Mutant Presenilin 2 Transgenic Mice A LARGE INCREASE IN THE LEVELS OF A b 42 IS PRESUMABLY ASSOCIATED WITH THE LOW DENSITY MEMBRANE DOMAIN THAT CONTAINS DECREASED LEVELS OF GLYCEROPHOSPHOLIPIDS AND SPHINGOMYELIN *

From the ‡Department of Neuropathology, Faculty of Medicine, University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033, Japan, the §Department of Membrane Biochemistry, Tokyo Metropolitan Institute of Gerontology, 35-2 Sakaecho, Itabashi-ku, Tokyo 173-0015, Japan, the ¶Central Institute for Experimental Animals, 1430 Nogawa, Miyamae-ku, Kawasaki 216-0001, Japan, the iDivision of Neuropathology, Department of Pathology, and Center for Neurologic Diseases, Brigham and Women’s Hospital, Harvard Medical School, Boston, Massachusetts 02115, the **Genetics and Aging Research Unit, Massachusetts General Hospital, Harvard Medical School, Charlestown, Massachusetts 02129, the ‡‡Department of Demyelinating Disease and Aging, National Institute of Neuroscience (National Center of Neurology and Psychiatry), 4-1-1 Ogawahigashi, Kodaira, Tokyo 187-8502, Japan, and the §§Core Research for Evolutional Science and Technology (CREST), Japan Science and Technology Corporation, Kawaguchi 332-0012, Japan

A subset of early onset familial Alzheimer's disease is inherited as an autosomal dominant trait, and the genes for presenilin 1 and 2 (PS1 and PS2), 1 in addition to the gene for ␤-amyloid precursor protein (APP), were identified as the causative genes.PS1 is mapped to chromosome 14 (1,2), whereas PS2 is mapped to chromosome 1 (3)(4)(5).These PS1 and PS2 genes encode multispanned transmembrane proteins showing high degrees of homology (4,5).Both proteins are located predominantly in the endoplasmic reticulum and partly in the Golgi apparatus and other compartments (6 -8), but their physiological functions remain unclear.More than 50 different mutations have thus far been identified in PS1 (9), whereas only two mutations have been found in PS2 (10).The residue at position 141 (Asn) in PS2, which is conserved in human and mouse PS1 and PS2, is substituted by Ile (N141I) in the Volga German kindred.Another missense mutation (M239V) in PS2 has been found in Italian familial Alzheimer's disease families (5).
Although the pathogenetic mechanism of how Alzheimer's disease is developed by PS mutations remains unknown, mutations of PS1 and PS2 are known to have similar effects on the production of amyloid ␤ protein (A␤) 42, the initially deposited A␤ species in senile plaques (11)(12)(13)(14).Although A␤42 is normally secreted in much lower quantities than A␤40, these aforementioned mutations induce elevation of the A␤42 levels in cultured cells and transgenic mice (15)(16)(17)(18)(19)(20).It has also been reported that in primary neuronal cultures derived from PS1 knockout mouse embryos, A␤ secretion was remarkably decreased, concomitant with the accumulation of the C-terminal fragment of APP (21).The mutation in the two particular Asp residues in the PS1-transmembrane domains induced a profound decrease in the A␤ production and an increase in the levels of the C-terminal fragment of APP (22).These observations indicate that PS1 may have a direct or indirect role in the ␥-cleavage of APP.
We previously reported that the mutant PS2 transgenic mice showed increases in the A␤42 levels in the Tris-saline (TS)soluble fractions in an age-dependent manner during the period of 2-8 months of age (20).On the other hand, a series of A␤ quantitation studies of autopsied human tissues has clearly shown that A␤42 already accumulates to a significant extent in the TS-insoluble fraction of normal aged brains even before senile plaques become immunocytochemically detectable (23)(24)(25).The presence of A␤ in the TS-insoluble fraction has recently been demonstrated in cultured human neuroblastoma cells (26).Thus, these findings raise the question as to which species of A␤, soluble A␤ or insoluble A␤, is involved in the initial stage of A␤ deposition.A likely possibility is that the earliest abnormality in ␤-amyloidogenesis is the altered intracellular trafficking of the specific membrane domains, because i) a substantial fraction of A␤ is present in the detergentinsoluble membrane compartment, which is rich in cholesterol and sphingolipids, in cultured cells and rat brains (26,27); ii) A␤, generated from apically missorted APP in a cholesterol-dependent manner in Madin-Darby canine kidney cells, serves as a seed for A␤ fibrillogenesis (28); iii) GM1 ganglioside-bound A␤ is detected in Down's syndrome brains exhibiting exclusively diffuse plaques, the earliest stage of senile plaques (29); and iv) the addition of GM1-containing vesicles to A␤ solution dramatically accelerates A␤ fibril formation in vitro (30).
