The Zebrafish Equivalent of Alzheimer’s Disease-Associated PRESENILIN Isoform PS2V Regulates Inflammatory and Other Responses to Hypoxic Stress

Abstract

Dominant mutations in the PRESENILIN genes PSEN1 and PSEN2 cause familial Alzheimer’s disease (fAD) that usually shows onset before 65 years of age. In contrast, genetic variation at the PSEN1 and PSEN2 loci does not appear to contribute to risk for the sporadic, late onset form of the disease (sAD), leading to doubts that these genes play a role in the majority of AD cases. However, a truncated isoform of PSEN2, PS2V, is upregulated in sAD brains and is induced by hypoxia and high cholesterol intake. PS2V can increase γ-secretase activity and suppress the unfolded protein response (UPR), but detailed analysis of its function has been hindered by lack of a suitable, genetically manipulable animal model since mice and rats lack this PRESENILIN isoform. We recently showed that zebrafish possess an isoform, PS1IV, that is cognate to human PS2V. Using an antisense morpholino oligonucleotide, we can block specifically the induction of PS1IV that normally occurs under hypoxia. Here, we exploit this ability to identify gene regulatory networks that are modulated by PS1IV. When PS1IV is absent under hypoxia-like conditions, we observe changes in expression of genes controlling inflammation (particularly sAD-associated IL1B and CCR5), vascular development, the UPR, protein synthesis, calcium homeostasis, catecholamine biosynthesis, TOR signaling, and cell proliferation. Our results imply an important role for PS2V in sAD as a component of a pathological mechanism that includes hypoxia/oxidative stress and support investigation of the role of PS2V in other diseases, including schizophrenia, when these are implicated in the pathology.

Keywords

Gene regulatory networks neurodegenerative diseases transcriptome profiling zebrafish

INTRODUCTION

Alzheimer’s disease (AD) is a neurodegenerative disorder with several pathological characteristics, including severe neuronal loss, glial proliferation, extracellular deposition of senile plaques composed of amyloid-β (Aβ), and deposition of intracellular neurofibrillary tangles (reviewed in [1, 2]). The overwhelming majority of AD cases are sporadic (sAD), with more than 15 million people affected worldwide. However, the cause(s) of sAD is poorly understood and therapeutic approaches are not yet well developed.

The PRESENILIN1 (PSEN1) and PRESENILIN2 (PSEN2) genes are associated with the relatively rare, dominantly inherited form of AD, familial AD (fAD, reviewed in [3]). These genes encode transmembrane proteins embedded in the membranes of the endoplasmic reticulum (ER), Golgi, plasma membrane, endosomes, and lysosomes [4]. Intriguingly, although the PRESENILIN proteins provide the γ-secretase enzyme activity that is critical for production of Aβ peptide (regarded by many as central to pathological processes causing AD [5]), no genetic variation at the PSEN1 or PSEN2 loci has yet been discovered to contribute to the risk for developing sAD [6]. This, combined with the failure of γ-secretase inhibitors as AD-therapeutics and other observations inconsistent with the idea that Aβ is a pathogenic agent, has encouraged a search for other changes in PRESENILIN gene function that may contribute to AD pathogenesis (reviewed in [7]).

One PRESENILIN-related difference observed between normal and sAD brains is expression of a truncated isoform of PSEN2 protein, PS2V [8]. PS2V is produced by alternative splicing of PSEN2 transcripts that is induced specifically by hypoxia (see below). PS2V protein accumulates in intracellular inclusion bodies termed PS2V bodies. These are observed in pyramidal cells of the cerebral cortex and in the hippocampus of sporadic AD patients during early stages [9]. In vitro experiments indicate that PS2V protein impairs the unfolded protein response thereby sensitizing cells to various ER stresses and significantly stimulating the production of Aβ₄₀ and Aβ₄₂ peptides [9]. PS2V also changes the conformation of the protein tau, which is a major component of neurofibrillary tangles [10]. Interestingly, PS2V was seen to be elevated in brains from people with schizophrenia to an even greater extent than in sporadic AD brains (although the sample size was small [11]). This suggests that PS2V may play an important role in the pathogenesis of AD and other diseases in which hypoxia and oxidative stress are implicated.

Using human neuoblastoma SK-N-SH cells, Sato et al. showed that the exclusion of exon 5 from PSEN2 transcripts to form PS2V mRNA is induced by hypoxia but not by a number of other cellular stressors [9]. The oxidative stress produced by hypoxia induces expression of the protein HIGH MOBILITY GROUP AT-HOOK 1 isoform a (HMGA1a) in neurons and this binds to a sequence in exon 5 of PSEN2 transcripts. This interferes with spliceosome complex function causing exon 4 to be ligated to exon 6 and generating a frameshift that leads to a premature termination codon in exon 6 sequence [8]. Remarkably, although PS2V lacks critical amino acid residues required for γ-secretase catalysis, it nevertheless can increase γ-secretase activity and production of Aβ peptide from the amyloid-β protein precursor (AβPP). To do this, PS2V probably requires the presence of full-length PSEN protein [12]. Interestingly, the pathogenic K115Efx10 mutation of PSEN2 (causing fAD) truncates the open reading frame in a very similar position to the frameshift that generates PS2V [13, 14] and our previous analyses have shown that a peptide generated from K115Efx10 transcripts would have similar effects to PS2V on γ-secretase activity and the UPR [14]. This strongly supports that PS2V plays a role in development of sAD brain pathology.

We recently documented that PS2V is not a uniquely human isoform but is widely conserved among most mammals although not in mice or rats [15]. Indeed, we showed that zebrafish possess a PS2V-cognate isoform, PS1IV, regulated by a conserved mechanism (i.e., involving a zebrafish orthologue of HMGA1a) but generated from the fish’s PSEN1 orthologous gene (psen1) rather than its PSEN2 orthologue (psen2). Thus, human PS2V and zebrafish PS1IV appear to share a common origin in an hypoxia-induced isoform of the ancestral PSEN gene that duplicated to form the contemporary vertebrate PSEN1 and PSEN2 genes [14]. Like human PS2V, zebrafish PS1V can stimulate γ-secretase activity and suppress the unfolded protein response [14]. However, aside from these activities, we have no idea how this PRESENILIN isoform modulates cellular responses to hypoxia and oxidative stress.

We recently demonstrated an ability to inhibit, specifically, the induction by hypoxia of the PS1IV isoform in zebrafish embryos. When embryos are placed in a sodium azide solution to mimic hypoxia, expression of Hmga1a is induced [16] but binding of this protein specifically to nascent psen1 transcripts can be inhibited by the presence of a morpholino antisense oligonucleotide that binds competitively to the Hmga1a binding site. This prevents the alternative splicing event that forms PS1IV mRNA under hypoxia while nevertheless allowing formation of translatable mRNA encoding full-length Psen1 protein [14]. Since so little is known regarding the function of PS1IV or its cognate human isoform PS2V, we exploited this ability to investigate changes in the zebrafish transcriptome caused by absence of PS1IV during hypoxia. We used a comprehensive microarray detecting zebrafish mRNAs to observe transcriptome-level changes in gene expression. These data were then subjected to various forms of genetic network analysis to reveal greater biological meaning from the observed expression changes. The results provide evidence supporting modulation by PS1IV of numerous processes that, in humans, are implicated in AD pathology and also in other diseases. In particular, PS1IV is required for upregulation of the IL1B gene that is a recognized AD risk locus and is central to neuroinflammation. Our results support that PS2V plays an important role in a pathological process in sAD possibly involving chronic hypoxia.

MATERIALS AND METHODS

Ethics statement

All animal experiments including experiments on genetically modified organisms were conducted under the auspices of the University of Adelaide’s Animal Ethics Committee and Institutional Biosafety Committee.

Zebrafish husbandry, morpholino injection, and exposure to NaN₃

Danio rerio were bred and maintained at 28°C on a 14 h light/10 h dark cycle. Embryos were collected from natural mating of the Tübingen strain (Tu), grown in embryo medium (E3), and staged [17]. MoHmga1aBindBlock morpholino was synthesized by GeneTools LLC (Corvallis, OR, USA) and injected at 0.5 mM (diluted to this concentration with negative control morpholino MoCont as previously described [18]) at the one-cell stage.

Sodium azide, NaN₃, has previously been used for mimicry of hypoxia [19, 20] including in zebrafish [16, 21]. Exposure of embryos to NaN₃ (Sigma-Aldrich CHEMIE Gmbh, Steinheim, Germany) was performed at 100 μM from 6 h post fertilization (hpf) until embryos reached the developmental stage equivalent to that attained under normoxia at 48 hpf at 28.5°C. 30 embryos were used for each treatment and replicate.

Oligonucleotides

The sequences of the microRNA primers and qPCR primers used in this study are given in Supplementary Material 1. The sequence of morpholino oligonucleotide MoHmga1aBindBlock was 5’- CTTGTAGAGCACCACCAGCACCAGG-3’.

Microarray experiment, experimental design, and statistical analysis

For microarray analysis, total RNA was extracted from batches of injected embryos using the RNeasy Mini Kit according to the manufacturer’s specifications (Qiagen GmbH, Hilden, Germany). For each sample, the RNA concentration was determined using a NanoVuetrademark UV–vis spectrophotometer (GE Healthcare Life Sciences, Fairfield, USA). RNA integrity and quality were then estimated using an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA) and the RNA integrity number (RIN) index was calculated for each sample. Only RNAs with a RIN number >7.0 were processed further.

To evaluate genome-wide changes in gene transcription related to the presence or absence of PS1IV under mimicry of hypoxia, we performed hybridization analysis using the Zebrafish Gene 1.0 ST Array (Affymetrix Inc. Santa Clara, CA). Briefly, 300 ng of total RNA derived from embryos was converted to amplified sense strand cDNA using the Ambion WT Expression Kit (Life Technologies, Carlsbad, CA, USA). The resulting sense cDNAs were fragmented and Biotin end labeled using the Affymetrix Genechip WT Terminal Labeling Kit prior to hybridization to the arrays at 45°C for 16 h. Finally, the signal intensity of the chip was scanned using a GeneChipR Scanner 3000TG and analyzed using Expression Console software (http://www.Affymetrix.com). CEL files were imported and intensities adjusted by RMA background correction and quantile normalization [22 –24]. Each treatment had four biological replicates. The CEL files that did not pass quality control were removed based on their quality assessments obtained by Expression Console reports.

The final list of genes significantly (p≤0.05) over-expressed or under-expressed as a result of blocking PS1IV formation under hypoxia was generated using the following formula:

Impact of PS1IV splicing blockage under hypoxia = [Comparison of (MoHmga1aBindBlock) vs (MoCont) under chemical mimicry of Hypoxia] – [Comparison of (MoHmga1aBindBlock) vs (MoCont) under Normoxia] – [Comparison of (MoCont) vs (Uninjected) under chemical mimicry of Hypoxia] – [Comparison of (MoHmga1aBindBlock) under chemical mimicry of Hypoxia vs (MoHmga1aBindBlock) under Normoxia].

A Bayesian t-test using the FlexArray package (McGill University, Canada) was employed to find significantly over-expressed and under-expressed genes (p≤0.05) for each of the dual comparisons in the above formula as previously described. Removal of genes with significant alteration in the control comparisons increases the reliability of discovery of genes influenced specifically by PS1IV expression. We note that the formula used excludes any genes that show altered expression under mimicry of hypoxia relative to normoxia independent of PS1IV action. Expression of some such genes may, nevertheless, be modulated by PS1IV.

