Meta-Analysis of Transcriptome Data Related to Hippocampus Biopsies and iPSC-Derived Neuronal Cells from Alzheimer’s Disease Patients Reveals an Association with FOXA1 and FOXA2 Gene Regulatory Networks

Abstract

Although the incidence of Alzheimer’s disease (AD) is continuously increasing in the aging population worldwide, effective therapies are not available. The interplay between causative genetic and environmental factors is partially understood. Meta-analyses have been performed on aspects such as polymorphisms, cytokines, and cognitive training. Here, we propose a meta-analysis approach based on hierarchical clustering analysis of a reliable training set of hippocampus biopsies, which is condensed to a gene expression signature. This gene expression signature was applied to various test sets of brain biopsies and iPSC-derived neuronal cell models to demonstrate its ability to distinguish AD samples from control. Thus, our identified AD-gene signature may form the basis for determination of biomarkers that are urgently needed to overcome current diagnostic shortfalls. Intriguingly, the well-described AD-related genes APP and APOE are not within the signature because their gene expression profiles show a lower correlation to the disease phenotype than genes from the signature. This is in line with the differing characteristics of the disease as early-/late-onset or with/without genetic predisposition. To investigate the gene signature’s systemic role(s), signaling pathways, gene ontologies, and transcription factors were analyzed which revealed over-representation of response to stress, regulation of cellular metabolic processes, and reactive oxygen species. Additionally, our results clearly point to an important role of FOXA1 and FOXA2 gene regulatory networks in the etiology of AD. This finding is in corroboration with the recently reported major role of the dopaminergic system in the development of AD and its regulation by FOXA1 and FOXA2.

Keywords

Alzheimer’s disease energy metabolism forkhead box proteins gene expression,induced pluripotent stem cells meta-analysis microarray analysis transcription factors

INTRODUCTION

Alzheimer’s disease (AD) is estimated to affect about 35 million of the world’s population and due to continuous increase in our aging population, it is obvious that its prevalence will increase [1]. Epidemiology shows only slight variation in the prevalence with a band from 5% to 7% in most regions of the world—however with some noticeable outliers of about 2% in West Sub-Saharan Africa [2]. A dichotomy of rare genetic variants leading to early-onset AD (EOAD) and more frequent but less malignant variants leading to late-onset (LOAD) has been proposed [3]. The most prominent discovery for LOAD has been the association of a higher risk with the APOE 4 allele [4]. However, the effect is mediated by its action on the amyloid-β product of a variant APP gene. Nonetheless, age and gender remain the major risk factors for sporadic forms of AD.

The interplay of genetic and environmental influences leading to the disease is only partially understood. There is now increasing awareness that age-associated metabolic impairment might also have a contributory effect on LOAD. In our earlier studies on aging in mouse brain, heart, and kidney, we postulated that metabolic stability of gene regulatory networks modulating glutathione metabolism, oxidative phosphorylation, insulin signaling, and inflammation which are biological processes intimately associated with aging are under constraint and should be tightly controlled to attain healthy aging [5, 6]. The metabolic stability theory which is the ability of cells to maintain stable concentrations of reactive oxygen species (ROS) and other critical metabolites is in fact further supported by emerging evidence that associates type 2 diabetes mellitus (T2DM), brain insulin resistance, oxidative stress, and cognitive impairment with AD [7]. More recent data also imply an interplay between metabolic homeostasis and neurogeneration [8].

Pistollato et al. hypothesize that progress in AD research has been hampered by overuse of animal models inadequate for the study of LOAD which accounts for about 95% of AD cases [9]. Demetrius and Driver consider the sporadic late-onset forms of AD a metabolic disease [10]: the amyloid cascade hypothesis postulating the imbalance of amyloid production and clearance due to missense mutations can explain the familial EOAD forms but cannot stringently explain the etiology of sporadic forms as it ignores the impact the brain energy metabolism has on neuronal dynamics. The simplistic view of the brain energy metabolism consisting of glucose as main energy substrate has been refined by the findings that neurons acquire their energy predominantly from oxidative phosphorylation and astrocytes from glycolysis via anaerobically produced lactate [10, 11]. Moreover, a dichotomy of AD pathological changes between the youngest-old (individuals at about 75 years) and the oldest-old (individuals at about 95 years) has been reported in view of the fact that for the youngest-old the severity of dementia correlates very well with neural plaques and neurofibrillary tangles while that correlation is reduced for the oldest-old [12, 13]. However, correlation between cerebral atrophy and dementia consistently is strong for the youngest-old and oldest-old [13]. These observations for the oldest-old are hard to explain with the amyloid-cascade hypothesis but in particular the correlation to brain atrophy is very well in line with the metabolic model. Inter-relations between metabolism, brain atrophy, and age have already been reported in several publications, e.g., by Yoshii et al. [14], and cerebral glucose metabolism measures have even been suggested as AD markers [15]. Besides the age-dependence the metabolic model also referred to as Inverse Warburg Hypothesis can elucidate two further hallmarks of sporadic AD which are in conflict with the amyloid cascade model: the selective vulnerability of neurons in different brain regions and the inverse cancer comorbidity [16].

As a consequence of our fragmentary understanding of the molecular basis of AD, fully functional therapies are not available. Furthermore, unsatisfying accuracy of diagnoses has been reported, e.g., in a large study 39% of clinically diagnosed non-AD dementia patients showed postmortem histopathology consistent with AD [17].

Meta-analyses have been performed related to polymorphisms [18], secreted cytokines [19], and cognitive training [20]. Here, we performed meta-analysis based on hierarchical clustering analysis of transcriptome data of a reliable training set of hippocampus biopsies which led to a gene expression signature distinguishing AD from healthy control samples. This gene expression signature was also applied to various test sets of brain biopsies and induced pluripotent stem cells (iPSC)-derived neuronal cell transcriptome models to demonstrate its ability to distinguish AD from control in various experimental setups. Besides the EOAD PSEN1 mutation brain biopsy dataset from Antonell et al. [21] and the iPSC-derived sporadic AD model from Hossini et al. [22], the biopsy datasets we analyzed are representative of the LOAD form, while the iPSC-based AD models are based on EOAD-related APP- and PSEN-mutations.

