[Skip to Content]
Sign In
Individual Sign In
Create an Account
Institutional Sign In
OpenAthens Shibboleth
[Skip to Content Landing]
Figure 1.
Transverse Relaxation Rate (R2) Regions Associated With Cognitive Decline
Transverse Relaxation Rate (R2) Regions Associated With Cognitive Decline

Two separate views are shown of the 4 white matter regions (colorized) in which the R2 was associated with late-life cognitive decline (global cognition). To provide spatial context, the 3-dimensional regions are superimposed on a 2-dimensional sagittal slice of an ex vivo brain hemisphere template, shown in gray scale. The same 2-dimensional slice is used in both views, with the arrow oriented in the anterior direction in both cases. The left image is captured from a more medial camera position, while the right image is captured from a more lateral camera position.

Figure 2.
Genomic Associations With the Transverse Relaxation Rate (R2) Measure
Genomic Associations With the Transverse Relaxation Rate (R2) Measure

A, Significance level vs the regression estimates for individual cytosine-guanine dinucleotide (CG) sites. B, Significance level vs the regression estimates for individual genes. Orange dots represent CGs (A) and genes (B) that reached the significance level of P < .001. In both cases, most of the top signals are anticorrelated with R2 such that the higher the methylation (or gene expression), the slower the relaxation rate.

