K. Endres, Institute of Biochemistry, Johannes Gutenberg-University, Johann-Joachim-Becherweg 30, 55128 Mainz, Germany Fax: +49 6131 3925348 Tel: +49 6131 3926182 E-mail: email@example.com
Although ADAM10 is a major α-secretase involved in non-amyloidogenic processing of the amyloid precursor protein, several additional substrates have been identified, most of them in vitro. Thus, therapeutical approaches for the prevention of Alzheimer’s disease by upregulation of this metalloproteinase may have severe side effects. In the present study, we examined whether the ErbB receptor ligand neuregulin-1, which is essential for myelination and other important neuronal functions, is cleaved by ADAM10. Studies with β- and γ-secretase inhibitors, as well as with the metalloproteinase inhibitor GM6001, revealed an inhibition of neuregulin-1 processing in human astroglioma cell line U373; however, specific RNA interference-induced knockdown of ADAM10 remained without effect. In vivo investigations of mice overexpressing either ADAM10 or dominant negative ADAM10 showed unaltered cleavage of neuregulin-1 compared to wild-type animals. As a consequence, the myelin sheath thickness of peripheral nerves was unaffected in mice with altered ADAM10 activity. Thus, although the β-secretase BACE-1 acts as a neuregulin-1 sheddase, ADAM10 does not lead to altered neuregulin-1 processing either in cell culture or in vivo. Adverse reactions of an ADAM10-based therapy of Alzheimer’s disease due to neuregulin-1 cleavage are therefore unlikely.
Neuregulin-1 (NRG-1) belongs to a family of growth factors that transduce cellular signals by binding to ErbB receptors [1,2]. At least sixteen different gene products of NRG-1 have been identified [3,4], which display a wide range of functions in the developing as well as in the adult organism. Besides organs such as the heart [5,6] or breasts , certain isoforms of NRG-1 mediate important properties in the central and peripheral nervous system: synapse formation  and transmission , expression of neurotransmitter receptors [10–12] and synaptic plasticity . Additionally, general features of neurones or Schwann cells, such as proliferation, differentiation, migratory processes and regeneration, depend on NRG-1 activity [14–17].
Although some of these functions are restricted to the developing embryonic brain, expression of NRG-1 or at least some of its isoforms [18,19] and the ErbB receptors [13,17,20,21] persists throughout the adult rodent and human nervous system. Within hippocampal synapses of adult mice, for example, NRG-1β is implicated in activity dependent remodulation by reversing long-term potentiation . Moreover, it induces neurite extension and arborization of primary cultures derived from adult murine hippocampi . There is substantial genetic evidence that single nucleotide polymorphisms of NRG-1 are associated with the pathogenesis of schizophrenia [23–25]. In addition, NRG-1 has been found to be involved in the pathogenesis of other diseases such as multiple sclerosis [26,27] or breast cancer .
How proteins derived from the gene for NRG-1 fulfil their different functions exactly remains elusive: isoforms of type I and III exist as transmembrane forms or can be proteolytically processed [8,28–30] to release soluble fragments. It is not known in detail whether the transmembrane protein or its proteolytic products are mainly responsible for the different functions. Because recombinant soluble NRG-1 often is sufficient to induce morphological or biochemical phenotypes [21,31] and shedding of NRG-1 is activity dependent, as shown for electrically stimulated neurones , an important role of the cleavage fragments is implicated.
The proteinases involved in NRG-1 proteolysis have been partly characterized: cleavage by the amyloid precursor protein (APP)-processing γ-secretase [32,33] and, more recently, β-secretase BACE-1 [29,34] was analyzed both in vitro as well as in vivo. Additional data have also been reported with respect to metalloproteinase-derived proteolysis of NRG-1 isoforms. For example, ADAM19 was shown to participate in NRG-1-β shedding, whereas NRG-1-α2 was not affected by coexpression of this enzyme . Cleavage of the α2 isoform of NRG-1, on the other hand, was impaired in fibroblasts with catalytically inactive ADAM17 .
