β-Amyloid Peptide Variants in Brains and Cerebrospinal Fluid from Amyloid Precursor Protein (APP) Transgenic Mice

In this study, we report a detailed analysis of the different variants of amyloid-β (Aβ) peptides in the brains and the cerebrospinal fluid from APP23 transgenic mice, expressing amyloid precursor protein with the Swedish familial Alzheimer disease mutation, at different ages. Using one- and two-dimensional gel electrophoresis, immunoblotting, and mass spectrometry, we identified the Aβ peptides Aβ(1–40), -(1–42), -(1–39), -(1–38), -(1–37), -(2–40), and -(3–40) as well as minor amounts of pyroglutamate-modified Aβ (Aβ(N3pE)) and endogenous murine Aβ in brains from 24-month-old mice. Chemical modifications of the N-terminal amino group of Aβ were identified that had clearly been introduced during standard experimental procedures. To address this issue, we additionally applied amyloid extraction in ultrapure water. Clear differences between APP23 mice and Alzheimer disease (AD) brain samples were observed in terms of the relative abundance of specific variants of Aβ peptides, such as Aβ(N3pE), Aβ(1–42), and N-terminally truncated Aβ(2/3–42). These differences to human AD amyloid were also noticed in a related mouse line transgenic for human wild type amyloid precursor protein. Taken together, our findings suggest different underlying molecular mechanisms driving the amyloid deposition in transgenic mice and AD patients.

The major histopathological hallmarks of Alzheimer disease (AD) 2 are neuritic plaques and neurofibrillary tangles that are composed of aggregates of the amyloid-␤ (A␤) peptide and hyperphosphorylated Tau protein, respectively (1)(2)(3)(4)(5). A␤ peptides are generated from the amyloid precursor protein (APP) by two proteolytic enzymes called ␤and ␥-secretases (for reviews see Refs. 6 -8). The resulting A␤ peptides represent a heterogeneous group of peptides with different lengths. A␤  is the predominant form released from cultured cells (9) and in biological fluids, such as blood (10) and cerebrospinal fluid (CSF) (11). Longer forms ending at amino acid 42 are believed to be particularly important in the pathogenesis of AD due to a higher propensity to aggregate (12). A␤ accumulation in the AD brain likely starts many years before cognitive deficits become evident (13,14). The reported inverse association of in vivo cortical binding of Pittsburgh Compound-B and CSF A␤42 suggests that brain amyloid deposition and low CSF A␤42 are mechanistically related (13,15). Based on the amyloid cascade hypothesis stating a central role of A␤ in AD, several therapeutic strategies targeting A␤ and aiming for inhibition of disease progression are pursued. The more advanced of these potential therapeutic approaches include inhibition or modulation of cellular A␤ production and A␤-targeted immunotherapy (16,17).
Animal models of the human disease are important research tools to facilitate the study of the pathophysiology, and they have been widely used for testing potential therapeutic approaches. The APP23 mouse line (18) represents one out of several well established transgenic mouse models that display key features of the amyloid pathology observed in patients with Alzheimer disease (for review, see Ref. 19). Human APP751 carrying the Swedish double mutation (K670N/M671L) is overexpressed in APP23 mice under control of the murine Thy-1 promoter, which drives neuron-specific expression (20). At ϳ6 months of age, the first amyloid deposits are observed in heterozygous animals (18).
We present here a detailed immunoblot and mass spectrometry analysis of the different variants of A␤ peptides that can be detected after one-and two-dimensional electrophoretic separation in brain extracts and CSF from heterozygous APP23 mice at different ages. In addition to standard amyloid preparation with formic acid, extraction in pure water was applied for preparing amyloid fibrils in the absence of potentially modifying agents (21). Our study reveals clear differences between APP23 mice and human AD brain samples in terms of the relative abundance of specific variants of A␤ peptides. To explore whether the observed differences might be attributed to the expression and processing of a familial Alzheimer disease mutant APP form, we furthermore investigated brain extracts and CSF from APP51 mice that are transgenic for human wild type APP751 (22).

EXPERIMENTAL PROCEDURES
Transgenic Mice-The transgenic mice investigated express the human APP751 isoform under the control of the murine Thy-1 promoter. The mice were on a C57BL/6 background and were hemizygous for the transgene. Only male mice were included in this study. APP23 mice express APP with the KM670/671NL "Swedish" mutation (18), whereas APP51 mice express APP wild type (22). All animal experiments were in compliance with protocols approved by the Swiss Animal Care and Use Committees. CSF was collected from the cisterna magna under anesthesia with 3% isoflurane. Thereafter, the mice were decapitated, and the brains were removed, frozen on dry ice, and stored at Ϫ80°C until used.
Human Samples-The human brain tissue was obtained from The Netherlands Brain Bank, Netherlands Institute for Neuroscience, Amsterdam. All material was collected from donors for or from whom a written informed consent for a brain autopsy and the use of the material and clinical information for research purposes had been obtained by the Netherlands Brain Bank. Three of the five donors used for the study were neuropathologically diagnosed with Alzheimer disease and the other two were diagnosed as nondemented controls. The post-mortem delay ranged between 3 h, 40 min and 5 h, 30 min. Demographic data and neuropathological findings provided by the Netherlands Brain Bank are summarized in Table 1.
