Regional Age-related Atrophy After Screening For Preclinical Alzheimer Disease

Contact: emily.li@wecistanche.com

Lauren N. Koenig, et al

Abstract:

Brain atrophy occurs in aging even in the absence of dementia, but it is unclear to what extent this is due to undetected preclinical Alzheimer's disease. Here we examine a cross-sectional cohort (ages 18-88)free from the confounding influence of preclinical Alzheimer's disease, as determined by amyloid PET scans and three years of clinical evaluation post-imaging. We determine the regional strength of age-related atrophy using linear modeling of brain volumes and cortical thicknesses with age. Age-related atrophy was seen in nearly all regions, with the greatest effects in the temporal lobe and subcortical regions. When modeling age with the estimated derivative of smoothed aging curves, we found that the temporal lobedeclined linearly with age, subcortical regions declined faster at later ages, and frontal regions declined slower at later ages than during midlife. This age-derivative pattern was distinct from the linear measure of age-related atrophy and signifificantly associated with a measure of myelin. Atrophy did not detectably differ from a preclinical Alzheimer's disease cohort when age ranges were matched.

Keywords: Normal Aging, Volumetrics, Preclinical Alzheimer's disease, Magnetic Resonance Imaging (MRI)

Click here to get more information about Cistanche for Anti Alzheimer's Disease

1. Introduction

Older adults constitute an increasingly large fraction of our society, making research on brain aging important for public health. Cerebral atrophy associated with aging is, in particular, a concern due to its association with cognitive decline, independent of known neurodegenerative diseases (Armstrong et al., 2020; Fletcher, Gavett, et al., 2018). Previous studies have shown regional variability and non-linear changes in this atrophy occurring with age. In general, these studies show the strongest atrophy in frontal and temporal regions, and a pattern of accelerated atrophy in temporal regions (Irwin et al., 2018; Lockhart & DeCarli, 2014). It has been hypothesized that these non-linear regional patterns may in part be due to mid-life increases in cerebral myelination causing the appearance of reduced gray matter density (Irwin et al., 2018). However, myelin may also be acting as a proxy for other regional properties of the brain such as intracortical circuit complexity and aerobic glycolysis levels (Glasser et al., 2014).

Measures of age-related atrophy are complicated by abundant confounding factors inherent within studies of aging. One major factor is cardiovascular disease, with atrophy correlating with white matter hyperintensities (Coutu et al., 2017; Habes et al., 2021), high blood pressure (Armstrong, An, et al., 2019; Lockhart & DeCarli, 2014), and diabetes (Hamed, 2017; Suzuki et al., 2019). Some studies indicate sex or gender differences in age-related atrophy, with greater atrophy in men for select regions (Armstrong, Huang, et al., 2019; Chételat et al., 2010; Jack et al., 2015; Lockhart & DeCarli, 2014; Wang et al., 2019). Additionally, apolipoprotein E ε4 (APOE4) - the greatest genetic risk factor for sporadic Alzheimer's disease (AD) - has also been associated with greater rates of atrophy even in the unimpaired (Armstrong, An, et al., 2019; Erten-Lyons et al., 2013; Irwin et al., 2018; Kelly et al., 2018; Mishra et al., 2018; Raz et al., 2010; Smith et al., 2012). A previous study has shown that this effect of APOE4 is linked to increasing amyloid levels, indicative of the preclinical stage of Alzheimer's disease (Mishra et al., 2018).

Preclinical AD is characterized by the absence of cognitive symptoms and the presence of parenchymal deposits of amyloid- β peptide, one of the hallmarks of AD. Despite its association with atrophy (Becker et al., 2011; Chételat et al., 2012; Dickerson et al., 2009; Fagan et al., 2009; Fjell et al., 2010; Fletcher et al., 2016; Fletcher, Filshtein, et al., 2018; Oh et al., 2014; Pettigrew et al., 2017; Schott et al., 2010; Storandt et al., 2009; Xie et al., 2020), preclinical AD can only be detected on an individual basis using measures of amyloid. As such, it often goes undetected in studies of aging populations and may be contaminating results. For example, screening out participants with preclinical AD has been shown to reduce variability and age-related decline in neuropsychological tests (Hassenstab et al., 2016) and resting-state functional connectivity measures (Brier et al., 2014). However, it is unclear if this confound extends to measures of atrophy

