Tools and Resources

Neuroscience

Neuropathological and transcriptomic characteristics of the aged brain

Allen Institute for Brain Science, United States
University of Washington, United States
Kaiser Permanente Washington Health Research Institute, United States
University of California, Los Angeles, United States

Nov 9, 2017

Abstract
Introduction
Results
Discussion
Materials and methods
Data availability
References
Decision letter
Author response
Article and author information
Metrics

Abstract

As more people live longer, age-related neurodegenerative diseases are an increasingly important societal health issue. Treatments targeting specific pathologies such as amyloid beta in Alzheimer’s disease (AD) have not led to effective treatments, and there is increasing evidence of a disconnect between traditional pathology and cognitive abilities with advancing age, indicative of individual variation in resilience to pathology. Here, we generated a comprehensive neuropathological, molecular, and transcriptomic characterization of hippocampus and two regions cortex in 107 aged donors (median = 90) from the Adult Changes in Thought (ACT) study as a freely-available resource (http://aging.brain-map.org/). We confirm established associations between AD pathology and dementia, albeit with increased, presumably aging-related variability, and identify sets of co-expressed genes correlated with pathological tau and inflammation markers. Finally, we demonstrate a relationship between dementia and RNA quality, and find common gene signatures, highlighting the importance of properly controlling for RNA quality when studying dementia.

https://doi.org/10.7554/eLife.31126.001

Introduction

The population of the United States is aging, with the fastest growth in the very oldest part of the population where the number of nonagenarians and centenarians are expected to increase from 2 million to 10 million by 2050 (Corrada et al., 2012). This creates a significant public health challenge due to the increased health issues related to age, most notably the debilitating effects of neurodegenerative diseases. Dementia is thought to affect 11% of the US population over the age of 65 overall, and the incidence of dementia onset roughly doubles every five years to ~40–60% after age 90 (Corrada et al., 2012; Gardner et al., 2013; Alzheimer's Association, 2016). Approximately 2/3 of dementia cases are diagnosed as Alzheimer's disease (AD), representing an estimated 13.8 million cases by 2050 (Alzheimer's Association, 2016), with most of the remaining cases diagnosed as vascular dementia or mixed dementia from multiple etiologies. AD is characterized by stereotyped progressive neurodegeneration and accumulation of two misfolded proteins in brain regions important for cognition and memory. Hyperphosphorylated tau is thought to form intracellular neurofibrillary tangles (NFTs) initially in projection neurons of the entorhinal cortex, which then spread to the hippocampus and neocortex (Braak and Braak, 1991). Similarly, Aβ forms into extracellular plaques in cortical and deep brain structures (Mirra et al., 1991). In addition, Lewy bodies can be identified in cerebral cortex and deep nuclei as well as brainstem, and microvascular lesions can occur throughout the brain. While diagnostic grading systems for these pathologies have been developed that variably correlate with cognitive and behavioral function, there is no consensus on whether these microscopically observed disease pathologies are causal or effects of other underlying processes. Despite enormous efforts, no anti-tau or anti-amyloid therapies have been successful, and those limited treatment options targeting symptoms are based on acetylcholine or NMDA metabolism (Allgaier and Allgaier, 2014).

To further complicated diagnosis and treatment, pathologies associated with dementia are widespread in the aged brain even in the absence of dementia; for example, nearly half of non-demented participants in the 90+ Autopsy Study met pathological criteria for AD (Corrada et al., 2012). Among individuals with dementia, pathological phosphorylated tau (pTau) and amyloid beta (Aβ) pathology findings are actually lower in the 90+ year old age group than in the 60–80 year old group (Haroutunian et al., 2008), whereas other neuropathological conditions such as Lewy bodies and hippocampal sclerosis were only identified in individuals with dementia (Corrada et al., 2012). Indeed, while the overall incidence of dementia increases with age, pathology becomes much more variable, the relationship between disease pathologies and cognition weakens (Haroutunian et al., 2008; Corrada et al., 2012), and the relevance of canonical risk factors for AD, including APOE genotype, decreases with advancing age (Gardner et al., 2013). Identification of biological and environmental factors critical to the etiology and progression of neurodegenerative processes will be critical to developing preventive and therapeutic strategies in the aged brain.

Genome-wide gene expression analyses have been applied to identify molecular pathways affected by aging and dementia. Transcriptomics shows robust and stereotyped gene expression patterning in the brain, including spatial (brain region) (Hawrylycz et al., 2012) and temporal variation over the lifespan from development through adulthood into aging (Colantuoni et al., 2011) (http://brainspan.org). The aged brain shows increased variability in this transcriptional patterning compared to younger brains (Colantuoni et al., 2011). Comparing brains of people who died with a clinical diagnosis of AD to brains of people who died with no dementia, a number of studies have identified dysfunction of pathways and biological processes including synaptic transmission, energy metabolism, inflammation, cytoskeletal dynamics, signal transduction, transcription factors, and cell proliferation (Colangelo et al., 2002; Blalock et al., 2004; Webster et al., 2009; Miller et al., 2013). Many of these same pathways show disrupted gene expression in older compared with younger individuals not diagnosed with dementia (Miller et al., 2008), although in many cases to a lesser extent and in different brain regions (Avramopoulos et al., 2011). Since many of the gene expression studies to date focus on somewhat younger cohorts and have only limited information on disease pathologies, it is unclear whether robust relationships between gene expression and disease pathology or cognition extend to older individuals.

To better understand the relationship between cognition, brain pathology and injury, and gene expression in the aged brain, we created the Aging, Dementia, and Traumatic Brain Injury (TBI) Study, which is a detailed neuropathological, molecular, and transcriptomic characterization of brains of 107 people from the Adult Changes in Thought (ACT) cohort. The ACT study was designed as a population-based, prospective study of normal brain aging and dementia, and incorporates extensive medical history and postmortem characterization (Kukull et al., 2002; Larson et al., 2006; Crane et al., 2013). ACT participants entering the study are at least 65 years old and free of dementia, and the median age at death of the cohort used for this study is 90. This freely available resource (http://aging.brain-map.org/) presents a systematic and extensive dataset of study participant metadata, quantitative histology and protein measurements of neuropathology, and RNA sequencing (RNA-seq) analysis of hippocampus and neocortex. Here we describe this resource and initial analyses to understand features of the aged brain, the relationship between dementia and pathology, and transcriptional signatures of dementia, neuropathology and aging. Code and additional files required to reproduce all analyses are available in Github (https://github.com/AllenInstitute/agedbrain; Miller, 2017; copy archived at https://github.com/elifesciences-publications/agedbrain).

Results

A multimodal atlas of aging and dementia

The Aging, Dementia, and TBI Study was initially designed to study the long term effects of mild-to-moderate TBI, but we focus the current analysis on aging and dementia and present our results with respect to TBI elsewhere. This study includes 55 participants of the ACT study self-reporting TBI with loss of consciousness, along with 55 individuals matched for age, sex, and year of death who did not report a TBI with loss of consciousness. Donors in the exposure cohort reported between 1–3 lifetime TBIs with loss of consciousness ranging from <10 s to >1 hr (Figure 1A). Most participants were male (63 males, 44 females), with a wide range of educational backgrounds, and quite old (77–102 years old at time of death, median = 90), representing one of the oldest cohorts of its kind to date. Around half of the donors were diagnosed with dementia, including 30 with AD, 12 with dementia of multiple etiologies, and four with vascular dementia. More APOE ε4-positive participants had dementia (65%) than APOEε4 negative participants (40%), consistent with the role of this gene as a primary genetic risk factor for AD. Deposition of the disease pathologies pTau, in the form of NFTs (Braak stage), and Aβ, in the form of neuritic plaque density (CERAD score), ranged from absent through severe with relatively equal frequency (Figure 1A), and was generally higher in donors with dementia compared to donors without dementia, as expected (Table 1). It should be noted that this cohort is not representative of the ACT cohort as a whole (i.e., it is older, more heavily male, and all deceased). Analyses can be extrapolated back to the entire ACT cohort using weights (Haneuse et al., 2009) available on the online resource (http://aging.brain-map.org/download/index); however, since we did not observe substantially different results using these weights (data not shown) we choose not to use them in the analyses presented here.

Figure 1 with 2 supplements see all

Download asset Open asset

Experimental design and cohort characteristics.

(A) Demographics for all 106 donors (after excluding one outlier; Materials and methods). Histograms are shown for Age at death, Education (yrs), and Age at first TBI. All other metrics (except sex) are sorted from lowest to highest, with white corresponding to none (or 0 or control), and red corresponding to the highest severity of the condition or pathology. (B) Summary of all data available for each donor included in this study, including IHC on fresh and frozen tissue, RNA-seq analysis, and Luminex protein and isoprostane quantification. (C) Examples of histology from fresh frozen temporal cortex of donor H14.09.075 using IHC for Ab6E10 and AT8 and ThioS labeling, showing severe Ab and pTau pathology. Numbers indicate cortical layers. Descriptions of each metric (including abbreviations used) are included as a downloadable file on http://aging.brain-map.org/download/index. See also Figure 1—figure supplements 1–2.

https://doi.org/10.7554/eLife.31126.002

Table 1

Summary of demographics for donors with and without dementia.

P-values for upper seven and lower two metrics are uncorrected significance values from T-tests and hypergeometric tests, respectively (*p<0.05 after Bonferroni correction for multiple comparisons). Demographic summary includes 106 donors used in analysis (Materials and methods).

https://doi.org/10.7554/eLife.31126.005

Category	Non-demented		Demented
Category	Mean	SD	Mean	SD	P-value
Age at death (yrs)	89	7	90	6	2.8E-01
Education (yrs)	15	3	14	3	4.9E-02
Number of TBIs	0.6	0.7	0.7	0.7	8.0E-01
Age at first TBI	23	31	24	32	8.2E-01
Braak stage	2.8	1.5	4.1	1.7	8.9E-05*
NIA Reagan	1.4	0.7	1.9	0.9	1.8E-03*
CERAD score	1.2	0.9	1.8	1.2	1.4E-02
	Count		Count		P-value
Sex	20 F / 36 M		23 F / 27 M		1.0E-01
>0 APOE ɛ4 alleles	47 No/7 Yes		32 No/13 Yes		2.1E-02

For each donor, we collected tissue from four brain regions known to show neurodegeneration and pathology as a result of AD and Lewy body disease (LBD; hippocampus and temporal and parietal cortex) (Braak and Braak, 1991; McKeith et al., 1996; Hyman et al., 2012; Montine et al., 2012), hippocampal sclerosis and phospho (p) TDP-43 pathology (hippocampus and temporal cortex) (Nelson et al., 2016), chronic traumatic encephalopathy (CTE; temporal cortex and parietal cortex and white matter) (McKee et al., 2016), and microvascular brain injury (multiple regions) (Flanagan et al., 2016), and characterized each tissue in a highly standardized manner using a broad set of informative data modalities (Figure 1B). We used immunohistochemistry (IHC) on both fresh frozen and formalin fixed paraffin embedded (FFPE) tissue to stain and quantify proteins marking dementia-related pathologic findings, including pTau, Aβ, α-synuclein (Lewy bodies), and pTDP-43, as well as microglia (IBA1) and astrocytes (GFAP). For example, donor H14.09.075 (78 year old female with dementia) shows significant pathology of intracellular pTau (AT8) and extracellular Aβ plaques (Ab6E10 and ThioS) based on IHC of fresh frozen (Figure 1C) and fixed (Figure 1—figure supplement 1) tissue. Negligible amounts of α-synuclein were observed in these brain regions. In addition, we used multiplexed Luminex assays for protein molecular quantification of tau and pTau variants and Aβ species, as well as for α-synuclein, inflammatory mediators (cytokines and chemokines), neurotrophic factors, and other targets. We determined free radical injury in parietal and temporal cortex using GC/MS quantitation of isoprostanoids. We used in situ hybridization (ISH) to detect expression for canonical marker genes for astrocytes (AQP4, GFAP), oligodendrocytes (MOBP), and neuronal subtypes (RELN, SLC6A1, SLC17A7) to provide insight into the cellular makeup of tissues used for neuropathological and transcriptomic analysis. Finally, we used RNA-Seq to assess genome-wide expression levels of >50,000 coding and non-coding transcripts on macrodissected tissue sections from the same blocks used for fresh-frozen IHC and ISH. Brain region and sample RIN represent the largest source of transcriptional variation (Figure 1—figure supplement 2), as shown previously (Preece and Cairns, 2003; Li et al., 2004; Tomita et al., 2004; Mexal et al., 2006; Vawter et al., 2006; Hawrylycz et al., 2012); therefore, we treat RNA-seq data from each brain region independently and correct for RIN. After excluding poor quality or otherwise unusable tissue, data from 377 tissue blocks across four brain regions in 107 donors are available as part of the resource.

Widespread tau and amyloid beta pathology in the aged brain

NFTs and amyloid plaques are thought to progress in a stereotyped anatomical pattern with increased AD severity, but also appear to show a more general increased load in advanced aging (Mungas et al., 2014). We assessed disease pathology in this resource using two approaches. First, we used standard global metrics of disease severity by including NFT distribution (extent; i.e., Braak stage) (Braak and Braak, 1991) and neuritic amyloid plaque cortical density (i.e., CERAD score) (Mirra et al., 1991). Amyloid plaque distribution (Thal phase) (Thal et al., 2002) was not routinely available in the ACT study until 2012 and therefore was not included in the analysis. In addition to standard diagnostic neuropathological endpoints routinely assessed for each case, we used local measurements in the blocks used for RNA-seq. These included soluble protein (Luminex) and histological (IHC) analysis on adjacent frozen sections, as well as more standard histology on FFPE tissue sections from the same brain regions from the same or opposite hemisphere. To quantify pathology load in IHC we calculated the fraction of labeled pixels in representative regions of interest for each stain from each case, using a modification of a technique we previously developed for quantification of ISH signal (Dang et al., 2007; Lein et al., 2007). Quantitative scores are consistent with qualitative observations using this technique (Figure 2—figure supplement 1). Pathology quantifications based on frozen and fixed tissue are highly correlated (r = 0.78 for pTau and r = 0.67 for Aβ), although in some cases the threshold and dynamic ranges showed some variation between the two tissue preparations (Figure 2A and Figure 2—figure supplement 2). These quantified values also show regional patterns consistent with Braak stage and CERAD scores (Figure 2—figure supplement 3).

Figure 2 with 6 supplements see all

Download asset Open asset

Amyloid beta and tau pathologies show a relationship with age, but with high variability.

(A) Distributions of values for quantitative pathology metrics, separated by brain region (colors in legend). Lines are density plots (y-axis) of distributions of each metric (specified in x-axis label), with triangles indicating the average value. Note that several metrics have higher values in cortex than hippocampus, or vice versa. (B) Donors with higher levels of tau pathology (defined as Braak stage; top row) and of Aβ (defined as CERAD score; bottom row) were older on average (y-axis) than donors with lower measures of pathology. (C) Donors with dementia have higher levels of tau and Aβ pathology on average than donors without dementia, as measured both by global metric (Braak stage, CERAD score; left column), and local IHC quantifications in hippocampus (AT8, Ab6E10; right column). (D) Donors with higher levels of dementia pathology (x-axes; same metrics as in C) also tend to have lower cognitive scores (y-axes). For bar plots in **B-D**, dots indicate specific donors, and boxes and whiskers represent 25%/75% and 5%/95%, respectively. For scatterplots, dots indicate donors, with specific metrics shown on axes. See also Figure 2—figure supplements 1–6.

https://doi.org/10.7554/eLife.31126.006

We observe a wide distribution of pTau and Aβ pathology loads in this aged cohort ranging from no pathology to extremely high (Figure 2A). As expected, pathological tau (AT8 IHC and pTau Luminex) tended to be higher in hippocampus while Aβ (Ab6E10 IHC and Aβ42 Luminex) is higher in cortex, consistent with known AD pathological distributions and progression. Both pTau and Aβ pathologies (as measured by Braak stage and CERAD score) are found more widely distributed in the brain among people who died at older ages compared to those who died at younger ages, despite the relatively compressed age range of this cohort (Figure 2B). We find a statistically significant relationship between dementia and pTau pathology, shown in Figure 2C as both an increased anatomical distribution (Braak stage) and hippocampal pathology load based on multiple protein quantification metrics (e.g., AT8 IHC). Despite the known differences in pathological signatures in different disorders of dementia, we find the same results for all dementia cases and the subset of cases diagnosed with probable or possible AD (Figure 2—figure supplement 4). This is likely due to the predominance of AD cases in our cohort, particularly when including donors diagnosed with dementia of multiple etiologies (only eight donors were diagnosed with vascular or other dementia with no clinical diagnosis of AD). We also observed a correlation between pTau levels and age (but not Aβ) in donors without dementia (Figure 2—figure supplement 5). Phrased differently, there is a significant difference in pTau pathology between donors with and without dementia in the younger (<90 years) but not the older (90 + years) donors (Figure 2—figure supplement 6), consistent with a general increase in pTau pathology with age (Haroutunian et al., 2008; Corrada et al., 2012). These results provide support for the idea that different pathways and progressions may be involved in pathological processes in the oldest old (i.e., individuals > 90 years old).

As part of the ACT Study, participants are assessed every other year for cognitive status using the Cognitive Abilities Screening Instrument (CASI) to determine whether further assessment for dementia is necessary (Teng et al., 1999). Tau is significantly correlated with an Item Response Theory version of CASI score (CASI_irt; Figure 2D) that accounts for the uneven distribution of CASI item difficulty levels across the ability spectrum (Crane et al., 2008; Ehlenbach et al., 2010); furthermore, local metrics of pathology (AT8 IHC) associate more strongly with cognitive metrics than does Braak stage, demonstrating the value of these quantitative metrics. Measures of Aβ pathology show nominally significant, but less robust associations with disease and cognitive status (Figure 2B–D, bottom row), aligning with previous reports that pTau may be a better indicator of AD severity than Aβ (Nelson et al., 2007).

While pTau and Aβ pathologies were more common in people with dementia than cognitively normal older adults, these pathologies were highly variable across this cohort among those with and without dementia. For example, eight of the 32 donors (25%) with severe NFT pathology (Braak stage >= 5) and six of the 25 donors (24%) with severe amyloid pathology (CERAD score = 3) did not have dementia. This disconnect is dramatic in individual cases where extremely high or low pTau (AT8) pathology is found in donors with and without dementia (Figure 3A). Similar cases are found with Aβ pathology in the cortex (Figure 3B). These findings are consistent with and extend previous observations in the ACT cohort (Sonnen et al., 2007; Sonnen et al., 2011) and other community-based samples (Sonnen et al., 2011), and suggest that some individuals with very AD pathology are resilient to the effects of these pathologies, while others may develop dementia through other mechanisms. Identifying gene expression signatures of resilience and tolerance will be an important area of future study.

Figure 3

Download asset Open asset

Disconnect between pathology and dementia status.

