Abstract
Human β-amyloid precursor protein (APP) transgenic mice are commonly used to test potential therapeutics for Alzheimer's disease. We have characterized the dynamics of β-amyloid (Aβ) generation and deposition following γ-secretase inhibition with compound LY-411575 [N2-[(2S)-2-(3,5-difluorophenyl)-2-hydroxyethanoyl]-N1-[(7S)-5-methyl-6-oxo-6,7-dihydro-5H-dibenzo[b,d]azepin-7-yl]-l-alaninamide]. Kinetic studies in preplaque mice distinguished a detergent-soluble Aβ pool in brain with rapid turnover (half-lives for Aβ40 and Aβ42 were 0.7 and 1.7 h) and a much more stable, less soluble pool. Aβ in cerebrospinal fluid (CSF) reflected the changes in the soluble brain Aβ pool, whereas plasma Aβ turned over more rapidly. In brain, APP C-terminal fragments (CTF) accumulated differentially. The half-lives for γ-secretase degradation were estimated as 0.4 and 0.1 h for C99 and C83, respectively. Three different APP transgenic lines responded very similarly to γ-secretase inhibition regardless of the familial Alzheimer's disease mutations in APP. Amyloid deposition started with Aβ42, whereas Aβ38 and Aβ40 continued to turn over. Chronic γ-secretase inhibition lowered amyloid plaque formation to a different degree in different brain regions of the same mice. The extent was inversely related to the initial amyloid load in the region analyzed. No evidence for plaque removal below baseline was obtained. γ-Secretase inhibition led to a redistribution of intracellular Aβ and an elevation of CTFs in neuronal fibers. In CSF, Aβ showed a similar turnover as in preplaque animals demonstrating its suitability as marker of newly generated, soluble Aβ in plaque-bearing brain. This study supports the use of APP transgenic mice as translational models to characterize Aβ-lowering therapeutics.
Deposits of the Aβ peptide known as amyloid plaques are one of the defining pathological hallmarks of Alzheimer's disease, and aggregated Aβ species are considered to play a key role in disease pathogenesis (Hardy and Selkoe, 2002). Generation of Aβ from the membrane-bound β-amyloid precursor protein (APP) involves consecutive cleavage by the β-secretase BACE1 and the γ-secretase complex (Wolfe, 2006). BACE catalyzes the cleavage at the N terminus of Aβ, releasing a soluble form of APP (sAPPβ) and leaving a C-terminal fragment (C99) in the membrane. C99 is then processed by γ-secretase, which possibly involves three successive cleavage steps (Zhao et al., 2005), finally yielding a set of Aβ peptides heterogeneous at the C terminus, with the most abundant ends at positions 40, 42, and 38. The other product of this cleavage is the APP intracellular domain thought to be a regulator of gene expression (Wolfe, 2006). The γ-secretase is a complex composed of presenilins (PS1 or PS2), nicastrin, PEN-2 (presenilin enhancer-2), and anterior pharynx-defective protein 1 (APH-1). Considerable evidence suggests that the presenilins contain the active site of this intramembrane aspartyl protease (Wolfe, 2006). A large number of presenilin mutations are associated with familial AD (Tandon and Fraser, 2002). They usually shift the Aβ generation to the longer Aβ42 isoform as demonstrated in cellular and transgenic models (Citron et al., 1997) as well as in AD patients (Scheuner et al., 1996). γ-Secretase was shown to also cleave a number of other type I transmembrane proteins (Selkoe and Kopan, 2003), the best characterized one being Notch, an important regulator of cell fate. Accordingly, mice deficient in PS1 (Wong et al., 1997) or nicastrin (Li et al., 2003) are not viable, whereas PS2-knockout mice survive and lack an overt phenotype (Donoviel et al., 1999).
Because of its pivotal role in Aβ generation, γ-secretase has been a target of drug development, and several substances have been described to inhibit γ-secretase in vitro (Shearman et al., 2000; Dovey et al., 2001) in rats (Best et al., 2005), in guinea pigs (Anderson et al., 2005), and in several APP transgenic mouse models of Alzheimer's disease (Dovey et al., 2001; Cirrito et al., 2003; Lanz et al., 2004; Wong et al., 2004; Anderson et al., 2005; Barten et al., 2005; Best et al., 2005, 2007; Churcher et al., 2006). Inhibition of γ-secretase in rodents resulted in a substantial reduction of Aβ in brain and cerebrospinal fluid (CSF) within a few hours after substance application. In several in vivo models, Aβ reduction by either γ-secretase inhibition or conditional presenilin knockout was accompanied by a beneficial effect on cognitive function (Comery et al., 2005; Dash et al., 2005; Saura et al., 2005). However, severe side effects concerning lymphatic and intestinal cell differentiation have been observed for γ-secretase inhibitors seemingly due to inhibition of Notch cleavage (Searfoss et al., 2003; Wong et al., 2004). Application to humans has been described for one γ-secretase inhibitor only, which showed reduction of Aβ in plasma but not in CSF (Siemers et al., 2007a).
In this study, we used the γ-secretase inhibitor LY-411575 (May et al., 2001) to characterize the dynamics of Aβ in brain, CSF, and plasma in several APP transgenic mouse models before onset of amyloid plaque formation. The corresponding changes in the γ-secretase substrates, the APP C-terminal fragments, were analyzed in comparison. Subfractionation of brain was used to distinguish Aβ populations with different turnover. We also analyzed the effect of chronic reduction of Aβ production in a preventive as well as a therapeutic paradigm, i.e., in preplaque and plaque-bearing mice. Our study describes key characteristics of Aβ generation and deposition in APP transgenic mice.
Materials and Methods
Experimental Animals. The transgenic mouse lines used express human APP751 under the control of the murine Thy1-promoter, resulting in neuron-specific expression. APP23 mice (Sturchler-Pierrat et al., 1997) express APP with the K670M/N671L “Swedish” mutation. In APP24 mice, the Swedish double mutation was combined with the “London” mutation V717I. APP51/16 mice express wild-type APP (Herzig et al., 2004). APP23 and APP51/16 overexpress human APP approximately 7 and 10 times over endogenous APP. Mostly compact amyloid plaques start appearing at the ages of approximately 6 and 12 to 15 months, respectively. In APP24 mice, overexpression of human APP is ∼3.5-fold. Mostly diffuse amyloid plaques start appearing at 8 months of age. The mice were on a C57BL/6 background and hemizygous for the transgene. All animal experiments were in compliance with protocols approved by the Swiss Animal Care and Use Committees.