Amyloid Extraction from Brain Tissue-Brain samples were weighed and homogenized by sonication in 1 ml per 100 mg wet tissue of a triple detergent buffer (50 mM HEPES, pH 7.4, 150 mM NaCl, 1.0% Nonidet P-40 ("Igepal CA 630") (v/v), 0.5% sodium deoxycholate (w/v), 0.1% SDS (w/v), and "Complete " protease inhibitor mixture (Roche Applied Science), 1 tablet per 10 ml). After centrifugation (16,000 ϫ g, 15 min, 4°C), the supernatant was collected, and the protein concentration was determined by the BCA assay. The pellet was washed twice with PBS and resolubilized under shaking for 20 min at room temperature with 70% formic acid (150 l per 100 mg of wet brain tissue). After centrifugation (16,000 ϫ g, 10 min, 4°C), the supernatant was collected, and the formic acid was evaporated in a vacuum concentrator. The formic acid-soluble material was then resolved under shaking for 10 min at 95°C in 50 l of SDS sample buffer (0.36 M bis-Tris, 0.16 M Bicine, 1% SDS (w/v), 15% sucrose (w/v), and 0.0075% bromphenol blue (w/v); 50 l per 100 mg wet brain tissue). Insoluble material was removed by centrifugation, and the supernatant was collected as the formic acid fraction. In an additional control experiment, stable isotope-labeled synthetic A␤  containing Arg( 13 C 6 / 15 N 4 ) and Lys( 13 C 6 / 15 N 2 ) (ClearPoint, Anaspec, Fremont, CA) was added to the detergent-insoluble pellet before extraction with 70% formic acid. Alternatively, amyloid fibrils from brain samples were extracted essentially according to Pras et al. (21). Briefly, the brain tissue samples were first homogenized by sonication in 1 ml per 100 mg wet tissue of saline (150 mM NaCl, 2 mM Tris-HCl, pH 7.4, protease inhibitor mixture) to extract soluble proteins. The insoluble material was pelleted by centrifugation at 4°C and 16,000 ϫ g for 30 min, and the procedure was repeated until no further changes in the UV absorption at 280 nm of the supernatants were observed. The insoluble material was then homogenized in ultrapure water (1 ml per 100 mg wet tissue) and centrifuged as described above. Again, this step was repeated until no further amyloid material could be extracted as indicated by a stable UV absorption at 280 nm.
Negative Staining of Amyloid Fibrils-Two different protocols were used for negative staining of ␤-amyloid water extracts from an APP23 mouse brain sample. First, a drop of protein solution was diluted 1:1 with 0.2% glutaraldehyde in water and immediately incubated with a Formvar-coated 100 hexagonal mesh copper EM grid (Plano, Germany) for 15 s. After removing the fluid from the grid with a pointed piece of filter paper, the grid was rinsed shortly on three droplets of water and incubated 15 s with 2% (w/v) uranyl acetate (SPI Supplies). Subsequently, the fluid was again removed with pointed filter paper, and the grid was dried. In the second approach, 9 l of water extract were placed on an EM grid and incubated 5 min after adding 1 l of 100 mM CaCl 2 in water. After removing excess fluid with filter paper, the grid was stained for 5 min with 1% (w/v) phosphotungstic acid (Merck) and adjusted to pH 5.7 with 1 M NaOH according to Ref. 23. Grids were analyzed with a LEO EM912 Omega (Zeiss, Oberkochen), and digital micrographs were obtained with an on-axis 2048 ϫ 2048-CCD camera (Proscan, Scheuring).

TABLE 1 Demographic and neuropathological data for the human donors
The demographic data, apoE status, and neuropathological diagnosis according to Braak and Braak (37,55) were provided by the Netherlands Brain Bank, Netherlands Institute for Neuroscience, Amsterdam.

␤-Amyloid Peptide Variants in APP23 Mice
Bicine/bis-Tris/Tris/Sulfate SDS-PAGE-For the analysis of A␤ peptides the Bicine/bis-Tris/Tris/sulfate SDS-PAGE was used (24). To separate the different C-terminal A␤ peptide variants, the urea version of the gel system was applied (25,26). The percentage of total acrylamide plus bisacrylamide (w/v) in the separation gel was 10%, and gel thickness was 0.5 mm. The gels were run at room temperature for 50 min at a constant current of 25 mA/gel. Each gel carried a dilution series of a synthetic standard A␤ peptide mix.
Two-dimensional A␤-immobilized pH Gradient (IPG)-PAGE-IPG electrofocusing was performed with dry IPG strips (GE Healthcare) on an Ettan TM IPGphor TM II isoelectric focusing (IEF) system (GE Healthcare). IPG strips were rehydrated in rehydration solution (8.3 M urea, 2% CHAPS (w/v), 0.5% (v/v) Pharmalyte pH 3-10, 20 mM DTT) containing the sample for 10 h at 20°C. IEF was performed as described elsewhere (27). For the urea-free IEF version, the rehydration solution was modified to 4% CHAPS (w/v), 0.5% Pharmalyte (v/v) pH 3-10, 20 mM DTT. After focusing, IPG strips were placed on a 1-mm thick A␤-SDS-polyacrylamide gel and embedded by a low melting agarose solution (1% low-melt agarose (w/v), 0.36 M bis-Tris, 0.16 M Bicine, 0.25% SDS, 0.002% bromphenol blue (w/v)). A Teflon spacer was inserted next to the IPG strip to form a one-dimensional slot for synthetic A␤ standards. Gels were run at a constant current of 15 mA per gel for 10 min, followed by 25 mA per gel for 10 min and continued with 35 mA per gel until the running front reached the bottom of the gel. Strip holders that had been used for IEF of large amounts of A␤ peptides were subjected to a harsh cleaning procedure, including prolonged treatment with proteinase K to avoid cross-contaminations in subsequent experiments (28).
