Cyclooxygenase-2 and Presenilin-1 Gene Expression Induced by Interleukin-1β and Amyloid β42 Peptide Is Potentiated by Hypoxia in Primary Human Neural Cells*

Lipid messengers and amyloid beta (Aβ) peptides generated by cyclooxygenase-2 (COX-2) and presenilin-1 (PS1) mediate pro-inflammatory signaling and neural degeneration in Alzheimer's disease (AD) brain. This study provides data showing that theCOX-2 and PS1 genes each transcribe rare, highly labile RNA species that display early response gene behavior in human neural (HN) cells in primary culture, down-regulation during human neural development, and up-regulation in AD neocortex and hippocampal CA1. Together, interleukin-1β and amyloid β42 peptide [IL-1β+Aβ42] synergistically activated COX-2 andPS1 gene expression preceded by increases in AP1-, STAT1α-, and in particular NF-κBp50/p65- and HIF-1α-DNA binding. These events were markedly potentiated by hypoxia and blocked by the antioxidant α-phenyl-N-tert-butyl nitrone. Broad transcription profiling further indicated that hypoxia-induced, [IL-1β+Aβ42]-treated HN cells display robust induction of COX-2 and PS1 as well as a pro-inflammatory gene family that includes NF-κBp50/p105,IL-1β precursor, and cytosolic phospholipase A 2 genes. These findings indicate a novel [IL-1β+Aβ42]-mediated, hypoxia-enhanced, free radical-triggered gene program that drives inflammatory gene signaling and suggest a mechanism by which hypoxia during aging contributes episodically to amyloidogenesis, inflammation, and AD pathophysiology.

During the microglial-mediated inflammatory response to A␤ deposition in AD, overexpression of IL-1␤ not only further promotes amyloidogenesis (20 -23) but also induces an oxygensensitive transcription factor (TF) series that includes AP1, STAT1␣, and NF-Bp50/p65 (24 -27). In the human association neocortex in AD, DNA-binding activities for these TFs, as well as IL-1␤, are sharply elevated (9,24,26). Therefore, IL-1␤triggered AP1-, NF-Bp50/p65-, and STAT1␣-sensitive genes have strong potential to propagate inflammatory signaling in the brain (20 -27). Although IL-1␤ up-regulates both ␤APP and COX-2 gene transcription and post-translational processing of ␤APP into neurotoxic A␤ peptides (20 -23, 27), IL-1␤-triggered neural inflammation may also be induced by hypoxia (28 -30). The hypoxia-inducible factor (HIF-1, a heterodimeric DNAbinding protein consisting of a cytokine-, oxygen-and growth factor-regulated HIF-1␣ and a constitutive HIF-1␤/aryl hydrocarbon receptor nuclear transporter element) normally promotes the expression of adaptive genes under hypoxic conditions. However, during pathophysiologic conditions involving ischemia and/or hypoxia, HIF-1␣ overexpression drives aberrant gene expression with rapid progression to the lethal phenotype (28,29). It is interesting that IL-1␤ partially mimics cellular hypoxia, leading to increased HIF-1␣-DNA binding and activation of COX-2 and PS2 in human gingival and syno-vial fibroblasts and in rat and monkey retinal neurons (30 -32). How hypoxia modulates oxygen-sensitive, pro-inflammatory transcription factor signaling and gene expression in human neural cells is not well understood. Because COX-2 regulates PG and lipid-messenger production and PS1 mediates the generation of A␤, which can trigger microglial activation and further release of IL-1␤, COX-2 and PS1 expression represents an exceptionally potent driver for pro-inflammatory signaling in the brain. Here we provide data showing that COX-2 and PS1 gene up-regulation in [IL-1␤ϩA␤42]-induced HN cells and in AD hippocampal CA1 contrasts with the coordinate down-regulation of basal COX-2 and PS1 gene transcription observed during human neurogenesis. These effects were greatly magnified in hypoxia-stressed, [IL-1␤ϩA␤42]-treated HN cells and blocked by the electron spin-trap agent and free radical scavenger ␣-phenyl-N-tert-butyl nitrone (PBN). The data also indicate that COX-2 and PS1 are basally rare, labile, and transient genetic participants in [IL-1␤ϩA␤42]-mediated, hypoxiaenhanced, pro-inflammatory signaling in HN primary cells, similar to those found in AD-affected brain. DNA microarray analysis further suggests that COX-2 and PS1 are under transcriptional control, in part by AP1-, NF-Bp50/p65-, STAT1␣-, and HIF-1␣-DNA binding, as are the related TATA-containing pro-inflammatory gene family members encoding NF-Bp50/ p105, IL-1␤ precursor, and cytosolic phospholipase A 2 (cPLA 2 ). These findings further suggest that episodic hypoxia during brain aging may drive intermittent COX-2, PS1, and related pro-inflammatory gene expression that contributes cumulatively to amyloidogenesis, brain inflammation, and AD neuropathology.