Recently, links between the monoaminergic/dopaminergic system and AD were described [23]. The predominant locations of this system (locus coeruleus, raphe nuclei, and tuberomammillary nucleus) are degenerated and thus impair functionality of hippocampus and cortical neurons. Furthermore, genome-wide association studies of AD patients detected polymorphisms in genes from the dopaminergic system, e.g., polymorphisms in Catechol-O-methyltransferase (COMT) are associated with availability of dopamine and, on the other hand, with loss of behavioral control and psychosis in AD patients [24]. Moreover, Ferri et al. [25] and Stott et al. [26] reported that Foxa1 and Foxa2 regulate development and maintenance of dopaminergic neurons in the mouse midbrain. This finding is in corroboration with the observation that FOXA1 and FOXA2 regulate dopaminergic neurons and that the dopaminergic system plays a major role in the development of AD.

MATERIALS AND METHODS

Determining an AD-correlated gene signature from a training set of transcriptomes from hippocampus-derived biopsies

Figure 1 illustrates the scheme of this meta-analysis. A sound dataset of hippocampus biopsies (GSE29378 downloaded from the public repository NCBI GEO) with a relatively high sample size of 72 produced on a state-of-the-art microarray platform (Illumina Beadchip) is used as training set [27]. The correlation of each gene to the AD phenotype is calculated on a binary scale (AD = 1, healthy control = 0) using Spearman correlation. Genes with the ‘n’ most significant correlations are determined based on the test for association between paired samples using the R function cor.test(). This implies that highly correlated and anti-correlated (down-regulated in AD) genes are integrated into the gene signature.

Recalibration of the gene signatureon an alternative platform

In order to place the gene signature on a more robust basis which is independent of the technical platform, it was recalibrated with the dataset GSE36980 downloaded from NCBI GEO [28] on an alternative technical platform - the Affymetrix Human Gene 1.0 ST Array. In this dataset, again the correlation with the AD phenotype was determined employing the approach used for the first dataset. The probesets found to be correlated with the AD phenotype in this probeset (p < 0.05) were intersected with the gene signature from the first step. The result is a reduced gene signature, which, however, is more robust as it works at least for the Affymetrix and the Illumina microarray platforms.

Cluster analysis of brain biopsies

The re-confirmed gene signature was employed for cluster analysis in several test sets of brain biopsy-derived microarray data. Test datasets for human brain biopsies were downloaded from NCBI GEO for Affymetrix and Illumina microarray platforms. These comprise the datasets: GSE4757 [29], GSE26927 [30], GSE39420 [21]. In detail, the raw data were processed via R/Bioconductor [31] packages affy [32], lumi [33] and oligo [34] and normalized using a platform-specific method (rma for Affymetrix, quantile normalization for Illumina). The probes or probesets matching genes from the gene signature are extracted from the normalized test dataset and subjected to cluster analysis via the R function heatmap.2.

Cluster analysis of iPSC-based AD models

As a complement to the brain biopsies, cluster analysis was also performed for several test sets of iPSCs-based AD models microarray data for the re-confirmed gene signature. Test datasets for iPSCs-based AD models were downloaded from NCBI GEO for Affymetrix and Illumina microarray platforms. These comprise the datasets: GSE28379 [35], GSE43326 [36], GSE34879 [37], and GSE42492 [22]. These test datasets were processed in the same way as described above for the brain biopsy test datasets.

Transcription factor analysis

Transcription factors were analyzed using the oPOSSUM-3 tool [38]. Single Site Analysis was performed on a locally installed oPOSSUM database (downloaded in June 2014) in order to detect over-represented conserved transcription factor binding sites within the AD gene signature. Parameters were set to the following values: species “human”, 2000 base pairs upstream and downstream each, use only JASPAR Transcription Factor Binding Site (TFBS) profile matrices which belong to the tax group “vertebrates”, a minimum relative TFBS position weight matrix (PWM) score of 0.85 and a minimum information content (specificity) of JASPAR TFBS profile matrices of 8.

For the network plot, we used the R package network [39] to connect transcription factors with an oPOSSUM Z-score >3 with their target genes. The oPOSSUM Z-score accounts for an enrichment of TFBS in the investigated up- and downstream regions of the set of genes compared to the background. Genes and transcription factors were distinguished by red (white in b/w) circles for genes and green (grey in b/w) circles with sizes corresponding to the Z-scores for the transcription factors.

Pathway and gene ontology over-representation analysis

KEGG pathways were analyzed via the hypergeometric test from the R package [40]. The qvalue package was employed to adjust for multiple testing [41]. Over-represented gene ontologies were determined employing the GOstats package [42] from the R/Bioconductor environment [31].

Gene expression analysis

Quantitative real-time polymerase chain reaction (qRT-PCR) was carried out as described [22]. Briefly the data presented fold induction based on the ΔΔCt calculation over fetal brain RNA for the adult and AD brain and undifferentiated H9 or AD5 cells for the iPSC model. Only one neuronal differentiation of a single well of embryonic stem cell line H9 and sAD iPSC line AD5 was used for this analysis.

RESULTS

The meta-analysis of the datasets listed in Table 1 was performed according to the flow chart illustrated in Fig. 1. We started with the determination of genes most significantly correlated or reversely correlated with the AD phenotype in the dataset from Miller et al.Figure 2 shows that the gene signature consisting of the 40 most significantly correlated or anti-correlated Illumina probes (mapping to 34 distinct gene symbols, Table 2) enables a good separation into two main clusters. One cluster (healthy control cluster) contains only healthy control brain biopsies while the other cluster (AD cluster) contains 35 (87.5%) AD brain biopsies and 5 outliers in 3 groups (one group contains two biopsies from the same patient). Table 2 lists these 34 genes most significantly correlated with the AD phenotype. 32 genes correlate positively (most significant: GEM, S100A6, YAP1, and PFKFB3) while only two genes (FXYD5 and WDFC1) are anti-correlated.

The good separation of AD and healthy controls by the gene signature is additionally demonstrated by the principal component analysis (PCA) plotted in Supplementary Fig. 1. While we used a supervised clustering approach (filtering genes by their correlation to the AD phenotype), unsupervised clustering analyses revealed no clear separation of AD and control clusters. For comparison, Supplementary Fig. 2 shows the result of a clustering analysis of genes from the KEGG AD pathway, and Supplementary Fig. 3 shows the result of a clustering analysis of genes from the secretome on the training dataset GSE29378. The results of the reconfirmation of the gene signature via the GEO dataset GSE36980 from Hokoma et al. [28] is presented in Fig. 3. The adjusted gene signature then is reduced to 29 genes.