Immunoblot Analysis and Quantification-The separated A␤ peptides were transferred onto Immobilon-P PVDF membranes at 1 mA/cm 2 under semidry conditions (Hoefer Semi-Phor) for 30 min (0.5-mm thick gels) or 45 min (1.0-mm thick gels) (26). The PVDF membranes were then boiled for 3 min in PBS in a microwave oven to improve antibody binding (29). Immunostaining with the monoclonal A␤-antibodies 1E8 (Bayer Schering Pharma AG, Berlin, Germany) and 6E10 (Covance, Emeryville, CA) was performed as described before (30). For the detection of A␤(N3pE) peptides, the blots were blocked in RotiBlock (Carl Roth GmbH, Karlsruhe, Germany) and incubated overnight at 4°C with the polyclonal antibody amyloid ␤ (N3pE) (IBL, Hamburg, Germany) at a concentration of 0.5 g/ml. The next day, the blots were washed three times in 0.075% Tween in PBS and were probed with peroxidase-coupled goat anti-rabbit IgG (1:50,000 in PBS-T). All blots were developed with ECL-Plus TM chemiluminescent substrate according to the manufacturer's instructions (GE Healthcare) and detected with an Intas Imager (Intas, Göttingen, Germany). Band intensities were quantified with the Quantity One Software 4.1 (Bio-Rad) relative to a dilution series of A␤  present in a mixture of A␤ standard peptides. The mouse brain extracts were prediluted to reach final A␤ concentrations that were in the range of the A␤(1-40) calibration curve. Each mouse brain extract was analyzed on two separate blots, and the mean values were calculated from the duplicates.
Gel Staining-For Coomassie staining of two-dimensional gels, a modified version of the "Blue Silver" method (31) was used. Gels were first fixed for 1 h at room temperature in fixation solution (30% ethanol, 2% phosphoric acid) and then stained for 16 h at room temperature in Coomassie staining solution (0.12% Coomassie Brilliant Blue G-250 (w/v), 20% ethanol (v/v), 10% ammonium sulfate (w/v), 10% phosphoric acid (v/v)). The next day the gel was washed several times in 20% ethanol.
Mass Spectrometry of Intact A␤-For mass spectrometric analyses, the spots of interest were excised from Coomassiestained two-dimensional gels and extracted with a mixture of ammonia (25%)/water/isopropyl alcohol, 1:3:2 (v/v/v). For extraction, the thoroughly minced gel pieces suspended in 50 l of solvent were vortexed at 37°C for 1 h, ultrasonicated in an ultrasonication bath for 10 min, and finally centrifuged. After removal of the supernatant, the procedure was repeated with another 50 l of solvent, and the remainder was extracted with 50 l of ethanol. The combined extracts were concentrated and used for MALDI mass spectrometry in the linear mode on a Bruker Ultraflex TOF/TOF mass spectrometer. Therefore, the samples were applied to AnchorChip targets (Bruker, Germany) using the dried droplet method with 2,5-dihydroxybenzoic acid as the matrix.
Statistical Analysis-Statistical tests were performed with GraphPad Prism 5.01 software (GraphPad Software Inc. La Jolla, CA). A level of p Ͻ 0.05 was regarded as statistically significant; the actual p levels are given for descriptive purposes.

Age-dependent A␤ Accumulation in the Brains of APP23
Mice-Hemibrains from 3-, 7-, 15-, and 24-month-old APP23 transgenic mice were extracted first with an extraction buffer containing 0.1% SDS, 1.0% Nonidet P-40, and 0.5% sodium deoxycholate to prepare the detergent-soluble fraction. In a second step, the remaining insoluble material was extracted with 70% formic acid to obtain the formic acid fraction. The samples were separated by SDS-PAGE in the presence of urea and analyzed by immunoblot with the monoclonal antibody 1E8. On immunoblots, this antibody recognizes A␤ peptides starting with Asp-1 or Ala-2 (30). One-dimensional separation revealed five different forms of A␤ that were tentatively identified as A␤(1-37), -(1-38), -(1-39), -(1-40), and -(1-42) according to their electrophoretic mobilities (Fig. 1A). The amount of the A␤ peptides in the different fractions was estimated from the blots by semi-quantitative image analysis and in comparison with a standard curve obtained with synthetic A␤(1-40). Detergent-soluble A␤ was found in all tested brain samples, and its total amount increased substantially between 7 and 24 months of age. Detergent-insoluble A␤ was not detected in mice younger than 7 months. The total amount of A␤ in the formic acid fraction augmented dramatically between 7 and 24 months of age (Table 2).
In all A␤-positive detergent and formic acid fractions and independent of age, the most prominent of the different A␤ variants was A␤(1-40) (Fig. 1A). To evaluate the relative contribution of A␤(1-42), which is known to be particularly prone to aggregation (32), the A␤(1-42)/(1-40) ratio was calculated.
