Address correspondence and reprint requests to Bart De Strooper, Neuronal Cell Biology and Gene Transfer, Center for Human Genetics, Flanders Interuniversity Institute for Biotechnology (VIB4) and K. U. Leuven, 3000 Leuven, Belgium. E-mail: Bart.DeStrooper@med.kuleuven.ac.be
Mutations in human presenilin (PS) genes cause aggressive forms of familial Alzheimer's disease. Presenilins are polytopic proteins that harbour the catalytic site of the γ-secretase complex and cleave many type I transmembrane proteins including β-amyloid precursor protein (APP), Notch and syndecan 3. Contradictory results have been published concerning whether PS mutations cause ‘abnormal’ gain or (partial) loss of function of γ-secretase. To avoid the possibility that wild-type PS confounds the interpretation of the results, we used presenilin-deficient cells to analyse the effects of different clinical mutations on APP, Notch, syndecan 3 and N-cadherin substrate processing, and on γ-secretase complex formation. A loss in APP and Notch substrate processing at ɛ and S3 cleavage sites was observed with all presenilin mutants, whereas APP processing at the γ site was affected in variable ways. PS1-Δ9 and PS1-L166P mutations caused a reduction in β-amyloid peptide (Aβ)40 production whereas PS1-G384A mutant significantly increased Aβ42. Interestingly PS2, a close homologue of PS1, appeared to be a less efficient producer of Aβ than PS1. Finally, subtle differences in γ-secretase complex assembly were observed. Overall, our results indicate that the different mutations in PS affect γ-secretase structure or function in multiple ways.
Missense mutations in the PS genes are a major cause of familial Alzheimer's disease. Although the known mutations, mostly missense mutations causing amino acid substitutions, are scattered over the PS protein, they all result in one major biochemical alteration, that is a relative increase in the ratio of the β-amyloid (Aβ)42 to Aβ40 peptides (Scheuner et al. 1996). In principle this can be caused either by increased Aβ42 or decreased Aβ40 generation, or a combination of both. A complication in interpreting the effects of clinical mutants is their possible direct interaction with wild-type PS. Indeed, it is not unlikely that every γ-secretase complex contains two (or more) PS subunits (Schroeter et al. 2003). The situation is further complicated by the fact that γ-secretase has many substrates, which raises the question of whether mutations affect all functions to a similar extent. Thus the debate about whether clinical mutations in PS cause gain of an ‘abnormal’ or loss of a ‘normal’ function is ongoing. Rescue experiments with PS containing clinical mutations in Ps-deficient mice (Ps denotes mouse presenilin), cells and Caenorhabditis elegans have provided quite opposite conclusions. Although loss of Notch cleavage or signalling has been demonstrated repeatedly in mammalian cell lines (Song et al. 1999; Schroeter et al. 2003) and in C. Elegans (Baumeister et al. 1997), expressing the clinical PS1-A246E mutant was able to partially rescue the Notch signalling-deficient phenotypes in Ps1-deficient mice (Davis et al. 1998; Qian et al. 1998). With regard to APP processing, the experimental data suggest two opposing possibilities: all PS mutations apparently increase the Aβ42/40 ratio, arguing for a gain of function, whereas total loss of Ps (De Strooper et al. 1998; Herreman et al. 2000) results in a total loss of Aβ generation. The crucial experiment to be done is to express clinical mutants in deficient cells and to evaluate to what extent they can rescue different aspects of loss of function. Our results demonstrated in all investigated cases a (partial) loss of function associated with clinical mutants. With regard to APP processing, the PS clinical mutations increased the Aβ42/40 ratio, even in the absence of wild-type Ps. This was, however, the consequence of either lowering Aβ40, or increasing Aβ42, or a combination of both.
Materials and methods
Cell culture and generation of stable cell lines
Ps1- and Ps2-deficient mouse embryonic fibroblast (MEF) cell lines (Ps–/–) were maintained in Dulbecco's modified Eagle's medium (Sigma, St Louis MO, USA) containing 10% fetal calf serum. At 50% confluency, MEFs were transduced using a replication-defective recombinant retroviral expression system (Clontech, Palo Alto, CA, USA) harbouring cDNA inserts encoding wild-type human PS1 and human PS2, for PS variants containing Alzheimer's disease-causing clinical mutations [PS1-Δ-exon9 (PS1-Δ9), PS1-L166P, PS1-G384A, PS1-A246E or PS2-N141I ;Table 1] or for mouse Ps1 and mouse Ps2 fused to an N-terminal double Tag (DT) consisting of 3 × Flag peptide and calmodulin-binding peptide (CBP) separated with a spacer. Stable transfected cell lines were selected using 5 µg/mL puromycin (Sigma).