Table 1.  
Characteristics of the Study Participants
Characteristics of the Study Participants
Table 2.  
Transverse Relaxation Rate (R2) and Cognitive Decline Controlling for Common Neuropathologic Indicesa
Transverse Relaxation Rate (R2) and Cognitive Decline Controlling for Common Neuropathologic Indicesa
Table 3.  
Top Genes Associated With the Transverse Relaxation Rate (R2) Measurea
Top Genes Associated With the Transverse Relaxation Rate (R2) Measurea
1.
Hebert  LE, Weuve  J, Scherr  PA, Evans  DA.  Alzheimer disease in the United States (2010-2050) estimated using the 2010 census.  Neurology. 2013;80(19):1778-1783.PubMedGoogle ScholarCrossref
2.
Boyle  PA, Yang  J, Yu  L,  et al.  Varied effects of age-related neuropathologies on the trajectory of late life cognitive decline.  Brain. 2017;140(3):804-812.PubMedGoogle Scholar
3.
White  LR, Edland  SD, Hemmy  LS,  et al.  Neuropathologic comorbidity and cognitive impairment in the Nun and Honolulu-Asia Aging Studies.  Neurology. 2016;86(11):1000-1008.PubMedGoogle ScholarCrossref
4.
Kryscio  RJ, Abner  EL, Nelson  PT,  et al.  The effect of vascular neuropathology on late-life cognition: results from the SMART project.  J Prev Alzheimers Dis. 2016;3(2):85-91.PubMedGoogle Scholar
5.
Troncoso  JC, Zonderman  AB, Resnick  SM, Crain  B, Pletnikova  O, O’Brien  RJ.  Effect of infarcts on dementia in the Baltimore Longitudinal Study of Aging.  Ann Neurol. 2008;64(2):168-176.PubMedGoogle ScholarCrossref
6.
Boyle  PA, Wilson  RS, Yu  L,  et al.  Much of late life cognitive decline is not due to common neurodegenerative pathologies.  Ann Neurol. 2013;74(3):478-489.PubMedGoogle ScholarCrossref
7.
Bauer  CM, Jara  H, Killiany  R; Alzheimer’s Disease Neuroimaging Initiative.  Whole brain quantitative T2 MRI across multiple scanners with dual echo FSE: applications to AD, MCI, and normal aging.  Neuroimage. 2010;52(2):508-514.PubMedGoogle ScholarCrossref
8.
House  MJ, St Pierre  TG, Foster  JK, Martins  RN, Clarnette  R.  Quantitative MR imaging R2 relaxometry in elderly participants reporting memory loss.  AJNR Am J Neuroradiol. 2006;27(2):430-439.PubMedGoogle Scholar
9.
Granziera  C, Daducci  A, Donati  A,  et al.  A multi-contrast MRI study of microstructural brain damage in patients with mild cognitive impairment.  Neuroimage Clin. 2015;8:631-639.PubMedGoogle ScholarCrossref
10.
Arfanakis  K, Gui  M, Tamhane  AA, Carew  JD.  Investigating the medial temporal lobe in Alzheimer’s disease and mild cognitive impairment, with turboprop diffusion tensor imaging, MRI-volumetry, and T2-relaxometry.  Brain Imaging Behav. 2007;1(1-2):11-21. doi:10.1007/s11682-007-9001-4Google ScholarCrossref
11.
Dawe  RJ, Yu  L, Leurgans  SE,  et al.  Postmortem MRI: a novel window into the neurobiology of late life cognitive decline.  Neurobiol Aging. 2016;45:169-177.PubMedGoogle ScholarCrossref
12.
Bennett  DA, Schneider  JA, Arvanitakis  Z, Wilson  RS.  Overview and findings from the Religious Orders Study.  Curr Alzheimer Res. 2012;9(6):628-645.PubMedGoogle ScholarCrossref
13.
Bennett  DA, Schneider  JA, Buchman  AS, Barnes  LL, Boyle  PA, Wilson  RS.  Overview and findings from the Rush Memory and Aging Project.  Curr Alzheimer Res. 2012;9(6):646-663.PubMedGoogle ScholarCrossref
14.
Dawe  RJ, Bennett  DA, Schneider  JA, Vasireddi  SK, Arfanakis  K.  Postmortem MRI of human brain hemispheres: T2 relaxation times during formaldehyde fixation.  Magn Reson Med. 2009;61(4):810-818.PubMedGoogle ScholarCrossref
15.
Dawe  RJ, Bennett  DA, Schneider  JA, Arfanakis  K.  Neuropathologic correlates of hippocampal atrophy in the elderly: a clinical, pathologic, postmortem MRI study.  PLoS One. 2011;6(10):e26286.PubMedGoogle ScholarCrossref
16.
Wilson  RS, Boyle  PA, Yu  L, Segawa  E, Sytsma  J, Bennett  DA.  Conscientiousness, dementia related pathology, and trajectories of cognitive aging.  Psychol Aging. 2015;30(1):74-82.PubMedGoogle ScholarCrossref
17.
Dawe  RJ, Bennett  DA, Schneider  JA,  et al.  Ex vivo T2 relaxation: associations with age-related neuropathology and cognition.  Neurobiol Aging. 2014;35(7):1549-1561.PubMedGoogle ScholarCrossref
18.
Ishigami  A, Ohsawa  T, Hiratsuka  M,  et al.  Abnormal accumulation of citrullinated proteins catalyzed by peptidylarginine deiminase in hippocampal extracts from patients with Alzheimer’s disease.  J Neurosci Res. 2005;80(1):120-128.PubMedGoogle ScholarCrossref
19.
Mastronardi  FG, Noor  A, Wood  DD, Paton  T, Moscarello  MA.  Peptidyl argininedeiminase 2 CpG island in multiple sclerosis white matter is hypomethylated.  J Neurosci Res. 2007;85(9):2006-2016.PubMedGoogle ScholarCrossref
20.
Sharma  S, Dimasi  D, Higginson  K, Della  NG.  RZF, a zinc-finger protein in the photoreceptors of human retina.  Gene. 2004;342(2):219-229.PubMedGoogle ScholarCrossref
21.
Yu  L, Boyle  P, Schneider  JA,  et al.  APOE ε4, Alzheimer’s disease pathology, cerebrovascular disease, and cognitive change over the years prior to death.  Psychol Aging. 2013;28(4):1015-1023.PubMedGoogle ScholarCrossref
22.
Yu  L, Boyle  PA, Leurgans  S, Schneider  JA, Bennett  DA.  Disentangling the effects of age and APOE on neuropathology and late life cognitive decline.  Neurobiol Aging. 2014;35(4):819-826.PubMedGoogle ScholarCrossref
23.
Musse  AA, Li  Z, Ackerley  CA,  et al.  Peptidylarginine deiminase 2 (PAD2) overexpression in transgenic mice leads to myelin loss in the central nervous system.  Dis Model Mech. 2008;1(4-5):229-240.PubMedGoogle ScholarCrossref
24.
Moscarello  MA, Pritzker  L, Mastronardi  FG, Wood  DD.  Peptidylarginine deiminase: a candidate factor in demyelinating disease.  J Neurochem. 2002;81(2):335-343.PubMedGoogle ScholarCrossref
25.
Rovira  À, Wattjes  MP, Tintoré  M,  et al; MAGNIMS Study Group.  Evidence-based guidelines: MAGNIMS consensus guidelines on the use of MRI in multiple sclerosis—clinical implementation in the diagnostic process.  Nat Rev Neurol. 2015;11(8):471-482.PubMedGoogle ScholarCrossref
26.
Calabrese  R, Zampieri  M, Mechelli  R,  et al.  Methylation-dependent PAD2 upregulation in multiple sclerosis peripheral blood.  Mult Scler. 2012;18(3):299-304.PubMedGoogle ScholarCrossref
27.
Yu  L, Chibnik  LB, Srivastava  GP,  et al.  Association of brain DNA methylation in SORL1, ABCA7, HLA-DRB5, SLC24A4, and BIN1 with pathological diagnosis of Alzheimer disease.  JAMA Neurol. 2015;72(1):15-24.PubMedGoogle ScholarCrossref
28.
Chen  S, Lu  FF, Seeman  P, Liu  F.  Quantitative proteomic analysis of human substantia nigra in Alzheimer’s disease, Huntington’s disease and multiple sclerosis.  Neurochem Res. 2012;37(12):2805-2813.PubMedGoogle ScholarCrossref
29.
Jang  B, Ishigami  A, Maruyama  N, Carp  RI, Kim  YS, Choi  EK.  Peptidylarginine deiminase and protein citrullination in prion diseases: strong evidence of neurodegeneration.  Prion. 2013;7(1):42-46.PubMedGoogle ScholarCrossref
30.
Hernandez  DG, Nalls  MA, Gibbs  JR,  et al.  Distinct DNA methylation changes highly correlated with chronological age in the human brain.  Hum Mol Genet. 2011;20(6):1164-1172.PubMedGoogle ScholarCrossref
31.
Yang  J, Yu  L, Gaiteri  C,  et al.  Association of DNA methylation in the brain with age in older persons is confounded by common neuropathologies.  Int J Biochem Cell Biol. 2015;67:58-64.PubMedGoogle ScholarCrossref
32.
Moscarello  MA, Lei  H, Mastronardi  FG,  et al.  Inhibition of peptidyl-arginine deiminases reverses protein-hypercitrullination and disease in mouse models of multiple sclerosis.  Dis Model Mech. 2013;6(2):467-478.PubMedGoogle ScholarCrossref
Original Investigation
December 2017

Association Between Brain Gene Expression, DNA Methylation, and Alteration of Ex Vivo Magnetic Resonance Imaging Transverse Relaxation in Late-Life Cognitive Decline

Author Affiliations
  • 1Rush Alzheimer’s Disease Center, Rush University Medical Center, Chicago, Illinois
  • 2Department of Neurological Sciences, Rush University Medical Center, Chicago, Illinois
  • 3Department of Diagnostic Radiology and Nuclear Medicine, Rush University Medical Center, Chicago, Illinois
  • 4Department of Behavioral Sciences, Rush University Medical Center, Chicago, Illinois
  • 5Department of Pathology, Rush University Medical Center, Chicago, Illinois
  • 6Department of Biomedical Engineering, Illinois Institute of Technology, Chicago
  • 7Center for Translational and Computational Neuroimmunology, Columbia University Medical Center, New York, New York
  • 8Program in Medical and Population Genetics, Broad Institute, Cambridge, Massachusetts
JAMA Neurol. 2017;74(12):1473-1480. doi:10.1001/jamaneurol.2017.2807
Key Points

Question  What are the associations between brain gene expression, DNA methylation, and alteration of ex vivo magnetic resonance imaging transverse relaxation in late-life cognitive decline above and beyond common neuropathologic conditions?