Substance Application and Brain Concentration. The γ-secretase inhibitor LY-411575 (for chemical structure, see Wong et al., 2004) was used. Mice received vehicle alone (0.25% methylcellulose in water) or 1 to 10 mg/kg LY-411575 suspended in vehicle solution by oral gavage at 20 ml/kg. The following compound concentrations (mean ± S.D. as nM) were determined in brain after treatment of male mice with 10 mg/kg LY-411575 (0.5 h, 111 ± 45; 1 h, 83 ± 8; 2 h, 52 ± 6; 4 h, 33 ± 4; and 6 h, below limit of quantification). For comparison, the cellular IC50 value of LY-411575 is ∼0.1 nM (unpublished results with APP23 and APP51/16 primary neurons) (Wong et al., 2004).
Mouse Sacrifice and Tissue Sampling. CSF was sampled from the cisterna magna under anesthesia with 3% isoflurane. In most cases, 3 to 10 μl of CSF could be obtained. Mice were decapitated, and EDTA-plasma was prepared from the blood. In experiments with chronic substance treatment, one brain hemisphere was fixed by immersion in 4% formaldehyde solution. For biochemical investigation, forebrain was prepared from the second hemisphere by removal of olfactory bulb and hind brain, frozen on dry ice, and stored at -80°C until used. After chronic treatment of APP24 mice, the olfactory bulb and the pons/medulla oblongata from one brain hemisphere were also isolated and frozen on dry ice.
Homogenization of Brain Tissue. Brain regions were weighed and homogenized by sonication in 10 volumes of Tris-buffered saline (TBS; 20 mM Tris-HCl, pH 7.6, 137 mM sodium chloride, and protease inhibitor cocktail Complete; Roche Diagnostics, Mannheim, Germany).
Determination of APP and Its Metabolites. Aβ was quantified using different methods (Western blotting, quantitative MALDI-TOF, ELISA, and an electrochemiluminescence linked immunoassay). To obtain standards for the quantifications, dilution series of the analytes [synthetic Aβ1–38/40/42 peptides (Bachem, Torrance, CA and r-Peptide, Bogart, GA), recombinant sAPPα and sAPPβ (Meso Scale Discovery, Gaithersburg, MD), APP (in transgenic mouse brain homogenates)] were prepared in nontransgenic mouse brain extract or plasma. The presence of nontransgenic CSF had no influence on ELISA calibration curves from synthetic Aβ standards. Therefore, standards for determination of CSF Aβ were prepared without addition of CSF from nontransgenic mice.
Immunoprecipitation of Aβ. For immunoprecipitation of Aβ from brain tissue, homogenates were extracted with 1% sodium dodecyl sulfate at 95°C for 3 min, diluted with 9 volumes of TBS, and cleared by centrifugation. Aβ peptides were immunoprecipitated using the monoclonal antibody β1 reacting with an epitope near the amino terminus of Aβ (Schrader-Fischer and Paganetti, 1996) and protein G-coated magnetic beads (Dynal Biotech, Hamburg, Germany). Antibody β1 does not recognize rodent Aβ under the conditions used. For immunoprecipitation of Aβ from plasma, antibody β1 covalently coupled to Sepharose beads was used.
Dephosphorylation of APP C-terminal Fragments. Forebrain homogenates were supplemented with 1% Nonidet P-40 (Fluka, Buchs, Switzerland), 50 mM HEPES (pH 7.5), 0.1 mM Na2EDTA, 5 mM dithiothreitol, 0.01% Brij 35, and 3 mM MnCl2 and were incubated with Escherichia coli λ-phosphatase (400 units/mg forebrain; New England Biolabs, Ipswich, MA) at 30°C for 1 h. The reaction was stopped by the addition of SDS sample buffer.
Gel Electrophoresis and Western Blotting. For detection of Aβ, immunoprecipitates, CSF, or brain homogenates were separated on 10% Tris-bicine acrylamide gels with 8 M urea as described previously (Klafki et al., 1996). C-terminal APP fragments were separated on 10 or 13% Tris-bicine acrylamide gels without urea and were detected with rabbit antiserum APP-C8 raised against the C-terminal amino acids of APP (Schrader-Fischer and Paganetti, 1996). Transgene-derived APP and Aβ were detected with the human-specific monoclonal antibody 6E10 (Signet, Dedham, MA). Total APP was detected with antibody 22C11 (Roche Molecular Biochemicals, Rotkreuz, Switzerland). The bands obtained on films were quantified using the software MCID (version M7 Elite; InterFocus Imaging Ltd, Cambridge, UK). Some of the CSF samples analyzed by Western blotting (Fig. 5) were slightly contaminated with blood and globin comigrated with Aβ1–38 on the gel, which precluded quantification of the Aβ1–38 band. For detection of presenilin 1, forebrain homogenates from five animals per group were separated on 13% Tris-glycine acrylamide gels, and PS1 CTFs were detected on Western blots with antibody R28 (Baumann et al., 1997) directed against the large cytoplasmic loop of presenilin 1.
Aβ Quantification by MALDI-TOF. Aβ immunoprecipitated with antibody β1 and protein G-coated magnetic beads as described above was eluted with a acetonitrile/water/formic acid solution (50: 40:10, v/v/v) saturated with α-cyano-4-hydroxy-cinnamic acid. As internal standard, 10 pM bovine insulin was added. Samples were loaded on a MALDI-TOF target plate (Applied Biosystems, Framingham, MA) and dried at ambient temperature. Mass spectra were acquired in the mass range of 3000 to 6000 Da using an Applied Biosystems Voyager-DE STR TOF mass spectrometer. Aβ peptides were located on the spectra by identifying the most prominent peak in a region of ±5 Da of the calculated mass of each peptide. For quantification, the peak height was used because the area under the curve yielded less reliable results when irrelevant peaks were present adjacently.