Human Neural Progenitor Cells in Primary Culture-HN cells derived from explanted human fetal brain tissue and cryo-preserved in primary passage as "spheroids" were grown to ϳ70% confluence (ϳ50,000 cells per 3.5-cm diameter well) in six-well Costar plates at 37°C, 5% CO 2 /20% O 2 /75% N 2 in humidified air at 1 atm (normoxic conditions) in NPMM medium (Clonetics CC-4241) supplemented with human fibroblast growth factor, neural survival factor-1, human epidermal growth factor, and G/A1000, as described by the manufacturer (Clonetics, Walkersville, MD). HN cells tested negatively for HIV-1, hepatitis B and C, mycoplasma, bacteria, yeast, and fungi and positively for the glial and neural markers GFAP, MAP2, and ␤-tubulin III (Clonetics). After 2 weeks of development, HN cells were exposed to IL-1␤, A␤42, A␤42s, [IL-1␤ϩA␤42], PBN, and/or normoxia or ambient hypoxic conditions (37°C, 5% CO 2 /5% O 2 /90% N 2 in humidified air at 1 atm) for 0, 1, and 3 h in an electronically metered hypoxia incubator (Model 1143, Series II, ThermoForma) in sterile Plexiglas chambers as previously described (31,32). Concentrations of O 2 were also continuously monitored using a dissolved O 2 meter (Corning Model 317) with the O 2 probe immersed in an adjacent well in NPMM culture medium. RNA and protein were isolated (see below) and stored at Ϫ81°C within minutes of removal of cells from normoxic or hypoxic conditions.
Control and Alzheimer Brain Tissues-Stringent criteria were used in the case selection of human brain tissues used in these studies. Hippocampal cornu ammonis 1 region (CA1) and superior temporal lobe neocortices (A22) from Consortium to Establish a Registry for AD and National Institutes of Aging-autopsy-confirmed control, and AD brains were obtained from hundreds of potential specimens from domestic and international brain banks. No AD tissues had a known clinical history of familial degenerative disease, and there were no statistically significant differences in age (control age, 68.8 Ϯ 1.9 years; AD, 69.2 Ϯ 1.5 years; p ϳ 0.8), post-mortem interval (control, 2.0 Ϯ 0.7 h; AD, 2.1 Ϯ 0.7 h; p ϳ 0.9), or tissue pH (control, 6.7 Ϯ 0.1; AD, 6.8 Ϯ 0.1; p ϳ 0.9) between the two brain groups (4,32,44).