AT8 (tau; A) and Ab6E10 (Aβ; B) IHC in high and low pathology donors with and without dementia, demonstrating individual variation in the relationship between pathology and dementia status. AT8 images of tau pathology are from the hippocampus (with matching Nissl-stained section below), while Ab6E10 images of Aβ pathology are from the parietal cortex. CA1, CA3: hippocampal subfields; DG: dentate gyrus. Numbers in B indicate cortical layers. Donor labels are indicated. Scale bar: 1 mm.

https://doi.org/10.7554/eLife.31126.013

Global and regional molecular signatures of inflammation do not correlate with age or dementia status

Inflammation occurs across a wide range of brain dysfunction, including acute TBI (Lu et al., 2009), AD (Akiyama et al., 2000) and normal aging (Franceschi and Campisi, 2014), due at least in part to disruption of the blood brain barrier (Popescu et al., 2009). We took several strategies to assess the range of inflammation across this cohort and the extent to which inflammation is generalized or shows regional specificity, including IHC for microglia (IBA1) and reactive and other types of astrocytes (GFAP), Luminex for cytokines and chemokines, and transcriptome data. Donors showed a continuous range of expression for IBA1 and GFAP, as well as inflammatory proteins based on Luminex assays (Figure 4A). A small number of cases (2–5 per region) showed exceptionally high levels, but the majority of cases showed continuous variation across a lower range (high cases excluded from Figure 4A to better show the distribution of lower values). Interestingly, individual inflammatory proteins showed regional specificity; for example, some proteins were enriched in hippocampus compared to cortex (TNF-A, IL-6, MIP-1A) or vice versa (MCP1, IL-7, RANTES) (Figure 4A). Regional heterogeneity in microglial gene expression has been described elsewhere, such as enrichment in TNF-A expression in rat hippocampus compared to cortex (Ren et al., 1999).

Figure 4

Download asset Open asset

Gene expression signatures of inflammation.

(a) Distributions of values for glia and for each Luminex variable marking inflammation, separated by brain region (labeling as in Figure 2A). (B) Correlation between gene metrics of inflammation (x-axis) and protein metrics of inflammation (y-axis) in each brain region. Gene signatures are defined as the module eigengene (ME) of the module with the largest enrichment for the GO term ‘inflammatory response’, and protein metrics are the truncated quantifications of the Luminex protein most highly correlated with each ME. (C) Gene (lower left) and protein (upper right) expression markers of inflammation are highly correlated between brain regions. Dots represent donors with x- and y-axes corresponding to the gene and protein values in B. Pairwise brain region correlations are shown below in each box. Blue circle and orange box correspond to donors in D. (D) IHC for IBA1 in a donor showing inflammation across regions (left, orange), and in a donor showing higher levels of inflammatory marker genes in hippocampus than cortex (right; blue). See also Figure 4—source datas 1–3.

https://doi.org/10.7554/eLife.31126.014

Figure 4—source data 1 Module assignments and associated module eigengene correlations for each gene in the four regional WGCNA networks.: https://doi.org/10.7554/eLife.31126.015
Download elife-31126-fig4-data1-v1.xlsx
Figure 4—source data 2 Module annotation for cell types. Cell type gene lists are derived from Zhang et al. (2014) and NeuroExpresso in the left and right panels, respectively, and enrichment p-values for each module are calculated using a hypergeometric test and are Bonferroni-corrected.: https://doi.org/10.7554/eLife.31126.016
Download elife-31126-fig4-data2-v1.xlsx
Figure 4—source data 3 Module comparison with demographic and pathology metrics. Module eigengene expression is compared with 24 demographic and pathology metrics (including Dementia/control, AD/control, sex, measures of tau and abeta pathology, and inflammatory markers) using SVA, and p-values are Bonferroni-corrected.: https://doi.org/10.7554/eLife.31126.017
Download elife-31126-fig4-data3-v1.xlsx

Despite this heterogeneity at the individual gene level, there is likely to be a generalized molecular pathway associated with inflammation that can be used to assess the degree of inflammation across regions. Indeed, previous genome-wide transcriptome studies have identified gene networks associated with microglia and inflammation in adult human brain (Oldham et al., 2008; Hawrylycz et al., 2012; Miller et al., 2013; Zhang et al., 2013). To identify similar networks in the current cohort we performed weighted gene co-expression network analysis (WGCNA) (Zhang and Horvath, 2005; Langfelder and Horvath, 2008) separately in each brain region (Materials and methods; see Figure 4—source data 1 for module assignments). This strategy identifies groups of genes with similar expression patterns in an unbiased manner, whose functional significance can be assigned by searching for overrepresented gene ontology (GO) terms. Here, we identified a network of genes in each region highly overlapping for markers of the GO term ‘inflammatory response’ (Benjamini and Hochberg corrected p<10⁻¹³ in all regions; ToppGene) and for cell-type specific markers of ‘microglia’ (Bonferroni corrected p<10⁻³⁵ in all regions) (Zhang et al., 2014) or ‘Microglial activation’ (Bonferroni corrected p<10⁻¹² in all regions) (Mancarci et al., 2016) (Figure 4—source data 2). As expected, these region-specific inflammation gene networks (HIP_M18, TCx_M18, PCx_M18, FWM_M13) show highly significant gene overlap (hypergeometric test, p<10⁻¹⁰⁰), despite being generated independently.

Coordinated gene expression levels within gene networks can be summarized by a module eigengene (ME). MEs of these gene networks are highly correlated with specific protein markers of inflammation in specific regions (Figure 4B; Figure 4—source data 3; p<10⁻⁵ in all regions; BH-corrected SVA p-values). For example, MIP-1α (gene CCL3) shows the best agreement between genes and protein in hippocampus, and IL-6 shows the most consistent patterning in cortex (despite having much higher protein levels in hippocampus). Furthermore, the well-known inflammatory gene STAT3, which is activated in mouse brain after induction of inflammatory responses using lipopolysaccharide (LPS) (Beurel and Jope, 2009), is one of the genes most highly correlated with the ME in each of these modules (Figure 4—source data 1). Using inflammation-related MEs as generalized measures of inflammation, we find high correlations (0.69–0.83; p<10⁻¹² for all comparisons) between regions indicating that inflammation is largely a global phenomenon (Figure 4C, lower panel). These correlations were significant but lower (0.45–0.88; p<4×10⁻⁴ for all comparisons) when correlating individual protein markers across regions (Figure 4C, upper panels). We also find some individual cases where inflammation shows regional specificity, both by MEs and using microglial immunohistochemical labeling with IBA1 (Figure 4D). This is not unexepcted, as several diseases show region-specific inflammatory responses: for example, the substantia nigra pars compacta is particularly susceptible to neurodegeneration due to inflammation in Parkinson's disease (Ji et al., 2008).

None of these inflammation-related gene networks have ME expression significantly correlated with any metrics for aging, cognition, dementia, or associated pathology in this cohort (Figure 4—source data 3; p=1 for all comparisons), although a link between inflammation and AD has been described in the literature (Akiyama et al., 2000). This discrepancy may be due to the advanced age of this cohort, as there is currently no consensus on the relationship between inflammation and dementia in the oldest old (Gardner et al., 2013).

Transcriptional markers of dementia-related pathology

Gene expression studies have identified dysfunction related to dementia phenotypes in a variety of biological pathways including synaptic transmission, energy metabolism, inflammation, myelin-axon interactions, protein misfolding, and transcription factors (Colangelo et al., 2002; Blalock et al., 2004; Webster et al., 2009; Miller et al., 2013), although most of these studies evaluated data from somewhat younger cohorts. In contrast with previous transcriptional studies of AD, we did not find any genes with significant differential expression between control and dementia (or AD) cases in any brain region (SVA, p<0.05, Bonferroni corrected; Figure 5A and Figure 5—figure supplement 1). We performed several sensitivity analyses that reinforced our conclusion that methodological details were not driving this result (see Materials and methods for details). In addition, none of the gene network MEs described above distinguish dementia cases from controls (Figure 4—source data 3).

Figure 5 with 1 supplement see all

Download asset Open asset

Gene expression signatures of dementia and related pathology.

(A) No significant gene expression differences between donors with and without dementia in hippocampus. The histogram shows the distribution of log₂(fold difference) expression levels (x-axis) between control and dementia donors. Numbers indicate how many genes have a fold change > 1.3 (red lines) and p<0.05. (B) Significant correlation between the ME of M16 (y-axis) and measures of tau (AT8 IHC) in hippocampus. (C) Significant correlation between protein quantification of IHC for GFAP (y-axis) and measures of pTau (AT8 IHC) in temporal cortex. (D) Genes show comparable relationships with tau in this and an earlier study of dementia. X-axis shows the correlation between gene expression and AT8 IHC in this study. Y-axis shows the correlation between quantifications of NFTs and gene expression in (Blalock et al., 2004). Dots represent genes, with black dots corresponding to genes in module M16. See also Figure 5—figure supplement 1.

https://doi.org/10.7554/eLife.31126.018

We extended our analysis to instead search for MEs significantly associated with transcriptional correlates of dementia-related pathologies. No MEs corresponded to any measure of Aβ or α-synuclein pathology, or to age. While the lack of gene expression differences in the brain with age may seem surprising, this result is in line with previous studies that have found a much larger difference in gene expression during middle age (approximately 40–70) than in aged adults (Lu et al., 2004; Berchtold et al., 2008). However, one gene network (M16, consisting of 660 genes) was significantly correlated with pTau burden in hippocampus (AT8 IHC levels; Figure 5B), and showed similar but not statistically significant trends to other pTau metrics (Figure 4—source data 3). This set of genes is expressed predominately in astrocytes (Bonferroni corrected p<10⁻⁷) and microglia (p<10⁻¹⁵), potentially marking increased gliosis in hippocampus with increasing tau pathology (Figure 4—source data 2); it should be noted that this is not the same gene module associated with inflammation and activated microglia discussed above. Interestingly, several of the hub genes of M16 (genes most highly correlated to the module eigengene) are known to be involved in Aβ processing. For example, ITPKB shows higher expression in human AD than control brain and increases apoptosis and Aβ peptide production in mouse Neuro-2a neuroblastoma cells (Stygelbout et al., 2014). Similarly, SNX33 increased cleavage of APP alpha-secretase in cultured cells at the cell surface (Schöbel et al., 2008), while LRP10 overexpression diverts accumulation of mature APP from the cell surface to the Golgi apparatus, reducing Aβ production (Brodeur et al., 2012). Why genes associated with tau metrics would be related to Aβ processing is unclear, but it provides an interesting link between these two pathologies.

Quantifications of IHC for GFAP protein are correlated with pTau burden in temporal cortex, supporting a relationship between tau pathology and reactive astrocytes in this region (Figure 5C). Several other modules show significant enrichment for markers of astrocytes, neurons, oligodendrocytes and microglia (Figure 4—source data 2), but the MEs of these modules are not associated with dementia or any related pathology (Figure 4—source data 3), in contrast to previously published studies (e.g., [Miller et al., 2013]).

To assess whether the gene-pTau trends observed here match prior reports, we calculated the correlation between each gene and AT8 IHC levels, and compared these values with correlations between gene expression and reported levels of pTau in the hippocampal CA1 region (Blalock et al., 2004). The two studies agree well, with a correlation over all genes of R = 0.49 (Figure 5D, genes in M16 in black). Together, these results recapitulate the reported relationship between astrocyte and microglia-related gene expression and pTau pathology in hippocampus, but fail to identify genes related to dementia status.

Dementia-related gene expression associated with variation in RNA quality

Our failure to identify genes significantly related to dementia status was surprising, given that many studies have shown differential gene expression with AD (Colangelo et al., 2002; Blalock et al., 2004; Liang et al., 2008; Webster et al., 2009; Avramopoulos et al., 2011; Miller et al., 2013; Zhang et al., 2013; De Jager et al., 2014; Satoh et al., 2014; Allen et al., 2016). However, we found an inconsistency in how these studies normalized for tissue quality as measured by the pH or RIN scores of the tissues analyzed; in fact, only a few of them corrected for RIN at all. Repeating our analysis without accounting for RIN, we find a large fraction of genes (11%) to be differentially expressed between dementia cases and controls in at least one region (B & H corrected p<0.05, log₂(FC) >1.3; Figure 6—source data 1), leading to apparently larger fold changes between conditions (Figure 6A; compare with Figure 5A). We once again find very few genes significantly associated with dementia when we perform this analysis on a subset of 70 donors in our cohort matched for RIN, sex, and dementia status (Materials and methods; two or fewer in each region). This result suggested a direct link between RNA quality and dementia status. Indeed, we find a substantially lower RNA quality in dementia cases vs. controls in all four brain regions (Figure 6B; Figure 6—figure supplement 1B), and this difference was not related to the time between death and autopsy (PMI <8 hr, all donors).

Figure 6 with 2 supplements see all

Download asset Open asset

Differences in RNA quality between dementia and controls greatly impact gene expression results.

(A) Gene expression differences between donors with and without dementia in uncorrected data. Histograms show the distribution of log₂(fold difference) expression levels (x-axis) between control and dementia donors in two brain regions (hippocampus, left; temporal cortex, right). Numbers indicate how many genes have a fold change > 1.3 (red lines) and p<0.05. (B) RNA quality in donors with dementia (Dem.; right bars) is significantly lower than in non-demented controls (CT; left bars) in all four brain regions. Y-axes are RIN values. Plots as in Figure 2B–D. (C) Gene expression levels for many genes are highly correlated with RIN, with more showing lower expression with lower RNA quality (positive values) than with higher RNA quality (negative values). Histograms show the distribution of RIN correlations in two brain regions. Numbers indicate how many genes have R > 0.5 (red lines) and p<0.05. (D) Rank order of fold differences between controls and dementia cases is largely unchanged after controlling for RNA quality. Ranked fold differences on the x- and y-axes correspond to Figure 6A and Figure 5A, respectively. Dots indicate genes and are color-coded by density. (E) Genes with higher or lower expression levels in people with dementia compared with cognitively normal older adults from 12 brain regions in eight previous studies (rows) are related to dementia diagnosis and RNA quality in this study. Horizontal tics show the 25th percentile, median, and 75th percentile rank of the indicated dementia-related list in our current data set. Gene expression levels from genes lower in low RIN samples are also lower in AD samples from the comparison studies (red, solid lines are shifted towards 1), while gene expression levels from genes higher in high RIN samples are also higher in AD samples from the comparison studies (red, dotted lines are shifted towards 0). Gene expression results accounting for RIN (green) generally agree less well between studies than results not accounting for RIN (blue). See also Figure 6—figure supplements 1–2 and Figure 6—source datas 1–6.

https://doi.org/10.7554/eLife.31126.020

Figure 6—source data 1 Log2 fold changes between AD vs. control and dementia vs. control for each gene in all four brain regions, along with associated SVA p-values. Results for both RIN corrected and uncorrected data are shown.: https://doi.org/10.7554/eLife.31126.023
Download elife-31126-fig6-data1-v1.xlsx
Figure 6—source data 2 Correlations between each gene and RNA quality (RIN) with associated SVA p-values in all four brain regions.: https://doi.org/10.7554/eLife.31126.024
Download elife-31126-fig6-data2-v1.xlsx
Figure 6—source data 3 Significant GO terms for genes increasing or decreasing expression with decreasing RNA quality. Up to 50 significant categories for molecular function (MF), cellular component (CC), and biological process (BP) are shown, after correction for multiple comparisons.: https://doi.org/10.7554/eLife.31126.025
Download elife-31126-fig6-data3-v1.xlsx
Figure 6—source data 4 Description of nine previous studies comparing AD vs. control, including details of how gene lists used in paper were derived.: https://doi.org/10.7554/eLife.31126.026
Download elife-31126-fig6-data4-v1.xlsx
Figure 6—source data 5 List of 26 gene lists from the nine above publications that are used in Figure 6E.: https://doi.org/10.7554/eLife.31126.027
Download elife-31126-fig6-data5-v1.xlsx
Figure 6—source data 6 Significance of association between each of the first 25 principal components and every assessed metric in the un-normalized and RIN-normalized data. SVA P-values (Bonferroni corrected) are shown. Only metrics with at least one significant association in any region in either the un-normalized or RIN-normalized data are included in the table.: https://doi.org/10.7554/eLife.31126.028
Download elife-31126-fig6-data6-v1.xlsx

We next repeated this comparison on data from four additional population-based cohorts as part of the AMP-AD knowledge portal (https://www.synapse.org/ampad; (Bennett et al., 2012a; Bennett et al., 2012b; Allen et al., 2016)), and compared these with previous reports (Colangelo et al., 2002; Preece and Cairns, 2003; Durrenberger et al., 2010; Zhang et al., 2014). In five of the eight additional data sets assayed, donors with AD had significantly lower RIN than donors without dementia (Table 2). Donors with AD had significantly higher RNA quality in only one study (Allen et al., 2016). Thus the link between dementia status and RNA-quality is a broader phenomenon that is not unique to the ACT cohort, but is also not ubiquitous.

Table 2

RNA quality assessment for donors with and without AD in multiple studies.

https://doi.org/10.7554/eLife.31126.029

Data set	Brain region	Count (Control\|AD)	RIN/pH in control	RIN/pH in AD	P-value
ACT cohort (current data set)	TCx	50\|29	6.87 ± 0.92	6.18 ± 1.27	p=1.48×10⁻²
ROS (Bennett et al., 2012a)	TCx	107\|136	7.21 ± 1.01	7.06 ± 0.95	p=0.23 (ns)
MAP (Bennett et al., 2012b)	TCx	94\|120	7.25 ± 1.07	6.79 ± 0.96	p=1.18×10⁻³
MSBB *	BA36	98\|169	6.51 ± 1.30	5.86 ± 1.59	p=1.10×10⁻³
Mayo Study (Allen et al., 2016) ^†	TCx	31\|82	7.64 ± 1.21	8.59 ± 0.55	p=1.98×10⁻⁴
(Colangelo et al., 2002)	CA1	6\|6	6.75 ± 0.1	6.76 ± 0.1	ns
(Preece and Cairns, 2003) ^$	Cortex	81\|90	~6.5	~6.4	p<10⁻²
(Durrenberger et al., 2010) ^‡,$	Brain	72\|12	~7.1	~5.9	p<10⁻⁴
(Zhang et al., 2014)	PFc	101\|129	7.31 ± 0.47 6.55 ± 0.34	7.12 ± 0.56 6.37 ± 0.32	p=6.78×10⁻³ p=1.22×10⁻⁴

*=we are defining AD as CDR >1 in this data set, but note that the result holds for other cutoffs, †=only control donors from the Mayo Brain Bank Dickson were considered, ‡=Tissue collected from multiple brain regions and multiple brain banks. RNA quality is assessed with either RIN (black text) or pH (blue text) in each study. $=RIN/pH values are estimated from plots. Calculated p-values in this table are two tailed student t-tests uncorrected for multiple comparisons, and p-values from previous studies are as reported.

