Elevation of β-Amyloid Peptide 2–42 in Sporadic and Familial Alzheimer's Disease and Its Generation in PS1 Knockout Cells*

Urea-based β-amyloid (Aβ) SDS-polyacrylamide gel electrophoresis and immunoblots were used to analyze the generation of Aβ peptides in conditioned medium from primary mouse neurons and a neuroglioma cell line, as well as in human cerebrospinal fluid. A comparable and highly conserved pattern of Aβ peptides, namely, 1–40/42 and carboxyl-terminal-truncated 1–37, 1–38, and 1–39, was found. Besides Aβ1–42, we also observed a consistent elevation of amino-terminal-truncated Aβ2–42 in a detergent-soluble pool in brains of subjects with Alzheimer's disease. Aβ2–42 was also specifically elevated in cerebrospinal fluid samples of Alzheimer's disease patients. To decipher the contribution of potential different γ-secretases (presenilins (PSs)) in generating the amino-terminal- and carboxyl-terminal-truncated Aβ peptides, we overexpressed β-amyloid precursor protein (APP)-trafficking mutants in PS1+/+ and PS1−/− neurons. As compared with APP-WT (primary neurons from control or PS1-deficient mice infected with Semliki Forest virus), PS1−/− neurons and PS1+/+ neurons overexpressing APP-Δct (a slow-internalizing mutant) show a decrease of all secreted Aβ peptide species, as expected, because this mutant is processed mainly by α-secretase. This drop is even more pronounced for the APP-KK construct (APP mutant carrying an endoplasmic reticulum retention motif). Surprisingly, Aβ2–42 is significantly less affected in PS1−/− neurons and in neurons transfected with the endocytosis-deficient APP-Δct construct. Our data confirm that PS1 is closely involved in the production of Aβ1–40/42 and the carboxyl-terminal-truncated Aβ1–37, Aβ1–38, and Aβ1–39, but the amino-terminal-truncated and carboxyl-terminal-elongated Aβ2–42 seems to be less affected by PS1 deficiency. Moreover, our results indicate that the latter Aβ peptide species could be generated by a βAsp/Ala-secretase activity.

Identification of ␥-secretase is of crucial importance because this is the last step before amyloidogenesis and because ␥-secretase-like activities are also involved in the processing of other proteins such as Notch (15). Even if the ␥-secretase has not yet been formally identified, most of the existing in vivo and in vitro data based on presenilin (PS) knockout analysis, mutagenesis of specific amino acids, and drug targeting point to the narrow relationship between presenilins and an aspartyl ␥-secretase activity (15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25). PSs are proteins with multiple transmembrane domains, which are essentially located in the ER (26). The precise cleavage sites of the ␥-secretase and hence the generation of the shorter or longer A␤ peptide are widely believed to have important pathological consequences. Diseaselinked mutations in PS1, PS2, and APP result in an increase in production of A␤1-42, which is the major component of the ␤-amyloid plaques deposited in the brains of AD patients (27)(28)(29)(30). Targeted deletion of the PS1 gene resulted in a greatly decreased production of A␤ peptides 1-42 as well as A␤1-40 (16). The ␥-secretase cleavage constitutes an obvious target for disease prevention. The relationship between generation of a wide range of A␤ species (with a 2-or 3-amino acid variation at both the amino-or carboxyl-terminal ends; see Fig. 1A) and the existence of potential different ␥-secretases and/or PS is not clear (31)(32)(33).
The location of the secretase activities within the cell has only been partially established. ␣-Cleavage occurs at the plasma membrane and also potentially occurs in intracellular post-Golgi compartments (13, 34 -37). ␤-Secretase is active in acidic, nonlysosomal compartments (38). There is some evidence that the carboxyl termini of A␤1-40 and A␤1-42 are generated in different cellular compartments (39,40).
Mass spectroscopy has indicated the existence of additional A␤ peptide species besides A␤1-40 and A␤1-42 (41)(42)(43). However, the absolute and relative quantities of these additional A␤ peptides are not known, nor has it been investigated whether they are regularly produced.
We recently established one-and two-dimensional electrophoretic methods 2 for high resolution expression profiling of APP metabolites (44,45) that are based on the N,NЈ-bis-(2hydroxyethyl)-glycine/Tris urea separation gels of Wiltfang et al. (47). The separation principle relies on conformational shifts of A␤ peptides in the presence of urea, which are specific for distinct A␤ peptide species (Fig. 1B). Interestingly, we observed a structure-function analogy in vivo and in vitro because the solubility of A␤ peptide species in vivo, e.g. increased aggregation of amino-terminal-truncated and carboxyl-terminalelongated A␤ peptides, was closely and positively correlated with the electrophoretic mobility of A␤ peptide species in vitro. By using monoclonal antibody 1E8, which is specific for the first 2 amino acids of the A␤ peptide amino terminus (Fig. 1A), we realized a detection sensitivity in the subpicogram range. 2 Here we combined immunoprecipitation (1E8) with the refined A␤ SDS-PAGE/immunoblot, 2 which allowed us to study the secretion of A␤ peptides into the supernatants of neuronal cell cultures and a neuroglioma cell line. In addition, we studied the APP metabolite profile in brain homogenates and the CSF of patients with AD, patients with non-AD dementias (nADs), and patients with various nondementive neuropsychiatric diseases (ONDs).
A comparable and highly conserved pattern of A␤ peptides, namely, A␤1-40/A␤1-42 and carboxyl-terminal-truncated A␤1-37, A␤1-38, and A␤1-39, was found. Besides A␤1-42, we also observed a striking and consistent elevation of aminoterminal-truncated A␤2-42 in the brains of subjects with AD. Because A␤2-42 was enriched in the detergent-soluble fraction of AD brain homogenates, it did not originate from ␤-amyloid plaque core. A␤2-42 was also specifically elevated in CSF samples of AD patients. To decipher the contribution of potential different ␥-secretases (presenilins) in generating the amino-terminal-and carboxyl-terminal-truncated A␤ peptides, we overexpressed APP-trafficking mutants in PS1ϩ/ϩ and PS1Ϫ/Ϫ neurons. As compared with APP-WT (primary neurons from control or PS1-deficient mice infected with Semliki Forest virus), PS1Ϫ/Ϫ neurons and PS1ϩ/ϩ neurons overexpressing APP-⌬ct (a slow-internalizing mutant) or APP-KK (APP mutant carrying an ER retention motif) show a decrease of all secreted A␤ peptide species. Surprisingly, A␤2-42 is significantly less affected in PS1-and endocytosis-deficient (APP-⌬ct) neurons.
Taken together, our results indicate that the amino-terminal-truncated and carboxyl-terminal-elongated A␤2-42 was generated within a post-ER secretory pathway, where most of the A␤2-42 is generated by a ␤ Asp/Ala -secretase activity, possibly as alternative BACE activity. The data confirm that PS1 is an essential ␥-secretase protein/cofactor responsible for the production of A␤1-40/A␤1-42 and the carboxyl-terminal-truncated A␤1-37, A␤1-38, and A␤1-39, but not for the aminoterminal-truncated and carboxyl-terminal-elongated A␤2-42.

Patients
We investigated a total of 104 patients by A␤ SDS-PAGE/immunoblot. The dementia group included 51 patients with probable AD (average age, 70.9 Ϯ 9.8 years) and 12 patients with dementia of other etiology (nAD group; average age, 53.9 Ϯ 20.1 years). Moreover, we investigated 41 patients with various ONDs (average age, 42.5 Ϯ 15.5 years). The patient groups nAD and OND can be summarized as neuropsychiatric disease controls (n ϭ 53). Neuropsychiatric diagnosis was established by International Classification of Diseases 10 (ICD10) and Diagnostic and Statistical Manual of Mental Disorders IV (DSM-IV) criteria (48). Patients with probable AD had to satisfy DSM-IV criteria for dementia of the Alzheimer's type and the National Institute of Neurological and Communicative Disorders and Stroke (NINCDS) and Alzheimer's Disease and Related Disorders Association (ADRDA) criteria (49).