Gene ontology analysis was carried out using the “Comparative GO” web application [25, 26].

Statistics-based subnetwork discovery and regulatory network analysis after loss of PS1IV under mimicry of hypoxia

To understand the consequences of loss of PS1IV under hypoxia, we used the integrative approach of “gene set enrichment” (based on Fisher’s exact test) [27] with union selected subnetworks [28]). This reveals significantly altered subnetworks as well as underlying regulatory mechanisms controlling alteration of gene expression patterns in an integrative network.

Subnetwork discovery

For determination of statistically altered subnetworks after loss of PS1IV splicing under mimicry of hypoxia, the following steps were performed:

1) An enriched mammalian database of gene/ protein/small RNA interactions, ResNet, was implemented in the Pathway Studio 10 package (Elsevier). ResNet is a database of biological relations, ontologies and subnetworks, compiled by Ariadne (Elsevier) for mammalian research [20, 21]. One of the major advantages of this database is inclusion of information on the subcellular localization of gene products, which provides the opportunity of evaluating the contribution of different cell compartments to biological networks.

A range of interaction sets are deposited in the ResNet database such as regulation, expression, promoter binding, molecular transport, protein modification, binding, molecular synthesis, chemical reaction, direct regulation, miRNA effect, protein complex, and small molecule function. Furthermore, we updated the network database Pathway Studio 10 using Medscan language programming via text mining [29] with the keywords of names of over-expressed and under-expressed genes after PS1IV loss of expression under mimicry of hypoxia, AD, and hypoxia to enrich the deposited networks in our reference database. MedScan is an NLP (Natural Language Processing) application for reading scientific literature and extracting protein-centric biological relations [29].

2) Using as input the set of genes showing significantly altered gene expression after loss of PS1IV, different possible subnetworks to which these genes may contribute were called from the updated database of the Pathway Studio package.

3) The gene set enrichment concept was used to distinguish and select important subnetworks. Gene set enrichment measures the enrichment of different subnetworks by an imported list of genes/miRNAs and highlights the statistically enriched subnetworks at p≤0.05, based on various statistical tests and particularly Fisher’s exact test [27]. As an example, if a particular subnetwork has 15 genes and 14 out of 15 genes of this subnetwork are present in the list of those displaying altered expression after loss of PS1IV, then this subnetwork has a high probability of existence according to Fisher’s exact test.

To identify subnetworks from genes with expression altered in sAD brains compared to normal brains, we used the microarray data presented by Antonell et al. [30]. The 50 most statistically significant differentially expressed genes among the late-onset AD brains analyzed by Antonell et al. were subjected to subnetwork discovery at p < 0.05.

Integrative regulatory network construction

For construction of an integrative regulatory network, various algorithms such as “add neighbors” (with varying ranges of directly and indirectly interacting neighbors – “expansions” of 1, 2 or 3), “add direct interaction”, “add shortest path”, “add common targets”, “add common regulators”, and “union selected subnetworks” [28 , 31–34] were compared.

The add neighbors algorithm adds a new interacting element (direct or indirect interaction) to an imported gene entry if that new element is within a defined interacting range. Here, expansions of 1, 2, and 3 were tested. Add direct interaction only adds direct interactions to the imported gene list and produces the smallest networks. The next algorithm was the add shortest path algorithm. Two elements in the network may interact with each other via many different routes. The add shortest path algorithm selects the shortest route as that is the most probable to occur. The add common targets algorithm searches for interactions that are affecting a shared target. Add common regulators looks for common regulators of imported genes such as transcription factors. Union selected subnetworks is a reliable algorithm which attempts to join together statistically significant subnetworks (from the previous stage of subnetwork discovery) and finds the underlying relationships between subnetworks based on the shared elements. Due to the ability of transcription factors, ligands, and microRNAs, to regulate multiple systems, it can be expected that these play central roles in joining various subnetworks. In fact, union selected subnetworks can be seen as a crosstalk discovery tool.

Topological analysis of regulatory networks

Topological analyses of the regulatory networks of over-expressed, under-expressed as well as the combined set of over- and under-expressed genes (constructed using union selected subnetworks) were performed using the cyto-Hubba package [35] in Cytoscape software [36]. Centrality indices including Degree, clustering coefficient, Edge Percolated Component (EPC), Maximum Neighborhood Component (MNC), Density of Maximum Neighborhood Component (DMNC), Maximal Clique Centrality (MCC) and centralities based on shortest paths, such as Bottleneck (BN), EcCentricity, Closeness, Radiality, Betweenness, and Stress were used to identify the highest-ranked genes (hubs) in the structure of the regulatory networks using the cyto-Hubba plugin [35, 37]. For details regarding the employed indices, please see the cyto-Hubba URL: http://hub.iis.sinica.edu.tw/cytoHubba/supplementary/index.htm. The genes receiving the highest scores in the above mentioned indices were selected as the highest-ranked genes (hubs) in the regulatory networks. The 10 highest-ranked genes according to Bottleneck score were visualized in Cytoscape for the regulatory networks of over-expressed and under-expressed genes, and for the combined set of over- and under-expressed genes.

Validation of network nodes by quantitative real-time PCR (qPCR)

The relative standard curve method for quantification was used to determine the expression of experimental samples compared to a basis sample. For experimental samples, target quantity was determined from the standard curve and then compared to the basis sample to determine fold changes in expression. Gene specific primers were designed for amplification of target cDNA and the cDNA from the ubiquitously expressed control gene eef1a1l1. The reaction mixture consisted of 50 ng/ μl of cDNA, 18 μM of forward and reverse primers and Power SYBR green master mix PCR solution (Applied Biosystems Foster City, CA, USA).

To generate the standard curve, cDNA was serially diluted (100 ng, 50 ng, 25 ng, 12.5 ng). Each sample and standard curve reaction was performed in triplicate for the control gene and experimental genes. Amplification conditions were 2 min at 50°C followed by 10 min at 95°C and then 40–45 cycles of 15 s at 95°C and 1 min at 60°C. Amplification was performed on an ABI 7000 Sequence Detection System (Applied Biosystems) using 96 well plates. Cycle thresholds obtained from each triplicate were averaged and normalized against the expression of eef1a1l1. Each experimental sample was then compared to the basis sample to determine fold changes of expression. For each gene, 3 biological replicates and 3 technical replicates were used in qPCR.

RESULTS

In the work described below, we used microarray analysis to assess changes in the expression of genes when PS1IV expression is blocked during hypoxia-like conditions in zebrafish embryos. However, since numerous human genes have “co-orthologues” in zebrafish (due to a whole genome duplication that occurred early in the teleost evolutionary lineage [38]), and since the GO and Pathway Studio databases contain predominantly mammalian genetic information, we refer to genes using their human orthologue names. Raw microarray expression data gene designations are given in the file “Supplementary Material 2 and 3”.

Genes with altered expression after failure to form PS1IV under hypoxia

PS2V is a truncated isoform of PSEN2 that shows elevated expression in the brains of people with sporadic AD. However, the cellular role of PS2V cannot be investigated using mouse or rat models since the rapid evolution of some of their AD-related genes has resulted in loss of their ability to induce this isoform [15]. In contrast, zebrafish produces a PS2V-cognate isoform named PS1IV [38]. We can block induction of PS1IV in embryos subjected to hypoxia by injection of the morpholino oligonucleotide MoHmga1aBindBlock [14]. Since so little is known about PS2V function, we exploited this phenomenon to investigate the effects on cellular function of loss of PS1IV under hypoxia using gene network analysis of relative expression data from microarrays.

Zebrafish embryos were injected at the 1-cell stage with MoHmga1aBindBlock or the negative control morpholino MoCont. They were then incubated in 100 μM sodium azide, NaN₃ to mimic the effects of hypoxia (a treatment hereafter referred to simply as “hypoxia”) from 36 hpf to 48 hpf before extraction of mRNA and analysis of relative gene expression on the Affymetrix Zebrafish Gene 1.0 ST Array. A number of control microarray comparisons were also made to identify genes for which expression is affected by the act of injection or by the non-specific effects of the presence of morpholino oligonucleotides or by hypoxia independent of PS1IV (Fig. 1). Any genes for which expression was significantly altered by these effects were excluded from subsequent analyses (see below).

Bayesian t-tests produced lists of genes with significantly increased or decreased transcript levels (“expression”) in our microarray comparisons. We describe these genes below as either “over-” or “under-expressed” in response to loss of PS1IV under hypoxia. The final list of genes with significant (p≤0.05) over- or under-expression was generated after removal of any genes showing significantly altered expression in the controls (see Fig. 1, Table 1, and the file “Supplementary Material 2 and 3”).

Among the genes showing over-expression under hypoxia without PS1IV (i.e., genes that would normally be repressed by the formation of PS1IV) were numerous genes involved in modulation of immune responses (HRH2, IL17RA, CD22, ACY1, IRF4, XCL2, etc.), in regulation of TOR signaling (RICTOR, PKD1), in regulation of blood pressure and formation of vascular structures (e.g., AGTR1 and ANGPTL2), and in cell proliferation (PIM3, ACY1, OLFM4) (File “Supplementary Material 2 and 3”). This set of genes also included two genes known to be involved in neurological conditions, GAK (associated with Parkinson’s disease) and CAMTA1 (associated with cerebellar ataxia, nonprogressive, with mental retardation, File “Supplementary Material 2 and 3”). Two non-coding RNAs were also seen in the list of genes over-expressed in the absence of PS1IV, si:dkey-167i21.2 and si:ch73-110p20.1.

A greater number of genes were identified as showing significant under-expression under hypoxia without PS1IV (i.e., genes that are normally dependent on PS1IV for stable or increased expression under hypoxia). These included genes encoding transcription factors or proteins involved in chromatin remodeling (e.g., NKX3-2, SRCAP, GBX2, TCF25, SMARCA4, POU4F2), genes involved in ER function (CALR), intracellular signaling (OCRL, AIDA, PLEKHA4, PLCB1, MAPKAPK5) and a considerable number of ribosomal proteins (see Table 1).

An overview of our genetic network analysis of transcriptome changes occurring due to failure to form the PS1IV isoform under hypoxia

Gene regulatory network analysis offers a more comprehensive level of understanding of biological mechanisms compared to analyses of individual (separate) gene expression changes.

Within cells, the transcripts and proteins encoded by genes can have many functionally interacting partners. “Subnetworks” are sets of genes with biologically important interactive relationships. The Pathway Studio 10 software package (based on the Fisher Exact Test) accepts a set of genes with significantly altered expression and evaluates the genes with respect to their contribution to all the possible subnetworks in its database. It then ranks these subnetworks in terms of statistical significance to create a list of “statistically significant subnetworks”. We used the Pathway Studio 10 software package to examine the sets of genes showing significantly increased or decreased transcript levels when formation of the PS1V isoform is blocked during hypoxia. Subnetworks with a statistical significance of p≤0.05 were then ranked.