Table 1. PS clinical mutants used in the study
Mean age at onset (years)
Mean age at death (years)
In PS1-Δ9, exon 9 is deleted which is accompanied by the substitution of Ser 290 by a cysteine (Steiner et al. 1999). HL, hydrophylic loop; TM, Transmembrane domain, ND, Not determined
Blue native gel electrophoresis
Blue native polyacrylamide gel electrophoresis (PAGE) (Schagger et al. 1994) allows the separation of membrane protein complexes in their native state according to their molecular weight. This method was performed as described previously (Nyabi et al. 2003), except that the polyacrylamide gradient ranged from 5 to 16% to allow better separation; total protein loaded was 20 µg. After transfer, previously described antibodies raised against Nct (9C3), PS1 (B14.5), PS2 (B24.2), Aph-1-A (B80.2) and Pen-2 (B126.2) were used to analyse the incorporation of γ-secretase components into the complex (Herreman et al. 1999; Nyabi et al. 2003; Esselens et al. 2004).
Syndecan 3, APP, N-cadherin and Notch processing analysis
The analysis of endogenous APP, syndecan 3 and N-cadherin processing was performed by direct western blotting. Briefly, proteins were extracted from different cell lines described above and were separated on Nu PAGE 10% Bis-Tris gel (Invitrogen, Carlsbad, CA, USA) transferred to nitrocellulose membranes and developed using polyclonal antibody B10.4, monoclonal antibody 2E9 (Schulz et al. 2003) and monoclonal antibody anti-N-cadherin (clone 32; BD Bioscience-pharmingen (San Jose, CA, USA)) respectively. C-terminal fragments (CTFs), obtained from three independent experiments, and produced after shedding of the full-length substrates, were quantified by densitometry and their accumulation levels were compared with those of the Ps–/– cell line.
For Notch processing, different stable cell lines were transduced with adenovirus bearing Notch ΔE construct. Similarly, expressed Notch ΔE protein, which is the direct substrate of γ-secretase, and Notch intracellular domain (NICD) corresponding to γ-secretase cleaved product were revealed by western blotting using anti-myc C-myc (A14) polyclonal antibody (Santa Cruz Biotechnology, Santa Cruz, CA, USA) and cleaved Notch1 (Val 1744) antibody (Cell Signaling Technology, Beverly, MA, USA) respectively. The amounts of NICD produced were compared, quantified and normalized to the level of expression of the uncleaved precursor protein.
Analogous fragments of APP were generated from exogenous APP-C99-Flag substrate in a cell-free system (Li et al., 2000). Briefly, microsomal membranes were isolated from different cell lines, resuspended in 50 mm HEPES pH 7, containing protease inhibitor cocktail (Roche, Indianapolis, IN, USA), and solubilized with an equal volume of 1% CHAPS. Equal amounts of synthetic substrate were added to 10 µg membrane proteins and the reaction mix was incubated overnight at 37°C. APP ɛ CTFs were revealed by western blotting using a Flag M2 monoclonal antibody (Sigma) and quantified by densitometry.
Stable MEF cell lines were transduced with human APP-Swedish-695 (APP695Sw) adenovirus (Michiels et al. 2002) for 6 h using an infection multiplicity of 50. Conditioned media were collected 48 h after infection and immunoprecipitated with B7/8 polyclonal antibodies raised against the N-teminus of Aβ (De Strooper et al. 1995). Immune precipitates and cell extracted proteins were separated on 10% MES gel (Invitrogen, Carlsbad, CA, USA) and analysed by western blotting with horseradish peroxidase-conjugated secondary antibodies and enhance chemiluminescence detection. B10.4 polyclonal antibody was used to detect full-length APP (De Jonghe et al. 2001) and CTFs of APP, and monoclonal antibody W02 was used to detect Aβ peptides (Ida et al. 1996). Finally, Aβ peptide signals were quantified by densitometry and data were normalized with respect to the level of expression of full-length APP.