β- and γ-secretase are responsible for processing of the Alzheimer associated APP and its paralogues amyloid precursor-like protein (APLP) 1 and APLP2 . Furthermore, the distribution of NRG-1 and the localization of its receptor ErbB4 have been found to be altered in Alzheimer’s disease patients [19,37] and a mouse model of the disease . ADAM10 was found to act as α-secretase in vitro and in cultured cells [38,39]. It competes with BACE-1 for the substrate APP and is able to prevent the formation of Aβ plaques in a mouse model of the disease . Moreover, ADAM10 restores long-term potentiation and increases cognitive function in transgenic mice [40,41] and enhances cortical synaptogenesis . Due to the overlap of substrate specificity of BACE-1 and ADAM10 with respect to substrates such as APP or the APLPs and partial phenotypic overlap of ADAM10 and NRG-1 knockout mice [43–45], it was considered important to investigate the possible role of ADAM10 in NRG-1 processing in cells and in the living animal.
Identification of NRG-1 isoforms expressed in the human astroglioma cell line U373
The expression and processing of NRG-1 was described previously in different astroglioma cell lines . Therefore, we chose the human astroglioma cell line U373 to examine the relevance of ADAM10 for NRG-1 shedding. The investigated cell line stably overexpresses the human neuron specific APP isoform 695 to provide an appropriate control substrate for α- as well as β- and γ-secretases [39,46].
Recently, sixteen different isoforms of NRG-1 generated by alternative promoter usage, transcription initiation sites or splicing [4,47,48] have been described, and a wide variety of these isoforms are found in brain-derived cell types [49–51]. To characterize the isoforms present in the U373 cell line, we performed RT-PCR with domain specific primers [52,53]. Type I as well as type III specific PCR products (schematically shown in Fig. 1A: immunoglobulin-like domain and glycosylation site or cysteine rich domain sequences) were produced, whereas those characteristic of the type II Kringle domain coding region were not detectable (Fig. 1B). Both α- as well as β-type indicating PCR products were generated. We amplified the juxtamembrane region coding sequence with primers independent of α- or β-type (primers jD_for and TM_rev; Table 1) and subcloned the resulting DNA fragments into pUC19. Sequencing analysis revealed that seven out of eight clones were the α2-type, whereas only one was identified as β2 (for sequences, see Fig. 1B; NM_013964 and NM_13957). Hence, the predominant NRG-1 isoform present in U373 cells is the α2-type with respect to the juxtamembrane region. Because ADAMs such as ADAM10 cleave their substrates in close proximity to the membrane, knowledge of this region of the putative substrate NRG-1 in the investigated cell line was mandatory.
Table 1. Primer sequences used for NRG-1 isoform analysis.
On the protein level, the NRG-1 antibody against the C-terminal domain detected a prominent band of approximately 90 kDA in the lysate of U373 cells (Fig. 1C). This is consistent with full length NRG-1 in its glycosylated state  for a panel of glioma cells. Protein bands with a higher molecular weight might indicate the immature proform and the band with a reduced molecular weight might represent an incompletely glycosylated intermediate. Additionally, between 40 and 50 kDa, two C-terminal fragments were detectable. After serum free incubation of cells for 4 h, a band of approximately 60 kDa (N-terminal fragment) was detected in cell supernatants using the pan-NRG-1 antibody against the ectodomain, which recognizes α- as well as β-isoforms (Fig. 1C). This protein reveals a slightly lower molecular weight compared to the results obtained by Ritch et al.  where secreted NRG-1 had a molecular mass of 70 kDa. Because at least two Asn residues and 11 Thr/Ser residues were identified as potential sites for N- or O-glycosylation of NRG-1 , the deviation in the size of the soluble protein fragment may depend on different glycosylation patterns in the investigated cell lines. In mouse brain membranes, a panel of proteins in the approximate range of 160–70 kDa was observed (Fig. 1C), which corresponds to NRG-1 species described for mouse brain as well as human brain material . Similar to cell supernatants, the soluble fraction of mouse brain contained a secreted form of NRG-1 of 55–60 kDa.
Proteolytical processing of NRG-1 in the human astroglioma cell line U373
The tripeptidic β-secretase inhibitor II led to an 80% reduction of soluble β-secretase cleaved APP (APPs-β) in cell culture supernatants and also diminished significantly the 60 kDa soluble NRG-1 in cell conditioned medium (N-terminal fragment; Fig. 2A). Furthermore, the γ-secretase inhibitor N-[N-(3,5-difluorophenacetyl)-l-alanyl ]-S-phenylglycine t-butyl ester (DAPT), which induced accumulation of C-terminal fragments of APP in cell lysates (C-terminal fragment; Fig. 2B), increased the NRG-1 C-terminal fragment of approximately 50 kDa six-fold as compared to solvent-treated cells. These results demonstrate cleavage of NRG-1 in U373 cell line by both β- and γ-secretase.
