Abstract
Familial forms of Alzheimer's disease (FADs) are caused by the expression of mutant presenilin 1 (PS1) or presenilin 2. Using DNA microarrays, we explored the brain transcription profiles of mice with conditional knock-out of PS1 (cKO PS1) in the forebrain. In parallel, we performed a transcription profiling of the hippocampus and frontal cortex of the FAD-linked ΔE9 mutant transgenic (TG) mice and matched controls [TG mice expressing wild-type human PS1 (hPS1)]. When the TG and cKO datasets were cross-compared, the majority of the 30 common expression alterations were in opposite direction, suggesting that the FAD-linked PS1 variant produces transcriptome changes primarily by gain of aberrant function. Our microarray studies also revealed an unanticipated inverse correlation of transcript levels between the brains of mice that coexpress ΔE9 hPS1+ amyloid precursor protein (APP)695 Swe and ΔE9 hPS1 single transgenic mice. The opposite directionality of these changes in transcript levels must be a function of APP and/or APP derivatives.
Introduction
Presenilins are highly homologous membrane proteins that play an essential role in processing of amyloid precursor protein (APP), leading to the production of β-amyloid (Aβ) peptides (for review, see Price et al., 1998; Selkoe, 2001; De Strooper, 2003; Van Gassen and Annaert, 2003). Mutations in mutant presenilin 1 (PS1) and presenilin 2 (PS2) lead to familial forms of Alzheimer's disease (FAD) (Rogaev et al., 1995; Sherrington et al., 1995) and are characterized by enhanced production of Aβ42 peptides (Borchelt et al., 1996; Duff et al., 1996; Scheuner et al., 1996; Borchelt et al., 1997). However, the cellular response(s) to FAD-linked PS1 variants are not fully understood and involves simultaneous “gain of function” and “loss of function” properties (Cai et al., 2003; Marjaux et al., 2004; Saura et al., 2004).
Over the last several years, we have seen a revolution in highthroughput expression profiling of diseased human brain tissue (Ginsberg et al., 2000; Ho et al., 2001; Loring et al., 2001; Colangelo et al., 2002; Mufson et al., 2002; Yao et al., 2003; Blalock et al., 2004) and various models of human brain disorders (Dickey et al., 2003, 2004; Marcotte et al., 2003; Mirnics et al., 2003; Gan et al., 2004; Wang et al., 2004). Although the expression profiling of AD tissue continues to generate data of enormous potential, the interpretation of the postmortem findings is greatly complicated by the nature of AD disease progress (Mirnics et al., 2001a; Marcotte et al., 2003; Mirnics and Pevsner, 2004).
In an attempt to identify genes for which transcription depends on normal PS1 function, we first performed a transcriptome profiling of the frontal cortex (FC) and hippocampus of mice with a conditional ablation of the PS1 gene. Once we defined the gene expression pattern of the PS1-deficient mice, the ΔE9-mutant human PS1 (hPS1) transgenic animal model attracted our attention (Lee et al., 1997; Lazarov et al., 2002). Inheritance of a single allele of PS1 with a deletion of exon 9 results in early onset FAD in several independent pedigrees. However, because the endogenous APP production in the mouse brain is low, these mice do not develop amyloid deposits or other pathology associated with AD, and as such, may mimic the premorbid phase of human AD. As a result, data obtained in brain transcriptome profiling of ΔE9 hPS1 mice would allow the assessment of the molecular consequences of the mutation but without the interference from cell loss or secondary transcriptome changes resulting from amyloid deposition or associated neuropathological features.
We focused our attention on the following questions in the current study. First, what are the gene expression consequences of PS1 ablation? Second, what are the transcriptome changes that arise from expression of human ΔE9 PS1 in mice? Third, are the genes for which expression level is modulated by conditional ablation of PS1 also regulated in mice carrying the ΔE9 hPS1 mutation? Finally, how do the transcriptome responses caused by expression of ΔE9 hPS1 alone relate to those in which this polypeptide is coexpressed with the FAD-linked APP695 Swedish variant?
