Frameshifted β-Amyloid Precursor Protein (APP+1) Is a Secretory Protein, and the Level of APP+1 in Cerebrospinal Fluid Is Linked to Alzheimer Pathology*

Molecular misreading of the β-amyloid precursor protein (APP) gene generates mRNA with dinucleotide deletions in GAGAG motifs. The resulting truncated and partly frameshifted APP protein (APP+1) accumulates in the dystrophic neurites and the neurofibrillary tangles in the cortex and hippocampus of Alzheimer patients. In contrast, we show here that neuronal cells transfected with APP+1 proficiently secreted APP+1. Because various secretory APP isoforms are present in cerebrospinal fluid (CSF), this study aimed to determine whether APP+1 is also a secretory protein that can be detected in CSF. Post-mortem CSF was obtained at autopsy from 50 non-demented controls and 122 Alzheimer patients; all subjects were staged for neuropathology (Braak score). Unexpectedly, we found that the APP+1 level in the CSF of non-demented controls was much higher (1.75 ng/ml) than in the CSF of Alzheimer patients (0.51 ng/ml) (p < 0.001), and the level of APP+1 in CSF was inversely correlated with the severity of the neuropathology. Moreover the earliest neuropathological changes are already reflected in a significant decrease of the APP+1 level in CSF. These data show that APP+1 is normally secreted by neurons, preventing intra-neuronal accumulation of APP+1 in brains of non-demented controls without neurofibrillary pathology.

amino acids 66 to 81 of the N terminus of APP (17). Antibody specificity was tested by staining the recombinant APP ϩ1 protein on a Western blot with pre-immune serum, the AMY6 antibody, and pre-adsorbed AMY6 antibody.
Cell Lines and Transfections-The human neuronal SH-SY5Y cell line (American Type Culture Collection number CRL-2260) (18) was cultured in high glucose Dulbecco's modified Eagle's medium supplemented with 10% (v/v) fetal calf serum, 100 IU/ml penicillin and 100 g/ml streptomycin (all media and chemicals from Invitrogen). For Western blot analysis the cells were seeded in 6-cm dishes (Nunc, Roskilde, Denmark). The next day, cells were transfected with 10 g of plasmid DNA per 6-cm dish according to the calcium-phosphate transfection method. The human APP 695 isoform (kindly provided by Dr. T. Hartmann, Heidelberg, Germany) and APP ϩ1 (APP 695 ⌬GA exon 9) were cloned into pcDNA3 (Invitrogen). The full sequence of both constructs was confirmed by sequencing.
Human Cerebrospinal Fluid-Post-mortem non-hemolytic ventricular CSF was obtained from 50 non-demented controls and 122 neuropathologically confirmed AD patients. In addition, brain homogenates from four non-demented controls and eight AD patients were obtained. The CSF samples and brain tissues were collected by the Netherlands Brain Bank, Amsterdam (coordinator Dr. R. Ravid). Sex, age, brain weight, post-mortem delay, pH of CSF, and clinicopathological data of the patients described in this study can be found in the supplemental material in the on-line version of this article. The concentration of protein content of the CSF was determined with a Bradford assay (19).
Recombinant His-tagged APP ϩ1 -Full-length human APP ϩ1 cDNA was cloned into pQE31 (Qiagen, Hilden, Germany). Escherichia coli XL1-Blue cells were transformed with pQE31-APP ϩ1 , and recombinant protein production was induced with 1 mM isopropyl-1-thio-␤-D-galactopyranoside. The recombinant His 6 -tagged APP ϩ1 was purified over a nickel-nitrilotriacetic acid column and used as a positive control on the Western blot and for determining the specificity of the APP ϩ1 radioimmunoassay (see below).
