|Home | About | Journals | Submit | Contact Us | Français|
Alzheimer’s disease (AD) is a devastating neurodegenerative disorder that threatens to reach epidemic proportions as our population ages. Although much research has examined molecular pathways associated with AD, relatively few such studies have focused on the disease’s critical early stages. In a prior microarray study we correlated gene expression in hippocampus with degree of Alzheimer’s disease and found close associations between upregulation of apparent glial transcription factor/epigenetic/tumor suppressor genes and incipient AD. The results suggested a new model in which AD pathology spreads along myelinated axons (Blalock et al., 2004). However, the microarray analyses were performed on RNA extracted from frozen hand-dissected hippocampal CA1 tissue blocks containing both gray and white matter, limiting the confidence with which transcriptional changes in gray matter could be distinguished from those in white matter. Here, we used laser capture microdissection (LCM) to exclude major white matter tracts while selectively collecting CA1 hippocampal gray matter from formalin-fixed, paraffin-embedded (FFPE) hippocampal sections of the same subjects assessed in our prior study. Microarray analyses of this gray matter-enriched tissue revealed many transcriptional changes similar to those seen in our past study and in studies by others, particularly for downregulated neuron-related genes. Additionally, the present analyses identified several previously undetected pathway alterations, including downregulation of molecules that stabilize ryanodine receptor Ca2+ release and upregulation of vasculature development. Conversely, we found a striking paucity of the upregulated changes in the putative glial and growth-related genes that had been strongly overrepresented in the prior mixed-tissue study. We conclude that FFPE tissue can be a reliable resource for microarray studies of brain tissue, that upregulation of growth-related epigenetic/transcription factors during incipient AD is predominantly localized in and around white matter (supporting our prior findings and model), and that novel alterations in vascular and ryanodine receptor-related pathways in gray matter are closely associated with incipient AD.
The major pathological hallmarks of Alzheimer’s disease (AD) include beta amyloid accumulation (Hardy and Selkoe, 2002; Klein et al., 2001; Morgan, 2003; Mucke et al., 2000; Mullan and Crawford, 1994; Price and Sisodia, 1998; Tanzi and Bertram, 2001), neurofibrillary tangles (NFTs) (Davies and Koppel, 2009; Johnson and Bailey, 2002; Morris et al., 2011; Noble et al., 2003) and synaptic dysfunction or loss (Masliah et al., 1994; Scheff and Price, 2001; Sze et al., 1997; Yao et al., 2003). In addition, AD is associated with other pathological processes, including failing mitochondrial function and oxidative stress (Bickford et al., 2000; Butterfield and Sultana, 2007; Perry et al., 2003; Wang et al., 2010), increased inflammatory response (Finch et al., 2002; Gemma et al., 2002; Ginsberg et al., 2006; Mrak and Griffin, 2001; Mucke et al., 2000; Rogers et al., 1996), protein misfolding (Forman et al., 2003; Stefani and Dobson, 2003), altered growth factor signaling (Mufson et al., 2007; Tuszynski and Gage, 1990; Williams et al., 2006), aberrant reentry of neurons into the cell cycle (Arendt et al., 2000; Bowser and Smith, 2002), lysosomal activation (Ginsberg et al., 2010; Nixon et al., 2001), endocrine alteration (Brinton, 2008; Landfield et al., 2007; Simpkins et al., 2005; Sohrabji and Lewis, 2006), insulin resistance (Craft, 2007; Gustafson, 2006; Whitmer et al., 2007; Yaffe et al., 2004), cholesterol dyshomeostasis (Petanceska et al., 2002; Puglielli et al., 2001), and calcium dysregulation. The latter plays an important role in normal brain aging as well as in some models of AD (Bezprozvanny and Mattson, 2008; Disterhoft et al., 1994; Foster and Norris, 1997; Gibson and Peterson, 1987; Khachaturian, 1989; Landfield, 1987; Michaelis et al., 1996; Nixon et al., 1994; Norris et al., 1998; Stutzmann, 2005; Stutzmann et al., 2007; Thibault et al., 2007; Wang et al., 2010), and may arise in part from a decrease in immunophilin-mediated stabilization of ryanodine receptors (Gant et al., 2011).