Materials and Methods
Experimental animals
Because we were interested in early transcriptome events, all transgenic animals used in the current study were killed at 3 months of age. All animals used in this study have been described previously. Briefly, ΔE9 hPS1 mice and mice carrying the human wild-type (wt) PS1 express similar levels of PS1 transcripts, and the encoded proteins accumulate to levels observed in nontransgenic mice (Borchelt et al., 1997; Lee et al., 1997; Lazarov et al., 2002) but result in enhanced production of Aβ42 peptides. Both cDNAs are driven by the mouse prion protein promoter and are expressed in a ubiquitous manner across the mouse brain. However, both the human wild-type and mutant PS1 polypeptides fully replace the mouse PS1 polypeptides by competing for limiting stabilization factors (Thinakaran et al., 1996; Lee et al., 1997). The ΔE9 hPS1 mice show increased production of Aβ42 peptides but do not develop FAD microscopic pathology.
Mice coexpressing FAD mutant human PS1-ΔE9 and a chimeric mouse-human APP695 harboring a human Aβ domain and mutations (K595N, M596L) linked to Swedish FAD pedigrees (APPswe) have been described previously (Borchelt et al., 1997). Hippocampi of eight mice were analyzed on the same microarray platform as part of an independent environmental enrichment study. We established previously that the expression of the mutant PS1 transgene-encoded polypeptide, ΔE9, in the APP-overexpressing mice is identical to the levels seen in mice that express PS1ΔE9 alone (data not shown).
Conditional PS1 ablation in the forebrain is achieved via cre-lox recombination under calmodulin kinase 2 control (Feng et al., 2001). We have shown previously by in situ hybridization that PS1 mRNA expression is no longer present in the forebrain of 10-month-old animals, and we subsequently established that forebrain ablation of PS1 expression occurs at 6 months of age. Furthermore, in a parallel study of conditional knock-out (cKO) WT mouse PS1 (mPS1), using the same CamK11cre transgene to conditionally ablate PS1, Yu et al. (2001) demonstrated that PS1 protein expression is reduced by >90-95% at 6 months of age. The remaining levels of PS1 most likely represent PS1 expression in glial cells and vascular-associated cell types and is expected, because the conditional ablation occurs in a neuron-specific manner. Furthermore, the same study reported that PS2 expression is not altered as a consequence of ablating PS1 expression, whereas the levels of Aβ40 and Aβ42 were reduced by ∼73 and ∼25%, respectively. Unlike PS1/PS2-deficient mice (Saura et al., 2004), our mice with a conditional ablation of PS1 are viable and do not show alterations in postenrichment contextual memory.
Sample preparation and hybridization
Hippocampi and frontal cortices were rapidly dissected, frozen on dry ice, and stored at -80°C until RNA isolation. Total RNA was isolated using the Trizol reagent. RNA quality was assessed using the Agilent Technologies (Palo Alto, CA) Bioanalyzer. Reverse transcription (RT), in vitro transcription, and fragmentation were performed according to the recommendation of the manufacturer. Samples were hybridized onto MOE430A mouse Affymetrix (Santa Clara, CA) GeneChips, which contained >22,000 probe sets, using the Affymetrix hybridization station. To avoid microarray batch variation, only microarrays from a single lot were used. Microarrays were considered for use only if the average 3′:5′ ratio for glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and actin did not exceed 1:1.2. Of the 40 samples, three failed to meet these criteria. These were excluded from additional analyses and are denoted by the gray outline in the experimental design (see Fig. 1). Segmentation of scanned microarray images was performed by Microarray Analysis Suite 5.0. (Affymetrix). Determination of expression levels and scaling were performed using robust multichip average (Irizarry et al., 2003a,b).