In the present study, we sought to obtain further insight into the effects of mutant PS2 on the A␤ levels in the TS-insoluble, guanidine hydrochloride-solubilized fraction of the mouse brain and to characterize the intracellular compartmentalization of insoluble A␤.

EXPERIMENTAL PROCEDURES
Transgenic Mice-The heterozygous PS2 transgenic mice used in this study were from the previously established lines W2 (wild-type PS2 transgenic mice) and M1 and M2 (N141I mutant PS2 transgenic mice) (20).Each line of transgenic mice was backcrossed to the C57BL/6J strain, and those mice carrying the PS2 transgene were selected using a transgene-specific polymerase chain reaction assay (20).Littermates without PS2 transgenes were used as the nontransgenic controls.The levels of PS2 mRNA and PS2 fragments and the levels of APP and its C-terminal fragments in the mouse brains were determined as described previously (20).
Another set of PS2 transgenic mice, including the N, K, and P lines was generated as follows (62).To construct the transgene vector, the cytomegalovirus promoter was excised from the expression vector pCI (Promega) and replaced with an oligonucleotide polylinker into which the platelet-derived growth factor ␤-chain promoter fragment was inserted (31).The wild-type and N141I mutant PS2 cDNAs (32) in the pcDNA3.1-Zeo(ϩ)backbone were subcloned into the above construct.DNA fragments containing PS2 transgenes were microinjected into pronuclei of fertilized zygotes from FVB/N mice.Mice expressing either wild-type (N line) or N141I mutant form (K and P lines) of PS2 were generated and genotyped using tail snip DNA.Transgene expression was confirmed by reverse transcription-polymerase chain reaction using a human PS2-specific primer set.Transgenes were maintained in the FVB/N background.
ELISA-ELISA for A␤ was carried out as described previously (33,35).Authentic rodent A␤ was purchased from AnaSpec (San Jose, CA).A microtiter plate was precoated with BNT77 as the capture antibody.The captured A␤ was detected with horseradish peroxidase-labeled BA27 or BC05 for the quantitation of A␤x-40 and A␤x-42, respectively.
The brain tissue (about 120 mg) from each mouse was homogenized in 4 volumes of Tris-saline (50 mM Tris-HCl, pH 7.4, 150 mM NaCl) by a motor-driven Teflon homogenizer, as described previously (20).After centrifugation, the pellet was resuspended by brief sonication in the same volume of 6 M guanidine hydrochloride in 50 mM Tris-HCl, pH 7.6.
The suspension was centrifuged at 265,000 ϫ g for 20 min in a TLX ultracentrifuge (Beckman, Palo Alto, CA).The supernatant was diluted 1:12 to reduce the concentration of guanidine hydrochloride to 0.5 M (36) and subjected to ELISA.
Preparation of the Detergent-insoluble, Low Density Membrane (LDM) Fractions-Detergent-insoluble membrane fractions were obtained according to the established protocol with minor modifications (26,37).The brain tissue (120 mg) from each mouse was homogenized in MES-buffered saline (MBS) (25 mM MES, pH 6.5, 0.15 M NaCl) containing 1% Triton X-100, 1 mM phenylmethylsulfonyl fluoride, 10 g/ml leupeptin, 1 g/ml pepstatin, and 10 g/ml aprotinin, followed by brief sonication on ice.The sucrose concentration of the extract was adjusted to 40% by the addition of 80% sucrose in MBS; the extract was then placed at the bottom of an ultracentrifuge tube and overlaid with a 5%/35% discontinuous sucrose gradient in MBS without Triton X-100 (4 ml of 5% sucrose/4 ml of 35% sucrose).The gradients were centrifuged at 39,000 rpm for 20 h in a SW41 rotor (Beckman, Palo Alto, CA).A light-scattering interface at 5%/35% sucrose was carefully aspirated.These low density and Triton-solubilized high density fractions (bottom 4 ml) were separately pooled, diluted 3-fold with MBS, and centrifuged.The resultant pellets and the pellet obtained by centrifugation of the original sucrose gradient were extracted with guanidine hydrochloride and subjected to ELISA, as described above.