Radioimmunoassay-Fifty microliters of cerebrospinal fluid was di-rectly measured in the radioimmunoassay (RIA) in duplicate. Standard peptides and samples were diluted in RIA buffer (50 mM Tris, 140 mM NaCl, 10 mM EDTA, and 10 g/liter bovine serum albumin, pH 8.0). Fifty microliters of standard or sample was incubated at 4°C for 48 h with 50 l of antiserum (AMY6; 1:10,000) and 10 l (10,000 cpm) of 125 I-AMY6 peptide. AMY6 peptide was iodinated by the chloramine T method. To precipitate the antibody peptide/protein complex, 50 l of cellulose coated with a secondary antibody against rabbit IgG (Saccel; IDS LTD, Boldon, UK) was added and incubated at 4°C for 1 h. The samples were centrifuged at 5,000 rpm for 15 min, and the pellets were counted in a Cobra ␥-counter for 5 min. The sensitivity of the RIA is 15 pg of APP ϩ1 per 50 l.
Characteristics of APP ϩ1 RIA-The AMY6 peptide was iodinated and subsequently purified on a Sepharose G25 column to separate free 125 I-sodium, non-iodinated AMY6, and 125 I-AMY6. The binding capacity of five different bleedings of the AMY6 antibody and the pre-immune serum to 125 I-AMY6 was tested. A maximum binding of 80% could be reached with an antibody dilution of 1:1000. A clear increase in binding capacity was observed between the first bleeding (170697) and the later bleedings of the AMY6 serum. Pre-immune serum did not bind to 125 I-AMY6 peptide at all. All subsequent RIAs were done using AMY6 bleeding 050897.
The optimal antibody-peptide binding for the APP ϩ1 RIA was reached with AMY6 bleeding 050897 at a dilution of 1:20,000. We performed displacement curves by adding an increasing amount of non-labeled AMY6 peptide or recombinant His 6 -APP ϩ1 to AMY6 (dilution 1:20,000, bleeding 050897; results not shown). The slopes of the graphs obtained with peptide and the recombinant protein are identical, indicating that this assay can reliably measure full-length APP ϩ1 . In our subsequent experiments we used the AMY6 peptide in the standard curves of the RIA.
Statistical Analysis-The Kruskal-Wallis nonparametric analysis of variance with multiple comparison of groups (20) was used to test differences between groups (program developed by J.M. Ruijter (Department of Anatomy and Embryology, Academic Medical Center, Amsterdam).

APP ϩ1
Is Secreted by Human Neuronal Cells-To determine whether APP ϩ1 is a secretory protein like sAPP␣ and sAPP␤, we transfected human neuronal SH-SY5Y cells with the APP and APP ϩ1 pcDNA3 constructs. The expression of APP and APP ϩ1 was driven by the cytomegalovirus (CMV) promotor to ensure high expression. The cells and their supernatants were harvested 1 day after transfection. Western blots of the cell pellets and the supernatants were probed with the 22C11 antibody ( Fig. 2A) directed against residues 66 -81 of APP as well as APP ϩ1 . The AMY6 antibody (Fig. 2B), directed against the unique C terminus of APP ϩ1 , was used to detect APP ϩ1 specifically. The Western blot analysis revealed the presence of both APP ( Fig. 2A, #) and APP ϩ1 ( Fig. 2A, *) in the cell lysate ( Fig.  2A, lys) and the cell culture medium ( Fig. 2A, med) with the 22C11 antibody. In addition, endogenous APP is visible in all transfection conditions ( Fig. 2A, arrowheads). A second blot of the same experiment was probed with APP ϩ1 specific antibody AMY6 and showed only staining of the APP ϩ1 proteins ( Fig.   FIG. 1. Schematic representation of the APP and APP ؉1 proteins. The epitopes recognized by the several different APP antibodies used in this study are indicated. The APP ϩ1 molecule depicted consists of 348 amino acids and is the ϩ1 protein of the 695 splicing variant, thus, without the Kunitz fragment. The distinct ϩ1 C-terminal amino acid sequence is indicated in italics. TM, transmembrane domain; SP, signal peptide. 22C11 is a monoclonal APP antibody directed against amino acids 66 -81 (17). The antibody directed against the AMY6 peptide is used for the RIA and Western blot. As this peptide contains an additional N-terminal tyrosine it can easily be iodinated.
