A Pathogenic Presenilin-1 Deletion Causes Abberrant Aβ42 Production in the Absence of Congophilic Amyloid Plaques*

Familial Alzheimer's disease (FAD) is frequently associated with mutations in the presenilin-1 (PS1) gene. Almost all PS1-associated FAD mutations reported so far are exchanges of single conserved amino acids and cause the increased production of the highly amyloidogenic 42-residue amyloid β-peptide Aβ42. Here we report the identification and pathological function of an unusual FAD-associated PS1 deletion (PS1 ΔI83/ΔM84). This FAD mutation is associated with spastic paraparesis clinically and causes accumulation of noncongophilic Aβ-positive “cotton wool” plaques in brain parenchyma. Cerebral amyloid angiopathy due to Aβ deposition was widespread as were neurofibrillary tangles and neuropil threads, although tau-positive neurites were sparse. Although significant deposition of Aβ42 was observed, no neuritic pathology was associated with these unusual lesions. Overexpressing PS1 ΔI83/ΔM84 in cultured cells results in a significantly elevated level of the highly amyloidogenic 42-amino acid amyloid β-peptide Aβ42. Moreover, functional analysis in Caenorhabditis elegans reveals reduced activity of PS1 ΔI83/ΔM84 in Notch signaling. Our data therefore demonstrate that a small deletion of PS proteins can pathologically affect PS function in endoproteolysis of β-amyloid precursor protein and in Notch signaling. Therefore, the PS1 ΔI83/ΔM84 deletion shows a very similar biochemical/functional phenotype like all other FAD-associated PS1 or PS2 point mutations. Since increased Aβ42 production is not associated with classical senile plaque formation, these data demonstrate that amyloid plaque formation is not a prerequisite for dementia and neurodegeneration.

Alzheimer's disease (AD) 1 is an age-dependent neurogenerative disorder. Although most AD cases occur sporadically, autosomal dominant inheritance has been recorded in numerous families (1). Mutations in four genes have been mapped to familial AD (FAD). These include the genes encoding the ␤-amyloid precursor protein (␤APP), presenilin 1 (PS1), PS2 (1), and ␣ 2 -macroglobulin (2). Functional analysis revealed that ␤APP and PS mutations affect endoproteolytic processing of ␤APP in a very similar manner. In the amyloidogenic pathway, ␤APP is first cleaved at the N terminus of the A␤ domain by the recently identified ␤-secretase (3). This generates a membraneretained C-terminal fragment, which is the substrate for the ␥-secretase. ␥-Secretase cleaves its substrate within the membrane, which results in the physiological secretion of A␤ (1). About 90% of secreted A␤ terminates at amino acid 40 (A␤40), while most of the remaining A␤ peptides are elongated by two amino acids (A␤42). The rare A␤42 appears to aggregate much faster than A␤40 (4,5) and is therefore the major constituent of senile plaques (6,7). FAD-associated mutations found within ␤APP, PS1, and PS2 all cause the increased production of this highly amyloidogenic A␤ variant and therefore increase the kinetics of A␤ aggregation and of its deposition in congophilic senile plaques (1).
PS proteins not only affect the ␥-secretase cleavage in FAD cases but are also required for physiological A␤ generation, since a PS1 ablation results in a dramatically reduced A␤ production (8). Moreover, mutagenesis of two critical aspartate residues located within transmembrane domains 6 and 7 (TM6 and -7) also results in an inhibition of A␤ generation (9). Similar mutations in human PS2 also reduce A␤ generation (10,11), and the critical aspartate residues are functionally conserved during evolution (12). In all cases, inhibition of PS function not only reduced A␤ generation but also concomitantly increased the corresponding membrane-retained ␤APP C-terminal fragments, which are the immediate precursors for A␤ generation. Since two critical aspartate residues are required within the catalytic center of aspartyl proteases and since ␥-secretase function can be blocked by aspartyl protease inhibitors (13), it was recently claimed that PS proteins may be identical with the ␥-secretase (14).