To visualize A␤42 in the LDM fraction by Western blotting, each LDM pellet was first extracted with chloroform/methanol (see below), and then the residue with formic acid.After centrifugation, aliquots of the supernatant were dried using a Speed Vac (Savant Instruments, Farmingdale, NY), solubilized with the Laemmli sample buffer containing 4 M urea, and subjected to Western blotting.
Western Blotting and Quantitation-Except in the case of A␤ and the C-terminal fragments of APP, Western blotting was performed on a polyvinylidene difluoride membrane (Immobilon, Nihon Millipore Ltd., Yonezawa, Japan).Bound antibodies were detected by the enhanced chemiluminescence (ECL) detection system (Amersham Pharmacia Biotech).For the detection of A␤ and the C-terminal fragments of APP, the proteins separated on a 16.5% Tris-Tricine gel were transferred onto a nitrocellulose membrane (Schleicher & Schuell).The membrane was immersed in boiling phosphate-buffered saline to enhance the detection sensitivity (38).The bound antibodies were detected by either an ECL or ECL plus detection system (Amersham Pharmacia Biotech).
Quantitation of the ECL bands of interest was performed using a densitometer (Model GS-700 imaging densitometer, Bio-Rad) as described previously (20).
Lipid Analysis-Total lipids were extracted from the pellet of each LDM fraction with chloroform/methanol (2: 1) and then with chloroform/methanol/water (1: 2: 0.8).The extracts were applied to DEAE-Toyopearl columns to separate the neutral lipids and acidic lipids (39).Aliquots of acidic lipids were subjected to acidic phospholipid analysis.To determine gangliosides, the rest of the acidic fractions were subjected to mild alkaline treatment and then applied to Toyopearl HW-40 columns for desalting (40).
To determine the cholesterol content, neutral lipid fractions were analyzed by an enzymatic method using an assay kit, Determiner FC-555 (Kyowa Medex, Tokyo, Japan) or F-Chol E (Wako, Osaka, Japan).Total lipid phosphorus was measured by a modified Bartlett method (41).Phospholipid composition was determined as follows.Each phospholipid was separated by TLC on high performance TLC plates (Merck, Darmstadt, Germany) with chloroform/methanol/acetone/acetic acid/water (32:25:14:4:2) as the solvent system (42).Each band was scraped off the TLC plates after visualizing with iodine vapor and then subjected to phosphorus assay.Total ganglioside contents were determined by the method of Miettinen and Takki-Lukkainen (43).
Histological and Immunocytochemical Analyses-Mouse brains were cut into halves and each half was fixed in 10% formalin and paraffinembedded for histological analysis.The tissue sections were stained with hematoxylin-eosin and by a modified Bielshowsky's method.They were also immunostained with various A␤ antibodies after formic acid treatment, as described elsewhere (13,14).
Other Methods-The obtained data were statistically analyzed by the Mann-Whitney's U test using the Stat View computer software (Version 5.0, Abacus Concepts Inc., Berkeley, CA).p values Ͻ 0.05 were considered to be significant.Protein concentrations were determined by the bicinchoninic acid protein assay (Pierce) in the presence of 1% SDS.

RESULTS
The A␤40 and A␤42 Levels in the TS-insoluble Fraction of PS2 Transgenic Mouse Brains-To precisely define the effect of N141I mutation of PS2 on the A␤ levels, we compared two transgenic mouse lines (W2 for wild-type, M1 for mutant) expressing comparable mRNA and protein levels of human wildtype and mutant PS2 (20).Between the two lines, there was no difference in the levels of endogenous mouse APP and PS2 that could potentially affect the A␤ levels (20).
We modified the extraction procedure for quantitating A␤ in the TS-insoluble fraction; guanidine hydrochloride was used instead of formic acid.This is because extraction with concentrated formic acid, which is widely used, requires large volumes of NaOH and Trizma (Tris base) base for neutralization, resulting in a marked dilution of the samples.This substantially reduces the ELISA sensitivity for quantitation of the A␤ levels in the brain (24,44).Guanidine hydrochloride was found to be even more effective for the extraction of low levels of A␤ than formic acid and provided reproducible results (26,36).Using this new extraction protocol, we were able to detect significant amounts of A␤40 and A␤42 in the TS-insoluble fraction of the nontransgenic and transgenic mouse brains (Table I and Fig. 1).The A␤40 and A␤42 levels in the TS-insoluble fraction were consistently higher than those in the TS-soluble fraction reported previously (20).The differences in the A␤ levels between the TS-insoluble and TS-soluble fractions may be even larger, because the authentic rodent A␤ used in the previous study was found to contain some impurities and might have led to an overestimation of the A␤ levels, in particular A␤42 level.In fact, severalfold more A␤ than TS-soluble A␤ was detectable in the TS-insoluble fraction when quantitated using an identical authentic A␤ as the standard.
