering in CSF Aβ is caused fundamentally by the inhibition of brain Aβ generation, the brain exposure was used to drive the effect in both brain and CSF compartments. Brain exposures were described by using compartmental PK modeling or the nonparametric, connect-the-dot interpolation approach, as appropriate.
By fitting calculated Aβ time courses to normalized Aβ data, the PK/PD model estimates the values of kout, Imax (maximum inhibition of Kin), IC50 (concentration at which 50% of the maximum inhibition of Kin is achieved), and γ (Hill coefficient). In some cases, the data did not cover sufficient Aβ dynamic ranges, preventing reliable estimation of all four parameters simultaneously. Thus, we reduced the number of unknown parameters by fixing γ to 1 or Imax to 1 as appropriate.
The set of parameters (Imax, IC50, γ, and kout) that seems biologically reasonable and yielded the lowest value of the objective function (i.e., best fit of Aβ time courses) was chosen to be the final. As such, from a set of exposure-Aβ data, an exposure-Aβ generation relationship, as described in eq. 3, can be derived:
This exposure-Aβ generation relationship was defined as the intrinsic PK/PD relationship of GSIs in this study. Unlike the observed exposure-Aβ relationship, this intrinsic PK/PD relationship is devoid of the confounding impacts of PK and Aβ turnover and hence reflects the true in vivo potency and efficacy of a compound.
All modeling was implemented using the PK/PD computer program NONMEM V (GloboMax, Hanover, MD).
Comparison of Intrinsic PK/PD Relationships.
Comparisons of intrinsic PK/PD relationships across strains and species were realized by plotting and visually inspecting the curves of calculated Aβ generation rates versus free brain concentrations.
Effects of GSIs on Plasma Aβx-40 in Mice.
All four compounds reduced plasma Aβx-40 in a dose-dependent manner in both Tg2576 and 129/SVE mice (Fig. 2). Two striking differences in the effects between the strains were observed: 1) the effects in the Tg2576 mice were much more pronounced and more long-lived than in the 129/SVE mice, as demonstrated with LY450139 and BMS-708163; and 2) significant plasma Aβx-40 rise above the concurrent vehicle control was observed in 129/SVE mice after BMS-708163, LY450139, and GSI-953 treatment. The rise seemed to be inversely related to dose. In the Tg2576 mice, however, no significant rise was demonstrated.
The minimal hysteresis in plasma Aβ response in the animals allowed PK/PD analyses via plotting the plasma Aβ versus free plasma exposures. As shown in Fig. 3, although the exposure ranges in the 129/SVE and Tg2576 mice were similar, the Aβ lowering was stronger in the Tg2576 mice; that is, the exposure-response curve in the Tg2576 mouse left-shifted compared with that in the 129/SVE mouse. Moreover, in the animals treated with BMS-708163, a robust increase in plasma Aβ occurred in the 129/SVE mice (up to 200% of vehicle level), but not in the Tg2576 mice. The exposure-response curve for BMS-708163 in the 129/SVE mouse largely overlapped that observed in the human (no hysteresis in human), suggesting that, at least for this compound, the 129/SVE mouse plasma PK/PD relationship is predictive of that in human. Although plasma Aβ efficacy data are available in human for LY450139 (Siemers et al., 2005), the significant hysteresis in human prevents a direct interspecies comparison on the free plasma concentration versus Aβ plot.
Effect of GSIs on CSF and Brain Aβ in Mice.
The effects of the GSIs on CSF Aβx-40 and brain Aβx-40 and Aβx-42 are summarized in Fig. 4. The effects were dose- and exposure-dependent for all endpoints in both CSF and brain compartments. Although the maximal effects on CSF and brain Aβx-40 were generally comparable in the 129/SVE mice (no CSF data collected from the Tg2576 mice), the effect on CSF Aβx-40 was much short-lived. An apparent Aβx-40 rise (∼130% of vehicle) in the CSF was noted in the groups treated with 30 mg/kg GSI-953 or LY450139 at 8 h and with 32 mg/kg Elan-X at 18 h. The rise, however, was not statistically significant (p > 0.05). In the brains of both strains, the profiles of Aβx-40 and Aβx-42 were largely parallel to each other, although the lowering of Aβx-40 seemed to be greater than that of Aβx-42. Unlike in plasma, no Aβ rise was noted in the brain of either strain.