Once ranked lists of statistically significant subnetworks had been determined, we then used the algorithm “union selected subnetworks” to attempt to join different subnetworks into larger aggregate networks (integrative networks). Networks were produced based on the set of genes with significantly increased transcript levels, the set of genes with significantly decreased transcript levels, and the set of all genes with either significantly increased or decreased transcript levels (see later). The union selected subnetworks algorithm looks for elements shared between subnetworks that could act as links between them. This procedure is valuable as a test of the veracity of the existence of the subnetworks identified within our experimental system since subnetworks that show greater numbers of shared elements are less likely to be spurious.

A number of other methods for network analysis were also applied to our sets of genes that are “over”- and “under”-expressed in the absence of PS1IV. These other methods are summarized in the Materials and Methods section. In general, these analyses confirmed the major features of the networks produced by union selected subnetworks analysis. Of all the methods used, union selected subnetworks proved to be the most efficient network prediction algorithm for combining the statistically significant subnetworks and revealing their regulatory components (transcription factors, microRNAs, and ligands). The veracity of the networks predicted by union selected subnetworks was confirmed by qPCR analysis of predicted changes of transcript levels for a number of genes identified as nodal within the networks (see the section “Validation of network nodes (hubs) by RT-qPCR” below and Fig. 2). In the description that follows, we have focused on the union selected subnetworks analysis since it is the method producing the most statistically rigorous output. (The union selected subnetworks algorithm has also been recently used in the discovery of transcriptomics-based regulatory networks of biological phenomena such as apoptosis and stem cell differentiation [32 , 39]).

Significant subnetworks formed by genes over-expressed after failure to form PS1IV under hypoxia

Genes over-expressed after failure to form PS1IV under hypoxia were examined for their contribution to formation of statistically significant subnetworks (p≤0.05) based on the “gene set enrichment” approach [27] using the Pathway Studio 10 software package. The identified significant subnetworks and their corresponding p-values are presented in Table 2. The genetic relationships underlying each subnetwork and citations upon which these relationships are based are presented in Supplementary Material 4. The subnetworks are: Neighbors of GABARAP (Gamma-Aminobutyric Acid Receptor-Associated Protein), PDLIM5 (PDZ And LIM Domain 5), MIR20A (microRNA 20a), LPS (lipopolysaccharides), BTK (Bruton A gamma globulinemia), and NOTCH2 (notch2). These subnetworks and their interactions (relationships) are visualized in Fig. 3. The gene GABARAP encodes a ligand-gated chloride channel that mediates inhibitory neurotransmission and interacts with the cytoskeleton. GABARAP protein has a direct regulatory effect on the angiotensin receptor AGTR1 that regulates blood pressure and vascular development. GABARAP also mediates membrane expression of AGTR1 (Fig. 3A) [40]. Interestingly (and with relevance to AD), GABARAP is also involved in apoptosis and autophagy [41 –46]. As presented in Fig. 3B, PDLIM5 may be involved in the observed over-expression of AGTR1 and another gene, PKD1, after loss of PS1IV [47, 48]. PDLIM5 has been implicated in bipolar disorder, depression, and schizophrenia via regulation of dendritic spine morphogenesis in neurons [49, 50]. This is especially interesting when one considers that the human homologue of PS1IV, PS2V, is greatly increased in the brains of some cases of schizophrenia [11]. It has recently been suggested that PDLIM5 plays an important role in heart development and cardiomyocyte expansion [51]. Since mutations in both human PSEN1 and PSEN2 can cause Dilated Cardiomyopathy (DC) this suggests that the PRESENILIN genes together with PDLIM5, PKD, and AGTR1 may form a genetic network involved in cardiovascular disease.

One subnetwork is formed by NOTCH2 together with FOXE3. NOTCH genes encode transmembrane receptors that signal directly to the nucleus. When NOTCH receptors are cleaved by PSEN proteins within γ-secretase complexes, their intracellular domains move into the nucleus to act as transcription factors [52]. NOTCH receptors have important roles in maintenance of undifferentiated stem cell states, particularly under hypoxic conditions [53, 54]. During zebrafish embryo development, the gene foxe3 is expressed mainly in the developing lens of eyes that are relatively much larger than in humans. The foxe3 gene also shows restricted expression in the brain and in the hatching gland [52], a structure not found in mammals. This reminds us that our microarray analysis involves all the cells of a developing zebrafish at 48 hpf and not only those with relevance to the brain. This means that the networks we subsequently identify should not be interpreted as representing single overall regulatory networks within a single cell type. Some of the subnetworks will exist only within particular cell types and some apparent linkages between putative subnetworks may be artifactual.

A subnetwork illustrating the ability of microRNAs to coordinate regulation of multiple gene transcripts is formed by the apparent inhibitory effect of the microRNA MIR20A on PKD1, AGTR1, and CAMTA1 (Fig. 3C). Since these three genes are over-expressed in the absence of PS1IV, we would predict MIR20A to be downregulated in this circumstance. However, MIR20A expression is not measured by the Affymetrix Zebrafish Gene 1.0 ST Array that we have used. Another microRNA, MIR1-1, is implicated in a putative subnetwork regulating NOTCH2 and GAK (Fig. 3F). Decreased NOTCH signaling has been suggested to underlie the development of AD pathology [55] so the downregulation of NOTCH2 by PS1IV under hypoxia may be significant. Decreased expression of CYCLIN G-ASSOCIATED KINASE, GAK has been shown to inhibit NOTCH signaling and to increase neural cell death in zebrafish [56]. GAK has also been associated with Parkinson’s disease in genome wide association studies (GWAS) [57], consistent with its interaction with the Parkinson’s disease-associated protein α–synuclein [58]. Previously, we observed increased expression of cyclin G2 gene expression when zebrafish psen1 or psen2 activity was suppressed in embryos in a morpholino-based study [59]. We have also argued that insufficient delivery of oxygen to the brain may play an important role in AD pathogenesis [60] and, in this light, it is interesting that both the MIR1-1- and MIR20A-coordinated subnetworks described here involve genes important for vascular function and development (e.g., see [61, 62]). The analysis, summarized in Fig. 4, also illustrates that both these subnetworks are potentially linked through the transcription factor ESTROGEN RECEPTOR 2, ESR2. The ESR2 gene shows homology to the Hairy/enhancer of split (HES)-related genes that are typically directly activated by NOTCH signaling. Also, the estrogen molecule estradiol has been shown to up-regulate Agtr1 in rat heart [63].

Finally, lipopolysaccharides are known to have positive regulatory effects on all the receptors seen to be over-expressed after loss of PS1IV (AGTR1, HRH2, IL17RA, CD22, and NOTCH2) [64 –66]. Together these form a statistically significant subnetwork (Fig. 3D). Lipopolysaccharides commonly induce inflammatory responses [67] while hypoxia (that induces PS2V and PS1IV formation) is also pro-inflammatory. The fact that PS1IV upregulates expression of the pro-inflammatory cytokine IL1B and the chemokine receptor CCR5 (see later) suggests that the truncated PRESENILIN transcript isoforms induced by hypoxia act as important modulators of inflammatory responses.

Prediction of a regulatory network underlying genes over-expressed after failure to form PS1IV

Transcription factors and microRNAs commonly establish the crosstalk and coordination between different regulatory subnetworks since any one transcription factor or microRNA usually has the ability to regulate multiple gene transcripts. To detect potentially coordinated subnetworks, genes significantly over-expressed after failure to form PS1IV under hypoxia were subjected to various network construction algorithms including “add interacting neighbors (interacting range of 1, 2, and 3, as mentioned in Material and Methods)”, “add direct interaction”, “add shortest path”, “add common targets”, “add common regulators”, and “union selected subnetworks”.

Figure 4 shows the network generated when the union selected subnetworks algorithm was applied to the list of statistically significant subnetworks given in Table 2. Significant features of this network are described below. Also, Supplementary Material 5 demonstrates the underlying relationships of the network.

The transcription factor encoded by ATF2 (also known as CREBP1) regulates PKD1 and ANGPTL2 within this network. Loss of ATF2 has been shown to affect mouse Psen1 expression and to cause cranial motoneuron degeneration during brain development [68]. The ATF2 protein is involved in regulation of cellular responses to numerous stresses [69]. The role of MIR20A as a key subnetwork coordinator controlling expression of CAMTA1, AGTR1 and PKD1 is evident (see below and Fig. 4). The CAMTA1 and AGTR1 genes are described above but PKD1 encodes a protein that has been observed in astrocytes and has been shown to activate the JAK-STAT signaling pathway [70]. Another microRNA suggested to coordinate a subnetwork, MIR1-1 (see above) regulates NOTCH2 and MRC2 receptor expression and expression of GAK in the Golgi apparatus. In the network analysis, the number of interactions for any gene product can be used as an index of that gene’s regulatory importance. The structure of our network of genes over-expressed after failure to form PS1IV highlights the products of PKD1, AGTR1, ANGPTL2, and NOTCH2 as central regulatory elements since these each show large numbers of interactions.

The results of an alternative strategy for network assembly— the shortest path method— are presented in Supplementary Material 6 and emphasize the importance of the genes CAMTA1 and AGTR1 as coordinating regulatory nodes. Calmodulin has a considerable number of interactions within that network. Noticeably, CAMTA1 expression is under the control of many microRNAs, such as MIR20A and MIR34A emphasizing both its functional importance and the selective pressure for its tight regulatory control. This is not unexpected when one considers the important role that Ca^2 + signaling and the PRESENILIN proteins play in interactions between the ER and mitochondria [71, 72]. CAMTA1 encodes a Ca^2 +-regulated transcription factor and changes in CAMTA1 expression have been associated with early-onset nonprogressive cerebellar ataxia and mild mental retardation [73 –75]. The relationships within this network are presented in Supplementary Material 7.

Significant subnetworks formed by genes under-expressed after failure to form PS1IV under hypoxia

When formation of PS1IV is suppressed during hypoxia, a number of genes show decreased expression. These genes require PS1IV for normal or increased expression during chemical mimicry of hypoxia and are listed in Table 1. We used these genes to construct subnetworks using the Pathway Studio 10 software (as described above for those genes with increased expression in the absence of PS1IV)

Significant subnetworks generated using the under-expressed genes are shown in Table 3 and Fig. 5. Many of the subnetworks are related to apoptosis and immune response, including genes such as CASP5, NLRP12, IL1B, TRIM22, and CCR2. The underlying relationships of these subnetworks are presented in Supplementary Material 8. It is apparent that IL1B is shared in many subnetworks. The regulatory subnetwork with the greatest significance (smallest p-value) is controlled by the transcription factor MEF2D (myocyte enhancer factor 2D) (Fig. 5C). MEF2D is a stress-induced gene involved in cardiac muscle development as well as neuronal differentiation and survival. MEF2D plays diverse roles in cell growth, survival and apoptosis via the p38 MAPK signaling pathway. It has been suggested that MEF2D is a critical protein in the regulation of neuronal apoptosis and autophagy [76 –79]. Another important subnetwork coordinates around Ca^2 + ion concentration (Fig. 5D) again emphasizing the relationship between PRESENILIN gene function and this ion.

Prediction of a regulatory network coordinating genes under-expressed after failure to form PS1IV

A regulatory network for under-expressed genes was constructed based on the union selected subnetworks algorithm as discussed above. This network is presented in Fig. 6 and its underlying relationships are given in Supplementary Material 9. A variety of cellular organelles and structures respond to modulation by PS1IV including the mitochondrion, endoplasmic reticulum, Golgi apparatus, nucleolus, and plasma membrane emphasizing the functional significance of this PRESENILIN isoform (Fig. 6).