Intriguingly, neither the well-described AD-related genes APP and APOE nor the later discovered presenilin1/2 (PSEN1, PSEN2) were within the signature because their gene expression profiles showed a lower correlation to the disease phenotype than genes from the signature. This is in line with the distinct characteristics of the disease as EOAD or LOAD or with or without genetic predisposition and the low frequency of the AD-causing mutations in APP, APOE, PSEN1, and PSEN2. To investigate the gene signature and its relevance to the AD phenotype in more detail, associated pathways, gene ontologies, and transcription factors were analyzed. Finally, we demonstrated that our gene signature could better distinguish AD from healthy controls than a collection of genes encoding secretory factors. Thus, it may be a good base for determination of biomarkers that are urgently needed to overcome the current shortfall in accurate diagnostics.

Cluster analysis of hippocampus biopsy datasets

The good correlation of the clustering of dataset GSE36980 from Hokama et al. [28] with the AD phenotype is illustrated by Fig. 3. Also with other hippocampus datasets, the gene signature performed very well. Using the gene signature for cluster analysis of the hippocampus biopsy dataset GSE39420 with EOAD and PSEN1 mutations from Antonellet al. [21] facilitated a good separation of one pure AD cluster and one healthy control clusters containing two AD cases (Supplementary Fig. 4). Additionally, the dataset GSE28146 from Blalock et al. [44] derived from formalin-fixed, paraffin-embedded (FFPE) hippocampus biopsies was subjected to cluster analysis and enabled separation into one AD cluster and one mixed cluster with healthy controls and not-severe AD cases (Supplementary Fig. 5). The results also showed that in most datasets, there was a high level of variation between the various transcript variants and not all correlated with the AD phenotype. Using the variants with the highest correlation to the AD phenotype unveiled the results shown in (Supplementary Fig. 5).

Cluster analysis of entorhinal cortex biopsy datasets

Cluster analysis of entorhinal cortex biopsies (Supplementary Figs. 6 and 7) did not yield as good results as for hippocampus biopsies (Supplementary Figs. 4 and 5). Cluster analysis with the gene signature on the test set of entorhinal cortex biopsies AD dataset GSE26927 from Durrenberger et al. [30] resulted in three clusters one pure AD cluster containing only two samples and two incoherent clusters containing AD as well as control samples (Supplementary Fig. 6). Red bars at the bottom indicate healthy control while blue bars indicate AD. The dataset GSE4757 from Dunckley et al. [29] consists of healthy controls and entorhinal cortex biopsies with neurofibrillary tangles which are considered an early event in AD pathology (Supplementary Fig. 7). Cluster analysis with the gene signature on this test set resulted in three clusters that were more or less mixed up and only had a few coherent sub-clusters. Red bars at the bottom indicate healthy control while blue bars indicate neurofibrillary tangles. The inferior performance of the gene signature on entorhinal cortex biopsies might be due to its specificity to hippocampus. AD-related gene expression and regulatory networks exhibit distinct behavior between hippocampus and entorhinal cortex.

Cluster analysis of iPCS-derived AD model datasets

In comparison to brain biopsy-derived datasets, a major disadvantage of iPSC-derived AD models is the smaller sample size of the datasets due to the complex process of patient-specific iPSC-derivation and differentiation into neuronal cells. On the other hand, the advantages of the iPSC-based models are obvious: patient-specific in-vitro models of AD can be perturbed in order to test new hypotheses. A summary of the analyzed iPSC-derived AD models with their mutationand the neural protocols is provided in Table 1B.

The cluster analysis with the gene signature for the dataset GSE43326 from Kondo et al. [36] with relatively high sample size for iPSC-AD-models showed good results: it separated one familial AD (APP mutations) from the healthy control samples (APP wild type) as demonstrated in Fig. 4.

As mentioned above, the small sample size has major impact on the results of the cluster analysis: in most datasets it is so low that a reasonable clustering is barely possible. Dataset GSE28379 from Yagi et al. [35] contains two iPSC-derived AD model cell lines carrying PSEN2 mutations and two control iPSC-derived neuron samples—one of them from a Parkinson’s disease (PD) patient (Supplementary Fig. 8). Here, one PSEN2 mutation sample clusters with the PD sample while the other PSEN2 mutation sample clusters separately from the other samples. In dataset GSE34879 from Israel et al. [37], there is only one AD patient-specific iPSC sample and one iPSC-derived neuron sample. Nevertheless, these samples showed some differences with the healthy control samples in the cluster analysis of the gene signature (Supplementary Fig. 9). The dataset from Hossini et al. [22] comprised two iPSC-derived neuron cell cultures derived from a single sporadic AD (sAD) and one healthy control derived from embryonic stem cell line H9 (Supplementary Fig. 10). Applying the gene signature to this test set clustered the two AD samples separate from the healthy control revealing some prominent differences between the two AD samples and the control, e.g., the S100 calcium binding protein A6 (S100A6). In summary, our AD gene signature could separate AD samples from healthy controls in the iPSC-based AD models, with the exception of the Yagi et al. (GSE28379, [35]) dataset which was probably masked by the similarity of the PD control to the AD phenotype.

Characterization of the gene signature

The gene signature was further investigated with the aim to identify associated gene regulatory networks employing gene set and transcription factor analyses. Table 3 shows the 20 most significant terms from gene ontology over-representation analysis of the AD gene signature. These gene ontology terms point to a major role of regulation of glial cells and development of nervous system. Table 4 shows the results of a KEGG pathway over-representation analysis. Protein processing in endoplasmatic reticulum, amyotrophic lateral sclerosis and Neurotrophin signaling pathway are found as significant with a p-value <0.05 and at least two genes in the pathway. KEGG pathway Protein processing in endoplasmic reticulum (Supplementary Fig. 11) obviously is related to AD associated misfolded amyloid-β (Aβ) and tau proteins. Furthermore, within this pathway involvement of the ubiquitin ligase complex via FBP (FBXO2) and induction of apoptosis (BCL2 and ASK1:MAP3K5) are fundamentalsof AD.

Figure 5 shows the network resulting from a transcription factor analysis with the gene signature using the oPOSSUM-3 tool [38]. Factors from the FOX family (FOXA1, FOXA2, FOXD1, FOXO3, FOXQ1, FOXI1) have the highest significance (see also Supplementary Fig. 12, Supplementary Table 1). The graphs point to the Forkhead/winged helix transcription factors that regulate the gene signature. Transcription factors FOXA1 and FOXA2 and others from the Forkhead/winged helix family were most significant with a z-score greater than mean plus two standard deviations. Foxa1 and Foxa2 have been reported to regulate development and maintenance of mouse dopaminergic neurons [25, 26] and the dopaminergic system in midbrain has been associated with AD [23]. The FOXA1 and FOXA2 mediated regulation of dopaminergic neurons has been reported for the midbrain whereas the hippocampus is part of the forebrain. Thus the question emerged if the midbrain-located dopaminergic system also affects the hippocampus. Influences of the dopaminergic system on the hippocampus have been described [45]: dopaminergic neurons promote hippocampal reactivation and axons expressing dopaminergic markers have been found within the hippocampus. Our results suggest that the gene regulatory network orchestrated by FOXA1 and FOXA2 not only affects the dopaminergic neurons in the midbrain but also in the hippocampus where dopaminergic axons are located. However, this hypothesis needs further validation using an AD-iPSC cell model.

