Postmortem Brain MRI Is Related to Cognitive Decline, Independent of Cerebral Vessel Disease in Older Adults

Abstract

The purpose of this study was to determine whether metrics of brain tissue integrity derived from postmortem MRI are associated with late life cognitive decline, independent of cerebral vessel disease. Using data from 554 older adults, we employed voxelwise regression to identify regions where the postmortem MRI transverse relaxation rate constant R₂ was associated with the rate of decline in global cognition. We then used linear mixed models to investigate the association between a composite R₂ measure and cognitive decline, controlling for neuropathology including three indices of vessel disease: atherosclerosis, arteriolosclerosis, and cerebral amyloid angiopathy. This composite R₂ measure was associated with the rate of decline (0.049 unit annually per R₂ unit, p<0.0001) and accounted for 6.1% of its variance, beyond contributions from vessel disease indices and other prominent age-related neuropathologies. Thus, postmortem brain R₂ reflects disease processes underlying cognitive decline that are not captured by vessel disease indices or other standard neuropathologic indices and may provide a measure of brain tissue integrity that is complementary to histopathologic evaluation.

1. INTRODUCTION

Cognitive decline, one of the most common and feared experiences of late life (Cutler, 2002), is a costly phenomenon that will increasingly strain healthcare resources as the population ages (Hayden et al., 2011, Zhu et al., 2013). Despite being the focus of intense research efforts, gaps in our understanding of the underlying neurobiology of late life cognitive decline impede the development of effective interventions to prevent or delay its onset. In particular, although common age-related neuropathologies are important determinants of cognitive impairment and decline, they account for less than half of the heterogeneity observed in rates of decline (Boyle et al., 2013), suggesting that other factors are involved. This highlights a pressing need to identify novel factors that influence late life cognitive decline.

MRI is a powerful tool that may detect important brain changes related to cognitive decline. Recently, we reported that a metric of tissue integrity derived from postmortem MRI, the transverse relaxation rate constant R₂ (inverse of T₂), accounted for approximately 10% of the variation in the rate of late life cognitive decline, beyond the contributions of neuropathologic indices of the three most common causes of dementia – Alzheimer’s disease (AD), cerebrovascular disease (CVD), and Lewy body disease (LBD) (Dawe et al., 2016). While the specific mechanisms linking lower R₂ values to faster cognitive decline are unknown, one plausible hypothesis is that hypoxic-ischemic events are particularly injurious to myelinating oligodendrocytes and their progenitors (Back et al., 2002, Pantoni et al., 1996), and that demyelination stemming from loss of these cells impairs cognition by disrupting communication within various brain networks (Felts et al., 1997, Pievani et al., 2011). This type of damage is likely accessible to postmortem R₂ imaging because the tissue’s R₂ value depends largely on the ratio of free water molecules, which exhibit a slower transverse relaxation (smaller R₂ value), to those that are trapped between myelin layers and therefore exhibit a faster transverse relaxation (larger R₂ value) (Laule et al., 2006). Thus, brain tissue whose free water content increases due to some form of degeneration is expected to have a smaller R₂ imaging metric. Although previous analyses examining the relation of R₂ with cognitive decline controlled for micro- and gross infarcts, other vascular pathologies may also result in ischemic conditions leading to demyelination or axonal loss. In particular, pathologies underlying cerebral vessel disease, namely atherosclerosis, arteriolosclerosis, and cerebral amyloid angiopathy (CAA), are recognized as prominent contributors to hypoxic-ischemia (Bijanki et al., 2013, Chung et al., 2009, Gurol et al., 2013, Kalback et al., 2004, Román et al., 2002).