Experimental summary. Top, The experimental series was performed using 40 MOE430A GeneChip oligonucleotide arrays with >22,000 gene probe sets. ΔE9 hPS1 transgenic animals were compared with mice carrying the human wild-type PS1 gene, and conditional knock-out mice were compared with mice that did not undergo cre-lox recombination. The analysis was performed on hippocampus and frontal cortex tissue. Thirty genes were identified as differentially expressed across the HC and FC of both ΔE9 hPS1-wt hPS1 and cKO mPS1-wt mPS1 comparisons. Bottom, Two-way clustering of the normalized expression levels for these 30 genes separated the mouse genotypes according to their expression phenotype. In the vertical dendrogram, each arm represents a single animal (red, cKO; blue, ΔE9 hPS1; light green, wild-type mouse PS1; dark green, wild-type hPS1), and rows denote gene probe sets with National Center for Biotechnology Information accession numbers. Note that the cKO mice and ΔE9 mice show the largest Euclidian distance, whereas the wt mPS1 and wt hPS1 animals cluster adjacently. For gene names and statistical parameters, see Table 1.
Data analysis
Identification of differentially expressed genes. We identified genes as differentially expressed in the wt hPS1-ΔE9hPS1 and wt mPS1-cKO mPS1 comparisons if they, first, reported >0.2 average log ratio in both the hippocampal cortex (HC) and FC comparisons and, second, if the p value for both the FC and HC comparisons was <0.05. These combined criteria further increased the reliability of the data by elimination of significant, but very small, expression changes that may have a marginal biological effect (Mirnics and Pevsner, 2004).
Calculation of pooled significance. Because the expression levels are different between brain regions, the outcomes of the comparisons have to be compared independently. For calculating the combined significance between the transgenic (TG) and cKO comparisons (Table 1), the χ distribution was calculated as follows: -2*[ln(p value for TG comparison) + ln(p value for cKO comparison)]. The calculation was performed with four degrees of freedom.
Common TG-cKO changes
Correlations. Correlations were calculated using Pearson's r value for the log2 ratios between the two compared conditions.
False discovery ratio estimate. False discovery ratio (FDR) was assessed in a permutation test, in which equal numbers of experimental and control microarrays were randomly assigned to two groups (supplemental material 1, available at www.jneurosci.org). These groups were assessed for gene expression differences with the same analysis as in the experimental-control comparison. FDR was established for each of the five permutations performed, and the mean FDR was calculated by averaging the FDR obtained in each of the permutation tests.
Clustering. Two-way clustering (sample and gene vectors) was performed on RMA-generated log2-transformed expression levels using Euclidian distance measurement in Genes@Work developed by IBM (IBM Corporation, White Plains, NY) (Lepre et al., 2004).
Custom database. RMA normalized data and statistical measurements were imported into Microsoft Access (Microsoft, Seattle, WA). This database is searchable by significance, accession number, log ratio, and gene name. The database displays individual RMA normalized data points across all experimental conditions.
Data sharing. The Microsoft Access database with all data points (∼130 MB) is available on request. All of the raw microarray data have been deposited into a Gene Expression Omnibus (Edgar et al., 2002; Wheeler et al., 2004) in a Minimum Information about a Microarray Experiment/Microarray Gene Expression Data Society (Brazma et al., 2001; Ball et al., 2002; Causton and Game, 2003) compliant format and are publicly available without any restrictions.
In situ hybridization
In situ hybridization was performed by methods described previously (Mirnics et al., 2001b). Briefly, after designing gene-specific primers, 600-900 bp amplicons were obtained in a standard PCR. The resulting products were cloned into a vector by T/A cloning. All clones were verified by sequencing. 35S-labeled antisense riboprobes were generated using T7-SP6 in vitro transcription and were cleaned and hybridized overnight to the tissue sections (2,000,000 dpm/slide). After washing, slides were exposed to x-ray film for up to 3 d and subsequently dipped in photoemulsion. Dipped slides were developed after 3-14 d, depending on the strength of the radioactive signal observed in film exposures. In all experiments, sense riboprobe was used as a specificity control.