Phorbol 12-myristate 13-acetate, a known inducer of shedding events, significantly elevated soluble APPs-α, as well as the soluble N-terminal fragment of NRG-1 in those cells, by 200% and 150% (Fig. 2C). For this reason, we analyzed metalloproteinase dependent shedding of NRG-1: APPs-α that acted as a control was reduced to 40% by the broad spectrum metalloproteinase inhibitor GM6001 as compared to solvent-treated cells, and NRG-1 cleavage also was reduced significantly, although to a lower extent (65% of control cells; Fig. 2C). Therefore, metalloproteinases appear to be involved in NRG-1 processing in the astrocytoma cell line U373, as previously described for other cell lines [28,55].
RNA interference (RNAi)-induced knockdown of ADAM10 has no influence on NRG-1 shedding
Because GM6001, which was used for inhibitory studies, is a broad spectrum inhibitor of MMPs as well as ADAMs, we chose the RNAi approach to analyze in particular the role of ADAM10 in NRG-1 cleavage. As a control for unspecific RNAi-induced effects, MMP2 knockdown was examined as well. RNAi treatment targeted against endogenous ADAM10 of the U373 cells resulted in a 60% reduction of mature ADAM10, whereas MMP2 targeted oligomers had no influence (Fig. 3A). The decrease of ADAM10 due to RNAi was accompanied by a 30% decrease in APPs-α shedding (Fig. 3B) serving as an internal control. Because APP is not only a substrate for ADAM10, but also, for example, TACE, the absolute effect of ADAM10 knockdown was small but reached significance. By contrast, for NRG-1, we observed no alteration of soluble NRG-1 in cell culture supernatants, as well as for the membrane-bound protein species. Because of a potential compensation of reduced NRG-1 cleavage by other secretases, we cannot exclude the possibility that ADAM10 might have an effect on proteolytic processing of NRG-1 in U373 cells but, if this is the case, ADAM10 at least is not a major sheddase of this protein (Fig. 3C).
In vivo effect of ADAM10 on NRG-1 proteolysis
Because ADAM10 was not implicated in the shedding of distinct NRG-1 isoforms of cultured human astroglioma cells (α2 and β2; Fig. 1B), we analyzed NRG-1 processing in ADAM10 overexpressing mice to take into account all of the expressed isoforms. Two transgenic mouse lines with different expression levels of ADAM10 (moderate, ADAM10mo; high, ADAM10hi) and a mouse line transgenic for a dominant negative ADAM10 mutant (ADAM10dn) were included in this investigation. All mouse lines have been examined in detail elsewhere with respect to APP processing, learning and behaviour [40,41,56]. The expression of the proteinase itself is illustrated in Fig. 4A (lower part). In soluble protein fractions of brains derived from the three transgenic lines, the amount of the N-terminal fragment of NRG-1 (approximately 60 kDa; Fig. 4) was not changed compared to the wild-type. Additionally, neither full length NRG-1, nor C-terminal fragments in the brain membrane fraction were influenced by an altered ADAM10 amount or activity (Fig. 4). We therefore conclude that, in vivo, the proteolytic processing of NRG-1 does not depend on the α-secretase ADAM10.
For further confirmation of these findings, we examined the myelination of peripheral nerves in ADAM10 transgenic mice and mice overexpressing dominant negative ADAM10. Because myelination strongly depends on NRG-1-ErbB signalling of Schwann cells and neurones , any relevant change of this pathway induced by altered ADAM10 activity should be observable as a physiological consequence. Again, the ADAM10 transgenes remained without effect in all investigated mouse lines (Fig. 5A). G-ratios of ADAM10mo as well as of ADAM10dn mice at postnatal day 17 were identical to nontransgenic littermates. Furthermore, Akt-phosphorylation, which also is partly controlled by NRG-1 signaling [31,58,59], was unaffected in both mouse lines (Fig. 6). In adult mice with a high expression level of ADAM10 (ADAM10hi), G-ratios were also unaltered (Fig. 5B), but tomacula-like structures (local myelin thickenings ) were observed. Additionally, in the mouse line with higher ADAM10 expression (ADAM10hi), Akt-phosphorylation was significantly reduced to 40% compared to wild-type mice. This probably reflects effects that do not depend on NRG-1 cleavage.