COX-2 and PS1 RNA Basal Abundance and Kinetic Analysis in HN Cells and in Control and AD Human Brain-Basal abundance and decay kinetics were established for human-specific COX-2, PS1, and PS2 RNA message by an assay for total RNA in both HN cells and in control human neocortex and hippocampal CA1 using Northern dot blots and RT-PCR. No inhibitors of transcription or translation were used. Nuclei within these latter tissues still actively support run-on RNA polymerase II transcription to ϳ85% of control values up to the 3-h post-mortem interval (4). Aged HN cells were incubated at 37°C, and COX-2, PS1, and PS2 RNA message levels were compared with the RNA message levels for control ␤-actin signal in the same sample for 0 -24 h. RNA adenine-uridine-rich element (ARE) analysis was performed by Hitachi DNASIS MAX DNA sequence analysis software (Hitachi Genetics group) and/or TRANSFAC version 5.4 (BIOBASE, Wolfenbuttel, Germany). COX-2 and PS1 RNA message stabilities were used as a stringent guide in the selection of post-mortem human brain tissues for RNA analysis (4).
Isolation of Total and Poly(A) ϩ RNA, Northern, and RT-PCR Analyses and DNA Microarray Panel Hybridizations-Total RNA from HN cells or human brain tissues was isolated with a phenol-guanidine isothiocyanate reagent (TRIzol, Invitrogen). RNA isolation reagents were prepared from 0.2-m-filtered, diethyl pyrocarbonate-treated water, and human RNase inhibitors (Ambion 2682 and 2690), anti-RNase (Ambion 2692, 1 unit/l), and RNasin (Promega N2115, 1 unit/l) were used throughout the isolation procedures. Total RNA samples were analyzed on an RNA Bioanalyzer (Agilent Technologies) and had A 260 / A 280 ratios typically of Ն2.0. No significant differences in the spectral purity, rate of degradation, or molecular size or yield of total RNA between control and AD brain tissues or between control HN and ligand-treated RNA message populations were noted. Packed HN cells and brain RNA yields were ϳ100 g of total RNA from each 100-mg wet weight tissue/cell sample. Total RNA was fractionated into an enriched poly(A) ϩ fraction by RNase-free DNase treatment followed by incubation with biotinylated oligo(dT) (CLONTECH BD Biosciences, Palo Alto, CA, K1038). Enriched poly(A) ϩ mRNA was recovered on streptavidin-coated paramagnetic beads in a magnetic separator (Magne-Sphere Z5341, Promega Biotech). Poly(A) ϩ mRNA was converted into cDNA by Moloney murine leukemia virus RT in the presence of [␣-32 P]dATP (10 Ci/l, 100 Ci per reaction, 3000 Ci/mmol, Amersham Biosciences), nucleospin-column purified, and hybridized onto dot blots or DNA array panels in ExpressHyb hybridization solution (CLON-TECH BD Biosciences, S1135). DNA array panels (1184 human genes; CLONTECH BD Biosciences, 7850/7852; lots 0030637 and 0010880), each containing 10 ng of cDNA, were pretreated, prehybridized, hybridized, washed, and exposed according to the manufacturer's protocol PT3140 (CLONTECH BD Biosciences, Palo Alto, CA). RT-PCR was performed by the superscript one-step RT-PCR system (Invitrogen) and human-specific COX-1, COX-2, PS1, NF-Bp50/p105, IL-1␤, and cPLA 2 primers as previously described (4,9,31,32).
Statistical Analysis-For Western, Northern dot blot, RT-PCR, and DNA array analyses, signal-intensity data were gathered and averaged from three to five independent experiments by phosphorimaging onto molecular imaging screens (FUJI Film) for 3-36 h using a Typhoon Molecular Imager system (Amersham Biosciences). Statistical significance of the Western, Northern dot blot, and RT-PCR data was determined by two-way factorial analysis of variance (p, analysis of variance); error bars in Figs. 1, 2, 5-7, and 9 represent standard errors from the mean. For DNA array quantitation a "standard analysis" used 9 ϫ 10 6 cpm of probe per 96-cm 2 DNA array, an 18-h hybridization, and a 12-h exposure onto phosphor storage screens. Resulting data sets were analyzed, and features were extracted using Atlas version 2.0 (CLON-TECH BD Biosciences) and GeneSpring version 4.1.5 (Silicon Genetics, Redwood City, CA) bioinformatics algorithms. Signal parameters (signal ratio, difference, and DNA filter background correction) were adjusted so that gene targets showing only the largest treated versus control changes, that is, exceeding a factor of 2.7-fold (p Ͻ 0.05), were reported.