2B, *) in the cell lysate ( Fig. 2B, lys) and the cell culture medium (Fig. 2B, med). Recombinant APP (from the transfected cells; Fig. 2A, #), endogenous APP and sAPP ( Fig. 2A, arrowheads) was stained with 22C11 and not at all with the APP ϩ1 -specific antibody AMY6. The intracellular APP ϩ1 is present in two forms, probably reflecting the non-glycosylated and O-glycosylated forms of APP ϩ1 . Such a doublet has also been observed by others for APP 695 , and they also argue that the doublet consists of an O-glycosylated and non-O-glycosylated form (21). In addition, Hersberger et al. (16) have also discussed the O-glycosylation of APP ϩ1 . Only the higher band of APP ϩ1 is present in the cell culture medium, which is likely to be the fully processed, glycosylated form of APP ϩ1 . This band does not run exactly at the same height as the higher APP ϩ1 band in the cell pellet because of the presence of albumin in the cell culture medium. In the lanes of the mock transfected cells as well as the APP transfected cells, two bands, which represent proteins with an apparent mass between 50 and 60 kDa, reacted with the 22C11 antibody. These proteins might be degradation products of APP or N-terminal fragments. Because AMY6 does not react with these bands, it is excluded that these proteins represent endogenous APP ϩ1 .
Quantification of APP ϩ1 -The RIA developed to measure APP ϩ1 in CSF was validated by analyzing temporal and frontal cortex homogenates of four non-demented controls and eight AD patients (Table I). The latter have a confirmed GA deletion in either exon 9 or exon 10 in part of the APP transcripts (4). In the cortex homogenates of AD patients a 3.4-fold increase in intracellular APP ϩ1 levels could be measured compared with the non-demented control group (Fig. 3A). To establish the presence of APP ϩ1 in CSF and determine the differences between non-demented controls and AD patients, we assayed post-mortem ventricular CSF collected by the Netherlands Brain Bank. In contrast with the data on the cortex homogenates, we measured significantly lower levels of APP ϩ1 in CSF samples of AD patients (Fig. 3B). CSF of non-demented controls contained 3.4 times more APP ϩ1 . With respect to the age of the subject and pH of the CSF, there was no significant difference between the control subjects and the Alzheimer patients. The value of the pH of CSF is an indication of the agonal state of the patients (22,23). A significant difference between the groups was found in the post-mortem delay; however, no correlation was found between post-mortem delay and APP ϩ1 levels or protein levels. Furthermore, a significant decrease in brain weight of AD patients, which is an inevitable characteristic of the disorder, as well as a significant decrease in CSF total protein content were observed (Table II).
Western Analysis of APP ϩ1 in CSF-The nature of the APP ϩ1 immunoreactivity in CSF was determined by a Western blot. First, we determined whether the AMY6 antibody, which is directed against the AMY6 peptide, will stain purified His 6 -APP ϩ1 on a Western blot. This protein is produced in a prokaryotic expression system, and, therefore, no glycosylation of APP ϩ1 takes place. Furthermore, six histidines are added to the N terminus, and the signal peptide sequence is not cleaved off from His 6 -APP ϩ1 . Consequently, His 6 -APP ϩ1 (Fig 4A, lane  3) will run at a slightly higher molecular mass than the secreted endogenous APP ϩ1 (ϳ50 -60-kDa band indicated with an asterisk in Fig. 4B, lane 2). The same blot probed with a buffer without the first antibody or with pre-adsorbed AMY6 showed no signal at all (Fig. 4A, lanes 1 and 2). In human CSF of a non-demented control (NBB92-030, female, 78 year-old, ApoE 33), a banding pattern was observed after staining with the AMY6 antibody (Fig. 4B, lane 2). The band of ϳ50 -60 kDa (asterisk) is most likely the APP ϩ1 band, because this band has been significantly reduced in the blot probed with pre-adsorbed antibody (Fig. 4B, lane 1). In agreement with the findings in the culture medium of the APP ϩ1 transfected SH-SY5Y cells, only the fully processed form of APP ϩ1 protein is secreted. Fig.  4C provides a direct comparison between His 6 -APP ϩ1 , APP ϩ1 in lysate, and APP ϩ1 in a cell culture medium of APP ϩ1 transfected cells. On the same gel, the two different pre-stained molecular mass markers used in the study were loaded, i.e. in the left lane (Fig. 4C) with the Multimark® multi-colored standard of Invitrogen and in the right lane (Fig. 4C) with the Rainbow marker of Amersham Biosciences. It is clear from this  blot that the several different forms of APP ϩ1 all have an apparent molecular mass between 50 and 60 kDa.