The data obtained in the present study for the human astroglioma cell line U373 clearly reveal BACE-1 and γ-secretase dependent shedding of the endogenous ErbB receptor ligand, which we identified predominantly as type α2-NRG-1 and, to a lesser extent, as the β2 isoform. Additionally, GM6001, a broad spectrum metalloproteinase inhibitor, was able to reduce NRG-1 shedding but a specific knockdown of ADAM10 by RNAi remained without any effect within the cellular system. Therefore, the present study demonstrates that ADAM10 is not a major sheddase of neuregulin-1 and enhancement of ADAM10 will probably have no side effects due to NRG-1 cleavage.
Because catalytically active ADAM10 is found on the plasma membrane  and neuregulin-1β1 cleavage, for example, is restricted to the Golgi apparatus , it is plausible that distinct localization of ADAM10 and NRG-1 might inhibit a functional substrate–proteinase interaction. Furthermore, NRG-1 is mainly found in cholesterol rich lipid rafts [61,62], favouring its role as a BACE-1 substrate, whereas ADAM10 and its catalytic activity (at least for APP) were shown to be localized in cholesterol-poor nonraft regions of the membrane . Nevertheless, a possible in vivo relevance of ADAM10 to NRG-1 shedding required investigation due to the fact that NRG-1 proteolysis also depends on, for example, electric stimulation of cells , which might be accompanied by translocation within the cell. Furthermore, the animal model offers a more complex representation because of the wide variety of cells that express NRG-1 and which might interact.
Reconstitution experiments with transfection of ADAM10 in ADAM17−/− embryonic mouse fibroblasts  suggested only a minor influence of ADAM10 on neuregulin-1 shedding, but any positive proof in the living animal was still missing. Therefore, we investigated neuregulin-1 processing in mice with postnatal expression of ADAM10 or its dominant negative variant. The Thy.1-promoter driven expression of both ADAM10-constructs  occurs at postnatal day 1 (data not shown). Thy.1-based expression in general is predominantly found in postmitotic neurones of the perinatal period, but also occurs in dorsal root ganglia and in spinal cord . Hence, the animal model is sufficient to study early ontogenetic phenomena after birth without disturbances due to impeded embryonic development in the central or peripheral nervous systems.
In a recent study, the age-dependency of NRG-1 cleavage by BACE-1 was demonstrated . Although the accumulation of full length neuregulin-1 in BACE−/− mice aged 15 days confirmed previous data [29,34], mice at postnatal day 30 or even older (2 years) showed no abnormalities with respect to neuregulin-1 processing. In the case of ADAM10, investigations of adult mice moderately overexpressing ADAM10 or its dominant negative variant resulted in totally unchanged amounts of NRG-1 processing products.
Therefore, the influence of ADAM10 on NRG-1 was additionally analyzed in young mice (postnatal day 17) by the status of peripheral nerve ensheathment. In the second postnatal week, myelination is almost finished in mice (central nervous system ; peripheral nervous system ); therefore, alterations should be apparent. However, neither moderate ADAM10 overexpressing mice, nor mice with a restriction of enzyme activity by dominant negative ADAM10, revealed differences in axon myelination parameters compared to wild-type littermates at postnatal day 17.
Surprisingly, adult mice with high levels of ADAM10 overexpression showed myelin infoldings (tomacula-like structures). This observation has not been made in the context of reduced or enhanced NRG-1 signalling in mice . We therefore suggest that mechanisms beside NRG-1 signal transduction might be responsible for the neuropathological phenotype. It will be interesting to analyze these observations in future studies.
Additionally, Akt phosphorylation, a consequence of NRG-1-ErbB signalling [31,59], was unaltered by moderate overexpression or by inhibiting ADAM10 activity by its negative mutant form. However, mice with high overexpression of ADAM10 showed a strong decrease of phosphorylated Akt compared to nontransgenic mice. This observation may relate to recent findings demonstrating that a high level of ADAM10 overexpression in the mouse increases susceptibility to kainate-induced seizures and neuronal damage , whereas the neuroprotective properties of ADAM10 were only evident in mice with APP overexpression.