Prior studies have assessed the impact of undetected Alzheimer's pathology (Armstrong, Huang, et al., 2019; Fjell, McEvoy, et al., 2013; Fjell, Westlye, et al., 2014; Fjell, McEvoy, et al., 2014; Knopman et al., 2013), using either measure of amyloid pathology or longitudinal tracking to ensure no cognitive impairment develops. However, sample sizes were small in these studies and screening used longitudinal tracking or amyloid measures separately. In this study we use cognitively normal participants from longitudinal Alzheimer's disease studies, allowing us to screen a large cohort for preclinical Alzheimer's disease using both amyloid PET and longitudinal tracking of cognition in the same individuals. Using this screened cohort, we measure age-related volumetric changes across the brain that occur independently of preclinical Alzheimer's disease.

2. Methods

2.1. Participants

The n = 383 participants in the Normal Aging Cohort came from 2 open source databases: Open Access Series of Imaging Studies (OASIS) (LaMontagne et al., 2019) and the Dominantly Inherited Alzheimer Network (DIAN). The n = 115 participants in the Preclinical AD Cohort were all from OASIS. All procedures in this retrospective study were HIPAA compliant and approved by the Washington University Institutional Review Board; informed consent was gained for all participants.

Both the Normal Aging and the Preclinical Alzheimer's disease Cohorts only included participants who were evaluated as “Cognitively normal” or “No dementia” in their clinical assessment and who had a global Clinical Dementia Rating (CDROM) (Morris, 1993) of 0 within 1 year of Magnetic Resonance Imaging (MRI). The Normal Aging Cohort, which has been previously described (Koenig et al., 2020), only included participants who remained CDR 0 for a minimum of 3 years after MRI. Participants over age 45 were only included if they additionally had a negative amyloid PET scan (defined in section 2.4) within 1 year of their MRI. The longitudinal CDR and negative amyloid PET scan limited the possibility that the participants in the Normal Aging Cohort were in the preclinical stage of Alzheimer's disease. The Preclinical AD Cohort differed from the Normal Aging Cohort in that it required a positive amyloid PET scan and did not require longitudinal CDR assessment.

While the Normal Aging Cohort included participants from DIAN, a study on autosomal dominant Alzheimer's disease caused by rare mutations, only non-mutation carriers (control group) were included. DIAN was used due to its similarity to studies in the OASIS database and because DIAN has amyloid PET data available in the 45-60 age range. When compared to OASIS participants in the overlapping age range (age 42-59), there were no differences in volumetric data after multiple comparisons (section 2.6 and Supplemental Table S1). Both OASIS and DIAN include self-reported race and gender. We use the term gender and not sex to match the terminology of the questionnaire used, but participants were offered only “Male” and “Female” as options, and sex was not assessed separately.

2.2. Clinical assessment

Experienced clinicians, blinded to amyloid status, evaluated each participant for the possibility of a clinical diagnosis of dementia, and only those considered to be cognitively normal were included in this study. Their assessment, outlined previously (Morris et al., 2006), integrated results from a semi-structured interview conducted with the participant and a knowledgeable collateral source, a thorough neurological examination, and bedside measures of cognitive function (including Mini-Mental State Exam (MMSE) (Folstein et al., 1975) among others).

2.3. MR imaging

The MR imaging parameters for OASIS are approximate due to the variety of studies included. Scanner strength was primarily 3T (n = 19 were 1.5T) within OASIS, while DIAN was 3T. OASIS T1- weighted images (MPRAGE) primarily had 2 sets of parameters. The first used TR = 2.3 s, TE = 3.16 ms, TI = 1 s, a flip angle of 8 degrees, and a spatial resolution of 1 × 1 × 1. The second used TR = 2.3 s, TE = 2.95 ms, TI = 0.9 s, a flip angle of 9 degrees, and a spatial resolution of 1 × 1 × 1 or 1 × 1 × 1.2 mm3. DIAN T1 scans had TR = 2.3 s, TE = 2.95 ms, TI = 0.9 s, a flip angle of 9 degrees, and a spatial resolution of 1 × 1 × 1.2 mm3.