The A␤40 levels in the TS-insoluble fractions from 5-8month-old W2 or M1 mouse brains were lower than those from nontransgenic mice of corresponding ages (p Ͻ 0.05, Mann-Whitney's U test), a finding similar to the previous one with TS-soluble fractions (20).There was an age-dependent increase in the TS-insoluble A␤40 levels in W2 mice and a similar but smaller age-dependent increase in the case of M1 mice; between the ages of 8 and 12 months, the levels of A␤40 in the TSinsoluble fraction from W2 mice increased (p Ͻ 0.02) and reached the levels seen in the case of nontransgenic mice.The levels of A␤40 slightly but significantly decreased between the ages of 12  and 15 months in nontransgenic mice (p Ͻ 0.01).The A␤40 levels in W2 mice were similar to those in M1 mice up to the age of 8 months, whereas its levels in W2 mice were significantly higher than those in M1 mice at the ages of 12 and 15 months (p Ͻ 0.01).
In contrast to A␤40, the A␤42 levels in the TS-insoluble fractions of W1 mouse brains were quite similar to those of nontransgenic mice, although its levels were much lower (by ϳ5-fold) than the A␤40 levels.It is noteworthy that the levels of A␤42 in the TS-insoluble fractions of M1 mouse brains were significantly higher (about 2-3-fold) at each age than those in the case of other two lines (p Ͻ 0.02).The A␤42 levels in M1 mice exhibited no significant changes between the ages of 5 and 12 months and declined slightly at the age of 15 months (p Ͻ 0.02).In W2 and nontransgenic mice, a very small increase in the A␤42 levels occurred between the ages of 5 and 8 months (p Ͻ 0.02).
The ratio of A␤42 to A␤40 in M1 mice was much larger, in all the age groups examined, than those in the case of nontransgenic or W2 mice (p Ͻ 0.02; see Table I).The ratio remained the same in M1 mice between the ages of 5 and 12 months and declined between the ages of 12 and 15 months (p Ͻ 0.02).In W2 and nontransgenic mice, the ratio increased between the ages of 5 and 8 months (p Ͻ 0.02) and then declined.
A remarkable increase in the A␤42 levels (p Ͻ 0.05), but not A␤40 levels, was similarly observed in another independent mutant PS2 transgenic mouse line, M2, at the age of 2 months (Table I).
Although A␤42 accumulated in the TS-insoluble fraction in M1 mice 2-3-fold more than in W2 or nontransgenic mice, A␤ deposition was immunocytochemically undetectable in the brain even at the age of 24 months (data not shown).
A Large Increase in the A␤42 Levels Is Associated with the LDM Fraction in Mutant PS2 Transgenic Mice-One of the principal membrane compartments in which A␤ resides is the detergent-insoluble, glycolipid-enriched membrane domain (DIG) (45,46).DIGs are rich in sphingolipids and cholesterol and serve as membranous rafts for recruiting proteins and lipids that collaborate in signaling.Because of their insolubility in Triton X-100, DIGs float readily into the low density fraction upon density gradient centrifugation.To learn whether the A␤ in the TS-insoluble fraction quantitated above is associated with this distinct membrane domain, we prepared LDM fractions from nontransgenic and PS2 transgenic mouse brains.
As shown in Fig. 3B, flotillin, a DIG-associated membrane protein abundantly expressed in neurons (47) was fractionated exclusively into the LDM fractions in both transgenic and nontransgenic mice.In all three lines, the A␤40 in the LDM fractions constituted half of the total amount of Triton X-100insoluble A␤40 (Fig. 2A).The A␤40 levels in the LDM fractions seemed to increase gradually with age, but this increase was not statistically significant.The A␤40 levels in each fraction were comparable among W2, M1, and nontransgenic mice at the ages of 2 and 8 months.The A␤40 levels in the LDM fractions in M1 mice at the age of 15 months, however, were significantly lower than those in W2 or nontransgenic mice of the same age (p Ͻ 0.05).