Intrinsic PK/PD Relationship in Mice.
The PK/PD modeling describes all Aβ data adequately. A representative example of model fitting is shown in Fig. 5. The CSF Aβx-40 and brain Aβx-42 time courses in the 129/SVE mouse treated with Elan-X are replicated by model predictions (Fig. 5). The visual predictive check plots for brain Aβx-42 fitting (Supplemental Fig. S3, A-C) also indicate reasonable performance of the model. The modeling-derived intrinsic PK/PD (free brain exposure − Aβ generation rate) relationships of brain Aβ for GSI-953, LY450139, and BMS-708163 in both strains are juxtaposed in Fig. 6. With the exception of brain Aβx-42 for GSI-953, Fig. 6 indicates that, in general, the GSIs have similar or slightly higher potency and greater efficacy in inhibiting brain Aβ generation in 129/SVE mice than in Tg2576 mice.
In 129/SVE mice, as shown in Fig. 7, with the exception of GSI-953, the intrinsic PK/PD relationship of CSF Aβx-40 was generally similar to that of brain Aβx-42. Compared with CSF Aβx-40 and brain Aβx-42, the intrinsic PK/PD curve of brain Aβx-40 was left-shifted.
Intrinsic PK/PD Relationship in Humans Versus 129/SVE Mice.
Comparisons of the intrinsic PK/PD relationship for CSF Aβ in humans and 129/SVE mice are illustrated in Fig. 8. Although the shapes of the curves for 129/SVE mice and humans are somewhat different, for a 50% reduction from baseline of Aβ generation in CSF, the required concentrations in the two species are comparable (up to 3-fold difference).
PD Parameters from PK/PD Modeling.
The key PD parameters, Imax, γ, kout, and IC50, in Tg2576 and 129/SVE mouse and human inferred from PK/PD modeling are listed in Tables 3 to 5. In some cases, the Imax was estimated to be less than 1. This may be either pharmacologically plausible or limited by the data that did not cover sufficient pharmacodynamic range to allow adequate estimation of Imax. Nevertheless, given the observation that a moderate reduction in Aβ generation attenuates plaque burden in the brain (Li et al., 2007), we believe that the upper portion (with approximately 0–60% inhibition of the Aβ generation rate) of the intrinsic PK/PD curves is more pharmacologically relevant, and hence reasonable for interendpoint and interspecies comparisons. As such, the uncertainty in Imax does not affect the conclusions.
IVIVC of the 129/SVE Mouse.
An in vitro-in vivo potency correlation (IVIVC) was evaluated for the 129/SVE mouse based on LY450139, BMS-708163, and Elan-X. GSI-953 was not included because its effect on brain Aβ42 was abnormally low; the reason is not yet understood. As shown in Fig. 9, the free brain concentration causing 50% of reduction from baseline in the generation rate of brain Aβx-42 or CSF Aβx-40 (IC50%) in 129/SVE mice was positively correlated with the in vitro IC50 against Aβ42. We selected IC50% instead of IC50 for IVIVC analyses because a fair comparison using IC50 values is prevented by the difference in Imax inferred from the data. With these IVIVCs, from a measure of in vitro IC50, it is possible to predict for a new GSI its IC50% for brain Aβx-42 or CSF Aβx-40 generation in 129/SVE mice. The similarity between the regression equations for the two endpoints agrees with the observation above (Fig. 7) that the intrinsic PK/PD relationships of brain Aβx-42 and CSF Aβx-40 generally track with each other. The total brain IC50%, however, does not seem to positively correlate with the in vitro IC50, suggesting total brain concentration is not likely to be indicative of the concentration at the target site.