To further investigate if there is contribution of the FOX genes to AD, we analyzed expression of several FOX genes by qualitative real-time PCR analysis as shown in Fig. 6. We analyzed the genes encoding FOXA2 (see Fig. 5 and Supplementary Fig. 12), FOXD1 (involved in diencephalon development [46]; see Supplementary Fig. 12), FOXF1, and FOXG1 (associated with forebrain interneuronal development [47]). For a general overview of these FOX genes in neuronal patterning, we documented their expression in healthy adult and AD brain (normalized to fetal brain; all commercial RNA from Amsbio^®), Fig. 6A. In the AD-affected brain, FOXA2 is slightly downregulated, beyond that the medial ganglionic eminence-progenitor marker FOXG1 is barely detectable as expressed in both healthy adult and AD brain. Further analysis in the iPSC models of Hossini et al. [22], reveals a general depression of all FOX genes in the neural network derived from the sAD patient with the exception of FOXA2, which seems to be upregulated (Fig. 6B).Furthermore, in Supplementary Fig. 13, significant transcription factors FOXA1, FOXA2, FOXO3, FOXD1, and PAX6 were analyzed together with genes from the AD signature in the iPSC-based AD model from Hossini et al. [22]. FOXO3 shows the highest expression over all samples and clusters together with PAX6. FOXA1 and FOXA2 have high differences between AD and healthy control and are in the same cluster together with GLIS3, CXCR4, NUPR1, and EMP1.

Consideration of AD as metabolic disease

As described extensively in our introduction, it is evident that metabolism should be included when studying the etiology of AD. To this end, we carried out analyses targeting metabolic processes within the current datasets.

Table 5 lists the over-representation of central metabolic KEGG pathways in the analyzed AD datasets tested via hypergeometric test. We found TCA-cycle significant (p < 0.05) in the dataset from Hokama et al. [28] as well as oxidative phosphorylation which also is significant in Blalock et al. [44], while insulin signaling is significant in Hokama et al. [28] and also in the iPSC AD model from Kondo et al. [36]. Additionally, other metabolism-related pathways are detected as significant, for example, purine metabolism (Supplementary Table 2). Furthermore, gene ontology over-representation analysis demonstrates that response to stress, regulation of cellular metabolic processes, and reactive oxygen species are significantly over-represented in the AD gene signature (Supplementary Table 3). In summary, these findings show that several metabolic processes and particularly oxidative stress are associated and intimately involved in the pathogenesis of AD.

DISCUSSION

With our meta-analysis we found a gene signature of 34 distinct genes. For most of them an association with AD has been reported (Table 2). Trying to find common functionality in this gene set via literature search identified the TGFβ pathway, which is represented by the genes SMAD9 and TGFBR3 and the genes BCL2, FBXO2, and MAP3K5 related to protein processing. Tesseur et al. described that a deficiency in TGFβ signaling can initiate AD pathology [48]. Paradis et al. found that Aβ peptide of AD downregulates Bcl-2 [49]. Furthermore the astrocytic calcium/zinc binding gene S100A6 which is part of our AD signature has been reported to be overexpressed in AD and in PS1/APP transgenic mouse models [50]. Another striking gene from the AD signature was GFAP. Its expression differed from the brain-specific- to the iPSC-derived AD models. It seems that neuronal loss and the inflamed region of the AD-affected brain induces GFAP gene expression. Typically, an increased expression of GFAP represents astroglia activation and gliosis, especially during neurodegeneration [51]. In the iPSC-derived and AD neuronal network models, GFAP seemed to be expressed at a lower level than in the brain biopsies. However, in almost all AD models, e.g., Kondo et al. (Fig. 4), the expression of GFAP is higher in AD patients compared to healthy control individuals.

Our results clearly point to an important role of FOXA1 and FOXA2 in regulating biological processes, which are significantly dysregulated in AD. This finding is in corroboration with reported observations implying that Foxa1 and Foxa2 regulate development of dopaminergic neurons and that the dopaminergic system plays a major role in the development of AD. Thus, the results from this analysis have unveiled a gene regulatory network controlled by FOXA1 and FOXA2. This gene network is involved in development and maintenance of dopaminergic neurons and is associated with AD. The influence of FOXA1 and FOXA2 on the dopaminergic neurons has been described only for the midbrain but not for hippocampus where we found the dysregulation of these transcription factors. However, McNamara et al. previously discussed the influence of dopaminergic neurons on the hippocampus [45] revealing that dopaminergic neurons promote hippocampal reactivation and that axons expressing dopaminergic markers have been found in hippocampus. Furthermore we have elucidated the FOX gene expression in the AD-affected brain and in our iPSC model. FOXA1/FOXA2 are regulating ventral midbrain neural development [25], but beyond that FOXA2 directly induce sonic hedgehog, a key player in forebrain development of inhibitory interneurons. Interneurons express FOXG1 upon progenitor state [47] and are impaired in AD patients [52].

Our results highlight the important role of metabolic processes, particularly oxidative phosphorylation and oxidative stress in the pathology of AD. Furthermore, the FOX transcription factors have already been reported to play a major role in metabolism and induction of oxidative stress. Interestingly, Gao et al. find an inverse regulation of metabolism-related and neuronal genes in Foxa1 and Foxa2 compound knockdown mouse beta cells [53]. Kittappa et al. already discussed that impairment of mitochondrial function via oxidative stress induced by FOXA2 leads to neuronal loss [54]. However, this study was based on dopaminergic neurons in PD but similar mechanisms may also exist in AD. An important difference to PD might be the contribution of other FOX family transcription factors such as the FOXG and FOXO families.