Neuropathological Diagnosis and Brain Samples
Brains were fixed in 10% formalin for at least 14 days for neuropathological studies. Representative blocks of cortex, cerebellum, and brainstem were cut and embedded in paraffin. Paraffin sections (4 m) were stained with hematoxylin and eosin and with Bielschowsky's silver impregnation methods. Immunohistochemistry was performed with monoclonal antibodies against ␤-amyloid (clone 6F/3D; Dako, Glostrup, Denmark), phosphorylated tau (AT8; Innogenetics, Ghent, Belgium), and ␣-synuclein (15G7; Connex, Munich, Germany) on paraffin-embedded sections (4 m). Neuropathological diagnosis was performed using international criteria. AD-related pathology was classified using the Consortium to Establish a Registry for Alzheimer's Disease (CERAD) pathologic criteria based on semiquantitative analysis of neuritic plaques (Bielschowsky stain and ␤A4 immunostain) as well as the Braak and Braak classification based on the distribution of neurofibrillary tangles and neuropil threads (Bielschowsky stain and AT8 immunostain) (50,51). Distribution and frequency of Lewy bodies (LBs) were evaluated according to the Consensus Criteria for diagnosing dementia with LBs (52). The number of LBs was counted in the brain regions determined in the protocol (␣-synuclein immunostain) and converted into scores of A (no LBs), B (1-4 LBs), and C (Ͼ5 LBs) for each area. Based on the total score, dementia with LB cases were divided into three subtypes: brainstem dominant, limbic, and neocortical. Diagnosis of Pick disease and dementia lacking distinctive histopathology was made according to Cooper et al. (53). Diagnosis of tangle predominant form of senile dementia was made according to Jellinger and Bancher (54). Frozen brain tissue (frontal cortex and cerebellum) of the neuropathologically examined cases was used for immunoprecipitation with one-and two-dimensional A␤ SDS-PAGE/immunoblot.

Neuronal Cell Culture and Neuroglioma Cell Line
Generation of PS1 knockout mice has been described previously (16). For mouse neuronal primary cultures, 14-day-old embryos were taken from heterozygote (PS1ϩ/Ϫ) breedings. Total brains were dissected and trypsinized. Cells were resuspended, plated on 95-cm 2 tissue culture dishes (Nunc) precoated with 1 mg/ml poly-L-lysine (Sigma), and finally incubated in minimum Eagle's medium (Life Technologies, Inc.) supplemented with 10% horse serum medium (Perbio Sciences). After 4 h of culture, cells were incubated in neurobasal medium (Life Technologies, Inc.) supplemented with the B27 complement mixture (Life Technologies, Inc.) for 24 h before the addition of 5 M cytosine arabinoside for inhibition of glial cell growth.

Semliki Forest Virus (SFV) Constructs
The cDNA coding for human APP695 and containing a Myc tag 3 amino acids after the signal sequence (inserted at the KpnI site) cloned in pSP65 was kindly provided by Drs. P. Tienari and K. Beyreuther (Zentrum fü r Molekulare Biologie der Universitat Heidelberg, Heidelberg, Germany). This construct either (i) had the last 43 amino acids of the cytoplasmic tail deleted (from Tyr 653 to Asn 695 , APP-⌬ct) as de-scribed in Ref. 56 or (ii) was modified by site-directed mutagenesis (Stratagene) to add a di-lysine motif allowing its retention in the ER at position 692 and 693 (QM mutated to KK, APP-KK). To obtain SFVs containing those sorting mutants, all constructs were moved to pSFV-1 vector (provided by Dr. C. Dotti, EMBL) and linearized with SpeI. mRNAs were produced using the SP6 polymerase run-off transcription, and a mixture of mRNA from APP constructs and from pSFV-helper were cotransfected in BHK cells by electroporation to obtain recombinant SFV. BHK cells were grown in Dulbecco's modified Eagle's medium/F-12 (Life Technologies, Inc.) supplemented with 5% fetal calf serum (Perbio Sciences), 2 mM L-glutamine, 100 units/ml penicillin, and 100 mg/ml streptomycin. The cell culture medium (7.5 ml) containing infective recombinant SFV was collected 24 h after transfection, spun at 3500 rpm for 15 min, aliquoted (500 l), frozen directly in liquid nitrogen, and stored at Ϫ70°C until use. Each batch of virus preparation was analyzed for APP expression levels, and only batches that gave similar levels of expression were used in subsequent experiments.

Viral Infection and Metabolic Labeling
Mouse primary neuronal cultures were incubated in neurobasal medium containing SFV (100 l of stock solution) expressing the pSFV-1 plasmid bearing APP-WT, APP-⌬ct, or APP-KK. After 1-h infection at 37°C, cells were incubated in neurobasal medium without viruses for 2 h at 37°C. Cells were then metabolically labeled for 4 h at 37°C by incubation in methionine-free minimum Eagle's medium (Life Technologies, Inc.) supplemented with the B27 complement mixture and containing 200 Ci of [ 35 S]methionine (ICN). At the end of the incubation, culture media were recovered, and cells were washed once with PBS and finally lysed in DIP buffer (20 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% Triton X-100, 1% sodium deoxycholate, and 0.1% SDS). Cell extracts and media were spun for 10 min at 14,000 rpm in a microfuge (Eppendorf) to remove DNA and cell debris.
Cells extracts and cell media (supernatants) were immunoprecipitated by mixing them with 25 l of protein G-Sepharose and with an antibody against the APP ectodomain (diluted 200-fold; generously given by Dr. Greenberg, Cephalon Inc.). After an overnight incubation on a rotating wheel at 4°C, immunoprecipitates were washed four times with DIP buffer and washed once with PBS/water (1:3). Samples were finally solubilized in 25 l of Tris-tricine sample buffer (450 mM Tris-HCl, pH 8.45, 12% glycerol, 4% SDS, 0.0025% Coomassie Blue G, and 0.0025% phenol red) and boiled for 10 min before loading on gel. SDS-PAGE was performed on precasted 10 -20% acrylamide Tristricine gels (Novex). At the end of the run, gels were fixed in methanol/ water (1:1) for 2 h at room temperature and dried (Bio-Rad, type 583). Radiolabeled material was detected using a PhosphorImager (Molecular Dynamics), and quantification was performed by ImagQuaNT 4.1.

Preparation of Samples
CSF Samples-After obtaining informed consent from patients and/or family members, CSF (3-10 ml) was drawn by lumbar puncture from patients and sampled in polypropylene vials. After centrifugation (1000 ϫ g, 10 min, 4°C), CSF samples were processed within 12 h, and aliquots of 150 l were stored at Ϫ80°C for subsequent one-and two-dimensional A␤ SDS-PAGE/immunoblot.