RNA Kinetic Analysis: Quantitation of Human COX-2, PS1
, and PS2 RNA Abundance and Stability-Preliminary inspection of COX-2 and PS1 RNA sequence structures suggested limited longevity. Experimentally observed and predicted RNA message-decay kinetics, derived from Northern blot and RT-PCR studies in HN cells in primary culture and computerassisted AU-rich element (ARE) sequence analysis (33)(34)(35) of COX-2, PS1 variants 374, 463, and 467, and PS2 variants 1 and 2 RNA messages, showed that the mean order of basal RNA message abundance at 0 time was ␤-actin ϾϾϾ PS1 Ͼ Ͼ basal COX-2 Ͼ PS2 (Fig. 1, A and B). Both the observed and predicted RNA message stability analyses indicated that COX-2 and PS1 RNA messages have relatively short half-lives, on the order of 3 and 4 -5 h, respectively; PS2 RNA message half-life was Ͼ10 h, and, in agreement with previous reports, brain ␤-actin RNA half-life was Ͼ12 h (4, 24).

COX-2 and PS1 RNA Exhibited Coordinate Down-regulation during Human Brain Development and Aging and Coordinate
Up-regulation in AD Neocortex and Hippocampal CA1-Using RNA message and cDNA libraries generated from embryonic, adult, and aged normal human brain controls (4,9) and ATCC collections of control human brain-specific cDNA libraries (Ex-press-Check, ATCC, Rockville, MD), we report a developmental down-regulation of PS1 and basal COX-2 RNA levels during human brain development, from embryonic week 7 (E7) to at least 75 years ( Fig. 2A; n ϭ 5; correlation of COX-2 and PS1 over eight time points, r 2 ϭ 0.93; p Ͻ 0.01). Developmental down-regulation of PS1, but not of basal COX-2 gene expression, has been reported previously (36). To evaluate their potential contribution to pro-inflammatory signaling, COX-2 and PS1 RNA message levels were next quantitated by Northern dot blot analysis and RT-PCR in AD and age-matched control superior temporal lobe neocortex (n ϭ 27) and hippocampal CA1 (n ϭ 8). In AD, COX-2 and PS1 RNA message levels were coordinately increased by 1.8-and 1.3-fold (p Ͻ 0.03) and 3.5and 3.1-fold (p Ͻ 0.01) in neocortex and CA1, respectively, over age-matched controls (Fig. 2B).

FIG. 3. TF-DNA binding structure of the human-specific COX-2, PS1, and PS2 gene promoters, indicating major TF-DNA consensus sequences as predicted by DNASIS and TRANS-FAC databases.
There is a noticeable homology between the nucleotide structure in COX-2 and PS1 (Ϫ1000 to ϩ1 bp) 5Ј regulatory regions (49.5% and 50.5% CϩG, respectively) and enrichment for a specific set of pro-inflammatory TFs (AP1, NF-Bp50/p65, STAT1␣, and HIF-1␣) in their cis transactivation domains. Some TF targets have been skewed or omitted due to space constraints.