The complexity and number of changes associated with AD has impeded attempts to disentangle the processes important for pathogenesis or to define the roles of specific cell types in disease progression. Microarray analysis of the simultaneous expression of thousands of genes is well-suited to address complex processes and has been used effectively to provide overviews of the gene networks disturbed in AD (Colangelo et al., 2002; Ginsberg et al., 2006; Ginsberg et al., 2000; Liang et al., 2008; Loring et al., 2001; Pasinetti, 2001; Wang et al., 2010; Yao et al., 2003). Nevertheless, expression profiles vary substantially across cell types and regions (Bishop et al., 2010; Blalock et al., 2010; Burger, 2010; Ginsberg et al., 2006; Zahn et al., 2007; Zhao et al., 2001), and it is not yet clear how the expression signatures of specific cell/tissue types are related to early-stage AD.
In a prior study, we combined microarray technology with statistical correlation analyses to identify hippocampal gene expression changes associated with cognitive dysfunction and neurofibrillary tangles across a spectrum from incipient to severe AD (Blalock et al., 2004). Results of that study led us to propose that incipient AD was in part driven by aberrant activation of growth factors in oligodendrocytes and myelinated fiber tracts and, in turn, suppressive responses in other cell types (Blalock et al, 2004). However, because our study was conducted on hand-dissected frozen hippocampal CA1 blocks containing both white and gray matter, it was not possible to clearly distinguish their transcriptional changes. Further, when heterogeneous cell types/regions are mixed together, alterations in one component may oppose or dilute those in another, potentially obscuring important tissue-specific changes.
To overcome tissue heterogeneity problems, laser capture microdissection (LCM) technology and other techniques to isolate specific subregions and individual cell types have been employed in several AD studies (Ginsberg et al., 2010; Ginsberg et al., 2006; Liang et al., 2008). Here, we used a similar LCM approach to address some of the tissue-specific questions that were not clearly resolved in our prior study. In the present LCM study, we laser dissected formalin-fixed, paraffin-embedded (FFPE) sections from the same subjects (N = 30) analyzed in our earlier study (Blalock et al., 2004) and captured hippocampal CA1 gray matter (neuropil/neuronal somata) regions while largely excluding prominent white matter tracts (the perforant path, fimbria and alveus). Thus, this re-sampling of the same subjects provided a unique opportunity to directly compare results from similar samples containing or lacking significant white matter components, as well as to compare data from two disparate tissue fixation/collection approaches. If RNA integrity were generally comparable across these methods, we predicted many findings would correspond between the two studies, as both methods collected neuropil. However, we also hypothesized that, compared to our past analysis of white-matter-enriched hand-dissected CA1 tissue, the present analysis of white-matter-sparse tissue should detect fewer upregulated AD-associated genes, while also uncovering some previously obscured neuronal-specific genes/processes. The data reported here support these hypotheses, strengthening our model of white matter pathogenesis and revealing novel gray matter alterations with potential mechanistic and/or therapeutic implications.
Descriptions and categorization of the subjects were reported previously (Blalock et al, 2004). Brain sections from a total of 30 of the original 31 subjects were analyzed here. As in the prior study, we correlated gene expression with quantitative values on two AD marker scales (MiniMental Status Exam (MMSE) and neurofibrillary tangle (NFT) density), irrespective of clinical diagnosis. This strategy relied on quantitative markers to assess extent of disease progression rather than on categorization by clinico-pathologic criteria, which can be uncertain in borderline cases or when diagnostic markers disagree. However, for comparison and validation, the 30 subjects were also categorized into four categories of varying AD severity. Briefly, subjects (11 male, 19 female; average age 86.3 ± 1.4 years) were separated into four categories (control n = 8; incipient n = 7; moderate n = 8; severe n = 7) largely based on MMSE results (control- 27.6 ± 0.6; incipient- 24.3 ± 1.1; moderate- 16.5 ± 0.6; severe- 6.0 ± 1.4). For borderline cases, NFT count (control- 3.0 ± 1.1; incipient- 17.5 ± 8.2; moderate- 25.6 ± 3.5; severe- 32.7 ± 3.2) and Braak staging (control- 2.3 ± 0.4; incipient- 5.0 ± 0.5; moderate- 5.5 ± 0.2; severe- 5.8 ± 0.2) also informed categorization decisions.