Additional A␤ peptide variants displaying shifts in their pI values were revealed by two-dimensional urea-SDS-PAGE followed by immunoblot analysis. Fig. 1B shows examples of twodimensional immunoblots of detergent and formic acid extracts from a 24-month-old APP23 mouse. Major A␤ variants detected by mAb 1E8 displayed a pI of ϳ5.4. In addition, A␤ peptides with estimated pI values of ϳ6.2 or ϳ4.7 were regularly observed in the formic acid fractions and, to a smaller extent, also in the detergent extracts from 24-month-old APP23 mice.
Mass Spectrometric Characterization of A␤ Variants in Formic Acid Fractions-To provide unequivocal identification of the A␤ peptides isolated from the brains of 24-month-old APP23 mice and to search for possible chemical peptide modifications, we applied different mass spectrometric methods. To this end, intact A␤ peptides were extracted from spots that were excised from Coomassie-stained two-dimensional gels (Fig. 1C). Compared with immunoblots, the general Coomassie staining revealed only one additional peptide spot (spot 5 in Fig.  1C) in the region of the two-dimensional gel where A␤ peptides were expected. The gel extracts were subjected to MALDI-TOF MS, which allowed the assignments shown in Table 3. Most of FIGURE 1. A␤ peptide variants in APP23 mice. A, representative A␤ peptide blots are shown. Samples are derived from detergent and formic acid fractions of APP23 mouse brains at different ages. A␤ peptides were separated by urea-SDS-PAGE followed by immunoblot with mAb 1E8. The total protein amount loaded for the different age groups was adjusted to obtain comparable signals on the blots (detergent fractions: 10 g for 3 months and 7 months, 500 ng for 15 months, and 100 ng for 24 months; formic acid fractions: 5 l of a 1:10, a 1:2000, or a 1:10,000 dilution for 7-, 15-, and 24-month-old mice, respectively). S, mixture of synthetic A␤ peptides. B, representative two-dimensional Western blots of detergent and formic acid fractions from the brain of a 24-month-old APP23 mouse. IPG strips with a linear pH gradient from 4 to 7 were used. The estimated pI values of the observed A␤ peptides are indicated. The figure shows those sections of the two-dimensional separations in which the A␤ peptides were resolved. C, Coomassie-stained two-dimensional gel of formic acid-extracted A␤ peptides from the brain of a 24-month-old APP23 mouse. The spots that were excised for mass spectrometric analysis are encircled. The numbers correspond to those in Table 3. a Brain tissue samples were subjected to A␤ urea-SDS-PAGE followed by immunoblot with the mAb 1E8. The blots were run in duplicate, and the A␤ peptides were quantified by comparison with a dilution series of synthetic A␤(1-40) standards. The values for the distinct A␤ variants were added to obtain total A␤. For each group, medians and 25-75% percentiles are shown. The statistical significance between age groups (rank sums) was analyzed with the nonparametric Kruskal-Wallis test and Dunn's multiple comparison post-test. The A␤(1-42)/(1-40) ratio in the detergent fractions from 24-month-old mice was statistically significantly lower than in 7-and 15-month-old mice (p Ͻ 0.01 for 24 months versus 7 months, and p Ͻ 0.05 for 24 months versus 15 months, respectively). In the formic acid fractions, the A␤(1-42)/(1-40) ratio was statistically significantly lower in 24-month-old animals than in 7-month-old mice (p Ͻ 0.001). This peptide migrated above all other A␤ spots, which is consistent with the finding that oxidized A␤ peptides have a retarded electrophoretic mobility under these conditions (33,34). Accordingly, A␤(1-40)ox* was most probably oxidized before the electrophoretic separation. Other A␤ peptides were presumably partially oxidized during experimental procedures, e.g. electrophoresis, fixation, staining, spot extraction, or MS analysis. The N-terminally truncated peptides A␤(2-40) and -(3-40) migrated as a single spot at an estimated pI of ϳ6.2 (spot 12 in Fig. 1C and Table 3). The shifted pI value can be easily explained by the absence of the aspartic acid residue Asp-1. Several additional peptide spots displayed a more acidic pI of ϳ4.7. Spot 5 migrated faster than all other peptides. It was detected by Coomassie staining but not on immunoblots probed with the mAb 1E8, which is directed against human A␤. For this peptide, we obtained a measured mass of 4235.8, which is in reasonable agreement with the theoretical mass of 4233.8 for endogenous mouse A␤ . The masses we observed for the remaining peptides displaying an acidic pI shift indicated the presence of modified forms of human A␤(1-40) and A␤ , with mass differences of ϩ28 or ϩ43 (Table 3). By mass spectrometric sequencing, the underlying chemical modifications could be unequivocally assigned to the N terminus of Asp-1 and were identified as formylation (ϩ28) and carbamoylation (ϩ43), respectively (supplemental Fig. S1). To further explore whether N-formylation and N-carbamoylation of A␤ had occurred artificially during the experimental procedures, we repeated the sequential amyloid extraction from the brain of a 24-month-old APP23 mouse and added stable iso-tope-labeled A␤ as an internal standard prior to the addition of formic acid. Formylation (ϩ28) and carbamoylation (ϩ43) of the N-terminal Lys-C fragment were found in both the human A␤ from the APP23 mouse and in the stable isotope-labeled internal standard (supplemental Fig. S2). Thus, the observed modifications had occurred at the stages of extraction (formylation by formic acid) and separation (carbamoylation by urea) and therefore have to be considered experimental artifacts. A fraction of the N-terminal Lys-C fragments from the pI ϳ4.7 A␤ was apparently unmodified.