The A␤42 levels in the LDM fractions of the brain did not differ significantly between W2 and nontransgenic mice at all FIG. 2. Sucrose density gradient fractionation of the Triton-insoluble A␤ in the brains from nontransgenic (non-Tg) and transgenic mice (W2 and M1).Each brain from nontransgenic or transgenic mice was homogenized in the presence of 1% Triton X-100 and fractionated by sucrose density gradient centrifugation as described under "Experimental Procedures."The resultant fractions were collected from the top.After centrifugation of the pooled fractions, the pellet was extracted with 6 M guanidine hydrochloride, and the extract was subjected to BNT77-based ELISA for the determination of A␤40 (A) and A␤42 levels (B).L, H, and P represent the LDM fractions, high density fractions, and pellets, respectively.Data are expressed as mean Ϯ S.E.(three independent experiments).
the ages examined (Fig. 2B).Most importantly, the A␤42 levels in the LDM fractions of M1 mouse brain were remarkably higher at all ages examined than those in the case of W2 and nontransgenic mice (p Ͻ 0.05).The A␤42 levels in the high density fractions and pellets were also significantly increased in M1 mice compared with nontransgenic or W2 mice at the age of 8 months (p Ͻ 0.05) but were significantly higher compared with only nontransgenic mice at the age of 15 months (p Ͻ 0.05).The A␤42 levels in each of the fractions showed no significant difference between the ages of 2 and 8 months in W2, M1, or nontransgenic lines.The A␤42 levels in the LDM fraction at the age of 15 months were significantly higher than those at the age of 8 months in M1 mice (p Ͻ 0.05).Similar increases in the A␤42 levels during this period were also observed in the high density fractions and pellets in M1 mice (p Ͻ 0.05).A similar age-dependent increase during the same period was also found in the LDM fractions and pellets obtained from W2 or nontransgenic mice (p Ͻ 0.05).
To further confirm the increase in A␤42 levels in the LDM fractions in M1 mice, Western blotting was performed using BC05, an A␤42-specific monoclonal antibody.As expected, a BC05-immunoreactive band migrating to 4 kDa was detected in the LDM fraction from M1 mice but not in that from W2 or nontransgenic mice (Fig. 3A).This indicates that the transgenic mice harboring mutant PS2 have markedly increased levels of A␤42 in the LDM fraction.
The N-and C-terminal Fragments of PS2, and APP and Its C-terminal Fragments in the LDM Fraction-As previously reported (20), the levels of the N-terminal fragment (NTF) at 34 kDa and C-terminal fragment (CTF) at 22 kDa of PS2 did not differ significantly between W2 and M1 mice at each age and above the age of 15 months (data not shown).These results indicate that the increased levels of A␤42 in the brains of mutant PS2 transgenic mice are not brought about by altered metabolism or stability of the PS2 protein in vivo.
We next examined whether the PS2 fragments and APP and its fragments were present in the LDM fraction and, if these were present, whether these protein levels differed among W2, M1, and nontransgenic mice.The NTF was mainly present in the high density fraction in W2 and M1 mice, but a fraction of the NTF was associated with the LDM fraction (Fig. 3B).Similarly, a small proportion of the CTF was fractionated into the LDM fraction, although this fragment was also prominent in the high density fraction (Fig. 3B).CTFs derived from endogenous mouse PS2 were only faintly labeled in nontransgenic mice, and their distribution was similar to that of human PS2.The profiles of immunolabeled bands of both the NTF and CTF were nearly identical between W2 and M1 mice in each fraction.This indicates that the mutation did not alter the subcellular localization of the PS2 fragments, especially their partition into the LDM fraction.
Full-length mouse APP at 120 kDa was most predominantly located in the high density fraction, but to a smaller extent in the other fractions, in particular the LDM fraction (Fig. 3B).There was again no significant difference in the levels of fulllength APP in the LDM fraction between W2 and M1 mice.The cleavage of APP by a putative ␣or ␤-secretase generates two kinds of C-terminal fragments, C83 and C99 (9).C83, with an apparent molecular mass of ϳ9 kDa, was located mainly in the high density fraction and barely in the LDM fraction (Fig. 3B).C99, at ϳ12 kDa, an amyloidogenic fragment, was predominantly present in the LDM fraction (Fig. 3B).The levels of these fragments in each fraction were comparable between W2 and M1 mice (Fig. 3B).