FIG. 4. Experimental design. HN cells were grown under normoxic
conditions to ϳ70% confluence and were then exposed for 3 h to conditions of normoxia or hypoxia. Cells were treated ϮA␤42, ϮA␤42s, or ϮIL-1␤ at the times indicated. After 3 h of normoxia or hypoxia, nuclear TFs, total RNA, poly(A) ϩ RNA, and protein were isolated and analyzed. When used, PBN was preincubated with HN cells for 1 h before 0 h time. *, sampling times for AP1-, HIF-1␣-, NF-Bp50/p65-, and STAT1␣-DNA binding in untreated HN cells exposed to hypoxia (Fig.  7A). A␤42s is a reverse-order 42-amino acid peptide used as an inactive control.
ing hypoxia, were strongly attenuated under these conditions in the presence of the electron spin-trap and free radical scavenger PBN (Fig. 7B). These data suggest the involvement of free radicals in the COX-2 and PS1 gene activation mechanism at the levels of AP1-, NF-Bp50/p65-, STAT1␣-, and HIF-1␣-DNA binding (Fig. 7C), resulting, after PBN treatment (Fig. 4), in suppressed levels of COX-2 and PS1 RNA message (Fig. 7D) and COX-2 and PS1 protein abundance (Fig. 7E). As in previous experiments, [IL-1␤ϩA␤42s] was not able to elicit these changes, and there was no significant induction in either COX-1 RNA message or protein under these experimental conditions (Fig. 7, D and E).
Inflammatory Gene Expression in [IL-1␤ϩA␤42]ϩHypoxiatreated HN Cells Assessed by DNA Arrays-High density human DNA array panels containing 1184 gene targets were next used to assess thematic patterns of gene expression in [IL-1␤ϩA␤42]ϩhypoxia-treated HN cells (Fig. 8).  Fig. 8, A and B). Of the strongest signal differences detected, [IL-1␤ϩA␤42]ϩhypoxia-treated HN cells showed statistically significant (p Յ 0.05), Ն2.7-fold increases in five potentially pro-inflammatory transcripts, including COX-2 (but not COX-1), PS1, NF-Bp50/p105, IL-1␤ precursor, and cPLA 2 (Fig. 8C). ϩHypoxia-treated HN Cells-We again searched the promoters (Ϫ1200 to ϩ100 bp) of a family of pro-inflammatory genes that are up-regulated in [IL-1␤ϩA␤42]ϩhypoxiatreated HN cells and found them to contain an overrepresentation of AP1-, HIF-1␣-, NF-Bp50/p65-, and STAT1␣-DNA binding sites (Fig. 9A). Although other TFs may be involved, we observed a strong correlation (r 2 ϭ 0.81; p Ͻ 0.05) between the number of these four pro-inflammatory TF-DNA binding sites delimited to the Ϫ1200 to ϩ1 regions of these promoters and the degree to which each gene was up-regulated. Independent RT-PCR analysis again suggested that these five genes represent members of a specifically targeted pro-inflammatory brain gene family in [IL-1␤ϩA␤42]ϩhypoxia-triggered HN cells. A comparison of RT-PCR and DNA array data is shown in Fig. 9B.
Episodic Hypoxia: Implications for Alzheimer's Disease-Although AD neuropathology demonstrates progressive A␤ deposition, microglial and cytokine signaling activation, and neuronal and synaptic loss, the events initiating AD remain unclear. Environmental factors, gene mutations, cerebrovascular disease, and ischemia against a background of aging have emerged as risk factors for AD (59 -65). When compared with transformed human neural cell lines and brain cells from other species, non-transformed HN cells in primary culture in this study were exceptionally sensitive to gene expression triggered by IL-1␤, A␤42, and hypoxia, and delayed cell death was apparent after Ն6 h of hypoxia in the presence of [IL-1␤ϩA␤42] (Figs. 5-7). These results also suggest synergism in the activation of AP1-, STAT-1␣-, and in particular HIF-1␣-and NF-Bp50/p65-DNA binding and COX-2 and PS1 gene expression by [IL-1␤ϩA␤42] after hypoxia. These data further suggest that episodic hypoxia during aging may drive COX-2, PS1, and related genes by way of complementary, interdependent TF signaling that may contribute, cumulatively, to inflammatory responses and amyloidogenic neuropathology.