For FOXO transcription factors which we also found significantly enriched in the Opossum analysis, Manolopoulos et al. reported an involvement in the pathogenesis of AD and in insulin resistance via oxidative stress [55]. They hypothesize that ROS activating c-Jun N-terminal kinases (JNK) and inhibiting Wnt-signaling may lead to amyloid-β plaques and tau protein phosphorylation. Additionally, inhibition of Wnt-signaling may further activate FOXO proteins thus leading to a feedback loop of oxidative stress, insulin resistance, ROS, and neurodegeneration. In summary, we hypothesize that the FOXA- and FOXO-mediated mechanisms leading to neuronal loss in AD (Fig. 7) might be as follows: FOXA is involved in the aging-induced dysregulation of metabolic processes including glucose homeostasis [56], carbohydrate metabolism [53], and oxidative phosphorylation [54, 57]. This metabolic dysregulation leads to oxidative stress, the abundant increase of ROS. ROS activates JNK through MAP3K5 (ASK1) [58, 59]. JNK regulates several processes including a cascade of JUNB and BCL2 connected to mitochondrial stress and FOXO-mediated processes involved in neuronal loss: (i) FOXO3 regulates the pro-apoptotic BCL2 [60]. (ii) Aberrant neuronal cell cycle re-entry and dysregulation by FOXO together with its regulators JNK and PI3K-PKB [61] eventually results in neuronal loss [10 , 63].

Finally, we propose an AD-iPSC cell model to test our hypothesis, which implies an important role of FOXA1, FOXA2, and FOXO gene regulatory networks in the etiology of AD.

Footnotes

ACKNOWLEDGMENTS

We acknowledge support from the European Union funding/FP7 (FP7/2007-2013)/Grant Agreement n° 305299 (AgedBrainSYSBIO) and from the medical faculty of the Heinrich-Heine Universität Düsseldorf.

Authors’ disclosures available online ().

Appendix

References

James

, Doraiswamy

, Borges-Neto

(2015) PET imaging of tau pathology in Alzheimer’s disease and tauopathies. Front Neurol 6, 38.

Prince

, Bryce

, Albanese

, Wimo

, Ribeiro

, Ferri

(2013) The global prevalence of dementia: A systematic review and metaanalysis. Alzheimers Dement 9, 63–75.e2.

Bertram

, McQueen

, Mullin

, Blacker

, Tanzi

(2007) Systematic meta-analyses of Alzheimer disease genetic association studies: The AlzGene database. Nat Genet 39, 17–23.

Corder

, Saunders

, Strittmatter

, Schmechel

, Gaskell

, Small

, Roses

, Haines

, Pericak-Vance

(1993) Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer’s disease in late onset families. Science 261, 921–923.

Brink

, Demetrius

, Lehrach

, Adjaye

(2009) Age-related transcriptional changes in gene expression in different organs of mice support the metabolic stability theory of aging. Biogerontology 10, 549–564.

Brink

, Regenbrecht

, Demetrius

, Lehrach

, Adjaye

(2009) Activation of the immune response is a key feature of aging in mice. Biogerontology 10, 721–734.

de la Monte

, Wands

(2008) Alzheimer’s disease is type 3 diabetes–evidence reviewed. J Diabetes Sci Technol 2, 1101–1113.

Ngo

, Steyn

(2015) The interplay between metabolic homeostasis and neurodegeneration: Insights into the neurometabolic nature of amyotrophic lateral sclerosis. Cell Regen (Lond) 4, 5.

Pistollato

, Cavanaugh

, Chandrasekera

(2015) A human-based integrated framework for Alzheimer’s disease research. J Alzheimers Dis 47, 857–868.

10.

Demetrius

, Driver

(2013) Alzheimer’s as a metabolic disease. Biogerontology 14, 641–649.

11.

Pellerin

, Magistretti

(2004) Neuroenergetics: Calling upon astrocytes to satisfy hungry neurons. Neuroscientist 10, 53–62.

12.

Haroutunian

, Schnaider-Beeri

, Schmeidler.

, Wysocki

, Purohit

, Perl

, Libow

, Lesser

, Maroukian

, Grossman

(2008) Role of the neuropathology of Alzheimer disease in dementia in the oldest-old. Arch Neurol 65, 1211–1217.

13.

Savva

, Wharton

, Ince

, Forster

, Matthews

, Brayne

(2009) Age, neuropathology, and dementia. N Engl J Med 360, 2302–2309.

14.

Yoshii

, Barker

, Chang

, Loewenstein

, Apicella

, Smith

, Boothe

, Ginsberg

, Pascal

, Duara

(1988) Sensitivity of cerebral glucose metabolism to age, gender, brain volume, brain atrophy, and cerebrovascular risk factors. J Cereb Blood Flow Metab 8, 654–661.

15.

Mosconi

, Sorbi

, de Leon

, Li

, Nacmias

, Myoung

, Tsui

, Ginestroni

, Bessi

, Fayyazz

, Caffarra

, Pupi

(2006) Hypometabolism exceeds atrophy in presymptomatic early-onset familial Alzheimer’s disease. J Nucl Med 47, 1778–1786.

16.

Demetrius

, Magistretti

, Pellerin

(2015) Alzheimer’s disease: The amyloid hypothesis and the Inverse Warburg effect. Front Physiol 5, 522.

17.

Beach

, Monsell

, Phillips

, Kukull

(2012) Accuracy of the clinical diagnosis of Alzheimer disease at National Institute on Aging Alzheimer Disease Centers, 2005-2010. J Neuropathol Exp Neurol 71, 266–273.

18.

Lambert

, Ibrahim-Verbaas

, Harold

, Naj

, Sims

, Bellenguez

, DeStafano

, Bis

, Beecham

, Grenier-Boley

, Russo

, Thorton-Wells

, Jones

, Smith

, Chouraki

, Thomas

, Ikram

, Zelenika

, Vardarajan

, Kamatani

, Lin

, Gerrish

, Schmidt

, Kunkle

, Dunstan

, Ruiz

, Bihoreau

, Choi

, Reitz

, Pasquier

, Cruchaga

, Craig

, Amin

, Berr

, Lopez

, De Jager

, Deramecourt

, Johnston

, Evans

, Lovestone

, Letenneur

, Morón

, Rubinsztein

, Eiriksdottir

, Sleegers

, Goate

, Fiévet

, Huentelman

, Gill

, Brown

, Kamboh

, Keller

, Barberger-Gateau

, McGuiness

, Larson

, Green

, Myers

, Dufouil

, Todd

, Wallon

, Love

, Rogaeva

, Gallacher.