Findings  From analysis of data from longitudinal clinicopathologic cohort studies of aging, 4 genes were identified (ie, PADI2, ZNF385A, PSD2, and A2ML1), of which higher expression was associated with a slower transverse relaxation rate. Furthermore, an anticorrelation pattern between the transverse relaxation rate and DNA methylation in many of the cytosine-guanine dinucleotides was observed.

Meaning  Gene expression and DNA methylation are implicated in the alteration of brain tissue properties associated with cognitive decline.

Abstract

Importance  Alteration of ex vivo magnetic resonance imaging transverse relaxation is associated with late-life cognitive decline even after controlling for common neuropathologic conditions. However, the underlying neurobiology of this association is unknown.

Objective  To investigate the association between brain gene expression, DNA methylation, and alteration of magnetic resonance imaging transverse relaxation in late-life cognitive decline.

Design, Setting, and Participants  Data came from 2 community-based longitudinal cohort studies of aging and dementia, the Religious Orders Study, which began in 1993, and the Rush Memory and Aging Project, which began in 1997. All participants agreed to undergo annual clinical evaluations and to donate their brains after death. By October 24, 2016, a total of 1358 individuals had died and had brain autopsies that were approved by board-certified neuropathologists. Of those, 552 had undergone ex vivo imaging. The gene expression analysis was limited to 174 individuals with both imaging and brain RNA sequencing data. The DNA methylation analysis was limited to 225 individuals with both imaging and brain methylation data.

Main Outcomes and Measures  Maps of ex vivo magnetic resonance imaging transverse relaxation were generated using fast spin echo imaging. The target was a composite measure of the transverse relaxation rate (R2) that was associated with cognitive decline after controlling for common neuropathologic conditions. Next-generation RNA sequencing and DNA methylation data were generated using frozen tissue from the dorsolateral prefrontal cortex. Genome-wide association analysis was used to investigate gene expression and, separately, DNA methylation for signals associated with the R2 measure.

Results  Of the 552 individuals with ex vivo imaging data, 394 were women and 158 were men, and the mean (SD) age at death was 90.4 (6.0) years. Four co-expressed genes (PADI2 [Ensembl ENSG00000117115], ZNF385A [Ensembl ENSG00000161642], PSD2 [Ensembl ENSG00000146005], and A2ML1 [Ensembl ENSG00000166535]) were identified, of which higher expressions were associated with slower R2. The association of R2 with cognitive decline was attenuated when the gene expression signals were added to the model, such that the mean (SE) coefficient of association was reduced from 0.028 (0.008) (P < .001) to 0.019 (0.009) (P = .03). The DNA methylation scan did not detect a genome-wide significant signal, but it revealed an anticorrelation between R2 and DNA methylation in many of the cytosine-guanine dinucleotides.

Conclusions and Relevance  Brain gene expression and DNA methylation dysregulations are implicated in the alteration of brain tissue properties associated with late-life cognitive decline above and beyond the influence of common neuropathologic conditions.

Introduction

As approximately 14 million Americans 65 years of age or older will have Alzheimer disease (AD) dementia by 2050,1 prevention and treatment of AD dementia becomes increasingly important to public health. Alzheimer disease, cerebrovascular disease, Lewy body disease, hippocampal sclerosis, and the more recently observed transactive response DNA-binding protein 43 (TDP-43) are common neuropathologic conditions that contribute to late-life cognitive decline, the clinical hallmark of AD dementia.2-5 Yet, a majority of person-to-person variations in cognitive decline are not attributable to these pathologic conditions, suggesting that additional determinants of cognitive decline remain to be identified.6 Ex vivo brain magnetic resonance imaging (MRI) complements common neuropathologic measures by providing information on brain structure and tissue integrity that is not captured by traditional histopathologic findings. In particular, transverse relaxation properties measure the integrity of neural tissue and are important correlates of cognition and memory.7 A slower transverse relaxation rate (R2) in white matter regions is associated with poorer cognitive performance and faster cognitive decline.8-11 The association between R2 and cognitive decline is in addition to the influence of key neuropathologic measures.11 Investigation of the neurobiology underlying the R2 alteration in cognitive decline may lead to novel therapeutic targets to combat late-life cognitive impairment and AD dementia.

Genes dictate cell function, and gene expression profiling has proved to be fundamental in understanding the genetic architecture of complex human diseases. In addition, gene expression can be regulated by epigenetic markers. DNA methylation, the addition of a methyl group to the cytosine base of cytosine-guanine dinucleotides (CGs), is one of the most common epigenetic markers that regulate access to DNA sequence and, consequently, gene expression. Identifying differentially expressed genes and differentially methylated CGs that are implicated in R2 and cognitive decline holds the potential to better understand the transcriptomic and epigenetic footprints underlying the association between R2 and cognitive impairment.

In this study, we investigated gene expression and DNA methylation in postmortem human brains, examined their association with R2 alteration, and explored the degree to which these markers contribute to the association between R2 and cognitive decline. Postmortem specimens came from autopsied participants from 2 ongoing clinicopathologic cohort studies of aging and dementia. Genome-wide gene expression and DNA methylation profiles were generated using frozen tissue from the dorsolateral prefrontal cortex (DLPFC). We targeted a composite ex vivo R2 measure that was associated with cognitive decline after controlling for common neuropathologic indices.