The following nodes are central in the generation of the regulatory network of under-expressed genes: (1) IL1B, (2) CCR5, (3) TH, (4) CALR, (5) SMARCA4, and (6) MEF2. According to the Gene Ontology classification of CCR5 and based on our ComparativeGO database [25, 26], this gene, CCR5 is involved in aging and promotion of apoptosis. CCR5 has previously been suggested to have a modulatory effect on the activated microglia seen in AD brains [80] and its expression is elevated in peripheral blood mononuclear cells [81] and in other cells of the immune system in AD patients [82]. In rats, CCR5 is expressed by microglia in the developing brain and predominantly in the endothelial cells of the adult brain [83]. It is highly up-regulated in the brain by hypoxia/ischemia supporting that hypoxia is an important element of AD pathology [83]. However, some researchers have reported that hypoxia can inhibit upregulation of CCR5 in mouse macrophages [84].

As can be seen in Fig. 6, TH and IL1B represent intensive nodes in the network. TH encodes the enzyme TYROSINE HYDROXYLASE that is rate limiting in the catecholamine biosynthetic pathway required for production of dopamine. Interestingly, hypoxia is known to regulate TH activity in selected areas of the brain [85]. It has recently been documented that disruption of TH activity in humans causes Segawa syndrome, an inherited form of infantile Parkinsonism [86, 87]. TH has also been seen to modulate the release of inflammatory cytokines so the observed coupling of the TH and IL1B nodes (below) in our network is not unexpected. Also intriguing is the existence of another gene associated with Parkinson’s disease as a node in our network, LRRK2. Mutations in LRRK2 are one of the most common causes of familial Parkinson’s disease. The protein product of LRRK2 binds to the product of the TUBULIN, BETA gene, TUBB, that is the most under-expressed gene when PS1IV is absent under hypoxia. LRRK2 has also been seen to stimulate protein expression by acting on Argonaut proteins to repress microRNA action [88].

Cytokines are signaling compounds and major mediators of immune responses which are involved in stress responses, cell differentiation, and apoptosis. The IL1B protein is a member of the interleukin 1 cytokine family which is proteolytically processed to its active form by CASPASE 1 (CASP1/ICE). The importance of IL1B in sporadic AD is demonstrated by identification of polymorphisms at this gene locus associated with increased AD risk [89]. IL1B mRNA is highly significantly under-expressed in the absence of PS1IV and this gene forms a major node in our genetic network. Thus PS1IV contributes significantly to induction of IL1B expression under hypoxia. In our network, IL1B communicates with CCR5 the product of which has important roles in mediation of neuroinflammatory responses to excitotoxicity (see above and [90]). Our predicted network also shows the transcription factor encoded by STAT4 as increasing the expression of CCR5 and IL1B. STAT4 protein is well recognized as playing a stimulatory role in inflammation including in the brain [91].

The family of MEF transcription factors figures prominently in our network of under-expressed genes. These transcription factors control neuronal survival, apoptosis, and autophagy [78, 79]. We also see involvement of the gene encoding CREATINE KINASE, MITOCHONDRIAL 1B, (CKMT1B), a protein of the inner membrane of mitochondria involved in lipid transfer between membranes. The gene CALR, that encodes a stress response protein binding Ca^2 + ions in the ER and that is also found in the nucleus possibly acting as a transcription factor [92] is an important node in this network. Interestingly, in the ER, CALR binds to PROGRANULIN (a protein implicated in frontotemporal dementia), where it forms a complex together with protein disulfide isomerases [93]. The mitochondrial associated membranes (MAM) are foci for oxidative protein folding (disulphide bond formation) and are a major subcellular location for PRESENILIN protein expression [94]. Changes in PRESENILIN expression affect the lipid constitution of the MAM that is a site of oxygen use and that produces high levels of reactive oxygen species. Reactive oxygen species induce expression of PS2V via upregulation of HMGA1a to alter PSEN2 transcript splicing [16]. Numerous mutations in PSEN1 are linked to frontotemporal dementia (usually in association with AD). Thus, our network analysis supports involvement of the PSEN proteins in frontotemporal dementia via regulatory interactions with PROGRANULIN in the MAM under hypoxic conditions.

SMARCA4 is another node in the network (Fig. 6). SMARCA4 encodes a calcium signaling related protein with helicase and ATPase activities that can regulate the expression of genes by altering their chromatin structure [95]. Interestingly, significant down regulation of SMARCA4 has been reported in severe AD [96].

MAPKAPK5, encoding a protein of the serine/threonine kinase family, governs another node in the network. This kinase is expressed in response to cellular stress and proinflammatory cytokines [97, 98]. As a tumor suppressor protein, the major functions of MAPKAPK5 are activation of apoptosis and mediation of senescence [99]. Interestingly, MAPKAPK5 is also involved in amyloidosis and yellow fever diseases [100]. This supports the idea that PS2V might be important for the accumulation of amyloid in AD (e.g., through its suppression of the UPR, [9, 14]).

In our network of under-expressed genes, the microRNAs MIR137 and MIR26A1 potentially negatively regulate the nodal gene, PHOSPHOLIPASE C, BETA-1, (PLCB1), that encodes a key protein regulating signal transduction pathways involving IP₃ and diacylglycerol. Deletions of PLCB1 have been observed in the brains of people with schizophrenia [101] which may be significant since PS2V has also been observed to be greatly elevated in schizophrenia brains [11]. This supports that neural hypoxia may be an important environmental influence in schizophrenia [102] via PS2V.

Finally, a considerable number of genes encoding ribosomal proteins are under-expressed in the absence of PS1IV, for example RPL9, RPS15A, RPL28, RPL26, and RPS19. The stimulation of expression of these proteins by PS1IV under hypoxia appears consistent with the action of this truncated PRESENILIN protein and its cognate PS2V isoform to repress the UPR [103]. The UPR usually acts to suppress protein synthesis when cellular stresses (e.g., lack of oxygen for oxidative protein folding) lead to increased levels of unfolded proteins [103]. As mentioned above, the stimulation of LRRK expression by PS1IV is also consistent with its opposition to the UPR since this should increase protein synthesis. Notably, among the genes normally suppressed by PS1IV expression, the gene with most significantly altered expression is HSPA8 encoding the chaperone HEAT-SHOCK 70-KD PROTEIN 8. Upregulation of chaperone protein expression is an important element of the UPR. However, the action of PS1IV on other UPR actors is less easy to explain. PS1V stimulates CALR expression (see above), the product of which is involved in the UPR and is normally upregulated by ER stress [104]. The CALRETICULIN protein can form complexes with the Ca^2 + pump ATPase, Ca(2+)-TRANSPORTING, FAST-TWITCH 1 (ATP2A1, also known as SERCA) and disulphide isomerases required for oxidative protein folding [105]. PRESENILIN1 holoprotein binds ATP2A1 and is, itself, upregulated by unfolded protein stress [106]. Indeed, the PRESENILIN1 holoprotein has been observed to act as a chaperone [107, 108].

Topological analysis of regulatory networks

To identify those genes and factors central to the structure of the regulatory networks, Bottleneck, DMNC, MNC, and MCC centrality indices were calculated. The results are presented in Supplementary Material 10 and the genes and factors are ranked based on overall scores of importance in various indices. Also, the 10 highest-ranked genes and factors according to Bottleneck are presented in Fig. 7.

AGTR1, NOTCH2, LPS, are PKD1 are central hubs in the network structure of genes over-expressed after failure to form PS1IV under hypoxia (Supplementary Material 10). MIR20A received a high rank in network topological indices (Supplementary Material 10, Fig. 7).

The Inflammatory response genes IL1B and CCR5 are central in the regulatory network of genes under-expressed after failure to form PS1IV under mimicry of hypoxia (Supplementary Material 10). Also, STAT4, a transcription factor involved in stimulation of brain inflammation [91], forms a hub in this network (Supplementary Material 10, Fig. 7).

The raw expression value within different biological replicates of the treatments in this study is presented in Supplementary Material 11. Low standard deviations within different biological treatments show that gene expression is consistent across the biological replicates of each treatment.

In the regulatory network constructed from the combined set of over- and under-expressed genes after loss of PS1IV, the highest ranked genes and factors (hubs) were: IL1B, TH, LPS, CCR5, AGTR1, HSPA8, PKD1, CALR, and MAPKAPK5. Ca²⁺ was also identified as a hub. Overall, the topological analyses support the importance of these genes/factors in the function exercised by PS1IV.

A global regulatory network of both over- and under-expressed genes after failure to induce PS1IV during hypoxia

The genes observed to be over- or under-expressed in our study were detected after a only one form of experimental manipulation, i.e., blockage of PS1IV expression under hypoxia. Therefore, we can expect both the over- and under-expressed genes to show regulatory interactions within a global network. To examine this, the combined set of over- and under-expressed genes was subjected to subnetwork discovery using the “gene set enrichment” approach followed by integrative network construction using the union selected subnetworks algorithm. Supplementary Material 12 shows the subnetworks enriched by this global gene set. The network underlying these genes regulated by PS1IV is presented in Supplementary Material 13 and its constructing relationships in Supplementary Material 14.

The combined, global network does not reveal very significant new nodes but does highlight the intensity of the nodal roles played by IL1B, PLCB1, NOTCH2, AGTR1, SMARCA4, CALR, PKD1, and HSPA8. The regulation by PS1IV of processes in the ER, mitochondrion, nucleus and on the plasma membrane is evident. The contribution of the ER and mitochondrial dysfunction to induction and progression of AD has been discussed recently [107 , 110]. Our results support that PS2V may play a significant role in this.

Validation of network nodes (hubs) by RT-qPCR

Finally, to confirm the validity of our constructed networks, RT-qPCR was performed for a subset of those genes located at network nodes (Fig. 2). The results showed high correspondence between the microarray and RT-qPCR results. CAMTA1 showed more than a 2-fold increase in mRNA level in both microarray and qPCR analyses. On the other hand, IL1B was down regulated by more than 2-fold in both analyses. Both methods of analysis showed PKD1 to be significantly over-expressed after inhibition of PS1IV formation under hypoxia while MAPKAPK5 and SMARCA4 were significantly under-expressed.

Several factors have inhibited analysis of the correspondence between changes in mRNA expression and consequent protein expression changes. There is a dearth of commercially available antibodies that detect zebrafish proteins and proteomic analyses of zebrafish embryos can be problematic due to interference from the high levels of yolk vitellogenins present at these stages [111]. Hopefully, the results of our analyses will direct future proteomic experimentation in cultured human cells that will support our findings.