Immunoprecipitation of Cell Culture Media-Cell culture media were removed after 24 h of incubation and analyzed by immunoprecipitation for A␤ peptides. For immunoprecipitation, 400 l of media were added to 100 l of 5-fold concentrated RIPA detergent buffer (5ϫ RIPA ϭ 2.5% Nonidet P-40, 1.25% sodium deoxycholate, 0.25% SDS, 750 mM NaCl, 250 mM HEPES, and 1 tablet of protease inhibitor mixture Complete TM Mini per 2 ml of 5ϫ RIPA, pH adjusted to 7.4 with NaOH) and 25 l of magnetic microparticles (DynaBeads, Dynal, Germany) coated with monoclonal antibody 1E8 (1 g of mAb 1E8/1.68 ϫ 10 7 beads). The protease inhibitor mixture Complete TM Mini was obtained from Roche Molecular Biochemicals, and mAb 1E8 was provided by T. Dyrks. Samples were incubated under rotation for 15 h at 4°C. Beads were washed four times with PBS/0.1% bovine serum albumin and one time with 10 mM Tris-HCL, pH 7.4. For A␤ SDS-PAGE/immunoblot, bound A␤ peptides were eluted by heating the sample to 95°C for 5 min with 25 l of sample buffer (0.36 M bis-Tris, 0.16 M N,NЈ-bis-(2-hydroxyethyl)-glycine, 1% (w/v) SDS, 15% (w/v) sucrose, and 0.004% (w/v) bromphenol blue). In case of A␤ IPG-2D-PAGE/immunoblot, A␤ peptides were eluted in the presence of 25 l of 40% (v/v) formic acid by sonification for 10 min at 37°C. Finally, the formic acid was removed by evaporation (SpeedVac, 40°C).
Immunoprecipitation of Brain Homogenates-The RIPA-soluble fraction of A␤ peptides was extracted from homogenates of the frontal lobe and cerebellum of each patient as follows: brain samples (ϳ50 mg) were homogenized in 1.5-ml polypropylene tubes in the presence of 1 ml of 1ϫ RIPA by ultrasonification (2 ϫ 25 strikes, output 20%, Bronson sonifier 250). After centrifugation (20,000 ϫ g, 5 min, 4°C), the protein content of supernatants was adjusted to 3 mg/ml by dilution with 1ϫ RIPA. Protein concentration was determined by the BCA assay (57). 50 l of the magnetic microparticles (2 g mAb 1E8/3.36 ϫ 10 7 beads) was added to 1 ml of homogenate. Except for two additional washes with 1ϫ RIPA, immunoprecipitation was performed as described previously. For immunoprecipitation with mAb 6E10 (Senetek, St. Louis, MO), we pooled the brain homogenates of eight AD patients (3 mg/ml protein content). 50 l of the magnetic microparticles (2 g mAb 6E10/3.36 ϫ 10 7 beads) was added to 1 ml of homogenate, and immunoprecipitation was performed as described for mAb 1E8.
One-dimensional A␤ SDS-PAGE-For the separation of A␤ peptides, we applied the urea version of the N,NЈ-bis-(2-hydroxyethyl)-glycine/ bis-Tris/Tris/sulfate SDS-PAGE of Wiltfang et al. (47). This system was used for the separation of A␤ peptides for the first time and without further modification by Klafki et al. (44). Due to urea-induced differential shifts in conformation, A␤ peptides that differ in only 1-2 amino acids can be separated (45). The composition of the separation gel initially applied for the analysis of A␤ peptides was modified from 15% T%/5% C/8 M to 12% T%/5% C/8 M urea (T% ϭ percentage (w/v) of total acrylamide monomer; C% ϭ percentage (w/w) of bisacrylamide/total acrylamide monomer), and gel thickness was reduced to 0.5 mm. 2 Gels were run at room temperature for 2 h at a constant current of 12 mA/gel, using the MiniProtean II electrophoresis unit (Bio-Rad). For analysis of CSF, an equivalent volume of sample was added to sample buffer (0.12 M bis-Tris, 0.053 M N,NЈ-bis-(2-hydroxyethyl)-glycine, 5% sucrose, 0.5% SDS, and 0.0025% bromphenol blue containing 1 tablet of proteinase inhibitor mixture Complete TM Mini per 10 ml), which had been dried before in Eppendorf vials by vacuum evaporation. Subsequently, samples were vortexed until complete solubilization of the sample buffer and heated at 95°C for 5 min after the addition of ␤-mercaptoethanol to a final concentration of 2.5% (v/v). 10 l of sample was loaded per lane. All samples were run as quadruplicates, and each gel carried a dilution series of synthetic A␤ peptides. Mean values were used for subsequent calculations.
Two-dimensional A␤ IPG-2D-PAGE-A␤ IPG-2D-PAGE was done as described by Wiltfang et al. 2 For IPG, we used dry strips (linear pH gradient, 4 -7; length, 7 cm) according to the protocol of the manufacturer (Amersham Pharmacia Biotech). Dry strips were rehydrated overnight at room temperature to 0.5-mm gel height using the following rehydration solution: . 30 l of sample in IPG sample buffer was applied to the rehydrated dry strip at pH 6.5 (cathodic site) using sample cups, and isoelectric focusing was performed for 30 min/300 V, 30 min/800 V, 30 min/1400 V, and 5 h/2000 V (⌺ 12,500 voltageϫhours). Subsequently, the strips were equilibrated for the second analytical dimension (A␤ SDS-PAGE) in the following buffer for 10 min at room temperature: 6 M urea, 20% (w/v) glycerol, 2.0% (w/v) SDS, 0.36 M bis-Tris, 0.16 M bicin, and 1.0% (w/v) dithiothreitol. Dithiothreitol was added just before equilibration. Equilibrated IPG strips were placed on top of the A␤ SDS-PAGE stacking gel (1-mm thick) and embedded by a low gelling temperature agarose solution (1.0% (w/v) agarose, 0.16 M bicin, 0.36 M bis-Tris, 0.25% (w/v) SDS, and 0.002% (w/v) bromphenol blue). Next to the IPG strip, a Teflon tooth was inserted to form a track for synthetic standard A␤ peptides or a one-dimensional reference separation of the SDS/heat-denatured sample. Subsequently, A␤ SDS-PAGE/immunoblot was performed as described, but separation gels were run at constant voltage for 15 min/60 V and for 1 h and 30 min/120 V.
Western Blotting, Immunostaining, and Quantification-Western blotting, immunostaining, and quantification were performed as described by Wiltfang et al. 2 A␤ peptides were transferred for 30 min at 1 mA/cm 2 and room temperature under semidry conditions (Hoefer Semi-phor) onto Immobilon-P PVDF membranes according to Wiltfang et al. (45).
For immunostaining, Immobilon-P PVDF membranes were washed for 30 s in double distilled H 2 O and boiled for 3 min in PBS using a microwave. Blocking was performed for 1 h at room temperature in the presence of RotiBlock. Incubation with primary mAb 1E8, which was diluted 4000-fold (stock, 0.25 mg/ml), was done overnight at 4°C. For mAb 6E10 as primary antibody (dilution, 1000-fold; stock, 0.1 mg/ml), membranes were blocked in 2.5% nonfat dry milk in PBS-T. After a brief wash in PBS-T (0.075% (v/v) Tween 20), membranes were further washed for 30 min, 15 min, and 2 ϫ 10 min. Next, membranes were incubated for 1 h at room temperature with an anti-mouse biotinylated IgG (H ϩ L) antibody (1.5 mg/ml), which was diluted 3000-fold in PBS-T. A second PBS-T wash was done for 3 ϫ 10 min at room temperature. The membranes were then incubated for 1 h at room temperature with streptavidin biotinylated horseradish peroxidase complex diluted 3000-fold with PBS-T. After a wash for 3 ϫ 10 min at room temperature, the membranes were developed for 5 min at room temperature with ECLPlus TM solution according to the protocol of the manufacturer. Detection of the emitted light signal was performed by a charge-coupled device camera (FluorSMax MultiImager; Bio-Rad), using a series of 1, 5, 20, 60, 120, and 300 s for data acquisition. Band intensities were quantified relative to an internal dilution series of the A␤ peptide standard mix using Quantity One software (Version 4.1; Bio-Rad). Detection sensitivity (mAb 1E8) was 0.6 and 1 pg for A␤1-40 and A␤1-42, respectively (data not shown). Signal acquisition was linear within a range of 3.8 orders of magnitude. The high detection sensitivity was due to mAb 1E8 and an optimization of the former immunoblot procedure (45) because the latter mAb was compatible with a synthetic reagent (Roti-Block) instead of nonfat milk powder for blocking the PVDF membrane. The method allowed quantification of A␤ peptides in only 10 l of CSF. The inter-and intra-assay coefficients of variation for 80 and 20 pg of synthetic A␤ peptides were below 10%.