APP ϩ1 Levels in CSF and Association with Neuropathology-The analysis of APP ϩ1 in CSF showed a particularly striking inverse correlation between the concentration of APP ϩ1 in the CSF and the neuropathological stage of the patients, i.e. the Braak stages (24). Braak staging of the autopsy brain material was performed previously by W. Kamphorst. Fig. 5A shows that a higher Braak stage, i.e. with more severe cortical Alzheimer changes, is strongly related to a decline in the amount of APP ϩ1 in the CSF. No such correlation was found in total protein content, although a significant decrease was measured in the CSF samples of AD patients  compared with non-demented controls (Braak 0 -2) (Fig. 5B). After correction for protein content of the CSF, the amount of APP ϩ1 per milligram of total protein shows a rapid and significant decline in APP ϩ1 levels between Braak stages 0 and 1, but the levels remain constant at higher Braak stages (Fig. 5C). These findings imply that measuring APP ϩ1 levels in CSF indeed can be used as an ante-mortem test to diagnose early AD changes before the onset of clinical manifestation of the disease. DISCUSSION In this study we demonstrate that the APP ϩ1 protein is normally secreted by neuronal cell lines transfected with APP ϩ1 . Furthermore, we show that APP ϩ1 is present in CSF of non-demented controls with no neuropathology and that the concentration of APP ϩ1 is decreased in patients with AD pathology. In earlier studies we have shown that APP ϩ1 accumulates in the neurons in the cortex and hippocampus of AD patients (4), indicating that the neuronal secretion of APP ϩ1 in these patients has stopped. This process of intraneuronal retention and accumulation of APP ϩ1 already starts in clinically non-demented controls with initial neuropathology. 2 Moreover, we show that the concentration of APP ϩ1 in CSF is closely related to the grade of neurofibrillary pathology in AD patients. Even minor pathological changes are reflected in a decrease in the APP ϩ1 level in CSF. Hence, the APP ϩ1 RIA can potentially be used for diagnosing AD in patients with initial neuropathology.
APP is a type I trans-membrane protein that is cleaved by secretases, resulting in secretory forms of APP that are either secreted by the constitutive (25) or, as a recent study suggests, by the regulated pathway (26). Given that APP ϩ1 consists of the first 329 N-terminal amino acids of APP and therefore contains the signal peptide moiety, we anticipated that APP ϩ1 would also be secreted by neurons and could even be detected in CSF. In the present study, we show by transfecting human neuronal cells with APP ϩ1 that APP ϩ1 is indeed a secretory protein with an apparent molecular mass between 50 and 60 kDa. The calculated molecular mass of APP ϩ1 , however, is 38 kDa. Still, this difference in calculated and observed molecular mass is expected, because a similar deviant SDS-PAGE migration pattern has been observed for the several different splice forms of APP (27). The aspartate/glutamate-rich acidic region at amino acid positions 230 -260 causes the aberrant migration pattern on the SDS-PAGE gel. In our earlier study (4) on APP ϩ1 in human brain homogenate, we have reported on a 38-kDa APP ϩ1 band. In this study, we used a different APP ϩ1 antibody (AMY1) to detect the protein on a Western blot. The antibody used in the present study (AMY6) is much more sensitive on a Western blot. By the current transfection studies we have proven that the observation of the 38 kDa band in the earlier study does not represent the full-length APP ϩ1 . It might be a C-terminal fragment of APPϩ1, similar to the 38 kDa band in Fig 4, A and B. Despite the observation that neuronal cells transfected with APP ϩ1 secrete APP ϩ1 readily, we were surprised to find APP ϩ1 in the CSF of non-demented controls.  Thus far we had only observed the presence of APP ϩ1 mRNA in the hippocampus and cortex of AD and Down syndrome patients and the APP ϩ1 protein in the neuritic plaques and neurofibrillary tangles of these brain areas in AD and Down syndrome patients (4,6).