Methods
Study Participants

Participants came from the Religious Orders Study and the Rush Memory and Aging Project.12,13 Participants were community-based older persons without known dementia who agreed to annual, detailed clinical and cognitive evaluations and to donate their brains after death. Both studies were approved by the Rush University Medical Center institutional review board. Written informed consent was obtained from each participant, and each participant signed an Anatomical Gift Act consent donating his or her brain.

The Religious Orders Study started in 1993, and the Rush Memory and Aging Project started in 1997. At the time of the analyses (October 24, 2016), 3308 participants were enrolled, 1599 had died, and 1383 had undergone autopsy. The autopsy rate was 86.5%. Of the autopsied participants, 1358 cases were reviewed and approved by board-certified neuropathologists. The collection of ex vivo imaging, RNA sequencing, and DNA methylation data started later. As a result, ex vivo imaging was available for 552 autopsy-approved participants. The analysis for genome-wide gene expression was limited to 174 participants with available imaging and RNA sequencing data. The analysis for genome-wide DNA methylation was limited to 225 participants who had both imaging and brain DNA methylation data. The Venn diagram (eFigure 1 in the Supplement) describes the sample sizes across different data sets. The demographic and neuropathologic characteristics were largely comparable across the samples (Table 1).

Ex Vivo MRI and R2 Associated With Cognitive Decline

Brains were removed and bisected after death following a standard procedure. Hemispheres chosen for imaging were fixed in 4% paraformaldehyde and stored in refrigerators at 4°C. Scans were performed a mean (SD) of 41.1 (19.2) days (range, 23-235 days) after death, and each scan included a 2-dimensional turbo spin echo sequence with multiple echo times. The original R2 value was estimated voxel by voxel from the turbo spin echo data by fitting a monoexponential decay function, then normalized within each group of hemispheres imaged on a given MRI scanner.14,15 Details on scanner-specific normalization are provided in the eAppendix in the Supplement.

We considered but elected not to restrict the R2 metric to the DLPFC, the actual tissue collection site where RNA sequencing and DNA methylation data were obtained. The literature extensively supports the functional and structural connectivity of the DLPFC with other brain regions, and genes expressed or methylated in 1 region are closely linked to, and possibly even influence, imaging signatures throughout the brain. Therefore, we focused on a composite R2 measure that captures the overall relaxation associated with cognitive decline above and beyond the influence of commonly measured neuropathologic conditions.11

In brief, person-specific rates of cognitive decline (ie, slopes) were estimated using longitudinal data from annual cognitive assessments prior to death. Next, voxelwise linear regressions were performed by regressing person-specific rates of decline against the R2 value of each of approximately 400 000 tissue-containing voxels, controlling for the demographics and 9 common neuropathologic indices. A false discovery rate of 0.05 was used to correct for multiple testing. These analyses identified 4 clusters of voxels, each consisting of at least 100 contiguous voxels (100 mm3) that were associated with cognitive decline (Figure 1). A composite score of R2 was obtained by averaging individual voxel R2 values first within and then across clusters and standardized using the overall mean and SD. The use of a composite R2 measure was supported by the positive intercorrelations among individual clusters and the result from exploratory factor analysis (eFigure 2 and 3 in the Supplement). Specifically, pairwise Pearson correlations between clusters ranged from 0.43 to 0.63 (P < .001 for all correlations). Furthermore, only 1 principal component was retained in the factor analysis, which contributes to most (65%) of the total variance.

Gene Expression and DNA Methylation

Details on gene expression and DNA methylation data processing are provided in the eAppendix in the Supplement. In brief, RNA and DNA samples were extracted from frozen DLPFC tissue, and RNA sequencing was performed on the Illumina HiSeq (Illumina Inc). The gene expression level was measured, with higher value corresponding to higher expression. Genome-wide DNA methylation data were generated using the Illumina Infinium HumanMethylation450 BeadChip (Illumina Inc). The methylation level at each CG site was measured by a β value ranging from 0 (unmethylated) to 1 (fully methylated).

Neuropathologic Evaluations

At autopsy, systematic evaluations assessed burdens of common age-related neuropathologic conditions, including AD, macroscopic infarcts, microinfarcts, Lewy bodies, hippocampal sclerosis, cerebral amyloid angiopathy, TDP-43, atherosclerosis, and arteriolosclerosis. Details on the quantification of these pathologic conditions are provided in the eAppendix in the Supplement.

Longitudinal Cognitive Assessments

Uniform cognitive assessments were administered each year at baseline and at follow-up visits. The cognitive testing battery included 17 cognitive tests. Scores on each test were standardized using the baseline mean and SD of the entire cohort. A composite score of global cognition was obtained by averaging individual standardized scores across the tests.16 Higher score indicates better cognition.

Statistical Analysis

Bivariate associations between the R2, neuropathologic, and cognitive measures were examined using pairwise correlations, 2-sided t tests, or analysis of variance, as appropriate. Multivariable linear regression models were used to examine gene expression. In these models, the composite R2 measure was the continuous outcome, and individual gene expression was the predictor. The corresponding regression coefficient estimates the difference in standardized R2 value with every 1-SD increase in gene expression level. Genome-wide statistical significance for gene expression analysis was determined at an α level of 2 × 10−6 (ie, Bonferroni correction for approximately 23 000 genes targeted).

Similar linear regression models were used to interrogate the association of DNA methylation with R2. In these models, the R2 measure was the continuous outcome, and the methylation level of individual CG was the predictor. The regression coefficient estimates the difference in standardized R2 value with every 1-SD increase in methylation. Genome-wide statistical significance for DNA methylation was determined at an α level of 1 × 10−7 (ie, Bonferroni correction for approximately 500 000 CGs targeted). For both analyses, the models were adjusted for age, sex, postmortem interval to imaging, and other technical confounders, as well as neuropathologic indices for AD, macroscopic infarcts, microinfarcts, Lewy bodies, hippocampal sclerosis, TDP-43, cerebral amyloid angiopathy, atherosclerosis, and arteriolosclerosis. All analyses were conducted using SAS/STAT software, version 9.3 (SAS Institute Inc), and R software, version 3.3.1 (R Foundation for Statistical Computing). Unless otherwise noted, P < .05 was considered significant.