PS proteins not only support the intramembraneous endo-proteolysis of ␤APP but are also required for the similar cleavage of Notch (10,(15)(16)(17). The endoproteolytic cleavage of Notch appears to be required for the generation of the Notch intracellular cytoplasmic domain (18), which translocates to the nucleus, where it is involved in transcriptional regulation (19). A function of PS in Notch signaling is also supported by the phenotypes observed in various PS1/PS2 deletions in mice (20 -23), which resemble that observed upon the deletion of the Notch gene. Moreover, several mutant alleles of the Caenorhabditis elegans PS homolog sel-12 cause an egg-laying phenotype, which is due to a functional deficit in Notch signaling (24). The failure in Notch signaling in worms can be functionally rescued by transgenic expression of human PS1 or PS2 (10,25,26). FAD-associated PS mutations occur frequently within the PS1 gene and are associated with the most aggressive AD phenotype (27). Out of the numerous PS mutations described to date, only three deletions (28 -31) have been observed so far. However, none of the deletions are directly associated with a pathological function. We have shown previously that the pathological activity of the PS1 ⌬exon9 splicing mutation (28) is independent of the large deletion and rather due to a single amino acid exchange at the aberrant splice junction at codon 290 (32). A genomic deletion of the exon 9-encoded domain (⌬exon9 Finn; see below) was reported as well (30). However, due to the aberrant splicing of exon 8 with exon 10, the same amino acid exchange is introduced at codon 290 as observed in the original PS1 ⌬exon9 splicing mutation. Therefore, the amino acid sequence of PS1 ⌬exon9 Finn is identical to the PS1 ⌬exon9 splicing mutation. In analogy to the PS1 ⌬exon9 splicing mutation (32), it would therefore be expected that this mutation (PS1 ⌬exon9 Finn) produces A␤42 independent of the exon 9 deletion. The unusual genomic exon 9 deletion of PS1 in the Finnish pedigree is associated with Alzheimer's disease and spastic paraparesis. In contrast to all other AD cases, these patients as well as the patients with the PS1 ⌬exon9 splicing mutation develop "cotton wool" plaques, which lack a congophilic dense core and plaque-related neuritic pathology (30). 2 Finally, the deletion produced by the intron 4 mutation of PS1 could not be associated with an increased A␤42 production (29). It rather turned out that a single amino acid insertion, which is generated by aberrant splicing, is responsible for the pathological activity of this mutation (29). Therefore, no PS1 deletion has so far been associated with increased A␤42 generation.
We have now analyzed the function of a novel PS1 deletion (PS1 ⌬I83/⌬M84; Fig. 1), which is also associated with early onset AD and spastic paraparesis. A potentially pathological function in A␤ generation and Notch signaling was specifically investigated. We found that PS1 ⌬I83/⌬M84 causes increased A␤42 production like all other FAD-associated PS1/PS2 mutations. The PS1 ⌬I83/⌬M84 mutation is associated with A␤ deposition in noncongophilic cotton wool plaques, widespread cerebral amyloid angiopathy, neurofibrillary tangles, and neuropil threads, although tau-positive abnormal neurites are rare.
Histology and Immunohistochemistry-Brains from the patient with the PS1 ⌬I83/⌬M84 mutation and from a patient with a PS1 T115C mutation were collected at postmortem and fixed in 10% formalin in phosphate-buffered saline. Blocks from the major anatomical areas, including the hippocampal formation, were processed in paraffin wax. Tissue sections were stained with hematoxylin and eosin and Bielschowsky's silver impregnation methods. Congo red and thioflavine S methods were used to detect A␤ deposits in ␤-sheet conformation. For immunohistochemistry, 4-, 7-, or 20-m sections were deparaffinized in xylene and rehydrated using graded alcohols. For PHF1, AT8, and CR3/43 immunohistochemistry, sections were pretreated in a microwave oven in sodium citrate buffer for 20 min, for GFAP immunohistochemistry in trypsin for 10 min, and for A␤, A␤40, and A␤42 immunohistochemistry in formic acid for 10 min followed by treatment in a pressure cooker in citrate buffer for 10 min. After washes in phosphatebuffered saline and 10% milk, sections were incubated with the PHF1 antibody at 4°C or with the GFAP, AT8, CR3/43, A␤, A␤40, and A␤42 antibodies at room temperature. Detection of antibody binding was either performed with the ABC or the alkaline phosphatase anti-alkaline phosphatase system (DAKO) according to the manufacturer's instructions. Either diaminobenzidine/H 2 O 2 or neufuchsin was used as chromogen.