To assess the impact of RNA quality on gene expression levels, we compared gene expression to RIN. We find that 47% of expressed genes are correlated with RIN (Benjamini and Hochberg corrected p<0.05, RIN correlation >0.5; Figure 6C; Figure 6—source data 2)—in some cases accounting for 80% of the variation in gene expression—and that failure to account for RNA quality reduces our ability to separate samples by brain region (Figure 6—figure supplement 2). This result mirrors other studies that have shown RNA quality dramatically impacts measured gene expression levels (Preece and Cairns, 2003; Li et al., 2004; Tomita et al., 2004; Mexal et al., 2006; Vawter et al., 2006; Atz et al., 2007). However, we extend this result to show that the set of genes positively correlated with RIN are enriched for pathways found to be disrupted with AD in prior gene expression studies (Colangelo et al., 2002; Blalock et al., 2004; Webster et al., 2009; Miller et al., 2013), including the GO terms ‘mitochondrion organization’ and ‘RNA processing’ (Benjamini and Hochberg corrected p<10⁻³⁰; Figure 6—source data 3). But is the entire relationship of genes to dementia a function of RIN? To test this, we rank ordered genes by fold change difference between controls and dementia cases with and without correcting for RIN. In all regions, we found a significant correlation (R ~ 0.8) between RIN-corrected and uncorrected rankings (Figure 6D), indicating that much of the variance is not explained by RIN.

These results suggest that differential gene expression in neurodegeneration may include both the contributions of chronic conditions (e.g., dementia) and acute conditions (e.g., agonal stress), and/or other factors impacting RNA quality. To test whether previously published studies of AD could show similar effects, we compared gene measures of RNA quality (RIN) and of dementia status (before and after accounting for RIN) in our data set with AD-related genes from nine previous studies (Figure 6—source datas 4–5) (Colangelo et al., 2002; Blalock et al., 2004; Liang et al., 2008; Webster et al., 2009; Avramopoulos et al., 2011; Miller et al., 2013; Zhang et al., 2013; Satoh et al., 2014; Allen et al., 2016). First and foremost, we found that many of the genes most highly associated with dementia status were shared between studies, whether we controlled for RIN or not (Figure 6E, green and blue), confirming previous reports. However, genes that had lower levels of expression among people with AD tended to also have lower expression in donors with low RNA quality (Figure 6E, red), in some cases agreeing better with RIN reported herein than with dementia status reported herein. The converse was also true, although the effect was less robust. Furthermore, in nearly every case, our dementia-related genes identified without accounting for RIN agree better with gene lists from previous studies than those identified when we did account for RIN. The same results hold when we repeat this entire analysis considering only the subset of dementia cases with AD (Figure 6—figure supplement 1). These results, which were consistent across studies implementing a wide variety of experimental designs and strategies for controlling for RNA quality (or not), demonstrate a strong relationship between transcriptional changes in neurodegenerative disease and RNA quality, although it is important to note that at least two of these studies (Zhang et al., 2013; Allen et al., 2016) identified several hundred dementia-related genes after controlling for RIN, indicating that relative contribution of neurodegenerative disease and RNA quality to gene expression may differ between studies. Finally, we note that all our results hold when considering the top 25 principal components instead of differential genes or WGCNA modules (Figure 6—source data 6). For example, exactly one PC (PC1 or PC2 for all regions) is significantly associated with dementia status in the unnormalized data, while none are associated with dementia in the RIN-normalized data.

Discussion

The Aging, Dementia and TBI Study aims to provide the research community with an open access multimodal resource for studying relationships between cognition, neurodegeneration, inflammation, and injury in the aged brain (http://aging.brain-map.org/). This resource includes gene expression, protein quantification of neuropathology and inflammatory molecules, and histology on markers for cell type and neuropathology in four brain regions from 107 well-characterized donors from the ACT cohort. This unusually broad and systematic study allows a variety of analyses to correlate these features and identify associated molecular pathways. We confirm known associations between pTau and Aβ pathologies and dementia, and identify sets of co-expressed genes correlated with tau pathology and inflammation. The advanced age of the ACT cohort presented much higher variability than is seen in somewhat younger cohorts, which may have led to our difficulties in identifying significant gene signatures associated with aging, dementia and neuropathological markers, since these relationships appear to be stronger in younger individuals. Furthermore, we confirm a systemic relationship between dementia and RNA quality, showing a strong overlap between genes whose levels are affected by RNA quality and genes previously reported to be associated with AD. This study illustrates the challenges posed by the high degree of biological variation in the aged brain, and provides a resource that will facilitate future efforts to understand the nature of this variation.

We find a relationship between pTau (and to a lesser extent Aβ) neuropathology and cognitive status, based both on global pathology metrics (e.g. Braak score) and local hippocampal and cortical quantification of pathological tau (e.g., AT8 IHC), even in patients older than 90 years old. Interestingly, the strongest correlations between tau pathology and cognitive scores, dementia status and gene expression were observed with pTau protein quantification based on automatic image analysis of IHC data. This strategy for image quantification, which has been successfully applied at a large scale to study gene expression patterns in mouse (Lein et al., 2007), has the potential to provide an unbiased and informative method for studying pathology. A prior study of 390 donors also found a strong relationship between antemortem global cognitive ability (MMSE score) and counts of NFTs, and a weaker correlation with neuritic plaques (Nelson et al., 2007). However, the connection between neuropathology and cognitive status was weaker in the current cohort and there were many cases with a striking disconnect. For example, 25% of donors with Braak stage >= 5 did not have dementia. A major reason for the weaker relationship between pTau pathology and cognitive status appears to be higher levels of tau pathology associated with advanced age that is not related to dementia (Haroutunian et al., 2008; Corrada et al., 2012). This finding could be related to primary age-related tauopathy (PART), a recently described age-associated pathologic entity, usually in non-demented individuals, characterized by pTau-related neurofibrillary degeneration in the absence or paucity of Aβ pathology (Crary et al., 2014). As a new entity, PART has not been widely assessed in the ACT study autopsy cohort, and the co-existence of PART and AD, particularly in the oldest old, is poorly characterized, although there is some evidence of co-occurrence of these diseases (Mungas et al., 2014). The influence of mixed pathologies on cognition is poorly understood and an area of active investigation. The relationship between tau pathology and dementia status in our study was not observed in donors > 90 years old, highlighting the complexities of neuropathological processes in the oldest old and demonstrating resilience to the effects of pathology (‘tolerance’) in a substantial subset of aged individuals.

Neuroinflammation, predominantly in the form of innate immune activation via microglia, has been reported to occur across normal aging and in many neurodegenerative disorders, including AD (Akiyama et al., 2000; Franceschi and Campisi, 2014). Here, we took advantage of the wide range of molecular analyses to search for such correlations in the aged brain. A challenge we found is that individual protein markers of inflammatory pathways behave differently in different structures and differently from each other (Ren et al., 1999; Ji et al., 2008; Wang et al., 2015). To identify more global signatures of inflammatory pathways we used gene network analysis to identify sets of co-expressed genes enriched for inflammatory genes. This approach identified inflammatory gene networks in each brain region that were correlated with the most informative individual protein markers in those regions. The gene expression levels of these networks were highly correlated across brain regions, indicating that neuroinflammation is in many cases largely a global phenomenon across brain regions, although there were individuals with regional enrichment of inflammatory signatures. Neither the individual inflammatory markers nor gene expression levels of the inflammatory gene network correlated with either aging or dementia status.

We identified important variation in the RNA quality of tissues analyzed. Surprisingly, we found a statistically significant relationship between RNA quality and dementia status in donors from several independent cohorts. Why might RNA quality vary with dementia status? One possibility is that RNA quality reflects antemortem conditions, and control donors are more likely to expire from sudden, unexpected causes than donors with neurodegenerative conditions requiring long-term care (Monoranu et al., 2009; Mills et al., 2014). A number of studies have shown that gene expression levels can vary dramatically based on end-of-life conditions (Li et al., 2004; Tomita et al., 2004; Atz et al., 2007; Monoranu et al., 2009; Durrenberger et al., 2010). For example, Li and colleagues used unbiased clustering of brain tissue from multiple regions to group donors into two types differing by brain pH and agonal duration (Li et al., 2004). Remarkably, 30–50% of all genes differentiated these two types, including many markers for oxidative stress which have lower expression in donors with lower RNA quality (Vawter et al., 2006). We find similar results: 47% of expressed genes are correlated with RIN, including many in mitochondrial-related pathways whose levels decrease with RIN, although information regarding end of life conditions is unavailable for ACT cohort donors.

Relationships between gene expression, RNA quality, and agonal state markedly complicate studies of neurodegenerative diseases (Monoranu et al., 2009; Mills et al., 2014). Here we find a strong overlap between signatures of AD and RNA quality, with most (but not all) of the transcriptional variability accounted for by RIN rather than disease status. To the best of our knowledge, this is the first study directly comparing genes associated with RNA quality and dementia status. Furthermore, we find that nearly all reported AD gene lists (regardless of treatment for RNA quality) are strongly enriched for gene expression in genes associated with both RIN and AD status in this study, and that accounting for RNA quality in this study markedly decreases the agreement between studies. Importantly, all previous studies included in our analysis that did control for RNA-quality identified many dementia-associated genes, suggesting that the relative impact of RNA-quality in our data set may be exaggerated. Nevertheless, these results suggest a potential convergence of gene pathways involved in agonal state and dementia, and highlight the importance of carefully accounting for technical variables—particularly RNA quality—when studying neurodegenerative diseases.

Materials and methods

Participant information and consent

Request a detailed protocol

All work was performed according to guidelines for the research use of human brain tissue. Participants signed informed consent forms at enrollment that includes permission for sharing de-identified data, and signed additional consent forms for the autopsy that included data and tissue sharing. Autopsy consents were updated for all subjects with the legal next of kin after death. All study procedures were reviewed and approved by Institutional Review Boards at Kaiser Permanente Washington and the University of Washington. Non-identifying information about each 107 participants (i.e., age, sex, etc.) is publicly available under the ‘Specimens’ tab at http://aging.brain-map.org/.

ACT cohort

Request a detailed protocol

ACT is a prospective, longitudinal study of randomly-selected, cognitively normal participants of Kaiser Permanente Washington in the Seattle area that were willing to volunteer for the study (Kukull et al., 2002; Larson et al., 2006; Crane et al., 2013). Kaiser Permanente Washington is an integrated staff-model Health Maintenance Organization (HMO). At enrollment and at follow-up visits every two years, ACT research staff members administer the Cognitive Abilities Screening Instrument (CASI) (Teng et al., 1999) and participants with CASI ≤85 receive secondary follow-up with a clinical evaluation and a comprehensive neuropsychological battery. Results of these evaluations and clinical data are reviewed in a multidisciplinary consensus conference which uses standardized criteria to diagnose incident dementia (American Psychiatric Association, 1994) and AD as well as other neurodegenerative diseases when applicable (McKhann et al., 1984). Total enrollment as of December 2015 was approximately 5100 people, including more than 600 participants who have donated their brains. Requests to access other data from the ACT cohort should be addressed to KPWA.actproposals@kp.org.

Tissue collection and utilization

Request a detailed protocol

A team from the University of Washington (UW) Neuropathology (NP) Core is contacted soon after death to perform brain autopsies of ACT subjects. ACT study staff ask participants at enrollment and every study visit whether they have experienced a loss of consciousness (LOC) and, if so, what caused it, such as electrocution, near drowning, or head injury (TBI). If updated autopsy consent is obtained and the brain can be removed and dissected with a postmortem interval (PMI) <8 hr, a rapid autopsy is performed. During a rapid autopsy, ventricular cerebrospinal fluid (CSF) is taken, the brain is hemisected along the mid-sagittal plane and dissected and ~60 flash frozen tissue samples from at least 12 brain regions are collected, flash-frozen in liquid nitrogen, and stored at −80°C. The unsampled hemibrain, and all remaining non-frozen tissue from the contralateral hemisphere is then fixed in 10% normal buffered formalin for approximately 2–3 weeks. Fixed tissue from every brain undergoes a thorough neuropathological examination where 22 standard samples, in addition to samples of any focal lesions or abnormalities, are dissected and submitted for routine processing for formalin-fixed paraffin-embedded (FFPE) sections.

This project was initially designed to study the long-term effects of TBI exposure, and participants were selected on the basis of exposure or lack of exposure to TBI. All ACT subjects with a TBI with loss of consciousness (LOC) and rapid autopsy with available banked frozen tissue were identified, and then each TBI with LOC donor was matched for sex, age, year of death, and finally PMI to an individual in the ACT autopsy sample without a history of TBI with LOC. Once a subject was included in this study, two adjacent flash frozen tissue blocks from parietal lobe, temporal lobe, and hippocampus were removed from the NP Core repository; one was sent to the Allen Institute for IHC, ISH and RNA -seq, and the other processed at the University of Washington (UW) for immunoassays (Luminex) and gas chromatography-mass spectrometry (GC/MS). If two blocks were not available, the remaining block was either divided (cortex) or prioritized for Allen Institute studies (hippocampus).

Tissue processing for histology and immunoassays

Request a detailed protocol

Frozen tissue at UW was divided evenly (while frozen, in the sagittal plane through the long axis of the gyrus) for GC/MS (isoprostanes), where the entire piece was used, and for immunoassays (Luminex), which was run on 1 cm punch biopsies from the depth of sulcus cortex (gray matter) in parietal and temporal lobe and through deepest subcortical white matter (in the parietal lobe sample). Due to relative paucity of available tissue, the entire hippocampus tissue block was submitted for immunoassays. For IHC of FFPE tissues, blocks were taken from either the same (cortex) or the opposite (hippocampus) side of the brain that was sampled for frozen tissues (although not from adjacent blocks), and were submitted for sectioning, histochemical, and immunohistochemical staining. Slides were then sent to the Allen Institute for scanning and image analysis as described below. Frozen tissue sent to the Allen Institute was cryosectioned into a series of 25-micron-thick sections that were designated for histological staining (Nissl, ISH, IHC, and Thioflavin-S) and for RNA-seq (see below). Following sectioning, histological stains were processed according to standard protocols as previously described (Sun et al., 2002; Lein et al., 2007). GC/MS was quantified as described previously (Montine et al., 2005). Specific assays run for GC/MS, Luminex, ICH, and ISH are presented in Figure 1. IHC markers for paired helical filament pTau (AT8) and Aβ (Ab6E10) were processed on both fresh frozen and FFPE tissue. Note that a broader marker for pTau, Tau2, was processed only for FFPE tissue (see Figure 1—figure supplement 1 for an example staining).

Image processing and quantification

Request a detailed protocol

Nissl, H&E-LFB, ISH and IHC slides were scanned at 10x full resolution using a Leica ScanScope scanner, while Thioflavin-S slides were scanned at 10x full resolution using an Olympus VS110 scanner. An Informatics Data Pipeline (IDP) managed image preprocessing, image QC, IHC expression detection and measurement, Nissl processing, annotation QC and public display of information via the web application, as described previously (Dang et al., 2007), with some modifications and additions for processing images for this project. For ISH and IHC slides, respectively, masks highlighting areas with enriched gene expression or immunoreactivity were generated using adaptive detection/segmentation image processing algorithms. Images that were out of focus after rescanning or with technical or tissue artifacts obscuring the target anatomical region were then failed and excluded from public release. For each set of gene images available in the online viewer, the nearest set of Nissl-stained sections (and other histological data) can be accessed and viewed. To generate quantitative image metrics for IHC, macrodissection sites as delineated on the Nissl images were used to annotate regions of interest (ROIs) on each of the near-adjacent IHC images. The ROI was then adjusted if there were technical artifacts that would affect the evaluation of pathology. The expression density, defined as the percentage of area within the ROI that was occupied by the IHC reaction product, was then assessed using an adaptive detection/segmentation technique which algorithmically determines whether each pixel in an ROI contains the IHC stain (see the Expression Detection Module section in the Informatics Data Processing paper in the Allen Mouse Brain Atlas Documentation tab for more details; http://mouse.brain-map.org/static/docs). For stains with very low expression densities, ROIs that were identified as outliers were visually inspected and then adjusted or excluded as necessary. Good correlations were seen between quantifications of antibodies for amyloid beta and pTau in FFPE and fresh frozen tissue, indicating good agreement between these two measures of pathology (Figure 2—figure supplement 2).

RNA-Seq tissue and RNA processing

Request a detailed protocol

Collection of tissue samples from temporal and parietal neocortex, parietal white matter, and hippocampus was done by manual macrodissection. Specific areas for macrodissection were identified by neuroanatomists using images of Nissl-stained tissue sections immediately adjacent to the sampled tissue, and were excised from the remaining tissue frozen tissue block using a scalpel. Tissue was immediately transferred to prepared tubes where RNA was isolated using the RNeasy Lipid Tissue Mini Kit (Qiagen #78404) as per manufacturer’s instructions. RNA was then quantified on a Nanodrop 8000 spectrophotometer (Thermo Scientific, Wilmington, DE) and normalized to 5 ng/μl before RNA QC was performed using a Bioanalyzer (Agilent Technologies) and RNA Integrity Number (RIN) was recorded. Total RNA (250 ng) was used as input into the Illumina TruSeq Stranded Total RNA Sample Prep Kit (RS-122–2203), which uses random hexamer first strand cDNA synthesis and includes rRNA depletion (Ribo-Zero Gold rRNA depletion kit to remove both cytoplasmic and mitochondrial rRNA) and fragmentation. At the time of project inception, this sequencing strategy provided the most reliable option for quantification of transcriptomic reads from tissue of widely varying quality, allowing the broadest inclusion of donors from the ACT cohort. External RNA Controls Consortium (ERCCs) (Baker et al., 2005) at a 1:10,000 dilution were spiked into each sample. RNA sequencing was done on Illumina HighSeq 2500 using v4 chemistry, producing a minimum of 30M 50 bp paired-end clusters per sample. Expression Analysis, Inc. (Morrisville, NC) performed both the TruSeq Stranded Sample Prep as well as the Illumina sequencing. All samples, regardless of RNA quality, were sent for sequencing. In total 377 samples from 107 donors passed all QC metrics and are included as part of the resource. Nearly all of the missing 51 samples were excluded because tissue was unavailable from the brain bank or because it completely failed in sequencing. A few samples were failed because their average inter-array correlation across all genes was several (usually but not always 3) standard deviations below the mean of all other samples from the same brain region. This strategy has been used to fail samples in other Allen Brain Atlases and is useful for ensuring that results are not driven by outliers. For this analysis, we removed one additional sample from the data set that showed high expression of Y chromosome genes but that was collected from a genetically-confirmed female who was documented to have previously given birth (donor H14.09.011).

RNA-Seq data alignment and normalization

Request a detailed protocol

Raw read (fastq) files were aligned to the GRCh38.p2 human genome (current as of 01/15/2016). Illumina sequencing adapters were then clipped using the fastqMCF program (Aronesty, 2011), and then mapped to the transcriptome using RNA-Seq by Expectation-Maximization (RSEM) (Li et al., 2010) using default settings except for two mismatch parameters: bowtie-e (set to 500) and bowtie-m (set to 100). RSEM aligns reads to known isoforms and then calculates gene expression as the sum of isoform expression for a given gene, assigning ambiguous reads to multiple isoforms using a maximum likelihood statistical model. Reads that did not map to the transcriptome were then aligned to the hg38 genome sequence using Bowtie with default settings (Langmead et al., 2009), after which remaining unmapped reads were mapped to ERCCs. Anonymized BAM files (where sequence-level information has been removed) for both transcriptome- and genome-mapped reads, and gene-level quantification (transcripts per million (TPM), fragments per kilobase per million (FPKM), and number of reads) are available as part of the resource (see Download tab).