The failure to detect a GA deletion in the mRNA of APP in non-demented controls can be explained by the relatively low sensitivity of the immunoscreening assay we used in our earlier study (4). In the cDNAs from APP mRNA isolated from the cortex and hippocampus of AD brains, we found between 2 and 12 APP ϩ1 immunopositive clones of 20,000. It is therefore conceivable that we missed the mutation in the two non-demented control patients we screened in that study. Recent extensive studies on the frequency of molecular misreading of APP in the cell lines and temporal cortex of non-demented control, AD, and Down syndrome patients showed that a low frequency of GA deletions in APP mRNA, i.e. 1:100,000, occur in all studied tissues. 3 Furthermore, our group did report on the presence of dinucleotide deletions in mRNAs of ubiquitin B RNA isolated from the cortex of elderly non-demented controls, indicating that molecular misreading is not restricted to AD patients (4). Recent stainings of vibratome sections of the cortex and hippocampus of non-demented controls also showed that APP ϩ1 immunoreactivity is present in beaded fibers in these brain areas. 4 This immunocytochemical technique is more sensitive compared with the staining on paraffin sections that was applied in our initial study on APP ϩ1 (4). Moreover, with the vibratome technique we also observed APP ϩ1 in the neurofibrillary tangles and neuritic fibers in plaques of clinically non-demented controls with initial neuropathology. Therefore, the accumulation of APP ϩ1 in the neuropathological hallmarks of AD is likely to be caused by a deficiency in the secretion of this protein, given our present findings and the earlier publication of Hersberger et al. (16) that APP ϩ1 and the enhanced green fluorescent protein-tagged APP ϩ1 are secreted.
The analysis of the CSF samples of non-demented controls and AD patients showed clearly that the APP ϩ1 level was significantly decreased in neuropathologically confirmed AD patients . The RIA technique to measure APP ϩ1 is highly specific, because there is direct competition between endogenous APP ϩ1 and iodinated-peptide for the same antibody. Fig. 4B shows that the APP ϩ1 antibody used in this study, AMY6, recognized several different proteins on Western antibody. One specific band was clearly visible with an apparent molecular mass of ϳ50 -60 kDa. B, endogenous APP ϩ1 detection in CSF with the AMY6 antibody showed a pattern of several bands; the band assigned with the asterisk (*) is APP ϩ1 , because this band specifically disappears in a blot probed with pre-adsorbed AMY6. C, His 6 -APP ϩ1 (6ϫHis-APP ϩ1 ), cell lysate of APP ϩ1 , transfected SH-SY5Y cells, and the culture medium of these cells were put on one gel to compare directly the differences in molecular mass between these different APP ϩ1 proteins. The blot is stained with the AMY6 antibody. The molecular mass marker on the left is the MultiMark® multi-colored standard of Invitrogen, and the molecular mass marker on the right is the Rainbow marker of Amersham Biosciences.
blot, but pre-adsorption with the same peptide as used in the RIA showed a diminished intensity of mainly a ϳ50 -60 kDa band, indicating that this band likely represents the APP ϩ1 which is measured in the RIA. The decrease in APP ϩ1 in the CSF of AD patients supports the idea that neuropathology is preceded by a deficiency in protein secretion by neurons. Also, other hormones and secretory proteins, such as melatonin (28) and sAPP (12), are decreased in CSF of AD patients. Another explanation for the drop in the concentration of APP ϩ1 is that the enlargement of the ventricles may cause a subsequent dilution of CSF proteins, because the volume of ventricular CSF in AD patients has been shown to be twice as much as in non-demented controls (29). Furthermore, ventricular dilatation and reduced CSF production will presumably result in a progressive reduction in CSF turnover during aging (30). An alternative reason for the reduced levels of APP ϩ1 in the CSF of AD patients could, therefore, be the proteolytic breakdown of APP ϩ1 in CSF. From the patients we analyzed, there is no information available on the degree of ventricular dilatation. Therefore, we decided to measure the total protein content of the CSF, which will provide information on the dilution of CSF and, therefore, these data might reflect the degree of ventricular dilatation. As can be observed in Fig. 5B, the protein content of CSF obtained from AD patients with Braak stage 4 -6 is significantly less compared with that from the patients with Braak 0 -2. It is also clear from Fig. 5 that dilution of the CSF is not the cause of the decrease in APP ϩ1 levels. A strong argument in favor of a specific reduction of APP ϩ1 secretion in patients with early AD changes comes from the Braak stage 1 group. In these controls less than half of the APP ϩ1 levels of the Braak stage 0 group has been found (Fig. 5A), but no decline in total protein was observed (Fig. 5B). It is highly unlikely that a ventricular dilatation by more than a factor two would already have occurred in these non-demented controls with only very mild AD changes.