Genetic Analysis and cDNA Encoding PS1 ⌬I83/⌬M84 -Genomic DNA and mRNA were extracted from blood and frozen brain, respectively. All exons of the PS1 gene were analyzed by polymerase chain reaction amplification of genomic DNA and Big Dye sequencing. Sequence analyses of PS1 exon 4 revealed a heterozygous deletion of ATCATG at codons 83 and 84 (isoleucine-methionine) of the gene. The corresponding cDNA encoding PS1 ⌬I83/⌬M84 was cloned into pcDNA3.1-zeo(ϩ) expression vector (Invitrogen).
Analysis of PS by Combined Immunoprecipitation/Western Blotting-Cell lysates from stably transfected K293 cells were prepared and subjected to immunoprecipitation using the polyclonal antibody 3027 to PS1 or 3711 to PS2 (32). Following gel electrophoresis, immunoprecipitated PS proteins were identified by immunoblotting using the monoclonal antibody BI.3D7 (PS1) or BI.HF5c (PS2) (32). Bound antibodies were detected by enhanced chemiluminescence (ECL; Amersham Pharmacia Biotech).
Analysis of A␤ by ELISA-Conditioned media (2 ml) were collected from confluent K293 cells in six-well dishes for 24 h. The media were assayed for A␤40 and A␤42 according to a previously described enzymelinked immunosorbent assay (36).
Transgenic Lines of C. elegans and Rescue Assays-To construct a new sel-12 expression vector, a 3.0-kilobase pair fragment of cosmid C08A12 was amplified by polymerase chain reaction using primers CCC GGC TGC AGC TCA ATT ATT CTA GTA AGC and GTC TCC ATG GAT CCG AAT TCT GAA ACG TTC AAA TAA C and cloned into pPD49.26 (37). The resulting plasmid contains only nontranscribed sequences from the 5Ј region of the C. elegans sel-12 gene. PS1 derivatives were cloned into this vector as a BamHI/SalI fragment. Transgenic lines were established by microinjection of plasmid DNA mixtures into the C. elegans germ line to create extrachromosomal arrays (26). Four independent lines from the progeny of F2 generation animals were established. Since the sel-12(ar171) animals never lay eggs (26), rescue of the sel-12 defect can be quantified by scoring egg-laying behavior in transgenic animals (26). 50 transgenic animals of each line were analyzed for their ability to lay eggs. The numbers of eggs laid by individual transgenic animals were counted and placed into four categories: Eglϩϩϩ, robust egg laying, more than 30 eggs laid; Eglϩϩ, 15-30 eggs laid; Eglϩ, 5-15 eggs laid; EglϪ, 0 -5 eggs laid. 2 H. Houlden and J. Hardy, unpublished data.

Genetic Analysis of a Family with Autosomal Dominant
Early Onset AD with Spastic Paraparesis-Several families were identified by Houlden et al. (38) with autosomal dominant early onset AD with spastic paraparesis. Here we present the detailed pathological, biochemical, and functional analysis of one of these families. Neuropathological examination of a large Scottish family (Fig. 1A) revealed the presence of large cotton wool plaques (see below) similar to those seen in the Finnish family (30,39). Sequencing revealed the presence of an exon 4 deletion (ATC-ATG; isoleucine-methionine) of codons 83 and 84 of the PS1 gene. The mutation was not present in 100 controls. The PS1 ⌬I83/⌬M84 deletion occurs within TM1 of PS1 (Fig.  1B). TM1 may be functionally important, since other mutations were previously located in that region. Interestingly, the PS1 ⌬I83/⌬M84 deletion is located immediately C-terminal to the V82L mutation (40). Moreover, a third mutation has been observed in TM1, which results in the exchange of valine at position 96 to phenylalanine (41).