IVIVC of Human.
The IVIVC for inhibition of human CSF Aβ generation based on LY450139 and BMS-708163 is summarized in Table 6. To achieve a 50% reduction in CSF Aβ generation, the average CSF exposure had to reach 50 to 100 times of in vitro IC50.
In this study, we investigated the suitability of 129/SVE mice as a preclinical pharmacology model for identifying GSIs by systematically evaluating the translatability of the intrinsic PK/PD relationships of four structurally diverse GSIs across Tg2576 and 129/SVE mice and humans. Our results indicate that 129/SVE mice can be used for identification of GSIs from an Aβ-lowering perspective.
PK/PD Modeling Is Critical for Rational Interpretation of Dose-Exposure-Aβ Response Data.
Because of the hysteresis often observed in Aβ response after GSI treatment, the interpretation of dose-exposure-Aβ response data is complex. The current practice generally involves such empirical approaches as qualitative or semiquantitative dose-effect assessment, single time-point exposure-effect assessment, area under the concentration curve (AUC) versus maximum effect, or AUC versus area under the effect curve assessment. These approaches, albeit convenient to implement, have serious flaws and limitations.
First, they compress a set of time-course PK/PD data to a single pair of dose or concentration versus effect values, underusing valuable information, e.g., temporal relationship.
Second, they prevent rational evaluation of in vivo potency of a compound. The AUC/maximum effect or AUC/area under the effect curve ratios seem to be an indicator of in vivo potency. These ratios, however, are controlled not only by potency, but also by dose, PK properties, and Aβ turnover. These ratios can lead to deficient estimation of in vivo potency. Illustrated in Supplemental Table S1 are two issues with these ratios: 1) they are dose-dependent, contradicting the notion that in vivo potency, as a compound's intrinsic property, is independent of dose; and 2) Elan-X seems to be ∼5-fold less potent than BMS-708163, inconsistent with the observation that they are approximately equipotent in vitro.
Third, they prevent reasonable translations across species and dosing regimens. These approaches fail to dissect PK, intrinsic PK/PD relationship, and Aβ turnover from Aβ data collected under given conditions; hence, they fail to allow proper scaling and reintegration of the elements to yield projections for targeted conditions, e.g., different dosing regimens, another species.
Finally, they prevent researchers from gaining insights from complex PK/PD data. For example, the profiles of CSF Aβx-40 and brain Aβx-40 and Aβx-42 are different after a GSI treatment in the 129/SVE mouse (Fig. 4). The empirical approaches will interpret the data such that the potency of a GSI against CSF Aβx-40 and brain Aβx-40 and Aβx-42 descends in order. However, these approaches cannot assess the validity of the interpretation or offer mechanistic explanation of the apparent disparity in potency.
Given the flaws and limitations, those qualitative or semiquantitative empirical approaches should be avoided.
We propose to use a semimechanistic PK/PD model to interpret Aβ data, as demonstrated in this study. An observed Aβ time course is a composite of PK, intrinsic PK/PD relationship, and Aβ turnover (Fig. 10). From a set of dose-concentration-Aβ response data, a PK/PD model can separate these elements and yield a quantitative description of each. As such, the in vivo potency of a GSI can be properly defined, and understanding can be attained in how PK behavior and Aβ turnover affect Aβ profile.