For this study, formalin fixed, paraffin embedded (FFPE) specimens were used. Post mortem interval (PMI: 3.7 ± 0.6 hours); duration of direct formalin exposure (4.0 ± 1.2 days); and time in paraffin block (7.3 ± 0.2 years) did not differ significantly among groups (p > 0.05 in all tests; 1-ANOVA). Eight-μm sections from FFPE hippocampal blocks were mounted on PALM 1mm polyethylene naphthalate (PEN) membrane slides (Zeiss, Germany). Slides were de-waxed (2 × 10 min xylene, 2 × 10 min 100% EtOH, 1 × 10 min 90% EtOH, 1 × 10 min 70% EtOH, 1 × dH20 5 min) and dehydrated (70% EtOH 10 min, 90 % EtOH 2 min, 2 × 100% EtOH 3 & 5 min, 2 × 100% xylene 5 & 10 min) to facilitate dissection and laser capture. Neuropathologists defined the CA1 region for each specimen using photomicrographs.
A Zeiss AxioObserver PALM Microbeam with RoboMover cap system was used to image, cut and capture specimens (Woods Hole, MA; courtesy of Zeiss- Fig. 1). The pathologist-identified CA1 region for each specimen (Fig. 1A) was used as a guide during laser capture. Specimen sections were targeted using a 5x cutting objective and were collected from regions of grossly defined gray matter (as defined in The Human Brain, Nolte, 2002) in hippocampal sub-field CA1, comprising the pyramidal cell layer and surrounding neuropil, largely excluding regions containing major white matter tracts (e.g., fimbria, alveus and perforant path) (Fig. 1B). Note that gray matter still contains astrocytes, capillaries, and other non-neuronal cells, as well as myelinated fibers (Vercellino et al., 2009), albeit considerably fewer of the latter than found in whole tissue dissections. The laser was set to cut through the PEN membrane to which the specimens were adhered, and then defocused and activated to capture (using the light pressure catapult) the cut regions of each sample. After each catapult, the capture area (siliconized eppendorf-style 0.5 ml centrifuge tube cap- one per slide) was visualized (Fig. 1B inset) to validate collection. On average, ~20 individual capture attempts were performed on each specimen and 1-2 capture attempts failed per sample. Captured material was stored dry at room temperature prior to RNA extraction and analysis.
RNA was extracted using RecoverAll Total Nucleic Acid Isolation Kit for FFPE (Ambion) according to manufacturer’s instructions (3h incubation at 55° C followed by glass fiber filtration). This system has recently been shown to outperform other FFPE methods/ kits regarding yield of amplifiable RNA (Okello et al., 2010). Quality assessment of extracted material was performed with the Paradise Reagent Quality Assessment Kit (Molecular Devices), as well as via NanoDrop (Thermoscientific). All samples yielded sufficient genetic material (>50 ng) for subsequent reactions. 50 ng of extracted purified nucleic acid underwent RNA amplification using WT-Ovation FFPE System (NuGen) followed by FL-Ovation cDNA Biotin Module V2 (NuGen) for labeling and microarray (Affymetrix HGU133 v2) hybridization. All 30 microarrays (one per each subject sample) performed within acceptable limits (Scaling factor: 32.6 +/− 3.7; RawQ: 1.28 +/− 0.01; GapDH 3′:5′: 1.48 +/− 0.08; % present 35.4 +/− 1.5) and were not significantly different across condition (p < 0.5 for all measures, 1-ANOVA). In comparison to prior work on frozen samples from the same subjects (Scaling factor: 5.9 +/− 0.6; RawQ 2.7 +/− 0.04; % present: 44.6 +/− 1.1), the increased scaling factor, decreased RawQ, and reduced % present all indicate reduced signal intensity, consistent with other reports of the dynamics of small FFPE sample extraction (Turner et al., 2011). Finally, the % present call, while lower than found in frozen tissue, is much greater than would be expected by chance (5%),suggesting that the extracted material contains substantial amounts of valid mRNA.