Results
Common Neuropathologic Conditions, Cognitive Decline, and R2

At autopsy, 376 of the study participants (68.1%) had a pathologic AD diagnosis, 190 (34.4%) had macroscopic infarcts, and 171 (31.0%) had microinfarcts (Table 1). Lewy bodies were present in 127 of the participants (23.0%), hippocampal sclerosis in 65 (11.8%), and TDP-43 in 312 (56.5%). Other neuropathologic conditions are described in Table 1. The mean (SD) Mini-Mental State Examination score of these participants decreased from 27.6 (2.8) at baseline to 20.3 (9.3) proximate to death, consistent with an overall decline in cognition; the annual mean (SE) rate of change in global cognition, as measured by annual cognitive assessments spanning a mean of 9 years, was –0.100 (0.004) standard units (P < .001).

The bivariate association between the R2 measure and neuropathologic indices was consistent with a previous report.17 As shown in eFigure 4 in the Supplement, slower R2 was correlated with higher burden of AD pathology. Individuals with macroscopic infarcts or hippocampal sclerosis had slower R2. Slower R2 was also associated with more severe TDP-43, atherosclerosis, and arteriolosclerosis. On the other hand, we did not observe a strong association between R2 and microinfarcts, Lewy body, or cerebral amyloid angiopathy. In a regression analysis with all the neuropathologic measures included in the same model, AD, macroscopic infarcts, hippocampal sclerosis, TDP-43, and arteriolosclerosis remained associated with slower R2 (eTable 1 in the Supplement). After adjustment for demographics and all 9 neuropathologic indices, slower R2 was strongly associated with faster decline in cognition (estimate [SE], 0.023 [0.004]; P < .001) (Table 2).

Brain Gene Expression and R2

We investigated the RNA sequencing data by regressing the R2 measure associated with cognitive decline on individual gene expression. Four gene expression signals (ie, PADI2 [Ensembl ENSG00000117115], ZNF385A [Ensembl ENSG00000161642], PSD2 [Ensembl ENSG00000146005], and A2ML1 [Ensembl ENSG00000166535]) reached genome-wide significance. In all cases, higher expression of these genes was associated with slower R2 (Table 3). Partial residuals plots support a robust negative association between R2 and individual gene expression (eFigure 5 in the Supplement). The lead gene is PADI2, which encodes a peptidyl arginine deiminase enzyme that promotes posttranslational protein citrullination. The overexpression of PADI2 has been implicated in neurodegeneration and demyelination.18,19 The other 3 genes are lesser known. ZNF385A encodes a regulatory zinc finger protein that functions as a transcription factor, binds RNA, or interacts with other proteins.20PSD2 is involved in phospholipid binding based on gene ontology. A2ML1 encodes an N-glycosylated monomeric protein that acts as a protease inhibitor. These 4 genes were highly coexpressed (eFigure 6 in the Supplement), such that the pairwise Pearson correlation coefficients ranged from 0.56 to 0.74 (P < .001 for all correlations). Additional top hits are shown in eTable 2 in the Supplement.

Brain DNA Methylation and R2

We investigated genome-wide DNA methylation data for differentially methylated CGs in relation to the R2 measure associated with cognitive decline. A summary of the results is illustrated in eFigure 7 in the Supplement. None of the CGs survived the Bonferroni correction for multiple testing. A total of 6 CGs (eTable 3 in the Supplement) reached a suggestive significance level of α = 1 × 10−5. Methylation in all 6 CGs was anticorrelated with the R2 measure such that higher methylation was associated with slower R2. Furthermore, the anticorrelation pattern was evident in many of the CGs investigated and was robust against different choices of statistical cutoffs. The results from proportions (χ2) tests suggest that the number of CGs anticorrelated with R2 was highly enriched regardless of the cutoffs. To illustrate, Figure 2A shows that, of the 1221 CGs that reached the significance level of α = 0.001, 979 (80.2%) were anticorrelated (P < .001). A similar pattern was observed in gene expression (Figure 2B).

R2 Association With Cognitive Decline Attenuated by Gene Expression Signals

We assessed the extent to which R2 association with cognitive decline was attenuated by the top gene expression signals. To do so, we fit a series of linear mixed models with annual global cognitive scores as the continuous longitudinal outcome. The reference model included only demographics as the predictors; the second model (eTable 4 in the Supplement; model A) included demographics and the neuropathologic indices. The reduction of the variance estimates for random slope between the 2 models captures the variability of person-specific rates of cognitive decline explained by common neuropathologic conditions. In this sample, neuropathologic conditions explained 43% of the variance, and the result is consistent with a previous report for the overall cohorts.6 The third model included demographics, neuropathologic indices, and the composite R2 measure (eTable 4 in the Supplement; model B), in which R2 explained an additional 6% of the variance. Next, we augmented the model by including gene expression signals identified from the RNA sequencing (eTable 4 in the Supplement; model C). The mean expression averaged across the 4 genes was used, considering the strong coexpression. With the gene expression signal added to the model, we observed an attenuation of R2 association with cognitive decline such that the mean (SE) coefficient was reduced from 0.028 (0.008) (P < .001) to 0.019 (0.009) (P = .03). Furthermore, after accounting for the gene expression, the percent variance explained by R2 decreased to 1%.