Stable cell lines expressing wild-type and different PS clinical mutants were transduced for 6 h with APP695Sw adenovirus. Some 24 h after infection conditioned media were cleared by centrifugation and assayed immediately for Αβ40 and Αβ42 using specific ELISAs purchased from Biosource International (Camarillo, CA, USA) and Innogenetics Inc. (Alpharetta, GA, USA) respectively, according to the manufacturer's protocol. Data from two independent ELISAs (n = 6) were used for calculations. For statistical analysis of the data, one-way anova was performed with a Bonferroni correction to determine significance.
Secreted APP fragments (APPs) were measured in the same samples by western blotting using 22C11 monoclonal antibody (Chemicon International, Temecula, CA, USA) and quantified by densitometry using the Odyssey Infrared Imaging System (LI-COR Biosciences, Lincoln, NE, USA).
Surface-enhanced laser desorption/ionization (SELDI) time-of-flight mass spectrometry
Monoclonal antibody W02 (1 µg) was diluted in phosphate-buffered saline (PBS) and applied to all spots of the SELDI protein chip coated with a pre-activated surface (Ciphergen Biosystems, Fremont, CA, USA). After antibody coupling, free reactive sites were blocked by incubation for 30 min at room temperature (20°C) with 0.5 m ethanolamine, pH 8. The chip was washed with PBS/0.5% Triton X-100 and rinsed with PBS. Some 200 µL conditioned medium was incubated with the spots in a bioprocessor overnight at 4°C. After washing the unbound material, 2 µL 20% saturated energy absorbing molecule (α-cyano-4-hydroxy cinnamic acid) was added to each spot. The spots were allowed to air dry completely before analysing the samples on a SELDI mass analyser PBS II with a linear time-of-flight mass spectrometer (Ciphergen Biosystems, Freemont, CA, USA).
Urea gel electrophoresis
Conditioned media were immunoprecipitated using 25 µL dynabeads (Dynal Biotech, Hamburg, Germany) coated with 1E8 monoclonal antibody (Nanotools antikörpertechik GmbH & Co., KG-Teningen, Germany). Immune precipitates were separated on 12% Bicine/Tris gel containing 8 m urea (Wiltfang et al. 2001). Different Aβ peptide species were revealed by western blotting using monoclonal antibody 1E8. Synthetic Aβ peptides of different size were run in parallel in the same gel system and under the same conditions for the identification and quantification of Aβ peptide species by densitometry.
We stably transduced Ps–/– MEFs with wild-type human PS1 or wild-type human PS2 or with PS containing the following mutations causing familial Alzheimer's disease: PS1-Δ9, PS1-L166P, PS1-A246E, PS1-G384A and PS2-N141I (Table 1). All proteins were stably expressed to a similar level, as demonstrated by the staining of PS1 and PS2 N-terminal fragment (NTF) and CTF variants (Fig. 1a). Moreover, all proteins were processed by ‘presenilinase’ into NTF and CTF except for the PS1-Δexon9 mutant that lacks the presenilinase endoproteolysis site (Perez-Tur et al. 1995) (Fig. 1a). All mutants were able to rescue the Nct glycosylation and Pen-2 expression deficiencies observed in Ps–/– deficient cells to a similar extent as wild-type PS1 and PS2. Blue native gel electrophoresis (Fig. 1b) showed that all clinical mutants were incorporated in a 440-kDa complex. Higher molecular weight complexes were observed in some lanes. With PS1-G384A, PS1-A246E, PS2 and PS2-N141I additional bands were observed reproducibly; bands of approximately ∼220 kDa stained for Nct, PS and Aph-1-A, and bands of approximately ∼140 kDa stained for Nct alone.
PS clinical mutations affect processing of substrates
APP is first cleaved by α- or β-secretase in the ectodomain, generating APP CTFs. These are further cleaved by γ-secretase. In a similar way, N-cadherin and syndecan 3 are first cleaved in their ectodomain generating CTFs. These fragments are substrates for γ-secretase and accumulate in the absence of PS (Fig. 2). We analysed the ability of PS clinical mutants to rescue the accumulation of endogenously expressed APP, syndecan 3 and N-cadherin CTFs in the PS-deficient fibroblasts. PS1-L166P appeared by far the strongest loss of function mutation, being completely inefficient in restoring the CTF turnover of the three substrates analysed (Figs 2a, c and e). The PS2-N141I mutation, on the other hand, was as inefficient as PS1-L166P in cleaving syndecan 3 and N-cadherin fragments but was able to reduce APP CTFs to wild-type levels (Figs 2b, d and f). The PS1-Δ9 mutant showed only a mild loss of function in processing of APP and syndecan 3 CTFs but dramatically affected the processing of N-cadherin substrate (Figs 2a, c and e). Finally, the PS1-G384A and PS1-A246E clinical mutants appeared more or less as efficient as the wild type in processing the three different substrates (Figs 2a, c and e). These results indicate that PS clinical mutants have different effects on γ-secretase activity and suggest that clinical mutations do not necessarily have the same effect on different substrates. Moreover, in some cases, dramatic accumulations of CTF were observed that were even stronger then those seen in the Ps knockout fibroblasts.