␤-Amyloid
Characterization of Amyloid Prepared in Water-As a more gentle approach to the preparation of amyloid, we additionally employed a method that was first described by Pras et al. (21) for the isolation of amyloid fibrils from the spleen of a patient with primary amyloidosis. In brief, following the removal of proteins soluble in saline, the amyloid fibrils were extracted from APP23 mouse brain by repeated homogenization in deionized ultrapure water. Fig. 2A shows the UV absorptions of the saline and the water fractions and Fig. 2B the corresponding immunoblot analysis. As expected, A␤ peptides were found in the initial saline extracts, representing a highly soluble fraction. Only trace amounts of A␤ peptides were detected in the first extract prepared in ultrapure water, which presumably still contained substantial amounts of ions. Appreciable amounts of deposited A␤ peptides were finally extracted in the water fractions 2-5. This elution pattern for amyloid is in agreement with the original report by Pras et al. (21). One defining general feature of amyloid, irrespective of the specific amyloid protein under investigation, is the fibrillar ultrastructure revealed by electron microscopy (35,36). In the water extract from an APP23 mouse brain, we could observe negatively stained coiled fibrils by electron microscopy, confirming that the A␤ peptides in the water extract most probably represented fibrillar amyloid (Fig. 2C). Two-dimensional immunoblot analysis of the second water fraction revealed a pattern of A␤ variants that was similar to the formic acid fraction and contained N-terminally truncated (pI ϳ6.2) A␤ forms as well as variants displaying an acidic pI shift (pI ϳ4.7) (Fig. 2D). To further exclude the possibility of artificial carbamoylation, we subjected an aliquot of water extracted amyloid to isoelectric focusing in the absence of urea followed by SDS-PAGE (without urea) and immunoblot analysis. In the absence of urea, the C-terminal A␤ variants are not resolved in the second dimension (25,26). Again, a major fraction of A␤ with a pI of ϳ5.4 and minor amounts of additional A␤ peptides displaying either more alkaline or more acidic pI values were detected (Fig. 2E). Our attempts to determine the masses of the water-extracted A␤ peptides and/or to elucidate the exact nature of their N termini were not successful due to low yields and a number of co-purified and in part presumably amyloid-associated proteins that were detected with general protein stains, such as Coomassie Blue or Silver.
A␤ Patterns in APP23 Mouse Brains Exhibit Noticeable Differences to Human AD Brain Samples-For comparing the A␤ peptide variants in APP23 mice with human AD amyloid pathology, we subjected five human temporal lobe tissue samples to sequential extraction as described above. Three of the samples were from AD patients and two from age-matched nondemented controls with and without neuropathologically a The spots were numbered according to Fig. 1C. Several A␤ peptides were found with and without oxidation at methionine as indicated by a mass difference of 16. For the sake of clarity, however, only the observed masses of nonoxidized peptides are listed except for A␤(1-40)ox*, which was most likely oxidized prior to separation (see text). SEPTEMBER (Table 1). A␤ peptides were readily detected by immunoblot analysis in the detergent and formic acid fractions from all three AD samples and also in control sample NDC1 (amyloid stage B) (Fig. 3, A and B). The detergent and formic acid fractions from APP23 mice displayed fairly similar patterns of different A␤ variants. In both, A␤(1-40) was clearly the most abundant species with minor amounts of codeposited A␤(1-37), -(1-38), -(1-39), and - . In all three human temporal lobe samples from AD patients and in NDC1, the predominant detergent-soluble A␤ peptide was A␤(1-42) (Fig. 3A). The second most abundant A␤ variant in the AD samples co-migrated with synthetic A␤(2-42). The formic acid preparations from the human tissue samples displayed substan-tial inter-individual differences. In sample AD1, the presumed A␤(1-40) was the most prominent peptide followed by A␤ . In samples AD2, AD3, and NDC1, the most abundant A␤ peptide was A␤(1-42) and was accompanied by varying amounts of additional A␤ peptides co-migrating with synthetic A␤(1-40), A␤(2-42), or A␤(1-38) (Fig. 3B). The biochemical detection of appreciable amounts of A␤ in the control sample NDC1 confirms the neuropathological classification amyloid stage B (Table 1), which is characterized by medium densities of amyloid deposits in isocortical areas (37). In the second control, NDC2 (amyloid stage 0, see Table 1), A␤ peptides were below the detection sensitivity of direct immunoblotting (Fig. 3, A and  B). For comparing the detergent fractions from NDC1 and

␤-Amyloid Peptide Variants in APP23 Mice
NDC2, A␤ peptides were thus pre-enriched by immunoprecipitation prior to immunoblot analysis. Electrophoretic separation in the absence of urea indicated a comparatively low amount of total A␤ in NDC2, as expected (Fig. 3C). Most probably, this fraction represented a soluble pool of A␤ in this brain tissue sample devoid of amyloid deposits. Resolution of C-terminal A␤ variants by urea-SDS-PAGE revealed A␤  to constitute the predominant A␤ species in NDC2 (Fig. 3D). In contrast, in NDC1 the most abundant A␤ variant was clearly A␤(1-42), confirming the findings from direct immunoblot analysis (see above).