IR-MALDI Mass Analysis and Gas-phase Sequencing
For mass analysis and generation of amino-terminal sequence tags, RIPA-soluble A␤ peptides from the frontal lobe (1 g, wet weight) of AD patient 638 were immunoprecipitated with 1 ml of DynaBeads (20 g of mAb 1E8, 6.7 ϫ 10 8 beads) as described. The sample was split into two aliquots, which were separated by A␤ SDS-PAGE and blotted onto PVDF membranes (Immobilon-P; Millipore). Five blot lanes of each PVDF membrane were used for subsequent IR-MALDI mass analysis and amino-terminal sequence tags. For this application, the semidry Western blot procedure was modified because an additional PVDF membrane was inserted at the cathodic side of the separation gel. The stack of cathodic PVDF membrane, gel, and anodic PVDF membrane was pinhole-perforated at three sites. This guaranteed exact realignment of both membranes after the stack had been dismantled. Before electroblotting, the stack was left for 5 min to allow cathodic contact blotting. For subsequent IR-MALDI mass analysis, one of the two anodic PVDF membranes was incubated in an aqueous matrix solution (0.3 M succinic acid) for 20 min at room temperature directly after the Western blot, while the membrane was still wet. This was followed by slow drying at room temperature. The other anodic PVDF membrane was washed for 3 ϫ 5 min with double distilled H 2 O, dried, and used for subsequent gas-phase sequencing. Meanwhile, the cathodic PVDF membrane was immunostained (mAb 1E8) as described above but developed with diaminobenzidine (peroxidase substrate kit; Vector Laboratories) instead of enhanced chemiluminescence. This membrane served as a template to excise the areas of the anodic PVDF membrane corresponding to A␤1-42 and the unknown A␤ peptide.
IR-MALDI mass analysis (wavelength, 2.94 m; spot, 100 m; pulse, 90 ns) was performed directly from the pieces of matrix-embedded PVDF membrane (58). For gas-phase sequencing of electroblotted A␤ peptides, spots were excised from the PVDF membrane and applied to a Procise cLC protein sequencer (Applied Biosystems). For amino-terminal sequencing, standard protocols were used according to the manufacturer's instructions.

Statistics
Groups were characterized by mean values and S.D. Group differences were tested for significance (p Ͻ 0.05, two-sided level) by the Mann-Whitney U test (adjusted for small sample size) and, in case of dichotomic variables, by 2 analysis according to Fisher's exact test. Computations were performed using the statistical software package Statistica for Windows, Version 5.1 F.

RESULTS
High Resolution Analysis of A␤ Peptides-Whereas A␤ peptides migrate as a single species by conventional A␤ SDS-PAGE, the urea-based separation gels resolve a complex but highly conserved A␤ peptide pattern. By A␤ SDS-PAGE/immunoblot, we were able to demonstrate that in addition to A␤ peptides 1-40 and 1-42, the carboxyl-terminal-truncated spe-FIG. 1. Amyloidogenic processing of human APP-generating amino-terminal-and carboxyl-terminal-truncated A␤ peptide species. A, schematic representation of the human ␤-amyloid precursor protein and the amino-terminal-and carboxyl-terminal-truncated A␤ peptides analyzed in this study using monoclonal antibody 1E8 (bar, recognizing the first 2 amino acids of A␤ peptides). The site of proteolytic activity responsible for the amyloidogenic (␤-and ␥-secretase) and nonamyloidogenic pathway (␣-secretase) is depicted. ␤-Secretase was recently identified as BACE (4). ␥-Secretase activity is dependent on presenilins (reviewed in this study). B, synthetic A␤ peptides migrate as a single band during conventional SDS-PAGE, whereas by urea-based separation gels (A␤ SDS-PAGE), the carboxyl-terminal-truncated A␤ peptides 1-37, 1-38, and 1-39 and the amino-terminal-truncated A␤2/3-42 were resolved in addition to A␤1-40/42. A␤2-42 and A␤3-42 migrate as a single band, but A␤3-42 was not detected by Western immunoblot due to the amino-terminal specificity of mAb 1E8 (see A). We observed a structure-function analogy in vitro and in vivo because the amyloidogenic potential of the single A␤ peptide species in vivo is closely correlated to its urea-induced conformational shifts in vitro. cies 1-37, 1-38, and 1-39 are regularly found in human CSF (Fig. 1B). This A␤ peptide quintet was observed in all human CSF samples (n Ͼ 500) investigated thus far ( Fig. 2A, lane 4).
An Additional Amino-terminal-truncated A␤ Species in Cell Secretions and Patients with Alzheimer's Disease-Surprisingly, in neuroglioma H4 cells transfected with human APP carrying the Swedish double mutation and in primary mouse neurons virus-transfected with human APP, a minor A␤ peptide species became prominent in addition to the A␤ peptide quintet 1-37, 1-38, 1-39, 1-40, and 1-42 ( Fig. 2A, lanes 2 and  3) migrating below (anodically) A␤1-42. Interestingly, in a subset of AD patients, a reduced CSF concentration of A␤1-42 was paralleled by a detectable level of the additional A␤ peptide species (Fig. 2A, lane 6), whereas other AD patients presented only reduced CSF levels of A␤1-42 ( Fig. 2A, lane 5). In contrast, in the CSF of nAD or OND patients, A␤1-42 was usually not decreased, and the additional A␤ peptide species was not detectable (Fig. 2A, lane 4). Moreover, the additional A␤ peptide species was highly abundant in the detergent (RIPA)-soluble fraction of brain homogenates from the frontal lobe of patients with Alzheimer's disease (Fig. 2B, lane 2*).
Identification of the New A␤ Peptide as A␤2-42-RIPA-soluble A␤ peptides from the frontal lobe of a patient with sporadic AD were immunoprecipitated (1E8), separated by A␤ SDS-PAGE, and blotted onto PVDF membranes as described previously. IR-MALDI mass analysis was performed directly from the pieces of matrix-embedded PVDF membrane and yielded a molecular weight of 4523.3 Ϯ 7 for the PVDF area corresponding to the R F value of synthetic A␤1-42 (M r 4514.14). The PVDF area containing the unknown A␤ peptide yielded a major peak with a M r of 4405.6 Ϯ 10 and a minor peak (Ͻ10%) with a M r of 4334.5 Ϯ 12, corresponding most probably to A␤ peptides 2-42 (M r 4399.01) and 3-42 (M r 4327.93), respectively.
Gas-phase sequencing of the PVDF membrane area corresponding to the R F of synthetic A␤1-42 yielded a major sequence of DAEFRHDSGY, corresponding to A␤1-x, and a minor sequence (Ͻ10%) of AEFR, corresponding to A␤2-x. The PVDF area corresponding to the R F value of the unknown A␤ peptide migrating anodically of A␤1-42 gave the sequence AEFRHDS, corresponding to A␤2-x.