In the current study, we extended our previous work on the association between postmortem R₂ and late life cognitive decline by examining whether this association is independent of cerebral vessel disease. Further, because cognitive decline tends to accelerate in the years just prior to death (i.e., terminal decline), we also investigated whether R₂ is primarily associated with the relatively early, gradual decline in cognition or the more rapid terminal decline phase (Wilson et al., 2010). Participants were 554 individuals from the Rush Memory and Aging Project (MAP) and the Religious Orders Study (ROS) who underwent at least two annual cognitive evaluations. Upon death, one cerebral hemisphere from each individual underwent MRI and neuropathologic evaluation. Using voxelwise regression analyses, we first delineated the brain regions where R₂ exhibited significant association with the rate of cognitive decline. Using linear mixed models, we quantified the degree to which R₂ in these regions accounted for the variance in cognitive decline after adjustment for three hallmark cerebral vessel diseases: atherosclerosis, arteriolosclerosis, and CAA. Finally, in a subset of participants who had at least five cognitive evaluations, we employed random changepoint models to examine specific components of the cognitive trajectory (i.e., preterminal and terminal slopes), to elucidate the temporal associations of cognitive decline with R₂.

2. METHODS

2.1. Participants and Specimens

Participants were individuals enrolled in two longitudinal clinical pathologic studies of aging and dementia, MAP and ROS (Bennett et al., 2012a, Bennett et al., 2012b), with harmonized cognitive, neuropathologic, and postmortem MRI data collection procedures. The Institutional Review Board of Rush University Medical Center approved these studies, which are carried out in accordance with the Declaration of Helsinki. At enrollment, participants are free of known dementia, sign an anatomical gift act, and agree to brain donation after death. At the time of analyses, a total of 3202 individuals had been enrolled (1858 MAP, 1344 ROS), 1535 were deceased, and 1327 had undergone autopsy. Since 2006 when postmortem MRI began, 610 autopsied brain specimens had undergone postmortem R₂ imaging and had passed quality control and post-processing. Of these decedents, 554 had undergone at least two cognitive evaluations during life, forming the sample of the current work.

2.2. Cognitive Evaluation and Clinical Diagnosis

Participants underwent an annual battery of tests to assess cognitive abilities, including the MMSE (used for descriptive purposes only) and 17 other tests that are common between the two studies: Logical Memory I for immediate recall, Logical Memory II for delayed recall, East Boston immediate recall, East Boston delayed recall, word list memory, word list recall, word list recognition, Boston naming, category fluency, National Adult Reading, digit span forward, digit span backward, digit ordering, Symbol Digits Modalities, number comparison, judgment of line orientation, and progressive matrices tests. The individual scores for the 17 tests were converted to z-scores based on the mean and standard deviation of the combined MAP-ROS cohort at baseline, which were then averaged to form a composite score for global cognition. A clinical diagnosis of AD or MCI was rendered by a clinician at every assessment based on these cognitive test scores and clinical judgment by a neuropsychologist. At death, a neurologist with expertise in dementia reviewed all available clinical data (but remained blinded to all postmortem data) and rendered a summary diagnostic opinion of the most likely clinical diagnosis at the time of death.

2.3. Postmortem MRI and R₂ Image Processing

Postmortem MRI procedures have been described previously (Dawe et al., 2014). Upon death of a participant, the brain was removed and hemisected during rapid autopsy. The cerebral hemisphere with more visible pathology was immersed in 4% paraformaldehyde solution and refrigerated at 4 degrees Celsius. At approximately one month postmortem, this hemisphere was imaged using a 3-Tesla MRI scanner. R₂ maps were generated from the fast spin echo images using in-house software to fit an exponential decay of the form S=S₀·exp(−R₂·TE) to the data in each voxel, where S is the magnitude of the MRI signal measured at known echo times of TE. R₂ maps were spatially registered to a cerebral hemisphere template, first using linear and then nonlinear transformations as previously described (Dawe et al., 2016). To minimize differences in R₂ stemming from the use of four different MRI scanners throughout the postmortem MRI data collection, we carried out a voxel-by-voxel normalization of R₂ values within each group of hemispheres imaged on a given scanner by subtracting the median and dividing by the interquartile range of R₂ for that scanner, to guard against undue influence from outliers.