Aβ Quantification by ELISA. For determination of total Aβ40 or Aβ42, forebrain homogenates were extracted for 15 min at 4°C with 70% formic acid. The extracts were neutralized by the addition of 19 volumes 1 M Tris base and centrifuged for 15 min at 20,000g. For preplaque mice, the supernatant was directly applied to ELISA plates (IBL Human Amyloid β (1–40) JP27713 or (1–42) JP27711 Assay; IBL, Hamburg, Germany; and Innotest β-Amyloid 1–42 80177 Assay; Innogenetics, Gent, Belgium). When forebrains with high plaque load were analyzed, the extracts were further diluted in 70% formic acid, neutralized, and diluted in sample dilution buffer from the ELISA kit as necessary to achieve an Aβ concentration within the standard curve. For analysis of TBS-soluble Aβ, forebrain homogenates were centrifuged at 100,000g for 15 min. For Triton X-100 soluble Aβ, Triton X-100 was added to the forebrain homogenate (final concentration 1%), and samples were extracted on ice for 15 min before centrifugation at 100,000g. Samples were mixed every 5 min. The supernatants were diluted (total dilution 1:80) with sample dilution buffer from the ELISA kit (see above), as undiluted extracts caused high background and a low signal to noise ratio. For determination of Aβ in CSF, the IBL human amyloid β-(1–40) and the Innotest β-amyloid-(1–42) assays were used.
Aβ Quantification by Electrochemiluminescence-Linked Immunoassay. MSD 96-Well MULTI-ARRAY Human (6E10) Abeta 40 or Abeta 42 Ultra-Sensitive Kits were used (Meso Scale Discovery, Gaithersburg, MD). For Aβ40 determination, plasma samples were cleared by centrifugation (1 min, 20,000g), mixed with an equal volume of 3% Blocker A solution included in the assay kit, and further processed according to the manufacturer's instructions. Signals were measured on a SECTOR Imager 6000 reader (Meso Scale Discovery). Triton X-100-soluble Aβ was prepared as described for ELISAs and diluted to a final dilution of 1:100 with 3% Blocker A. This method was also used to verify the Aβ reductions in some experiments and the Aβ concentrations by direct comparison of all preplaque control groups.
Determination of sAPPα and sAPPβ by Electrochemiluminescence-Linked Immunoassay. Forebrain homogenates were extracted with 1% Triton X-100 on ice for 15 min. Samples were centrifuged (15 min, 4°C, 20,000g). The supernatants were further diluted with 1% Triton X-100 in TBS (total dilution 1:2000), loaded on a MS6000 sAPPα/sAPPβ multiplex plate (Meso Scale Discovery) and further processed as described for Aβ.
Estimation of Aβ and CTF Half-Life and Production Rate. The time course of Aβ concentrations after γ-secretase inhibition is described by the equation where [Aβ]t and [Aβ]0 are the concentrations of Aβ at time t and start (in vehicle-treated mice), respectively, k is the first-order rate constant of Aβ removal, and offset is the amount of Aβ not affected by treatment. Data pairs (t, [Aβ]t) were fitted to the above equation to determine k (Origin 7.5 software; Origin Lab Corporation, Northampton, MA). The half-life was calculated using
Under steady-state conditions, which can be assumed to exist in preplaque mice, the rate of production, k′, of Aβ equals its rate of clearance. It can be calculated from
An independent estimate of Aβ production rates was obtained from average Aβ loads measured in 11- and 18.3-month-old APP23 mice with the formula
Half-lives of C99 and C83 were estimated based on their relative concentration in vehicle-treated mice (steady-state concentration) and their production rates p (increase of concentration per time as determined between 0.5 and 2 h after substance treatment) according to
Immunocytochemistry. Formaldehyde-fixed brain halves were embedded into paraffin and sagittally cut into 4-μm sections at three different anatomical levels. Sections were processed as described previously (Sturchler-Pierrat et al., 1997) and immunohistochemically stained with antibodies. Aβ-specific antibodies were rabbit serum NT11 for total Aβ and end-specific mouse monoclonal antibodies 25H10 for Aβ40 and 29C12 for Aβ42 (a gift of P. Paganetti, Novartis Institutes for BioMedical Research). APP was stained with rabbit sera APP474 raised against purified secreted APP from rat cells or with APP-C8. Glial fibrillary acidic protein (GFAP) as a marker of astrocytes was detected with a rabbit antiserum (Dako M0761; Dako, Glostrup, Denmark). The microglial cells were stained with rabbit antiserum Iba1 obtained from Wako Pure Chemicals (Tokyo, Japan). Bound antibody was visualized by using the avidinbiotin-peroxidase technique (ABC-Elite Kit PK6100; Vector Laboratories, Burlingame, CA). Finally, sections were reacted with diaminobenzidine metal-enhanced substrate (Roche Diagnostics) and counter-stained with hemalum.
Amyloid Plaque Load Quantification. Amyloid plaque load as percentage of total area and plaque numbers was measured in neocortex and caudate putamen of NT11-stained sections at three different anatomical levels using a microcomputer-assisted imaging software (MCID, version M7 Elite). Microscopic images were digitized by use of a Roper black and white CCD TV camera (Roper Scientific Photometrics, Tucson, AZ) and stored with 1124 × 1124 pixel resolution at 256 gray levels. Diffuse amyloid was captured with a detail enhancing function of the software. The pixel size was calibrated using an object micrometer at 5× magnification (Leica Neoplan Objective; Leica, Wetzlar, Germany). Using a motor driven microscope stage for exact positioning of adjacent object fields, the entire brain region was analyzed. For each object field, the anatomical area was defined by manual outline. Isolated tissue artifacts were excluded by manual outline.
Statistical Analyses. For analysis of APP and its C-terminal fragments of soluble Aβ and of total Aβ in preplaque mice, Student's t-tests (usually one-tailed) or analysis of variance (ANOVA) followed by post hoc analysis were done (Tukey's test for pairwise comparison of all groups or one-tailed Dunnett's test for comparison of all treated groups to vehicle). To compare magnitudes of the substance effect on different Aβ species or between different extractions, ANOVA and one-tailed Dunnett's test were performed on the decadic logarithm of the ratio of the two values. Given that amyloid loads in plaque-bearing mice do not usually show a normal distribution, we used nonparametric tests for these data (Mann-Whitney U test or the Kruskal-Wallis test followed by post hoc Tukey's test on ranks; p < 0.05 considered significant for all tests, analyses done with Systat for Windows 11; Systat Software Inc., San Jose, CA).