Real-time quantitative PCR
For selected genes, quantitative PCR (qPCR) was performed on pooled samples from the five groups of animals (wt hPS1, ΔE9 hPS1, wt mPS1, cKO mPS1, and ΔE9 hPS1 × APP695). After primer validation for amplification efficiency (95-100%), the experiment was performed using standard Δ Ct-Sybr Green measurement protocols with two independent reverse transcriptions and four replicates per RT (Mimmack et al., 2004). GAPDH was used as a standard normalizer. Primer sequences are available in supplemental material 2 (available at www.jneurosci.org).
Results
RNA was harvested from HCs and FCs from five CKO mPS1 and five control wt mPS1 animals. Conditional ablation is achieved by cre-lox recombination under calmodulin kinase 2 control previously generated in our laboratory (Fig. 1, top right). These cKO mice are viable and do not develop any obvious pathology with aging. Each sample was hybridized to a single MOE430AGeneChip (20 arrays in total). The obtained intensity measurements were normalized by RMA (Irizarry et al., 2003a,b) and imported into a custom-made Microsoft Access database. Because we were interested in the overall influence of the PS1 ablation on the transcriptome of two AD-affected brain regions, we focused our attention to expression changes that were present in both HC and FC. This analysis identified 85 genes that were differentially expressed across both regions (Fig. 1) (supplemental material 3, available at www.jneurosci.org) and will be referred to as the “KO comparison.” For these 85 genes, the correlation between the outcome of the HC and FC comparisons was r = 0.96 (p < 0.0001). The number of genes with increased and decreased expression was evenly distributed, with transcript reductions showing a greater magnitude than transcript inductions.
Once we defined a PS1-dependent transcript network, we were interested to determine whether the expression profiles of the cKO animal share any commonalities with that of humanized mice carrying the ΔE9 hPS1 mutation. As a result, we chose to perform a microarray analysis on the HC and FC of the ΔE9 hPS1 mutants using a similar experimental design used in the cKO experiment. We hypothesized that the most critical expression changes produced the conditional ablation of PS1 might also appear in the ΔE9 hPS1 variant. We speculated that if the mutant human PS1 polypeptide was a hypermorph, or “gain of function” species, differences in ΔE9 hPS1 mice would be in the opposite direction than in the PS1 cKO animals. However, if both the ΔE9 hPS1 and cKO PS1 animals show expression changes in the same direction, this would suggest that these changes are related to the “loss of function” effect of the human mutant PS1. When compared with the wt hPS1 transgenic animal (also referred to as a “TG comparison”), ΔE9 hPS1 mice reported 71 gene expression changes (supplemental material 4, available at www.jneurosci.org). The expression ratios of these 71 genes identified as differentially expressed were highly correlated between the HC and FC comparisons (r = 0.97; p < 0.0001). Twenty-two genes were downregulated, whereas 49 genes were upregulated in these animals. Contrary to the cKO comparison, the ΔE9 hPS1 mice showed more robust transcript upregulations than mRNA level reductions. Selected expression changes were verified for several genes on two additional pairs of transgenic animals using in situ hybridization (Fig. 2). In these experiments, maternally expressed gene 3 (MEG3), exostoses-like 2 (EXL2), hippocampus-abundant transcript 1 (HIAT1), activity-related cytoskeletal protein (ARC), and amylase 1A (AMY1A) all validated the microarray-predicted expression changes between the two sets of transgenic animals.
Verification of TG comparison data by in situ hybridization. Low-magnification, dark-field composite micrographs of ΔE9 hPS1 (right column) and wt hPS1 (left column) mice brain sections. Riboprobes for EXL2, HIAT1, MEG3, and AMY1A were hybridized to 20-μm-thick coronal sections using methods described previously. For the investigated genes, the in situ hybridization data were in concordance with the microarray findings.