We next analysed the processing of NotchΔE at site 3 cleavage (S3 cleavage) versus ɛ cleavage in APP (ɛ site). The signalling protein Notch is constitutively cleaved by Furin (S1 cleavage). Upon binding of its ligands (Delta or Jagged), the Notch receptor is cleaved at a second site (S2 cleavage) by a metalloprotease of the disintegrin and Metalloprotease (ADAM) family (Brou et al. 2000; Mumm and Kopan 2000). The remaining membrane-bound fragment is then rapidly cleaved by γ-secretase (S3 cleavage) to yield a NICD (De Strooper et al. 1999). The artificial construct Notch ΔE is a truncated Notch receptor that becomes constitutively cleaved by γ-secretase and mimics the S3 cleavage of wild-type Notch. With this substrate the clinical mutants displayed a variable loss of function. Several mutants, including PS1-P166L and PS2-N141I, failed to produce any detectable level of NICD as shown by quantification of three independent experiments (Fig. 3a, lower panel). Processing of APP at the ɛ cleavage site was analysed in a cell-free assay. Absence of PS or addition of γ-secretase inhibitor X to the reaction mix completely abolished the production of the APP ɛ-CTF whereas transfection with wild-type PS1 restored this activity (Figs 3b and c). All the PS-containing clinical mutants, including wild-type PS2, showed a severe loss of APP cleavage at this site. PS1-L166P and PS1-A246E had differential and contradictory effects on APP and Notch at these analogous cleavage sites. PS1-L166P showed a complete defect in Notch processing, but produced significant levels of APP ɛ-CTF. The PS1-A246E mutation on the other hand displayed good activity in the Notch assay but not in the ɛ-CTF assay (Figs 3a and b).
We finally investigated the effects of clinical PS mutations on production of Αβ species. We transduced the fibroblasts with adenoviral vectors driving human APP695Sw expression. APP CTFs accumulated to a variable extent in the cell lines expressing PS clinical mutations and displayed a similar pattern to endogenous APP CTF (Figs 2a and b). Total Aβ generation more or less followed the same profile, being higher when APP CTF accumulation was lower and vice versa (Fig. S1). Furthermore familial Alzheimer's disease-associated mutations did not affect α- or β-secretase activity as demonstrated by the quantification of APPs (Fig. 4b). We next analysed Aβ40 and Aβ42 production using a specific ELISA (Fig. 4a). The PS clinical mutants caused variable loss in Αβ40 production, which was clear for PS1-Δ9, PS1-L166P and PS2-N141I mutations, but statistically not significant in the case of PS1-G384A and PS1-A246E. Aβ42 cleavage, on the other hand, was less affected by the mutations, except that the PS1-G384A mutant produced increased levels of Aβ42. The PS2-N141I mutant had the most dramatic effects, causing a severe loss of Aβ40 production and a significant increase in Aβ42 (3-fold increase compared with wild-type PS2) whereas PS1-A246E was overall the less severe mutation. Overall these results confirm previous conclusions that clinical mutations increase the Aβ42/Aβ40 ratio (Scheuner et al. 1996; Citron et al. 1997). In case of PS1-Δ9 and PS1-L166P, a decrease in Aβ40 is observed, while in case of PS1-G384A an increase in Aβ42 is observed. In the case of PS2-N141 mutation, both loss in Aβ40 and gain in Aβ42 can explain the dramatic change in Aβ42/Αβ40 ratio. It is remarkable that the cell line expressing wild-type PS2 produced less Aβ40 (and also Aβ42) than wild-type PS1, suggesting that PS2 is a less efficient γ-secretase than PS1 with regard to APP processing.