To provide further information regarding the relative abundance of N-truncated A␤ peptides in the APP23 mice as compared with an example of human AD amyloid pathology, we applied two-dimensional separation followed by immunoblotting with mAb 6E10. This antibody recognizes N-terminal A␤ variants from A␤  to A␤   (38). In the AD3 formic acid sample, a substantial proportion of the A␤ peptides displayed pI values greater than pH 6 suggesting N-terminal truncations. In contrast, only small amounts of corresponding N-truncated A␤ forms were observed in the murine sample (Fig. 4A). The most prominent N-terminal truncated A␤ peptide in human amyloid plaques starts with Glu-3 modified to pyrrolidone 2-carboxylic acid (pyroglutamate) (A␤(N3pE)) (39,40). A␤(N3pE) was readily detected by immunoblotting with a specific antibody in the detergent and formic acid fractions from the human AD samples and also in the extracts from sample NDC1 (Fig. 4B). The modified peptides displayed higher electrophoretic mobilities than the A␤ peptides starting at Asp-1, and synthetic A␤(N3pE40) was clearly resolved from A␤(N3pE42). Reprobing of the blot with antibody 6E10 confirmed that the pyroglutamate-modified peptides represented a major fraction in the human AD samples (Fig. 4B, bottom  panel). In the APP23 mice, we were able to detect multiple peptide bands with the A␤(N3pE) antibody in formic acid extracts from 15-and 24-month-old mice when relative large protein amounts were applied (Fig. 4C). For appreciable signals with the A␤(N3pE) antibody, ϳ1000 times more total protein was loaded on the gel than for detection of N-terminally unmodified peptides with antibody 1E8. Trace amounts of A␤(N3pE) were also observed in the detergent fractions from 24-month-old mice (data not shown). From the band intensities on the blots and by comparison with dilutions of a synthetic A␤(N3pE40) reference peptide, the amount of A␤(N3pE) peptides in the murine formic acid fractions was estimated to be ϳ50 Ϯ 14 pg/mg wet tissue (mean Ϯ S.D., n ϭ 4) at 15 months FIGURE 3. Comparison of A␤ patterns between human and APP23 brain samples. The relative abundance of different variants of A␤ in detergent and formic acid preparations from human temporal lobe samples and 24-month-old APP23 brains was analyzed by urea-SDS-PAGE/immunoblot. For comparison, the indicated synthetic A␤ peptides were loaded (S). AD1, AD2, and AD3, temporal lobe samples from three different AD patients; NDC1 and NDC2, temporal lobe samples from nondemented controls neuropathologically classified as amyloid stage B and 0, respectively. A, A␤ immunoblots of the detergent fractions probed with mAb 1E8. For the human samples 500 ng and for the murine samples 100 ng of total protein of each was loaded. B, A␤ immunoblots of the formic acid fractions probed with mAb 1E8. 5 l of a 1:1000 dilution (human samples) or 5 l of a 1:10,000 dilution (mouse samples) were loaded. C, A␤ peptides from detergent fractions of NDC1 and NDC2 (6 mg of total protein each) were immunoprecipitated with mAb 1E8 and separated by Bicine/bis-Tris/Tris/sulfate SDS-PAGE followed by immunoblotting with mAb 1E8. Note that in the absence of urea the N-and C-terminal A␤ variants migrate in one band. D, C-terminal A␤ variants in immunoprecipitates from NDC1 and NDC2 were separated by urea-SDS-PAGE and visualized by immunoblot with mAb 1E8. For the visualization of NDC1 a short exposure time and for NDC2 a short and a long exposure time are shown. S ϭ mixture of synthetic A␤ peptides. and ϳ469 Ϯ 138 pg/mg wet tissue (mean Ϯ S.D., n ϭ 8) at 24 months of age, respectively. Thus, A␤(N3pE) increased with age in parallel with total A␤ and accounted for roughly 0.1% of the total A␤ in the formic acid fractions.
A␤ Variants in Aged APP51 Mice Overexpressing Wild Type APP-The findings presented so far revealed clear differences in terms of the composition of A␤ deposits between APP23 mice and human AD pathology. To explore whether these differences can be primarily attributed to the Swedish mutant form of APP overexpressed in the APP23 mice, we extended our study to the closely related APP51 mouse line, which is transgenic for human wild type APP (22). In APP51 mice, amyloid plaques appear between 12 and 15 month of age, which is ϳ6 -9 months later than in the APP23 line (41). Detergent and formic acid brain fractions from 6-, 15-, and 24-month-old APP51 mice were prepared and analyzed as described for the APP23 mice. In pre-plaque, 6-month-old APP51 mice, only low levels of A␤ peptides could be detected. The total A␤ content increased dramatically between 15 and 24 months of age, resembling the exponential plaque growth observed in the APP23 line between ages 7 and 24 months (see above). Again, in detergent as well as in formic acid fractions, the predominant A␤ peptide was A␤(1-40) (Fig. 5A and Table 4). Similar to the findings in APP23 mice, the relative contribution of A␤  was comparatively high in the formic acid fractions at an early plaque state and decreased with ongoing amyloid deposition. The A␤(1-42)/(1-40) ratio (rank sums) in the formic acid fractions was significantly lower in 24-month-old mice than in 15-monthold animals (p Ͻ 0.01) according to the nonparametric Kruskal-Wallis test with Dunn's multiple comparison post-test. A two-dimensional separation of a formic acid brain extract from a 24-month-old APP51 mouse indicated minor amounts of N-terminal truncated A␤ peptides (pI ϳ6.2) (Fig. 5B). Based on semiquantitative image analysis from anti-A␤(N3pE) immunoblots (Fig. 5C), we estimate A␤(N3pE) to account for roughly 1% of total formic acid extracted A␤ in 24-month-old APP51 mice. This proportion is higher than in the APP23 line but still substantially below that in human AD samples (40).