, St George-Hyslop

, Clarimon

, Lleo

, Bayer

, Tsuang

, Yu

, Tsolaki

, Bossù

, Spalletta

, Proitsi

, Collinge

, Sorbi

, Sanchez-Garcia

, Fox

, Hardy

, Deniz Naranjo

, Bosco

, Clarke

, Brayne

, Galimberti

, Mancuso

, Matthews

, European Alzheimer’s Disease Initiative (EADI), Genetic and Environmental Risk in Alzheimer’s Disease, Alzheimer’s, Disease Genetic Consortium, Cohorts for Heart and Aging Research in Genomic Epidemiology, Moebus

, Mecocci

, Del Zompo

, Maier

, Hampel

, Pilotto

, Bullido

, Panza

, Caffarra

, Nacmias

, Gilbert

, Mayhaus

, Lannefelt

, Hakonarson

, Pichler

, Carrasquillo

, Ingelsson

, Beekly

, Alvarez

, Zou

, Valladares

, Younkin

, Coto

, Hamilton-Nelson

, Gu

, Razquin

, Pastor

, Mateo

, Owen

, Faber

, Jonsson

, Combarros

, O’Donovan

, Cantwell

, Soininen

, Blacker

, Mead

, Mosley

, Bennett

, Harris

, Fratiglioni

, Holmes

, de Bruijn

, Passmore

, Montine

, Bettens

, Rotter

, Brice

, Morgan

, Foroud

, Kukull

, Hannequin

, Powell

, Nalls

, Ritchie

, Lunetta

, Kauwe

, Boerwinkle

, Riemenschneider

, Boada

, Hiltuenen

, Martin

, Schmidt

, Rujescu

, Wang

, Dartigues

, Mayeux

, Tzourio

, Hofman

, Nöthen

, Graff

, Psaty

, Jones

, Haines

, Holmans

, Lathrop

, Pericak-Vance

, Launer

, Farrer

, van Duijn

, Van Broeckhoven

, Moskvina

, Seshadri

, Williams

, Schellenberg

, Amouyel

(2013) Meta-analysis of 74,046 individuals identifies 11 new susceptibility loci for Alzheimer’s disease. Nat Genet 45, 1452–1458.

19.

Swardfager

, Lanctôt

, Rothenburg

, Wong

, Cappell

, Herrmann

(2010) A meta-analysis of cytokines in Alzheimer’s disease. Biol Psychiatry 68, 930–941.

20.

Sitzer

, Twamley

, Jeste

(2006) Cognitive training in Alzheimer’s disease: A meta-analysis of the literature. Acta Psychiatr Scand 114, 75–90.

21.

Antonell

, Lladó

, Altirriba

, Botta-Orfila

, Balasa

, Fernández

, Ferrer

, Sánchez-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.

22.

Hossini

, Megges

, Prigione

, Lichtner

, Toliat

, Wruck

, Schröter

, Nuernberg

, Kroll

, Makrantonaki

, Zouboulis

, Zoubouliss

, Adjaye

(2015) Induced pluripotent stem cell-derived neuronal cells from a sporadic Alzheimer’s disease donor as a model for investigating AD-associated gene regulatory networks. BMC Genomics 16, 84.

23.

Trillo

, Das

, Hsieh

, Medina

, Moghadam

, Lin

, Dang

, Sanchez

, De Miguel

, Ashford

, Salehi

(2013) Ascending monoaminergic systems alterations in Alzheimer’s disease translating basic science into clinical care. Neurosci Biobehav Rev 37, 1363–1379.

24.

Borroni

, Agosti

, Archetti

, Costanzi

, Bonomi

, Ghianda

, Lenzi

, Caimi

, Luca

, Padovani

(2004) Catechol-O-methyltransferase gene polymorphism is associated with risk of psychosis in Alzheimer disease. Neurosci Lett 370, 127–129.

25.

Ferri

ALM

, Lin

, Mavromatakis

, Wang

, Sasaki

, Whitsett

, Ang

S-L

(2007) Foxa1 and Foxa2 regulate multiple phases of midbrain dopaminergic neuron development in a dosage-dependent manner. Dev Camb Engl 134, 2761–2769.

26.

Stott

SRW

, Metzakopian

, Lin

, Kaestner

, Hen

, Ang

S-L

(2013) Foxa1 and foxa2 are required for the maintenance of dopaminergic properties in ventral midbrain neurons at late embryonic stages. J Neurosci 33, 8022–8034.

27.

Miller

, Woltjer

, Goodenbour

, Horvath

, Geschwind

(2013) Genes and pathways underlying regional and cell type changes in Alzheimer’s disease. Genome Med 5, 48.

28.

Hokama

, Oka

, Leon

, Ninomiya

, Honda

, Sasaki

, Iwaki

, Ohara

, Sasaki

, LaFerla

, Kiyohara

, Nakabeppu

(2014) Altered expression of diabetes-related genes in Alzheimer’s disease brains: The Hisayama study. Cereb Cortex 24, 2476–2488.

29.

Dunckley

, Beach

, Ramsey

, Grover

, Mastroeni

, Walker

, LaFleur

, Coon

, Brown

, Caselli

, Kukull

, Higdon

, McKeel

, Morris

, Hulette

, Schmechel

, Reiman

, Rogers

, Stephan

(2006) Gene expression correlates of neurofibrillary tangles in Alzheimer’s disease. Neurobiol Aging 27, 1359–1371.

30.

Durrenberger

, Fernando

, Magliozzi

, Kashefi

, Bonnert

, Ferrer

, Seilhean

, Nait-Oumesmar

, Schmitt

, Gebicke-Haerter

, Falkai

, Grünblatt

, Palkovits

, Parchi

, Capellari

, Arzberger

, Kretzschmar

, Roncaroli

, Dexter

, Reynolds

(2012) Selection of novel reference genes for use in the human central nervous system: A BrainNet Europe Study. Acta Neuropathol (Berl) 124, 893–903.

31.

Gentleman

, Carey

, Bates

, Bolstad

, Dettling

, Dudoit

, Ellis

, Gautier

, Ge

, Gentry

, Hornik

, Hothorn

, Huber

, Iacus

, Irizarry

, Leisch

, Li

, Maechler

, Rossini

, Sawitzki

, Smith

, Smyth

, Tierney

, Yang

JYH

, Zhang

(2004) Bioconductor: Open software development for computational biology and bioinformatics. Genome Biol 5, R80.

32.

Gautier

, Cope

, Bolstad

, Irizarry

(2004) affy–analysis of Affymetrix GeneChip data at the probe level. Bioinformatics 20, 307–315.

33.

, Kibbe

, Lin

(2008) lumi: A pipeline for processing Illumina microarray. Bioinformatics 24, 1547–1548.