For analysis, the FPKM data matrix was first adjusted for the total transcript count using TbT normalization (Kadota et al., 2012), which scales each sample based on the summed expression of all genes that are not differentially expressed. The differential expression vector was defined as TRUE if a sample was from either temporal or parietal cortex, and FALSE otherwise. Sample data were then log-transformed and scaled such that the total log₂(FPKM + 1) across the entire data set remained unchanged after normalization. The result was that all expression levels for a particular sample were multiplied by a scalar close to 1 (in most cases between 0.9–1.2).

The amount of variation explained by each demographic and tissue source was estimated using MDMR (Zapala and Schork, 2012). Specifically, Pearson correlation-based distances were calculated between each pair of samples (D_xy = 1-corr(x,y)), and a matrix of these values and of each demographic and tissue source variable was input into the MDMR R function as a univariate model with 100 permutations. Resulting percent variance explained and associated p-values are presented in Figure 1—figure supplement 2.

As brain region and RIN were identified as the largest sources of variability, log-normalized quantifications of each gene were corrected for RNA quality independently within each brain region. This was done as follows: (1) exclude outliers (>3 standard deviations from the mean) and zero values, (2) determine whether expression data is best fit by one or two Gaussians using Mclust (Fraley and Raftery, 2002), (3) model RIN as a quadratic variable in each of the one or two groups, taking the sum of the residual and the mean as the normalized value, and then setting any negative values to 0. In most cases this normalization is equivalent to correcting the log-transformed data for RIN + adjusted RIN, and in other cases can additionally account for bimodalities in the data (e.g., gender) that are unrelated to RNA quality. This strategy of regressing out RIN is conceptually similar to one previously published (Gallego Romero et al., 2014). Other strategies accounted for RNA quality in RNA-seq data assume cDNA synthesis based on poly-A priming (Wan et al., 2012; Sigurgeirsson et al., 2014); these models break down in our data set where cDNA synthesis is based on a random hexamer method.

We note that this is a different final normalization step from that performed in the online data resource, where data were corrected for RIN + adjusted RIN + batch in linear space. The current strategy of excluding outliers from the normalization retains realistic expression values for biologically-relevant processes such as inflammation, and accounting for bimodalities removes effects of sex (which we sought to retain on the web resource). Similarly, batch correction is not included in this analysis as donors with the most severe TBIs were front loaded in the first two batches due to experimental constraints. Using data normalized on the website to assess differential gene expression between donors with and without dementia (as described below) produced comparable results.

Assessment of differential and co-expression

Request a detailed protocol

For pathology and demographic information, significance of differential expression between groups was assessed with analysis of variance (ANOVA) tests using the ‘aov’ function in R (Chambers et al., 1992). Correlations between continuous variables were calculated using the ‘cor’ function in R and are Pearson correlations with Bonferroni-corrected p-values of p<0.05, unless otherwise specified. Distributions of quantitative metrics are displayed using a smoothed density curve, with no associated statistical tests performed. Two tailed student t-tests were used to compare RIN between control and AD donors from multiple studies. We used surrogate variable analysis (SVA) (Leek and Storey, 2007) to quantify significance of gene expression with respect to dementia status (in combination with fold-difference thresholds) and RNA quality (in combination with correlation thresholds). P-values of p<0.05, after Benjamini and Hochberg correction, were considered significant unless otherwise noted. SVA was also used for assessing significance in gene clusters, as discussed below.

To assess the robustness of our result that few if any genes are significantly associated with dementia (or AD) status after controlling for RIN, we performed additional analyses to quantify significance of gene expression with respect to dementia status, in all cases defining significance as p<0.05 after Bonferroni correction. First, we repeated our analyses using additional statistical tests including 1) two tailed student t-tests, 2) ANOVA, 3) and limma (Ritchie et al., 2015), in all cases defining two groups based on dementia (or AD) status. Second, we repeated the SVA analysis described above on the RIN-normalized RNA-Seq data available for download from the website (which uses a slightly different normalization schema, as described above). Third, we performed principal component analysis (PCA) independently on each region using all genes, and used SVA to assess whether any of the top 25 PCs showed significant associate with dementia. Finally, we subsampled our data set to 70 donors who are matched for RIN, sex, and dementia status and repeated the SVA analysis using data that is not RIN-corrected to determine whether our particular RIN-normalization strategy could be biasing our ability to identify genes associated with dementia. In all cases we found two or fewer total genes associated with dementia or AD, indicating that our negative result is not due to improper statistical assessment.

We used weighted gene co-expression network analysis (WGCNA) (Zhang and Horvath, 2005; Langfelder and Horvath, 2008) to generate unbiased gene co-expression networks separately for each brain region. Initial networks were generated using an automated strategy with the following function call in R: blockRun = blockwiseModules(datExprRun, checkMissingData = TRUE, maxBlockSize = 17500, power = 14, networkType = ‘signed’, deepSplit = 2, minModuleSize = 20, minCoreKMESize = 7, minKMEtoStay = 0.4, mergeCutHeight = 0.1, numericLabels = TRUE, verbose = 1) where datExprRun is the log₂ normalized RNA-seq data from the top 9615 (50%) most variable genes (in each region). We then calculated module eigengenes (ME), defined as the first principal component of genes in the module. If the resulting network contained more than 20 modules, the module pairs with the most highly correlated ME were iteratively merged until 20 modules remained. Each expressed gene was then reassigned to the module to which it is most highly correlated to the ME (referred to as the gene’s module membership, or kME). Genes with maximum kME <0.4 were left unassigned (defined as module 0), as we have done in previous analyses (Hawrylycz et al., 2012). Since networks are unaffected by changes in labelling, we then re-labeled modules by percent of neuron-enriched genes so that those with the highest percentage of neuronal markers (Zhang et al., 2014) have lower numbers and those with the lowest percentage have higher numbers, as described previously (Hawrylycz et al., 2012).

We compared ME expression with 24 pathological and demographic measures, and used SVA to assess significance, defined here as Bonferroni-corrected p<0.05. In addition to the modules discussed in the Results, we found a single module of Y-chromosome genes in each network with nearly exclusive expression in males, as expected.

Gene set comparison between studies

Request a detailed protocol

In order to compare our differential expression results with prior work, we first assembled lists of genes differentially expressed between donors diagnosed with AD and matched controls from nine previous studies (Colangelo et al., 2002; Blalock et al., 2004; Liang et al., 2008; Webster et al., 2009; Avramopoulos et al., 2011; Miller et al., 2013; Zhang et al., 2013; Satoh et al., 2014; ). Figure 6—source data 4 describes in more detail specifically how we derived each gene list. Figure 6—source data 5 includes all gene lists. We then sorted and ranked all genes in our analysis with respect to fold difference (for dementia vs. control) or correlation with RIN in hippocampus and temporal cortex, scaling from 0 to 1. We then noted the ranks of external gene lists in our sorted lists, including the 25th, 50th, and 75th percentile values. We calculated p-values using a two-sided Wilcoxon rank sum test to measure divergence from a random distribution, with the R function ‘wilcox.test’ (Bauer, 1972). This schema is a modification of one previously described (Miller et al., 2013).

We compared the relationships between gene expression and local quantifications of tau pathology in our study with one previous study (Blalock et al., 2004). Correlations between gene expression (defined by the probe for each gene with maximal expression) and ‘NFT Score’ were calculated for the comparison study using publicly available data (GEO: GSE1297). Correlations between gene expression and AT8 IHC quantification from fresh frozen tissue were calculated in this study and the resulting correlations were themselves correlated for comparison.

Enrichment for gene ontology (GO) categories was performed using ToppFun with default parameters, which is available as part of the ToppGene Suite (Chen et al., 2009) (https://toppgene.cchmc.org/). Cell type-specific expression levels were collected from a published data set of selective expression in human neurons, astrocytes, microglia, and oligodendrocytes (Zhang et al., 2014) (http://web.stanford.edu/group/barres_lab/brain_rnaseq.html). Cell type enrichment was calculated by comparing gene lists in this study against genes with 2-fold enrichment in one verses all other cell types and FPKM >1. Cell type enrichment was largely confirmed by comparison with mouse-derived gene sets downloaded from NeuroExpresso (neuroexpresso.org) (Mancarci et al., 2016) using the same strategy.

Data and software availability

Request a detailed protocol

All images and most data presented in this manuscript are freely available from the resource website, http://aging.brain-map.org/. Code and remaining files required to reproduce all analyses and associated figure panels are available as part of the Github repository (https://github.com/AllenInstitute/agedbrain; Miller, 2017; copy archived at https://github.com/elifesciences-publications/agedbrain). Raw RNA-Seq data (FASTQ) and the output files after alignment (bam/FASTQ) are available for controlled access at NIAGADS: https://www.niagads.org/datasets/ng00059. TbT-normalized data (both before and after controlling for RIN) are also available through GEO (GSE104687).

Additional resources

Request a detailed protocol

Technical documentation describing the ACT cohort, all experimental procedures (i.e., tissue collection, tissue processing, quantitative data generation), and weighted analysis in more detail are freely available at http://help.brain-map.org/display/aging/Documentation.

Data availability

The following data sets were generated

(2017) Aging, Dementia, and TBI Study
Publicly available at the National Institute on Aging Genetics of Alzheimer's Disease Data Storage Site (accession no. ng00059).

https://www.niagads.org/datasets/ng00059
(2017) Aging, Dementia, and TBI Study
Publicly available at the NCBI Gene Expression Omnibus (accession no: GSE104687).

https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE104687

References

1. Akiyama H
2. Barger S
3. Barnum S
4. Bradt B
5. Bauer J
6. Cole GM
7. Cooper NR
8. Eikelenboom P
9. Emmerling M
10. Fiebich BL
11. Finch CE
12. Frautschy S
13. Griffin WS
14. Hampel H
15. Hull M
16. Landreth G
17. Lue L
18. Mrak R
19. Mackenzie IR
20. McGeer PL
21. O'Banion MK
22. Pachter J
23. Pasinetti G
24. Plata-Salaman C
25. Rogers J
26. Rydel R
27. Shen Y
28. Streit W
29. Strohmeyer R
30. Tooyoma I
31. Van Muiswinkel FL
32. Veerhuis R
33. Walker D
34. Webster S
35. Wegrzyniak B
36. Wenk G
37. Wyss-Coray T
(2000) Inflammation and Alzheimer's disease
Neurobiology of Aging 21:383–421.
https://doi.org/10.1016/S0197-4580(00)00124-X
- PubMed
- Google Scholar
1. Allen M
2. Carrasquillo MM
3. Funk C
4. Heavner BD
5. Zou F
6. Younkin CS
7. Burgess JD
8. Chai HS
9. Crook J
10. Eddy JA
11. Li H
12. Logsdon B
13. Peters MA
14. Dang KK
15. Wang X
16. Serie D
17. Wang C
18. Nguyen T
19. Lincoln S
20. Malphrus K
21. Bisceglio G
22. Li M
23. Golde TE
24. Mangravite LM
25. Asmann Y
26. Price ND
27. Petersen RC
28. Graff-Radford NR
29. Dickson DW
30. Younkin SG
31. Ertekin-Taner N
(2016) Human whole genome genotype and transcriptome data for Alzheimer's and other neurodegenerative diseases
Scientific Data 3:160089.
https://doi.org/10.1038/sdata.2016.89
- PubMed
- Google Scholar
1. Allgaier M
2. Allgaier C
(2014) An update on drug treatment options of Alzheimer's disease
Frontiers in Bioscience 19:1345–1354.
https://doi.org/10.2741/4285
- PubMed
- Google Scholar
1. Alzheimer's Association
(2016)
2016 Alzheimer's disease facts and figures

Alzheimer's & Dementia : The Journal of the Alzheimer's Association 12:459–509.
- PubMed
- Google Scholar
Book
1. American Psychiatric Association
(1994)
Diagnostic and statistical manual of mental disorders: DSM-IV

Washington, DC: American Psychiatric Press.
- Google Scholar
Software
1. Aronesty E
(2011) Command-line tools for processing biological sequencing data
Ea-Utils.

https://githubcom/ExpressionAnalysis/ea-utils
(2007) Methodological considerations for gene expression profiling of human brain
Journal of Neuroscience Methods 163:295–309.
https://doi.org/10.1016/j.jneumeth.2007.03.022
- PubMed
- Google Scholar
(2011) Gene expression reveals overlap between normal aging and Alzheimer's disease genes
Neurobiology of Aging 32:2319.e27–34.
https://doi.org/10.1016/j.neurobiolaging.2010.04.019
- PubMed
- Google Scholar
1. Baker SC
2. Bauer SR
3. Beyer RP
4. Brenton JD
5. Bromley B
6. Burrill J
7. Causton H
8. Conley MP
9. Elespuru R
10. Fero M
11. Foy C
12. Fuscoe J
13. Gao X
14. Gerhold DL
15. Gilles P
16. Goodsaid F
17. Guo X
18. Hackett J
19. Hockett RD
20. Ikonomi P
21. Irizarry RA
22. Kawasaki ES
23. Kaysser-Kranich T
24. Kerr K
25. Kiser G
26. Koch WH
27. Lee KY
28. Liu C
29. Liu ZL
30. Lucas A
31. Manohar CF
32. Miyada G
33. Modrusan Z
34. Parkes H
35. Puri RK
36. Reid L
37. Ryder TB
38. Salit M
39. Samaha RR
40. Scherf U
41. Sendera TJ
42. Setterquist RA
43. Shi L
44. Shippy R
45. Soriano JV
46. Wagar EA
47. Warrington JA
48. Williams M
49. Wilmer F
50. Wilson M
51. Wolber PK
52. Wu X
53. Zadro R
54. External R
55. External RNA Controls Consortium
(2005) The External RNA Controls Consortium: a progress report
Nature Methods 2:731–734.
https://doi.org/10.1038/nmeth1005-731
- PubMed
- Google Scholar
1. Bauer DF
(1972) Constructing Confidence Sets Using Rank Statistics
Journal of the American Statistical Association 67:687–690.
https://doi.org/10.1080/01621459.1972.10481279
- Google Scholar
(2012a) Overview and findings from the religious orders study
Current Alzheimer research 9:628–645.
https://doi.org/10.2174/156720512801322573
- PubMed
- Google Scholar
(2012b) Overview and findings from the rush Memory and Aging Project
Current Alzheimer research 9:646–663.
https://doi.org/10.2174/156720512801322663
- PubMed
- Google Scholar
1. Berchtold NC
2. Cribbs DH
3. Coleman PD
4. Rogers J
5. Head E
6. Kim R
7. Beach T
8. Miller C
9. Troncoso J
10. Trojanowski JQ
11. Zielke HR
12. Cotman CW
(2008) Gene expression changes in the course of normal brain aging are sexually dimorphic
PNAS 105:15605–15610.
https://doi.org/10.1073/pnas.0806883105
- PubMed
- Google Scholar
1. Beurel E
2. Jope RS
(2009) Lipopolysaccharide-induced interleukin-6 production is controlled by glycogen synthase kinase-3 and STAT3 in the brain
Journal of Neuroinflammation 6:9.
https://doi.org/10.1186/1742-2094-6-9
- PubMed
- Google Scholar
(2004) Incipient Alzheimer's disease: microarray correlation analyses reveal major transcriptional and tumor suppressor responses
PNAS 101:2173–2178.
https://doi.org/10.1073/pnas.0308512100
- PubMed
- Google Scholar
1. Braak H
2. Braak E
(1991) Neuropathological stageing of Alzheimer-related changes
Acta Neuropathologica 82:239–259.
https://doi.org/10.1007/BF00308809
- PubMed
- Google Scholar
(2012) LDLR-related protein 10 (LRP10) regulates amyloid precursor protein (APP) trafficking and processing: evidence for a role in Alzheimer's disease
Molecular Neurodegeneration 7:31.
https://doi.org/10.1186/1750-1326-7-31
- PubMed
- Google Scholar
(1992)
Statistical Models in S

Analysis of variance; designed experiments, Statistical Models in S, Wadsworth & Brooks/Cole.
- Google Scholar
1. Chen J
2. Bardes EE
3. Aronow BJ
4. Jegga AG
(2009) ToppGene Suite for gene list enrichment analysis and candidate gene prioritization
Nucleic Acids Research 37:W305–W311.
https://doi.org/10.1093/nar/gkp427
- PubMed
- Google Scholar
1. Colangelo V
2. Schurr J
3. Ball MJ
4. Pelaez RP
5. Bazan NG
6. Lukiw WJ
(2002) Gene expression profiling of 12633 genes in Alzheimer hippocampal CA1: transcription and neurotrophic factor down-regulation and up-regulation of apoptotic and pro-inflammatory signaling
Journal of Neuroscience Research 70:462–473.
https://doi.org/10.1002/jnr.10351
- PubMed
- Google Scholar
1. Colantuoni C
2. Lipska BK
3. Ye T
4. Hyde TM
5. Tao R
6. Leek JT
7. Colantuoni EA
8. Elkahloun AG
9. Herman MM
10. Weinberger DR
11. Kleinman JE
(2011) Temporal dynamics and genetic control of transcription in the human prefrontal cortex
Nature 478:519–523.
https://doi.org/10.1038/nature10524
- PubMed
- Google Scholar
(2012) A population-based clinicopathological study in the oldest-old: the 90+ study
Current Alzheimer Research 9:709–717.
https://doi.org/10.2174/156720512801322537
- PubMed
- Google Scholar
1. Crane PK
2. Narasimhalu K
3. Gibbons LE
4. Mungas DM
5. Haneuse S
6. Larson EB
7. Kuller L
8. Hall K
9. van Belle G
(2008) Item response theory facilitated cocalibrating cognitive tests and reduced bias in estimated rates of decline
Journal of Clinical Epidemiology 61:1018–1027.
https://doi.org/10.1016/j.jclinepi.2007.11.011
- PubMed
- Google Scholar
(2013) Glucose levels and risk of dementia
New England Journal of Medicine 369:540–548.
https://doi.org/10.1056/NEJMoa1215740
- PubMed
- Google Scholar
1. Crary JF
2. Trojanowski JQ
3. Schneider JA
4. Abisambra JF
5. Abner EL
6. Alafuzoff I
7. Arnold SE
8. Attems J
9. Beach TG
10. Bigio EH
11. Cairns NJ
12. Dickson DW
13. Gearing M
14. Grinberg LT
15. Hof PR
16. Hyman BT
17. Jellinger K
18. Jicha GA
19. Kovacs GG
20. Knopman DS
21. Kofler J
22. Kukull WA
23. Mackenzie IR
24. Masliah E
25. McKee A
26. Montine TJ
27. Murray ME
28. Neltner JH
29. Santa-Maria I
30. Seeley WW
31. Serrano-Pozo A
32. Shelanski ML
33. Stein T
34. Takao M
35. Thal DR
36. Toledo JB
37. Troncoso JC
38. Vonsattel JP
39. White CL
40. Wisniewski T
41. Woltjer RL
42. Yamada M
43. Nelson PT
(2014) Primary age-related tauopathy (PART): a common pathology associated with human aging
Acta Neuropathologica 128:755–766.
https://doi.org/10.1007/s00401-014-1349-0
- PubMed
- Google Scholar
Book
1. Dang C
2. Sodt A
3. Lau C
4. Youngstrom B
5. Ng L
6. Kuan L
7. Pathak S
8. Jones A
9. Hawrylycz M
(2007)
The Allen Brain Atlas: Delivering Neuroscience to the Web on a Genome Wide Scale