In conclusion, ADAM10 was excluded both in cell culture and in the animal model as a major candidate secretase for neuregulin-1 shedding. We cannot rule out that other secretases, such as TACE or ADAM19, which were identified formerly as NRG-1 sheddases [28,30,55], compensate for the lack of ADAM10 in RNAi-treated cells or animals with overexpression of the dominant negative mutant. In breast cancer cells, ADAM10 was described to mediate the shedding of the receptor ErbB2 ; therefore, an influence on NRG-1-ErbB signalling could in principal have also occurred by ErbB2 cleavage in our transgenic mice. However, because we did not observe an influence on myelination in ADAM10 transgenic mice, this observation might be restricted to tumour cells.
We also cannot exclude that, in non-neuronal tissue, embryonic development or pathological stages ADAM10 itself, or cleavage products of its other substrates, might be involved in NRG-1-ErbB cross-talk. In summary, however, we present evidence demonstrating that, in the healthy early postnatal and adult mouse, moderate alterations in the amount of ADAM10 do not interfere with neuregulin-1 signalling. Accordingly, ADAM10 will have no impact on downstream physiological functions such as nerve remyelination or the schizophrenia-resembling psychiatric changes as observed for BACE-1 knockout mice [34,69]. The results obtained in the present study therefore suggest that a moderate upregulation of ADAM10 expression and its α-secretase activity with a preventive or therapeutical intention is not impaired by side effects resulting from the NRG-1-ErbB signalling network.
Antibodies, inhibitors and RNAi oligomers
The primary antibodies used were: 6E10 for the detection of APPs-α (Senetek, St Louis, MO, USA; dilution 1 : 1000), anti-neuregulin-1 (H-210; dilution 1 : 200) for the detection of secreted NRG-1 fragments, anti-neuregulin-1α/β1/2 (C-20; dilution 1 : 500) (both Santa Cruz Biotechnology, Santa Cruz, CA, USA) for the detection of membrane bound NRG-1, anti-ADAM10 (Chemicon, Temecula, CA, USA; dilution 1 : 1000) for the detection of the proteinase in cells and 6687 (C. Haass, LMU Munich, Germany) for the detection of full length APP and C-terminal protein fragments. Anti-P-Akt and anti-total Akt were purchased from Cell Signaling [PhosphoPlus Akt (Ser473) Antibody Kit; Cell Signaling, Danvers, MA, USA]. Overexpressed ADAM10 in mouse brain membranes was visualized by HA-antibody Y-11 (Santa-Cruz Biotechnology). The secondary antibodies were coupled to alkaline phosphatase (Tropix, Bedford, MA, USA; dilution 1 : 10000) or horseradish peroxidase (Pierce, Rockford, IL, USA; dilution 1 : 3000) and were used in combination with their substrates CDP-Star (Tropix) or SuperSignalECL (Pierce).
The β-secretase-inhibitor II (Calbiochem, Bad Soden, Germany) was applied at a concentration of 25 μm and the γ-secretase inhibitor DAPT (B. Schmitt, Clemens Schöpf-Institute of Organic Chemistry and Biochemistry, Technische Universität Darmstadt, Germany) was applied at a concentration of 2 μm. GM6001 (Calbiochem, San Diego, CA, USA) was used at a final concentration of 10 μm and phorbol 12-myristate 13-acetate (Sigma, Deisenhofen, Germany) was used at a concentration of 1 μm. All substances were dissolved in dimethylsulfoxide as stock solutions.
For the RNAi experiments, the Stealth RNAis ADAM10 HSS165, HSS166, HSS167 and MMP2 HSS106612, HSS106613, HSS106614 (Invitrogen, Karlsruhe, Germany) were used. Transfections were performed with Opti-MEM and Lipofectamine2000 (Invitrogen).
RNA preparation and RT-PCR
The RNA of U373 cells was isolated by using confluent 6 cm culture plates and the RNA isolation kit with on-column DNA digestion as recommended in the manufacturer’s protocol (Macherey-Nagel, Düren, Germany). Four hundred nanograms of RNA were reverse transcribed in a 20 μL reaction volume by the reverse-it-RT-PCR-kit from ABgene (Hamburg, Germany) with intron-spanning specific primers (0.2 μm each). Amplificates were analyzed on 1% agarose gels and GAPDH amplification served as a control for the RT-PCR reaction and PCR conditions. Primer sequences and the amplificate length are provided in Table 1.