Probe sets were annotated, and transcriptional profiles were generated, using the MAS5 algorithm and annotation data sets (Affymetrix GCOS v. 1.1; HGU133 annotation October, 2003) in order to facilitate comparison with our prior work (Blalock et al., 2004). Raw data are available from the Gene Expression Omnibus (Barrett and Edgar, 2006) under accession ID GSE28146. Results were filtered for presence, redundancy, and annotation status and analyzed by Pearson’s test for correlation with each subject’s Mini-Mental Status Exam score and Neurofibrillary Tangle count. The false discovery rate (FDR) (Hochberg and Benjamini, 1990) was used to estimate the error of multiple testing’s contribution to False Positives (see FDR, Fig 2). The DAVID suite of bioinformatic tools (Huang da et al., 2009), which identifies functional categories and biological processes/pathways that are statistically overrepresented by genes of interest, was used to identify processes/pathways from the Gene Ontology (Ashburner et al., 2000) associated with the various lists of AD-correlated genes, using the ‘table cluster’ option. To reduce redundancy, only one representative pathway (p ≤ 0.05; between 3 and 100 genes) from within each cluster is reported.
The comprehensive laser dissection data set was acquired from formalin-fixed, paraffin-embedded microscope slide specimens (from the same subjects as the original frozen tissue block study). Quality control results (Methods 2.3) showed that the laser dissection samples averaged 35% presence calls, ~7x the presence call rate that would be expected by chance (5%). Further, concordance at the presence call level among the 30 laser dissection arrays was also quite high. Genes rated present/absent on a single microarray tended to agree across all arrays (78.4% agreement).
Microarray signal intensities and presence calls were transferred to Excel and analyzed as in prior work (Blalock et al., 2004) (Fig. 2). Briefly, results were filtered to remove probe sets that were rated absent (< 4 presence calls across 30 chips) or poorly annotated (lacking or redundant gene symbol). The remaining 12,665 genes were correlated (Pearson’s test) with MMSE scores as well as NFT counts across all subjects (Overall correlation). Genes with significant (p ≤ 0.05) overall correlations were also sub-correlated with MMSE and NFT across only the control and incipient subjects (Incipient correlation).
Results for all significantly correlated genes are provided as supplementary information (Supplemental Table 1). As seen in Fig. 2, correlation with MMSE appeared to explain a greater proportion of transcriptional profile variability than correlation with NFT, although there was a strong, statistically significant overlap between the two (747 genes; p < 0.0001, binomial test). However, 470/ 1566 (30%) of the NFT-correlated genes were also correlated across the incipient and control groups, whereas only 335/ 2646 (12.6%) of MMSE-correlated genes showed a similar incipient AD correlation. Hence, a greater proportion of NFT-correlated genes were associated with early disease progression, while cognitive measures appeared more strongly tied to later disease stages.
Table 1 shows the functional processes found by DAVID analysis to be overrepresented by either downregulated or upregulated overall AD-correlated genes. Downregulated processes were strongly associated with neuronal function (synaptic components, ion channel activity, neurotransmitter receptors, axon, Ca2+ signaling) and, to a lesser extent, with energy production (glycolysis, mitochondrial components) and cell development (differentiation, transport, and filament structure). Conversely, upregulated categories included the now-stereotypical inflammatory (inflammatory response and complement activation) and iron homeostasis responses often associated with glial activation, as well as strong apoptotic signals. These patterns of downregulated neuronal and mitochondrial processes in conjunction with upregulated inflammatory and apoptotic pathways are highly similar to results reported in our prior work analyzing AD-associated microarray signatures from excised blocks of frozen hippocampal CA1 tissue (Blalock et al, 2004), as well to results from a number of studies by others (Wang et al., 2010). All significantly correlated genes in the present study are listed in Supplemental Table 1.
Nonetheless, some notable differences were found between processes identified in this study vs. those identified in our prior study. Among the most prominent was the present study’s paucity, relative to the prior study, of upregulated genes related to growth, tumor suppression, transcription, chromatin remodeling, lipid metabolism, extracellular matrix, cytoskeletal organization, oxidative stress, and some immune processes. For downregulated processes, the absence here of ubiquitin-dependent protein catabolic categories was particularly evident. Conversely, potentially important processes seen in laser dissected gray matter samples, including decreased regulation of Ca2+ release from ryanodine receptors, and increased vascular development, were not found in our prior work.