A number of PS1IV-regulated subnetworks are also altered in sAD brains

The human equivalent of PS1IV, PS2V, is upregulated in sAD brains [8, 11] and may modulate gene expression in the disease. Therefore, we asked whether similar subnetworks are affected by changes in PS1IV expression and in human sAD. A microarray analysis of human sAD versus normal control brains has previously been performed by Antonell and co-workers using RNA isolated from the posterior cingulate area of brains at the thalamus level [30]. We took the 50 most statistically significant differentially expressed genes in sAD reported by Antonnell et al. and subjected these to subnetwork discovery at p < 0.05. These were then compared to the significant subnetworks identified in our study of PS1IV in this paper. Thirteen shared subnetworks were identified including those related to expression of the genes IL1B, HSPA8, CALR, and TH (subnetworks Neighbors of MAPKAPK2 and Neighbors of MEF2A) and subnetworks Neighbors of NOTCH2 and Neighbors of hypoxia (see Table 4). This supports that PS2V is responsible for a subset of the changes in gene expression seen in sAD brains. We are currently preparing a detailed analysis of Antonell et al.’s microarray data to examine the contribution of hypoxia to changes in gene expression in sAD brains.

Literature mining-based network analysis of under-expressed genes after failure to form PS1IV highlights human PS2V’s possible involvement in cancer and cardiomyopathy

During our construction of various networks, one was of particular interest. By using the shortest path algorithm to analyze genes under-expressed after failure to form PS1IV and by incorporating disease terms (to extract relationships between these genes and different diseases from the scientific literature, see Material and Methods) we identified a network that emphasizes the pathological importance of changes in expression of IL1B, HSPA8, TH and CALR (see Supplementary Material 15 and Supplementary Material 16). This suggests that PS1IV’s human homologue, PS2V, may play roles in the progression of cancer and in preeclampsia. Changes in the expression of the under-expressed genes TH, IL1B, PLCB1, and AMPD1 are also associated with heart failure. This is interesting since mutations in both human PSEN1 and PSEN2 exist that are thought to cause dilated cardiomyopathy [112].

DISCUSSION

Despite their importance as fAD loci, genome-wide association studies have failed to identify the PRESENILIN genes as risk loci in the predominant, late onset form of AD, sAD. This has raised concern that the pathological mechanisms underlying these two forms of AD may differ. However, a particular isoform of PSEN2 upregulated by hypoxia [8] and high cholesterol [15], PS2V, shows increased expression in sAD brains. Therefore it is important to understand the role this isoform plays in gene regulation. Investigation of PS2V function in the genetically manipulable rodent models is thwarted by the accelerated evolution of the fAD genes in these animals and their consequent loss of ability to express PS2V [113]. However, the ability to express this isoform in response to hypoxia has been retained by zebrafish and the molecular mechanism controlling induction of zebrafish PS1IV and human PS2V under hypoxia is conserved [14]. We have also previously observed that the pathological function of the PRESENILIN genes cannot be correctly understood in physiologically abnormal systems [14]. Our ability to inhibit specifically the induction of the endogenous PS1IV isoform under hypoxia-like conditions in developing zebrafish embryos enables us to examine its function in an otherwise normal cellular state (in contrast to the abnormal cell types and growth conditions commonly involved in studies using tissue culture).

Aside from the abilities of PS2V/PS1IV to suppress the UPR and stimulate γ-secretase activity, nothing is known regarding the function of this isoform, why this function has been conserved over nearly half a billion years since the tetrapod/teleost divergence, or why this isoform was lost from the rodent family Muridae but not other rodents or mammals [14]. Disconcertingly, our results show that this isoform regulates expression of a number of genes and cellular systems associated with a variety of neurodegenerative conditions including AD (see below), Parkinson’s disease (e.g., LRRK2, TH, and GAK), schizophrenia (e.g., PLCB1, PDLIM5), early-onset nonprogressive cerebellar ataxia (CAMTA1), and others. This reinforces concerns that mouse models of AD and other neurological diseases do not accurately represent the human conditions [114].

Our microarray analysis has shown that loss of PS1IV function in zebrafish embryos under hypoxia-like conditions significantly alters the expression of a number of genes implicated in AD pathology in humans. We have listed a number of these genes in this paper but a greater depth of analysis is required in order to understand the changes in cell biology that these gene expression changes represent. Gene regulatory network analysis can provide greater insight into function than simply identifying genes with greatest and/or most significantly changed expression. This is because small changes in the expression of genes at nodal positions in networks can have greater systemic effects on cell biology than large changes in the expression of non-nodal genes. One example of this is the TUBB gene that shows the most statistically significant and greatest fold change in expression of any gene in our analysis but that is not very enlightening in terms of understanding the functional consequences of PS1IV loss. Interestingly, however, problems with tubulin polymerization into microtubules have been observed in AD brains [115] and, of course, AD pathology is associated with abnormal aggregation of MICROTUBULE-ASSOCIATED PROTEIN TAU, MAPT into neurofibrillary tangles. In contrast, it was significant to see that PS1IV is required to maintain or increase the expression of the nodal gene CALR under hypoxia since the CALR protein acts as a chaperone in the maturation of glycoproteins including AβPP [116]. One previous report observed CALR protein as down-regulated in neurons in AD brains [117]. However, we have performed a detailed meta-analysis of gene expression data from multiple late-onset AD and normal brains and this appears to support the existence of a statistically significant upregulation of CALR transcripts in AD brains (Ebrahimie et al., manuscript in preparation.) It may be that pre- and post-transcriptional regulation of CALR is differentially affected in AD. In any case, our analysis supports that expression of PS2V in AD brain is an attempt by neural cells to cope with the stresses that are driving the development of AD pathology. CALR protein is known to exist in complexes with the protein PGRN implicated in frontotemporal dementia (FTD) and has also been shown to bind the ATP2A1 protein that binds PSEN1 [118]. This supports a putative functional link between the major genetic locus mutated in familial AD (PSEN1) and the second most commonly mutated gene in familial FTD (PGRN). Interestingly, there are mutations in PSEN1 that are thought to cause FTD rather than FAD and these have been observed to differ in their effects on subcellular calcium transport [119, 120]. This emphasizes the importance of Ca^2 + in PRESENILIN function. Since the PRESENILINs and ATP2A1 are all MAM-resident proteins, the data above imply that PGRN will also show localization in the MAM. Of course, since PGRN possesses a very high number of disulphide bonds [121], the oxidative protein folding that occurs in the MAM must be central to PGRN function. Reassuringly, the involvement of CALR, a protein important for quality control of protein folding and for signal transduction [121], showed that our microarray and network analyses are consistent with the known roles of PS1IV and PS2V in modulation of the UPR.

Other observations from our network analysis further support that the human equivalent of PS1IV, PS2V, plays a role in AD pathology. Somewhat unexpectedly, we saw that PS1IV is important for the upregulation of 1L1B and CCR5 that, in humans, have been strongly implicated in the inflammation characterizing AD brains. The hypoxia that upregulates PS1IV can also drive inflammatory, innate immune responses (e.g., [122]), possibly by increasing the amount of incorrectly folded proteins in a cell [123]. In our recent paper describing PS1IV we argued that such truncated PSEN isoforms function to inhibit the over-activation of the UPR under hypoxia that can lead to cell death [14]. By inhibiting the UPR, PS1IV may permit higher levels of mis-folded proteins under hypoxia and thus stimulate inflammatory responses including IL1B expression. We note also that the NLRP3 inflammasome that can promote cleavage of pro-IL1β to form active IL1β protein is apparently localized to the MAM [124]. Recently, activation of the innate immune response was shown to be suppressed in the brains of mice lacking expression of the fAD gene AβPP [125]. In terms of AD, the AβPP protein is the most important substrate for the γ-secretase activity of the PSEN proteins and both PS2V and PS1IV can act to increase γ-secretase activity in cells. Thus, another molecular subnetwork connecting PS2V/PS1IV to increased IL1B expression in sAD may be via the Aβ and/or AICD production from AβPP that would increase with increased γ-secretase activity.

Upregulation of PS2V mRNA expression [8, 11] and accumulation of PS2V protein [126] in the human brain are markers of sAD. Since hypoxia is important for induction of PS2V via oxidative stress, this supports that insufficient brain oxygenation may be a fundamental driver of sAD pathology. This is consistent with evidence that serum markers of hypoxia can distinguish people who progress from mild cognitive impairment to AD versus from those who do not [127]. Also, hypoxia has been seen to upregulate production of the AD-associated peptide Aβ in cultured human neuroblastoma cells and to greatly raise Aβ in serum [128]. Thus, in humans, PS2V may form a facilitating link between hypoxia and a number of pathological processes in AD brains. It is possible that blockage of PS2V action in AD brains might inhibit some of these pathological processes such as the inflammation provoked by IL1B expression. It might relieve suppression of the UPR that normally acts to restrict generation of inappropriately folded, aggregation-prone proteins. On the other hand, PS2V suppression might increase the cell death that results from over-stimulation of the UPR [14] while decreasing expression of CALR with negative consequences for protein folding (see above). Understanding the multifaceted actions in sAD of the human PS2V isoform will require using the data we have obtained from zebrafish to generate hypotheses that can be tested in human cellular systems such as in neurons derived from induced pluripotent stem cells [129].

Our analyses in this paper show that extending examination of the consequences of gene/isoform dysregulation from simplistic observation of statistically significant changes in gene expression to construction of statistically significant theoretical gene interaction networks enhances our understanding of the normal functions of dysregulated genes and protein isoforms. Thus, gene network analysis can contribute greatly to our understanding of the pathological mechanisms underlying complex disorders such as AD. The fact that, despite publication of over 80,000 scientific papers on AD (excluding reviews), there is no firm consensus on either its underpinning molecular mechanisms or disease treatment reinforces the need to implement more advanced analytical strategies such as machine learning (data mining), network analysis, promoter architecture analysis, promoter-binding based network analysis, and gene ontology illustration to reveal the cellular changes and stresses underlying AD pathology [26 , 130–133].

Footnotes

ACKNOWLEDGMENTS

This work was supported by the Judith Jane Mason & Harold Stannett Williams Memorial Foundation [grant 13033 to ML]; and by the Australian National Health and Medical Research Council [NHMRC, Project Grant APP1061006 to ML, MN, and EE, ] (awarded to A/Professor Michael Lardelli, Dr. Morgan Newman, and Dr. Esmaeil Ebrahimie); and the School of Molecular and Biomedical Science of the University of Adelaide [support of ML and MN]. We are grateful for the support to MN given by the family of Lindsay Carthew. We would like to thank Professor Greg Goodall, Centre for Cancer Biology, Adelaide, Australia for his help in RNA expression analysis. We are also thankful for the assistance of Dr. Manijeh Mohammadi Dehcheshmeh, School of Agriculture, Food and Wine of The University of Adelaide, Australia in preparation of figures and graphics.

Authors’ disclosures available online ().

Appendix

References

Blennow

, de Leon

, Zetterberg

(2006) Alzheimer’s disease. Lancet 368, 387–403.

Selkoe

(2001) Alzheimer’s disease results from the cerebral accumulation and cytotoxicity of amyloid beta-protein. J Alzheimers Dis 3, 75–80.

Katayama

, Bandoh

, Kishibe

, Takahara

, Ogino

, Nonaka

, Harabuchi

(2004) Expressions of matrix metalloproteinases in early-stage oral squamous cell carcinoma as predictive indicators for tumor metastases and prognosis. Clin Cancer Res 10, 634–640.

van der Kant

, Goldstein

(2015) Cellular functions of the amyloid precursor protein from development to dementia. Dev Cell 32, 502–515.

Barage

, Sonawane

(2015) Amyloid cascade hypothesis: Pathogenesis and therapeutic strategies in Alzheimer’s disease. Neuropeptides 52, 1–18.