Since the presence of the apolipoprotein E (APOE [Ensembl ENSG00000130203]) ε4 is strongly associated with cognitive decline and neuropathologic conditions, we conducted secondary analyses to examine whether our results were affected by the presence of APOE ε4. Consistent with findings in prior reports,21,22 the phenotypic associations of ε4 were diminished after adjusting for AD and other neuropathologic conditions. Moreover, the primary results for R2 association with cognitive decline were unchanged after the models were controlled for APOE ε4 (eTable 5 and 6 in the Supplement).

Discussion

Much of the person-to-person variation in late-life cognitive decline is not due to AD or other common neuropathologic conditions.6 Identification of additional factors associated with cognitive decline may lead to novel therapeutic targets to prevent loss of cognition in old age. Alteration of R2 measured using ex vivo MRI captures change in brain tissue integrity, which provides a unique window into the neurobiology of cognitive decline.11 We previously showed that the association of R2 with cognitive decline persisted after controlling for multiple neuropathologic indices, suggesting that slowing R2 represents a separate determinant of cognitive decline.11 In this study, we explored genome-wide transcriptomic and epigenetic correlates that underlie this association. New findings and implications are discussed.

First, multiple genome-wide significant gene signals are identified. The top gene, PADI2, is of particular relevance. PADI2 encodes a posttranslational modification enzyme that catalyzes citrullination, a process that converts arginine residues into citrullines in proteins. Citrullination in myelin basic protein in particular destabilizes myelin structure. A transgenic mouse model shows that overexpressed PADI2 leads to demyelination in the central nervous system.23 Thus, the gene plays a key role in demyelinating diseases, such as multiple sclerosis,24 for which T2 (reciprocal of R2) lesions are a radiologic hallmark.25 Evidence shows that, in brains of individuals with multiple sclerosis, PADI2 expression was elevated and the PADI2 promoter was hypomethylated.19,26 We targeted CG sites in the PADI2 locus but did not find a significant association with R2 using an omnibus test.27 Upregulations of PADI2 expression and citrullinated proteins have also been implicated in neurodegenerative diseases, including prion diseases and AD.18,28,29 Using Western blot analyses, a prior study detected citrullinated proteins in hippocampal tissues from patients with AD but not healthy controls, and the amount of PAD2 was significantly higher in ADs.18 The other 3 genes, namely ZNF385A, PSD2, and A2ML1, are strongly coexpressed with PADI2; however, their biological connections with R2 alteration are unclear.

Second, the investigation of individual CGs did not yield a genome-wide significant signal for R2 associated with cognitive decline. One contributing factor is that, compared with gene expression, epigenomic markers, such as DNA methylation, are further away from the phenotype along the biological causal chain. Consequently, the association is likely weaker and requires a larger sample size to detect. Rather than targeting site-specific signals, an alternative approach focusing on methylation module networks helps to reduce the burden for multiple testing and to reveal a clearer picture of the role of DNA methylation. These works are currently ongoing. We observed that higher DNA methylation tends to be associated with slower R2; the direction of this association was consistent across a massive number of CGs. This pattern suggests that the association between DNA methylation and R2 may not be loci specific. A popular hypothesis for the independent association of R2 with late-life cognitive decline is that a prolonged brain transverse relaxation time represents brain pathologies not captured by our current neuropathologic indices. Considering that R2 alteration is a manifestation of tissue damage due to pathologic conditions, global DNA hypermethylation may be a biological consequence of such an insult. Similar findings have been previously reported with respect to chronological age,30,31 in which brain DNA methylation is elevated with aging and the association tends to be global.

Finally, the R2 association with cognitive decline is attenuated by the lead gene expression signals. This finding lends support to the idea that dysregulation of genes, including PADI2, is associated with late-life cognitive impairment above and beyond the influence of common neuropathologic conditions. If confirmed, our findings could provide potential gene targets for neuropathologic conditions that cause R2 alteration and, in turn, cognitive impairment and AD dementia. Earlier evidence from the multiple sclerosis mouse model demonstrated that protein hypercitrullination is reversible and disease progression is attenuated by inhibiting the peptidyl-arginine deiminase enzymes.32

Transverse relaxation properties can be measured in vivo and ultimately may serve as a biomarker in research on aging. A pilot study compared in vivo and ex vivo T2 maps using data from 28 individuals.17 The results showed strong linear correlations between the 2 sets of values across multiple brain regions, suggesting that ex vivo findings in this study may be generalizable to in vivo cases. Furthermore, while ex vivo R2 is the most readily available metric in our current data pool that can be matched with autopsied cases from which brain RNA sequencing, DNA methylation, and neuropathologic data are also available, other imaging markers might account for additional variation in cognitive decline above and beyond that explained by neuropathologic findings. In vivo MRI measures will eventually be available for a greater number of autopsied cases, but at the moment, these numbers are fewer than 100. Using preliminary data, we observed positive correlations between R2 and in vivo hippocampal volume and entorhinal cortex thickness. Other postmortem MRI modalities and metrics, such as R2*, diffusion imaging, and volumetrics, have not yet been developed as extensively as R2, although work is also ongoing to expand into these areas. Thus, as sufficient data accrue, we will be able to examine the association between various imaging markers, neuropathologic conditions, genomics, and cognitive decline.

Limitations

This study has some limitations. We applied the voxelwise analyses to empirically identify the regions of interest for the R2 measure. Because the genomic data are generated using DLPFC tissue, the current approach lacks the intuitive appeal of spatially matching with the actual tissue collection site. Future studies on DLPFC-specific R2 metrics are warranted to better understand the association of ex vivo brain integrity with brain genomics. Our DNA methylation data, generated using the Illumina 450 assay, preferentially target regions with a high frequency of CG sites, and the assay probes only a small fraction of the epigenome. In addition, the assay does not differentiate between cytosine methylation and hydroxymethylation. Furthermore, both the Religious Orders Study and the Rush Memory and Aging Project are voluntary cohorts, and the participants are generally older and have higher education levels. Our findings await replications from other clinicopathologic studies.