Volumetric T1-weighted images underwent regional tissue segmentation with FreeSurfer (version 5.0 or 5.1 for 1.5T scans and version 5.3 for 3T scans) (Fischl, 2012). Regional volumes (cortical and subcortical) were adjusted for head size with a regression approach using intracranial volume (Buckner et al., 2004). Left and right hemispheric data were combined by summing volumes and averaging cortical thicknesses.

2.4. PET imaging

[11C]-Pittsburgh compound B was used as the amyloid tracer for Indian participants, with a dosage ∼of 15 mCi, and data was collected 40-70minutes post-injection. Within OASIS, 287 participants were imaged using [11C]-Pittsburgh compound B, with a dosage ∼of 13 mCi and data collected 30-60 minutes post-injection. The remaining 75 participants were imaged using Florbetapir ([18F]-AV45), with a dosage ∼of 10 mCi, and data was collected 50-70 minutes post-injection.

PET images were processed with an in-house pipeline using FreeSurfer-derived regions (Su, 2014/2021) and a cerebellar cortex reference region. Signal spillover was addressed with partial volume correction, specifically with a regional spread function (geometric transfer matrix) technique based on the scanner point spread function and the relative distance between regions (Su et al., 2013, 2015). The mean cortical Standard Uptake Value Ratio with Regional Spread Function applied (SUVR rsf) was de- defined as the average SUVR rsf from the precuneus, prefrontal cortex, gyrus rectus, and lateral temporal regions (Su et al., 2019).

A negative amyloid PET scan was defined as having a mean cortical SUVR rsf<1.42 (Centiloid<16.4) for [11C]-Pittsburgh compound B PET or SUVR rsf<1.19 (Centiloid<20.6) for Florbetapir PET. The Centiloid conversion process, used to more easily compare the two amyloid tracers, is documented in detail in the initial Centiloid paper (Klunk et al., 2015), with specifific equations in follow-up papers (Su et al., 2018, 2019). Harmonization procedures such as this are imperfect, and so to remain as accurate as possible we used cutoffs determined individually for each tracer and then converted into Centiloid, as opposed to a unified Centiloid cutoff.

2.5. T1w/T2w myelin maps

This study uses a spatial map of the ratio of T1w/T2w image intensities in a cohort of 1071 healthy young adults (ages 22- 37, mean 29) from the Human Connectome Project (Glasser et al., 2014, 2016a,b; Glasser and Van Essen, 2011). The original map was averaged within each region of the Desikan-Killiany atlas used by FreeSurfer to allow comparison. Prior work has shown that this ratio correlates with cerebral cortical myelin content due to differences in lipids, free and myelin-bound water, and iron content (Glasser & Van Essen, 2011).

2.6. Statistics

We first examined if gender, MMSE, APOE4, race, and education influenced linear models of each regional volume (after normalization for intracranial volume) and each cortical thickness in the normal Aging Cohort. A separate linear model was run for every factor and regional volume or thickness pairing, with a BonferroniHolm, corrected p < 0.05 considered signifificant. Bonferroni-Holm, which progressively adapts the significance threshold, was done separately for each of the five factors, and across the 101 examined brain regions. Race in this study was self-reported and binarized to Caucasians and non-Caucasian due to the relatively small percentage of non-Caucasian participants. As few signifificant correlations were observed for any of these factors, we did not include these covariates in the remaining analyses.

We next modeled each regional volume and thickness by age. We used the resulting standardized coefficients (β weights) to compare the strength and directionality of age-related atrophy across regions. We then addressed non-linear changes that occur with age using the estimated derivative of normal aging curves. Normal aging curves were determined by smoothing the Normal Aging Cohort’s data for each FreeSurfer region with a locally weighted scatter-plot smoother regression, resulting in a nonlinear estimate of age-related atrophy. By correlating age with the estimated derivative at each age, we estimate the pattern of age-related atrophy across the lifespan. We display examples of these normal aging curves and their estimated derivatives in Fig. 1. As these are cross-sectional data, the estimated derivative is the change in the region’s smoothed average by age, not an individual participant’s trajectory over time. As with the previous analysis, we again corrected each set of p-values for multiple comparisons across the 101 regions using Bonferroni-Holm.

