Neuronal DNA damage response‐associated dysregulation of signalling pathways and cholesterol metabolism at the earliest stages of Alzheimer‐type pathology

Aims Oxidative damage and an associated DNA damage response (DDR) are evident in mild cognitive impairment and early Alzheimer's disease, suggesting that neuronal dysfunction resulting from oxidative DNA damage may account for some of the cognitive impairment not fully explained by Alzheimer‐type pathology. Methods Frontal cortex (Braak stage 0–II) was obtained from the Medical Research Council's Cognitive Function and Ageing Study cohort. Neurones were isolated from eight cases (four high and four low DDR) by laser capture microdissection and changes in the transcriptome identified by microarray analysis. Results Two thousand three hundred seventy‐eight genes were significantly differentially expressed (1690 up‐regulated, 688 down‐regulated, P < 0.001) in cases with a high neuronal DDR. Functional grouping identified dysregulation of cholesterol biosynthesis, insulin and Wnt signalling, and up‐regulation of glycogen synthase kinase 3β. Candidate genes were validated by quantitative real‐time polymerase chain reaction. Cerebrospinal fluid levels of 24(S)‐hydroxycholesterol associated with neuronal DDR across all Braak stages (r s = 0.30, P = 0.03). Conclusions A persistent neuronal DDR may result in increased cholesterol biosynthesis, impaired insulin and Wnt signalling, and increased GSK3β, thereby contributing to neuronal dysfunction independent of Alzheimer‐type pathology in the ageing brain.


Introduction
Population-based studies have shown that age-associated cognitive decline often occurs in individuals with low levels of either classical Alzheimer-type neuropathology or other neurodegenerative pathologies [1]. Oxidative damage to nucleic acids is evident in mild cognitive impairment (MCI) and early Alzheimer's disease (AD) [2][3][4], suggesting that oxidative stress is an early contributor to neuronal dysfunction, which either independently of, or interacting with, early Alzheimer-type pathology results in cognitive impairment. Oxidative DNA damage induces a DNA damage response (DDR), which is characterized by the activation of DNA-protein kinase catalytic subunit (DNA-PKcs) and the phosphorylation of the histone H2AX (γH2AX) [5]. Previous investigation of the Medical Research Council's Cognitive Function and Ageing Study (MRC CFAS) cohort has demonstrated high levels of oxidative stress and an associated DDR occur at the earliest stages of Alzheimer pathology [6], as defined by low Braak and Braak neurofibrillary tangle stage [7], and an increased neuronal DDR at these early pathological stages is associated with cognitive impairment and neuronal senescence [8].
Various candidate mechanisms may underlie neuronal vulnerability and initiate adverse cellular processes several years before eventual cognitive decline. Signalling dysregulation has been implicated in the pathogenesis of AD [9,10], and large-scale genome-wide association study (GWAS) studies have identified potential novel pathways [11]; however, the contribution of these cell processes to cognitive impairment remains to be defined. Population-based neuropathology studies allow unbiased assessment of the relationship of cellular and molecular pathologies to dementia and enables early stages of classical pathologies to be investigated. By identifying novel associations, these population-based approaches have the potential to identify new biomarkers and therapeutic strategies [12,13].
Having demonstrated the population variation in the neuronal DDR in individuals with little or no Alzheimertype pathology and the association with cognitive impairment [8], this study aimed to identify the gene expression changes associated with the neuronal DDR, which may contribute to neuronal dysfunction in the ageing brain. Laser capture microdissection (LCM) of neurones from human post-mortem material combined with microarray technology enables the neuronal transcriptome of samples to be simultaneously assessed, and changes in gene expression identified. Although a number of studies have reported microarray data in normal ageing and AD [14][15][16][17][18], to date, no study has specifically characterized the gene expression profile of neurones with high levels of a DDR (DNA-PKcs and γH2AX) at low Braak and Braak stages. Our findings identify a major transcriptional response with potentially important implications for neuronal dysfunction and cognitive decline independent of classical Alzheimer-type pathology in the ageing brain.