The amplificates from RT-PCR with primers NRG-jD_for and NRG-TM_rev were ligated into pUC19 (Fermentas, St Leon-Rot, Germany) and plasmid DNA from eight positive clones (identified by blue–white selection) was sequence analyzed with M13 universal primer.
Preparation of mouse brain samples
The generation of transgenic mice has been described previously . Seven- to 10-week-old mice were sacrificed, the brains were dissected and stored on dry ice. Ice-cold Tris buffer (20 mm Tris/HCl, pH 8.5) containing proteinase inhibitors (Inhibitor Complete Mini; Roche Diagnostics Corp., Mannheim, Germany) was added and tissue was homogenized in a tissue lyser (Eppendorf, Hamburg, Germany) with a frequency of 20 Hz for 2 min. The supernatants resulting from centrifugation at 13 500 g for 1.75 h were used for the detection and quantification of soluble NRG-1. NRG-1 membrane bound full length protein and membrane bound NRG-1 fragments, as well as the proteinase ADAM10 itself, were detected in the membrane fractions prepared from centrifugation pellets.
For total-Akt and phospho-Akt detection, brain hemispheres were homogenized in lysis buffer supplemented with additional phosphatase inhibitors (2.5 mm Na-pyrophosphate, 1 mmβ-glycerophosphate and 1 mm Na3VO4) and soluble proteins were isolated from membrane fractions by centrifugation at 20 800 g for 20 min.
Cell culture, treatment with inhibitors and RNAi silencing
U373 cells overexpressing human wild-type APP were maintained in MEM (Sigma, Taufkirchen, Germany) supplemented with 10% fetal bovine serum, 1% sodium pyruvate and 1% glutamine.
To inhibit β-secretase or metalloproteinases, cells were pretreated with the appropriate inhibitor (β-secretase-inhibitor II or GM6001) for 22 h. Then, the serum containing culture medium was removed and medium without serum supplemented with fatty acid free BSA (1 mg·mL−1) and fresh inhibitor was added for an additional 4 h. Phorbol 12-myristate 13-acetate-induced shedding was performed for 4 h in serum free BSA supplemented medium. For inhibition of γ-secretase, cells were treated with DAPT for 48 h in culture medium.
For RNAi experiments, cells were transfected with 750 pmol of RNAi oligomers (250 pmol each) in six-well plates. After 5 h of transfection, the medium was replaced by culture medium. After 44 h, cells were covered with serum-free medium and incubated for 4 h for collection of secreted proteins and cell lysates.
Proteins of cell culture supernatants were precipitated with trichloroacetic acid and normalized by the protein content of the cell lysates. Adequate amounts of soluble or membrane tethered proteins were separated on 8% SDS-gels or 4–12% Bis–Tris NuPAGE gels (Invitrogen) and blotted on a poly(vinylidene difluoride) membrane or nitrocellulose (total-Akt and phospho-Akt). Proteins were then detected with the appropriate primary antibodies. Chemiluminescent signals from alkaline phosphatase or horseradish peroxidase coupled secondary antibodies were visualized with a charge-coupled device camera and the software versa doc (Bio-Rad, Munich, Germany) and were quantified with aida 3.5 (Raytest, Straubenhardt, Germany).
Electron microscopy and G-ratio determination
Seventeen-day-old or adult transgenic mice (15–17 months old) and nontransgenic littermates were perfused with NaCl/Pi containing 60 μg·mL−1 heparin followed by 2.5% glutaraldehyde/4% parafomaldehyde in 0.1 m phosphate buffer. Afterwards, sciatic nerves were removed, postfixed and contrasted with osmium tetroxide and processed for electron microscopy . biovision software (Soft Imaging System GmbH, Münster, Germany) was used for determination of the G-ratio by measuring the inner and outer myelinated fibres.
All animal procedures were performed according to the German guidelines for the care and the use of laboratory animals and in accordance with the European Communities Council Directive of 24 November 1986 (86/609/EEC).
We thank C. Griffel (MDC, Berlin) for excellent technical support in the analysis of sciatic nerve myelination; A. Schröder (ZVTE, Mainz) for coordination of animal husbandry and M. Willem (LMU, Munich) for fruitful discussion. This work was supported by grants from the DFG priority program 1040 (to F.F.) and from the NGFN integrated consortium ‘Gene Identification and Functional analyses in Alzheimer’s disease’ funded by the Federal Ministry of Education and Research (to A.N.G.).