Conclusions

Taken together, these results suggest that both DNA methylation and gene expression dysregulations are involved in R2 alteration in late-life cognitive impairment. To our knowledge, this is the first study to investigate the correlates of genome-wide gene expression and DNA methylation with transverse relaxation in postmortem brains. Ex vivo imaging techniques provide novel metrics for brain structure and tissue integrity that are complementary to traditional histopathologic indices. By leveraging both transcriptomic and epigenomic data measured from the same participants in whom we have detailed clinical, neuropathologic, and imaging data, this study offers further insight into the genetic footprints that underlie the association of brain tissue properties with cognitive impairment and dementia.

Back to top
Article Information

Corresponding Author: David A. Bennett, MD, Rush Alzheimer’s Disease Center, Rush University Medical Center, 1750 W Harrison, Ste 1000, Chicago, IL 60612 (david_a_bennett@rush.edu).

Accepted for Publication: July 24, 2017.

Published Online: October 30, 2017. doi:10.1001/jamaneurol.2017.2807

Author Contributions: Dr Yu had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Yu, Dawe, Bennett.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Yu.

Critical revision of the manuscript for important intellectual content: Dawe, Boyle, Gaiteri, Yang, Buchman, Schneider, Arfanakis, De Jager, Bennett.

Statistical analysis: Yu, Dawe, Gaiteri, Yang.

Obtained funding: Boyle, Buchman, Arfanakis, Bennett.

Administrative, technical, or material support: Dawe, Buchman, Schneider, Arfanakis, Bennett.

Study supervision: Arfanakis, Bennett.

Conflict of Interest Disclosures: None reported.

Funding/Support: This study was supported by grants P30AG10161, R01AG15819, R01AG17917, R01AG36042, R01AG34374, and U01AG46152 from the National Institute on Aging.

Role of the Funder/Sponsor: The funding source had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, and approval of the manuscript; or decision to submit the manuscript for publication.

Additional Contributions: The study is not possible without contributions of all the participants in the Religious Orders Study and the Rush Memory and Aging Project. We also thank the investigators and staff at the Rush Alzheimer’s Disease Center and the De Jager Laboratory of the Broad Institute of Harvard University and the Massachusetts Institute of Technology.

Additional Information: To obtain data from the Religious Orders Study and the Rush Memory and Aging Project for research use, please visit the Rush Alzheimer’s Disease Center Research Resource Sharing Hub (http://www.radc.rush.edu).