Results
The compound used in this study, LY-411575, has been described previously (May et al., 2001; Wong et al., 2004). To determine optimal inhibition, preplaque APP23 mice expressing human APP with the K670M/N671L Swedish mutation received oral doses between 1 and 10 mg/kg LY-411575 (Supplemental Fig. 1S). The highest dose of 10 mg/kg was slightly more effective than 3 mg/kg and was chosen for further acute studies.
Kinetics of Aβ and CTFs in Preplaque APP23 Mice after a Single γ-Secretase Inhibitor Dose. To analyze the kinetics of Aβ removal in vivo, preplaque APP23 mice received a single oral dose of 10 mg/kg LY-411575. After sacrifice at different times thereafter, total Aβ concentrations in forebrains were determined after formic acid extraction, whereas CSF and plasma Aβ were analyzed directly. Reduced Aβ40 concentrations were observed in all three compartments as early as 30 min after treatment (Fig. 1A). Thereafter, Aβ40 continued to decrease reaching a minimum at 4 to 8 h. CSF and plasma Aβ40 decreased faster and to a lower minimum than total Aβ40 in forebrain. Even at 24 h, the Aβ40 concentrations in forebrain (-57%, p < 0.001) and CSF (-48%, p < 0.01) remained below vehicle values. Of note, in plasma, Aβ40 recovered more rapidly. At 8 h, its concentration was only 18% below vehicle compared with at least 80% for CSF and forebrain Aβ40. Thereafter, no difference from vehicle-treated animals was detectable in plasma. Total Aβ42 in forebrain was less reduced than Aβ40 at all of the time points up to 24 h [significant between 2 (p = 0.01) and 8 h (p < 0.001) Fig. 1B]. However, 53 h after treatment, a significant (-19%, p < 0.01) reduction persisted, whereas Aβ40 was back to vehicle concentrations. The kinetic difference between the Aβ peptides was confirmed in a separate experiment where Aβ was analyzed by Western blotting (data not shown). In another study, the substance was applied intravenously (1 mg/kg), which did not further accelerate the reduction of Aβ in forebrain (data not shown).
The effect of γ-secretase inhibition on APP C-terminal fragments was analyzed on Western blots of forebrain homogenates (Fig. 1C). For C99, a gradual increase was apparent already at 30 min but leveled off rapidly to reach a maximum at 8 h. After 24 h, C99 was still elevated, whereas it had declined to vehicle values at 53 h. In contrast, C83 showed a lag phase followed by a longer increase reaching a maximum at 8 to 24 h. This demonstrates inhibition of γ-secretase at least until 8 h. Even at 53 h, C83 remained considerably above vehicle levels. Overall, the relative increase of C83 was larger than that of C99.
Similar Reduction of Aβ following γ-Secretase Inhibition in APP24 and APP51/16 Mice. To test for potential differences between transgenic lines, we analyzed APP24 mice containing the London mutation (V717I) in the γ-secretase cleavage region of APP (in addition to the K670M/N671L mutation). Compared with APP23, these mice show the expected increased Aβ42/40 ratio but generate less APP and total Aβ. Forebrain and CSF Aβ were measured at 1 and 6 h after a 10 mg/kg LY-411575 treatment of preplaque APP24 mice (Fig. 2A). Total Aβ40 and Aβ42 were significantly reduced already after 1 h. At both time points, the relative decrease of each of the Aβ species was very similar to the one in APP23 mice (Fig. 1A). The smaller reduction of Aβ42 compared with Aβ40 found in APP23 was confirmed (p < 0.05 at 6 h). Average Aβ42 in CSF (1.0 ± 0.1 pmol/ml) was reduced by 33% (p < 0.05) at 1 h and 89% (p < 0.001) at 6 h, which is a stronger reduction than in forebrain, similar to the observation made for Aβ40 in APP23.
We also studied mice expressing human wild-type APP (preplaque APP51/16) treated with 10 mg/kg LY-411575 and sacrificed 6 h later (Fig. 2A). Forebrain showed a 66% decrease in Aβ40 following formic acid extraction, very similar to APP24. For Aβ42, a reduction was also found but could not be quantified reliably. In plasma, Aβ1–40 was reduced by 89% (p < 0.01). These data demonstrate that the three different APP transgenic mouse lines carrying different mutations respond with similar Aβ reductions to the acute inhibition of γ-secretase.
Differential Extraction of a Brain Aβ Subpopulation Undergoing Rapid Turnover. The maximal reduction of total Aβ40 or Aβ42 achieved after γ-secretase inhibition was lower in brain than in CSF. Considering that CSF Aβ originates from brain, this observation hints to brain Aβ species, which undergo a slower turnover resulting in an incomplete removal. We tried to enrich a subpopulation of Aβ, which is completely removed upon acute γ-secretase inhibition. Forebrains of APP23 mice from the kinetic study were extracted with formic acid, 1% Triton X-100, or TBS, and Aβ40 was quantified (Fig. 1A). In vehicle-treated mice, Triton X-100 and TBS extracted ∼30 and ∼2%, respectively, of total formic acid-soluble Aβ40. The relative reduction of Aβ40 was stronger in the Triton X-100 extract than in the formic acid extract at all of the time points of the descending part of the kinetics (p = 0.05 at 4 h and p = 0.016 at 6 h). The kinetics of Aβ40 in Triton X-100 extracts closely reflected those of CSF. In the TBS extracts, the reduction of Aβ40 was even higher than in Triton X-100 extracts; however, the variation was larger (Fig. 1A). Confirming these observations, LY-411575-treated APP24 and APP51 mice also showed a stronger Aβ40 reduction in the Triton X-100 compared with the formic acid extract (Fig. 2A).
APP24 mice were used to test for an enrichment of an Aβ42 population with a more rapid turnover. Triton X-100-soluble forebrain Aβ42 was more reduced after LY-411575 treatment than total formic acid extracted Aβ42 (Fig. 2A, p < 0.05 at 1 h) and paralleled Aβ42 in CSF better, similar to Aβ40 in APP23. A larger Aβ42 reduction in the Triton X-100 extract (-33% at 1 h, below limit of quantification at 6 h) compared with the formic acid extract (Fig. 1B) was also observed for APP23 mice. These results indicate that subpopulations of Aβ40 and Aβ42 exist in brain of preplaque APP transgenic mice, which differ in stability. They also confirm the similarity of the Aβ reduction in the different transgenic mouse models following γ-secretase inhibition.
Estimation of Aβ Half-Life and Production Rate. The decrease of Aβ after γ-secretase blockade allowed us to estimate its half-life by fitting data from the initial inhibition period (Figs. 2B and 3A) using a first-order model for Aβ reduction (see eq. 1 under Materials and Methods). This approach assumes immediate and complete block of Aβ generation, in agreement with brain substance levels that are several orders of magnitude above cellular IC50 values (see Materials and Methods). Although the good fit indicates that γ-secretase inhibition was obtained rapidly, the calculated values are upper limits of the half-life. For APP23 mice, the half-life of formic acid extracted total Aβ40 in forebrain was determined as 1.1 h. This predicts an almost complete disappearance after 4 h, which was not observed, probably reflecting the presence of Aβ subpopulations with considerably longer half-lives. Triton X-100-soluble Aβ40 disappeared with a half-life of 0.7 h and obviously reflects a population with a more rapid turnover. The same value was obtained for CSF Aβ40. Its almost complete loss after 4 h suggests the presence of only one Aβ40 population, probably reflecting the soluble and removable fraction of brain Aβ40. The Aβ40 half-life in plasma was calculated as 0.5 h based on a simple first-order decline (Fig. 3A) but may be shorter assuming an initial lag phase suggested by the 30-min time point.
For formic acid extracted brain Aβ40 from APP24 mice, a shorter half-life (1.3 h) was calculated than it was for Aβ42 (2 h; Fig. 2B). The amount of Aβ42 left after 6 h (50%) was far more than predicted from this half-life (12.5%), concurring with the presence of more stable Aβ42 species as well. However, Aβ42 in Triton X-100 brain extracts showed a faster turnover (t1/2 = 1.7 h), with the same half-life as Aβ42 in CSF.
As the brain concentration of Aβ remains approximately constant in preplaque APP transgenic mice, a steady state between generation and removal of Aβ can be assumed. This allows estimation of the approximate Aβ production rate in forebrain of APP23 mice at 9.3 pmol/g/h for Aβ40 and 0.7 pmol/g/h for Aβ42 (see eq. 3 under Materials and Methods). To test the plausibility of these values, we made the simplifying assumption that newly generated Aβ is quantitatively deposited during the phase of exponential plaque growth. We measured average forebrain Aβ in female APP23 mice at 11 and 18.3 months at 5447 and 65,653 pmol/g (Aβ1–40) and at 1668 and 9820 pmol/g (Aβ1–42), respectively. The increase of Aβ concentrations over time allows to roughly estimate the production rates of Aβ40 and 42 in APP23 forebrain as k′= 11 and 1.5 pmol/g/h, respectively (see eq. 4 under Materials and Methods). These values are in good agreement with production rates calculated from the Aβ removal kinetics. Overall, these estimations indicate a half-life for soluble, nondeposited Aβ40 and 42 of approximately 0.7 and 1.7 h or even less. Additional species exist in brain (e.g., Aβ bound to proteins or molecules in an oligomeric state) with a much longer half-life, which remains to be determined.
The kinetics of APP carboxyl-terminal fragments after blockade of γ-secretase were also used to estimate their degradation by this enzyme. We first determined the production rates for both C83 and C99 from the increase between 0.5 and 2 h. This interval accounts for a possible initial lag phase (Fig. 3B). As described in eq. 5 under Materials and Methods, the half-lives of C99 and C83 for degradation by γ-secretase were then calculated as approximately 0.4 and 0.1 h, respectively.
Subchronic Inhibition of γ-Secretase in Preplaque APP23 Mice. To study the effects of longer γ-secretase inhibition, preplaque male APP23 mice were treated for 2 weeks once daily with 10 mg/kg LY-411575 without signs of overt toxicity. A second set of animals received only a single dose. Mice were sacrificed 6 or 24 h after the last substance application. Brain Aβ levels were significantly more reduced after the 14 day than after a single treatment at the 6-h sacrifice point (Supplemental Fig. 2S). This additive effect is consistent with the observation that a single inhibitor dose-reduced Aβ for longer time than the treatment interval of 24 h (Fig. 1). Six and 24 h after the last of 14 substance applications, forebrain Aβ40 was decreased by 94 and 55%, whereas the Aβ42 reduction reached 89 and 32%, respectively. CTFs accumulated differentially. Although the C99 increase in the forebrain was only slightly higher than after a single treatment, the C89/C83 band was considerably stronger after the 14-day treatment (Fig. 4A). Dephosphorylation allowed us to identify C83 as the main polypeptide in this band (Fig. 4C). These data indicate that γ-secretase was still efficiently inhibited after subchronic LY-411575 treatment.
Chronic Inhibition of γ-Secretase in APP23 Mice during the Early Phase of Amyloid Plaque Formation. Using a preventive paradigm, we next tested whether chronic reduction of Aβ production also reduced amyloid plaque formation. Female APP23 mice (20 per group) were treated for 3 months with LY-411575 and sacrificed at 6 and 24 h after the last treatment. Treatment was started at an age of 6 months when plaque formation just begins. Mice received 10 mg/kg/day of drug for 7 days and 3 mg/kg thereafter to improve tolerability because, after the 1st week, two (10%) substance-treated animals showed weight reduction and sickness behavior and were lost from the study (one animal died, one was sacrificed). The side effects were not further analyzed but may have been related to intestinal goblet cell hyperplasia (Wong et al., 2004; Hyde et al., 2006). The body weight of the remaining animals was not reduced during chronic treatment and developed similarly to the nontreated controls. Yet, after approximately 14 days, gray patches started to appear in the normally black coat of substance-treated animals (Supplemental Fig. 3SC), and thymus atrophy was found at sacrifice (data not shown) as described previously (Wong et al., 2004; Hyde et al., 2006). The average forebrain weight of the treated animals was slightly reduced by approximately 5% (p < 0.05), and a 2.7 times larger volume of CSF could be collected from these animals (p < 0.001; Supplementary Fig. 3SB).