2.4. Neuropathologic Indices

The imaged hemisphere was then sliced into 1-cm thick coronal slabs, and neuropathologic indices were derived for AD, CVD, LBD, hippocampal sclerosis (HS), and TDP-43 as previously described (Boyle et al., 2013, Nag et al., 2015, Wilson et al., 2015) and summarized here. AD: Amyloid load and the density of paired helical filament tau tangles were quantified using antibody-specific immunohistochemistry. We computed summary measures for each by averaging the observed values across seven cortical regions and the hippocampus and taking the square root to account for skewness in the distributions. CVD: Chronic gross infarcts were coded as present or absent. We did not include microinfarcts because previous work demonstrated no association of microinfarcts with cognitive decline. LBD: Lewy bodies were coded as present if they were identified in any of seven brain regions using a monoclonal antibody to alpha-synuclein. HS: Typical hippocampal sclerosis was coded as present or absent based on the observation of severe neuronal loss and gliosis in CA1 or the subiculum. TDP-43: A rat phosphorylated monoclonal TAR5P-1D3 TDP-43 antibody (pS409/410, 1:100, Ascenion, Munich, Germany) applied to six brain regions provided a four-level staging of the extent of TDP-43 pathology: a value of 0 for no TDP-43 or stages of 1 to 3 corresponding to TDP-43 present only in the amygdala, also extending to other limbic structures, and in neocortical regions as well.

Indices of cerebral large and small vessel disease were derived as previously described. Atherosclerosis was rated on a semiquantitative scale based on the degree of occlusion of arteries in the Circle of Willis, with values 0-3 corresponding to none, mild, moderate, or severe (Arvanitakis et al., 2016). Similarly, arteriolosclerosis was graded on a semiquantitative scale with values 0-3 corresponding to no, mild, moderate, or severe occlusion arterioles of the anterior basal ganglia (Arvanitakis et al., 2016). For CAA, meningeal and parenchymal vessels in four neocortical regions were graded for amyloid deposition on a semiquantitative scale, which were then averaged and converted to a summary measure with values 0-3 corresponding to no, mild, moderate, and severe CAA (Boyle et al., 2015).

2.5. Statistical Analyses

We first conducted a voxelwise regression analysis in which the R₂ value for each voxel served as the explanatory term of interest and the person-specific slope of global cognitive decline estimated from linear mixed models was the outcome. This voxelwise analysis controlled for demographics and all available neuropathologic indices, namely AD pathology, CVD, LBD, HS, TDP-43, atherosclerosis, arteriolosclerosis, and CAA. Correction for multiple testing (approximately 400,000 tissue-containing voxels) was accomplished by using a false discovery rate (FDR) of 0.05 and also performing spatial clustering (minimum cluster size of 100 voxels), as previously described (Dawe et al., 2016). The mean R₂ value for each of the resultant clusters was computed for all brain specimens, and their correlations were examined using Pearson correlation and principal component analysis before combining them into a composite R₂ measure for use in subsequent analyses.

We then carried out a series of linear mixed models implemented in SAS 9.4 (SAS Institute, Inc.) to assess the contribution of the composite R₂ measure to the variance in the rate of global cognitive decline when controlling for different types of neuropathology. Scores of global cognition from all available assessments served as the longitudinal outcome in these models. The reference model included only an intercept and a term for time in years prior to death, which estimated the mean rate of decline (i.e., slope). We sequentially added explanatory terms for demographics, the three common causes of dementia (AD, CVD, and LBD) and two other neuropathologies (HS and TDP-43), and finally the three indices of cerebral vessel disease (atherosclerosis, arteriolosclerosis, and CAA). We conducted these models first without and then with the composite R₂ measure. For each model, we computed the reduction in variance relative to a previous model, thereby quantifying the contribution of the most recently entered explanatory terms to the variation in rate of cognitive decline.