The clinical diagnosis of probable and possible AD is largely based on neuropsychological examinations. Although the accuracy of the clinical diagnosis has improved, a definite diagnosis still can only be made after autopsy (1). A biomarker that can aid the clinical diagnosis and even detect early neuropathological changes would be extremely valuable (31)(32)(33)(34). In this respect, CSF markers are expected to be useful, because the biochemical processes in the brain are likely to be reflected by the proteins that are present in the CSF. A␤ 42 and sAPP levels have often been reported to be decreased in CSF of AD patients, whereas A␤ 40 levels were unchanged. The former finding also indicates that there is an initial problem with protein secretion in neurons. Another protein that plays an important role in the pathogenesis of AD is the microtubule-associated protein tau (35). The level of tau protein is increased in CSF of AD patients (36) and patients with mild cognitive impairments (37), probably reflecting neuronal death. Recently, it has been shown that altered tau and 〈␤42 concentration can help to diagnose AD patients in subjects with mild cognitive impairments (38). In post-mortem ventricular CSF it has been shown previously that the melatonin concentration is closely correlated to the Braak stage. In patients with more severe neuropathology a lower concentration of melatonin was measured (28), which is in agreement with our findings. In conclusion, the measurement of AD-related proteins in CSF, including APP ϩ1 , can be of great use in improving the clinical diagnostic accuracy of AD (24), which is, at present, only definite after autopsy.
In this manuscript we show that APP ϩ1 is a ϳ50 -60 kDa secretory protein. Furthermore, we provide evidence that APP ϩ1 is already retained in neurons in the brains of nondemented controls with initial AD pathology. This retention probably reflects an impaired capacity of protein secretion and other early pathological changes in affected neurons or dying neurons. Measuring levels of secretory proteins, like APP ϩ1 , in CSF can help to reveal these early deficits in the function of neurons. Above all, the strong correlation between APP ϩ1 levels in CSF and observed pathological changes could help in diagnosing AD at an early stage.  (24) independently and prior to CSF detection of the APP ϩ1 levels. The non-demented control group consisted of 19 Braak stage 0 (no neurofibrillary changes), 19 Braak stage 1, and 12 Braak stage 2 patients (neurofibrillary changes confined to transentorhinal region). The AD group consisted of 19 Braak stage 4 (severe neurofibrillary changes in entorhinal and transentorhinal regions), 56 Braak stage 5, and 47 Braak stage 6 patients (isocortical destruction). Braak 3 patients were left out of these analyses because they already have senile cortical changes, and some of them suffer from dementia. The boxes depict the 25th and 75th percentile values with the median value given by the horizontal line in each box. The whiskers range from the 10th to the 90th percentiles. The statistical analysis of APP ϩ1 concentration and protein content in CSF samples are shown in the crosstables; significance tested with Kruskal-Wallis (20) test; *, p Ͻ 0.05; NS: not-significant. For example, in panel A the Braak 0 group shows a significantly higher concentration of APP ϩ1 in CSF compared with Braak stages 4, 5, and 6. No significant difference was found between the Braak stage 0 group and Braak stages 1 and 2. This multiple comparison of the different Braak stages showed a significant decrease of the absolute APP ϩ1 concentration in Braak stages 4, 5, and 6 (see cross-table in panel A) and a significant decrease in the APP ϩ1 levels corrected for protein content in Braak stages 1-6 (see cross-table in panel C).