Analysis of total Aβ from forebrains showed that all of the isoforms measured (Aβ1–38, 40, and 42) were reduced by approximately 80% after LY-411575 treatment (Fig. 5). Data from the subgroups sacrificed 6 and 24 h after the last treatment were pooled because they did not differ significantly. Nonetheless, a trend for higher total Aβ values at the 24-h time point indicated an additional small contribution of the acute reduction of newly synthesized Aβ. Consistent with the shorter treatments, accumulated C-terminal APP fragments were found (Fig. 4). The strongest accumulation was found for C83, whereas C89 remained a minor portion of the accumulated CTFs (Fig. 4C). Very similar to acute LY-411575 treatment, Aβ1–38, Aβ1–40, and 1–42 concentrations in CSF were reduced by ∼85, 78, and 61%, respectively, 6 h after the last substance application (Fig. 5B). A significant reduction of Aβ in CSF by 30% (Aβ1–38), 43% (Aβ1–40), and 52% (Aβ1–42) was still seen at 24 h. In plasma, analyzed 6 h after the last substance application, all three Aβ species (Aβ1–38, 40, and 42) were reduced by ∼80% in LY-411575-treated animals (data not shown).
To test for an effect of the treatment on APP expression and its metabolism, we quantified total APP (full-length plus sAPP) and sAPPα and sAPPβ. Total APP on Western blots was slightly decreased in substance-treated animals (-13%, p = 0.008, two-tailed t-test). Average levels of sAPPα (15 ± 1 μg/g in controls) and sAPPβ (53 ± 5 μg/g in controls) were also significantly reduced (-6%, p < 0.01 and -18%, p < 0.001, respectively, two-tailed t-test). Although these changes may indicate a general treatment effect, they are much smaller than the Aβ reductions observed in the different compartments. We also tested for a compensatory up-regulation of presenilin during the long-term γ-secretase inhibition and probed forebrain Western blots with antibodies against the presenilin 1 CTF but did not detect a difference between the LY-411575 and vehicle group (data not shown).
Quantitative immunohistochemistry of the neocortical amyloid load revealed no difference between animals sacrificed at 6 or 24 h after the last treatment, hence both groups were combined for analysis. Compared with the vehicle-treated animals, a 80% reduction of the neocortical plaque area was found in the treated group (Fig. 6, A–D), which was identical in size to the biochemically determined effect. Although the neocortical area covered by plaques exceeded 0.1% in 65% of the vehicle-treated animals, all mice receiving LY-411575 had a plaque area below 0.1%. The median cortical plaque number was reduced by 66% from 4.6 to 1.6 plaques/mm2, i.e., less than the median plaque area, in agreement with a stronger effect on Aβ deposition than on de novo plaque formation.
After 3 months of LY-411575 treatment, a morphological change of intraneuronal Aβ was observed in cortical pyramidal cells. Whereas the Aβ distribution was mostly homogeneous in control mice, treated animals showed strongly stained clusters (Fig. 6, E and F). This change in Aβ distribution was found in cerebral cortex and subiculum, regions forming the first amyloid plaque deposits. It was not observed in the hippocampus proper where plaque formation starts somewhat later. End-specific Aβ40 and Aβ42 antibodies also reacted with these clusters (Supplemental Fig. 4S), confirming their Aβ content rather than a cross-reaction with APP. Consistent with this notion, staining of neuronal cell bodies for APP with antibodies recognizing its N or C terminus was reduced in LY-411575-treated animals (Fig. 7). This included neocortical as well as hippocampal neurons, suggesting a reduction of APP in certain brain regions of treated animals, in accordance with the reduced APP measured biochemically. In contrast, APP C-terminal antibodies strongly stained nerve fibers in several brain regions of LY-411575-treated animals (Fig. 7D). This staining most probably reflected accumulated C-terminal fragments of APP rather than full-length APP, because these structures were not stained by an N-terminal APP antibody (Fig. 7, E and F).
To analyze for an effect on plaque-associated gliosis, we quantified astrocyte (GFAP) and microglia (Iba1) stainings. At the low amyloid plaque load in 9-month-old APP23 mice, these markers were not generally increased compared to nontransgenic animals but rather redistributed resulting in an association with plaques. Accordingly, the GFAP- and Iba1-labeled surface areas in the cerebral cortex or hippocampus remained unchanged by the treatment. However, the Iba1-positive area was increased (+31%, p = 0.01, two-tailed t-test) in the cerebellum of LY-411575-treated mice, indicating some microglia activation. This probably is a substance effect unrelated to the plaque reduction, because plaques are not formed in the cerebellum of APP23 mice.
Chronic γ-Secretase Inhibition of Plaque-Bearing Mice. Given that AD patients already contain amyloid deposits in the brain at diagnosis, we also determined whether chronic γ-secretase inhibition would be effective in the presence of pre-existing plaques. Furthermore, to reproduce the effect in an independent mouse model, we used APP24 mice with a more diffuse, AD-like plaque morphology. A baseline group was sacrificed at 15 months, whereas other animals were treated for 2 months with daily oral doses of 3 mg/kg LY-411575 or vehicle and sacrificed 6 h after the last substance application. None of the mice died during the treatment but animals receiving LY-411575 also showed the coat color changes as described above.
In forebrain of the vehicle group, median Aβ1–38, 40, and 42 concentrations were 29, 37, and 57%, respectively, higher than at baseline, reflecting the increase in Aβ deposition during the 2 months of treatment (Fig. 8A). In contrast, in the LY-411575-treated group, forebrain Aβ remained at baseline level. The olfactory bulb also showed a significant increase of Aβ peptides in the vehicle group, which was largely inhibited by substance treatment (data not shown). Both forebrain and olfactory bulb already contained a considerable amyloid load at the start of the experiment. To analyze the treatment effect at the onset of plaque formation as in the previous study, Aβ was also determined in pons/medulla oblongata, which contains only very few plaques in 15 to 17-month-old APP24 mice. In this brain region, Aβ1–38 and 1–40 remained at baseline in the vehicle group, suggesting that they are not yet deposited (Fig. 8B). Confirming this notion, LY-411575 reduced both peptides below baseline, which indicates turnover. In contrast, Aβ1–42 was three to four times higher in the vehicle group compared to baseline, reflecting the formation of early amyloid deposits. This increase was only partially inhibited in LY-411575-treated animals. In CSF, average Aβ42 (0.9 ± 0.2 pmol/ml) was reduced by 77% (p < 0.001, two-tailed t-test) compared with vehicle-treated animals, demonstrating that an acute Aβ reduction in CSF still occurs in plaque-bearing mice.