␤-Amyloid Peptide Variants in APP23 Mice
A␤ Peptide Patterns in CSF-A␤ peptides in CSF from APP23 and related transgenic mice display a rapid turnover and appear to represent a pool of newly generated soluble A␤ in brain (41). In both transgenic mouse lines, patterns of 5 A␤ peptides were observed that were fairly similar to those reported for human CSF (11). In all cases, A␤  represented the predominant species and was accompanied by A␤ peptides co-migrating with synthetic A␤(1-37), -(1-38), - , and -(1-42) (Fig. 6). In the small number of CSF samples studied here, we did not observe statistically significant differences in the median total CSF A␤ levels or median A␤(1-42)/ (1-40) ratios between the tested age groups in either mouse line (Table 5). It is clear, however, that a substantially larger sample size will be required for an in-depth analysis able to reliably detect even subtle changes.
The overall mean total A␤ CSF levels in APP23 mice (independent of age) was 68 Ϯ 10 ng/ml (mean Ϯ S.D., n ϭ 14) and 52 Ϯ 18 ng/ml (n ϭ 14) in APP51 mice. Thus, the mean total A␤ CSF concentration in APP transgenic mice was about 6-fold higher in APP23 and about 5-fold higher in APP51 mice than in human CSF analyzed in a comparable fashion (34). Two-dimensional immunoblot analysis with mAb 1E8 did not indicate the presence of appreciable amounts of additional A␤ species with pI values other than ϳ5.4 (data not shown).

DISCUSSION
We report here a comprehensive comparison of A␤ peptides extracted from the brains of APP23 transgenic mice before and after onset of brain amyloid deposition and human temporal lobe samples from individuals with and without neuropathologically confirmed amyloid plaques. In the transgenic mice, a massive accumulation of A␤ was observed between 7 and 24 months of age. The time course of amyloid deposition involving a lag phase of several months is in line with the suggested mechanism of nucleation-dependent polymerization for amyloid formation (42). The most abundant form of A␤ in both detergent and formic acid fractions from the brains of APP23 mice at all stages analyzed was A␤ . Additionally, the A␤ variants FIGURE 5. A␤ peptide patterns in brains from APP51 mice overexpressing human wild type APP751. A, representative A␤ peptide blots of detergent and formic acid fractions from brains of 15-and 24-month-old APP51 mice. A␤ peptides were separated by urea-SDS-PAGE followed by immunoblot with mAb 1E8. The total protein amount loaded for the different age groups was adjusted to obtain comparable signals on the blots. B, two-dimensional immunoblot of a formic acid fraction from the brain of a 24-month-old APP51 mouse, probed with mAb 6E10. C, A␤(N3pE) peptides in formic acid fractions from the brains of 24-month-old APP51 mice. The blots were probed with the polyclonal anti-A␤(N3pE) antibody. FIGURE 6. A␤ patterns in CSF from APP23-and APP51 mice. CSF samples were separated by urea-SDS-PAGE followed by immunoblot with mAb 1E8. Representative blots for each age group from APP23 and APP51 mice are shown. A␤ , -(1-37), -(1-39), -(2-40), and -  and minor amounts of pyroglutamate-bearing (A␤(N3pE)) forms and endogenous murine A␤(1-40) were found to be co-deposited. A large excess of formic acid-soluble A␤40 over A␤42 in the cortex of 20-month-old APP23 mice has been reported before (43). The comparison of the amyloid preparations from APP23 transgenic mice with those from human AD brain samples revealed substantial differences in terms of the relative abundance of specific N-and C-terminal A␤ variants. In the human samples, A␤ , the presumed N-truncated A␤(2/3-42) and A␤(N3pE) peptides were much more prevalent than in the APP23 mice. N-truncated A␤ species were reported to account for more than 60% of the A␤ peptides in early and later stages of human AD amyloid pathology. Thus, N-truncated forms of A␤ ending at residue Ala-42 were proposed to be of particular importance in the development of AD neuropathology (44). In line with this hypothesis, in a human control brain sample devoid of amyloid plaques (amyloid stage 0), we observed a substantially different pattern of A␤ peptides than in the human brain samples classified as amyloid stages B or C. In the sample without plaques, the total amount of A␤ was comparatively small and presumably represented soluble A␤. The predominant A␤ variant was A␤ , which is consistent with A␤(1-40) representing the main soluble A␤ species in cell culture supernatants (9) and biological fluids (10,11). Thus, the soluble A␤ species appear similar in humans and transgenic mice. The low abundance of A␤ , N-truncated A␤(2/3-42), and A␤(N3pE) we observed in the amyloid deposits in aged APP23 transgenic mice suggests differences in the underlying biochemical mechanism driving the amyloid deposition in transgenic mice as compared with human AD brain. The A␤(1-42)/(1-40) ratio in the mice was found to be relatively high in the formic acid brain extracts by 7 months of age, which represents an early plaque stage. Kinetic in vitro aggregation studies have revealed that A␤(1-42) is more prone to aggregation than the shorter variants A␤(1-40) and A␤  and that amyloid formation by the kinetically soluble A␤  can be nucleated by