eTable 1. Effect of glucose, insulin, and insulin resistance on brain AD pathology using continuous mixed model analysis
eTable 2. Effect of glucose, insulin, and insulin resistance values on brain 11C-PiB binding using continuous mixed-model analysis
eTable 3. Effect of glucose, insulin, and insulin resistance on brain AD pathology using only data from midlife OGTT
Thambisetty M, Metter EJ, Yang A, Dolan H, Marano C, Zonderman AB, Troncoso JC, Zhou Y, Wong DF, Ferrucci L, Egan J, Resnick SM, O’Brien RJ. Glucose Intolerance, Insulin Resistance, and Pathological Features of Alzheimer Disease in the Baltimore Longitudinal Study of Aging. JAMA Neurol. 2013;70(9):1167-1172. doi:10.1001/jamaneurol.2013.284
Copyright 2013 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.
Peripheral glucose homeostasis has been implicated in the pathogenesis of Alzheimer disease (AD). The relationship among diabetes mellitus, insulin, and AD is an important area of investigation. However, whether cognitive impairment seen in those with diabetes is mediated by excess pathological features of AD or other related abnormalities, such as vascular disease, remains unclear.
To investigate the association between serial measures of glucose intolerance and insulin resistance and in vivo brain β-amyloid burden, measured with carbon 11–labeled Pittsburgh Compound B (11C-PiB), and AD pathology at autopsy.
Scores calculated from the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) and Braak criteria were correlated with measures of hyperglycemia, hyperinsulinemia, glucose intolerance, and insulin resistance in 197 participants who underwent autopsy after death and who had undergone 2 or more oral glucose tolerance tests (OGTT) using grouped analyses and a continuous mixed-models analysis. The same measures of glucose intolerance and insulin resistance were also correlated with brain 11C-PiB retention in an additional 53 living subjects from the Baltimore Longitudinal Study of Aging neuroimaging study.
Prospective, serially assessed cohort of community-dwelling subjects.
Cohort 1 consisted of 197 participants enrolled in the Baltimore Longitudinal Study of Aging who had 2 or more OGTTs during life and a complete brain autopsy after death. Cohort 2 consisted of 53 living subjects who had 2 or more OGTTs and underwent brain 11C-PiB positron emission tomography.
Autopsy and 11C-PiB positron emission tomography.
Main Outcomes and Measures
The correlation of brain markers of AD, including CERAD score, Braak score, and 11C-PiB retention, with serum markers of glucose homeostasis using grouped and continuous mixed-models analyses.
We found no significant correlations between measures of brain AD pathology or 11C-PiB β-amyloid load and glucose intolerance or insulin resistance in subjects who had a mean (SD) of 6.4 (3.2) OGTTs during 22.1 (8.0) years of follow-up. Thirty subjects with frank diabetes mellitus who received medications also had AD pathology scores that were similar to those of the cohort as a whole.
Conclusions and Relevance
In this prospective cohort with multiple assessments of glucose intolerance and insulin resistance, measures of glucose and insulin homeostasis are not associated with AD pathology and likely play little role in AD pathogenesis. Long-term therapeutic trials are important to elucidate this issue.
Glucose intolerance and diabetes mellitus are proposed risk factors for the development of Alzheimer disease (AD), but evidence of this assertion is not consistent. Some studies have shown excess cognitive impairment and lower cognitive performance in subjects with diabetes, impaired glucose tolerance, and insulin resistance,1- 4 especially in those with poor control of glucose levels,5 whereas others have not.6,7 In the Rotterdam Study, initial work indicated that diabetes was associated with a 2-fold increase in the risk of a clinical diagnosis of dementia and AD,8 although subsequent findings did not confirm the initial report.9 The Religious Orders Study10 and the Mayo Clinic Alzheimer Registry11 reported an increased risk of AD in subjects with diabetes, and the recent Hisayama Study12 correlated an increased risk of developing AD with 2-hour postload glucose values. These studies, in combination with recent evidence that long-term intranasal insulin administration during a 4-month period improved cognitive function in patients with mild cognitive impairment and early AD,13 have established that the relationship among diabetes, insulin values, and AD is an important area of investigation.14 Whether cognitive impairment seen in patients with diabetes is mediated by excess pathological features of AD or other related abnormalities, such as vascular disease, remains unclear.15 In addition, few studies have addressed the relationships between longitudinal changes in measures of glucose tolerance and/or insulin resistance and AD pathology or brain β-amyloid (Aβ) burden.