Fig. 1. Example Regions in the Normal Aging Cohort. Fig. 2 displays regional maps of the standardized β weights from the linear models used to assess age-related atrophy. Fig. 2A displays the overall age effect, taken from a direct comparison of participants’ ages and regional volume/thickness (blue line in Fig. 1A-D). A darker purple indicates more atrophy with age, while yellow indicates a lack of atrophy. Fig. 2B displays the pattern of atrophy with age, taken from the association of the age-derivative with age (blue line in Fig. 1E-F). Blue in Fig. 2B indicates regions whose rate of atrophy becomes less severe as age increases, while red indicates regions whose atrophy accelerates at later ages. To maintain the color schemes, the lateral ventricles are displayed with a reversed sign.

The 34 beta values for each of the four resulting cortical maps(from linear models of cortical volumes or thicknesses; as predicted by age or the age-derivative) were correlated with the myelin map described above. As this is a spatial correlation, Spearman’s rank correlation was used. To maintain consistency, these pvalues were also corrected for multiple comparisons across the 4pairings using Bonferroni-Holm.

Finally, we used linear models to assess the impact of amyloid on regional volumes and thicknesses. The Preclinical AD Cohort (amyloid positive) and the participants above age 60 in the normal Aging Cohort were combined, and a linear model was run for each region using age, amyloid positivity, and their interaction. This process was also repeated by replacing amyloid positivity with a continuous measure of amyloid (Centiloid). Each set of p-values was corrected for multiple comparisons across the 101 regions using Bonferroni-Holm.

3. Results

3.1. Demographics

Demographics for both cohorts and the subset of the Normal Aging Cohort above age 60 are listed in Table 1. As expected, the Normal Aging Cohort had a lower frequency of APOE4 alleles and lower amyloid levels than the Preclinical AD Cohort. No regions in the Normal Aging Cohort showed signifificant associations with APOE4 status, MMSE, or years of education, and few regions showed signifificant associations with gender or race after correction for multiple comparisons (Supplemental Table S1). As such, the later analyses did not adjust for these factors.

Signifificant differences by gender were observed in intracranial volume (β = 0.601, corrected p < 0.001), fusiform volume(β = 0.200,corrected corrected p = 0.008), frontal pole volume (β = 0.182,p = 0.03), lateral occipital volume (β = 0.178, p = 0.04),amygdala volume (β = 0.230, corrected p < 0.001), and lateralventricle volume (β = -0.197, p = 0.01). Signifificant differencesby race were in cuneus volume (β = 0.186, p = 0.03), inferiortemporal volume (β = 0.182, p = 0.04), lateral occipital volume(β = 0.198, p = 0.01), middle temporal volume (β = 0.220, p =0.001), and optic chiasm volume (β = -0.195, p = 0.01). In thesemodels, a positive β weight indicates larger volumesor thicknessesin men or Caucasians, respectively.

3.2. Regional variation in the strength of age-related atrophy

Almost all regions showed a signifificant association between atrophy and age in the Normal Aging Cohort (Supplemental Table S2). The only non-signifificant regional measures were caudal anterior cingulate thickness, entorhinal volume, temporal pole volume, corpus callosum posterior volume, intracranial volume, total subcortical gray matter volume, and fifth ventricle (cavum septum pallidum) volume. While volumetric measures of the remaining regions were signifificantly associated with age, the strength of that relationship varied. The strongest age effects were seen in the temporal lobe and subcortical regions (Fig. 2A). Of the regions and composites not pictured in Fig. 2A, summary measures such as total cortex volume and total gray matter volume also showed some of the strongest age effects (Supplemental Table S2).

Fig. 2. Regional Maps of Age-Related Atrophy. Fig. 2 displays regional maps of the standardized β weights from the linear models used to assess age-related atrophy. Fig. 2A displays the overall age effect, taken from a direct comparison of participants’ ages and regional volume and/or thickness (blue line in Fig. 1A-D). A darker purple indicates more atrophy with age, while yellow indicates a lack of atrophy. Fig. 2B displays the pattern of atrophy with age, taken from the association of the age-derivative with age (blue line in Fig. 1E-F). Blue in Fig. 2B indicates regions whose rate of atrophy becomes less severe as age increases, while red indicates regions whose atrophy accelerates at later ages. To maintain the color schemes, the lateral ventricles are displayed with a reversed sign. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

3.3. Regional variation in non-linear patterns of age-related atrophy

The previous section used standardized β weights from linear models to compare the strength of the relationship between age and regional volumetrics. Select regions declined in a linear fashion. Many regions showed non-linear patterns, with atrophy appearing to accelerate or decelerate at older ages. We assessed the non-linear pattern of each region by smoothing our data to create normal aging curves and then estimating the derivative of that curve at each age. Fig. 1 displays examples of these normal aging curves and the corresponding estimated derivatives.

Almost all regions’ age-derivative showed a signifificant association with age (Supplemental Table S2). The regions showing nonsignificant correlations of age were banks of the superior temporal sulcus thickness, fusiform thickness, and pars opercularis volume. Non-significance, in this case, indicates no relationship, i.e. rate of atrophy did not change linearly across the age range suggesting linear decline or no atrophy with age. The strength of the association between age and the age-derivative again represented using β weights, is displayed spatially in Fig. 2B and appears distinct from the age-association pattern in Fig. 2A. Of those regions that showed the most age-related atrophy, the temporal cortex showed an overall linear decline with age, while atrophy in subcortical regions appears to accelerate with age. In contrast, frontal regions appear to show higher rates of atrophy in midlife as opposed to late life. Of the regions not pictured in Fig. 2B, the corpus callosum stood out as a region stable at younger ages that atrophies rapidly in old age.

3.4. Relationship of T1w/T2w myelin content and slope of age-related atrophy

To quantify if the spatial patterns we observed in Fig. 2 related to myelin levels, we correlated each set of 34 cortical beta weights in Fig. 2 to an average T1w/T2w myelin map. This myelin map was generated on a separate cohort of healthy young adults (ages 22- 37, mean 29) and was and is displayed in Fig. 3. The regional pattern of the strength of age-related atrophy was not signifificantly associated with the regional map of myelin (rho = -0.060, corrected p = 0.74 for cortical volumes; rho = -0.348, corrected p = 0.09 for cortical thicknesses). However, the regional pattern of the estimated derivative βs was signifificantly associated with the regional map of myelin (rho = -0.640, corrected p < 0.001 for volume; rho = -0.546, corrected p = 0.003 for thickness). The directionality of the correlation is such that regions with higher myelin content are more likely to follow the pattern shown in Fig. 1C, with atrophy that accelerates in late life. Conversely, lower myelin regions were more likely to show the pattern in Fig. 1D: atrophy greatest in midlife and tapered at older ages. While this result emphasizes the distinctness of the two patterns, the moderate correlation suggests other factors are also at play.

3.5. Atrophy in preclinical AD vs normal aging

The impact of amyloid was assessed using those over age 60 in the Normal Aging Cohort (amyloid negative) and the Preclinical AD Cohort (amyloid positive). Linear models used age, amyloid, and age∗amyloid to predict regional volumes or thicknesses. No significant effects of amyloid or amyloid∗age were found after accounting for age and correcting for multiple comparisons (Supplemental Table S3, with examples in Fig. 4).

Fig. 4. Example Regions for Normal Aging Cohort vs. Preclinical AD Cohort. Fig. 4 displays the overlap of the Normal Aging Cohort (Amyloid Negative, black) and the Preclinical Cohort (Amyloid Positive, red), indicating our non-signifificant fifindings for amyloid and age∗amyloid. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

4. Discussion

In this paper, we report regional variation in age-related atrophy, with different spatial patterns for the effect size of age-related atrophy and in the non-linear pattern observed across the lifespan. Temporal regions showed the greatest association with age, while frontal and cingulate areas showed deceleration of atrophy with age (i.e. higher atrophy in mid-life than late-life). This reduced rate of atrophy in late life contrasted to most regions which showed accelerating atrophy in late life. This pattern of non-linearity was spatially related to myelin levels determined by T1w/T2w intensity ratio. As this ratio was determined in a separate cohort of healthy adults, this suggests that the observed pattern is the end result of a fundamental organizational property of the brain. The lack of correlation between myelin and the direct association with age further supports that the two observed patterns are unique. The direction of the myelin and age-derivative correlation suggests that regions that characteristically have higher myelin content in midlife are more vulnerable to accelerated atrophy in later life. While causality is not clear, this could in part be due to the greater vulnerability of myelinating cells to oxidative stress (Nasrabady et al., 2018). No differences were detected between our Normal Aging Cohort and our Preclinical AD Cohort, though a larger sample may reveal subtle differences.