In: Cohen-Boulakia S, Tannen V, editors. Data Integration in the Life Sciences: 4th International Workshop, DILS 2007, Philadelphia, PA, USA, June 27-29, 2007 Proceedings. Berlin, Heidelberg: Springer Berlin Heidelberg. pp. 17–26.
- Google Scholar
1. De Jager PL
2. Srivastava G
3. Lunnon K
4. Burgess J
5. Schalkwyk LC
6. Yu L
7. Eaton ML
8. Keenan BT
9. Ernst J
10. McCabe C
11. Tang A
12. Raj T
13. Replogle J
14. Brodeur W
15. Gabriel S
16. Chai HS
17. Younkin C
18. Younkin SG
19. Zou F
20. Szyf M
21. Epstein CB
22. Schneider JA
23. Bernstein BE
24. Meissner A
25. Ertekin-Taner N
26. Chibnik LB
27. Kellis M
28. Mill J
29. Bennett DA
(2014) Alzheimer's disease: early alterations in brain DNA methylation at ANK1, BIN1, RHBDF2 and other loci
Nature Neuroscience 17:1156–1163.
https://doi.org/10.1038/nn.3786
- PubMed
- Google Scholar
1. Durrenberger PF
2. Fernando S
3. Kashefi SN
4. Ferrer I
5. Hauw JJ
6. Seilhean D
7. Smith C
8. Walker R
9. Al-Sarraj S
10. Troakes C
11. Palkovits M
12. Kasztner M
13. Huitinga I
14. Arzberger T
15. Dexter DT
16. Kretzschmar H
17. Reynolds R
(2010) Effects of antemortem and postmortem variables on human brain mRNA quality: a BrainNet Europe study
Journal of Neuropathology & Experimental Neurology 69:70–81.
https://doi.org/10.1097/NEN.0b013e3181c7e32f
- PubMed
- Google Scholar
1. Ehlenbach WJ
2. Hough CL
3. Crane PK
4. Haneuse SJ
5. Carson SS
6. Curtis JR
7. Larson EB
(2010) Association between acute care and critical illness hospitalization and cognitive function in older adults
JAMA 303:763–770.
https://doi.org/10.1001/jama.2010.167
- PubMed
- Google Scholar
1. Flanagan M
2. Larson EB
3. Latimer CS
4. Cholerton B
5. Crane PK
6. Montine KS
7. White LR
8. Keene CD
9. Montine TJ
(2016) Clinical-pathologic correlations in vascular cognitive impairment and dementia
Biochimica et Biophysica Acta (BBA) - Molecular Basis of Disease 1862:945–951.
https://doi.org/10.1016/j.bbadis.2015.08.019
- PubMed
- Google Scholar
1. Fraley C
2. Raftery AE
(2002) Model-based clustering, discriminant analysis, and density estimation
Journal of the American Statistical Association 97:611–631.
https://doi.org/10.1198/016214502760047131
- Google Scholar
1. Franceschi C
2. Campisi J
(2014) Chronic inflammation (inflammaging) and its potential contribution to age-associated diseases
The Journals of Gerontology Series A: Biological Sciences and Medical Sciences 69 Suppl 1:S4–S9.
https://doi.org/10.1093/gerona/glu057
- PubMed
- Google Scholar
(2014) RNA-seq: impact of RNA degradation on transcript quantification
BMC Biology 12:42.
https://doi.org/10.1186/1741-7007-12-42
- PubMed
- Google Scholar
(2013) Dementia in the oldest old: a multi-factorial and growing public health issue
Alzheimer's Research & Therapy 5:27.
https://doi.org/10.1186/alzrt181
- PubMed
- Google Scholar
1. Haneuse S
2. Schildcrout J
3. Crane P
4. Sonnen J
5. Breitner J
6. Larson E
(2009) Adjustment for selection bias in observational studies with application to the analysis of autopsy data
Neuroepidemiology 32:229–239.
https://doi.org/10.1159/000197389
- PubMed
- Google Scholar
(2008) Role of the neuropathology of Alzheimer disease in dementia in the oldest-old
Archives of Neurology 65:1211–1217.
https://doi.org/10.1001/archneur.65.9.1211
- PubMed
- Google Scholar
1. Hawrylycz MJ
2. Lein ES
3. Guillozet-Bongaarts AL
4. Shen EH
5. Ng L
6. Miller JA
7. van de Lagemaat LN
8. Smith KA
9. Ebbert A
10. Riley ZL
11. Abajian C
12. Beckmann CF
13. Bernard A
14. Bertagnolli D
15. Boe AF
16. Cartagena PM
17. Chakravarty MM
18. Chapin M
19. Chong J
20. Dalley RA
21. David Daly B
22. Dang C
23. Datta S
24. Dee N
25. Dolbeare TA
26. Faber V
27. Feng D
28. Fowler DR
29. Goldy J
30. Gregor BW
31. Haradon Z
32. Haynor DR
33. Hohmann JG
34. Horvath S
35. Howard RE
36. Jeromin A
37. Jochim JM
38. Kinnunen M
39. Lau C
40. Lazarz ET
41. Lee C
42. Lemon TA
43. Li L
44. Li Y
45. Morris JA
46. Overly CC
47. Parker PD
48. Parry SE
49. Reding M
50. Royall JJ
51. Schulkin J
52. Sequeira PA
53. Slaughterbeck CR
54. Smith SC
55. Sodt AJ
56. Sunkin SM
57. Swanson BE
58. Vawter MP
59. Williams D
60. Wohnoutka P
61. Zielke HR
62. Geschwind DH
63. Hof PR
64. Smith SM
65. Koch C
66. Grant SGN
67. Jones AR
(2012) An anatomically comprehensive atlas of the adult human brain transcriptome
Nature 489:391–399.
https://doi.org/10.1038/nature11405
- PubMed
- Google Scholar
1. Hyman BT
2. Phelps CH
3. Beach TG
4. Bigio EH
5. Cairns NJ
6. Carrillo MC
7. Dickson DW
8. Duyckaerts C
9. Frosch MP
10. Masliah E
11. Mirra SS
12. Nelson PT
13. Schneider JA
14. Thal DR
15. Thies B
16. Trojanowski JQ
17. Vinters HV
18. Montine TJ
(2012) National Institute on Aging-Alzheimer's Association guidelines for the neuropathologic assessment of Alzheimer's disease
Alzheimer's & Dementia 8:1–13.
https://doi.org/10.1016/j.jalz.2011.10.007
- PubMed
- Google Scholar
1. Ji KA
2. Eu MY
3. Kang SH
4. Gwag BJ
5. Jou I
6. Joe EH
(2008) Differential neutrophil infiltration contributes to regional differences in brain inflammation in the substantia nigra pars compacta and cortex
Glia 56:1039–1047.
https://doi.org/10.1002/glia.20677
- PubMed
- Google Scholar
(2012) A normalization strategy for comparing tag count data
Algorithms for Molecular Biology 7:5.
https://doi.org/10.1186/1748-7188-7-5
- PubMed
- Google Scholar
1. Kukull WA
2. Higdon R
3. Bowen JD
4. McCormick WC
5. Teri L
6. Schellenberg GD
7. van Belle G
8. Jolley L
9. Larson EB
(2002)
Dementia and Alzheimer disease incidence: a prospective cohort study

Archives of neurology 59:1737–1746.
- PubMed
- Google Scholar
1. Langfelder P
2. Horvath S
(2008) WGCNA: an R package for weighted correlation network analysis
BMC Bioinformatics 9:559.
https://doi.org/10.1186/1471-2105-9-559
- PubMed
- Google Scholar
(2009) Ultrafast and memory-efficient alignment of short DNA sequences to the human genome
Genome Biology 10:R25.
https://doi.org/10.1186/gb-2009-10-3-r25
- PubMed
- Google Scholar
1. Larson EB
2. Wang L
3. Bowen JD
4. McCormick WC
5. Teri L
6. Crane P
7. Kukull W
(2006) Exercise is associated with reduced risk for incident dementia among persons 65 years of age and older
Annals of Internal Medicine 144:73–81.
https://doi.org/10.7326/0003-4819-144-2-200601170-00004
- PubMed
- Google Scholar
1. Leek JT
2. Storey JD
(2007) Capturing heterogeneity in gene expression studies by surrogate variable analysis
PLoS Genetics 3:e161–1735.
https://doi.org/10.1371/journal.pgen.0030161
- PubMed
- Google Scholar
1. Lein ES
2. Hawrylycz MJ
3. Ao N
4. Ayres M
5. Bensinger A
6. Bernard A
7. Boe AF
8. Boguski MS
9. Brockway KS
10. Byrnes EJ
11. Chen L
12. Chen L
13. Chen TM
14. Chin MC
15. Chong J
16. Crook BE
17. Czaplinska A
18. Dang CN
19. Datta S
20. Dee NR
21. Desaki AL
22. Desta T
23. Diep E
24. Dolbeare TA
25. Donelan MJ
26. Dong HW
27. Dougherty JG
28. Duncan BJ
29. Ebbert AJ
30. Eichele G
31. Estin LK
32. Faber C
33. Facer BA
34. Fields R
35. Fischer SR
36. Fliss TP
37. Frensley C
38. Gates SN
39. Glattfelder KJ
40. Halverson KR
41. Hart MR
42. Hohmann JG
43. Howell MP
44. Jeung DP
45. Johnson RA
46. Karr PT
47. Kawal R
48. Kidney JM
49. Knapik RH
50. Kuan CL
51. Lake JH
52. Laramee AR
53. Larsen KD
54. Lau C
55. Lemon TA
56. Liang AJ
57. Liu Y
58. Luong LT
59. Michaels J
60. Morgan JJ
61. Morgan RJ
62. Mortrud MT
63. Mosqueda NF
64. Ng LL
65. Ng R
66. Orta GJ
67. Overly CC
68. Pak TH
69. Parry SE
70. Pathak SD
71. Pearson OC
72. Puchalski RB
73. Riley ZL
74. Rockett HR
75. Rowland SA
76. Royall JJ
77. Ruiz MJ
78. Sarno NR
79. Schaffnit K
80. Shapovalova NV
81. Sivisay T
82. Slaughterbeck CR
83. Smith SC
84. Smith KA
85. Smith BI
86. Sodt AJ
87. Stewart NN
88. Stumpf KR
89. Sunkin SM
90. Sutram M
91. Tam A
92. Teemer CD
93. Thaller C
94. Thompson CL
95. Varnam LR
96. Visel A
97. Whitlock RM
98. Wohnoutka PE
99. Wolkey CK
100. Wong VY
101. Wood M
102. Yaylaoglu MB
103. Young RC
104. Youngstrom BL
105. Yuan XF
106. Zhang B
107. Zwingman TA
108. Jones AR
(2007) Genome-wide atlas of gene expression in the adult mouse brain
Nature 445:168–176.
https://doi.org/10.1038/nature05453
- PubMed
- Google Scholar
1. Li JZ
2. Vawter MP
3. Walsh DM
4. Tomita H
5. Evans SJ
6. Choudary PV
7. Lopez JF
8. Avelar A
9. Shokoohi V
10. Chung T
11. Mesarwi O
12. Jones EG
13. Watson SJ
14. Akil H
15. Bunney WE
16. Myers RM
(2004) Systematic changes in gene expression in postmortem human brains associated with tissue pH and terminal medical conditions
Human Molecular Genetics 13:609–616.
https://doi.org/10.1093/hmg/ddh065
- PubMed
- Google Scholar
1. Li B
2. Ruotti V
3. Stewart RM
4. Thomson JA
5. Dewey CN
(2010) RNA-Seq gene expression estimation with read mapping uncertainty
Bioinformatics 26:493–500.
https://doi.org/10.1093/bioinformatics/btp692
- PubMed
- Google Scholar
1. Liang WS
2. Dunckley T
3. Beach TG
4. Grover A
5. Mastroeni D
6. Ramsey K
7. Caselli RJ
8. Kukull WA
9. McKeel D
10. Morris JC
11. Hulette CM
12. Schmechel D
13. Reiman EM
14. Rogers J
15. Stephan DA
(2008) Altered neuronal gene expression in brain regions differentially affected by Alzheimer's disease: a reference data set
Physiological Genomics 33:240–256.
https://doi.org/10.1152/physiolgenomics.00242.2007
- PubMed
- Google Scholar
1. Lu T
2. Pan Y
3. Kao SY
4. Li C
5. Kohane I
6. Chan J
7. Yankner BA
(2004) Gene regulation and DNA damage in the ageing human brain
Nature 429:883–891.
https://doi.org/10.1038/nature02661
- PubMed
- Google Scholar
1. Lu J
2. Goh SJ
3. Tng PY
4. Deng YY
5. Ling EA
6. Moochhala S
(2009) Systemic inflammatory response following acute traumatic brain injury
Frontiers in Bioscience 14:3795–3813.
https://doi.org/10.2741/3489
- PubMed
- Google Scholar
Preprint
1. Mancarci BO
2. Toker L
3. Tripathy S
4. Li B
5. Rocco B
6. Sibille E
7. Pavlidis P
(2016) NeuroExpresso: A cross-laboratory database of brain cell-type expression profiles with applications to marker gene identification and bulk brain tissue transcriptome interpretation
bioRxiv.
https://doi.org/10.1101/089219
- Google Scholar
1. McKee AC
2. Cairns NJ
3. Dickson DW
4. Folkerth RD
5. Keene CD
6. Litvan I
7. Perl DP
8. Stein TD
9. Vonsattel JP
10. Stewart W
11. Tripodis Y
12. Crary JF
13. Bieniek KF
14. Dams-O'Connor K
15. Alvarez VE
16. Gordon WA
17. TBI/CTE group
(2016) The first NINDS/NIBIB consensus meeting to define neuropathological criteria for the diagnosis of chronic traumatic encephalopathy
Acta Neuropathologica 131:75–86.
https://doi.org/10.1007/s00401-015-1515-z
- PubMed
- Google Scholar
1. McKeith IG
2. Galasko D
3. Kosaka K
4. Perry EK
5. Dickson DW
6. Hansen LA
7. Salmon DP
8. Lowe J
9. Mirra SS
10. Byrne EJ
11. Lennox G
12. Quinn NP
13. Edwardson JA
14. Ince PG
15. Bergeron C
16. Burns A
17. Miller BL
18. Lovestone S
19. Collerton D
20. Jansen EN
21. Ballard C
22. de Vos RA
23. Wilcock GK
24. Jellinger KA
25. Perry RH
(1996) Consensus guidelines for the clinical and pathologic diagnosis of dementia with Lewy bodies (DLB): report of the consortium on DLB international workshop
Neurology 47:1113–1124.
https://doi.org/10.1212/WNL.47.5.1113
- PubMed
- Google Scholar
1. McKhann G
2. Drachman D
3. Folstein M
4. Katzman R
5. Price D
6. Stadlan EM
(1984) Clinical diagnosis of Alzheimer's disease: report of the NINCDS-ADRDA Work Group under the auspices of department of health and human services task force on Alzheimer's disease
Neurology 34:939–944.
https://doi.org/10.1212/WNL.34.7.939
- PubMed
- Google Scholar
1. Mexal S
2. Berger R
3. Adams CE
4. Ross RG
5. Freedman R
6. Leonard S
(2006) Brain pH has a significant impact on human postmortem hippocampal gene expression profiles
Brain Research 1106:1–11.
https://doi.org/10.1016/j.brainres.2006.05.043
- PubMed
- Google Scholar
(2008) A systems level analysis of transcriptional changes in Alzheimer's disease and normal aging
Journal of Neuroscience 28:1410–1420.
https://doi.org/10.1523/JNEUROSCI.4098-07.2008
- PubMed
- Google Scholar
(2013) Genes and pathways underlying regional and cell type changes in Alzheimer's disease
Genome Medicine 5:48.
https://doi.org/10.1186/gm452
- PubMed
- Google Scholar
Software
1. Miller JA
(2017) agedbrain, version 02c2d07
Github.