Random changepoint models capture the nonlinear trajectory of cognitive decline and facilitate testing for differential associations of R₂ with decline during preterminal and terminal phases (Wilson et al., 2010). In these models, the cognitive trajectory is modeled using four parameters: a first slope, which represents the rate of cognitive decline during the preterminal phase; the changepoint, which represents the time when the rate of cognitive decline accelerates; the second slope, which represents the rate of decline during the terminal phase; and the intercept, which represents the level of cognitive function proximate to death. To fit changepoint models, we required that participants have at least five cognitive evaluations with the last evaluation occurring within two years of death, narrowing the sample to 419 individuals. We included demographic terms, the nine available neuropathologic indices (amyloid, tangles, CVD, LBD, HS, TDP-43, atherosclerosis, arteriolosclerosis, and CAA), and the composite R₂ measure as predictors, with global cognition as the outcome. The changepoint models were implemented in OpenBugs software using a Bayesian Monte Carlo Markov Chain approach.

3. RESULTS

3.1. Descriptive Data

Demographic and clinical characteristics of the sample (n=554) are shown in Table 1. Participants were on average 90.4 years of age (SD=6.0, range=65-108) at the time of death and had completed 9.3 annual cognitive exams (SD=4.6, range=2-22). Based on neuropathologic evaluation, 379 individuals (68.4%) met the NIA-Reagan intermediate or high likelihood criteria for a pathologic diagnosis of AD, 189 (34.1%) had one or more gross infarcts (the index of CVD), Lewy bodies were present in 127 (22.9%), HS in 67 (12.1%), and some degree of TDP-43 in 314 (56.7%). Concerning the indices of cerebral vessel disease, moderate to severe atherosclerosis was noted in 152 individuals (27.4%), arteriolosclerosis in 146 (26.4%), and CAA in 195 (35.2%). Correlations among demographics and pathologies are shown in Table 2.

Table 1.

Demographic, Cognitive, and Pathologic Characteristics of Participants

Variable	Mean (SD) or N (%)
Total Participants	554 (100%)
MAP	286 (51.6%)
ROS	268 (48.4%)
Demographic
Age at Death (years)	90.4 (6.0)
Female	395 (71.3%)
Education (years)	15.8 (3.6)
White, Non-Hispanic	536 (96.9%)
Clinical Diagnosis
Cognitive Impairment at Baseline
Mild Cognitive Impairment (MCI)	151 (27.3%)
Alzheimer’s Disease (AD)	33 (6.0%)
Cognitive Impairment at Death
Mild Cognitive Impairment (MCI)	143 (26.3%)
Alzheimer’s Disease (AD)	230 (42.4%)
Cognitive
Number of Annual Cognitive Evaluations	9.3 (4.6)
Baseline MMSE (score out of 30)	27.7 (2.8)
Proximate to Death MMSE	20.5 (9.2)
Global Cognition, Baseline (composite of 17 z-scores)	−0.07 (0.59)
Global Cognition, Proximate to Death	−0.85 (1.11)
Global Cognition, Estimated Linear Rate of Change (per year)	−0.10 (0.10)
Pathologic
Postmortem Interval (hours)	8.7 (5.8)
Amyloid Load (square root transform)	2.04 (1.19)
Tangles (square root transform)	2.28 (2.12)
Gross Infarcts (present)	189 (34.1%)
Lewy Bodies (present)	127 (22.9%)
Hippocampal sclerosis (present)	67 (12.1%)
TDP-43 (stage 0-3)	1.11 (1.13)
Cerebral amyloid angiopathy (semiquantitative 0-3)	1.13 (1.05)
Atherosclerosis (semiquantitative 0-3)	1.11 (0.84)
Arteriolosclerosis (semiquantitative 0-3)	1.03 (0.86)

Table 2.