When the two datasets were cross-compared, 30 genes showed a robust and significant change in both TG and cKO comparisons. When the normalized adjusted intensities for the whole dataset were two-way clustered for the commonly changed 30 genes, the four experimental groups (wt mPS1, hPS1, ΔE9 hPS1, and cKO mPS1) separated into distinct clusters (Fig. 1, bottom). The ΔE9 hPS1 and the cKO mPS1 mice showed the greatest Euclidian distance, with the two control groups (hPS1 and mPS1) clustering in between. For the group of 30 genes, there was a high correlation in the expression ratio between the HC and FC across the two conditions and within the ΔE9 and cKO mice (Fig. 3). FDR assessment by permutation of the dataset suggested that by random chance, less than one gene would show the observed gene expression changes (<2% of the observed changes) (supplemental material 1, available at www.jneurosci.org)
Coregulation of genes across the TG and cKO comparisons. A, B, Expression changes in the hippocampus (A) and frontal cortex (B) for the 23 genes changed in the opposite direction. The x-axis represents TG comparison log2 ratio, and the y-axis denotes cKO comparison log2 ratio. C, D, Expression changes of seven genes that were regulated in a similar direction in both the TG and cKO comparisons. Graph layout is similar to that in A and B. Inserted tables in A and C denote statistical correlations across the TG and cKO comparisons and across the two brain regions (HC and FC). Note the high degree of transcript coregulation across both the TG and cKO comparisons and HC and FC.
Of the 30 genes, 23 genes were regulated in the opposite direction (updown or downup), whereas seven showed expression changes that were similar in direction (Table 1). For all genes, HC and FC data reported a consistent change of direction. Of the 23 transcripts that were regulated in an opposite direction in the TG and cKO mice comparisons, 11 genes reported increases in the ΔE9 hPS1 mice and decreases in the cKO mPS1 mice, and 12 genes reported a decrease in the ΔE9 hPS1 mice and increases in the cKO mPS1 mice. Of the seven genes that were changed in the same direction in both the ΔE9 and cKO mice, four genes reported transcript induction and three genes reported transcript repression. In addition to the 30 genes that showed unequivocal regulation in both systems, we identified an additional 72 genes that showed some evidence for regulation in both the ΔE9 and cKO comparisons (supplemental material 5, available at www.jneurosci.org). Although these gene probes satisfied only three of four statistical criteria we used in the current study [average log2 ratio (ALR) >0.20 and p < 0.05 in both TG and cKO comparisons], this group very likely contains biologically meaningful data, because of the following: (1) the majority of the probes in this group still reported an overall significance across the combined dataset; (2) eight genes were represented with multiple, independent probe sets that showed consistent findings; (3) 63 of the 72 probes showed consistent results between the HC and FC comparison; (4) two-way clustering of the 102 putatively changed genes resulted in an outstanding separation of the ΔE9 hPS1 and cKO mPS1 experimental groups (supplemental material 6, available at www.jneurosci.org); and (5) FDR assessment suggested that ∼80% of the observed gene expression changes can be attributed to real data discovery (supplemental material 1, available at www.jneurosci.org).
Although the expression of ΔE9 hPS1 leads to elevated Aβ42 production, the overall level of Aβ production is extremely low in the brains of nontransgenic mice. As a result, we felt it was critical to investigate ΔE9 hPS1-driven expression changes in the context of high-APP levels. To achieve this, we compared the expression in the hippocampus of five ΔE9 hPS1 mutant mice to that of eight ΔE9 hPS1 × APP695 Swe mice generated in our previous study (Lazarov et al., 2002). Surprisingly, the commonly regulated 30 genes that were the most strongly regulated in both the TG and cKO comparisons (Table 1) showed a strong inverse correlation (r =-0.72; p < 0.001) in the ΔE9 hPS1 × APP695 Swe versus ΔE9 hPS1 comparison: 25 of the 30 genes examined showed an expression change that was opposite in direction than observed in the ΔE9 hPS1 versus wt hPS1 comparison (Fig. 4).