PS2 is a less efficient generator of Aβ peptide than PS1
The possibility remained that PS2 expression is lower than PS1 expression in these cell lines, because the antibodies used to detect PS1 or PS2 were different (Fig. 1, compare lane 2 with lane 7). Therefore, we stably transduced Ps–/–fibroblasts with double-tagged versions of mouse Ps1 or mouse Ps2 (DT-Ps1 and DT-Ps-2 respectively). The double-tagged peptide consists of a CBP and a 3 × Flag peptide (Fig. 5a) that is inserted between the first Met and the second amino acid Thr or Leu in the amino terminal part of the Ps1 and Ps2 proteins respectively. Staining with an Flag antibody allowed us to directly compare the levels of expression of the double-tagged proteins in both cell lines. Quantification of the signal band intensities by densitometry revealed that the DT-Ps2 signal was 1.5-fold stronger than the signal from DT-Ps1; this was confirmed by serial dilution (Fig. 5c). We next analysed the repercussions of the DT on Ps1 and Ps2 incorporation into the γ-secretase complex. Like untagged human PS1 and PS2 (see above), the tagged proteins were able to restore Nct glycosylation and Pen-2 stability (Fig. 5b) and were incorporated in a 440-kDa complexes as assessed by blue native gel electrophoresis (result not shown), both of which are strongly affected in Ps-deficient cells. Probably because of high levels of expression, not all DT-Ps1 and DT-Ps2 was processed by presenilinase (Fig. 5b). These results confirm the results shown in Fig. 1, i.e. that Ps2 apparently acts as efficiently as Ps1 in reconstituting γ-secretase assembly. This was further confirmed by demonstrating that similar amounts of Nct, Aph-1-A and Pen-2 were co-immunoprecipitated with DT-Ps1 and DT-Ps2 (result not shown). We finally repeated the analysis of APP, syndecan 3, N-cadherin and Notch processing in the two stable transfected cell lines as discussed above. DT-Ps1 and DT-Ps2 were able to rescue CTFs of APP, syndecan 3 and N-cadherin to a similar extent (Figs 5d–f). Surprisingly, Aβ generation was much less efficient in cells transfected with DT-Ps2 (Fig. 5g).
Analysis of γ-secretase cleavage sites
Accumulating evidence indicates that γ-secretase can cleave APP at several positions (Zhao et al. 2004; Qi-Takahara et al. 2005). We therefore included an analysis of the cleavage profiles of secreted Aβ in conditioned media from our different cell lines. We analysed Aβ profiles both by mass spectrometry (Fig. 6a) and in a high- resolution urea gel (Wiltfang et al. 2001) (Fig. 6b). Both types of analysis provided qualitatively similar results, but the urea gel appeared to be more sensitive. Aβ1−38, Aβ1−40 and Aβ1−42 were the most abundant species. The urea gel also resolved the Aβ1−33/34, Aβ1−37, Aβ1−38, Aβ1−39, Aβ1−40 and Aβ1−42 species. Loss of γ-secretase activity affected apparently all cleavages to a fairly similar extent, except for the cleavage at residue 42, which was much less affected than other cleavages in PS1-Δ9 and PS1-L166P, and was increased in absolute amounts in PS1-G384A, PS1-A246E and PS2-N141I.
From the analysis of PS-containing clinical mutants it is fairly clear that these mutations induce loss of function in APP and Notch processing at the ɛ and S3 cleavage sites respectively. The Notch data are in line with previous reports (Baumeister et al. 1997; Song et al. 1999). Interestingly, the PS1-A246E mutation is relatively mild, displaying only a 20% reduction in cleavage of Notch. This resolves the long-standing discrepancy that clinical mutations of PS are not able to rescue Notch deficiency in C. Elegans, whereas two reports demonstrated that the PS1-A246E mutation rescued, at least partially, the Notch-deficient phenotype in Ps1 knockout mice (Davis et al. 1998; Qian et al. 1998). Apparently the latter conclusion was based on experiments with a relatively mild PS1 mutant and the 20% reduction in biochemical activity observed here is probably insufficient to cause major phenotypic alterations in the mouse. We conclude that if more severe loss of function mutants of PS had been used in those experiments, it is unlikely that rescue of the mouse knockout phenotype would have been observed, in line with the data from other approaches.