Taken together, IR-MALDI mass analysis and gas-phase sequencing identified the unknown A␤ peptide migrating anodically of A␤1-42 as A␤2-42. According to IR-MALDI mass analysis, a small amount of A␤3-42 was coimmunoprecipitated with A␤2-42, which was not detected by gas-phase sequencing. The minor amount of A␤2-x comigrating with A␤1-42 most probably corresponded to a contamination with A␤2-42.
Two-dimensional electrophoretic separation of the latter A␤ peptides showed that synthetic A␤ peptides 2-40/42 share a shift of their isoelectric points of one pH unit (5.37 f 6.37) during IEF due to the missing amino-terminal aspartate. The same shift is observed for A␤ peptides 3-40/42 because alanine at position 2 is not charged. A␤2-40 and A␤2-42 can be separated within the second analytical dimension due to differential shifts in conformation in the presence of urea (Fig. 3A). A␤ peptides 3-40 (data not shown) and 3-42 were barely detected by mAb 1E8 (Fig. 4B; cross reactivity Ͻ 5%). Synthetic pyroglutamtate derivates A␤3p-40/42 did not comigrate with A␤2/ 3-42 (Fig. 4C) and were not detected by mAb 1E8 (data not shown; cross reactivity Ͻ 5%).
Comparison of the Relative Abundance of Additional Aminoterminal-truncated A␤ Peptide Species in AD Brain Homogenates-We immunoprecipitated (mAb 6E10) a pool of the detergent-soluble fraction from the frontal lobes of AD patients (n ϭ 8). Immunoprecipitates were analyzed by A␤ SDS-PAGE/ immunoblot using either mAb 6E10 (Fig. 4, A and C) or 1E8 (Fig. 4B). For comparison, the Western immunoblots contained a dilution series of synthetic A␤2-42 and A␤3-42 (Fig. 4, A and  B) or A␤3p-40 and A␤3p-42 (Fig. 4C).
Analysis of the synthetic A␤ peptides revealed that mAb 1E8 barely detected A␤ peptides that were amino-terminal by more than 1 amino acid (cross reactivity Ͻ 5%). Interestingly, the high specificity of mAb 1E8 for the A␤ peptide amino terminus was restricted to our A␤ SDS-PAGE/immunoblot because A␤2-42 and A␤3-42 were immunoprecipitated with approxi-mately the same yield by mAb 1E8 (data not shown). Synthetic A␤2/3-42 showed an identical migration pattern and migrated faster than A␤1-42 (Fig. 4, A and B). The highly amyloidogenic A␤ peptides 3p-40 and 3p-42 migrated even faster than A␤2/ 3-42 but were not further separated by A␤ SDS-PAGE/immunoblot (Fig. 4C).
According to Fig. 4D, the enhanced chemiluminescence signal was additive for A␤2/3-42 if both peptides were run as a mix (duplicate; 500 pg each) but differentiated by quantification of one blot membrane by mAb 6E10 and the other by mAb 1E8. This differential Western immunoblot technique allowed us to estimate the amount of A␤3-42 comigrating with A␤2- 42. It is noteworthy that the avidity of mAb 6E10 for A␤3-42 during A␤ SDS-PAGE/immunoblot is 2.8-fold lower than that for A␤2-42. Again, we did not observe a significant difference with regard to the avidity of mAb 6E10 for A␤2/3-42 if the antibody was used during immunoprecipitation (results not shown).
A␤2-42 Is Specifically Elevated in Brain and CSF Samples of AD Patients-After A␤1-42, A␤2-42 was the next most prominent species in the RIPA-soluble fraction for the whole set of AD brain homogenates investigated (n ϭ 9), and both A␤ peptides were grossly elevated as compared with A␤1-38 and A␤1-40. Interestingly, a patient with familial AD due to PS1 mutation T115C did show the same A␤ peptide pattern as described for the sporadic AD cases. In line with the comparatively low ␤-amyloid plaque density of the cerebellum in AD, the cerebellar concentration of A␤1-42 and the additional A␤ peptide species was more than 10-fold lower than that present in the frontal lobe (Fig. 2B, lane 2; Table I). This difference between frontal lobe and cerebellum was most pronounced for A␤2-42. The additional A␤ peptide species and A␤1-42, A␤1-38, and A␤1-40 were also present in the RIPA-soluble fraction of brain homogenates from patients with frontotemporal dementia lacking distinctive histopathology (n ϭ 2), Pick's disease (n ϭ 3), tangle predominant form of senile dementia (n ϭ 1), or nondemented controls (n ϭ 4), but to a much lower extent (Fig. 2B, lanes 3 and 4). Interestingly, patients with Lewy body dementia (LBD) and pronounced ␤-amyloid plaque load (LBD CERAD C; n ϭ 2) also showed the striking elevation of A␤2-42 and the A␤ peptide pattern of AD patients, whereas a LBD patient without a concomitant ␤-amyloid pathology (LBD CERAD A; n ϭ 1) did not show these changes. The concentrations of A␤ peptide species 2-42, 1-42, 1-40, and 1-38 in the brain homogenates from AD patients (frontal lobe and cerebellum), patients with other dementias (frontal lobe), and controls (frontal lobe) are summarized in Table I.
Moreover, A␤2-42 was detected in the CSF of a subset of patients with AD (n ϭ 18 of 51; 35%). In contrast, A␤2-42 was detected significantly less frequently in nAD and OND patients ( Fig. 2A, lane 4). Only 2 of the 12 nAD patients (17%) and 4 of the 41 OND patients (10%) had detectable CSF levels of A␤2-42 (AD versus nAD/OND, p ϭ 0.005). Among the two nAD patients with detectable CSF levels of A␤2-42, one 71-year-old female was diagnosed with probable LBD, and a 61-year-old female was classified as having unspecified dementia (International Classification of Diseases 10 F.03) due to atypical symptomatology. According to our neuropatholical data, the former patient may represent a LBD CERAD C case. Interestingly, three of the four OND patients with A␤2-42-positive CSF presented with cognitive impairment as part of their psychiatric symptomatology.
The subgroup of AD patients with A␤2-42-positive CSF samples was characterized by significantly lower CSF concentrations of A␤1-37 (p Ͻ 0.05) as compared with the AD patients without detectable CSF concentrations of A␤2-42 (Fig. 5A). This drop in A␤1-37 was even more pronounced if A␤1-37 was    Results are expressed as mean Ϯ standard deviation (pg/mg total protein). expressed as percentage of total A␤ peptides (percentage of A␤1-37, p Ͻ 0.01) and was paralleled by a significant increase in the percentage of A␤1-39 (p Ͻ 0.025; Fig. 5B).

Amino-and Carboxyl-terminal-truncated A␤ Species in Presenilin-1-deficient Neurons and APP-trafficking Mutants-We
wanted to investigate to what extent the generation of aminoterminal-and carboxyl-terminal-truncated A␤ peptides was dependent on presenilin activity and intracellular trafficking. For this purpose, we infected primary mouse neurons from control animals and presenilin-1-deficient mice with Semliki Forest Virus-expressing human APP. In addition, we expressed human APP, which is retained in the ER due to an ER retention motif (APP-KK) and another APP variant lacking the cytoplasmic domain essential for reinternalization (APP-⌬ct).
According to Fig. 6, the expression level of holo-APP was comparable in both PS-1-deficient and control neurons as well as in the two trafficking mutants. The infected cells overexpressed holo-APP severalfold as compared with the endogenous level (Fig. 6).