Human CNS tissue
CNS tissue was obtained from MRC CFAS, following Research Ethics Committee approval (REC ref 12/EM/ 0118). The structure of the CFAS neuropathology cohort has been previously described and reviewed [10]. Brains were dissected following a standard protocol [19] and neuropathological lesions assessed as part of the core CFAS neuropathology study using a modified protocol from the Consortium to Establish a Registry of Alzheimer's Disease (CERAD) [20] (http://www.cfas.ac.uk). Braak and Braak stage 0-II, frozen frontal cortex (FCx) was available for 39 cases with an average age at death of 82 years (range 70-102 years), median post-mortem delay of 17.5 h [interquartile range (IQR 10-24 h] and tissue pH 6.6 (IQR pH 6.3-6.9). Individuals in the study were regularly interviewed and underwent extensive cognitive assessment [1,21]. Dementia status at death was determined on the basis of all information available for each participant [22]. Within this cohort, 14 participants had dementia, 22 had no dementia and 3 participants had an unknown dementia status at death due to the lack of information in the years preceding death.
using laser capture microscopy (LCM) (Figure 1), based on previous investigation of the neuronal DDR (DNA-PKcs and γH2AX) in this cohort [8]. Details of cases used in the microarray study are shown in Table 1. Validation experiments were performed on LCM-isolated neurones from the remaining 29 CFAS cases. Frozen sections (7 μm) were fixed in ice cold acetone for 3 min and stained with Toludine Blue for 1 min. Sections were dehydrated in a graded series of ethanol (70, 95 and 100%), extensively cleared in xylene and air dried for 1 h. Approximately, 1000 pyramidal neurones were isolated from each case using a PixCell II LCM system (Life Technologies, Paisley, UK). Total RNA was extracted using the PicoPure RNA extraction kit according to the manufacturer's protocol (Life Technologies), with typical yields of 100 ng of RNA. The quantity (NanoDrop 1000 spectrophotometer, Thermo Scientific, Wilmington, DE USA) and quality (Agilent Bioanalyser 2100, RNA 6000 Pico LabChip; Agilent, Palo Alto, CA, USA) of the starting RNA were analysed. Sterile solutions made with diethylpyrocarbonate (DEPC)-treated water and RNAsefree conditions were used throughout this protocol.

RNA amplification and microarray hybridization
Total RNA was annealed to an oligo-d(T) primer with a T7 polymerase binding site. After generation of doublestranded cDNA, copy RNA (cRNA) was transcribed which then formed the RNA template for a second round of amplification. At the end of this round, after synthesis of double-stranded cDNA, biotin-labelled cRNA was prepared using the Affymetrix Gene Chip (Affymetrix, Santa Clara, CA, USA) in vitro transcription labelling kit. Following clean-up of the biotin-labelled cRNA the material was assayed (Agilent Bioanalyser 2100) to ensure sufficient RNA of appropriate quality had been prepared. Labelled cRNA (12.5 μg) was fragmented and applied to HGU133 Plus 2.0 gene microarrays and hybridized over 16 h at 45°C in a rotating oven at 60 rpm. Post hybridization washing and sample staining was carried out using the Fluidics Station 400 and the Gene Chip Operating System.  Gene chips were scanned using the GC3000 7G scanner and data processed for quality control using Expression Console software (Affymetrix) and further analysis was carried out using Qlucore Omics Explorer (Qlucore, Lund, Sweden).

Microarray analysis
The Robust Multi-array Average (RMA) algorithm was used for data normalization and univariate and principal component analyses to determine intensity distribution and eliminate sample outliers [24]. Significant differentially expressed genes (P < 0.001) were analysed using Qlucore Omics Explorer. The Database for Annotation Visualisation and Integrated Discovery bioinformatics programme (DAVID) was used to group genes according to their function and to assign genes to specific functional pathways [25].