Correlations among Demographics and Pathologic Indices

	Amyloid	Tangles	CVD	LBD	HS	TDP-43	Athero	Arteriolo	CAA
Age	0.12c	0.21e	0.12c	−0.03	0.07	0.20e	0.16d	0.06	0.15d
Sex	−0.08a	−0.16e	0.01	0.04	−0.02	−0.10b	−0.02	−0.09b	−0.06
Education	−0.10b	−0.07a	−0.12c	−0.02	0.04	−0.01	−0.06	−0.08a	0.04
Amyloid		0.50e	0.05	0.09b	0.05	0.09b	0.01	0.02	0.33e
Tangles			0.04	0.10b	−0.01	0.24e	0.04	0.07a	0.34e
CVD				0.01	0.05	0.03	0.20e	0.13c	0.02
LBD					0.05	0.04	−0.02	0.00	0.03
HS						0.38e	0.02	0.03	0.03
TDP-43							0.05	0.05	0.10b
Athero								0.31e	0.02
Arteriolo									0.01

^ap<0.10,

^bp<0.05,

^cp<0.01,

^dp<0.001,

^ep<0.0001.

3.2. Heterogeneity in Rates of Cognitive Decline

We first employed linear mixed models to estimate the rate of decline in global cognition, which was 0.10 unit per year (SE = 0.004, p<0.0001), with substantial heterogeneity observed across individuals. The variance of person specific rates of decline was 0.0083 (p<0.0001).

3.3. Brain Regions in which R₂ is Associated with Rate of Cognitive Decline

Next we sought to identify specific brain regions in which R₂ was most strongly associated with cognitive decline, independent of known pathology. In voxelwise models controlling for pathologic indices, R₂ was associated with the rate of global cognitive decline in six regions located primarily within white matter, as illustrated in Figure 1 and summarized in Table 3. The regression coefficients of R₂ in these regions were all positive, meaning that persons exhibiting slower (smaller) R₂ values tended to decline more rapidly (a more negative slope of decline). The mean R₂ values of the regions exhibited moderate to strong correlation with each other (Table 3) and principal component analysis revealed a single dominant eigenvalue accounting for 59% of their total variance. Therefore, we averaged the R₂ values from all six regions into a single composite R₂ measure, in order to maximize parsimony of subsequent analyses.

Figure 1.

Sagittal, coronal, and axial views of clusters of voxels for which R₂ was associated with the estimated slope of decline in global cognitive ability. These images were generated by a voxelwise analysis controlling for demographics, AD pathology , CVD, LBD, HS, TDP-43, and the indices of atherosclerosis, arteriosclerosis, and CAA. The region identifiers A-F correspond to those in Table 3.

Table 3.

Summary of Brain Regions in which R₂ is Associated with Rate of Cognitive Decline

Region Identifier	Location	Volume (mm3)	Correlations with R2 of other Regions (Pearson’s r)
			B	C	D	E	F
Region A	Anterior Temporal Lobe WM	134	0.62	0.58	0.46	0.43	0.36
Region B	Hippocampus/Parahippocampal WM	191	-	0.52	0.44	0.38	0.32
Region C	Posterior Temporal Lobe WM	1,094		-	0.62	0.74	0.48
Region D	Majority of Frontal Lobe WM	13,243			-	0.73	0.46
Region E	Parietal/Occipital Lobe WM	3,613				-	0.50
Region F	Parietal Lobe GM	140					-

All correlations significant at p<0.0001.

3.4. Variance in Rate of Cognitive Decline Accounted for by Neuropathology and R₂

We conducted a series of linear mixed models to assess the variance in rate of cognitive decline accounted for by the composite R₂ measure (Table 4). In models that controlled for demographics, amyloid, tangles, CVD, LBD, HS, and TDP-43, the addition of terms for the composite R₂ measure led to an increase in the portion of the random slope variance accounted for by the model from 41.3% to 48.1% (Models 1a and 1b). Thus, R₂ explained 6.7% of the total variance in rate of cognitive decline, beyond the contributions from demographics and neuropathologic indices other than cerebral vessel diseases. After further adjustment for indices of atherosclerosis, arteriolosclerosis, and CAA, R₂ explained 6.1% of the total variance (Models 2a and 2b).