The comprehensive analysis (section 3.1) of laser-dissected gray matter from the same subjects generally confirmed and extended many of our prior observations from frozen tissue blocks of CA1 material (Blalock et al., 2004). However, to more systematically test the degree of similarity between the two data sets (as well as to identify unique or missing signatures), we statistically compared the two studies as follows: The same probe level algorithm (MAS5) and annotation files were used for both the “laser dissection” and “tissue block” data. Further, testing for significant correlations was restricted to 10,475 probe sets that were perfect matches between the laser microdissection (Affymetrix HG U133 v.2) and tissue block (Affymetrix HG U133A) microarrays. At the presence call level, 59.4% (6,227/10,475) of probe sets were present in laser dissection, and 60.1% (6,338/ 10,475) were present in tissue block, with 5,065 present in both (~81% agreement). These results are highly unlikely by chance (p < 0.0001; binomial test), supporting the idea that detection sensitivities for the laser dissection and tissue block samples, at least at the presence call level prior to evaluation of AD-related gene signatures, were highly similar. Because presence in either study indicates a potential for correlation in at least one dataset, the list of common annotated genes available for testing correlations was expanded from 5,065 present in both studies to include genes present in at least one of the two studies, for a total of 7,537 testable genes.
All genes found to correlate overall with at least one AD marker (NFT or MMSE) were partitioned according to the data set in which they correlated (“tissue block”, “laser dissection”, or “both”; Fig. 3-Venn diagram) and the complete list of correlated genes is provided (Supplemental Table 2). Functional overrepresentation analysis (see Methods) was used to identify biological processes overrepresented by significant genes. The primary processes identified for each of the three Venn component gene lists are shown below (Tables (Tables22--44).
(Overlap in Venn diagram, Fig. 3 inset center; Table 2; Supplemental Table 2A). If the results from laser dissection samples are generally valid, then the laser dissection and tissue block data should have many significantly correlated genes in common. That is, the number of AD-correlated overlapping genes should be much higher than expected by chance. Further, that overlap should be relatively enriched in neuron-specific genes and relatively impoverished in white matter genes, compared with results that are unique to the tissue block analysis. As in prior work examining similarities among gene signatures (Blalock et al., 2010), we plotted the found:expected ratio for number of overlapping genes as a function of the p-value cutoff (α level) used to test for correlation in both datasets (Fig. 3). There was very strong agreement between the two data sets and at a p-value cutoff of 0.05, 509 genes, or more than 34x as many genes as would be expected by chance, were found in the overlap (Fig. 3 Venn diagram). This highly significant result (p < 1.0×10−16; binomial test) shows that the overlapping genomic signature is very unlikely to have been identified by chance. Of the 509 genes that exhibited correlation with overall AD in both studies (Supplemental Table 2A), nearly twice as many also exhibited correlation with incipient AD in the laser capture study (202 genes, ~40%) as in the tissue block study (110 genes, ~22%) (p = 1.9×10−14, binomial test), suggesting that laser captured gray matter provided higher resolution and better signal-to-noise than mixed samples of gray and white matter.
Table 2 shows the functional processes found by DAVID analysis to be overrepresented by the 509 genes correlated with overall AD in both studies. It can be seen that these functional processes are generally similar to the processes identified in the comprehensive analysis of all genes in the laser capture study (Table 1).
(Fig. 3 Venn diagram left; Table 3; Supplemental table 2B). The largest number of significantly correlated genes (1905, 1159 of which were upregulated), was found among those correlated exclusively within tissue block samples (Supplemental Table 2B). Finding more significant gene expression changes in tissue block specimens is perhaps not surprising, given that the blocks contained a more varied cell type population. In addition, the functional categories identified by DAVID analysis for these 1905 tissue block genes tended to have the strongest statistical significance (lowest p-values). The tissue block study’s relative abundance of upregulated AD-associated genes is reflected statistically by a highly significant reduction (p = 8.7×10−6, binomial test) in the ratio of upregulated to downregulated AD-correlated genes in the laser dissection (1.26) vs. tissue block (1.46) studies (Fig. 3), indicating that many upregulated correlated genes were related to the white matter tracts included exclusively in those tissue block samples. Among downregulated categories, the 5 most significant (Table 3, left) appeared closely related to mitochondrial bioenergetics, while other downregulated categories were associated with axonal and synaptic activity. The upregulated AD gene signature in tissue block appeared heavily enriched with epigenetic and transcription factor-related processes (Table 3 right- chromosome organization, chromatin assembly, promoter binding, etc.).