References
1.
Hebert  LE, Weuve  J, Scherr  PA, Evans  DA.  Alzheimer disease in the United States (2010-2050) estimated using the 2010 census.  Neurology. 2013;80(19):1778-1783.PubMedGoogle ScholarCrossref
2.
Boyle  PA, Yang  J, Yu  L,  et al.  Varied effects of age-related neuropathologies on the trajectory of late life cognitive decline.  Brain. 2017;140(3):804-812.PubMedGoogle Scholar
3.
White  LR, Edland  SD, Hemmy  LS,  et al.  Neuropathologic comorbidity and cognitive impairment in the Nun and Honolulu-Asia Aging Studies.  Neurology. 2016;86(11):1000-1008.PubMedGoogle ScholarCrossref
4.
Kryscio  RJ, Abner  EL, Nelson  PT,  et al.  The effect of vascular neuropathology on late-life cognition: results from the SMART project.  J Prev Alzheimers Dis. 2016;3(2):85-91.PubMedGoogle Scholar
5.
Troncoso  JC, Zonderman  AB, Resnick  SM, Crain  B, Pletnikova  O, O’Brien  RJ.  Effect of infarcts on dementia in the Baltimore Longitudinal Study of Aging.  Ann Neurol. 2008;64(2):168-176.PubMedGoogle ScholarCrossref
6.
Boyle  PA, Wilson  RS, Yu  L,  et al.  Much of late life cognitive decline is not due to common neurodegenerative pathologies.  Ann Neurol. 2013;74(3):478-489.PubMedGoogle ScholarCrossref
7.
Bauer  CM, Jara  H, Killiany  R; Alzheimer’s Disease Neuroimaging Initiative.  Whole brain quantitative T2 MRI across multiple scanners with dual echo FSE: applications to AD, MCI, and normal aging.  Neuroimage. 2010;52(2):508-514.PubMedGoogle ScholarCrossref
8.
House  MJ, St Pierre  TG, Foster  JK, Martins  RN, Clarnette  R.  Quantitative MR imaging R2 relaxometry in elderly participants reporting memory loss.  AJNR Am J Neuroradiol. 2006;27(2):430-439.PubMedGoogle Scholar
9.
Granziera  C, Daducci  A, Donati  A,  et al.  A multi-contrast MRI study of microstructural brain damage in patients with mild cognitive impairment.  Neuroimage Clin. 2015;8:631-639.PubMedGoogle ScholarCrossref
10.
Arfanakis  K, Gui  M, Tamhane  AA, Carew  JD.  Investigating the medial temporal lobe in Alzheimer’s disease and mild cognitive impairment, with turboprop diffusion tensor imaging, MRI-volumetry, and T2-relaxometry.  Brain Imaging Behav. 2007;1(1-2):11-21. doi:10.1007/s11682-007-9001-4Google ScholarCrossref
11.
Dawe  RJ, Yu  L, Leurgans  SE,  et al.  Postmortem MRI: a novel window into the neurobiology of late life cognitive decline.  Neurobiol Aging. 2016;45:169-177.PubMedGoogle ScholarCrossref
12.
Bennett  DA, Schneider  JA, Arvanitakis  Z, Wilson  RS.  Overview and findings from the Religious Orders Study.  Curr Alzheimer Res. 2012;9(6):628-645.PubMedGoogle ScholarCrossref
13.
Bennett  DA, Schneider  JA, Buchman  AS, Barnes  LL, Boyle  PA, Wilson  RS.  Overview and findings from the Rush Memory and Aging Project.  Curr Alzheimer Res. 2012;9(6):646-663.PubMedGoogle ScholarCrossref
14.
Dawe  RJ, Bennett  DA, Schneider  JA, Vasireddi  SK, Arfanakis  K.  Postmortem MRI of human brain hemispheres: T2 relaxation times during formaldehyde fixation.  Magn Reson Med. 2009;61(4):810-818.PubMedGoogle ScholarCrossref
15.
Dawe  RJ, Bennett  DA, Schneider  JA, Arfanakis  K.  Neuropathologic correlates of hippocampal atrophy in the elderly: a clinical, pathologic, postmortem MRI study.  PLoS One. 2011;6(10):e26286.PubMedGoogle ScholarCrossref
16.
Wilson  RS, Boyle  PA, Yu  L, Segawa  E, Sytsma  J, Bennett  DA.  Conscientiousness, dementia related pathology, and trajectories of cognitive aging.  Psychol Aging. 2015;30(1):74-82.PubMedGoogle ScholarCrossref
17.
Dawe  RJ, Bennett  DA, Schneider  JA,  et al.  Ex vivo T2 relaxation: associations with age-related neuropathology and cognition.  Neurobiol Aging. 2014;35(7):1549-1561.PubMedGoogle ScholarCrossref
18.
Ishigami  A, Ohsawa  T, Hiratsuka  M,  et al.  Abnormal accumulation of citrullinated proteins catalyzed by peptidylarginine deiminase in hippocampal extracts from patients with Alzheimer’s disease.  J Neurosci Res. 2005;80(1):120-128.PubMedGoogle ScholarCrossref
19.
Mastronardi  FG, Noor  A, Wood  DD, Paton  T, Moscarello  MA.  Peptidyl argininedeiminase 2 CpG island in multiple sclerosis white matter is hypomethylated.  J Neurosci Res. 2007;85(9):2006-2016.PubMedGoogle ScholarCrossref
20.
Sharma  S, Dimasi  D, Higginson  K, Della  NG.  RZF, a zinc-finger protein in the photoreceptors of human retina.  Gene. 2004;342(2):219-229.PubMedGoogle ScholarCrossref
21.
Yu  L, Boyle  P, Schneider  JA,  et al.  APOE ε4, Alzheimer’s disease pathology, cerebrovascular disease, and cognitive change over the years prior to death.  Psychol Aging. 2013;28(4):1015-1023.PubMedGoogle ScholarCrossref
22.
Yu  L, Boyle  PA, Leurgans  S, Schneider  JA, Bennett  DA.  Disentangling the effects of age and APOE on neuropathology and late life cognitive decline.  Neurobiol Aging. 2014;35(4):819-826.PubMedGoogle ScholarCrossref
23.
Musse  AA, Li  Z, Ackerley  CA,  et al.  Peptidylarginine deiminase 2 (PAD2) overexpression in transgenic mice leads to myelin loss in the central nervous system.  Dis Model Mech. 2008;1(4-5):229-240.PubMedGoogle ScholarCrossref
24.
Moscarello  MA, Pritzker  L, Mastronardi  FG, Wood  DD.  Peptidylarginine deiminase: a candidate factor in demyelinating disease.  J Neurochem. 2002;81(2):335-343.PubMedGoogle ScholarCrossref
25.
Rovira  À, Wattjes  MP, Tintoré  M,  et al; MAGNIMS Study Group.  Evidence-based guidelines: MAGNIMS consensus guidelines on the use of MRI in multiple sclerosis—clinical implementation in the diagnostic process.  Nat Rev Neurol. 2015;11(8):471-482.PubMedGoogle ScholarCrossref
26.
Calabrese  R, Zampieri  M, Mechelli  R,  et al.  Methylation-dependent PAD2 upregulation in multiple sclerosis peripheral blood.  Mult Scler. 2012;18(3):299-304.PubMedGoogle ScholarCrossref
27.
Yu  L, Chibnik  LB, Srivastava  GP,  et al.  Association of brain DNA methylation in SORL1, ABCA7, HLA-DRB5, SLC24A4, and BIN1 with pathological diagnosis of Alzheimer disease.  JAMA Neurol. 2015;72(1):15-24.PubMedGoogle ScholarCrossref
28.
Chen  S, Lu  FF, Seeman  P, Liu  F.  Quantitative proteomic analysis of human substantia nigra in Alzheimer’s disease, Huntington’s disease and multiple sclerosis.  Neurochem Res. 2012;37(12):2805-2813.PubMedGoogle ScholarCrossref
29.
Jang  B, Ishigami  A, Maruyama  N, Carp  RI, Kim  YS, Choi  EK.  Peptidylarginine deiminase and protein citrullination in prion diseases: strong evidence of neurodegeneration.  Prion. 2013;7(1):42-46.PubMedGoogle ScholarCrossref
30.
Hernandez  DG, Nalls  MA, Gibbs  JR,  et al.  Distinct DNA methylation changes highly correlated with chronological age in the human brain.  Hum Mol Genet. 2011;20(6):1164-1172.PubMedGoogle ScholarCrossref
31.
Yang  J, Yu  L, Gaiteri  C,  et al.  Association of DNA methylation in the brain with age in older persons is confounded by common neuropathologies.  Int J Biochem Cell Biol. 2015;67:58-64.PubMedGoogle ScholarCrossref
32.
Moscarello  MA, Lei  H, Mastronardi  FG,  et al.  Inhibition of peptidyl-arginine deiminases reverses protein-hypercitrullination and disease in mouse models of multiple sclerosis.  Dis Model Mech. 2013;6(2):467-478.PubMedGoogle ScholarCrossref
×