Table 4.

Contributions to Variance in Cognitive Decline from Demographics, Neuropathologic Indices, and R₂

Variables	Model Identifier
	1a	1b	2a	2b
Time	−0.015a (0.010)	−0.0041 (0.0099)	0.037d (0.011)	0.015 (0.011)
Age	0.0026e (0.0006)	0.0029e (0.0006)	0.0029e (0.0006)	0.0032e (0.0006)
Sex	0.011 (0.008)	0.0087 (0.0077)	0.010 (0.0080)	0.0082 (0.0077)
Education	0.0019a (0.0010)	0.0019b (0.0009)	0.0016 (0.0010)	0.0017a (0.0009)
Amyloid	−0.0042 (0.0035)	−0.0043 (0.0033)	−0.0047 (0.0035)	−0.0045 (0.0034)
Tangles	−0.036e (0.004)	−0.030e (0.003)	−0.035e (0.0035)	−0.030e (0.0035)
Gross Infarcts	−0.034e (0.008)	−0.025d (0.007)	−0.027d (0.008)	−0.019b (0.008)
Lewy Bodies	−0.020b (0.009)	−0.020b (0.008)	−0.021b (0.009)	−0.021b (0.008)
Hippocampal Sclerosis	−0.066e 0.012	−0.055e (0.012)	−0.067e (0.012)	−0.056e (0.012)
TDP-43	−0.0062a (0.0036)	−0.0040 (0.0035)	−0.0059 (0.0036)	−0.0039 (0.0035)
Atherosclerosis			−0.018e (0.005)	−0.017d (0.005)
Arteriolosclerosis			−0.0057 (0.0044)	−0.0030 (0.0043)
Cerebral Amyloid Angiopathy			−0.0004 (0.0036)	−0.0018 (0.0035)
Composite R2		0.051e (0.007)		0.049e (0.007)
% of Variance Accounted for by all Effects	41.3%	48.1%	43.7%	49.8%
% of Variance Accounted for only by Effects in Boldface	-	6.7%	2.4%	6.1%
Model Identifier for Comparison	-	1a	1a	2a

^ap<0.10,

^bp<0.05,

^cp<0.01,

^dp<0.001,

^ep<0.0001.

The variables listed in the first column represent model terms consisting of the interaction of that variable with time. Effect sizes for those terms are shown, along with their standard errors in parentheses.

3.5. Association of R₂ with preterminal and terminal phases of decline

Changepoint models provide an indication of the time-dependent relations of R₂ with decline in global cognition. In this analysis, which included data from 419 individuals who had at least five cognitive exams and adjusted for all available neuropathologies including indices of cerebral vessel diseases, the composite R₂ measure was associated with the slope of the preterminal phase of decline (0.012 cognitive unit per year per standard deviation of R₂, 95% credible interval = 0.005 to 0.020) and also with the location of the changepoint (0.74 year per standard deviation of R₂, 95% credible interval = 0.51 to 0.97), but we did not observe R₂ association with the terminal slope. Thus, an individual with a composite R₂ value at the 10^th percentile (i.e., a relatively low, or slow, R₂) would typically exhibit a 0.016-unit per year faster rate of preterminal decline and a transition to the terminal phase (i.e., changepoint) 0.98 year earlier than an individual with R₂ equal to the sample median (Fig. 2).

Figure 2.

Changepoint trajectories illustrating the association of the composite postmortem R₂ measure with a nonlinear model of cognitive decline. The solid line corresponds to the typical decline trajectory for an individual with R₂ and AD pathology equal to the sample medians. The dotted and dashed lines represent decline trajectories for individuals with R₂ at the 10^th and 90^th percentiles, respectively. The displayed trajectories also control for age, sex, education, amyloid, tangles, CVD, LBD, HS, TDP-43, atherosclerosis, arteriolosclerosis, and CAA.