(Fig. 3 Venn diagram right; Table 4; Supplemental table 2C). Downregulated genes that correlated with AD exclusively in laser dissected material were similar to those seen in both the comprehensive analysis (Table 1) as well as the overlapping region of the Venn diagram, suggesting that different genes from the same processes may be identified in the different analyses (see Discussion). Nonetheless, proportionately more downregulated genes identified exclusively in LCM gray matter samples represented processes associated with declining protein transport and processing, compared to those in the prior tissue block study. Among upregulated genes, the more than 500 that were associated with secretory, epigenetic, stress and transcription processes were correlated with AD at much weaker average significance levels than in the tissue block samples. However, LCM sample genes uniquely identified previously undetected alterations correlated with incipient AD, including upregulated vasculature development Table 4), as well as downregulation of genes important for stabilization of ryanodine receptor-related intracellular Ca2+ release (the latter in the comprehensive laser capture analysis, Table 1).
The laser dissection protocol’s gray matter-enriched collection technique appears to have facilitated identification of novel AD-associated processes, including downregulated stabilization of ryanodine-sensitive Ca2+ release and upregulated vasculature development. Although the laser microdissected region of CA1 is heavily vascularized (Marinkovic et al., 1992) and contains high densities of ryanodine receptors (Gant et al., 2011), these signals may have been diluted or masked by white matter expression patterns in the prior tissue block study. Accordingly, the appearance of novel processes in the present study confirms and extends results of others (Ginsberg et al., 2010; Ginsberg et al., 2006; Liang et al., 2008) showing that laser dissection techniques can provide a higher resolution assessment of region-specific transcriptional profiles. Moreover, when contrasted with our prior data on hand-dissected CA1 that included white matter tracts, these selective gray matter analyses provide strong support for the view that the previously reported upregulated transcriptional, epigenetic, lipid transport, glial-immune and tumor suppressor responses in AD (Blalock et al., 2004), which were largely absent in the present study, are predominantly localized in and around the excluded white matter.
Comparison of the two studies also highlighted important technical issues. While there have been numerous investigations demonstrating that RNA extracted from FFPE tissue is suitable for quantitative assessment (Abdueva et al., 2010; Coudry et al., 2007; Farragher et al., 2008; Scicchitano et al., 2006; Turner et al., 2011), to our knowledge the present study is the first to make head-to-head comparisons of brain microarray data between FFPE tissue and frozen tissue blocks from the same human subjects. Apart from the apparent white-matter-related differences, there was substantial concordance between the two microarray studies, given that they were performed several years apart on samples collected and preserved very differently (laser dissected, formalin-fixed vs. hand dissected, frozen). Multiple gene networks altered in incipient AD and correlated with MMSE cognitive scores agreed closely across these studies, at both the gene and pathway levels (Overlap, Tables Tables22--4),4), particularly for downregulated neuronal processes.
Thus, this general concordance strongly indicates that the FFPE data are valid, and that the expression differences found between tissue block and laser dissection reflect, at least in part, differences in the cell types and sub-regions collected, rather than technical issues related to RNA integrity or extraction efficiency. Nonetheless, some non-specific procedural effects cannot be ruled out.
The conclusion that the FFPE material provides valid data is supported further by general agreement between the present LCM study and other gene expression analyses. One of the first studies of transcriptional profiles in AD selectively isolated tangle-bearing neurons (Ginsberg et al., 2000). Results showed significant reductions in genes associated with canonical AD neuropathology, including phosphatases/ kinases, glutamate and dopamine receptors, and cytoskeletal proteins, as well as upregulation of Cathepsin D. A subsequent study that surveyed multiple brain regions (Loring et al., 2001) found now-canonical upregulated inflammation, cell adhesion, cell proliferation and protein synthesis pathways in AD, as well as downregulated signal transduction, energy metabolism, stress response, synaptic vesicle synthesis/ function, calcium binding, and cytoskeletal pathways. An analysis of temporal gyrus found that downregulated synaptophysin directly correlated with, while downregulated alpha synapsin expression appeared to precede, AD-related changes (Pasinetti, 2001). Microarray analysis in the CA1 hippocampal subregion (Colangelo et al., 2002) found decreased transcription factor, neurotrophic factor, signaling, and synaptic activity along with increased inflammation in AD. A focused study of synapse-related gene expression changes in superior frontal gyrus (Yao et al., 2003) found downregulated dynamin 1 and syntaxin 1A expression. Studies of individually isolated brain neurons in AD have shown widespread downregulation of synaptic and cytoskeletal genes, as well as mitochondrial genes associated with reduced metabolism and altered expression of pathological hallmarks (Ginsberg et al., 2006; Liang et al., 2008). Moreover, a recent study in laser-dissected tangle-bearing neurons (Ginsberg et al., 2010) reported that Rab 5 and 7 upregulation correlated with cognitive decline during AD progression, supporting the hypothesis that increased endosomal activity, as reflected by Rab component upregulation, may enhance TrkB degradation and lead to neuronal dysfunction. Thus, among microarray studies of AD, there appears to be relatively good agreement across analyses, at least for downregulated changes in neurons and gray matter and for upregulated inflammatory responses. However, the expression signatures of different regions/tissues/ cell types, and their selective vulnerability in AD remain relatively unexplored (reviewed in Wang et al., 2010).