We sought to find a particular protein(s) the level of which is significantly elevated in the LDM fraction of M1 mouse brains.Such protein may interact specifically with A␤42, and anchor A␤42 into the LDM domain, thereby increasing the levels of A␤42 within the distinct membrane domain.The proteins in the LDM fraction from each line were separated by SDS-PAGE, followed by Coomassie Brilliant Blue or silver staining.We were, however, unable to find such a LDM resident protein in M1 mouse brain (data not shown).
Lipid Composition of the LDM Fraction-Finally, we investigated the lipid composition of the LDM fraction from nontransgenic, W2, M1, and M2 mice.Table II shows the contents of major lipid species in the LDM fractions from the four lines of mice at the age of 2 months.The molar ratio of cholesterol to total phospholipids was found to be 0.48 -0.63.The total cholesterol levels in W2 mice were higher than those in nontransgenic mice and M1 mice (Mann-Whitney's U test, p Ͻ 0.05) but did not differ from those in M2 mice.The total phospholipid levels in M2 mice were significantly lower than those in W2 mice (p Ͻ 0.05).Those in M1 mice were also lower than those in W2 mice, but the difference was not statistically significant.The levels of each glycerophospholipid species in M1 and M2 mice, such as phosphatidylcholine, phosphatidylethanolamine, phosphatidylserine, and phosphatidylinositol, were significantly lower than those in W2 mice (p Ͻ 0.05), except for the level of phosphatidylcholine and phosphatidylinositol in M1 mice.The levels of sphingomyelin in the LDM fraction from M1

FIG. 3. Confirmation of A␤42 in LDM fraction from M1 mice (A) and the presence of flotillin, PS2, and APP in LDM fractions from nontransgenic and transgenic mice (B).
A, the lipids and proteins in the LDM fraction from each mouse line were successively extracted with chloroform/methanol and formic acid, as described under "Experimental Procedures."The proteins were separated on a Tris-Tricine gel (16.5% T, 3% C) and subjected to sensitive Western blotting using BC05.B, after sucrose density gradient centrifugation, the pellet was resuspended in 1 ml of MBS by homogenization and solubilized with the Laemmli sample buffer.Four g of proteins from the LDM fractions (L), high density fractions (H), and pellets (P) were loaded onto a 15 or 7.5% polyacrylamide gel and subjected to Western blotting using anti-flotillin; anti-PS2 (2972N and PS2L1); 6E10 for C99; and C4 for APP, C99, and C83.C99 was recovered in the LDM fraction, whereas C83 was found exclusively in the high density fraction.and M2 mice were significantly lower than those in nontransgenic mice (p Ͻ 0.05) and tended to be lower than those from W2 mice.As a result, the ratios of cholesterol to sphingomyelin appeared to be increased in M1 and M2 mice.The total ganglioside levels did not differ among nontransgenic, W2, M1, and M2 mice.Thus, it is likely that mutant PS2 leads to significant decreases in the contents of glycerophospholipids and sphingomyelin in the LDM fraction, as compared with wild-type PS2.
To confirm our observations regarding the effects of mutant PS2, we analyzed the lipid composition of the LDM fraction from another set of wild-type (N line) and N141I mutant (K and P lines) PS2 transgenic mouse lines at the age of 12.5 months.The transgene-derived PS2 levels in N, K, and P lines were 4.7-, 5.8-, and 2.4-fold, respectively, those of the endogenous PS2, as quantitated by the CTF levels (data not shown).However, no decrease in the levels of glycerophospholipids and sphingomyelin in the LDM fraction from K and P lines was observed, as compared with N line (see Table III).The A␤42 levels in the LDM fraction from K line tended to be increased, and those from P line were not altered, but in either case, the difference was not statistically significant, as compared with N line (see Table III).Thus, the effects of PS2 mutation on the lipid composition of the LDM fraction appear to parallel its effects on the A␤42 levels.