4. DISCUSSION

We investigated the association of postmortem brain R₂ with late life cognitive decline after controlling for several common neuropathologic indices including three cerebral vessel diseases, namely atherosclerosis, arteriolosclerosis, and CAA. In voxelwise analyses of MR images from over 500 participants, quantitative R₂ was associated with the linear rate of cognitive decline in several white matter brain regions, the most extensive of which encompassed much of the frontal lobe white matter. A composite measure of R₂ formed from these regions accounted for approximately 6% of the variance in decline beyond that attributable to neuropathologic indices. The variance accounted for by R₂ remained similar when cerebral vessel disease indices were included in the analysis. Further, changepoint models showed that R₂ was related to preterminal changes in global cognitive function and the timing of the onset of terminal decline. These findings highlight the independent association of postmortem R₂ with late life cognitive decline, particularly in the preterminal phase of decline, and have implications for the clinical translation of results generated from postmortem R₂ studies.

Considerable evidence links different forms of vessel disease to multiple neuroimaging measures, such as white matter hyperintensities on T₂-weighted or fluid-attenuated inversion recovery (FLAIR) sequences (Erten-Lyons et al., 2013, Gurol et al., 2013, Holland et al., 2008, Moroni et al., 2016, Pico et al., 2002) and lower fractional anisotropy of white matter on diffusion tensor imaging (DTI) (Bijanki et al., 2013, Bos et al., 2012, Jolly et al., 2013, Salat et al., 2006). A related line of research links cerebral vessel disease indices to dementia and lower performance in certain cognitive domains in late life (Arvanitakis et al., 2011, Arvanitakis et al., 2016, Boyle et al., 2015, Greenberg et al., 2004, Reijmer et al., 2016, Smith, 2017, Thal et al., 2003, van Oijen et al., 2007). In addition, our own prior work showed a strong relationship between postmortem R₂ (the inverse of T₂) in white matter and cognitive decline (Dawe et al., 2016). A logical hypothesis unifying these bodies of research is that cerebral vessel diseases contribute to late life cognitive decline via tissue damage that manifests on MRI as alterations to quantitative R₂ (Abraham et al., 2016, Banerjee et al., 2016, Prins and Scheltens, 2015). According to this hypothesis, associations of R₂ with late life cognitive decline would be expected to weaken substantially after controlling for indices of vessel disease, which plausibly contribute to myelin degradation or axonal loss thought to be responsible for both a decrease in the tissue’s R₂ and cognitive decline. Instead, the portion of the variance in the rate of cognitive decline accounted for by R₂ was essentially unchanged (6.1% vs. 6.7%) when indices of atherosclerosis, arteriolosclerosis, and CAA were included in the analysis. The effect size of the R₂ composite also remained nearly constant (0.049 vs. 0.051 cognitive unit per year per standard deviation of R₂). Thus, the relation of R₂ to cognitive decline was largely independent of the three vessel disease indices considered in this work.

One explanation for these findings is that the R₂ alterations associated with cognitive decline reflect tissue damage stemming from conditions separate from those pathologies that were controlled for in the current work, including cerebral vessel diseases. This does not necessarily rule out a variety of other vascular etiologies, such as brain hypoperfusion (Brickman et al., 2009) stemming from, for example, decreased cardiac output (Alosco et al., 2013, van der Velpen, Isabelle F et al., 2017) or obstructive sleep apnea (Avci et al., 2017, Macey et al., 2008). Quantitative R₂ imaging may therefore be a valuable tool for assessment of damage to white matter inflicted by a variety of pathologies or conditions that are not captured by standard neuropathologic evaluation, but future work is needed to examine this issue.