Validation of microarray data: quantitative real-time polymerase chain reaction (qPCR)
RNA was extracted from LCM-ed neurones in additional frozen FCx cases, which had not been used for the microarray studies. qPCR was performed using IDT PrimeTime qPCR assays (Integrated DNA Technologies, Glasgow, UK) containing 50 ng cDNA, 500 nM primers, 250 nM probe, and Brilliant qPCR mix (Agilent, Stockport, UK) in a total volume of 10 μl, for more details of primer sequences see Table 2. Following denaturation at 95°C for 10 min the products were amplified (40 cycles at 95°C for 30 s and 60°C for 60 s) using an MX3000P RT PCR System (Stratagene, Paisley, UK). β-actin was amplified on each plate to normalize expression levels of target genes between different samples using the ΔΔCt calculation (ABI) and to assess assay reproducibility.

Immunohistochemistry
To confirm neuronal expression of proteins encoded by the candidate genes, immunostaining was performed using a standard avidin biotinylated enzyme complex (ABC) method, and the signal visualized with diaminobenzidine (Vector Laboratories, Peterborough, UK). A summary of the primary antibodies used is shown in Table 3.

Statistical analysis
Spearman's correlation coefficient was calculated to test associations between continuous variables. Analyses were performed and graphs obtained using IBM® spss® Statistics 21 (Armonk, NY, USA) and statistical package stata, version 12 (College Station, TX, USA). To test the possibility of 24(S)-OHC levels as a risk factor for dementia, logistic regression was used, and to test if possession of the ApoEε4 allele was a risk factor for decreased 24(S)-OHC, we used multiple linear regression analysis with 24(S)-OHC as a dependent variable. All analyses were adjusted by age at death and sex. All tests were two-tailed, 95% confidence intervals (CI) were calculated for the odds ratio (OR) and for the linear regression coefficients (β).

DDR-associated changes in the neuronal transcriptome
The transcription profiles of laser-captured neurones from high-and low-DDR cases were generated using Human Genome U133 Plus 2.0 Arrays, which comprise 1. Gene expression analyses are sensitive to the presence of sample outliers, therefore rigorous quality control procedures were used to ensure the highest possible level of quality for both datasets. One low-DDR sample outlier was identified by visual inspection after clustering the samples using hierarchical clustering and one high DDR case with Pick's disease was excluded on review of the neuropathology, both cases were removed from subsequent analysis (Figure 2). The RMA was re-analysed after these two cases were removed and principal components analysis was performed on the significant, differentially expressed genes (the expression dataset is freely available at Gene Expression Omnibus, accession number GSE66333). Transcript clusters showed no association with age, gender, post-mortem delay or pH.
Two thousand three hundred seventy-eight genes (1690 up-regulated, 688 down-regulated) were significantly differentially expressed (P < 0.001) in cases with a high DDR (Supplementary Tables 1 and 2). Functional grouping analysis identified the dysregulation of genes associated with the cellular response to stress, the synapse, signalling, ubiquitin-mediated proteolysis, transport, the cytoskeleton and transcription. Pathway analysis identified significant up-regulation of Wnt signalling transcripts (P = 0.004) and down-regulation of insulin signalling associated transcripts (P = 0.029) in cases with a high neuronal DDR. NFL transcripts had the greatest fold change (FC) (NFL probe set 221805_at, FC = 36.03, P = 0.0002; probe set 221801_x_at, FC = 13.87, P = 0.0001; probe set 221916_at, FC = 5.89, P = 0.0001). Glycogen synthase kinase 3β (GSK3B) transcripts were also significantly up-regulated in cases with a high neuronal DDR (probe set 226191_at, FC = 4.19, P = 0.00004).