Chouraki

, Seshadri

(2014) Genetics of Alzheimer’s disease. Adv Genet 87, 245–294.

Stiller

, Lizak

, Banhegyi

(2014) Physiological functions of presenilins; beyond gamma-secretase. Curr Pharm Biotechnol 15, 1019–1025.

Sato

, Hori

, Yamaguchi

, Lambert

, Chartier-Harlin

, Robinson

, Delacourte

, Schmidt

, Furuyama

, Imaizumi

, Tohyama

, Takagi

(1999) A novel presenilin-2 splice variant in human Alzheimer’s disease brain tissue. J Neurochem 72, 2498–2505.

Sato

, Imaizumi

, Manabe

, Taniguchi

, Hitomi

, Katayama

, Yoneda

, Morihara

, Yasuda

, Takagi

, Kudo

, Tsuda

, Itoyama

, Makifuchi

, Fraser

, St George-Hyslop

, Tohyama

(2001) Increased production of beta-amyloid and vulnerability to endoplasmic reticulum stress by an aberrant spliced form of presenilin 2. J Biol Chem 276, 2108–2114.

10.

Nishikawa

, Manabe

, Katayama

, Kudo

, Matsuzaki

, Yanagita

, Okuda

, Bando

, Tohyama

(2004) Novel function of PS2V: Change in conformation of tau proteins. Biochem Biophys Res Commun 318, 435–438.

11.

Smith

, Sharples

, Evin

, McLean

, Dean

, Pavey

, Fantino

, Cotton

RGH

, Imaizumi

, Masters

, Cappai

, Culvenor

(2004) Expression of truncated presenilin 2 splice variant in Alzheimer’s disease, bipolar disorder, and schizophrenia brain cortex. Mol Brain Res 127, 128–135.

12.

Newman

, Wilson

, Verdile

, Lim

, Khan

, Moussavi Nik

, Pursglove

, Chapman

, Martins

, Lardelli

(2014) Differential, dominant activation and inhibition of Notch signalling and APP cleavage by truncations of PSEN1 in human disease. Hum Mol Genet 23, 602–617.

13.

Jayadev

, Leverenz

, Steinbart

, Stahl

, Klunk

, Yu

, Bird

(2010) Alzheimer’s disease phenotypes and genotypes associated with mutations in presenilin 2. Brain 133, 1143–1154.

14.

Moussavi Nik

, Newman

, Wilson

, Ebrahimie

, Wells

, Musgrave

, Verdile

, Martins

, Lardelli

(2015) Alzheimer’s disease-related peptide PS2V plays ancient, conserved roles in suppression of the unfolded protein response under hypoxia and stimulation of γ-secretase activity. Hum Mol Genet 24, 3662–3678.

15.

Sharman

, Moussavi Nik

, Chen

, Ong

, Wijaya

, Laws

, Taddei

, Newman

, Lardelli

, Martins

, Verdile

(2013) The guinea pig as a model for sporadic Alzheimer’s disease (AD): The impact of cholesterol intake on expression of AD-related genes. PLoS One 8, e66235.

16.

Moussavi Nik

, Newman

, Lardelli

(2011) The response of HMGA1 to changes in oxygen availability is evolutionarily conserved. Exp Cell Res 317, 1503–1512.

17.

Westerfield

(2007) The zebrafish book. A guide for the laboratory use of zebrafish (Danio rerio), University of Oregon Press.

18.

Nornes

, Newman

, Verdile

, Wells

, Stoick-Cooper

, Tucker

, Frederich-Sleptsova

, Martins

, Lardelli

(2008) Interference with splicing of Presenilin transcripts has potent dominant negative effects on Presenilin activity. Hum Mol Genet 17, 402–412.

19.

Gao

, Okunieff

, Han

, Ding

, Wang

, Liu

, Zhang

, Yang

, Chen

, Underhill

, Kim

, Zhang

(2005) Hypoxia-induced alterations in hyaluronan and hyaluronidase. In Oxygen Transport to Tissue XXVI, Okunieff

, Williams

, Chen

eds. Springer US, pp, 249–256.

20.

Nik

SHM

, Newman

, Ganesan

, Chen

, Martins

, Verdile

, Lardelli

(2014) Hypoxia alters expression of Zebrafish Microtubule-associated protein Tau (mapta, maptb) gene transcripts. BMC Res Note 7, 767.

21.

Moussavi Nik

, Wilson

, Newman

, Croft

, Mori

, Musgrave

, Lardelli

(2012) The BACE1-PSEN-AbetaPP regulatory axis has an ancient role in response to low oxygen/oxidative stress. J Alzheimers Dis 28, 515–530.

22.

Bolstad

, Irizarry

, Astrand

, Speed

(2003) A comparison of normalization methods for high density oligonucleotide array data based on variance and bias. Bioinformatics 19, 185–193.

23.

Irizarry

, Bolstad

, Collin

, Cope

, Hobbs

, Speed

(2003) Summaries of Affymetrix GeneChip probe level data. Nucleic Acids Res 31, e15.

24.

Irizarry

, Hobbs

, Collin

, Beazer-Barclay

, Antonellis

, Scherf

, Speed

(2003) Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics 4, 249–264.

25.

Fruzangohar

, Ebrahimie

, Ogunniyi

, Mahdi

, Paton

, Adelson

(2013) Comparative GO: A web application for comparative gene ontology and gene ontology-based gene selection in bacteria. PLoS One 8, e58759.

26.

Fruzangohar

, Ebrahimie

, Adelson

(2014) Application of global transcriptome data in gene ontology classification and construction of a gene ontology interaction network. bioRxiv, doi: http://dx.doi.org/10.1101/004911.

27.

Subramanian

, Tamayo

, Mootha

, Mukherjee

, Ebert

, Gillette

, Paulovich

, Pomeroy

, Golub

, Lander

(2005) Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102, 15545–15550.

28.

Nikitin

, Egorov

, Daraselia

, Mazo

(2003) Pathway studio—the analysis and navigation of molecular networks. Bioinformatics 19, 2155–2157.

29.

Novichkova

, Egorov

, Daraselia

(2003) MedScan, a natural language processing engine for MEDLINE abstracts. Bioinformatics 19, 1699–1706.

30.

Antonell

, Llado

, Altirriba

, Botta-Orfila

, Balasa

, Fernandez

, Ferrer

, Sanchez-Valle

, Molinuevo

(2013) A preliminary study of the whole-genome expression profile of sporadic and monogenic early-onset Alzheimer’s disease. Neurobiol Aging 34, 1772–1778.

31.

Hosseinpour

, HajiHoseini

, Kashfi

, Ebrahimie

, Hemmatzadeh

(2012) Protein interaction network of Arabidopsis thaliana female gametophyte development identifies novel proteins and relations. PLoS One 7, e49931.

32.

Alanazi

, Ebrahimie

, Hoffmann

, Adelson

(2013) Combined gene expression and proteomic analysis of EGF induced apoptosis in A431 cells suggests multiple pathways trigger apoptosis. Apoptosis 18, 1291–1305.

33.

Ebrahimie

, Esmaeili

, Cheraghi

, Houshmand

, Shabani

, Ebrahimie

(2014) Efficient and simple production of insulin-producing cells from embryonal carcinoma stem cells using mouse neonate pancreas extract, as a natural inducer. PLoS One 9, e90885.

34.

Alisoltani

, Fallahi

, Ebrahimi

, Ebrahimie

(2014) Prediction of potential cancer-risk regions based on transcriptome data: Towards a comprehensive view. PLoS One 9, e96320.

35.

Lin

C-Y

, Chin

C-H

, Wu

H-H

, Chen

S-H

, Ho

C-W

, Ko

M-T

(2008) Hubba: Hub objects analyzer— a framework of interactome hubs identification for network biology. Nucleic Acids Res 36, W438–W443.

36.

Shannon

, Markiel

, Ozier

, Baliga

, Wang

, Ramage

, Amin

, Schwikowski

, Ideker

(2003) Cytoscape: A software environment for integrated models of biomolecular interaction networks. Genome Res 13, 2498–2504.

37.

Khosravi

, Gazestani

, Asgari

, Law

, Sadeghi

, Goliaei

(2014) Network-based approach reveals Y chromosome influences prostate cancer susceptibility. Comput Biol Med 54, 24–31.

38.

Newman

, Lardelli

, Ebrahimie

(2014) Using the zebrafish model for Alzheimer’s disease research. Front Genet 5, 189.

39.

Mansouri

, Esmaeili

, Nejatpour

, Houshmand

, Shabani

, Ebrahimie

(2014) Differentiation of P19 embryonal carcinoma stem cells into insulin-producing cells promoted by pancreas-conditioned medium. J Tissue Eng Regen Med, doi: 10.1002/term.1927.

40.

Cook

, Re

, deHaro

, Abadie

, Peters

, Alam

(2008) The trafficking protein GABARAP binds to and enhances plasma membrane expression and function of the angiotensin II type 1 Receptor. Circ Res 102, 1539–1547.

41.

Tanji

, Mori

, Kakita

, Takahashi

, Wakabayashi

(2011) Alteration of autophagosomal proteins (LC3, GABARAP and GATE-16) in Lewy body disease. Neurobiol Dis 43, 690–697.

42.

Nowak

, Archange

, Tardivel-Lacombe

, Pontarotti

, Pébusque

M-J

, Vaccaro

, Velasco

, Dagorn

J-C

, Iovanna

(2009) The TP53INP2 protein is required for autophagy in mammalian cells. Mol Biol Cell 20, 870–881.

43.

Hemelaar

, Lelyveld

, Kessler

, Ploegh

(2003) A single protease, Apg4B, is specific for the autophagy-related ubiquitin-like proteins GATE-16, MAP1-LC3, GABARAP, and Apg8L. J Biol Chem 278, 51841–51850.

44.

Weidberg

, Shvets

, Shpilka

, Shimron

, Shinder

, Elazar

(2010) LC3 and GATE-16/GABARAP subfamilies are both essential yet act differently in autophagosome biogenesis. EMBO J 29, 1792–1802.

45.

Mizushima

, Ohsumi

, Yoshimori

(2002) Autophagosome formation in mammalian cells. Cell Struct Funct 27, 421–429.

46.

Yoshimori

(2004) Autophagy: A regulated bulk degradation process inside cells. Biochem Biophys Res Commun 313, 453–458.

47.

Maturana

, Wälchli

, Iwata

, Ryser

, Van Lint

, Hoshijima

, Schlegel

, Ikeda

, Tanizawa

, Kuroda

(2008) Enigma homolog 1 scaffolds protein kinase D1 to regulate the activity of the cardiac L-type voltage-gated calcium channel. Cardiovasc Res 78, 458–465.

48.

De Mello

(2011) Novel aspects of angiotensin II action in the heart. Implications to myocardial ischemia and heart failure. Regul Pept 166, 9–14.

49.

Kato

, Iwayama

, Kakiuchi

, Iwamoto

, Yamada

, Minabe

, Nakamura

, Mori

, Fujii

, Nanko

, Yoshikawa

(2005) Gene expression and association analyses of LIM (PDLIM5) in bipolar disorder and schizophrenia. Mol Psychiatry 10, 1045–1055.

50.