The median plaque area of the 15-month-old baseline group was 1.8% in neocortex (Fig. 9D). Two months later, it increased to 5% in vehicle-treated animals. LY-411575 significantly lowered the neocortical plaque area to 4% corresponding to a reduction of further amyloid deposition by 31% (p = 0.01, two-tailed Mann-Whitney U test). In caudate putamen, which has a lower amyloid load than neocortex (median plaque area at baseline: 0.16%, vehicle group: 1.6%), LY-411575 had a more pronounced effect (plaque area of 0.5%, corresponding to a 79% reduction of further amyloid deposition, p = 0.001, two-tailed Mann-Whitney U test). Both plaque area and treatment effect were similar to the corresponding data for neocortex of the 9-month-old APP23 mice described above. The effects on plaque number were comparable to those on plaque area (Fig. 9E), indicating no specific effect on Aβ deposition versus de novo plaque formation. As expected, hardly any plaque could be detected in pons/medulla oblongata. No indication for a removal of pre-existing plaques during the 2-month treatment was found in the brain regions examined.
Discussion
In this study, we have blocked γ-secretase with a potent inhibitor to analyze Aβ turnover and deposition in different APP transgenic mouse models at several stages of amyloid formation. Acute γ-secretase inhibition in preplaque mice resulted in a dose-dependent decrease of brain Aβ and a corresponding increase in APP CTFs. These effects were more pronounced after subchronic treatment. Mice transgenic for wild-type human APP (APP51/16) responded to the same extent as APP23 carrying the Swedish mutation at the BACE cleavage site and APP24, which additionally contained the London mutation within the γ-secretase cleavage region.
The inhibitor used, LY-411575, rapidly achieves a high brain concentration (this study; Cirrito et al., 2003; Lanz et al., 2004) allowing for kinetic studies. Aβ reached a minimum in brain, CSF, and plasma at 4 to 8 h after substance application and recovered only partially until 24 h, indicating a long-lasting inhibitor effect. The turnover of Aβ40 was generally more rapid than that of Aβ42. CSF Aβ40 and 42 both declined slightly faster and to a lower level than total, formic acid-extracted forebrain Aβ. Their decline in TBS and Triton X-100-soluble brain extracts, however, matched the reduction in CSF well, suggesting at least two pools in brain: readily soluble Aβ with a half-life of ∼0.7 (Aβ40) and 1.7 h (Aβ42) as well as a less soluble, more stable pool. For both Aβ species, the half-lives in the soluble pool and in CSF were almost identical. It is likely that CSF Aβ directly originates from the soluble brain pool, which at least in part is rapidly transported from the parenchyma to CSF and then rapidly removed. The origin of Aβ from CNS neurons in these mice (Calhoun et al., 1998) further supports this notion. This first direct comparison of the dynamics of the different Aβ pools in brain and CSF indicates that CSF Aβ40 and Aβ42 are suited markers for the corresponding peptides in the soluble brain pool.
A rapid decline of detergent extracted brain Aβ after γ-secretase inhibition has also been found in rats (Best et al., 2005), guinea pigs (Anderson et al., 2005), and Tg2576 mice (Barten et al., 2005), where an Aβ40 half-life of 0.63 h was determined with another γ-secretase inhibitor (BMS-299897), in very good agreement with this study. Because these calculations all assume immediate and complete inhibition of Aβ generation, the real turnover may be even faster. Aβ half-life determinations in human CSF following γ-secretase inhibition are currently not available (see Siemers et al., 2007a). Longer Aβ half-lives have been measured in the brain interstitial fluid of LY-411575-treated PDAPP mice by in vivo microdialysis [∼2 h for young mice (Cirrito et al., 2003)] and in human CSF with an isotopic labeling method [clearance rate at ∼8%/h, corresponding half-life of ∼8–9 h (Bateman et al., 2006)]. The different techniques used may have contributed to the discrepancies.
The half-life of plasma Aβ40 (≤0.5 h) was shorter than in brain and CSF, in line with the faster turnover observed in other APP transgenic mouse models (Wong et al., 2004; Anderson et al., 2005; Barten et al., 2005). In plasma, Aβ returned to baseline levels more rapidly than in brain and CSF, although the human Aβ measured originates from brain. The APP overexpression in APP23 mice may have overloaded the export system to plasma. Under such conditions, a minor recovery of brain Aβ is sufficient to completely restore the plasma Aβ. Conversely, a lag phase is expected at the beginning of the decline. A rebound increase in plasma Aβ after γ-secretase inhibitor treatment has been described for humans (e.g., Siemers et al., 2007a), guinea pigs (Lanz et al., 2006), and in one (Prasad et al., 2007) but not other (e.g., Wong et al., 2004; Barten et al., 2005; Anderson et al., 2005) APP transgenic mouse studies. It is not accompanied by a change in brain or CSF Aβ suggesting a peripheral origin (as discussed in Siemers et al., 2007a). We have not detected a rebound increase in plasma Aβ, perhaps because human Aβ in APP23 originates almost exclusively from brain (Calhoun et al., 1998) but other possibilities (e.g., rebound between 24 and 53 h) cannot be excluded.
An estimate of the Aβ production rate in APP23 mice was obtained from its rate of removal, which is equal assuming Aβ steady state in preplaque mice. The production rate was independently estimated from the Aβ accumulation in plaque-bearing APP23 mice supposing that all produced Aβ is deposited. The values obtained with these entirely different approaches were remarkably similar (Aβ40, 9.3 and 11 pmol/g/h; Aβ42, 0.7 and 1.5 pmol/g/h), but in both cases they were higher for plaque-bearing mice. Therefore, it seems possible that Aβ generation increases during plaque formation. Moreover, the presence of Aβ in CSF of plaque-bearing mice demonstrates its incomplete deposition and suggests even higher actual production rates.