As noted, some of the most prominent incipient AD-related changes in our earlier study (increases of growth-related and oncogenic transcription factors, chromatin modifiers, lipid regulators and tumor suppressors (Blalock et al., 2004) were largely absent in the laser dissection material (Tables (Tables1,1, ,2,2, and and3;3; Supplemental Table 2). This result is consistent with our model of incipient AD (Blalock et al., 2004) in which aberrant transcriptional processes in oligodendrocytes are proposed to trigger cascades that affect other cell types, providing a potential mechanistic explanation for the apparent progression of AD pathology along myelinated tracts. Of course, additional studies will be needed to further elucidate the role of white matter transcriptional changes in incipient AD.
Another notable difference from our previous study and studies of others is the detection here of novel changes in potentially important functional processes, including upregulation of vasculature development and downregulation of ryanodine receptor stabilization, both of which correlated strongly with incipient AD. These processes may well have important mechanistic implications. For example, Ca2+ dyshomeostasis has long been suspected of a role in brain aging and AD (see Introduction) and it was recently found that downregulation or disruption of the immunophilin FKBP1b (and possibly, FKBP1a) can destabilize and increase ryanodine receptor Ca2+ release in hippocampal neurons, creating an aging-like phenotype (Gant et al., 2011). Here, FKBP1a, two junctophilins and CALM 3, which also participate in ryanodine receptor stabilization, were found to be significantly downregulated and correlated with incipient AD (Supplemental Table 1), as was FKBP1b in the previous study (Blalock et al, 2004; their Table 3). Thus, further studies investigating the significance of ryanodine receptor destabilization in the progression of AD appear warranted by the present results.
Similarly, the upregulation of genes related to vasculature development also may have mechanistic implications. Although many, or most, of the 48 correlated genes populating this category (see Supplemental Table 3) have roles in development and growth of other tissue types, all have also been linked to vasculature development. While the functional significance of this is difficult to discern at present, it may be relevant that angiogenesis is a well-established co-factor in tumor growth, and plays a key role in retinal macular degeneration (Ambati, 2011).
A major challenge for future research on molecular pathogenesis of AD will be to determine how, or if, these novel gray matter changes interact with the putative white matter alterations or, for that matter, with any of the multiple other processes previously identified in our studies and studies of others. Clearly, some of the alterations found may be secondary or incidental correlates of AD. Nonetheless, multiple processes identified in our studies were correlated with incipient as well as with late-stage AD, and consequently appear at least to be candidates for primary processes.
The overlap analysis (Fig. 3) reduces false positives, but does so at the cost of a likely increase in false negatives, in that a gene must meet the p ≤ 0.05 criterion in two studies (Blalock et al., 2005). Therefore, the ‘true’ agreement between these two studies may be substantially greater than reported here. Nonetheless, the more than 34x increase in genes agreeing between the two studies, compared to what should be expected by chance, illustrates that genes in the overlap analysis likely contain a very high proportion of true positive findings. It also should be noted that, because this approach assigns sets of genes to each exclusive Venn region based on a single p-value cutoff, genes attributed to one Venn region (Fig. 3, inset) may be borderline significant in another, effectively splitting a functional category across Venn regions. Upregulated immune activity, which appears in both the “overlap” and “tissue block” analyses may reflect this borderline effect.