https://github.com/AllenInstitute/agedbrain
1. Mills JD
2. Sheahan PJ
3. Lai D
4. Kril JJ
5. Janitz M
6. Sutherland GT
(2014) The alternative splicing of the apolipoprotein E gene is unperturbed in the brains of Alzheimer's disease patients
Molecular Biology Reports 41:6365–6376.
https://doi.org/10.1007/s11033-014-3516-8
- PubMed
- Google Scholar
1. Mirra SS
2. Heyman A
3. McKeel D
4. Sumi SM
5. Crain BJ
6. Brownlee LM
7. Vogel FS
8. Hughes JP
9. van Belle G
10. Berg L
(1991) The Consortium to Establish a Registry for Alzheimer's Disease (CERAD). Part II. Standardization of the neuropathologic assessment of Alzheimer's disease
Neurology 41:479–486.
https://doi.org/10.1212/WNL.41.4.479
- PubMed
- Google Scholar
1. Monoranu CM
2. Apfelbacher M
3. Grünblatt E
4. Puppe B
5. Alafuzoff I
6. Ferrer I
7. Al-Saraj S
8. Keyvani K
9. Schmitt A
10. Falkai P
11. Schittenhelm J
12. Halliday G
13. Kril J
14. Harper C
15. McLean C
16. Riederer P
17. Roggendorf W
(2009) pH measurement as quality control on human post mortem brain tissue: a study of the BrainNet Europe consortium
Neuropathology and Applied Neurobiology 35:329–337.
https://doi.org/10.1111/j.1365-2990.2008.01003a.x
- PubMed
- Google Scholar
(2005) F2-isoprostanes in Alzheimer and other neurodegenerative diseases
Antioxidants & Redox Signaling 7:269–275.
https://doi.org/10.1089/ars.2005.7.269
- PubMed
- Google Scholar
(2012) National Institute on Aging-Alzheimer's Association guidelines for the neuropathologic assessment of Alzheimer's disease: a practical approach
Acta Neuropathologica 123:1–11.
https://doi.org/10.1007/s00401-011-0910-3
- PubMed
- Google Scholar
(2014) A 2-process model for neuropathology of Alzheimer's disease
Neurobiology of Aging 35:301–308.
https://doi.org/10.1016/j.neurobiolaging.2013.08.007
- PubMed
- Google Scholar
1. Nelson PT
2. Jicha GA
3. Schmitt FA
4. Liu H
5. Davis DG
6. Mendiondo MS
7. Abner EL
8. Markesbery WR
(2007) Clinicopathologic correlations in a large Alzheimer disease center autopsy cohort: neuritic plaques and neurofibrillary tangles "do count" when staging disease severity
Journal of Neuropathology and Experimental Neurology 66:1136–1146.
https://doi.org/10.1097/nen.0b013e31815c5efb
- PubMed
- Google Scholar
1. Nelson PT
2. Trojanowski JQ
3. Abner EL
4. Al-Janabi OM
5. Jicha GA
6. Schmitt FA
7. Smith CD
8. Fardo DW
9. Wang WX
10. Kryscio RJ
11. Neltner JH
12. Kukull WA
13. Cykowski MD
14. Van Eldik LJ
15. Ighodaro ET
(2016) "New Old Pathologies": AD, PART, and Cerebral Age-Related TDP-43 With Sclerosis (CARTS)
Journal of Neuropathology & Experimental Neurology 75:482–498.
https://doi.org/10.1093/jnen/nlw033
- PubMed
- Google Scholar
1. Oldham MC
2. Konopka G
3. Iwamoto K
4. Langfelder P
5. Kato T
6. Horvath S
7. Geschwind DH
(2008) Functional organization of the transcriptome in human brain
Nature Neuroscience 11:1271–1282.
https://doi.org/10.1038/nn.2207
- PubMed
- Google Scholar
(2009) Blood-brain barrier alterations in ageing and dementia
Journal of the Neurological Sciences 283:99–106.
https://doi.org/10.1016/j.jns.2009.02.321
- PubMed
- Google Scholar
1. Preece P
2. Cairns NJ
(2003) Quantifying mRNA in postmortem human brain: influence of gender, age at death, postmortem interval, brain pH, agonal state and inter-lobe mRNA variance
Molecular Brain Research 118:60–71.
https://doi.org/10.1016/S0169-328X(03)00337-1
- PubMed
- Google Scholar
(1999) Differential expression of inflammatory mediators in rat microglia cultured from different brain regions
Molecular Brain Research 65:198–205.
https://doi.org/10.1016/S0169-328X(99)00016-9
- PubMed
- Google Scholar
1. Ritchie ME
2. Phipson B
3. Wu D
4. Hu Y
5. Law CW
6. Shi W
7. Smyth GK
(2015) limma powers differential expression analyses for RNA-sequencing and microarray studies
Nucleic Acids Research 43:e47.
https://doi.org/10.1093/nar/gkv007
- PubMed
- Google Scholar
1. Satoh J
2. Yamamoto Y
3. Asahina N
4. Kitano S
5. Kino Y
(2014) RNA-Seq data mining: downregulation of NeuroD6 serves as a possible biomarker for alzheimer's disease brains
Disease Markers 2014:123165.
https://doi.org/10.1155/2014/123165
- PubMed
- Google Scholar
1. Schöbel S
2. Neumann S
3. Hertweck M
4. Dislich B
5. Kuhn PH
6. Kremmer E
7. Seed B
8. Baumeister R
9. Haass C
10. Lichtenthaler SF
(2008) A novel sorting nexin modulates endocytic trafficking and alpha-secretase cleavage of the amyloid precursor protein
Journal of Biological Chemistry 283:14257–14268.
https://doi.org/10.1074/jbc.M801531200
- PubMed
- Google Scholar
(2014) Sequencing degraded RNA addressed by 3' tag counting
PLoS One 9:e91851.
https://doi.org/10.1371/journal.pone.0091851
- PubMed
- Google Scholar
1. Sonnen JA
2. Larson EB
3. Crane PK
4. Haneuse S
5. Li G
6. Schellenberg GD
7. Craft S
8. Leverenz JB
9. Montine TJ
(2007) Pathological correlates of dementia in a longitudinal, population-based sample of aging
Annals of Neurology 62:406–413.
https://doi.org/10.1002/ana.21208
- PubMed
- Google Scholar
1. Sonnen JA
2. Santa Cruz K
3. Hemmy LS
4. Woltjer R
5. Leverenz JB
6. Montine KS
7. Jack CR
8. Kaye J
9. Lim K
10. Larson EB
11. White L
12. Montine TJ
(2011) Ecology of the aging human brain
Archives of Neurology 68:1049–1056.
https://doi.org/10.1001/archneurol.2011.157
- PubMed
- Google Scholar
1. Stygelbout V
2. Leroy K
3. Pouillon V
4. Ando K
5. D'Amico E
6. Jia Y
7. Luo HR
8. Duyckaerts C
9. Erneux C
10. Schurmans S
11. Brion JP
(2014) Inositol trisphosphate 3-kinase B is increased in human Alzheimer brain and exacerbates mouse Alzheimer pathology
Brain 137:537–552.
https://doi.org/10.1093/brain/awt344
- PubMed
- Google Scholar
1. Sun A
2. Nguyen XV
3. Bing G
(2002) Comparative analysis of an improved thioflavin-s stain, Gallyas silver stain, and immunohistochemistry for neurofibrillary tangle demonstration on the same sections
Journal of Histochemistry & Cytochemistry 50:463–472.
https://doi.org/10.1177/002215540205000403
- PubMed
- Google Scholar
1. Teng EL
2. Hasegawa K
3. Homma A
4. Imai Y
5. Larson E
6. Graves A
7. Sugimoto K
8. Yamaguchi T
9. Sasaki H
10. Chiu D
11. White LR
(1999) The Cognitive Abilities Screening Instrument (CASI): a practical test for cross-cultural epidemiological studies of dementia
International Psychogeriatrics 6:45–58.
https://doi.org/10.1017/S1041610294001602
- PubMed
- Google Scholar
1. Thal DR
2. Rüb U
3. Orantes M
4. Braak H
(2002) Phases of A beta-deposition in the human brain and its relevance for the development of AD
Neurology 58:1791–1800.
https://doi.org/10.1212/WNL.58.12.1791
- PubMed
- Google Scholar
1. Tomita H
2. Vawter MP
3. Walsh DM
4. Evans SJ
5. Choudary PV
6. Li J
7. Overman KM
8. Atz ME
9. Myers RM
10. Jones EG
11. Watson SJ
12. Akil H
13. Bunney WE
(2004) Effect of agonal and postmortem factors on gene expression profile: quality control in microarray analyses of postmortem human brain
Biological Psychiatry 55:346–352.
https://doi.org/10.1016/j.biopsych.2003.10.013
- PubMed
- Google Scholar
1. Vawter MP
2. Tomita H
3. Meng F
4. Bolstad B
5. Li J
6. Evans S
7. Choudary P
8. Atz M
9. Shao L
10. Neal C
11. Walsh DM
12. Burmeister M
13. Speed T
14. Myers R
15. Jones EG
16. Watson SJ
17. Akil H
18. Bunney WE
(2006) Mitochondrial-related gene expression changes are sensitive to agonal-pH state: implications for brain disorders
Molecular Psychiatry 11:663–679.
https://doi.org/10.1038/sj.mp.4001830
- PubMed
- Google Scholar
1. Wan L
2. Yan X
3. Chen T
4. Sun F
(2012) Modeling RNA degradation for RNA-Seq with applications
Biostatistics 13:734–747.
https://doi.org/10.1093/biostatistics/kxs001
- PubMed
- Google Scholar
1. Wang WY
2. Tan MS
3. Yu JT
4. Tan L
(2015) Role of pro-inflammatory cytokines released from microglia in Alzheimer's disease
Annals of Translational Medicine 3:136.
https://doi.org/10.3978/j.issn.2305-5839.2015.03.49
- PubMed
- Google Scholar
1. Webster JA
2. Gibbs JR
3. Clarke J
4. Ray M
5. Zhang W
6. Holmans P
7. Rohrer K
8. Zhao A
9. Marlowe L
10. Kaleem M
11. McCorquodale DS
12. Cuello C
13. Leung D
14. Bryden L
15. Nath P
16. Zismann VL
17. Joshipura K
18. Huentelman MJ
19. Hu-Lince D
20. Coon KD
21. Craig DW
22. Pearson JV
23. Heward CB
24. Reiman EM
25. Stephan D
26. Hardy J
27. Myers AJ
28. NACC-Neuropathology Group
(2009) Genetic control of human brain transcript expression in Alzheimer disease
The American Journal of Human Genetics 84:445–458.
https://doi.org/10.1016/j.ajhg.2009.03.011
- PubMed
- Google Scholar
1. Zapala MA
2. Schork NJ
(2012) Statistical properties of multivariate distance matrix regression for high-dimensional data analysis
Frontiers in Genetics 3:190.
https://doi.org/10.3389/fgene.2012.00190
- PubMed
- Google Scholar
1. Zhang B
2. Horvath S
(2005) A general framework for weighted gene co-expression network analysis
Statistical Applications in Genetics and Molecular Biology 4:Article17.
https://doi.org/10.2202/1544-6115.1128
- PubMed
- Google Scholar
1. Zhang B
2. Gaiteri C
3. Bodea LG
4. Wang Z
5. McElwee J
6. Podtelezhnikov AA
7. Zhang C
8. Xie T
9. Tran L
10. Dobrin R
11. Fluder E
12. Clurman B
13. Melquist S
14. Narayanan M
15. Suver C
16. Shah H
17. Mahajan M
18. Gillis T
19. Mysore J
20. MacDonald ME
21. Lamb JR
22. Bennett DA
23. Molony C
24. Stone DJ
25. Gudnason V
26. Myers AJ
27. Schadt EE
28. Neumann H
29. Zhu J
30. Emilsson V
(2013) Integrated systems approach identifies genetic nodes and networks in late-onset Alzheimer's disease
Cell 153:707–720.
https://doi.org/10.1016/j.cell.2013.03.030
- PubMed
- Google Scholar
1. Zhang Y
2. Chen K
3. Sloan SA
4. Bennett ML
5. Scholze AR
6. O'Keeffe S
7. Phatnani HP
8. Guarnieri P
9. Caneda C
10. Ruderisch N
11. Deng S
12. Liddelow SA
13. Zhang C
14. Daneman R
15. Maniatis T
16. Barres BA
17. Wu JQ
18. Jq W
(2014) An RNA-sequencing transcriptome and splicing database of glia, neurons, and vascular cells of the cerebral cortex
Journal of Neuroscience 34:11929–11947.
https://doi.org/10.1523/JNEUROSCI.1860-14.2014
- PubMed
- Google Scholar

Decision letter

Sacha B Nelson

Reviewing Editor; Brandeis University, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

[Editors’ note: a previous version of this study was rejected after peer review, but the authors submitted for reconsideration. The first decision letter after peer review is shown below.]

Thank you for submitting your work entitled "Neuropathological and transcriptomic characteristics of the aged brain" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and a Senior Editor. We apologize for the delay in returning this decision to, but there was an extended period of discussion among the reviewers.

Our decision has been reached after consultation between the reviewers. Based on these discussions and the individual reviews below, we regret to inform you that the present manuscript will not be considered further for publication in eLife.

Although we recognize that the study represents a large and potentially valuable resource for the community, we are declining the present version of the manuscript because the reviewers found it difficult to evaluate the precise reasons for the discrepancies between the present negative results and those obtained from other carefully carried out prior studies. It is eLife policy that all data and analysis code needed to derive the analysis results from the data should be made available. It is possible that if the details (including code) of the analyses are made clearer in a revised version that the reasons for the discrepancies will be clearer and that we will be able to make a more informed decision about publication of the manuscript. Hence, despite the rejection, we encourage a resubmission, especially if the points raised by the reviewers can be addressed. Note that complete agreement between prior studies and this one is not required, but in the face of major disagreements, the present study must help to illuminate the cause of the discrepancies and should add to the ability of careful readers to come to a more integrated understanding. In the hopes of making the issues raised as clear as possible we are including both the original reviews and additional relevant comments made during discussion of the reviews.

Reviewer #1:

Considering this work as a data resource, my main consideration is in how well the results can be reproduced and repurposed, for the many other researchers who will be interested in replicating (or not) findings in their own cohorts. Since the conclusions of this study are somewhat at odds with several other studies, it is particularly important to make all the data needed for reprocessing available.

In particular, since many people might want to use their own normalization and RIN was emphasized in this paper, that would be an important variable to share, but I was unable to find that variable in the data-sharing website (nor could I find PMI). In addition, it would be helpful to provide other information that people would ostensibly use in a normalization – percentage aligned reads or other variables that you had access to that you ultimately chose not to use in this particular normalization such as number of ribosomal bases. Also, batch data was unavailable. If these are available and I'm just missing them, then this criticism dissolves.

However, it does raise a related point. The exact scripts used in your normalization will enable others to understand how additional variable change their results, in addition to the substantial time savings from having to figure out where everything is in the different files and trying to exactly infer your code from the text. Specifically, it should be possible to take the input files shared on the web – which is all any external researcher will be able to access – and arrive at your normalized data. Indeed, these are essential to providing a firm foundation for alternative normalizations of the data, which I suspect is going to be a major interest of many researchers given the particular results of this study.

I appreciate that you have provided your normalized output, but to actually consider modifications to the normalization, the scripts used in getting there are essential. There is a note that code for figures is available on GitHub – however I don't have a link for that and it's not clear it covers normalization. If so – great! – but I don't have a way of looking into this.

Reviewer #2:

The authors present an analysis of the ACT cohort of human brain tissue taken from aged individuals with and without various pathologies (AD, dementia and TBI). They analyze a broad range of assays such as protein markers and RNA-seq from several brain regions. While these efforts are impressive, in general it seems the findings primarily confirm what is already known about relationships between markers such as AB, pTau, inflammation, and aging, AD or dementia. While a new category (M16) of co-expressed genes was found to be correlated with pTau load, it is not clear whether this is particularly novel in relation to aging or dementia, or is informative for understanding the etiology of either. The paper initially seemed geared towards understanding the aging brain, but very little is presented to shed light on aging and most of what is presented regarding dementia and AD does not appear to be new. It is possible the study contains novel findings that were lost on this reader, but clearer descriptions and context are needed to make this apparent.

The investigation into RNA quality correlated with dementia is interesting and important, but seems unfinished. The finding that previous studies may be flawed due to a lack of accounting for RNA quality maybe significant enough for a short report. However, it also seems to undermine the current dataset. Given the discussion of brain pH and the greatest discordance between this study and a study which took this factor into account raises questions about the relationship between RIN and factors like pH. It is not clear from the analysis presented here that RIN is a sufficiently representative metric to account for all necessary variables. From the data in this study there appeared to be at least two potential issues in differential expression (DE) analysis. (1) Decreased RNA quality correlated with dementia but also (Allen et al., 2016) increased variance with age. The second factor was mentioned in the Discussion section but not clearly described in the Results section and so it is not clear how this may have impacted DE analysis. Thus, determining the correlation between RIN and dementia is an important issue, it is not clear that the authors are presenting an actual solution.

The main take-away seems to be that additional work is needed to optimize RNA isolation from pathological brain samples and perhaps a more in-depth investigation into increased expression variability with age. While decreased RNA quality and increase variance may be indicative of biological changes, this study does not seem to help clarify these possibilities. Improved methods to either remove or account for these factors might open up other possible lines of investigation. For example, the disconnect between pathological markers and dementia was interesting but only superficially investigated. Assuming the hurdles of RNA quality and high variance can be overcome, it would be a natural follow up to probe for gene expression signatures of "tolerance". While I agree that the ACT cohort seems to be a valuable resource, the limitations highlighted in this analysis make it unclear exactly what new information we might pull from it and how best to pull the information.

Reviewer #3:

This work represents an observational study that looks to characterize and associate various factors associated with Alzheimer's dementia (pathological markers, inflammatory markers, gene expression data and clinical dementia scores) by completing various measurements in brain specimens for which RNA sequencing data is publicly available through the ACT study (107 donors). Compared to prior studies, they evaluate an overall older sample population. They find an expected correlation between p-Tau levels and dementia. They find a positive correlation between astrocyte and microglia-related gene expression and hippocampal p-Tau levels, but no significant correlation with inflammatory related metrics and dementia.

The authors do not find a significant correlation between any gene expression network and dementia or other metric, which is troubling and pose serious doubts about the data – sample, processing, QC (batch effects), RNA-seq and downstream analyses. The most striking result that the authors present is the absence of any differentially expressed genes between AD and control samples in both hippocampus and cortex, which the authors agree is "surprising. One reason for this is likely the use of SVA which removes surrogate variables that may be confounders, or may not, especially if they are related to sources of variation like cell type. For example, we know at some stages that neurons are down, and there is astrogliosis, as the latter is a pathological feature. This signal has likely been removed by SVA. SVA, although widely used for some reason, has many limitations (e.g. Nature Methods 14, 218-219(2017)doi:10.1038/nmeth.4190). The authors should explore the major PCs in the data, understand what technical or biological factors they are related to. Removing technical confounders is important, but biological "confounders" may be of interest.

Unsurprisingly (as they did not find any DE genes) after performing WGCNA the authors found no modules (module-eigenegenes) to be correlated with diagnosis or AD pathological traits. Many publications have found hundreds of significantly differentially expressed genes between AD and control postmortem human samples (Colangelo et al., 2002, Blalock et al., 2004, Liang et al., 2008, Webster et al., 2009, Avramopoulos et al., 2011, Miller et al., 2013, Zhang et al., 2013, Narayanan et al. 2014, Satoh et al., 2014). The authors claim that many studies do not account for RIN quality, unlike this study, and as a result find gene expression changes between AD cases and controls. However, there are several issues with this:

a) It is unclear why there is discrepancy between the results from published studies which do account for RIN quality (like Zhang el al., 2013, Narayanan et al., 2014, etc. as they use linear regression to remove the effects of RIN) and this data? In this respect can the authors plot the logFC between AD and controls from their data and compare that with published data like Zhang et al. and Narayanan et al.?

b) It seems that the RIN values are heavily confounded with diagnosis in this study and removing the effect of RIN values using regression actually removes the signal. In fact, the authors claim "we find a substantially lower RNA quality in dementia cases vs. controls in all four brain regions". This is a major batch effect and technical confounder that limits the entire value of this study. In the end, the almost total confounding of low RIN with dementia status seriously undermines the conclusions that RIN is driving the changes.

Can the authors show distribution of RIN values (histogram) separately for controls and AD samples? Also, the authors can use the first 5 PCs of gene-expression (before and after regression – separately) and correlate them with various traits like diagnosis, age, sex, PMI, sequencing biases and most importantly RIN.

c) Can they take a subset of the study that is adequately powered relative to previous studies, where cases and controls are matched for RIN and show they get the same result?

d) It is also possible that the systematically lower RIN in demented cases resulted in sequencing biases like AT/GC bias, duplication rate, etc. but the authors seem to have not taken that into account while doing the analysis. Previous publications have shown that these sequencing biases are much better predictor of RNA quality than RIN itself. (Feng et al., 2015)

e) This is not a trivial issue that the authors do not find any DE genes, just based on the huge changes in cell-type proportion between AD and controls (neuronal loss and gliosis), we expect many differentially expressed genes in AD. In fact, the authors show huge microgliosis (Iba-1 staining, Figure 4), but cannot explain any lack of change in differential expression of any microglia related genes. Thus, the lack of any DE genes raises serious doubts about the analysis.

f) In fact, the widespread pathology in the AD samples as shown by AT8 staining, Nissl staining does not correlate with lack of any DE genes. Do the authors expect that pathological dementia cases have no changes in gene expression, – in this case what possibly accounts for the pathology?