34.

Carvalho

, Irizarry

(2010) A framework for oligonucleotide microarray preprocessing. Bioinformatics 26, 2363–2367.

35.

Yagi

, Ito

, Okada

, Akamatsu

, Nihei

, Yoshizaki

, Yamanaka

, Okano

, Suzuki

(2011) Modeling familial Alzheimer’s disease with induced pluripotent stem cells. Hum Mol Genet 20, 4530–4539.

36.

Kondo

, Asai

, Tsukita

, Kutoku

, Ohsawa

, Sunada

, Imamura

, Egawa

, Yahata

, Okita

, Takahashi

, Asaka

, Aoi

, Watanabe

, Kadoya

, Nakano

, Watanabe

, Maruyama

, Hori

, Hibino

, Choshi

, Nakahata

, Hioki

, Kaneko

, Naitoh

, Yoshikawa

, Yamawaki

, Suzuki

, Hata

, Ueno

S-I

, Seki

, Kobayashi

, Toda

, Murakami

, Irie

, Klein

, Mori

, Asada

, Takahashi

, Iwata

, Yamanaka

, Inoue

(2013) Modeling Alzheimer’s disease with iPSCs reveals stress phenotypes associated with intracellular Aβ and differential drug responsiveness. Cell Stem Cell 12, 487–496.

37.

Israel

, Yuan

, Bardy

, Reyna

, Mu

, Herrera

, Hefferan

, Van Gorp

, Nazor

, Boscolo

, Carson

, Laurent

, Marsala

, Gage

, Remes

, Koo

, Goldstein

LSB

(2012) Probing sporadic and familial Alzheimer’s disease using induced pluripotent stem cells. Nature 482, 216–220.

38.

Kwon

, Arenillas

, Worsley Hunt

, Wasserman

(2012) oPOSSUM-3: Advanced analysis of regulatory motif over-representation across genes or ChIP-Seq datasets. G3 (Bethesda) 2, 987–1002.

39.

Butts

(2008) network: A package for managing relational data in R. J Stat Softw 24, doi: 10.18637/jss.v024.i02

40.

Ihaka

, Gentleman

(1996) R: A language for data analysis and graphics. J Comput Graph Stat 5, 299–314.

41.

Storey

(2002) A direct approach to false discovery rates. J R Stat Soc Ser B Stat Methodol 64, 479–498.

42.

Falcon

, Gentleman

(2007) Using GOstats to test gene lists for GO term association. Bioinformatic 23, 257–258.

43.

Bertram

, Lill

, Tanzi

(2010) The genetics of Alzheimer disease: Back to the future. Neuron 68, 270–281.

44.

Blalock

, Buechel

, Popovic

, Geddes

, Landfield

(2011) Microarray analyses of laser-captured hippocampus reveal distinct gray and white matter signatures associated with incipient Alzheimer’s disease. J Chem Neuroanat 42, 118–126.

45.

McNamara

, Tejero-Cantero

, Trouche

, Campo-Urriza

, Dupret

(2014) Dopaminergic neurons promote hippocampal reactivation and spatial memory persistence. Nat Neurosci 17, 1658–1660.

46.

Herrera

, Marcus

, Li

, Williams

, Erskine

, Lai

, Mason

(2004) Foxd1 is required for proper formation of the optic chiasm. Dev Camb Engl 131, 5727–5739.

47.

Nicholas

, Chen

, Tang

, Southwell

, Chalmers

, Vogt

, Arnold

, Chen

Y-JJ

, Stanley

, Elefanty

, Sasai

, Alvarez-Buylla

, Rubenstein

JLR

, Kriegstein

(2013) Functional maturation of hPSC-derived forebrain interneurons requires an extended timeline and mimics human neural development. Cell Stem Cell 12, 573–586.

48.

Tesseur

, Zou

, Esposito

, Bard

, Berber

, Can

, Lin

, Crews

, Tremblay

, Mathews

, Mucke

, Masliah

, Wyss-Coray

(2006) Deficiency in neuronal TGF-beta signaling promotes neurodegeneration and Alzheimer’s pathology. J Clin Invest 116, 3060–3069.

49.

Paradis

, Douillard

, Koutroumanis

, Goodyer

, LeBlanc

(1996) Amyloid beta peptide of Alzheimer’s disease downregulates Bcl-2 and upregulates bax expression in human neurons. J Neurosci 16, 7533–7539.

50.

Boom

, Pochet

, Authelet

, Pradier

, Borghgraef

, Van Leuven

, Heizmann

, Brion

J-P

(2004) Astrocytic calcium/zinc binding protein S100A6 over expression in Alzheimer’s disease and in PS1/APP transgenic mice models. Biochim Biophys Acta 1742, 161–168.

51.

Heppner

, Ransohoff

, Becher

(2015) Immune attack: The role of inflammation in Alzheimer disease. Nat RevNeurosci 16, 358–372.

52.

Levenga

, Krishnamurthy

, Rajamohamedsait

, Wong

, Franke

, Cain

, Sigurdsson

, Hoeffer

(2013) Tau pathology induces loss of GABAergic interneurons leading to altered synaptic plasticity and behavioral impairments. Acta Neuropathol Commun 1, 34.

53.

Gao

, Le Lay

, Qin

, Doliba

, Schug

, Fox

, Smirnova

, Matschinsky

, Kaestner

(2010) Foxa1 and Foxa2 maintain the metabolic and secretory features of the mature beta-cell. Mol Endocrinol 24, 1594–1604.

54.

Kittappa

, Chang

, Awatramani

, McKay

RDG

(2007) The foxa2 gene controls the birth and spontaneous degeneration of dopamine neurons in old age. PLoS Biol 5, e325.

55.

Manolopoulos

, Klotz

L-O

, Korsten

, Bornstein

, Barthel

(2010) Linking Alzheimer’s disease to insulin resistance: The FoxO response to oxidative stress. Mol Psychiatry 15, 1046–1052.

56.

Friedman

, Kaestner

(2006) The Foxa family of transcription factors in development and metabolism. Cell Mol Life Sci 63, 2317–2328.

57.

Vatamaniuk

, Gupta

, Lantz

, Doliba

, Matschinsky

, Kaestner

(2006) Foxa1-deficient mice exhibit impaired insulin secretion due to uncoupled oxidative phosphorylation. Diabetes 55, 2730–2736.

58.

Tobiume

, Matsuzawa

, Takahashi

, Nishitoh

, Morita

, Takeda

, Minowa

, Miyazono

, Noda

, Ichijo

(2001) ASK1 is required for sustained activations of JNK/p38 MAP kinases and apoptosis. EMBO Rep 2, 222–228.