Our results provide strong reciprocal validation for two technically dissimilar microarray analyses of Alzheimer’s disease. Gene expression values derived from laser dissected gray matter in formalin fixed, paraffin embedded specimens yielded a signature highly similar in many aspects to that measured in frozen tissue block specimens, including calcium regulation, synaptic, neuronal, inflammatory, and endosomal pathways- most of which have been identified in multiple studies (see section 4.1). Therefore, the discrepancies between the tissue block and laser dissection signatures did not appear to be due to random error, but to differences in the expression profiles of gray and white matter (the latter perhaps including the synaptic fields and glial processes immediately adjacent to myelinated tracts) . Although technical differences to some extent probably contribute to discordance between the two data sets, such effects appear minor and, further, it is highly unlikely that technical influences would exert themselves at the functional grouping level. That is, there is no a priori reason to expect that a particular functional category of genes would be more sensitive to formalin, laser dissection, or amplification protocols- these effects should seemingly distribute randomly, or at least, not partition discreetly within biological pathways. Thus, we conclude that the targeted difference in tissue components is the most likely and most parsimonious explanation for many of those aspects of the transcriptional profile that differed between our two studies.
Supplemental Table 1. Complete list of 3,465 genes in the laser-captured tissue that were significant in at least one correlation analysis (Fig. 2). Genes are listed in alphabetical order by gene symbol. Column: (A) Probe set- Affymetrix probe set ID; (B) Symbol- gene symbol; (C) Description- gene description; (D- G) Overall- correlation results across all 30 subjects; (H- K) Incipient- correlation across the subset of control and incipient subjects only; (D, H) MMSE r- Pearson’s r value for MiniMental Status Exam correlation; (E, I) MMSE p- p-value for adjacent r value; (F, J) NFT r- Pearson’s r value for neurofibrillary tangle correlation; (G, K) NFT p- p-value for adjacent r value. Note: p-value results are color-coded if significant (p ≤ 0.05), red for increased, and blue for decreased expression with worsening AD.
Supplemental Table 2. Complete list of all 3,344 genes significantly correlated with overall AD from among the 7537 total genes used in the Venn diagram overlap analysis between tissue block and laser dissection results (Fig. 3). Genes are listed alphabetically by symbol and are separated by Venn region (overlap, unique tissue block, unique laser dissection) and direction in the following rows: 4-235 overlapping downregulated; 236-512 overlapping upregulated; 513-1258 tissue downregulated; 1259-2417 tissue upregulated; 2418-2821 laser downregulated; 2822-3347 laser upregulated. Columns (A) Venn and (B) Dir give Venn region and direction of change, and columns C-E give probe set ID, gene symbol and description information. All correlation results are provided. (F-M) Tissue Block Results; (N-U) Laser Dissection Results; (F-I; N-Q)- Overall correlation results; (J-M; R-U) Incipient correlation results); MMSE- delineates MiniMental Status Exam related results, NFT describes neurofibrillary tangle results (the suffix ‘r’ refers to Pearson’s r value, the suffix ‘p’ refers to p-value calculated based on adjacent r value). Note: p-values are color coded if significant (p ≤ 0.05), red for increased and blue for decreased expression with AD severity.
Supplemental Table 3. Complete list of all 48 genes comprising the vasculature response category. Genes are listed alphabetically by gene symbol, along with descriptions, and overall MMSE and NFT correlation R values and p-values. Significant results are color coded red.
We thank Sonya Anderson and Ela Patel from the Sanders-Brown Center on Aging for sectioning and mounting hippocampal specimens, Jeff Gilbert and Zeiss Microscopy, as well as Woods Hole Research Center for generously allowing us to use their Zeiss PALM Microbeam system, Anjali Chhabra for performing RNA isolation and amplification protocols, and Dr. Kuey Chu Chen and Donna Wall of the University of Kentucky Microarray Core Facility for microarray hybridization and scanning. This work was supported in part by grants from the University of Kentucky Alzheimer’s Disease Center (AG028383) and the NIH Research Resources Center (S10RR024704) (to E.M.B.), and by grants AG034605, AG004542 and P01AG010836 from the National Institute on Aging (to P.W.L.).
1Abbreviations: AD- Alzheimer’s disease; CA- Cornu Ammonis; FFPE: Formalin-fixed, paraffin-embedded; LCM- laser capture microdissection; MMSE- mini-mental status exam; NFT- neurofibrillary tangle
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.