A␤(1-42) (32). In APP24 transgenic mice, a related animal model overexpressing human APP carrying the Swedish K670N/M671L and the London V717I mutations, A␤(1-42) was shown to be the A␤ variant initially deposited (41). Our findings suggest that once the amyloid fibril formation has been nucleated (possibly by A␤ (1-42)), the A␤ accumulation in APP23 transgenic mice is mainly driven by the elevated production and high abundance of the five main A␤ peptides A␤(1-40), -(1-42), -(1-38), -(1-39), and -(1-37). Because of the large amounts of A␤ constantly being produced in the transgenic mice, the sustained A␤ deposition proceeds at a higher rate than in human AD brain, and differences between specific A␤ variants in their kinetic solubility or the ability to nucleate are thus probably less relevant. A correlation between interstitial A␤ concentration, as determined by in vivo microdialysis, and amyloid plaque growth has been demonstrated in APP/PS1 transgenic mice (45). Furthermore, in Tg2576 transgenic mice, the local interstitial A␤ concentrations in specific brain regions in young mice were recently reported to be associated with differences in the regional A␤ plaque burden in aged animals (46). In APP51 transgenic mice overexpressing human wild type APP, we observed patterns of deposited A␤ peptides fairly similar to those in the APP23 line. In both animal models, A␤(1-40) was the predominant variant at all stages of plaque development analyzed. N-truncated forms and A␤42 peptides were less abundant. Thus, the Swedish APP mutation does not provide an explanation for the observed differences in the brain A␤ patterns between aged APP23 mice and human AD pathology.
Two-dimensional gel electrophoresis indicated the presence of additional, modified A␤ peptides displaying acidic pI shifts in the amyloid preparations. We were able to identify N-formylation (ϩ28) and N-carbamoylation (ϩ43), which turned out to be artificially generated during the experimental procedures. These findings suggest that the primary amino group of A␤ is highly susceptible to artificial chemical modification and that extreme care must be taken during the analysis of A␤ peptide variants and potential post-translational modifications. A fraction of the A␤ peptides displaying an acidic pI shift was apparently unmodified, which might be explained by the following: (i) partially labile chemical modifications, (ii) the presence of conformational peptide variants (47) or oligomers, and (iii) tight interaction with distinct ampholyte molecules (48). Even when formic acid and urea were omitted from the work flow, these potential mechanisms also may have led to the appearance of residual A␤ peptides with low pI. As a relatively gentle approach to the preanalytical preparation of amyloid, we additionally employed "amyloid water extraction," a method originally introduced for the preparation of peripheral amyloid fibrils from a case of primary amyloidosis (20). The procedure has been successfully applied to the preparation of amyloid components from different sources (49 -54). However, to the best of our knowledge, it has not been reported before in the context of studies on AD-type amyloid. We confirmed the presence of typical amyloid fibrils in the water extract from an APP23 mouse brain by negative staining and electron microscopy. Immunoblot analysis revealed patterns of A␤ peptides

␤-Amyloid Peptide Variants in APP23 Mice
similar to those in detergent and formic acids extracts. Taken together, these observations suggest that water extraction represents an approach devoid of harsh chemicals for the preparation of amyloid fibrils from animal models of AD amyloid neuropathology. The A␤ peptides in CSF from APP23 transgenic mice undergo rapid turnover and appear to mirror newly generated soluble brain A␤ (41). Our semiquantitative immunoblot analysis indicated roughly 6-and 5-fold higher total CSF A␤ concentrations in APP23 and APP51 mice, respectively, compared with the reported levels in human CSF (34). In AD patients, CSF A␤42 concentrations are inversely correlated with in vivo amyloid imaging load (13,15). The selective reduction of A␤42 in CSF may thus represent a direct biomarker of A␤ deposition in human brain possibly reflecting the preferential deposition of A␤42. In the CSF samples from transgenic APP23 and APP51 mice analyzed here, we did not find evidence for statistically significant age-related or amyloid stage-related changes in the absolute A␤ CSF levels or in the relative distribution of the five C-terminal A␤ variants. It should be noted, however, that a larger sample size and additional time points will be required for an in-depth CSF analysis powerful enough to detect even subtle alterations.
Taken together, our findings reveal substantial differences in the A␤ peptide patterns between amyloid preparations from human AD temporal lobe samples and brains from two different APP transgenic mouse lines. The observations provide evidence for differences in the underlying molecular mechanisms driving the ongoing amyloid deposition in human AD brain and in the transgenic mouse models. This should be taken into consideration when translating observations from transgenic mice to the human disease.