Coregulation of gene expression between the ΔE9 hPS1 × APP695 double-transgenic and ΔE9 hPS1 single-transgenic animals. A, The 30 genes identified in the TG and cKO comparisons also showed robust transcription changes in the ΔE9 hPS1 × APP695 mice. The x-axis represents TG comparison log2 ratio (ΔE9 hPS1 vs wt hPS1), and the y-axis denotes ΔE9 hPS1 × APP695 versus ΔE9 hPS1 comparison log2 ratio in the hippocampus. The blue dashed line represents the trend line. Note that the 30 genes examined show a strong and inverse coregulation (r = -0.72; p < 0.001) across the two comparisons, suggesting a robust effect of APP on the ΔE9 hPS1 background. B-D, Individual gene expression changes across the ΔE9 hPS1 × APP695 versus ΔE9 hPS1 comparison. The x-axis represents sample class, and the y-axis denotes comparison log2 ratio. For gene abbreviations, see Table 1. Note that the majority of genes are regulated in different directions between the TG and ΔE9 hPS1 × APP695 versus ΔE9 hPS1 comparisons.
Finally, for seven of the 30 genes across the five different groups of mice (wt hPS1, ΔE9 hPS1, wt mPS1, cKO mPS1, and ΔE9 hPS1 × APP695 Swe), we decided to confirm the expression changes by real-time qPCR in the hippocampus (Fig. 5). In the qPCR assessment, Fos, DUSP1 (MAP kinase phosphatase-1, serine/threonine-specific protein phosphatase), ARC, EXL2, cyclin D1, C1Qb, and C1Qg expression levels were changed in the microarray-predicted direction. The microarray-reported expression changes (ALR) were highly correlated (r = 0.91; p < 0.001), with the qPCR reported expression changes (ΔΔCt) across the five different sample groups. The expression changes were more robust in the qPCR data than the GeneChip findings, suggesting that the microarray dataset may underestimate the actual expression differences.
RT-qPCR verification of microarray data for seven genes. A, The x-axis represents genes, and the y-axis denotes average ΔΔCt from two independent reverse transcriptions (4 replicates each). The green bars denote TG comparison (ΔE9 hPS1 vs wt hPS1), the red bars correspond to cKO comparison (cKO mPS1 vs wt mPS1), and the blue bars indicate expression change in the APP-PS1 comparison (ΔE9 hPS1 × APP695 vs ΔE9 hPS1). Error bars denote SD. For gene symbols, see Table 1. B, Correlation of micro array data with qPCR data. Thex-axis represents qPCR ΔΔCt, and the y-axis denotes the ALR established in the microarray comparisons. Colors denote different genes, and shapes denote comparisons (triangle, TG comparison; circle, cKO comparison; diamond, APP-PS1 comparison). The green dashed line denotes the linear trend. Note that the data show a strong and significant correlation (r = 0.91; p < 0.0001).
Discussion
In this study, we identified the transcriptome profile of ΔE9 hPS1, wt hPS1, cKO mPS1, and wt mPS1 mice. Transcriptome responses observed in both the TG and cKO comparisons were highly correlated between the HC and FC. Thirty genes that were changed in the ΔE9 hPS1-wt hPS1 comparison also showed an expression change in the mPS1-ablated and wild-type mice comparison, suggesting that these expression changes represent the critical effect of the ΔE9 hPS1 variant on the transcriptome. In the second part of this study, the comparison of ΔE9 hPS1 with the ΔE9 hPS1 × APP695 Swe mice revealed that the combined expression of ΔE9 hPS1 and APP695 Swe has a very strong influence on the expression of the same 30 genes identified in the TG and cKO comparisons: 25 of the 30 genes reported transcript level changes that were opposite in direction to those observed in the ΔE9 hPS1 single mutants.