As described previously (16), secretion of A␤ peptides was drastically reduced in PS1-deficient neurons. However, the PS1dependent reduction was different for the single A␤ peptide species studied. Interestingly, the carboxyl-terminal-truncated A␤ peptides were decreased to at least the extent of A␤1-40 and A␤1-42, in contrast to the much less pronounced reduction of A␤2-42 (Fig. 7). Comparable with the reduction of A␤ peptide secretion in PS1-deficient neurons, we also observed reduced levels of secreted A␤ peptides in both trafficking mutants, especially for APP-KK. For APP-⌬ct, we also observed a differential effect for the carboxyl-terminal-truncated A␤ peptides and the amino-terminal-truncated A␤2-42 (Fig. 7).
In the following text, A␤ peptide results will be expressed as ratios (Figs. 8 and 9) or percentage of change (see the body of text explaining Figs. 8 and 9). Either PS1Ϫ/Ϫ versus PS1ϩ/ϩ or slow-internalizing APP mutant versus APP wild-type transfected controls is shown (mean Ϯ S.D., four independent experiments).
According to Fig. 8C, total A␤ showed a PS1-dependent average reduction of 87 Ϯ 4.2% for the wild-type and 66 Ϯ 5.6% for APP-⌬ct, whereas A␤2-42 was reduced by only 43 Ϯ 15.8% (A␤2-42 versus total A␤ species, p ϭ 0.029) and 22 Ϯ 18% (A␤2-42 versus total A␤ species, p ϭ 0.029), respectively. The PS1-dependent reductions for the other single A␤ peptide species aside from A␤2-42 were similar to total A␤. They ranged between 92 Ϯ 4.2% (A␤1-38) and 84 Ϯ 4.6% (A␤1-42) for the wild-type and 70 Ϯ 5.4% (A␤1-40) and 57 Ϯ 5.0% (A␤1-42) for the carboxyl-terminal-deleted APP construct (Fig. 8C). Fig. 8, A and B, compares the effect of trafficking mutants on A␤ peptide secretion for PS-deficient neurons and controls. According to Fig. 8A, which depicts the slow-internalizing mutant (APP-⌬ct) versus wild-type APP in PS1ϩ/ϩ neurons, we observed an APP-⌬ct-dependent reduction in the range of 69 Ϯ 12.6% (A␤1-40) to 79 Ϯ 10.1% (A␤1-38) for all A␤ peptide species except A␤2-42, which was reduced in the order of only 25 Ϯ 29.1%. The APP-⌬ct-dependent reduction of A␤2-42 was significantly less pronounced as compared with total A␤ peptide species (p ϭ 0.029). Thus, a reduced secretion of A␤ peptides was observed for impaired endocytosis of APP in PS1ϩ/ϩ neurons (Fig. 8A) as well as PS1 deficiency (Fig. 8C), and both effects also differentially affected the secretion of amino-terminal-truncated A␤2-42 and the A␤1-x species. For the APP-KK construct, we observed low levels of A␤ peptides for all A␤ peptide species, which were too close to the limit of detection to allow quantification. Fig. 8B depicts the A␤ peptide ratios for APP-⌬ct versus wild-type APP in PS1Ϫ/Ϫ neurons. In contrast to PS1ϩ/ϩ neurons, the reduction of A␤ peptide secretion due to the slowinternalizing mutant was much less pronounced and ranged only from 14 Ϯ 25.1% (A␤1-37) to 26 Ϯ 14.3% (A␤1-42). In this case, the secretion of A␤2-42 was not reduced at all by the APP-⌬ct mutant, but the differential effect of the slow-internalizing mutant on the secretion of A␤2-42, as compared with that of the A␤1-x peptide species, was not significant any more.
In APP-KK-transfected and PS1-deficient neurons, decreased secretion of all A␤ peptide species was observed (Fig.  7). However, the validity of this effect is again limited by the low amount of A␤ peptides secreted. Fig. 9 compares the amino-terminal-truncated A␤2-42 with the other A␤ peptide species to highlight differential effects on APP processing. According to Fig. 9C, in both APP wild-typeand APP-⌬ct-transfected neurons, A␤2-42 is significantly less reduced due to PS1 deficiency as compared with the total of all other A␤ peptide forms (p ϭ 0.029). In PS1ϩ/ϩ control neurons (Fig. 9A), A␤2-42 is decreased by only 25 Ϯ 29.1%, whereas the total of all other A␤ forms is reduced by 71 Ϯ 10.9% (p ϭ 0.029).
Notably, as depicted in Fig. 9B, the differential effect of APP-⌬ct on the secretion of A␤2-42 as compared with A␤1-x was not significant any more in PS1-deficient neurons.

DISCUSSION
High Resolution Analysis of A␤ Peptides-One-and twodimensional A␤ SDS-PAGE/immunoblot combined with immunoprecipitation offers a detection sensitivity comparable to enzyme-linked immunosorbent assay methods and a high resolution separation of a complex mixture of APP metabolites. We further refined the separation gel matrix and detection sensitivity of the A␤ SDS-PAGE/immunoblot, which allows improved separation of the amino-and carboxyl-terminaltruncated A␤ peptide and detection of only 0.3-0.6 pg of A␤ peptides. 2 The monoclonal antibody 1E8 used in this study is highly specific for the A␤ peptide amino terminus, and A␤ peptides amino-terminally truncated by more than 2 amino acids are barely detected (cross reactivity Ͻ5%).
A␤ Peptide Quintet and Carboxyl-terminal-truncated A␤ Peptides-Our finding of a highly conserved A␤ peptide quintet in various biological fluids and PS1-dependent ␥-secretase processing of A␤1-37, A␤1-38, and A␤1-39 provides further evidence for alternative ␥-secretase activities (60 -62). Interestingly, the relative A␤ peptide quantities of the latter five A␤ peptides were nearly identical in human CSF and supernatants from human APP-transfected primary mouse neurons. In CSF samples, we identified disease-specific patterns of this A␤ peptide quintet in subjects with AD; specifically, the relative amounts of A␤1-38 and A␤1-42 were closely correlated. 2 In the RIPA fraction of brain homogenates, A␤1-38 was elevated in AD patients as compared with controls, whereas this was not observed for A␤1-37 and A␤1-39. This may be of pathophysiological relevance because A␤1-38 was reported to induce autoaggregation of ␣-synuclein (63,64).
Specific Elevation of an Additional Amino-terminal-truncated A␤ Peptide Species in AD and Its Identification as A␤2-42-Investigating the RIPA-soluble fraction of brain homogenates and CSF samples from AD patients, we observed an additional A␤ peptide species migrating below (anodically) A␤1-42. Using IR-MALDI mass analysis, gas-phase sequencing, and A␤ IPG-2D-PAGE/immunoblot comigration experiments with synthetic A␤ peptides, we identified the latter A␤ peptide as A␤2-42, i.e. an A␤ peptide species lacking the amino-terminal aspartate.
Like A␤1-42, this additional A␤ peptide species was strikingly and consistently up-regulated in the frontal lobe and up-regulated to a much lesser degree in the cerebellum of patients with AD. Interestingly, a patient with familial AD due to a PS1 mutation (T115C) also showed the striking elevation of A␤2-42 and an A␤ peptide pattern otherwise identical to the sporadic AD cases. An A␤ peptide pattern similar to the one observed for patients with sporadic AD was observed for LBD Representative analyses of full-length APP are shown. PS1ϩ/ϩ or PS1Ϫ/Ϫ primary neuronal cultures were infected with pSFV bearing APP-WT, APP-⌬ct, or APP-KK as described and metabolically labeled with [ 35 S]methionine for 4 h at 37°C. Cells extracts were immunoprecipitated using antibodies against the APP ectodomain. The precipitates were separated by SDS-PAGE on 10 -20% Tris-tricine gradient gels, and detection of radioactive material was performed with a PhosphorImager. Note that the APP-⌬ct holo-forms always run lower than the other constructs due to the absence of 42 amino acids of the cytoplasmic domain. patients, but only for those cases in which Lewy body neuropathology was paralleled by a heavy ␤-amyloid plaque burden (LBD CERAD C). A␤2-42 was also detectable on a very low level in patients with other dementias and in nondemented controls.