Our PK/PD modeling suggested that the difference in the lowering of CSF Aβx-40 and brain Aβx-40 and Aβx-42 in the 129/SVE mouse is caused predominantly by the different turnover of the three endpoints. The kout values of these endpoints descend in order with excellent consistency across the four compounds (Table 4); implications of kout on Aβ profile will be discussed elsewhere (Y. Lu, H. Barton, L. Leung, L. Zhang, E. Hajos-Korcsok, C.E. Nolan, J. Liu, S.L. Becker, K.M. Wood, A.E. Robshaw, et al., manuscript in preparation). The in vivo potencies against the three endpoints are actually similar (Fig. 7), with brain Aβx-40 slightly left-shifted. With this knowledge, we concluded that the apparent weaker effect on brain Aβx-42 probably is a result of the physiology of slower Aβx-42 turnover (compared with turnover of brain and CSF Aβx-40). We also concluded that the effect on CSF Aβx-40, interpreted appropriately, is predictive of the effect on brain Aβx-42.
Other values of the PK/PD modeling demonstrated in this study are:
1) The 129/SVE mouse IVIVC confirms that free brain exposure is a relevant exposure for Aβ lowering. Because the catalytic subunit of γ-secretase and the APP γ-cleavage site are membrane-embedded (Selkoe and Wolfe, 2007) it seems logical to expect the concentration of a GSI in the lipid bilayer, closely related to total brain concentration, to drive the effect. To the contrary, the 129/SVE mouse IVIVC confirms that free, rather than total, brain concentration is the driving force. Even though the free drug hypothesis holds in this case, we caution that this hypothesis should not be taken for granted. Violations of the hypothesis, albeit rare, do occur in our experience (Y. Lu, unpublished data). We advocate that, as a general rule, the relevant target site exposure should be identified early on for discovery programs. This exercise will have significant impacts on chemistry design strategy, definition of efficacious exposure, projection of clinical dose, and assessment of target engagement in the clinic.
2) The human IVIVC defines a quantitative relationship between in vitro IC50 against Aβ42 and in vivo IC50% against input of Aβ1-40 and Aβ1-42 into CSF. The in vivo IC50% is 50 to 100 times higher than the in vitro IC50. This relationship enables the determination of efficacious exposures and projection of human doses in silico based on in vitro potency and clearance. As such, the discovery of new GSIs can be expedited drastically.
It is noteworthy that PK/PD modeling is not without limitations. The quality of a modeling-based interpretation depends on the adequacy of the estimation of unknown pharmacodynamic parameters. The unknown parameters are estimated by fitting experimental data per statistical criteria. If a data set is too sparse (e.g., too few time points) or covers an insufficient dynamic range, unknown parameters may not be estimated with confidence. For instance, in this study, LY450139 data from 129/SVE mice do not allow adequate estimation of all parameters; fixing Imax to 1 allowed reasonable estimation of IC50 and γ. In this particular case, the limitation is not likely to compromise the conclusion. Nevertheless, we advocate that, in general, an experiment should be designed properly to allow PK/PD modeling-based analysis. To achieve that, inputs from modelers on study design should be sought and considered.
The 129/SVE Mouse Is a Suitable Model for Identifying GSIs.
Emerging laboratory findings and our analyses here argue that the Tg2576 mouse offers little advantage over the 129/SVE mouse for the discovery of GSIs.
Pathologically and biochemically, there are fundamental differences between Tg2576 mice and AD: 1) Tg2576 mice fail to duplicate the τ-phosphorylation and develop neurofibrillary tangles, one of the hallmarks of AD (Schwab et al., 2004); 2) the Aβ in Tg2576 mice is not subject to as extensive N-terminal post-translational modifications as in AD (Gravina et al., 1995; Kawarabayashi et al., 2001); 3) neuronal loss in Tg2576 mice is much less severe than in AD (Schwab et al., 2004); 4) the inflammatory response in Tg2576 mice is minor compared with that seen in AD (Schwab et al., 2004); and 5) the Swedish mutation renders the β-site drastically different β-cleavage kinetics (Hook et al., 2008) and compartmentalization of APP processing (Yamakawa et al., 2010) from that seen in the background species. These differences indicate that the Tg2576 mouse is not an ideal AD model as perceived.