The effect of PS1 mutation on the transcriptome: gain of function or loss of function?
FAD-linked PS1 variants have been primarily considered to lead to a gain of function. However, recent data suggest that these FAD-linked PS1 variants, in addition to gain of function effects, may also lead to loss of function properties (Cai et al., 2003; De Strooper and Woodgett, 2003; Marjaux et al., 2004; Saura et al., 2004). Our dataset is very informative in this regard: 23 of 30 genes (and 72 of 102 genes in the extended dataset) showed expression alterations that were opposite in direction between the TG and cKO comparisons. These data strongly support the view that, at least at the transcriptome level, the expression changes associated with the ΔE9 hPS1 mutation are a result of a gain of function. However, it should be noted that the ΔE9 hPS1 expression did not result only in transcript increases. Rather, expression increases and decreases were evenly distributed (11 upregulations and 12 downregulations), suggesting that transcript downregulations may be an equally important consequence of the ΔE9 hPS1 mutation as the upregulations; the obtained data argues that the mutant ΔE9 hPS1 is also a potent negative regulator of transcript levels.
The minority of genes (seven transcripts) were regulated in the same direction in the TG and cKO comparisons. The finding that the ΔE9 hPS1 mutation produces some of the same effects as seen in PS1 ablation is not unexpected and represents a strong argument for loss of function events that may occur in conjunction with this mutation.
The critical PS1-dependent transcript network
The expression levels of 30 genes that we identified in both the TG and cKO comparisons strongly depend on PS1 expression and function. Multiple genes of this group have been previously associated with AD, neurodegenerative diseases, or cognitive performance. Increased immunoreactivity for Fos-related proteins in Alzheimer's disease has been observed in multiple studies (Zhang et al., 1992; Anderson et al., 1994; Marcus et al., 1998), and activation of c-Fos contributes to amyloid β-peptide-induced neurotoxicity (Gillardon et al., 1996). In addition to Fos and Arc, several other genes we found have been implicated in the pathophysiology of AD and/or cognitive processes. For example, strong upregulation of Gadd45, indicating DNA damage, is an early event in Aβ cytotoxicity (Santiard-Baron et al., 1999). Furthermore, the presence of neuronal nitric oxide synthase (NOS) in pyramidal-like neurons is a distinct characteristic of AD (Fernandez-Vizarra et al., 2004), and familial PS1 mutations in the N-terminal fragment cause NOS inhibitor-sensitive neuronal cell death (Hashimoto et al., 2004). In addition, C1q has been postulated to play a significant role in AD pathogenesis by multiple studies (Matsuoka et al., 2001; Luo et al., 2003; Veerhuis et al., 2003; Fonseca et al., 2004). In a transgenic animal model, Matsuoka et al. (2001) found that C1q levels increase as a result of fibrillar Aβ production, whereas Veerhuis et al. (2003) argue that C1q-containing Aβ deposits precede, or occur commensurate with, neurodegenerative changes in AD. Furthermore, Luo et al. (2003) suggest that the C1q may mediate neuronal injury during AD by contributing to neuronal oxidative stress and neuronal demise. Finally, it is known that Tau protein and, in some cases, neurofilament subunits exhibit abnormal phosphorylation on specific serine and threonine residues in AD and frontotemporal dementia (for review, see Morfini et al., 2002), and this may be directly related to the expression changes in DUSP1 we observed.