DISCUSSION
One of the important findings in this study is that mutant PS2 linked to the Volga German families, N141I, had significant effects on the levels of A␤42 in the TS-insoluble fraction of mouse brain.The effects of the PS2 mutation on the A␤42 levels in the brain were more evident in the TS-insoluble fraction than in the TS-soluble fraction.It is quite possible that the TS-soluble A␤ in the previous study (20) was largely derived from TS-insoluble A␤ through disruption by homogenization: TS-soluble A␤ levels in the brain are severalfold lower than those of TS-insoluble A␤, and the former faithfully reflects the latter.The presence of TS-insoluble A␤ strongly suggests that such A␤ species represents a normal catabolic intermediate in the brain.Our investigation on human autopsy specimens using the new extraction protocol supports this view and further suggests that it is the level of this TS-insoluble A␤42 that becomes steeply elevated in the initial phase of A␤ deposition. 2hus, increased levels of TS-insoluble A␤42 in mutant PS2transgenic mice likely mimic this initial stage of ␤-amyloidogenesis in the human brain.
The data on the levels of TS-insoluble A␤ in the brain and the data on the levels in each fraction following sucrose density centrifugation appear to be contradictory.For example, the former showed that the levels of TS-insoluble A␤42 remained at a plateau or even declined slightly after the age of 8 months (Fig. 1); in contrast, the levels of A␤42 in each fraction from each mouse line appeared to increase from the age of 2 to 15 months (Fig. 2).It should be noted that the A␤ in Fig. 2 represents the amounts of the insoluble A␤ in each fraction that was recovered in the pellet after fractionation; following density gradient centrifugation of Triton X-100 homogenates, each fraction was pelleted for extraction.Thus, the TS-insoluble, Triton X-100-soluble A␤ that was fractionated into the high density fractions was probably lost.This may explain the above discrepancy.
The most striking feature of mutant PS2 transgenic mice is the large increase in the level of the A␤42 in the LDM fraction.This increase was obvious at the ages of 2-15 months in M1 mice (Fig. 2).It was even more remarkable at the age of 1 month, when the A␤42 levels in the other fractions were suppressed (data not shown).This indicates that it is a longstanding abnormality in M1 mice (and presumably also in M2 mice) and suggests that the initial abnormality in M1 and M2 mice is the increased levels of A␤42 in the LDM fraction.The origin of A␤40 and A␤42 in the high density fractions and the pellet and their interrelationships and relationships with the A␤ in the LDM fraction are entirely unknown.
At least a part of the intracellular A␤40 and A␤42 is generated in the trans-Golgi network (48,49) and in the endoplasmic reticulum/intermediate compartment (49 -51), respectively.On the other hand, LDM domains (DIGs) are considered to occur in the Golgi region, and some associated proteins, in particular glycosylphosphatidylinositol-anchored proteins, such as alkaline phosphatase, become incorporated into the LDM domains after being transported from the endoplasmic reticulum to the Golgi complex (52).Thus, it would be reasonable to speculate that substantial amounts of generated A␤42 are incorporated into the LDM domains occurring in the Golgi complex and delivered mostly to the plasma membrane.One of the plausible explanations is that aberrant partitioning of the generated A␤42 into the LDM domain may occur in the Golgi region in mutant PS2 transgenic mice for unknown reasons (see below).Another possibility would be that the LDM domain is one of the sites for A␤ generation, as previously suggested (27).Although each constitutes only a small fraction, the components that participate in or modulate the A␤ generation, namely APP, C99, and NTF and CTF of PS2, are localized in this rather limited membrane domain (Fig. 3).Thus, coexistence of these proteins in the LDM domain may facilitate their functional interaction and promote A␤ generation.It is also possible that secreted A␤42 is internalized through its specific interaction with a particular protein in the LDM domain.
The most unexpected finding of the present study is alterations of the lipid composition in the LDM fraction, likely induced by the overexpression of mutant PS2.The expression of mutant PS2 caused significant decreases in the contents of glycerophospholipids and sphingomyelin in the LDM fraction (Table II).It is unlikely that these effects originate from the small difference in the expression level rather than the mutation itself: The levels of wild-type (W2) and mutant (M1) PS2 mRNA are estimated to be ϳ8and ϳ9-fold that of the endogenous PS2, respectively (20).More importantly, similar alterations of the lipid composition were also observed in M2 mice, in which the levels of human mutant PS2 mRNA were 5-fold that of endogenous PS2 mRNA (20).Thus, the alterations of the lipid composition (decreased levels of glycerophospholipids and sphingomyelin) in the LDM fraction should be largely ascribed to the mutation itself.