The Baltimore Longitudinal Study of Aging (BLSA) is a prospective longitudinal cohort study of the effects of aging, including effects on cognition and dementia. Embedded within the BLSA are an autopsy study16 and a neuroimaging study.17 The intensity of the evaluations in the BLSA, including periodic oral glucose tolerance testing (OGTT), makes it an ideal study in which to examine the effects of glucose intolerance and insulin resistance on brain Aβ accumulation and neurofibrillary tangle formation. We report herein that no significant association exists between glucose intolerance, diabetes, or insulin resistance and pathological measures of AD or positron emission tomography detection of brain Aβ accumulation in this well-characterized sample.
The BLSA autopsy program consists of 579 participants from the main BLSA cohort who agreed to undergo postmortem brain examinations. The rate of dementia in the autopsy cohort is similar to that in the BLSA cohort as a whole.18 Two hundred thirty-two participants 69 years or older at the time of death who were cognitively and neurologically normal at entry into the study have died and undergone brain autopsy. Of this group, 6 participants were excluded because their cognitive deficits could be clearly attributed to non-neurodegenerative processes. Of the remaining 226 participants, 197 had undergone 2 or more OGTTs. None of the subjects used any medications to treat diabetes mellitus at the time of the OGTT. Sixteen of these participants eventually used medications to control their diabetes. An additional 14 participants from the autopsy cohort did not have any OGTT because they were taking medications to control diabetes. Of the 197 participants with OGTTs, 133 were men. Participants were predominantly white (94.7%), with a mean (SD) duration of education of 17.1 (4.3) years. The mean (SD) age at death was 88.3 (7.3) years (Table 1).
Fifty-three participants in the BLSA imaging study who were 69 years or older and had 2 or more OGTTs underwent in vivo assessment of fibrillar Aβ levels using carbon 11–labeled Pittsburgh Compound B (11C-PiB). Seven of these patients have died and undergone autopsies and are thus also part of the autopsy group. The mean (SD) age at the 11C-PiB scan was 79.2 (5.8) years. Thirty participants were male, with a mean (SD) duration of education of 16.8 (2.4) years.
Dynamic 11C-PiB positron emission tomography studies were performed as described previously.19 Volumetric T1-weighted magnetic resonance images were acquired coincident with the positron emission tomography and were coregistered to the positron emission tomography images using the mutual information method in the Statistical Parametric Mapping software (SPM 2; Wellcome Department of Imaging Neuroscience). Cerebellar gray matter was used as a reference region. Distribution volume ratio (DVR) parametric images of the 11C-PiB signal were estimated by simultaneous fitting of a simplified reference tissue model using linear regression with spatial constraints.20 We calculated the mean cortical DVR by averaging values from the orbitofrontal, prefrontal, superior frontal, parietal, lateral temporal, occipital, and anterior and posterior cingulate regions. We also calculated DVR values from 2 other regions of interest, the posterior cingulate/precuneus and medial temporal lobe, as described previously.17,20
Evaluations included neuropsychological testing, neurological examination, interval medical history, medication review, and a structured informant and subject interview.18 A diagnosis of dementia was made at a consensus conference18 with the conferees blinded to the pathological results. Most of the participants were examined annually after age 70 years, although approximately 25% of the cohort had gaps in their follow-up of several years’ duration. Studies of this cohort are conducted under the auspices of The Johns Hopkins School of Medicine and the MedStar Health Research Institute institutional review boards, and all participants provided written informed consent.
Oral glucose tolerance testing has been performed by the BLSA since 1959, and the details of this testing have been published.21 All patients in this study had at least 2 OGTTs during their annual or semiannual evaluations, and some had as many as 12. No differences in glucose levels or AD pathology could be discerned between participants with larger or smaller numbers of OGTTs. The variation in OGTT number per participant likely reflected age at entry into the study and administrative issues within the BLSA. All participants started fasting at 8 pm and received the glucose load between 7 and 8 am the following day. Blood samples were drawn at baseline and at 20, 40, 60, 80, 100, and 120 minutes. