Advanced Glycation End Product Formation in Human Cerebral Cortex Increases With Alzheimer-Type Neuropathologic Changes but Is Not Independently Associated With Dementia in a Population-Derived Aging Brain Cohort

Abstract

Diabetes mellitus is a risk factor for dementia, and nonenzymatic glycosylation of macromolecules results in formation of advanced glycation end-products (AGEs). We determined the variation in AGE formation in brains from the Cognitive Function and Ageing Study population-representative neuropathology cohort. AGEs were measured on temporal neocortex by enzyme-linked immunosorbent assay (ELISA) and cell-type specific expression on neurons, astrocytes and endothelium was detected by immunohistochemistry and assessed semiquantitatively. Fifteen percent of the cohort had self-reported diabetes, which was not significantly associated with dementia status at death or neuropathology measures. AGEs were expressed on neurons, astrocytes and endothelium and overall expression showed a positively skewed distribution in the population. AGE measures were not significantly associated with dementia. AGE measured by ELISA increased with Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) neurofibrillary tangle score (p = 0.03) and Thal Aβ phase (p = 0.04), while AGE expression on neurons (and astrocytes), detected immunohistochemically, increased with increasing Braak tangle stage (p < 0.001), CERAD tangle score (p = 0.002), and neuritic plaques (p = 0.01). Measures of AGE did not show significant associations with cerebral amyloid angiopathy, microinfarcts or neuroinflammation. In conclusion, AGE expression increases with Alzheimer’s neuropathology, particular later stages but is not independently associated with dementia. AGE formation is likely to be important for impaired brain cell function in aging and Alzheimer’s.

Advanced glycation end products, Alzheimer’s disease, Dementia, Diabetes, Tau

INTRODUCTION

Type II diabetes mellitus (DM) and the related metabolic syndrome, characterized by cellular insulin resistance resulting in impaired glucose utilization and hyperglycemia, are increasing problems in affluent societies. DM is a risk factor for dementia (1). This may be partly mediated via the common neuropathologic causes of dementia, as DM is a risk factor for Alzheimer’s disease (AD) and may contribute to vascular dementia through large vessel atherosclerosis and small vessel disease (2, 3). Population-based approaches suggest that up to one-third of AD may be attributable to modifiable risk factors. These include DM and midlife obesity with respective attributable risks for AD of 1.9% and 6.6% in the UK (4). However, the relationship between DM and AD remains unclear as several studies have shown that brains from individuals with DM do not show increased levels of AD neuropathology (3, 5, 6).

DM may impair the brain and contribute to dementia through several candidate mechanisms. Endothelial and neurovascular dysfunction occur in DM (7, 8), and microvascular dysfunction is also implicated early in AD pathogenesis (9-11). Islet amyloid polypeptide (amylin) accumulates in islet β-cells in DM and contributes to disease pathogenesis, and there is evidence that it also accumulates in AD brain where it interacts with Aβ to exacerbate the disease process (12). DM and AD share cellular insulin resistance, and AD patients have shown improved cognition with intranasal insulin (13-17). Peripheral insulin resistance and brain insulin resistance may be linked (18), but other mechanism may also contribute to neuronal and glial insulin resistance. Binding of Aβ-oligomers to the insulin receptor may contribute to impaired neuronal insulin signaling in neurons (19, 20), and more recently Aβ has been shown to reduce insulin levels in astrocytes (21). Dysregulation of expression of insulin signaling genes has been found in neurons associated with a DNA damage response (22), and downregulation of gene expression for insulin and other associated signaling pathways has been found in astrocytes with AD neuropathology progression (23). There is also an interesting question of reversed causality whereby brain disease, particularly AD, is a risk factor for DM (1, 24).

DM and the related metabolic syndrome are also associated with metabolic disturbances in carbohydrate, lipid, and amino acid metabolism (25, 26), suggesting more widespread metabolic derangement related to macromolecular damage and nutrient signaling. Glucose transport to neurons via non-insulin dependent glucose transporters means that elevated plasma glucose is also reflected in the brain and can contribute to neurodegeneration (27, 28). Hyperglycemia causes carbonyl stress, resulting in non-enzymatic glycation of proteins via intermediates such as methylglyoxal to form a complex variety of advanced glycation end-products (AGEs). Reactive dicarbonyls and AGE are increased in diabetic patients; they increase over lifetime and expression of AGE correlates with diabetic complications. AGE formation, including in the nervous system, reflects high glycemic states (29). Thus, AGE can be used as markers of glycosylation damage and affects cells by binding to receptors for AGE (RAGE).

Assessing the effects of DM on the brain and on cognition is difficult because of the difficulty in quantifying DM over the life-course (30). Furthermore, subclinical impaired glucose tolerance/insulin resistance is of increasing importance in the population, forming a component of the metabolic syndrome (31) but its effects are yet more difficult to quantify. Levels of glycated hemoglobin (HbA_1C) showed a relationship to cognitive impairment in the population-based Cognitive Function and Ageing Study (CFAS) (32), so that assessment of glycation may allow quantitative analysis, although HbA_1C may fluctuate over time with red cell turnover.

AGE formation in the CNS is, potentially, a means to assess one important aspect of the effects of DM and/or subclinical impaired glucose tolerance on brain cellular pathology, namely metabolic damage due to hyperglycemia and carbonyl stress, particularly as some of the neuronal proteins may have long half-lives. Therefore, we investigated the variation in AGE and the relationship between extent of glycation to Alzheimer-type neuropathology and other cellular pathologies, using an epidemiological neuropathology approach, in the CFAS aging population-representative neuropathology cohort (33, 34). CFAS is a longitudinal population-based study of cognitive impairment and frailty in the elderly (66 years and older) general UK population (33, 35). Studies using the neuropathology cohort of CFAS have previously shown that mixed pathology is common in dementia in a population setting, and that there is overlap in burdens of pathology between demented and non-demented individuals with a lack of thresholds for dementia, particularly at the oldest ages where the relationship between classical neuropathologic lesions and dementia status is attenuated (36-38). Using a population-based approach in this cohort avoids biases due preselection into diagnostic groups based on both clinical and pathological information, and provides complementary information to case-control studies based on well-defined groups. We have, therefore, examined the relationship of AGE to specific pathologies and to dementia separately.

MATERIALS AND METHODS

CFAS Cohort

Formalin-fixed paraffin-embedded postmortem tissue from the lateral temporal cortex (BA21/22) was obtained from brains donated to one center (Cambridge, n = 99), of the well-characterized aging population representative CFAS neuropathology cohort (33, 34, 39). Dementia status at death was established as present, absent, or uncertain, on the basis of AGECAT algorithm, death certification and Retrospective Informant Interview (33, 38, 40). Diabetes status was assessed by self-reporting of type 2 diabetes prior to death (self-reported DM [SR-DM]). Neuropathologists examined the brains and categorized pathology based on several measures: Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) stage (41) and Braak neurofibrillary tangle (NFT) stage (42). Additional measures of Aβ pathology were also available on the cohort (43), including Thal Aβ phase (44) and CAA. For the latter, the severity of vascular amyloid in cortical parenchymal and leptomeningeal vessels was assessed, each scored out of 3 according to the method of Love et al. (45). To obtain an aggregate measure of cortical CAA severity, we summed the scores for CAA severity in leptomeningeal and parenchymal vessels in the 4 cortical areas, giving a maximum score out of 24. Assessment of microinfarcts was performed as previously described, and analysis used to total number of cortical areas (out of 4) with a microinfarct (46). Data on markers of gliosis (47), DNA damage (48), and inflammation (49) were previously obtained in this cortical area in the same cohort. Blood-brain barrier changes were previously assessed using immunohistochemistry to albumin and fibrinogen, scored semiquantitatively (50). Ethical committee approval was granted for all procedures in this study (15/SW/0246).

AGE Enzyme-Linked Immunosorbent Assay

Total cortical AGE formation was assessed using enzyme-linked immunosorbent assay (ELISA). Twenty cases from the cohort were excluded from this analysis due to either low protein reads or lack of suitable frozen tissue, so that ELISA results are based on 79 cases. The amount of protein per microliters was determined with a BCA protein assay. To determine the levels of AGE present in each sample, OxiSelect Competitive ELISA kits (STA-817, Cambridge Bioscience, Cambridge, UK) were used according to the manufacturer’s instructions, loading 50 μL of each sample per well. The amount of AGE per micrograms of protein was calculated using the original protein concentration. Controls were included on each plate to ensure consistency between plates.

Cell-Associated Expression of AGE

Cell-associated expression of AGE was assessed using immunohistochemistry using all 99 cases. Lateral temporal cortex was sectioned at a thickness of 5 μm, sampled from a formalin-fixed paraffin-embedded block. Immunostaining was performed with a standard avidin biotinylated enzyme complex method, and the signal visualized with diaminobenzidine (Vector Laboratories, Peterborough, UK). The anti-AGE antibody (rabbit polyclonal, ab23722, Abcam, Cambridge, UK) was used at a 1 in 2000 dilution following microwave antigen retrieval in tri-sodium citrate, pH6.5. Negative controls were sections incubated with either isotype controls (Vector Laboratories), or omission of the primary antibody.

Semiquantitative analysis was performed for expression on individual cell types (neurons, glia and endothelial cells/blood vessels). Staining for neurons and glia, respectively, was scored as no staining (0), mild (up to 25%, +1), moderate (25%−50%, ++2), or frequent expression (>50%, +++3). Vessel staining was scored as 0, +1 (50%), or ++2 (>50%) (Supplementary Fig. S1). Scoring was performed by 2 observers, and interobserver variance assessed by Cohen’s κ.

Statistical Analysis

Statistical analyses were performed using IBM-SPSS (version 24, IBM, Portsmouth, UK) and STATA version 14 (StataCorp, College Station, TX). Normality testing was performed using the Kolmogorov-Smirnov test. As most data were nonparametric, correlation analyses were performed using Spearman’s rank test and variation between groups was assessed by Kruskal-Wallis (KW) test, with p values from the post hoc testing corrected using the Bonferroni method. Mann-Whitney was used for comparison of continuous data between 2 groups. For univariate analyses of AGE with Braak NFT stage, the 6 stages were combined into 3 groups representing entorhinal stages (Braak stages 0–2; 30 cases), limbic stages (Braak stages 3–4; 51 cases) and isocortical stages (Braak stages 5–6; 18 cases). For Thal phase, similarly the phases were combined into 3 groups with no or only cortical Aβ (phases 0–1; 19 cases), hippocampal/limbic Aβ (phases 2–3; 33 cases), and midbrain/cerebellar involvement (phases 4–5; 27 cases). For CERAD plaque and tangle scores, the moderate and severe groups were combined so that analysis involved 3 groups: None, mild, and moderate-severe.

The relationships between the outcomes of SR-DM and AGE with dementia and neuropathology were investigated using logistic regression, controlling the analyses for age of death and sex. The relationships between SR-DM and AGE with markers of cellular injury were verified using linear regression, where markers of cellular injury were the dependent variables, and SR-DM and AGE the independent variables, controlling the analyses for age of death and sex.

Comparisons between AGE values among participants with and without SR-DM were performed using the Mann-Whitney test. For regression analysis, BRAAK stage and Thal scoring were grouped into high and low (low: 0–2, high ≥3). AGE-ELISA results were log-transformed due to exponential distribution. Effects of tissue preservation were assessed by correlation analysis with tissue pH and postmortem interval.

All tests were 2-tailed, with a p value of <0.05 regarded as significant.

RESULTS

Cohort Details

Of the 99 cases in this study, 63% were female, the mean age of death was 85.6 years. Dementia was present in 61% of the cohort. Dementia status could not be defined for 2 participants. These were included for those analyses when not investigating dementia outcome. The distribution between the different dementia status and the different types of predominant pathology is presented in Table 1. In 3 participants, SR-DM data were not available. For the remaining 96 individuals, SR-DM was present in 15%. The logistic regression to verify if SR-DM was associated with dementia, AD and vascular pathology is shown in Table 2. The distribution of dementia, AD and vascular neuropathology is illustrated in Supplementary Table S1 (for all 99 cases). For these analyses, Braak stage and Thal phase were grouped into high and low (low: 0–2, high: ≥3). SR-DM was not significantly associated with dementia status or neuropathology.

TABLE 1.

Distribution of Dementia Status and Predominant Neuropathology

n = 97	No Dementia With No Significant Vascular Pathology	No Dementia With Significant Vascular Pathology	Dementia With No Significant Vascular Pathology	Dementia With Significant Vascular Pathology
No dementia with no significant AD pathology	14	17	0	0
No dementia with significant AD pathology	2	5	0	0
Dementia with no significant AD pathology	0	0	9	17
Dementia with significant AD pathology	0	0	8	25

TABLE 2.

Relationship Between SR-DM, Dementia Status, and Neuropathology

	OR	95% CI (OR)	p
Dementia	0.30	(0.08–1.12)	0.074
BRAAK	0.65	(0.17–2.42)	0.518
Thal scoring	0.74	(0.23–2.39)	0.611
Vascular disease	0.43	(0.13–1.39)	0.158

SR-DM, self-reported diabetes mellitus.

AGE Measured by ELISA Increases at Higher Thal Phases and Local NFT Score

We used ELISA to quantify AGE in the temporal cortex (AGE-ELISA: mean 0.062 [SD 0.050], median 0.044 [IQR 0.032 − 0.074]), as this provides a more quantitative, linear assessment of a protein than immunohistochemistry, which does not provide linear quantification of protein content. Suitable tissue was available for 79 cases of our cohort, with exclusion based only on poor protein recovery or lack of suitable frozen material. The variation of AGE in the population showed a positively skewed distribution (Kolmogorov-Smirnov p < 0.001; Fig. 1A). There was no correlation with tissue pH (r_s=−0.07, p = 0.546), but there was a very weak positive correlation with PMI (r_s=0.274, p = 0.018).

FIGURE 1.

AGE measured by ELISA. (A) Histogram of AGE in the cohort shows a positively skewed distribution. (B) Boxplot showing variation in AGE with Thal phase (p = 0.03) with an increase in group 3 (phases 4 and 5, p = 0.03). (C) Variation in AGE between temporal cortex neurofibrillary tangle score groups, showing an increase in cases with moderate-severe tangles (p = 0.03). Outliers for boxplots are values >1.5 times the interquartile range above the 75th percentile.

In univariate analysis, AGE-ELISA showed variation with Thal phase (KW H = 6.869, 2df, p = 0.032), which was due to increased levels of AGE at the higher Thal phases (p = 0.026, group 3 compared with group 2; Fig. 1B). AGE-ELISA did not vary significantly between Braak group (KW H = 3, 2df p = 0.223). AGE-ELISA did not increase with severity of CAA (r_s = 0.116, p = 0.307). AGE-ELISA did not vary according the number of areas with cortical microinfarcts (KW H = 0.845 3df p = 0.839).

We then investigated whether AGE increased with local measures of AD pathology in the temporal cortex. There was an increase in AGE-ELISA with increasing CERAD NFT score (with moderate-severe combined) (KW H = 6.981 2df p = 0.03; Fig. 1C), but no variation for neuritic plaques and no significant correlations with either Aβ or AT8 percentage area. AGE-ELISA did not vary according to whether CAA was present (n = 36) or absent (n = 43) in the temporal cortex (Mann-Whitney p = 0.93). Only 2 individuals showed microinfarcts in temporal cortex, so the local relationship to this was not assessed.

We additionally performed longitudinal regression to verify if AGE were associated with dementia, AD and vascular pathology, adjusted for sex and age of death (Table 3). AGE expression was not significantly associated with dementia status, or global measures of AD (Braak stage and Thal phase, dichotomized into high and low) and vascular pathology, though the risk for dementia and Braak are quite high for AGE with large error.

TABLE 3.

Relationship Between AGE, Dementia Status, and Neuropathology

	Neuronal AGE Score			Log-AGE-ELISA
	OR	95% CI (OR)	p	OR	95% CI (OR)	p
Dementia	2.09	(0.91– 4.83)	0.083	1.72	(0.74–4.00)	0.208
BRAAK	1.95	(0.78–4.90)	0.153	1.86	(0.66–5.30)	0.243
Thal scoring	1.29	(0.63–2.65)	0.480	2.23	(0.98–5.06)	0.055
Vascular disease	0.87	(0.43–1.76)	0.690	1.38	(0.63–3.03)	0.425

Semiquantitative AGE Expression in Neurons Increases With Higher Burdens of NFT and Neuritic Plaques

An ELISA provides a measure of overall AGE in the cortex, but may mask differences between cell types. We therefore carried out immunohistochemistry for AGE on the temporal cortex (n = 99). Expression of AGE was found in the perikarya and proximal neurites of pyramidal neurons. Antibodies to AGE decorated blood vessels and also astrocytes, where AGE was present in cytoplasm and in the glial processes (Fig. 2). We semiquantified the immuno-expression associated with neurons, astrocytes, and endothelial cells. AGE expression in neurons correlated with glial expression (r_s=0.423, p < 0.001). There were also correlations between AGE expression in glia in cortex and WM (r_s = 0.276, p = 0.006) and between vessels in Cx and WM (r_s = 0.263, p = 0.009). Inter-observer assessment showed good agreement of scoring for cortical neurons (weighted κ=0.52) and glia (weighted κ=0.34).

FIGURE 2.

(A, B) Photomicrographs of immunohistochemistry for AGE showing expression in neurons. Occasional glial cells are also seen (B, arrow); (C) cortical astrocytes (arrows); (D) around a plaque (arrow). Several AGE-positive astrocytes can also be seen. (E) Cortical blood vessels (F) white matter astrocytes and vessels (arrow). Scale bar: 50 μm.

AGE in cortical neurons increased with Braak NFT stage in cortical neurons (KW H = 19.499 2df p < 0.001), with the increase occurring at the isocortical stages (isocortical vs entorhinal p = 0.002; isocortical vs limbic p < 0.002; Fig. 3A). AGE in cortical glia also increased with Braak NFT stage (KW H = 6.614 p = 0.037), with the increase again being at the isocortical stages (isocortical vs limbic p = 0.031; Fig. 3B). In contrast, WM AGE and vascular AGE scores did not change with Braak NFT stage. These measures did not vary significant with Thal phase.

FIGURE 3.

Boxplots of semiquantitative assessment of cell-type associated AGE and measures of AD pathology. (A) AGE score for cortical neurons between Braak NFT groups, (B) AGE score on cortical astrocytes between Braak NFT groups, (C) AGE score on cortical neurons between CERAD neuritic plaque groups, (D) AGE score on cortical neurons between CERAD NFT groups. Outliers for boxplots are values >1.5 times the interquartile range above the 75th percentile or below the 25th percentile.

Neuronal AGE expression score increased with the local cortical CERAD neuritic plaque score (KW H = 8.754 2df p = 0.013), with the increase being for the highest plaque score group (moderate-severe vs no plaques p = 0.009; Fig. 3C). Neuronal AGE expression score also increased with the local cortical CERAD NFT score (KW H = 12.776 2df p = 0.002), with the increase being for the highest NFT score group (moderate-severe vs no NFT p = 0.003; moderate-severe vs mild NFT p = 0.003; Fig. 3D). No significant relationships were found to percentage area immunoreactivities of Aβ or τ.

Measures of Diabetes Do Not Correlate With AGE Immunoreactivity in the Temporal Cortex

There were no significant relationships between SR-DM with neuronal AGE expression (no SR-DM: median 1, SR-DM present median 1, z = −0.76, p = 0.446) and AGE-ELISA (no SR-DM: median −3.18, SR-DM median −2.93, z 0.13, p=0.900). AGE-ELISA showed no significant correlation with HbA1c (r = 0.36, p = 0.150) on a limited number of cases (n = 17) for which HbA1c data was available.

AGE and Markers of Neuroinflammation and Injury

To further define how AGE might drive brain cell injury, we sought to determine whether increased expression was associated with markers of neuroinflammation (GFAP, CD68, MHC II) and a DNA damage response (γH2Ax). These markers did not vary according to SR-DM (Supplementary Table S2). We performed linear regression to test if AGE was associated with markers of cellular injury. For these analyses, AGE expression (assessed as immunohistochemical neuronal score and by ELISA) was investigated adjusting for sex and age at death. Cortical AGE did not show variation with these markers, except for a negative association between neuronal AGE score (but not AGE-ELISA) with γH2Ax (p = 0.002, Supplementary Table S3). Neuronal (r_s = 0.342; p = 0.001) and glial (r_s = 0.257; p = 0.014) scores for AGE (but not ELISA-AGE) correlated with albumin score but not fibrinogen suggesting a relationship to blood-brain barrier changes.

DISCUSSION

We investigated the expression of AGE using immunohistochemistry and ELISA in a well-characterized aging population-derived sample from the CFAS neuropathology cohort. AGE was expressed in neurons, astrocytes, and endothelium and increased with several measures of AD pathology. Overall temporal cortex AGE, measured by ELISA, was associated with higher NFT score and with increasing Thal Aβ phase. AGE immunohistochemistry scores in neurons and glia were associated with higher Braak NFT stages and greater CERAD scores for NFT and neuritic plaques. However, AGE did not show an association with dementia at death in the cohort. SR-DM was present in 15% of the cohort, but was not associated with brain AGE levels, and AGE did not correlate with Hb-A1c levels.

AGE, Diabetes, and AD Pathology

The variation in AGE expression in this population-sample was positively skewed. AGE was detected even at low Braak stages on cells without protein pathology, suggesting that AGE may contribute to brain cell dysfunction in the aging population independently of AD pathology. Although we did not demonstrate an independent relationship to dementia, this cannot be excluded as the sample size was relatively small. Contrary to our expectations, we did not find that AGE increased with SR-DM. This may reflect limitations in assessing diabetes and numbers with SR-DM, so our data do not exclude the possibility that cerebral AGE expression could be a good biomarker of the cumulative effects of DM on the brain. Other factors, however, may relate to AGE expression.

AGE expression has previously been associated with AD pathology. Increased AGE was found in case-control material in association with older age and with increased Braak NFT stage (51). AGE have been found by immunohistochemistry in association with plaques, CAA, and NFTs (52), and AGE are implicated in AD pathogenesis (53). Using a more quantitative ELISA method, as well as immunohistochemistry with semiquantification for cell-specific expression, we now show that AGE expression increases with AD neuropathologic change in a population derived cohort that includes a spectrum of aging-brain pathology. The relationship to AD neuropathology progression was found with several different AD neuropathy measures related to both Aβ and tau cellular pathologies, and we found that AGE increased particularly at the higher, later stages of pathology.

AGE have pleiotropic effects mediated via binding to their receptor, RAGE, a member of the immunoglobulin superfamily (54). AGE cause vascular damage and are expressed in neurons and glia, where they induce oxidative stress, neuroinflammation, interact with Aβ and tau and thus may exacerbate AD pathology (28). In cell culture, AGE formation affects both amyloid precursor protein and tau phosphorylation (55). Glycation of Aβ increases its toxicity, with upregulation of RAGE and activation of glycogen synthase kinase-3 (which phosphorylates tau) (56), so that AGE formation may contribute to NFT formation (57). Binding of AGE to APOE may play a role in plaque pathogenesis (58). AGE have also been shown to provide a mitogenic signal potentially driving aberrant cell cycle re-entry, thought to be a mechanism important in neuronal death, in AD (59). AGE formation may therefore provide a link between abnormal glucose metabolism and exacerbation of AD pathology. However, AGE formation may result from endogenous as well as exogenous sources, so that mechanisms such as oxidative stress and glutathione depletion resulting from AD molecular pathology may contribute to cellular AGE formation (28, 53, 59, 60). It is not possible from postmortem studies to demonstrate the direction of causality in the relationship between AGE and AD and there may be bidirectional or circular influences. The fact that we show consistently, with several measures, that AGE increase at higher stages of AD pathology may favor, at least some, AGE increase being secondary to AD pathology.

We did not demonstrate relationships to neuroinflammation (gliosis, MHC upregulation) or oxidative stress in this cohort. However, we did find that neuronal and glial AGE increased with increasing score for albumin, suggesting that they are associated with blood-brain barrier impairment, while AGE have been found to damage the blood-brain barrier in vitro (61). Unexpectedly, AGE expression was negatively associated with the DNA damage marker γH2Ax. It may be that, given the relatively weak association, this is a spurious association. However, we previously found that γH2Ax tended to decline with AD progression (62), so the apparent negative correlation between AGE and γH2Ax might be a confounding effect of the AD relationship.

Study Strengths and Limitations

Strengths of this study included the population-representative nature of the cohort, allowing an assessment of relationships of AGE to dementia and pathology measures without the biases inherent in case preselection on clinical criteria and the potential for ceiling-floor effects (33). AGEs were assessed by both immunohistochemistry, allowing study of relationships of expression on specific cell-types, and by ELISA providing a more quantitative assessment of expression. The study was limited in terms of the size of the cohort examined and the use of single brain area. Dementia, although defined using a validated algorithmic approach, was assessed as a binary variable so that mild cognitive impairment and variation in neuropsychological types of dementia were not defined. Only 15% of the cohort had SR-DM, and quantifying the cumulative “burden” of DM is inherently difficult (30), so that the definition of SR-DM was binary. In addition to the self-reported history, we had only a single-time-point measure of peripheral glycation, HbA1_c, in a small subset of cases, and there were no measures to allow assessment of the relationship to subclinical burdens of abnormal glycation or insulin resistance, such as the metabolic syndrome.

Concluding Comments and Translational Applications

Given the variation of AGE formation in the aging brain and the potential range of mechanisms through which AGE may affect brain cell function, defining the effects of AGE on brain cells remains an important objective to understand the effects of metabolic derangements on brain cell health. AGE formation is likely to be important as a mechanism for impairing brain cell function in aging and in AD. It may provide an important integral of long-time glycemic status, although endogenous generation of AGE independently of systemic glycemic status may occur. AGE formation is important as a potential mechanism but also in terms of pathology prevention, treatment target, and the recent suggestion that it may be an important biomarker (60, 63, 64). Defining the relationship of brain AGE formation and central glycemic status to peripheral glucose metabolism and insulin resistance therefore is an important question, requiring further definition.

This project was funded by the Alzheimer’s Society (AS-PG-14-015). C.J.G. was also funded by an Alzheimer’s Society fellowship. Work in the individual CFAS centers is supported by the UK NIHR Biomedical Research Centre for Ageing and Age—awarded to Newcastle-upon-Tyne Hospitals Foundation Trust; Cambridge Brain Bank supported by the NIHR Cambridge Biomedical Research Centre; Nottingham University Hospitals NHS Trust; University of Sheffield, Sheffield Teaching Hospitals NHS Foundation Trust and the Sheffield NIHR Biomedical Research Centre; The Thomas Willis Oxford Brain Collection, supported by the Oxford Biomedical Research Centre; The Walton Centre NHS Foundation Trust, Liverpool.

The authors have no duality or conflicts of interest to declare.

Supplementary Data can be found at academic.oup.com/jnen.

ACKNOWLEDGMENTS

We would like to thank the Alzheimer’s Society lay monitors for their support and advice during the project. We would like to acknowledge the essential contribution of the liaison officers, the general practitioners, their staff, and nursing and residential home staff. We are grateful to our respondents and their families for their generous gift to medical research, which has made this study possible.