Iga

J-i

, Ueno

S-i

, Yamauchi

, Numata

, Motoki

, Tayoshi

, Kinouchi

, Ohta

, Song

, Morita

, Rokutan

, Tanabe

, Sano

, Ohmori

(2006) Gene expression and association analysis of LIM (PDLIM5) in major depression. Neurosci Lett 400, 203–207.

51.

, Ponten

, dos Remedios

(2012) The interactome of LIM domain proteins: The contributions of LIM domain proteins to heart failure and heart development. Proteomics 12, 203–225.

52.

Swindell

, Zilinski

, Hashimoto

, Shah

, Lane

, Jamrich

(2008) Regulation and function of foxe3 during early zebrafish development. Genesis 46, 177–183.

53.

Gustafsson

, Zheng

, Pereira

, Gradin

, Jin

, Lundkvist

, Ruas

, Poellinger

, Lendahl

, Bondesson

(2005) Hypoxia requires notch signaling to maintain the undifferentiated cell state. Dev Cell 9, 617–628.

54.

Keith

, Simon

(2007) Hypoxia-inducible factors, stem cells, and cancer. Cell 129, 465–472.

55.

Ethell

(2010) An amyloid-notch hypothesis for Alzheimer’s disease. Neuroscientist 16, 614–617.

56.

Bai

, Seebald

, Kim

, Ding

, Szeto

, Chang

(2010) Disruption of zebrafish cyclin G-associated kinase (GAK) function impairs the expression of Notch-dependent genes during neurogenesis and causes defects in neuronal development. BMC Dev Biol 10, 7.

57.

Rhodes

, Sinsheimer

, Bordelon

, Bronstein

, Ritz

(2011) Replication of GWAS associations for GAK and MAPT in Parkinson’s disease. Ann Hum Genet 75, 195–200.

58.

Dumitriu

, Pacheco

, Wilk

, Strathearn

, Latourelle

, Goldwurm

, Pezzoli

, Rochet

, Lindquist

, Myers

(2011) Cyclin-G-associated kinase modifies alpha-synuclein expression levels and toxicity in Parkinson’s disease: Results from the GenePD Study. Hum Mol Genet 20, 1478–1487.

59.

Newman

, Tucker

, Nornes

, Ward

, Lardelli

(2009) Altering presenilin gene activity in zebrafish embryos causes changes in expression of genes with potential involvement in Alzheimer’s disease pathogenesis. J Alzheimers Dis 16, 133–147.

60.

Moussavi Nik

, Wilson

, Newman

, Croft

, Mori

, Musgrave

, Lardelli

(2012) The BACE1-PSEN-AβPP regulatory axis has an ancient role in response to low oxygen/oxidative stress. J Alzheimers Dis 28, 515–530.

61.

Herr

, Rodewald

, Fraser

, Hack

, Konrad

, Kreienberg

, Wulff

(2008) Regulation of endothelial proliferation by the renin-angiotensin system in human umbilical vein endothelial cells. Reproduction 136, 125–130.

62.

Quillien

, Moore

, Shin

, Siekmann

, Smith

, Pan

, Moens

, Parsons

, Lawson

(2014) Distinct Notch signaling outputs pattern the developing arterial system. Development 141, 1544–1552.

63.

Ricchiuti

, Lian

, Oestreicher

, Tran

, Stone

, Yao

, Seely

, Williams

, Adler

(2009) Estradiol increases angiotensin II type 1 receptor in hearts of ovariectomized rats. J Endocrinol 200, 75–84.

64.

Zouali

, Sarmay

(2004) B lymphocyte signaling pathways in systemic autoimmunity: Implications for pathogenesis and treatment. Arthritis Rheum 50, 2730–2741.

65.

Guo

, Zhang

, Liu

, Wang

, Cheng

, Qiao

(2004) Testosterone influenced the expression of Notch1, Notch2 and Jagged1 induced by lipopolysaccharide in macrophages. Exp Toxicol Pathol 56, 173–179.

66.

Matsuda

, Hattori

, Sakuraya

, Kobayashi

, Zhang

X-H

, Kemmotsu

, Gando

(2002) Hemodynamic significance of histamine synthesis and histamine H1- and H2-receptor gene expression during endotoxemia. Naunyn-Schmiedebergs Arch Pharmacol 366, 513–521.

67.

Wang

, Rousset

, Hagberg

, Mallard

(2006) Lipopolysaccharide-induced inflammation and perinatal brain injury. Semin Fetal Neonatal Med 11, 343–353.

68.

Ackermann

, Ashton

, Lyons

, James

, Hornung

, Jones

, Breitwieser

(2011) Loss of ATF2 function leads to cranial motoneuron degeneration during embryonic mouse development. PLoS One 6, e19090.

69.

Van Dam

, Wilhelm

, Herr

, Steffen

, Herrlich

, Angel

(1995) ATF-2 is preferentially activated by stress-activated protein kinases to mediate c-jun induction in response to genotoxic agents. EMBO J 14, 1798.

70.

Bhunia

, Piontek

, Boletta

, Liu

, Qian

, Xu

, Germino

(2002) PKD1 induces p21(waf1) and regulation of the cell cycle via direct activation of the JAK-STAT signaling pathway in a process requiring PKD2. Cell 109, 157–168.

71.

Area-Gomez

, Del Carmen Lara Castillo

, Tambini

, Guardia-Laguarta

, de Groof

, Madra

, Ikenouchi

, Umeda

, Bird

, Sturley

, Schon

(2012) Upregulated function of mitochondria-associated ER membranes in Alzheimer disease. EMBO J 31, 4106–4123.

72.

Simmen

, Lynes

, Gesson

, Thomas

(2010) Oxidative protein folding in the endoplasmic reticulum: Tight links to the mitochondria-associated membrane (MAM). Biochim Biophys Acta 1798, 1465–1473.

73.

Mattson

, LaFerla

, Chan

, Leissring

, Shepel

, Geiger

(2000) Calcium signaling in the ER: Its role in neuronal plasticity and neurodegenerative disorders. Trends Neurosci 23, 222–229.

74.

Clapham

(2007) Calcium signaling. Cell 131, 1047–1058.

75.

Timmins

, Ozcan

, Seimon

, Li

, Malagelada

, Backs

, Bassel-Duby

, Olson

, Anderson

, Tabas

(2009) Calcium/calmodulin-dependent protein kinase II links ER stress with Fas and mitochondrial apoptosis pathways. J Clin Invest 119, 2925–2941.

76.

Rebhan

, Chalifa-Caspi

, Prilusky

, Lancet

(1998) GeneCards: A novel functional genomics compendium with automated data mining and query reformulation support. Bioinformatics 14, 656–664.

77.

Safran

, Solomon

, Shmueli

, Lapidot

, Shen-Orr

, Adato

, Ben-Dor

, Esterman

, Rosen

, Peter

(2002) GeneCardstrademark 2002: Towards a complete, object-oriented, human gene compendium. Bioinformatics 18, 1542–1543.

78.

Yang

, She

, Gearing

, Colla

, Lee

, Shacka

, Mao

(2009) Regulation of neuronal survival factor MEF2D by chaperone-mediated autophagy. Science 323, 124–127.

79.

Youn

, Chatila

, Liu

(2000) Integration of calcineurin and MEF2 signals by the coactivator p300 during T-cell apoptosis. EMBO J 19, 4323–4331.

80.

Škuljec

, Sun

, Pul

, Bénardais

, Ragancokova

, Moharregh-Khiabani

, Kotsiari

, Trebst

, Stangel

(2011) CCL5 induces a pro-inflammatory profile in microglia in vitro. Cell Immunol 270, 164–171.

81.

Reale

, Iarlori

, Feliciani

, Gambi

(2008) Peripheral chemokine receptors, their ligands, cytokines and Alzheimer’s disease. J Alzheimers Dis 14, 147–159.

82.

Goldeck

, Larbi

, Pellicanó

, Alam

, Zerr

, Schmidt

, Fulop

, Pawelec

(2013) Enhanced chemokine receptor expression on leukocytes of patients with Alzheimer’s disease. PLoS One 8, e66664.

83.

Cowell

, Xu

, Parent

, Silverstein

(2006) Microglial expression of chemokine receptor CCR5 during rat forebrain development and after perinatal hypoxia–ischemia. J Neuroimmunol 173, 155–165.

84.

Bosco

, Reffo

, Puppo

, Varesio

(2004) Hypoxia inhibits the expression of the CCR5 chemokine receptor in macrophages. Cell Immunol 228, 1–7.

85.

Gozal

, Sachleben

, Czyzyk-Krzeska

, Li

, Guo

S-Z

, Gozal

, Shah

, Pequignot

J-M

, Pequignot

(2011) Tyrosine hydroxylase expression and activity in the rat. Am J Physiol Cell Physiol 300, C743–C751.

86.

Nakashima

, Ota

, Kaneko

, Mori

, Nagasaki

, Nagatsu

(2013) A possible pathophysiological role of tyrosine hydroxylase in Parkinson’s disease suggested by postmortem brain biochemistry: A contribution for the special 70th birthday symposium in honor of Prof. Peter Riederer. J Neural Transm 120, 49–54.

87.

Zhang

, Kanthasamy

, Anantharam

, Kanthasamy

(2011) Effects of manganese on tyrosine hydroxylase (TH) activity and TH-phosphorylation in a dopaminergic neural cell line. Toxicol Appl Pharmacol 254, 65–71.

88.

Gehrke

, Imai

, Sokol

, Lu

(2010) Pathogenic LRRK2 negatively regulates microRNA-mediated translational repression. Nature 466, 637–641.

89.

Sciacca

, Ferri

, Licastro

, Veglia

, Biunno

, Gavazzi

, Calabrese

, Martinelli Boneschi

, Sorbi

, Mariani

(2003) Interleukin-1B polymorphism is associated with age at onset of Alzheimer’s disease. Neurobiol Aging 24, 927–931.

90.

Louboutin

J-P

, Strayer

(2013) Relationship between the chemokine receptor CCR5 and microglia in neurological disorders: Consequences of targeting CCR5 on neuroinflammation, neuronal death and regeneration in a model of epilepsy. CNS Neurol Disord-Drug Targets 12, 815–829.

91.

Maier

, Kincaid

, Pagenstecher

, Campbell

(2002) Regulation of signal transducer and activator of transcription and suppressor of cytokine-signaling gene expression in the brain of mice with astrocyte-targeted production of interleukin-12 or experimental autoimmune encephalomyelitis. Am J Pathol 160, 271–288.

92.

Burns

, Duggan

, Atkinson

, Famulski

, Nemer

, Bleackley

, Michalak

(1994) Modulation of gene expression by calreticulin binding to the glucocorticoid receptor. Nature 367, 476–480.

93.

Almeida

, Zhou

, Gao

F-B

(2011) Progranulin, a glycoprotein deficient in frontotemporal dementia, is a novel substrate of several protein disulfide isomerase family proteins. PLoS One 6, e26454.

94.

Area-Gomez

, de Groof

, Boldogh

, Bird

, Gibson

, Koehler

, Yu

, Duff

, Yaffe

, Pon

(2009) Presenilins are enriched in endoplasmic reticulum membranes associated with mitochondria. Am J Pathol 175, 1810–1816.

95.