DDR-associated activation of the cholesterol biosynthesis pathway
Analysis of the transcription profiles of laser-captured neurones from high neuronal DDR cases compared with low neuronal DDR at the less stringent parameters   (Figure 3f) was confirmed by immunohistochemistry. Furthermore, significantly higher levels of NFL (P = 0.003) and significantly lower levels of Wnt3a (P = 0.007) were associated with a high neuronal DDR, as detected by Western blotting of protein extracts from the array cases ( Figure 4).

CSF levels of 24(S)-OHC associate with the neuronal DDR and inversely associate with Braak and Braak stage
As cholesterol plays a key role in synapse formation and function [26], the study was extended to assess CSF levels of 24(S)-OHC across all Braak and Braak stages in all available CSF samples from one well-characterized centre of the CFAS cohort [6]. Use of this larger cohort enabled the association between cholesterol turnover and the DDR, Braak and Braak stage, ApoEε4 genotype and cognitive impairment to be investigated.  (Figure 5b,c).

Discussion
Age-associated cognitive decline often occurs in individuals with low levels of conventional dementia-associated pathologies (Aβ plaques, neurofibrillary tangles, vascular lesions, synucleinopathy) [1,27]. Braak and Braak staging describes an anatomical progression of neurofibrillary tangles, in which tangles that are restricted to mesial temporal lobe structures may precede clinico-pathological AD by many years [7]. The current study investigated prefrontal cortex as it is affected relatively late in the AD process, thereby allowing us to identify potential gene expression changes at early Braak and Braak stages before the onset of classical AD-associated neuropathological changes. We have previously demonstrated an association between the neuronal DDR and cognitive impairment in cases with little or no Alzheimer's neuropathology, and shown that high levels of a neuronal DDR correlate with lower Mini Mental State Examination (MMSE) scores [8]. The present population-based study extends this previous finding and demonstrates that the neuronal response to DNA damage at the earliest stages of Alzheimer's pathology (Braak and Braak stage 0-II) is associated with dysregulation of intracellular signalling pathways and  cholesterol biosynthesis, which may contribute to neuronal dysfunction and cognitive impairment independent of established Alzheimer pathology in the ageing brain. Insulin resistance has been linked to several neurodegenerative diseases, including AD [9,[28][29][30][31][32]. Low concentrations of insulin, reduced receptor density and dysregulation of insulin signalling have all been reported in AD [33,34] and with increasing Braak and Braak stage in the ageing brain [10,35]. Furthermore, administration of insulin has been shown to improve memory in AD patients [36] suggesting a potential therapeutic avenue for dementia treatment. In the present transcriptomic study a high neuronal DDR was associated with the dysregulation of insulin/IGF and Wnt signalling pathways. Defects in both signalling pathways have been postulated to contribute to the pathogenesis of AD and dementia [9,37,38]. Although pathway analysis of the microarray data identified significant up-regulation of Wnt signalling associated transcripts in cases with a high neuronal DDR, protein expression was significantly lower in these cases,  suggesting post-transcriptional regulation of Wnt expression. Both insulin/IGF and Wnt signalling negatively regulate the activity of GSK3β [39], expression of which was significantly up-regulated in cases with a high neuronal DDR. The GSK3 hypothesis of AD proposes that GSK3β over-activation plays a key role in the hyperphosphorylation of tau, increased β-amyloid production and neuronal dysfunction [40]. We propose that increased levels of oxidative stress lead to a neuronal DDR, which is associated with the down-regulation of insulin/ IGF and Wnt signalling, resulting in increased GSK3β activity and ultimately leading to neuronal dysfunction and cognitive impairment, independent of local Alzheimer-type pathology in the ageing brain.