From our quantitative analyses, we found that:
1) 129/SVE, but not Tg2576 mice demonstrated significant and dose-dependent Aβ rise after an initial reduction in plasma after an acute treatment of BMS-708163, LY450139, and GSI-953, a phenomenon consistently observed in human (Siemers et al., 2005; Martone et al., 2009; Wang et al., 2010). The mechanism of Aβ rise in plasma remains elusive. Nevertheless, the similar pattern of plasma Aβ response in the 129/SVE mouse and human suggests that the pharmacology, at least in plasma, is similar between these two species.
2) The 129/SVE mouse, in general, is at least as sensitive as and more efficacious than the Tg2576 mouse to GSI treatment in terms of inhibition of brain Aβ generation and observed lowering of brain Aβ itself.
3) In the 129/SVE mouse, the intrinsic PK/PD for CSF Aβx-40 is largely similar to that for brain Aβx-42.
4) The intrinsic PK/PD for CSF Aβx-40 in the 129/SVE mouse is predictive of that for human CSF Aβ. Points 2 to 4 collectively suggest that the intrinsic PK/PD for brain Aβx-42 in the 129/SVE mouse is predictive of that for CSF Aβ in humans.
5) Although other species, e.g., rat (Best et al., 2005; Gillman et al., 2010), guinea pig (Lanz et al., 2006), dog (Gillman et al., 2010), and nonhuman primate (Cook et al., 2010), have also been used for characterizing GSI PK/PD, our analyses (Y. Lu, unpublished data) and practical restraints do not support these species to be superior to 129/SVE mice.
Taken together, the laboratory observations and our quantitative analyses, coupled with practical considerations, suggest the 129/SVE mouse is a suitable preclinical pharmacology model for identifying GSIs.
We acknowledge that the 129/SVE mouse is not a model of AD and it may not completely replace Tg2576 mouse or other transgenic models of AD for GSI discovery. Although preclinical demonstration of brain Aβ lowering is the primary evidence supporting a GSI advancing to clinical development, it is often desired to achieve preclinical efficacy in the reduction of brain plaque load and improvement of behavioral or functional endpoints, despite the predictability of these assays for the clinic remaining questionable. Because of the lack of excessive plaque load or behavioral or functional deficits in 129/SVE mice, these assays have to be established by using Tg2576 mice or other transgenic models.
In summary, we demonstrated that the intrinsic PK/PD relationship for brain Aβx-42 and CSF Aβx-40 in the 129/SVE mouse is indicative of that for human CSF Aβ. This result, along with practical considerations, strongly suggests that the 129/SVE mouse is a suitable model for the discovery of GSIs. We also illustrated the necessity and utilities of PK/PD modeling for the rational interpretation of Aβ data.
Participated in research design: Lu, Zhang, Nolan, and Riddell.
Conducted experiments: Nolan, Becker, Atchison, Robshaw, Pustilnik, Osgood, and Miller.
Contributed new reagents or analytic tools: Stepan, Subramanyam, Efremov, and Hallgren.
Performed data analysis: Lu, Zhang, Nolan, and Becker.
Wrote or contributed to the writing of the manuscript: Lu, Nolan, Atchison, Robshaw, Osgood, Miller, and Riddell.
This work was supported by Pfizer Worldwide Research and Development.
Article, publication date, and citation information can be found at http://jpet.aspetjournals.org.
- Alzheimer's disease
- amyloid β
- amyloid precursor protein
- area under the concentration curve
- free brain concentration
- cerebrospinal fluid
- γ-secretase inhibitor
- in vitro-in vivo potency correlation
- dimethyl sulfoxide
- enzyme-linked immunosorbent assay
- phosphate-buffered saline containing 0.05% Tween 20
- liquid chromatography-tandem mass spectrometry
- maximum inhibition of Aβ generation
- Hill coefficient
- concentration that causes 50% of maximum inhibition of Aβ generation
- Aβ clearance rate constant.
- Received August 9, 2011.
- Accepted September 15, 2011.
- Copyright © 2011 by The American Society for Pharmacology and Experimental Therapeutics