Recent studies of Dickey et al. (2003, 2004) are highly relevant for the interpretation of our dataset. Although these studies and our present dataset were generated with very different experimental parameters (age of mice, pooling, replicates, strain of mice, controls, and microarray platform used), they both report a downregulation of several genes (e.g., Arc, Egr, Fos) in the APP × PS1 double-mutant mice. Thus, the combined data suggest that the expression changes in at least some of the genes we observed is a persistent hallmark of the disease model. These changes appear to be present both before the pathology develops (3 months; our present dataset) and during the Aβ deposition phase (17-18 months) (Dickey et al., 2003). Furthermore, this downregulation is observed in two different lines of mutant mice [APPK670N,M671L × PS1-5.1 (Dickey et al., 2003) and APP695 Swe ×ΔE9 PS1 (our study)] and suggests that this process is directly related to the pathophysiological processes associated with the expression of mutant APP and PS1 variants, rather than the exact experimental model.
Directionality of expression changes
Regulation of the 30 genes across the different conditions also argues that these are part of a common, strongly coregulated, transcript network that depends on normal PS1 function. These genes indicate an intriguing expression pattern across the hippocampus of the TG, conditional KO, and ΔE9 hPS1 × APP695 Swe versus ΔE9 hPS1 comparisons, suggesting that they may be critically involved in pathophysiological changes that lead to experimental AD and perhaps human pathology. In this context, we interpret that the majority of expression changes in the TG comparison are the result of gain of function effects of the ΔE9 hPS1. However, because the level of endogenously generated Aβ42 in mouse brain is extremely low (Duff et al., 1996), the small increase in the levels of Aβ42 after expression of mutant PS1 may be stimulatory rather than toxic. Nevertheless, it is important to point out that in our system, the mutant PS1-dependent expression changes clearly depend on APP/Aβ levels. Once the same mutant PS1 is coexpressed with APP (in the ΔE9 hPS1 × APP695 Swe mice), the levels of APP, as well as the neurotoxic Aβ42 peptide, are markedly elevated, leading to expression changes that are opposite those observed in the ΔE9 hPS1 single-mutant mice. Based on our results, we argue that the physiological effects of expressing PS1 mutants have two essential components: one that is constant and is a direct result of the mutation, and a second that depends on APP and Aβ42 levels.
At present, the role of those genes that are subject to PS1-dependant regulation are not established, but we suggest that for at least a subset of these genes, the levels of expression may be critical for cognitive performance. For example, disruption of Arc, Fos, or Egr-1 expression, either by intrahippocampal administration of antisense oligonucleotides or by germline disruption, impairs consolidation of long-term memory formation (Guzowski, 2002; Steward and Worley, 2002; Vazdarjanova et al., 2002; Montag-Sallaz and Montag, 2003). Furthermore, elegant microarray studies on aging by Blalock et al. (2004) have also identified Arc and NR4A as important aging- and cognitionrelated genes. Thus, we speculate that the observed gene expression changes may be directly relevant for the pathophysiology of aging, cognition, and/or Alzheimer's disease.
Footnotes
This work was supported by a National Alliance for Research on Schizophrenia and Depression Young Investigator Award (K.M.), a National Institutes of Health (NIH) Training Grant and the Pittsburgh Institute for Neurodegenerative Diseases (Z.K.), a Craumer Endowment of Children's Hospital of Pittsburgh (N.F.S.), NIH Grant AG021494 (S.S.S., O.L., D.T.), and the Ellison Medical Foundation (S.S.S., O.L., D.T.). We thank Dr. Pat R. Levitt for valuable comments on this manuscript. We also thank Carmel F. Portugal for superb technical assistance with the experiments.
Correspondence should be addressed to either of the following: Károly Mirnics, Department of Psychiatry, University of Pittsburgh, School of Medicine, E1453 Biomedical Science, Pittsburgh, PA 15261, E-mail: karoly+{at}pitt.edu; or Sangram S. Sisodia, Center for Molecular Neurobiology, The University of Chicago, 947 East 58th Street, MC 0926, Chicago, IL 60637, E-mail: ssisodia{at}drugs.bsd.uchicago.edu.
Copyright © 2005 Society for Neuroscience 0270-6474/05/251571-08$15.00/0
↵* K.M. and Z.K. contributed equally to this work.