Deposition of A␤42 in Cotton Wool Plaques-Neuropathological investigation of the PS1 ⌬I83/⌬M84 case by hematoxylin/ eosin staining, Bielschowsky's silver staining, and A␤ immunohistochemistry revealed the presence of widespread cotton wool plaques (Figs. 2 and 3). These plaques were most frequently found in the neocortex, hippocampus, and striatum. Cotton wool plaques appeared in the neuropil as round, eosinophilic, and strongly A␤-positive structures often larger than 100 m in diameter. These frequently seemed to displace other elements such as neurons, a finding readily noticeable in the hippocampus (Fig. 2, A and B). Cotton wool plaques did not generally contain amyloid, since they were negative or occasionally very weakly stained by Congo red and weakly positive with thioflavine S (Fig. 2, C-E). Cerebral amyloid angiopathy was widespread, capillaries having thickened walls, which, together with the affected arterioles, showed apple green birefringence following Congo red staining and strong fluorescence with thioflavine S (Fig. 2, C-E). Bielschowsky silver staining (Fig. 2B) and tau immunohistochemistry (Fig. 2, F and H) revealed that neurofibrillary tangle pathology was widespread in the neocortex and hippocampal formation, although the dentate fascia was spared. Fine neuropil threads were a prominent feature within the cortices, and AT8 as well as PHF1 immunohistochemistry also revealed that the cotton wool plaques contained many thread-like processes but were only rarely associated with abnormal neurites (Fig. 2, F and H). In contrast, numerous tau (Fig. 2G) and silver-positive abnormal neurites (data not shown) were seen in association with the classical plaques found in the control case with a PS1 T115C mutation (42) (Fig. 2G). GFAP immunostaining demonstrated a relatively sparse astrocytic response to the cotton wool plaques (Fig. 2I). Furthermore, no significant microglial activation (CR3/43 immunostaining) was observed in association with cotton wool plaques (Fig. 2J), in contrast to widespread activated microglia that were clustered mostly around the amyloid plaques in the PS1 T115C case (Fig. 2K).
Deposition of A␤ species ending at position 42 is believed to be closely associated with neuritic plaque formation in previously reported cases with different PS1 mutations (6, 7). In our PS1 ⌬I83/⌬M84 case, the cotton wool plaques were strongly positive not only with antibodies raised against epitopes 8 -17 FIG. 1. A FAD-associated PS1 deletion. A, three generation family with five affected members. All affected individuals had signs of AD with spastic paraplegia; this was progressive over 6 -10 years, and all cases were wheelchair bound in the later stages of the disease. Pedigree is disguised for family protection, and the numbers to the right of individual symbols refer to the age at death or current age. Arrow, sequenced cDNA of affected patient. B, schematic representation of PS1 ⌬I83/⌬M84. The deletion of amino acids Ile 83 and Met 84 in TM1 is shown as well as neighboring point mutations associated with FAD (asterisks). The black box represents the cleavage site domain of PS1.
(6F/3D immunostaining) or 1-17 (6E10 immunostaining) of the A␤ peptide (Fig. 3, A and B), but also with an end-specific antiserum to position 42 (Fig. 3D). In contrast, the cotton wool plaques were only weakly reactive for A␤40, which was found to be the predominant A␤ species deposited in blood vessels (Fig. 3C). These findings show that the cotton wool plaques are predominantly composed of A␤ ending at position 42, and, since they were also positive with the 6E10 antibody recognizing amino acids 1-17 of A␤, at least some full-length A␤1-42 is deposited (Fig. 3B).
Stable Expression of PS1 ⌬I83/⌬M84 in Human Cell Lines-To further prove the pathological activity of PS1 ⌬I83/ ⌬M84 on A␤ production, we analyzed its function in a tissue culture system, which has previously been proven to be very sensitive for the detection of abnormal A␤42 generation caused by the expression of FAD mutant presenilins (see Ref. 43 and references therein).
As described above, cotton wool plaques were so far found to be associated with two independent FAD mutations (⌬exon9 and ⌬exon9 Finn), which both affect PS1 endoproteolysis. Since the PS1 ⌬I83/⌬M84 mutation also results in a very similar pathology, we first investigated endoproteolysis of this mutant PS1 variant. cDNAs encoding PS1 ⌬I83/⌬M84, PS1 ⌬exon9, and wt PS1 were stably transfected into human embryonic kidney 293 cells expressing Swedish mutant ␤APP (34). To prove ectopic expression and endoproteolysis of the transfected PS1 derivatives, cell lysates were immunoprecipitated with antibody 3027 to the cytoplasmic loop of PS1 (32). Immunoprecipitated PS1 proteins were identified by immunoblotting using the monoclonal antibody BI.3D7 to the C terminus of PS1 (32). Consistent with previous results (44), large amounts of uncleaved PS1 holoprotein accumulated in cells expressing PS1 ⌬exon9 (Fig. 4A, upper panel). Robust amounts of PS1 C-terminal fragments (CTFs) were observed in all other cell lines including those overexpressing PS1 ⌬I83/⌬M84 (Fig. 4A, upper  panel). This demonstrates that the novel deletion mutation does not affect endoproteolysis of PS1 like the PS1 ⌬exon 9 deletion.