Increasing evidence suggests AD represents a metabolic disease in which brain glucose utilization and energy production are impaired [32][33][34]41]. Dysfunction of the insulin/IGF signalling pathway impacts neuronal growth, survival, differentiation, synapse formation, energy metabolism, gene expression, protein synthesis and cytoskeletal assembly [31]. Cytoskeletal disruption is a prominent feature of neurodegenerative diseases, including AD [42,43]. Expression of NFL was significantly up-regulated in cases with a high neuronal DDR, both at the gene (ranging from 5-to 36-fold increase) and protein levels (fourfold increase), suggesting that abnormal accumulation of this cytoskeletal protein may contribute to neuronal dysfunction by impacting axonal transport.
Oxidative stress is one of the earliest events in AD, with implications as an important mediator in the onset, progression and pathogenesis of the disease [2][3][4]. The proposed sources of oxidative stress include abnormal mitochondria, redox transition metals and oxysterols [44,45]. In the brain, cholesterol is converted to the oxysterol 24(S)-OHC, which is considered a marker of cholesterol turnover [46,47], and which is elevated in the CSF of patients with MCI [48][49][50]. In the current study, CSF levels of 24(S)-OHC were highest in cases with little or no Alzheimer-type pathology, and correlated with the neuronal DDR. We propose that high levels of the prooxidant 24(S)-OHC are one potential source of oxidative stress, which may contribute to an elevated neuronal response to oxidative DNA damage, resulting in the activation of the neuronal cholesterol biosynthesis pathway to maintain axons, dendrites and synapses at low Braak and Braak stages [26,51]. However, while increased cholesterol biosynthesis may preserve neuronal integrity at low Braak stages, high levels of 24(S)-OHC may also promote the progression of tau pathology [52] and increase Aβ neurotoxicity [45] in the ageing brain. We also demonstrate a significant reduction in 24(S)-OHC expression associated with increasing Braak stage in the ageing brain, supporting the hypothesis that cholesterol turnover is highest during early neurodegenerative changes [53], and suggesting that this neuroprotective response fails with disease progression.
The pathogenesis of AD is linked to cholesterol metabolism, and is associated with genetic risk factors including apolipoprotein E (ApoE) genotype and polymorphisms in HMGCR [54][55][56][57]. However, no association between ApoE genotype and 24(S)-OHC was detected in this study, although it should be noted that the sample size for the CSF study was small. In the brain, cholesterol is almost exclusively derived from endogenous biosynthesis, and is regulated by the transcription factor SREBP2, which controls expression of the enzymes involved in cholesterol synthesis, including the rate-limiting enzyme HMGCR [58]. Although neurones are capable of synthesizing their own cholesterol, in the adult brain they rely on delivery of cholesterol from neighbouring astrocytes via ApoEcontaining lipoproteins [59,60], but can activate the neuronal cholesterol biosynthesis pathway in response to oxidative stress in vitro [61]. In support of this observation, we demonstrate the significant up-regulation of cholesterol biosynthesis genes in neurones with high levels of a DDR.
In summary, we have defined the molecular signature of a neuronal DDR, which associates with cognitive impairment in older individuals with only early stage Alzheimertype pathology. As these individuals did not have established Alzheimer's neuropathology, changes in the neuronal transcriptome are not attributable to established AD, and may be independent of Alzheimer's or interact with the earliest molecular stages of the disease, or may reflect brain ageing. No other significant brain pathologies were present. We demonstrate an association between a persistent neuronal DDR, increased cholesterol biosynthesis, impaired insulin/IGF and Wnt signalling, and increased GSK3β, which may contribute to neuronal dysfunction and cognitive impairment. As these mechanisms, operating at the earliest stages of Alzheimer's neuropathology, are potential therapeutic targets it is important to understand their role in cognitive decline and to develop biomarkers to identify individuals who may benefit from targeting such pathways at preclinical disease stages.

Supporting information
Additional Supporting Information may be found in the online version of this article at the publisher's web-site: Table S1. Up-regulated genes in high versus low neuronal DDR cases at low Braak and Braak stages (P < 0.001). Table S2. Down-regulated genes in high versus low neuronal DDR cases at low Braak and Braak stages (P < 0.001).