Reviewer 3:

The two main issues can be summarized as:

1) there is a tendency just to apply hidden covariate correction models such as SVA without much introspection and SVA is particularly problematic.

2) confounding of RIN with case control status.

This work while perhaps well done to some degree, is not properly interpreted. Use of SVA as it is applied here, almost certainly removes biological signal. The authors must spend some time better understanding what PCs/SVs have been removed, what they are related to -- this will lead to a very different set of conclusions, I suspect.

[Editors’ note: what now follows is the decision letter after the authors submitted for further consideration.]

Thank you for submitting your article "Neuropathological and transcriptomic characteristics of the aged brain" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by Sacha Nelson as the Reviewing Editor and Eve Marder as the Senior Editor. The 3 reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

The authors present a large multi modal analysis of brain specimens from a prior RNA sequencing study of 107 aged human donors including many with dementia and traumatic brain injury. The major findings of the study are negative with respect to associations between neuroinflammation and aging or dementia, and between these factors and gene expression. The paper suggests that RNA quality was correlated with dementia status limiting the ability to independently assess correlations between dementia and gene expression. Although there was disagreement between the reviewers, on balance the reviewers and editors felt that the value of the resource to the community, and the value of the cautions raised with respect to RNA quality, outweighed the lack of other clear findings.

Essential revisions:

All three reviews are included for the author's benefit, but only the major points raised by reviewer #3 and the minor points raised by reviewer's #2 need to be addressed in the revision. Please indicate how these have been addressed in the letter accompanying the revision. These changes should only require textual changes, addition of a supplemental table or figure, and further notes about how code used in analyses were deposited.

Reviewer #1:

I appreciate the authors' effort in making the code available, and in added/reanalysis to demonstrate that the initial finding of little to no association between specific genes and AD pathology were not simply due to the computational method. While this rules out the simplest explanation for the discrepancy between this study and others, it does not in my opinion address the issue fully enough. As the author's state the discrepancy could be due to biological or technical differences in the data set. I don't think ruling out the analytical method as the source of the discrepancy is sufficient to "illuminate the cause of the discrepancies" nor does it sufficiently "add to the ability of careful readers to come to a more integrated understanding" of this study in the context of other carefully carried out prior studies.

I don't mean to imply that I wanted the authors to figure out a way to make their data fit with previous studies, but rather to say that the study still feels unfinished. The finding that their cohort agrees with other studies in some ways but not in others is interesting, and the finding that RNA quality is correlated with AD pathology is also important. But, I don't see any significant light shed on why. I also don't see a straightforward way to address this question in the short time period expected for a resubmission. I am certainly open to hearing the thoughts/suggestions of the other reviewers in case they see some potential that I'm missing. If not, for me, it boils down to a correlation that exists in some studies (including theirs), and unexplained major discrepancies, which doesn't seem suitable for eLife.

Reviewer #2:

The authors have adequately addressed my previously expressed concerns regarding both the specific findings of this paper and the ability to reproduce them, through their inclusion of code and tables used in their analysis, and as long as these are included with the publication or in GitHub. There is definite benefit in releasing this analysis, as it is reasonably performed and now well-documented.

Reviewer #3:

In the current revision, the authors have thoroughly addressed concerns from the reviewers. The major concern raised by all the reviewers is that SVA most likely removed biological signal from the dataset. The authors need to more clearly state this in the final manuscript, and acknowledge the likely many issues with their data – this work does not convincingly undermine previous well controlled studies and that should also be stated. But, the data will be useful to the field as it is a large data set, and the authors have done a good job responding to the reviewers' concerns, so I am favorably inclined.

a) The authors have now looked at the pre-regressed data and found that PC1 is correlated with diagnosis in all brain regions except FWM (where PC2 is correlated to diagnosis). However, the PC1 in these cases is also correlated with RIN, which is a huge confounder that limits the entire analysis. So, unsurprisingly accounting for RIN removes all the signal. The authors should make a point to report the PC correlations with various biological variable like diagnosis, age, sex, RIN, etc. as a supplemental table or supplemental figure pre-regression and post-regression.

b) To address the major confounding issue, the authors take a subset of the study (that is adequately powered relative to previous studies) where cases and controls are matched for RIN, they get the same result – no significant DE genes between cases and controls. Moreover, the authors acknowledge that the lack of any gene-expression changes given severe inflammation and Tau pathology (AT8 staining) in the samples is perplexing. This is very surprising especially given the consistency of previous studies at the RNA and more recently in the AMP-AD data, at the protein level.

My general sense is that the current results are due to confounders, as well as the fact that the underlying biology involves changes in cell composition, which when corrected for, removes the major signal. However, in light of the thorough analysis of the dataset using multiple statistical methods, and its size, this reviewer feels that the dataset/manuscript should be made publicly available so other can use and address the issues with the dataset. As a condition of publication, the authors should set up a GitHub or similar code-deposit service and deposit all the analysis done for this study in a systematic manner. Moreover, the metadata and the pre-normalized and SVA regressed gene-expression data should be submitted to GEO in addition to raw fastqs which should be deposited to dbGAP or similar services

https://doi.org/10.7554/eLife.31126.035

Author response

[Editors’ note: the author responses to the first round of peer review follow.]

Reviewer #1:

Considering this work as a data resource, my main consideration is in how well the results can be reproduced and repurposed, for the many other researchers who will be interested in replicating (or not) findings in their own cohorts. Since the conclusions of this study are somewhat at odds with several other studies, it is particularly important to make all the data needed for reprocessing available.

We completely agree about the importance of reproducibility and repurposing, and that the reviewer for bringing up this issue. For that reason all data, code, and additional files required to reproduce all analyses available in the manuscript are now freely and publicly available, either at http://aging.brain-map.org or as part of a zip file submitted with this version of the manuscript. In addition, the raw data files for RNA-Seq is now available for controlled access through NIAGADS (https://www.niagads.org/datasets/ng00059).

In particular, since many people might want to use their own normalization and RIN was emphasized in this paper, that would be an important variable to share, but I was unable to find that variable in the data-sharing website (nor could I find PMI). In addition, it would be helpful to provide other information that people would ostensibly use in a normalization – percentage aligned reads or other variables that you had access to that you ultimately chose not to use in this particular normalization such as number of ribosomal bases. Also, batch data was unavailable. If these are available and I'm just missing them, then this criticism dissolves.

RIN and batch are available as part one the columns information file that available when downloading the RNA-Seq data from the “Download” tab at http://aging.brain-map.org, and we apologize for not making the location of this information clearer in the initial version of our manuscript. PMI is not available; however, as all of these cases are rapid autopsy cases, the PMI for every donor was <8 hours.

However, it does raise a related point. The exact scripts used in your normalization will enable others to understand how additional variable change their results, in addition to the substantial time savings from having to figure out where everything is in the different files and trying to exactly infer your code from the text. Specifically, it should be possible to take the input files shared on the web – which is all any external researcher will be able to access – and arrive at your normalized data. Indeed, these are essential to providing a firm foundation for alternative normalizations of the data, which I suspect is going to be a major interest of many researchers given the particular results of this study.

We had originally intended to make all code available upon publication as part of a GitHub repository and mentioned this in the original text, but we recognize that we should have made it available during the review process. This code is now available to reviewers as a zip file. We agree that reproducibility is particularly important with regards to this normalization strategy and would encourage others replicate our analysis, and to analyze or normalize the data in a different way.

I appreciate that you have provided your normalized output, but to actually consider modifications to the normalization, the scripts used in getting there are essential. There is a note that code for figures is available on GitHub – however I don't have a link for that and it's not clear it covers normalization. If so – great! – but I don't have a way of looking into this.

The code covers all of the analyses, including normalization, and is now available to the reviewers.

Reviewer #2:

The authors present an analysis of the ACT cohort of human brain tissue taken from aged individuals with and without various pathologies (AD, dementia and TBI). They analyze a broad range of assays such as protein markers and RNA-seq from several brain regions. While these efforts are impressive, in general it seems the findings primarily confirm what is already known about relationships between markers such as AB, pTau, inflammation, and aging, AD or dementia. While a new category (M16) of co-expressed genes was found to be correlated with pTau load, it is not clear whether this is particularly novel in relation to aging or dementia, or is informative for understanding the etiology of either. The paper initially seemed geared towards understanding the aging brain, but very little is presented to shed light on aging and most of what is presented regarding dementia and AD does not appear to be new. It is possible the study contains novel findings that were lost on this reader, but clearer descriptions and context are needed to make this apparent.

We appreciate the reviewer’s understanding of the work that went into creating this resource, and of the insightful critiques. The analyses presented in this manuscript with respect to aging and dementia are largely confirmatory in nature, as the reviewer indicates. Our intension was to present a resource that includes representative features and pathologies of the aged brain, rather than to specifically target changes in the brain that occur during aging (since we do not have young donors in this cohort). By including confirmatory analyses, our goal was to show the validity of the various components of our study. With regards to novelty, to the best of our knowledge we are the first to show a direct connection between genes associated with dementia and RNA quality.

The investigation into RNA quality correlated with dementia is interesting and important, but seems unfinished. The finding that previous studies may be flawed due to a lack of accounting for RNA quality maybe significant enough for a short report. However, it also seems to undermine the current dataset.

We thank the reviewer for these thoughts about our findings with respect to previous reports. Upon further evaluation of the literature, we have scaled back on discussions about how previous studies may be flawed, and instead focus on the importance of accounting for RNA quality. The overlap in genes related to RNA quality and dementia suggests that RNA quality genes could be mistaken for dementia genes if not properly controlled for.

Given the discussion of brain pH and the greatest discordance between this study and a study which took this factor into account raises questions about the relationship between RIN and factors like pH. It is not clear from the analysis presented here that RIN is a sufficiently representative metric to account for all necessary variables.

The question of how best to measure RNA quality is an interesting one, but one that we feel is relatively minor when compared against the relationship between RNA quality and dementia. We now include Table 2 which shows that donors with AD have lower RNA quality than non-demented controls in several studies, in some cases based on RIN and in other cases based on pH, suggesting that this relationship holds for both measures of RNA quality. For this study, pH is not available, but considering RIN can explain up to 80% of gene expression variation for some genes, RIN seems to be a good proxy for RNA quality.

From the data in this study there appeared to be at least two potential issues in differential expression (DE) analysis. (1) Decreased RNA quality correlated with dementia but also (2) increased variance with age. The second factor was mentioned in the Discussion section but not clearly described in the Results section and so it is not clear how this may have impacted DE analysis. Thus, determining the correlation between RIN and dementia is an important issue, it is not clear that the authors are presenting an actual solution.

We did not intend to state that gene expression has higher variance with increasing age, and apologize for the confusion. Instead, we find that the relationship between dementia diagnosis and pathology (or gene expression) is less defined in the oldest old. This means that pathology (or genes) that are expected to show differential patterns in younger donors with and without dementia may not show these differential patterns in an aged cohort. Both this effect as well as the overlapping set of genes associated with dementia and RNA-quality could have led to our findings with respect to differential expression between donors with and without dementia.

The main take-away seems to be that additional work is needed to optimize RNA isolation from pathological brain samples and perhaps a more in-depth investigation into increased expression variability with age. While decreased RNA quality and increase variance may be indicative of biological changes, this study does not seem to help clarify these possibilities. Improved methods to either remove or account for these factors might open up other possible lines of investigation.

We appreciate the reviewer’s feedback with respect to potential changes in experimental design, but respectfully disagree about specifics in methodology negatively impacting our results. The difference in RNA quality between normal and pathological samples is more likely to be a biological effect or something to do with differences in agonal state, than an issue of optimizing RNA isolation, and now include a table showing these differences in several additional cohorts. Having said this, methods for RNA isolation that are more robust to RNA degradation could dramatically improve future studies of the aged brain. In addition, we do not see increased gene expression variability with age (see previous response), and have attempted to update the text accordingly.

For example, the disconnect between pathological markers and dementia was interesting but only superficially investigated. Assuming the hurdles of RNA quality and high variance can be overcome, it would be a natural follow up to probe for gene expression signatures of "tolerance". While I agree that the ACT cohort seems to be a valuable resource, the limitations highlighted in this analysis make it unclear exactly what new information we might pull from it and how best to pull the information.

This is a great insight, and we agree that identifying gene signatures of tolerance would be an important next step. In fact, we are currently pursuing this as an additional study beyond the scope of this manuscript. We hope that by making this resource and our analysis freely available (see responses to Reviewer #1), additional insights into dementia and the aged brain could be obtained.

Reviewer #3:

This work represents an observational study that looks to characterize and associate various factors associated with Alzheimer's dementia (pathological markers, inflammatory markers, gene expression data and clinical dementia scores) by completing various measurements in brain specimens for which RNA sequencing data is publicly available through the ACT study (107 donors). Compared to prior studies, they evaluate an overall older sample population. They find an expected correlation between p-Tau levels and dementia. They find a positive correlation between astrocyte and microglia-related gene expression and hippocampal p-Tau levels, but no significant correlation with inflammatory related metrics and dementia.

The authors do not find a significant correlation between any gene expression network and dementia or other metric, which is troubling and pose serious doubts about the data – sample, processing, QC (batch effects), RNA-seq and downstream analyses. The most striking result that the authors present is the absence of any differentially expressed genes between AD and control samples in both hippocampus and cortex, which the authors agree is "surprising. One reason for this is likely the use of SVA which removes surrogate variables that may be confounders, or may not, especially if they are related to sources of variation like cell type. For example, we know at some stages that neurons are down, and there is astrogliosis, as the latter is a pathological feature. This signal has likely been removed by SVA. SVA, although widely used for some reason, has many limitations (e.g. Nature Methods 14, 218-219(2017)doi:10.1038/nmeth.4190). The authors should explore the major PCs in the data, understand what technical or biological factors they are related to. Removing technical confounders is important, but biological "confounders" may be of interest.

We appreciate the reviewer’s accurate summary of our study and suggestions for potential reasons for our relatively limited associations between gene expression and pathology-related metrics. As noted in the initial version of the text, we agree that our negative result (failing to find gene associated with dementia status) is surprising and could potentially reflect biological or technical differences in this data set compared with previous data (e.g., older cohort, large range in RIN). To assess whether SVA may be removing important gene expression signal, we have repeated our analysis using several additional strategies including some that do not use surrogate variables. In all cases we find the same result, indicating that this is not an artifact of our choice of statistical method. Specifically we add to the Materials and methods section:

“To assess the robustness of our result that few if any genes are significantly associated with dementia (or AD) status after controlling for RIN, we performed additional analyses to quantify significance of gene expression with respect to dementia status, in all cases defining significance as p<0.05 after Bonferroni correction. First, we repeated our analyses using additional statistical tests including 1) two tailed student t-test, 2) ANOVA, 3) and limma (Ritchie et al., 2015), in all cases defining two groups based on dementia (or AD) status. Second, we repeated the SVA analysis described above on the RIN-normalized RNA-Seq data available for download from the website (which uses a slightly different normalization schema, as described above). Third, we performed principal component analysis independently on each region using all genes, and used SVA to assess whether any of the top 25 PCs showed significant associate with dementia. Finally, we subsampled our data set to 70 donors who are matched for RIN, sex, and dementia status and repeated the SVA analysis using data that is not RIN-corrected to determine whether our particular RIN-normalization strategy could be biasing our ability to identify genes associated with dementia. In all cases we found two or fewer total genes associated with dementia or AD, indicating that our negative result is not due to improper statistical assessment.”

While not indicated in the text, we should note that SVA and ANOVA produce identical results using our current set of parameters. Code for reproducing all of these analyses are now included in this submission as a zip file.

Unsurprisingly (as they did not find any DE genes) after performing WGCNA the authors found no modules (module-eigenegenes) to be correlated with diagnosis or AD pathological traits. Many publications have found hundreds of significantly differentially expressed genes between AD and control postmortem human samples (Colangelo et al., 2002, Blalock et al., 2004, Liang et al., 2008, Webster et al., 2009, Avramopoulos et al., 2011, Miller et al., 2013, Zhang et al., 2013, Narayanan et al. 2014, Satoh et al., 2014). The authors claim that many studies do not account for RIN quality, unlike this study, and as a result find gene expression changes between AD cases and controls. However, there are several issues with this:

a) It is unclear why there is discrepancy between the results from published studies which do account for RIN quality (like Zhang el al., 2013, Narayanan et al., 2014, etc. as they use linear regression to remove the effects of RIN) and this data? In this respect can the authors plot the logFC between AD and controls from their data and compare that with published data like Zhang et al. and Narayanan et al.?

We thank the reviewer for these thoughts about our findings with respect to previous reports. After more carefully reviewing the literature we have scaled back on our claims that differences with previous studies are likely due to earlier studies not accounting for RNA quality because, as multiple reviewers correctly point out, some previous studies do control for RIN and still find significant genes. Having said this, one of the points of figure 6e is that, despite the lack of significant genes in our data set, we do find good agreement with genes differentially expressed in previous studies and genes with the highest (albeit non-significant) fold change differences in our current study. This suggests that there is concordance between studies, although we agree that the lower fold changes and non-significant results in our current study are surprising. We have added an additional large study of AD that controls for RIN (Allen et al) and find that it is one of the studies that best agrees with our current results. Finally, we note that nearly all of these studies also show a significant overlap between dementia related genes identified in previous studies and genes associated with RNA quality in this study, even in cases where RIN is corrected. Our interpretation of this is that many of the same genes are associated with both processes, which is how we now present our result.

b) It seems that the RIN values are heavily confounded with diagnosis in this study and removing the effect of RIN values using regression actually removes the signal. In fact, the authors claim "we find a substantially lower RNA quality in dementia cases vs. controls in all four brain regions". This is a major batch effect and technical confounder that limits the entire value of this study. In the end, the almost total confounding of low RIN with dementia status seriously undermines the conclusions that RIN is driving the changes.

We respectfully disagree that the result of lower RNA quality in dementia cases compared with controls is a confounding factor and instead argue that this is a feature that is commonly found in studies of AD. To back this up, we now present table 2, which shows that in 5 of 8 other AD data sets that we could find, the same significant relationship is seen. Specifically we write:

“This result suggested a direct link between RNA quality and dementia status. Indeed, we find a substantially lower RNA quality in dementia cases vs. controls in all four brain regions (Figure 6B; Figure 6—figure supplement 1B), and this difference was not related to the time between death and autopsy (PMI; less than 8 hours for all donors). We next repeated this comparison on data from four additional population-based cohorts as part of the AMP-AD knowledge portal (https://www.synapse.org/ampad; (Bennett et al., 2012a, Bennett et al., 2012b, Allen et al., 2016)), and compared these with previous reports (Colangelo et al., 2002, Preece and Cairns, 2003, Durrenberger et al., 2010, Zhang et al., 2014). In five of the eight additional data sets assayed, donors with AD had significantly lower RIN than donors without dementia (Table 2). Donors with AD had significantly higher RNA quality in only one study (Allen et al., 2016). Thus the link between dementia status and RNA-quality is a broader phenomenon that is not unique to the ACT cohort, but is also not ubiquitous.”