Reduced Facilitation of Notch Signaling-PS1 and PS2 are both required for Notch signaling (19) and functionally replace the defective C. elegans PS homolog sel-12 (10,25,26). We now expressed PS1 ⌬I83/⌬M84 in a mutant strain of C. elegans, which lacks a functional PS homolog (sel-12(ar171)). The sel12(ar171) animals show an egg-laying defective phenotype, which is due to a defect in Notch-signaling during vulva differentiation (24). This system can be used to monitor PS function in the facilitation of Notch signaling in an in vivo rescue assay by transgenic expression of the corresponding human cDNA constructs (24,26). We (10,26) and others (25) have previously shown that human wt PS1 and PS2 rescue the egg laying phenotype of the mutant worm, whereas FAD-associated PS point mutations showed a reduced rescuing activity. Consistent with previous results (25,26), transgenic expression of wt PS1 in the mutant worm lead to a rescue of the egg laying phenotype (Table I). In contrast, PS1 ⌬I83/⌬M84 showed significantly less rescuing activity (Table I). Therefore, similar to all other PS1-associated FAD mutations investigated so far (25,32), the PS1 ⌬I83/⌬M84 deletion also lost activity in Notch signaling in the in vivo rescuing assay. DISCUSSION The PS1 ⌬I83/⌬M84 mutation is to our knowledge the first FAD-associated deletion. Almost all FAD mutations described within PS1 or PS2 are point mutations exchanging single highly conserved amino acids (48,49). Although the recently identified intron 4 mutation can result in C-terminal truncated PS1 derivatives, these are pathologically inactive and do not cause increased A␤42 generation (29). This is consistent with the finding that C-terminal deleted artificial presenilins containing a FAD mutation are unable to induce A␤42 production as well. In that regard, mutant PS fragments have been overexpressed, which correspond to the proteolytically generated N-terminal fragment. Such PS fragments consistently failed to induce A␤42 generation (36,50,51), and even very minor deletions at the C terminus of PS1/PS2 inactivate PS function (52). The lack of pathological activity of truncated presenilins appears to be due to the failure of such PS derivatives to incorporate into the functionally required PS complex (36,45,50,51,(53)(54)(55)(56). Based on these findings, it would be unexpected that large deletions in the PS sequence would indeed be associated with early onset AD. However, one of the previously identified PS1 mutations results in the deletion of the complete exon 9-encoded domain due to a splicing error (28). Since exon 9 encodes the cleavage site of PS1 (43,44,57), this deletion mutation is associated with a lack of endoproteolysis and consequently the accumulation of the PS1 ⌬exon9 holoprotein (44). This very drastic phenotype was believed to be the cause for the pathological activity of PS1 ⌬exon9 in A␤42 generation. However, we have shown recently that the pathological activity of PS1 ⌬exon9 splicing mutation in regard of A␤42 generation is solely due to a single amino acid exchange of the conserved Ser 290 to cysteine (32). Therefore, this demonstrates that deletions of the PS amino acid sequence have so far not been associated with a pathological function in A␤ generation. Consistent with these findings, De Jonghe et al. (29) reported that the pathological activity of the intron 4 mutation of PS1 was surprisingly associated with an amino acid insertion generated FIG. 4. Endoproteolytic cleavage, replacement of endogenous PS, and elevated A␤42 production by stable expression of PS1 ⌬I83/⌬M84 in K293 cells. A, upper panel, cell lysates from K293 cells expressing endogenous presenilins or overexpressing the indicated PS variants were immunoprecipitated with antibodies specific to the large loop of PS1 (3027). PS1 holoproteins and PS1 CTFs were detected by immunoblotting using the monoclonal antibody BI.3D7 (to PS1). Endoproteolytic cleavage occurs in cell lines stably expressing PS1 ⌬I83/ ⌬M84 or wt PS1. As observed before, cell lines stably expressing PS1 ⌬exon9 replace endogenous PS1 and accumulate as uncleaved fulllength PS proteins (32,44). Lower panel, cell lysates from the cell lines in (A) were immunoprecipitated with antibodies specific to the large loop of PS2 (3711), and PS2 CTFs were detected by immunoblotting using the monoclonal antibody BI.HF5c (to PS2). Overexpression of all indicated PS1 variants results in efficient replacement of endogenous PS2. B, quantitation of the A␤42 and A␤40 concentrations in conditioned media of K293 cells expressing the indicated presenilin variants using a previously described highly specific enzyme-linked immunosorbent assay (36). Expression of PS1 ⌬I83/⌬M84 results in a 1.5-1.8 fold increase of A␤42 production. Consistent with previous results, the PS1 ⌬exon9 mutation causes the production of higher A␤42 levels than the majority of other FAD mutations (63). However, the PS1 ⌬exon9 also causes the generation of cotton wool plaques (30).