In AD brain samples, A␤2-42 was consistently increased, but detectable CSF levels were observed only in a subset of AD patients. Nonetheless, the presence of A␤2-42 in CSF was specific for AD because the peptide was observed significantly less frequently in non-AD patients. Interestingly, one of those patients was a LBD case, and four of the remaining five patients presented with cognitive impairment of unknown etiology as part of their psychiatric symptomatology. At present, we cannot explain why only a subset of AD patients had detectable CSF levels of A␤2-42. Further experiments will clarify whether A␤2-42 CSF levels are sensitive to the preanalytical handling of the samples. Alternatively, CSF A␤2-42 may correlate with the clinical course of AD or indicate a specific phenotype of the disease. It is noteworthy that the subgroup of AD patients with A␤2-42-positive CSF samples was also characterized by decreased and elevated relative abundances of A␤1-37 and A␤1-39, respectively.
An increased ratio of amino-terminally modified A␤x-42: A␤1-42 in CSF samples of AD patients has been suggested previously (65), whereas the ratio of A␤x-40:A␤1-40 did not discriminate between AD and non-AD patients. However, the latter enzyme-linked immunosorbent assay-based study did not further specify the nature of these amino-terminal modifications. Taken together, the data suggest that the ratio A␤2-42:A␤1-42 is a promising surrogate marker for the neurochemical diagnosis of AD.

Amino-terminal-and Carboxyl-terminal-truncated A␤ Species in PS1-deficient Neurons and APP-trafficking Mutants-
There is increasing evidence that APP processing in the transmembrane region by ␥-secretase(s) is closely related to presenilin activity (20). Therefore, we investigated to what extent the generation of the amino-terminal-and carboxylterminal-truncated A␤ peptides was also dependent on presenilin activity. In addition, we were interested in studying the generation of the amino-terminal-and carboxyl-terminal-truncated A␤ peptide species in addition to A␤1-40 and A␤1-42 in APP-trafficking mutants.
We observed a striking PS1-dependent reduction of the carboxyl-terminal-truncated A␤ peptides 1-37, 1-38, and 1-39 that was at least as pronounced as that initially described for A␤1-40/42 (16). Surprisingly, this was contrasted by a significantly less pronounced reduction of A␤2-42. Furthermore, the A␤ peptide secretion pattern of neurons carrying the slow-internalizing mutant (PS1ϩ/ϩ and APP-⌬ct) closely resembled the A␤ peptide secretion profile of PS1Ϫ/Ϫ neurons. By contrast, a striking reduction of all secreted A␤ peptides, including A␤2-42, was observed for the construct carrying the ER retention motif (APP-KK). However, this finding is less valid due to the low level of secreted A␤ peptides. Taken together, our data show that the generation of A␤2-42 is less dependent on PS1 activity and endocytosis of APP, and they suggest that both A␤ peptide species (A␤2-42 and A␤1-x) were not generated in the ER.
Amino-terminal-truncated A␤ Peptides-The generation of  Fig. 7 and from three other independent experiments were quantified. Data obtained for PS1ϩ/ϩ (A) or PS1Ϫ/Ϫ neurons (B) transfected with APP-⌬ct were compared with data obtained from cells expressing APP-WT, which constitutes the 100% reference. In C, original data obtained in A and B were expressed as the ratio of PS1Ϫ/Ϫ:PS1ϩ/ϩ for all APP constructs independently. ‫,ء‬ A␤2-42 versus total A␤ peptide species, p Ͻ 0.025; n.s., A␤2-42 versus total A␤ peptide species, p Ͼ 0.05. amino-terminal-truncated A␤2-42 implies alternative ␤-secretase activity.
Heterogeneity of ␤-secretase cleavage has been described previously (6 -8), and inhibition of the ␤-secretase activity generating Asp 1 of A␤ peptides gives rise to alternative aminoterminal cleavages (66,67). Heterogeneity of the amino terminus of A␤ peptides was also detected in the earliest purification of the ␤-amyloid plaque core (29). Interestingly, A␤2-39/40 is a major A␤ peptide species of cerebrovascular ␤-amyloid, which contains a higher relative amount of A␤2-x as compared with plaque core ␤-amyloid (68,69). Studies of the A␤ peptides secreted into the media of various cultured cells and cell lines transfected with differing APP constructs have also identified A␤2-x among other amino-terminal-truncated A␤ peptide species (7,70,71).
Alternative ␤-secretase activity could be due to alternative BACE activity or standard BACE Met/Asp activity and further processing of A␤1-x to A␤2-x by aminopeptidase A (aspartylaminopeptidase activity) and of A␤2-x to A␤3-x by aminopeptidase N (alanyl-aminopeptidase activity). Aminopeptidase A and aminopeptidase N belong to the class of membrane-bound cell surface peptidases, which have extracellular catalytic sites and can metabolize biologically active peptides at the cell surface, serving as local regulators of peptide concentrations (72). An additional member of this group of enzymes is neutral endopeptidase, which recently was identified as a major A␤1-42-catabolizing enzyme (73).
Accordingly, A␤2-42 is the direct precursor of A␤ peptides 3-40/42 and their pyroglutamate derivatives 3pyro-40/42, which are not amenable to further degradation by aminopeptidase(s) (see review in Ref. 74). Moreover, pyroglutamyl aminopeptidase activity is low in human cortical extracts, which explains the relative prominence of A␤3pyro-40/42 (75). The latter two amino-terminal-truncated A␤ peptide species are deposited before A␤ peptides 1-x and represent a dominant species in early stages of ␤-amyloid plaque formation (76,77).
Under physiological conditions, A␤2-42 should be metabolized rapidly to A␤3-42 due to a high activity of cortical aminopeptidase N, which exceeds the activity of the glutamyl-aminopeptidase by severalfold (75). The high activity of aminopeptidase N might explain why A␤2-40, among several aminoterminal-elongated or -truncated A␤ peptides (A␤x-40) that were generated as recombinant A␤ peptides in a cell culture system, was selectively prone to amino-terminal degradation (9).
In view of the comparatively high percentage of A␤2-39/40 in cerebrovascular ␤-amyloid and the vasoactive properties of A␤ peptides (78,79), it is of interest to note that pericytes and periendothelial cells of brain parenchyma vessels coexpress aminopeptidase N, aminopeptidase A, and nestin as part of the autonomous angiotensin system of the brain (80). Soluble aminopeptidase N (CSF and plasma) and aminopeptidase A (glutamyl-aminopeptidase activity; plasma) have been reported to be decreased in AD (75,81). Aminopeptidase N also has alanyl-aminopeptidase activity, which may be relevant for the elevated CSF levels of A␤2-42 in a subgroup of our AD patients.
The pathogenetic role of amino-terminal-truncated A␤ peptides has been reviewed recently (82,83), and Saido (84) proposed an aminopeptidase hypothesis for A␤ peptide catabolism; however, the pathophysiological role of A␤2-42 was not addressed.