Together with the Aβ reduction, acute γ-secretase inhibition caused a strong elevation of the APP C-terminal fragments from which the half-lives of C99 and C83 for degradation by γ-secretase were estimated to be approximately 0.4 and 0.1 h, respectively. The reason for this difference is not understood but may be related to different intracellular locations of both fragments. The Swedish mutation as present in APP23 mice is thought to elevate BACE1 cleavage and C99 generation in secretory compared with endocytic compartments (Haass et al., 1995). Future studies will have to show whether the same half-lives are obtained with wild-type APP transgenic mice. The continuous increase of C83 during the phase of γ-secretase inhibition indicated that this enzyme serves as its primary degradation pathway in brain neurons. In contrast, the decrease in C99 accumulation with time suggests that this fragment may undergo degradation by proteases different from γ-secretase or that its generation is reduced by feedback mechanisms. It also remains to be determined whether γ-secretase inhibition of wild-type mice, which express far less APP, leads to similar increases in APP CTFs.
Chronic inhibition of γ-secretase in a preventive setting, where APP23 mice had been treated between 6 and 9 months of age, resulted in an 80% reduction of the amyloid plaque area as well as that of all three Aβ peptides analyzed (Aβ1–38, 1–40, and 1–42). The plaque number was less reduced, consistent with a smaller effect on de novo plaque formation than on Aβ deposition. Comparable effects have been obtained in PDAPP mice (May et al., 2001). In a clinically more relevant therapeutic paradigm, plaque-bearing APP24 mice were treated from 15 to 17 months of age. Forebrain Aβ levels remained at baseline, whereas the amyloid load increased, yet approximately 30% less than in the vehicle group. Caudate/putamen, a region with a low plaque development, showed a much stronger reduction of the amyloid load than forebrain of the same mice, indicating that the efficacy differences were not age-related. Instead they inversely correlated with the initial amyloid load (Table 1), irrespective of the brain region and amyloid type. These results demonstrate that inhibition of Aβ generation can effectively slow down further amyloid formation in brains already containing plaques, even when the block of Aβ generation is not complete. The reduction is region-specific insofar because a strong effect can be expected for areas with a low amyloid load, whereas it may be much smaller in amyloid laden regions of the same brain. In agreement with our data, an approximate 50% reduction of plaques was observed after a 3-month treatment of aged Tg2576 mice with the γ-secretase inhibitor MRK-560 (Best et al., 2007). Our studies did not provide evidence for a clearance of pre-existing plaques in any brain region. Even chronic suppression of APP expression in transgenic mice during 6 months only halted further amyloid deposition (Jankowsky et al., 2005).
At 15 to 17 months, plaque formation just starts in pons/medulla oblongata of APP24 mice. In this region, the last γ-secretase inhibitor treatment acutely reduced Aβ1–38 and Aβ1–40 below baseline and vehicle levels similar to forebrain of preplaque animals. In contrast, Aβ1–42 increased in the vehicle group indicating deposition, which was partially reduced by the treatment. This supports the notion that Aβ42 initiates deposition in vivo and suggests some independence between the Aβ isoforms, at least initially. In line with these results, CSF Aβ from aged mice with a high brain amyloid load showed a similar reduction after the last γ-secretase inhibitor dose as in acutely treated preplaque mice. Therefore, CSF Aβ seems to be a suited marker in assessing treatment effects on Aβ generation in amyloid-bearing brain.
As suggested by the acute studies, chronic γ-secretase inhibition caused dramatic elevation of APP C-terminal fragments. Histological analysis of 3-month-old treated mice showed that CTFs accumulated primarily in nerve fibers, rather than in cell bodies. Toxic properties have repeatedly been attributed to these fragments (Lee et al., 2006), and their accumulation may lead to side effects. However, the major contribution to side effects of γ-secretase inhibitors is thought to originate from the interference with intestinal and lymphatic cell differentiation associated with the inhibition of Notch cleavage (Searfoss et al., 2003; Wong et al., 2004). Recently, the coat color changes have also been linked to the inhibition of Notch processing (Schouwey and Beermann, 2008). Interestingly, alterations in hair color have also been observed in a clinical study with γ-secretase inhibitor LY-450139 (http://www.alz.org/preventionconference/pc2007/releases/61107_130pm_latebreak.asp) (Siemers et al., 2007b).
Independent of the potential liabilities of γ-secretase inhibition, our study describes typical features of Aβ turnover and deposition in vivo, which are important to design studies for testing Aβ-lowering approaches. It demonstrates the suitability of APP transgenic mice as translational models for the evaluation of such strategies.
Acknowledgments
We thank Dr. Paolo Paganetti for antiserum NT11 and antibodies 25H10 and 29C12 and Albert Enz for the determination of LY-411575 compound concentrations.
Footnotes
-
Article, publication date, and citation information can be found at http://jpet.aspetjournals.org.
-
doi:10.1124/jpet.108.140327.
-
ABBREVIATIONS: Aβ, amyloid β; AD, Alzheimer's disease; APP, amyloid precursor protein; sAPP, soluble form of APP; BACE, β-site APP-cleaving enzyme; CSF, cerebrospinal fluid; CTF, C-terminal fragment; ELISA, enzyme-linked immunosorbent assay; GFAP, glial fibrillary acidic protein; MALDI-TOF, matrix-assisted laser desorption/ionization-time of flight mass spectrometry; PS, presenilin; TBS, Tris-buffered saline; BMS-299897, 2-[(1R)-1-[[(4-chlorophenyl)-sulfonyl](2,5-difluorophenyl)amino]ethyl]-5-fluoro-benzenepropanoic acid; LY-450139, (S)-2-hydroxy-3-methyl-N-[(S)-1-((S)-3-methyl-2-oxo-2,3,4,5,-tetrahydro-1H-benzo[d]azepin-1-carbamoyl)-ethyl]-butyramide; MRK-560, N-[cis-4-[(4-chlorophenyl)sulfonyl]-4-(2,5-difluorophenyl)cyclohexyl]-1,1,1-trifluoromethanesulfonamide.
-
↵ The online version of this article (available at http://jpet.aspetjournals.org) contains supplemental material.
- Received April 25, 2008.
- Accepted August 6, 2008.
- The American Society for Pharmacology and Experimental Therapeutics