Can the authors show distribution of RIN values (histogram) separately for controls and AD samples?

We appreciate the suggestion to display this data in multiple formats for the purposes of understanding what is going on with respect to RNA quality. Therefore, here is the distribution of RIN values for controls and dementia donors in histogram form:

Author response image 1

Download asset Open asset

This shows the same result that is currently shown in Figure 6B (higher RIN in controls), and therefore we feel that it would be redundant to include as part of the manuscript.

Also, the authors can use the first 5 PCs of gene-expression (before and after regression – separately) and correlate them with various traits like diagnosis, age, sex, PMI, sequencing biases and most importantly RIN.

As requested we have performed a comparison between the top 25 PCs (to explain a larger total fraction of the variance) and disease diagnosis on the RIN-controlled data, as described above, and found no PCs significantly related to dementia status. When we repeated this analysis using the data uncorrected for RIN we found exactly 1 PC corresponding to dementia in each brain region. This was the first PC in all brain regions except FWM, where it was the second PC (PC 1 in FWM likely corresponds to gene expression markers of white matter). These same PCs related with much more significant p-values to RIN (p~0 in all cases). We have expanded this PC analysis to other diagnostic traits and find that PCs significantly associated with tau in hippocampus, with sex, and with inflammation, both in the analyses before and after controlling for RNA quality, recapitulating the results described in out manuscript. We do not find any PCs associated with age at death.

Since this analysis is conceptually similar to what is done with WGCNA, in that vectors relating to large sources of variation are used in place of genes to reduce dimensionality for a comparison against sample metrics, we do not feel that the PC analysis adds significantly to the result of the paper, and currently do not include it. However, if the reviewer feels strongly about this subject, we could include it as a supplemental analysis. In either case, the code required to run this analysis is included as part of the submission.

c) Can they take a subset of the study that is adequately powered relative to previous studies, where cases and controls are matched for RIN and show they get the same result?

This is an excellent suggestion. We have done this and get the same result:

“Finally, we subsampled our data set to 70 donors who are matched for RIN, sex, and dementia status and repeated the SVA analysis using data that is not RIN-corrected to determine whether our particular RIN-normalization strategy could be biasing our ability to identify genes associated with dementia. In all cases we found two or fewer total genes associated with dementia or AD, indicating that our negative result is not due to improper statistical assessment.”

d) It is also possible that the systematically lower RIN in demented cases resulted in sequencing biases like AT/GC bias, duplication rate, etc. but the authors seem to have not taken that into account while doing the analysis. Previous publications have shown that these sequencing biases are much better predictor of RNA quality than RIN itself. (Feng et al., 2015)

Given that some genes show close to 80% of their variance explained by RIN, we find it unlikely that other features of RNA quality would better represent such biases. Having said this, we do find a bias in the relationship between a gene’s GC content and its correlation with RIN: genes with lower fractions of “C” nucleotide in their coding transcripts are more prone to decreased expression with RNA degradation.

Author response image 2

Download asset Open asset

This result was true independent of dementia status. We were unable to identify a strategy to account for these types of biases that led to results different from the ones presented in the manuscript.

e) This is not a trivial issue that the authors do not find any DE genes, just based on the huge changes in cell-type proportion between AD and controls (neuronal loss and gliosis), we expect many differentially expressed genes in AD. In fact, the authors show huge microgliosis (Iba-1 staining, Figure 4), but cannot explain any lack of change in differential expression of any microglia related genes. Thus, the lack of any DE genes raises serious doubts about the analysis.

In fact, the widespread pathology in the AD samples as shown by AT8 staining, Nissl staining does not correlate with lack of any DE genes. Do the authors expect that pathological dementia cases have no changes in gene expression, – in this case what possibly accounts for the pathology?

These final points raised by the reviewer are arguably the most important. In some cases, we do find genes associated with specific pathology and that pathology is associated with dementia, as expected (e.g., tau). In other cases, we find genes associated with a pathology, but find that the pathology itself does not appear to be related to dementia status in our data set (e.g., inflammation/microglia, neuronal loss, gliosis). These findings (as well as the additional analyses not relying on SVA) indicate that the analysis itself is sound, but that our cohort for some reason does not show the usual relationship between disease diagnosis and pathology. Our best explanation for this result is due to the aged nature of this cohort, where in many cases disease diagnosis and pathology do not correlate. For example, eight of the 32 donors (25%) with severe NFT pathology (Braak stage >=5) and six of the 25 donors (24%) with severe amyloid pathology (CERAD score = 3) did not have dementia. It is possible that in such a cohort compensations are also in place for things like inflammation, gliosis, and neuronal loss. A final set of variables (e.g., amyloid load) do not find any associated genes, which is quite surprising. We don’t have a good explanation for this, but want to reiterate the soundness of our analysis.

Reviewer 3: The two main issues can be summarized as:

1) there is a tendency just to apply hidden covariate correction models such as SVA without much introspection and SVA is particularly problematic.

2) confounding of RIN with case control status.

This work while perhaps well done to some degree, is not properly interpreted. Use of SVA as it is applied here, almost certainly removes biological signal. The authors must spend some time better understanding what PCs/SVs have been removed, what they are related to -- this will lead to a very different set of conclusions, I suspect.

We have discussed these two important points brought up by the reviewer above. In short, (1) repeating the analysis using different strategies leads to the same results, indicated that SVA is not negatively impacting our interpretation, and (2) we feel that the result that RIN is lower in dementia cases compared with controls is an important finding rather than a confounder. Having said this, we appreciate the reviewer bringing up these concerns so we can ensure that our analysis is as technically sound as possible.

https://doi.org/10.7554/eLife.31126.036

Article and author information

Author details

Jeremy A Miller

Allen Institute for Brain Science, Seattle, United States

Contribution
Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review and editing

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0003-4549-588X
Angela Guillozet-Bongaarts

Allen Institute for Brain Science, Seattle, United States

Contribution
Formal analysis, Supervision, Methodology, Project administration

Competing interests
No competing interests declared
Laura E Gibbons

Department of Medicine, University of Washington, Seattle, United States

Contribution
Formal analysis

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0001-5054-2543
Nadia Postupna

Department of Pathology, University of Washington, Seattle, United States

Contribution
Formal analysis, Investigation, Project administration

Competing interests
No competing interests declared
Anne Renz

Kaiser Permanente Washington Health Research Institute, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Allison E Beller

Department of Pathology, University of Washington, Seattle, United States

Contribution
Formal analysis, Project administration

Competing interests
No competing interests declared
Susan M Sunkin

Allen Institute for Brain Science, Seattle, United States

Contribution
Project administration

Competing interests
No competing interests declared
Lydia Ng

Allen Institute for Brain Science, Seattle, United States

Contribution
Formal analysis, Supervision, Methodology

Competing interests
No competing interests declared
Shannon E Rose

Department of Pathology, University of Washington, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Kimberly A Smith

Allen Institute for Brain Science, Seattle, United States

Contribution
Supervision, Methodology, Project administration

Competing interests
No competing interests declared
Aaron Szafer

Allen Institute for Brain Science, Seattle, United States

Contribution
Formal analysis, Methodology

Competing interests
No competing interests declared
Chris Barber

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Darren Bertagnolli

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Kristopher Bickley

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Krissy Brouner

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Shiella Caldejon

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Mike Chapin

Allen Institute for Brain Science, Seattle, United States

Contribution
Supervision, Investigation

Competing interests
No competing interests declared
Mindy L Chua

Department of Pathology, University of Washington, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Natalie M Coleman

Department of Pathology, University of Washington, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Eiron Cudaback

Department of Pathology, University of Washington, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Christine Cuhaciyan

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Rachel A Dalley

Allen Institute for Brain Science, Seattle, United States

Contribution
Formal analysis, Methodology

Competing interests
No competing interests declared
Nick Dee

Allen Institute for Brain Science, Seattle, United States

Contribution
Formal analysis, Investigation, Methodology

Competing interests
No competing interests declared
Tsega Desta

Allen Institute for Brain Science, Seattle, United States

Contribution
Formal analysis, Supervision

Competing interests
No competing interests declared
Tim A Dolbeare

Allen Institute for Brain Science, Seattle, United States

Contribution
Supervision, Investigation, Methodology

Competing interests
No competing interests declared
Nadezhda I Dotson

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0003-3414-3176
Michael Fisher

Allen Institute for Brain Science, Seattle, United States

Contribution
Formal analysis, Investigation

Competing interests
No competing interests declared
Nathalie Gaudreault

Allen Institute for Brain Science, Seattle, United States

Contribution
Supervision, Investigation

Competing interests
No competing interests declared
Garrett Gee

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Terri L Gilbert

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Jeff Goldy

Allen Institute for Brain Science, Seattle, United States

Contribution
Formal analysis, Investigation, Methodology

Competing interests
No competing interests declared
Fiona Griffin

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Caroline Habel

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Zeb Haradon

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Nika Hejazinia

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Leanne L Hellstern

Department of Pathology, University of Washington, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Steve Horvath

Department of Human Genetics, University of California, Los Angeles, Los Angeles, United States

Contribution
Formal analysis

Competing interests
No competing interests declared
Kim Howard

Department of Pathology, University of Washington, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Robert Howard

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Justin Johal

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Nikolas L Jorstad

Department of Pathology, University of Washington, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Samuel R Josephsen

Department of Pathology, University of Washington, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Chihchau L Kuan

Allen Institute for Brain Science, Seattle, United States

Contribution
Formal analysis, Investigation, Methodology

Competing interests
No competing interests declared
Florence Lai

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Eric Lee

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0002-7166-0909
Felix Lee

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation, Methodology

Competing interests
No competing interests declared
Tracy Lemon

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Xianwu Li

Department of Pathology, University of Washington, Seattle, United States

Contribution
Formal analysis, Investigation

Competing interests
No competing interests declared
Desiree A Marshall

Department of Pathology, University of Washington, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Jose Melchor

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Shubhabrata Mukherjee

Department of Medicine, University of Washington, Seattle, United States

Contribution
Formal analysis

Competing interests
No competing interests declared
Julie Nyhus

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation, Methodology

Competing interests
No competing interests declared
Julie Pendergraft

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0002-6066-4576
Lydia Potekhina

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Elizabeth Y Rha

Department of Pathology, University of Washington, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Samantha Rice

Department of Pathology, University of Washington, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
David Rosen

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Abharika Sapru

Department of Pathology, University of Washington, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Aimee Schantz

Department of Pathology, University of Washington, Seattle, United States

Contribution
Project administration

Competing interests
No competing interests declared
Elaine Shen

Allen Institute for Brain Science, Seattle, United States

Contribution
Project administration

Competing interests
No competing interests declared
Emily Sherfield

Department of Pathology, University of Washington, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Shu Shi

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Andy J Sodt

Allen Institute for Brain Science, Seattle, United States

Contribution
Supervision, Investigation

Competing interests
No competing interests declared
Nivretta Thatra

Allen Institute for Brain Science, Seattle, United States

Contribution
Formal analysis, Investigation

Competing interests
No competing interests declared
Michael Tieu

Allen Institute for Brain Science, Seattle, United States

Contribution
Investigation

Competing interests
No competing interests declared
Angela M Wilson

Department of Pathology, University of Washington, Seattle, United States

Contribution
Investigation, Methodology

Competing interests
No competing interests declared
Thomas J Montine

Department of Pathology, University of Washington, Seattle, United States

Contribution
Conceptualization, Methodology

Competing interests
No competing interests declared
Eric B Larson

Kaiser Permanente Washington Health Research Institute, Seattle, United States

Contribution
Conceptualization, Supervision

Competing interests
No competing interests declared
Amy Bernard

Allen Institute for Brain Science, Seattle, United States

Contribution
Formal analysis, Supervision, Investigation, Methodology

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0003-2540-1153
Paul K Crane

Department of Medicine, University of Washington, Seattle, United States

Contribution
Conceptualization, Supervision, Writing—original draft, Writing—review and editing

Competing interests
No competing interests declared
Richard G Ellenbogen

Department of Neurological Surgery, University of Washington, Seattle, United States

Contribution
Conceptualization, Supervision, Writing—review and editing

Competing interests
No competing interests declared
C Dirk Keene

Department of Pathology, University of Washington, Seattle, United States

Contribution
Conceptualization, Supervision, Investigation, Methodology, Writing—original draft, Project administration, Writing—review and editing

Contributed equally with
Ed Lein

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0002-5291-1469
Ed Lein

Allen Institute for Brain Science, Seattle, United States

Contribution
Conceptualization, Formal analysis, Supervision, Methodology, Writing—original draft, Writing—review and editing

Contributed equally with
C Dirk Keene

For correspondence
EdL@alleninstitute.org

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0001-9012-6552

Funding

National Institutes of Health (P50AG005136)

Eric B Larson
Paul K Crane

National Institutes of Health (U01AG006781)

Eric B Larson
Paul K Crane

Nancy and Buster Alvord Endowment

Richard G Ellenbogen
C Dirk Keene
Ed Lein

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank the Allen Institute for Brain Science founders, PG Allen and J Allen, for their vision, encouragement, and support. This work was funded by a grant to CD Keene, RG. Ellenbogen and Ed Lein from the Paul G Allen Family Foundation, and supported by National Institutes of Health grants U01AG006781 andP50AG005136, and the Nancy and Buster Alvord Endowment. We are grateful for the technical and administrative support of the staff members in the Allen Institute and from the University of Washington and Kaiser Permanente Washington Health Research Institute who are not part of the authorship of this paper. ROS/MAP study data were provided by the Rush Alzheimer’s Disease Center, Rush University Medical Center, Chicago. Data collection was supported through funding by NIA grants P30AG10161, R01AG15819, R01AG17917, R01AG30146, R01AG36836, U01AG32984, U01AG46152, the Illinois Department of Public Health, and the Translational Genomics Research Institute. Mayo RNA-Seq study data were provided by the following sources: The Mayo Clinic Alzheimers Disease Genetic Studies, led by Dr. Nilufer Taner and Dr. Steven G. Younkin, Mayo Clinic, Jacksonville, FL using samples from the Mayo Clinic Study of Aging, the Mayo Clinic Alzheimers Disease Research Center, and the Mayo Clinic Brain Bank. Data collection was supported through funding by NIA grants P50 AG016574, R01 AG032990, U01 AG046139, R01 AG018023, U01 AG006576, U01 AG006786, R01 AG025711, R01 AG017216, R01 AG003949, NINDS grant R01 NS080820, CurePSP Foundation, and support from Mayo Foundation. Study data includes samples collected through the Sun Health Research Institute Brain and Body Donation Program of Sun City, Arizona. The Brain and Body Donation Program is supported by the National Institute of Neurological Disorders and Stroke (U24 NS072026 National Brain and Tissue Resource for Parkinsons Disease and Related Disorders), the National Institute on Aging (P30 AG19610 Arizona Alzheimers Disease Core Center), the Arizona Department of Health Services (contract 211002, Arizona Alzheimers Research Center), the Arizona Biomedical Research Commission (contracts 4001, 0011, 05–901 and 1001 to the Arizona Parkinson's Disease Consortium) and the Michael J Fox Foundation for Parkinsons Research. MSBB data were generated from postmortem brain tissue collected through the Mount Sinai VA Medical Center Brain Bank and were provided by Dr. Eric Schadt from Mount Sinai School of Medicine.

Reviewing Editor

Sacha B Nelson, Brandeis University, United States

Version history

Received: August 9, 2017
Accepted: October 22, 2017
Version of Record published: November 9, 2017 (version 1)

Copyright

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

6,909

Page views
1,000

Downloads
65

Citations

Article citation count generated by polling the highest count across the following sources: Scopus, Crossref, PubMed Central.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Mendeley

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

Jeremy A Miller
Angela Guillozet-Bongaarts
Laura E Gibbons
Nadia Postupna
Anne Renz
Allison E Beller
Susan M Sunkin
Lydia Ng
Shannon E Rose
Kimberly A Smith
Aaron Szafer
Chris Barber
Darren Bertagnolli
Kristopher Bickley
Krissy Brouner
Shiella Caldejon
Mike Chapin
Mindy L Chua
Natalie M Coleman
Eiron Cudaback
Christine Cuhaciyan
Rachel A Dalley
Nick Dee
Tsega Desta
Tim A Dolbeare
Nadezhda I Dotson
Michael Fisher
Nathalie Gaudreault
Garrett Gee
Terri L Gilbert
Jeff Goldy
Fiona Griffin
Caroline Habel
Zeb Haradon
Nika Hejazinia
Leanne L Hellstern
Steve Horvath
Kim Howard
Robert Howard
Justin Johal
Nikolas L Jorstad
Samuel R Josephsen
Chihchau L Kuan
Florence Lai
Eric Lee
Felix Lee
Tracy Lemon
Xianwu Li
Desiree A Marshall
Jose Melchor
Shubhabrata Mukherjee
Julie Nyhus
Julie Pendergraft
Lydia Potekhina
Elizabeth Y Rha
Samantha Rice
David Rosen
Abharika Sapru
Aimee Schantz
Elaine Shen
Emily Sherfield
Shu Shi
Andy J Sodt
Nivretta Thatra
Michael Tieu
Angela M Wilson
Thomas J Montine
Eric B Larson
Amy Bernard
Paul K Crane
Richard G Ellenbogen
C Dirk Keene
Ed Lein

(2017)

Neuropathological and transcriptomic characteristics of the aged brain

eLife 6:e31126.

https://doi.org/10.7554/eLife.31126

Categories and tags

Research organism

Human

Share this article

Cite this article

Experimental design and cohort characteristics.

Summary of demographics for donors with and without dementia.

Amyloid beta and tau pathologies show a relationship with age, but with high variability.

Disconnect between pathology and dementia status.

Gene expression signatures of inflammation.

Figure 4—source data 1

Figure 4—source data 2

Figure 4—source data 3

Gene expression signatures of dementia and related pathology.

Differences in RNA quality between dementia and controls greatly impact gene expression results.

Figure 6—source data 1

Figure 6—source data 2

Figure 6—source data 3

Figure 6—source data 4

Figure 6—source data 5

Figure 6—source data 6

RNA quality assessment for donors with and without AD in multiple studies.

Author details

Jeremy A Miller

Contribution

Competing interests

Angela Guillozet-Bongaarts

Contribution

Competing interests

Laura E Gibbons

Contribution

Competing interests

Nadia Postupna

Contribution

Competing interests

Anne Renz

Contribution

Competing interests

Allison E Beller

Contribution

Competing interests

Susan M Sunkin

Contribution

Competing interests

Lydia Ng

Contribution

Competing interests

Shannon E Rose

Contribution

Competing interests

Kimberly A Smith

Contribution

Competing interests

Aaron Szafer

Contribution

Competing interests

Chris Barber

Contribution

Competing interests

Darren Bertagnolli

Contribution

Competing interests

Kristopher Bickley

Contribution

Competing interests

Krissy Brouner

Contribution

Competing interests

Shiella Caldejon

Contribution

Competing interests

Mike Chapin

Contribution

Competing interests

Mindy L Chua

Contribution

Competing interests

Natalie M Coleman

Contribution

Competing interests

Eiron Cudaback

Contribution

Competing interests