In conditioned media of PS1ϩ/ϩ neurons, A␤2-42 amounts only to ϳ10% of A␤1-42. Thus, we would expect a severalfold increase of A␤2-42 if the prominent decrease of A␤1-42 in PS1-and endocytosis-deficient neurons is due to its increased catabolism by aminopeptidase A generating A␤2-42. However, we observe an attenuated decrease of A2-42. Therefore, A␤1-42 is not likely to be a precursor of A␤2-42. This does not exclude the possibility that A␤2-42 might be a rapidly metabolized (e.g. by aminopeptidases) precursor of other amino-terminal-truncated A␤ peptides, e.g. the highly amyloidogenic A␤3-42 and A␤3pyro-42.
Taken together, our data suggest that A␤2-42 is generated due to an alternative ␤ Asp/Ala -secretase activity rather than a standard BACE Met/Asp and aminopeptidase A (aspartyl) activity, in as much as in the latter case, a reduction of the precursor A␤1-42 in PS1-deficient neurons should have been paralleled by a much more pronounced decrease of A␤2-42.
Investigating BACE-1 knockout mice by the methods described here will help to clarify whether the generation of A␤2-42 is due to BACE activity.
Secreted A␤ peptides were also analyzed by urea-based A␤ SDS-PAGE/immunoblot in human embryonic kidney cell lines stably expressing human APP mutants defective in endocytosis (9). The slow-internalizing mutants comprised APP lacking the entire cytoplasmic domain or APP with both tyrosine residues of the motif GYENPTY mutated to alanine. In response to the impaired endocytosis of APP, a reduction of A␤1-40 secretion was observed, which was paralleled by elevated levels of two amino-terminal-truncated A␤ peptides (A␤x/y-40). These were identified as A␤ peptides 3-40 (y-40) and 5-40 (x-40) by comi-  Fig. 7 and from three other independent experiments were quantified; results of secreted A␤2-42 were compared with those of total secreted other A␤ forms. Data obtained for PS1ϩ/ϩ (A) or PS1Ϫ/Ϫ neurons (B) transfected with APP-⌬ct were compared with data obtained from cells expressing APP-WT, which constitutes the 100% reference. In C, original data obtained in A and B were expressed as the ratio of PS1Ϫ/Ϫ:PS1ϩ/ϩ for all APP constructs independently. ‫,ء‬ A␤2-42 versus total all other A␤ peptide species, p Ͻ 0.025; n.s., A␤2-42 versus total all other A␤ peptide species, p Ͼ 0.05. gration with a large panel of recombinant standard A␤ peptides that included A␤2-40. Interestingly, only A␤2-40 was partially amino-terminally-degraded in vivo when it was generated as recombinant standard. Our synthetic A␤2-40 was stable and, according to its electrophoretic migration pattern, might also comigrate with A␤ peptide y-40. Thus, it is not clear to what extent human embryonic kidney cell lines may also produce A␤2-40 in response to slow-internalizing APP mutants. In accordance with our data from the APP-KK construct, the amino-terminal-truncated A␤ peptides were not generated when the APP mutants were retained in the ER by treatment with brefeldin. Cescato et al. (9) conclude from their data that cleavage at position 1 of A␤ peptides occurs predominantly in endosomes, whereas ␤-secretase cleavage at alternative sites takes place at the plasma membrane.
Interestingly, human embryonic kidney cells did not generate significant amounts of A␤2/3-42, and, vice versa, we did not identify elevated levels of A␤2/3-40 in the RIPA-soluble fraction of A␤ peptides in brain homogenates. The latter observations suggest that non-neuronal cells such as human embryonic kidney cells or cerebral endothelial cells favor the generation of amino-terminal-and carboxyl-terminal-truncated A␤x-39/40, whereas neuronal cells favor the processing of amino-terminaltruncated but carboxyl-terminal-elongated A␤ peptides, e.g. A␤2-42. This would be of pathophysiological relevance for fibrillogenesis because amino-terminal truncation and carboxylterminal elongation act synergistically in increasing the aggregation potential of A␤ peptides. Moreover, our finding of a prominent elevation of A␤2-42 in the RIPA-soluble fraction of brain homogenates indicates that A␤2-42 does not originate from the ␤-amyloid plaque core because the latter aggregates will not be disaggregated by mild detergents. Because ␤-sheet structure can be induced in amyloidogenic peptides by neutralization of aspartate (85,86), A␤2-42 may serve as a first nidus for ␤-amyloid nucleation preceding the formation of plaque core ␤-amyloid. Moreover, in neuronal cells, amino-terminal-truncated A␤ peptides may also be generated intracellularly (87). If A␤2/3-42 peptides are generated by alternative ␤-secretase activity in a mildly acid subcellular compartment, as known for BACE activity, then their aggregation potential should be specifically high at this site due to a less acidic isoelectric point of A␤2/3-42 (Ip ϭ 6.37) as compared with that of A␤1-42 (Ip ϭ 5.37).
We cannot exclude that A␤3-42 is also generated in our neuronal cell culture system because both peptides migrate at the same position and are not separated by isoelectric focusing. However, our quantification of A␤2-40/42 is not confounded by A␤3-40/42 because during A␤ SDS-PAGE/immunoblot, mAb 1E8 does not detect significant amounts of A␤ peptides that are truncated amino-terminally by more than 1 amino acid. A comparative analysis of the detergent-soluble fraction of immunoprecipitates from the frontal lobe of AD patients revealed that A␤2-42 is the most prominent A␤ peptide species aside from A␤1-42, exceeding the abundance of additional aminoterminal-truncated A␤ peptides such as 3p-40/42 and 3-42.
We observed a striking elevation of A␤2-42 in the RIPAsoluble fraction of A␤ peptides from the frontal lobe of a patient with a PS1 mutation (T115C), and according to more recent studies, the generation of amino-terminal-truncated A␤ peptides is a general feature of familial AD.
Russo et al. (88) also investigated the detergent-soluble fraction of A␤ peptides from brain homogenates of subjects with sporadic AD and familial AD, which were linked to mutations in either the PS1 or APP gene. In AD, amino-terminal-truncated A␤ peptides (3/4 -42 and 11-42) were more abundant than the full-length A␤ peptide 1-42, and the most prominent relative amounts of amino-terminal-truncated A␤ peptides were observed for the pathogenic PS1 mutations. The authors conclude that both ␥-secretase cleavage and ␤-secretase cleavage are affected by PS1 mutations, but they also consider that PS1 mutations might affect these secretases indirectly by interfering with the trafficking of APP in the cell.
Furthermore, Kumar-Singh et al. (46) recently described an aggressive form of familial AD caused by a novel missense mutation in APP (T714I). Thus, the ␥-secretase cleavage generating A␤1-42 is directly involved. This mutation resulted in the most drastic increase (11-fold) of the A␤1-42:A␤1-40 ratio reported thus far and coincided in brain with the deposition of abundant and predominant nonfibrillar preamyloid plaques composed primarily of A␤x-42 in the absence of A␤x-40. Interestingly, carboxyl-terminal-truncated A␤ peptides 1-37, 1-38, and 1-39 and x-37, x-38, and x-39 were also elevated and were most pronounced for A␤38. This pattern closely resembles the one we observed in our brain homogenates of AD patients, i.e. striking elevation of A␤x-42 and A␤1-42, surprisingly low amounts of A␤1-40, and elevated levels of A␤1-38. The authors conclude that A␤x-42 as diffuse nonfibrillar plaques has an essential but undetermined role in AD pathology.
Conclusions-Taken together, our data indicate that the amino-terminal-truncated and carboxyl-terminal-elongated A␤2-42 was generated within a post-ER secretory pathway, where most of the A␤2-42 is generated by ␤ Asp/Ala -secretase activity, possibly as alternative BACE activity, and not due to a standard BACE Met/Asp activity with secondary catabolic aminopeptidase activity. Moreover, the generation of A␤2-42 by ␥-secretase activity seems to be less PS1-dependent.