It should be noted that we were unable to remove myelin contaminating the LDM fraction, as detected by an antibody to myelin basic protein, and we should be cautious to interpret the lipid composition presented here (see Ref. 53).However, the molar proportion of cholesterol and total phospholipid is usually approximated in myelin (54), but it is actually 1:2 in the present LDM fraction (Table II).Accordingly, a contribution of contaminated myelin to the lipid composition may not be so large.Alternatively, substantial proportions of LDM fraction could come from DIGs of myelin sheath, and multilamellar structures observed in the LDM fraction (data not shown) may represent DIGs instead of myelin sheath (55).In either case, we consider that the lipid composition presented here reflect largely that of the LDM domain.
Currently, we do not know why overexpression of mutant PS2 in an additional two lines (K and P lines) of mice fails to cause similar alterations in the lipid composition (see Table III).One possibility is that the lipid composition is greatly influenced by the genetic background of mouse lines (C57BL/6J versus FVB/N).In fact, the levels of phospholipids in the LDM fraction from these FVB/N mouse lines were much lower (ϳ1700 nmol/mg of protein), which resulted in a greater increase in the ratio of cholesterol to total phospholipid (0.69 -0.72).Another possibility is that the extent of myelin contamination would be much greater in these mice because of their older ages, as reflected by greater ratio of sphingomyelin to total phospholipid (0.053-0.055), thereby masking the alterations of the lipid composition in the LDM domain.It is also possible that different promoters driving the transgenes (cytomegalovirus/␤-actin versus platelet-derived growth factor ␤) result in distinct cell type-specific expression of mutant PS2.Thus, further investigations are required to generalize the observed effects of mutant PS2 to other mouse lines with different genetic backgrounds.
In this context, it is extremely intriguing to note that similar alterations in phospholipids were repeatedly reported in Alzheimer's disease brains (56,57).Furthermore, a series of studies (58,59) showed increased concentrations of glycerophosphocholine and glycerophosphoethanolamine, the metabolites of phosphatidylcholine and phosphatidylethanolamine, respectively.This suggests that membrane phospholipid catabolism is increased in Alzheimer's disease brain.
One may speculate that decreased levels of sphingomyelin rather than of glycerophospholipids are critical for the large increase of A␤42 in the LDM fraction.This is because i) sphingomyelin is specifically located in the outer leaflet of the lipid bilayer, where A␤ is supposed to be located (60), and ii) most of the sphingomyelin in the plasma membrane is localized to the LDM domain and thus it is the major component of the domain.This potential decrease in the levels of sphingomyelin is unlikely to be due to decreased levels of ceramide in the brain cells, because the levels of gangliosides appear not to be altered among nontransgenic and transgenic mice (Table II).It is of note that the sphingomyelin depletion caused by sphingomyelinase blocks site 1 proteolysis of sterol regulatory elementbinding protein-2 (61).By analogy, a substantial decrease in sphingomyelin levels may affect the activity of ␥-secretase.This may be induced by formation of thinner LDM domains, which enhances the production of A␤42 and/or leads to aberrant sorting of A␤42 that may depend on the thickness of the membrane.
Altogether, the present observations point to a previously unrecognized link between the mutant PS2 and the metabolism of glycerophospholipids and sphingomyelin.

FIG. 1 .
FIG. 1. A␤40 and A␤42 levels in the TS-insoluble, guanidine hydrochloride-solubilized fraction of the brains from nontransgenic (non-Tg) and transgenic mice (W2 and M1).The A␤40 (A) and A␤42 (B) levels in the TSinsoluble fractions from three mouse lines were determined as described under "Experimental Procedures."Each data point represents one mouse brain.The ELISA was carried out in duplicate for each brain, and the average was plotted.The bar indicates the mean.

TABLE I
Levels of A␤40 and A␤42 in the TS-insoluble fraction from the mouse brain Data are mean Ϯ S.D. values.Tg, transgenic.

TABLE II
Lipid composition of LDM fractions from four mouse lines with C57BL/6J background Data are mean Ϯ S.D. values (n ϭ 3 or 4, at the age of 2 months).Mann-Whitney's U test was used for comparison.

TABLE III
Lipid composition and A␤ levels in LDM fractions from another three mouse lines with FVB/N background Data are mean Ϯ S.D. values (n ϭ 3, at the age of 12.5 months).Note that the A␤42 levels in the K and P lines are not significantly different from those in the N line (p Ͼ 0.05, Mann-Whitney's U test).PC, phosphatidylcholine; PE, phosphatidylethanolamine; PS, phosphatidylserine; PI, phosphatidylinositol; SM, sphingomyelin; PL, phospholipids. a