TABLE I
Reduced rescuing activity of the sel-12 egg laying defect by PS1 ⌬I83/⌬M84 For 50 transgenic animals each, the numbers of progeny were counted and were grouped in the following categories: ϩϩϩ, over 30 progeny laid by individual animal; ϩϩ, 15-30 progeny laid; ϩ, 5-15 progeny laid; 0 -5 progeny laid. The failure to lay eggs in the transgenic animals is the consequence of mosaic expression and/or absence of the PS1 gene, whereas egg laying in a sel-12(ar171) background can only be achieved by a functional presenilin rescuing the mutant defect. by the utilization of a cryptic splice site but not with the deleted PS1 derivatives. Therefore, the PS1 ⌬I83/⌬M84 mutation is the first pathogenic deletion of presenilins, which indeed is directly associated with a malfunction in A␤42 production. Although the novel PS1 ⌬I83/⌬M84 mutation is the first pathogenic PS1 deletion, it is associated with a pathological phenotype similar to that first described in association with the PS1 ⌬exon9 Finn mutation. The cotton wool plaques observed lack congophilia including an amyloid core and associated abnormal neurites (30). Non-neuritic parenchymal deposits in association with extensive neurofibrillary degeneration are, however, not a unique feature of variant AD as lesions resembling cotton wool plaques, but composed of different amyloidogenic peptides also occur in the BRI gene related diseases, in familial British dementia (64 -66) and familial Danish dementia (67). It is of interest that the clinical phenotype of familial British dementia and familial Danish dementia also resembles that seen in variant AD with cotton wool plaques and spasticity (38). Since A␤42 has a great propensity to accumulate in classical senile plaques (6, 7), one would have expected that PS1 mutations associated with noncongophilic cotton wool plaques might not affect A␤42 generation. However, our data demonstrate that A␤42 generation is significantly induced in cultured cells by the PS1 ⌬I83/⌬M84 mutation very similar to all other FAD associated PS mutations investigated so far. In addition, we showed immunohistochemically that the cotton wool plaques associated with the PS1 ⌬I83/⌬M84 mutation are predominantly composed of A␤42, which is similar to that seen in classical plaques in sporadic AD and AD caused by either APP or other PS1 mutations (6,58,59). PS1 ⌬I83/⌬M84 not only behaves in terms of A␤42 generation like a typical PS-associated FAD mutation but also exhibits a similar loss of function in Notch signaling in C. elegans. Therefore, the question arises whether amyloid plaques composed of A␤42 are the primary cause of AD and whether such amyloid plaques initiate neurodegeneration. Apparently, the lack of classical dense core congophilic plaques did not prevent the cotton wool plaque cases from developing neurological symptoms including dementia, suggesting that the potential pathological activity of A␤42 may be acting upstream of amyloid deposition. Although this could indicate that increased A␤42 production is an epiphenomenon of FAD-associated mutations, we think it is much more likely that the previously characterized protofibrils of A␤ (4, 5) may be the primary cause for the observed neurological deficits. Since A␤42 can be observed intracellularly (33,60,61), primary pathological consequences may be induced long before A␤ finally precipitates into amyloid plaques.
Our findings may also indicate that therapeutic strategies exclusively based on the reduction of the amyloid plaque burden (62) may not always be sufficient to prevent AD symptoms.