Another possibility is that R₂ alterations indeed stem from cerebral vessel disease or other pathologies already considered in our analyses, but that R₂ probes an additional facet of the disease process that accounts for variation in the rate of decline beyond the pathologic indices themselves. For example, although atherosclerosis, arteriolosclerosis, and CAA all disrupt cerebral hemodynamics, the severity and chronicity of the resulting hypoperfusion may also depend on factors like blood pressure and autoregulation (van der Velpen, Isabelle F et al., 2017) and the extent of collateral blood supply to the tissue (Campbell et al., 2013). Neuroimaging markers such as R₂, it seems, may provide a summary measure of the tissue damage resulting from the altered hemodynamics brought about by cerebral vessel diseases, such as myelin degradation, axonal loss, or edema (Gouw et al., 2010). Because of their more direct link to the disruption of action potentials, these types of tissue damage may have a stronger influence on cognition than do the vessel disease indices themselves, which are envisioned to reside upstream in the causal chain (Banerjee et al., 2016, Weller et al., 2009). In this regard, postmortem R₂ imaging can provide powerful measures of tissue integrity that are complementary to traditional histopathologic evaluation of cerebral vessel disease and other neuropathologies.

This study also extends prior work by examining for the first time the temporal association of R₂ with cognitive decline. This was elucidated in the current work using changepoint models, which account for the nonlinearity of late life cognitive decline by fitting a cognitive trajectory that has distinct preterminal and terminal phases with differing slopes. We observed an association of the composite R₂ measure with the rate of preterminal decline and also with the temporal location of the changepoint, which marks the onset of the more rapid terminal decline phase. Therefore, just as hallmark AD neuropathology may begin accumulating in the brain as early as the fifth decade of life (Braak et al., 2011), the underlying contributors to alterations in postmortem R₂ appear to be influencing the cognitive trajectory well before death, not just in the final few years. Future interventions targeting those contributors are therefore promising in their ability to decrease the rate of cognitive decline during the preterminal period and also to delay the onset of the more severe terminal decline that immediately precedes death, resulting in compression of morbidity.

This study has strengths as well as limitations. One strength is the combination of autopsy data, brain MRI, and up to 22 annual cognitive evaluations in older adults, which is rare and allows for investigation of the associations between neuroimaging and cognition independent of known neuropathology. This is most feasibly achieved in large, community-based studies, but such studies have some limitations, including that the cohort is composed of willing participants who may not be representative of the general population. In addition, while the postmortem approach to MRI employed in this study was the most viable means of collecting the several hundred imaging datasets with matching autopsy data, this technique may be regarded as a limitation because of the uncertainty of translating results derived from postmortem imaging to the antemortem case. However, recent work demonstrating a strong correlation between antemortem and postmortem R₂ values suggests that such a translation is promising, if not yet assured. Another limitation of the current work is related to the voxelwise analysis technique, which provides excellent spatial specificity, but at the expense of power. Work is underway to advance adaptive region-based approaches that improve statistical power without meaningful sacrifices to spatial resolution. This postmortem study is also limited in that we did not account for participants’ cause of death in analyses. Finally, we did not account for white matter hyperintensities (WMH) that are commonly visible in the brains of older adults on fluid-attenuated inversion recovery (FLAIR) MRI pulse sequences. FLAIR data collection is ongoing, and refined WMH segmentation methods are being developed toward the goal of elucidating the associations among R₂, WMH, and cognitive decline.

HIGHLIGHTS.

• Postmortem MRI transverse relaxation (R₂) is associated with the linear rate of global cognitive decline in late life and accounts for more than 5% of its variance.

• The association of R₂ with cognitive decline persists after accounting for indices of cerebral vessel disease, namely atherosclerosis, arteriolosclerosis, and cerebral amyloid angiopathy.

• R₂ is also associated with specific components of a nonlinear model of cognitive decline.

• R₂ reflects a dimension of brain tissue integrity not captured by current histopathologic indices and therefore may highlight important targets for future interventions against late life cognitive decline.