Concerning the ability of PS clinical mutants to restore the cleavage of APP, syndecan 3 and N-cadherin CTFs, the effect of clinical mutants is somewhat variable: PS1-L166P, PS1-Δ9 and PS2-N141I mutants display the most robust loss in this activity, whereas other mutants affect this aspect of γ-secretase activity to a lesser extent. N-cadherin and syndecan 3 CTFs accumulated even more than in the Ps knockout fibroblasts. The reason for this remains unclear. It should be noed that we also observed a discrepancy between the turnover of APP CTF (which is normal) and the generation of Aβ peptide (which is lower than with PS1) in PS2-expressing cells. When we analysed APP processing at the γ site we found that PS1-L166P, PS1-Δ9 and PS2-N141I display lowering of total Aβ production and accumulation of APP CTF, in line with an overall loss of function phenotype. For PS1-A246E and PS1-G384A the reduction is less clear and in the latter mutant a dramatic increase (in absolute amounts) of Aβ42 is observed.
Overall, based on the results obtained with the different substrates and data from previous studies (Schroeter et al. 2003; Song et al. 1999; Moehlmann et al. 2002) we suggest that most mutations of PS cause partial loss of function of the γ-secretase complex. The cleavage generating Aβ42 is apparently the exception, being unaffected, or even increased in case of PS1-G384A, and explaining the relative increase in Aβ42/Αβ40 ratio in all five mutants. Given that the mutations consistently decrease the processing of other substrates, we propose that Aβ42 is an incompletely digested form of the Aβ peptide. The changes in Aβ40 and Aβ42 occur in the absence of wild-type PS, and are thus intrinsic to the mutated PS.
Recent studies with solubilized γ-secretase activity that employed a variety of inhibitors have led to the conclusion that these compounds bind at different sites (Tian et al. 2003). Moreover, purification of the γ-secretase complex on an affinity column generated with a catalytic site-orientated inhibitor resulted not only in purification of the γ-secretase complex but also in an enrichment of APP CTF, the substrate of the complex (Esler et al. 2002). A model for γ-secretase, emerging from these and other studies (Esler et al. 2002; Berezovska et al. 2003; Kornilova et al. 2003) is that the substrate-binding site is distinct from the catalytic active site and that, once the substrate is ‘docked’, it is subsequently displaced to the catalytic site of the enzyme (Tian et al. 2003). This model was independently proposed and developed in studies investigating the binding sites of APP and telencephalin on the presenilin protein (Annaert et al. 2001). In addition, the Nct ectodomain provides a separate substrate docking site (Shah et al., 2005). Thus we suggest that the clinical mutations could cause loss of function not only by interfering with the catalytic efficiency of the protease, but also by subtly modifying the docking of substrates and/or by interfering with the mechanism that moves the substrate towards the catalytic site. This would explain the diverse effects on APP and other substrates of the PS mutants studied in this work. It is also possible that the mutations affect the stochiometry of the γ-secretase complex. Some preliminary evidence for this suggestion comes from our blue native electrophoresis experiments that indicate subtle alterations in complex generation with some of the clinical mutants, mainly reflected by the presence of ∼220-kDa complexes and not unbound Nct (Fig. 1).
Finally, we observed that PS2 and PS1 act differently with regard to Aβ generation. A previous report showing that Ps1-mediated γ-secretase activity displays considerably higher specific activity than Ps2-associated γ-secretase (Lai et al. 2003) is in line with this observation. It is probabe that different γ-secretase complexes exert different biological and biochemical activities (De Strooper 2003; Hebert et al. 2004; Shirotani et al. 2004).
We would like to thank Christine Van Broekhoven and Samir Kumar-Sing for providing us with cDNA of presenilin mutants A246E and G384A and for their help in setting up the ELISA assay, Christian Haass for the cDNA of PS1-L166P mutant, Wim Annaert and Guido David for monoclonal antibodies, Pascal Merchiers and Kurt Spittaels (Galapagos, Mechelen, Belgium) for APP695Sw and Notch-ΔE recombinant adenovirus, and Bart Landuyt for his help with SELDI time-of-flight experiments.
This research was supported by a Pioneer award from the Alzheimer's Association (BDS), the Fund for Scientific Research, Flanders the K. U. Leuven Geconcerteerde Onderzoeksactie, the European Union (APOPIS: LSHM-CT-2003-503330) and the Federal Office for Scientific affairs, Belgium (Interuniversitaire attractiepool P5/19).