59.

Kadowaki

, Nishitoh

, Urano

, Sadamitsu

, Matsuzawa

, Takeda

, Masutani

, Yodoi

, Urano

, Nagano

, Ichijo

(2005) Amyloid beta induces neuronal cell death through ROS-mediated ASK1 activation. Cell Death Differ 12,19–24.

60.

Dijkers

, Medema

, Lammers

, Koenderman

, Coffer

(2000) Expression of the pro-apoptotic Bcl-2 family member Bim is regulated by the forkhead transcription factor FKHR-L1. Curr Biol 10, 1201–1204.

61.

Eijkelenboom

, Burgering

BMT

(2013) FOXOs: Signalling integrators for homeostasis maintenance. Nat Rev Mol Cell Biol 14, 83–97.

62.

Yuan

, Becker

EBE

, Merlo

, Yamada

, DiBacco

, Konishi

, Schaefer

, Bonni

(2008) Activation of FOXO1 by Cdk1 in cycling cells and postmitotic neurons. Science 319, 1665–1668.

63.

Zhu

, Lee

, Perry

, Smith

(2007) Alzheimer disease, the two-hit hypothesis: An update. Biochim Biophys Acta 1772, 494–502.

64.

Orcholski

, Zhang

, Bredesen

(2011) Signaling via amyloid precursor-like proteins APLP1 and APLP2. J Alzheimers Dis 23, 689–699.

65.

Demetrius

, Simon

(2013) The inverse association of cancer and Alzheimer’s: A bioenergetic mechanism. J R Soc Interface 10, 20130006–20130006.

66.

Ghani

, Sato

, Lee

, Reitz

, Moreno

, Mayeux

, St George-Hyslop

, Rogaeva

(2013) Evidence of recessive Alzheimer disease loci in a Caribbean Hispanic data set: Genome-wide survey of runs of homozygosity. JAMA Neurol 70, 1261–1267.

67.

Jun

, Moncaster

, Koutras

, Seshadri

, Buros

, McKee

, Levesque

, Wolf

, St George-Hyslop

, Goldstein

, Farrer

(2012) δ-Catenin is genetically and biologically associated with cortical cataract and future Alzheimer-related structural and functional brain changes. PloS One 7, e43728.

68.

Atkin

, Hunt

, Minakawa

, Sharkey

, Tipper

, Tennant

, Paulson

(2014) F-box only protein 2 (Fbxo2) regulates amyloid precursor protein levels and processing. J Biol Chem 289, 7038–7048.

69.

Tseveleki

, Rubio

, Vamvakas

S-S

, White

, Taoufik

, Petit

, Quackenbush

, Probert

(2010) Comparative gene expression analysis in mouse models for multiple sclerosis, Alzheimer’s disease and stroke for identifying commonly regulated and disease-specific gene changes. Genomics 96, 82–91.

70.

Gentil

, Delphin

, Mbele

, Deloulme

, Ferro

, Garin

, Baudier

(2001) The giant proteinAHNAKis a specific target for the calcium- and zinc-binding S100B protein: Potential implications for Ca2+homeostasis regulation by S100B J Biol Chem 276, 23253–23252.

71.

Liang

, Han

, Feng

, Sun

, Duan

, Lei

(2012) Concerted perturbation observed in a hub network in Alzheimer’s disease. PloS One 7, e40498.

72.

Mann

, Wu

, Rostamkhani

, Tourtellotte

, MacMurray

, Comings

(2001) Phenylethanolamine N-methyltransferase (PNMT) gene and early-onset Alzheimer disease. Am J Med Genet 105, 312–316.

73.

Sekar

, McDonald

, Cuyugan

, Aldrich

, Kurdoglu

, Adkins

, Serrano

, Beach

, Craig

, Valla

, Reiman

, Liang

(2015) Alzheimer’s disease is associated with altered expression of genes involved in immune response and mitochondrial processes in astrocytes. Neurobiol Aging 36, 583–591.

74.

Cruchaga

, Kauwe

JSK

, Harari

, Jin

, Cai

, Karch

, Benitez

, Jeng

, Skorupa

, Carrell

, Bertelsen

, Bailey

, McKean

, Shulman

, De Jager

, Chibnik

, Bennett

, Arnold

, Harold

, Sims

, Gerrish

, Williams

, Van Deerlin

, Lee

VM-Y

, Shaw

, Trojanowski

, Haines

, Mayeux

, Pericak-Vance

, Farrer

, Schellenberg

, Peskind

, Galasko

, Fagan

, Holtzman

, Morris

, GERAD Consortium, Alzheimer’s Disease Neuroimaging Initiative (ADNI), Alzheimer Disease Genetic Consortium (ADGC), Goate

(2013) GWAS of cerebrospinal fluid tau levels identifies risk variants for Alzheimer’s disease. Neuron 78, 256–268.

75.

Parachikova

, Cotman

(2007) Reduced CXCL12/CXCR4 results in impaired learning and is downregulated in a mouse model of Alzheimer disease. Neurobiol Dis 28, 143–153.

76.

Yan

, Kim

, Nho

, Chen

, Risacher

, Moore

, Saykin

, Shen

, Alzheimer’s Disease Neuroimaging Initiative (2015) Hippocampal transcriptome-guided genetic analysis of correlated episodic memory phenotypes in Alzheimer’s disease. Front Genet 6, 117.

77.

Pey

, Pearce

RKB

, Kalaitzakis

, Griffin

WST

, Gentleman

(2014) Phenotypic profile of alternative activation marker CD163 is different in Alzheimer’s and Parkinson’s disease. Acta Neuropathol Commun 2, 21.

78.

Wilhelmus

MMM

, Otte-Höller

, Wesseling

, de Waal

RMW

, Boelens

, Verbeek

(2006) Specific association of small heat shock proteins with the pathological hallmarks of Alzheimer’s disease brains. Neuropathol Appl Neurobiol 32, 119–130.

79.

Chacón

, Rodríguez-Tébar

(2012) Increased expression of the homologue of enhancer-of-split 1 protects neurons from beta amyloid neurotoxicity and hints at an alternative role for transforming growth factor beta1 as a neuroprotector. Alzheimers Res Ther 4, 31.

80.

Anderson

, Cummings

, Cotman

(1994) Increased immunoreactivity for Jun- and Fos-related proteins in Alzheimer’s disease: Association with pathology. Exp Neurol 125, 286–295.