Limitations of this study include its cross-sectional design, the lack of diversity in our participants, and our inability to control for vascular inflfluences on structural brain measures in these analyses. Group averages in aging volumetrics have been shown to be commensurate across cross-sectional and longitudinal designs (Fjell, Westlye, et al., 2013; Fotenos et al., 2005). However, by looking only at a single time-point per participant, we were unable to assess possible subtypes of patterns of aging in individuals. Our Normal Aging Cohort, collated from several studies of aging and Alzheimer's disease, is predominantly highly educated and Caucasian. A more representative cohort may show greater age-related atrophy due to the association of social inequities with chronic health conditions and other social determinants of health. As such, our study may be closer to a measure of ‘healthy aging’ than the ‘normal aging’ of an average individual in our social experiences.

While our cohort may not be representative of the broader population, it does reduce the probability that some unmeasured factors are confounding our measures of aging. Vascular disease is one such unmeasured factor that is common within the population represented in this study and likely impacts our results. Differences in blood pressure, even in non-hypertensive individuals, have correlated with volumetric differences (Lockhart & DeCarli, 2014). Additionally, regional volumetrics may be influenced by other non-AD neurodegenerative pathologic processes that are less common and more difficult to detect (e.g. argyrophilic grain disease, primary aging-related tauopathy, hippocampal sclerosis of aging, limbic-predominant aging-related TDP-43 encephalopathy neuropathologic change, aging-related tau astrogliopathy, frontotemporal lobar degeneration, Lewy body disease). For our detected pathology, amyloid, we are limited in that we did not follow our Preclinical AD participants longitudinally. We would expect some but not all of these participants to develop AD in the near future, and these two subgroups would likely have different rates and patterns of atrophy. One fifinal limitation is that the FreeSurfer regions used in this study were relatively coarse regions defined based on gyral and sulcal landmarks that contain signifificant structural and functional heterogeneity. This limits the neurobiological interpretability of regional effects as compared to cortical areas based on multiple modalities (Glasser, Coalson, et al., 2016) or more homogeneous functional regions (Gordon et al., 2016).

Despite these limitations, our results indicate that age-related atrophy is a regionally heterogeneous process, with the severity of atrophy and lifespan pattern of atrophy varying independently across regions. We also showed that age-related atrophy is not significantly associated with amyloid positivity in the absence of cognitive symptoms. This suggests that volumetric studies in older adults do not need to include amyloid PET scans to screen for preclinical AD or track their participants longitudinally for dementia if they instead use the same rigorous dementia screening we used at baseline (integrating a comprehensive history with a trusted collateral source and neurological examination). Previous similar studies had smaller sample sizes and were unable to screen for both longitudinal cognition and amyloid levels, giving new weight to our negative findings. Future studies should further investigate the association we saw between myelin levels and lifespan pattern of atrophy, as well as the potential influence of non-AD neurodegenerative pathologies.

5. Conclusions

The amount of atrophy that occurs with age and the pattern of decline over the lifespan exhibit two unique spatial patterns, with only the second pattern associated with regional myelination. Broadly, these patterns indicated the greatest atrophy in the temporal lobe and subcortical regions, with linear patterns of decline in temporal lobe regions, accelerating decline in subcortical regions, and decelerating declines in frontal regions. Despite measures of amyloid PET and longitudinal CDR, our aging cohort and preclinical Alzheimer's disease cohort did not show measurable differences in atrophy.

Disclosure statement

Authors declare that data contained in the manuscript being submitted have not been previously published, have not been submitted elsewhere, and will not be submitted elsewhere while under consideration at Neurobiology of Aging.

From: ' Regional age-related atrophy after screening for preclinical Alzheimer's disease' by Lauren N. Koenig, et al

---L.N. Koenig, P. LaMontagne, M.F. Glasser et al. / Neurobiology of Aging 109 (2022) 43–51