Medina

, Carretero

, Ballestar

, Angulo

, Lopez-Rios

, Esteller

, Sanchez-Cespedes

(2005) Transcriptional targets of the chromatin-remodelling factor SMARCA4/BRG1 in lung cancer cells. Hum Mol Genet 14, 973–982.

96.

Kong

, Mou

, Liu

, Chen

, Vanderburg

, Rogers

, Huang

(2009) Independent component analysis of Alzheimer’s DNA microarray gene expression data. Mol Neurodegener 4, 1–14.

97.

Dityatev

, Schachner

, Sonderegger

(2010) The dual role of the extracellular matrix in synaptic plasticity and homeostasis. Nat Rev Neurosci 11, 735–746.

98.

Antoniu

(2012) Discontinued drugs 2011: Pulmonary, allergy, gastrointestinal and arthritis. Expert Opin Investig Drugs 21, 1607–1618.

99.

Sun

, Yoshizuka

, New

, Moser

, Li

, Liao

, Xie

, Chen

, Deng

, Yamout

(2007) PRAK is essential for ras-induced senescence and tumor suppression. Cell 128, 295–308.

100.

McDonald

, Bamberger

, Combs

, Landreth

(1998) β-Amyloid fibrils activate parallel mitogen-activated protein kinase pathways in microglia and THP1 monocytes. J Neurosci 18, 4451–4460.

101.

Vasco

VRL

, Cardinale

, Polonia

(2012) Deletion of PLCB1 gene in schizophrenia-affected patients. J Cell Mol Med 16, 844–851.

102.

Prabakaran

, Swatton

, Ryan

, Huffaker

, Huang

J-J

, Griffin

, Wayland

, Freeman

, Dudbridge

, Lilley

(2004) Mitochondrial dysfunction in schizophrenia: Evidence for compromised brain metabolism and oxidative stress. Mol Psychiatry 9, 684–697.

103.

Katayama

, Imaizumi

, Sato

, Miyoshi

, Kudo

, Hitomi

, Morihara

, Yoneda

, Gomi

, Mori

(1999) Presenilin-1 mutations downregulate the signalling pathway of the unfolded-protein response. Nat Cell Biol 1, 479–485.

104.

Zhang

, Szabo

, Michalak

, Opas

(2007) Endoplasmic reticulum stress during the embryonic development of the central nervous system in the mouse. Int J Dev Neurosci 25, 455–463.

105.

Molinari

, Helenius

(1999) Glycoproteins form mixed disulphides with oxidoreductases during folding in living cells. Nature 402, 90–93.

106.

Jin

, Sanjo

, Uchihara

, Watabe

, George-Hyslop

, Fraser

, Mizusawa

(2010) Presenilin-1 holoprotein is an interacting partner of sarco endoplasmic reticulum calcium-ATPase and confers resistance to endoplasmic reticulum stress. J Alzheimers Dis 20, 261–273.

107.

Roussel

, Kruppa

, Miranda

, Crowther

, Lomas

, Marciniak

(2013) Endoplasmic reticulum dysfunction in neurological disease. Lancet Neurol 12, 105–118.

108.

Lee

J-H

, Yu

, Kumar

, Lee

, Mohan

, Peterhoff

, Wolfe

, Martinez-Vicente

, Massey

, Sovak

(2010) Lysosomal proteolysis and autophagy require presenilin 1 and are disrupted by Alzheimer-related PS1 mutations. Cell 141, 1146–1158.

109.

Brodsky

(2012) Cleaning up: ER-associated degradation to the rescue. Cell 151, 1163–1167.

110.

Schon

, Area-Gomez

(2010) Is Alzheimer’s disease a disorder of mitochondria-associated membranes?. J Alzheimers Dis 20, 281–292.

111.

Petushkova

, Kuznetsova

, Larina

, Kisrieva

, Samenkova

, Trifonova

, Miroshnichenko

, Zolotarev

, Karuzina

, Ipatova

, Lisitsa

(2015) One-dimensional proteomic profiling of Danio rerio embryo vitellogenin to estimate quantum dot toxicity. Proteome Sci 13, 17.

112.

Friedrich

, Carrier

(2012) Genetics of hypertrophic and dilated cardiomyopathy. Curr Pharm Biotech 13, 2467–2476.

113.

Moussavi Nik

, Newman

, Wilson

, Ebrahimie

, Wells

, Musgrave

, Verdile

, Martins

, Lardelli

114.

Bales

(2012) The value and limitations of transgenic mouse models used in drug discovery for Alzheimer’s disease: An update. Expert Opin Drug Discov 7, 281–297.

115.

Boutté

, Neely

, Bird

, Montine

(2005) Diminished taxol/GTP-stimulated tubulin polymerization in diseased region of brain from patients with late-onset or inherited Alzheimer’s disease or frontotemporal dementia with parkinsonism linked to chromosome-17 but not individuals with mild cognitive impairment. J Alzheimers Dis 8, 1–6.

116.

Johnson

, Xiao

, Shanmugaratnam

, Fine

(2001) Calreticulin functions as a molecular chaperone for the β-amyloid precursor protein1. Neurobiol Aging 22, 387–395.

117.

Taguchi

, Fujii

, Fujino

, Tsujioka

, Takahashi

, Tsuboi

, Wada

, Yamada

(2000) Different expression of calreticulin and immunoglobulin binding protein in Alzheimer’s disease brain. Acta Neuropathol 100, 153–160.

118.

Ihara

, Kageyama

, Kondo

(2005) Overexpression of calreticulin sensitizes SERCA2a to oxidative stress. Biochem Biophys Res Commun 329, 1343–1349.

119.

Nelson

, Tu

, Lei

, Bentahir

, De Strooper

, Bezprozvanny

(2007) Familial Alzheimer disease–linked mutations specifically disrupt Ca 2+ leak function of presenilin 1. J Clin Invest 117, 1230–1239.

120.

Cheung

K-H

, Mei

, Mak

D-OD

, Hayashi

, Iwatsubo

, Kang

, Foskett

(2010) Gain-of-function enhancement of IP3 receptor modal gating by familial Alzheimer’s disease-linked presenilin mutants in human cells and mouse neurons. Sci Signal 3, ra22.

121.

Prell

, Lautenschläger

, Grosskreutz

(2013) Calcium-dependent protein folding in amyotrophic lateral sclerosis. Cell Calcium 54, 132–143.

122.

Yang

, Zhong

, Xian

, Yuan

(2015) Hypoxia induces microglia autophagy and neural inflammation injury in focal cerebral ischemia model. Exp Mol Pathol 98, 219–224.

123.

Shenderov

, Riteau

, Yip

, Mayer-Barber

, Oland

, Hieny

, Fitzgerald

, Oberst

, Dillon

, Green

, Cerundolo

, Sher

(2014) Cutting edge: Endoplasmic reticulum stress licenses macrophages to produce mature IL-1β in response to TLR4 stimulation through a caspase-8–and TRIF-dependent pathway. J Immunol 192, 2029–2033.

124.

Misawa

, Takahama

, Kozaki

, Lee

, Zou

, Saitoh

, Akira

(2013) Microtubule-driven spatial arrangement of mitochondria promotes activation of the NLRP3 inflammasome. Nat Immunol 14, 454–460.

125.

Carrano

, Das

(2015) Altered innate immune and glial cell responses to inflammatory stimuli in amyloid precursor protein knockout mice. PLoS One 10, e0140210.

126.

Manabe

, Katayama

, Sato

, Kudo

, Matsuzaki

, Imaizumi

, Tohyama

(2002) The cytosolic inclusion bodies that consist of splice variants that lack exon 5 of the presenilin-2 gene differ obviously from Hirano bodies observed in the brain from sporadic cases of Alzheimer’s disease patients. Neurosci Lett 328, 198–200.

127.

Orešič

, Hyötyläinen

, Herukka

, Sysi-Aho

, Mattila

, Seppänan-Laakso

, Julkunen

, Gopalacharyulu

, Hallikainen

, Koikkalainen

(2011) Metabolome in progression to Alzheimer’s disease. Transl Psychiatr 1, e57.

128.

Zeng

, Chen

, Zhao

(2004) Genistein ameliorates β-amyloid peptide (25–35)-induced hippocampal neuronal apoptosis. Free Radic Biol Med 36, 180–188.

129.

Liu

, Waltz

, Woodruff

, Ouyang

, Israel

, Herrera

, Sarsoza

, Tanzi

, Koo

, Ringman

, Goldstein

, Wagner

, Yuan

(2014) Effect of potent gamma-secretase modulator in human neurons derived from multiple presenilin 1-induced pluripotent stem cell mutant carriers. JAMA Neurol 71, 1481–1489.

130.

Mestizo Gutiérrez

, Herrera Rivero

, Cruz Ramírez

, Hernández

, Aranda-Abreu

(2014) Decision trees for the analysis of genes involved in Alzheimer’s disease pathology. J Theor Biol 357, 21–25.

131.

AL-Dlaeen

, Alashqur

(2014) Using decision tree classification to assist in the prediction of Alzheimer’s disease. pp. Computer Science and Information Technology (CSIT), 2014 6th International Conference on IEEE 122–126.

132.

Bakhtiarizadeh

, Moradi-Shahrbabak

, Ebrahimie

(2014) Transcriptional regulatory network analysis of the over-expressed genes in adipose tissue. Genes Genom 36, 105–117.

133.

Hosseinpour

, Bakhtiarizadeh

, Khosravi

, Ebrahimie

(2013) Predicting distinct organization of transcription factor binding sites on the promoter regions: A new genome-based approach to expand human embryonic stem cell regulatory network. Gene 531, 212–219.

The Zebrafish Equivalent of Alzheimer’s Disease-Associated PRESENILIN Isoform PS2V Regulates Inflammatory and Other Responses to Hypoxic Stress

Abstract

Keywords

INTRODUCTION

MATERIALS AND METHODS

Ethics statement

Zebrafish husbandry, morpholino injection, and exposure to NaN3

Oligonucleotides

Microarray experiment, experimental design, and statistical analysis

Statistics-based subnetwork discovery and regulatory network analysis after loss of PS1IV under mimicry of hypoxia

Subnetwork discovery

Integrative regulatory network construction

Topological analysis of regulatory networks

Validation of network nodes by quantitative real-time PCR (qPCR)

RESULTS

Genes with altered expression after failure to form PS1IV under hypoxia

An overview of our genetic network analysis of transcriptome changes occurring due to failure to form the PS1IV isoform under hypoxia

Significant subnetworks formed by genes over-expressed after failure to form PS1IV under hypoxia

Prediction of a regulatory network underlying genes over-expressed after failure to form PS1IV

Significant subnetworks formed by genes under-expressed after failure to form PS1IV under hypoxia

Prediction of a regulatory network coordinating genes under-expressed after failure to form PS1IV

Topological analysis of regulatory networks

A global regulatory network of both over- and under-expressed genes after failure to induce PS1IV during hypoxia

Validation of network nodes (hubs) by RT-qPCR

A number of PS1IV-regulated subnetworks are also altered in sAD brains

Literature mining-based network analysis of under-expressed genes after failure to form PS1IV highlights human PS2V’s possible involvement in cancer and cardiomyopathy

DISCUSSION

Footnotes

ACKNOWLEDGMENTS

Appendix

References

Zebrafish husbandry, morpholino injection, and exposure to NaN₃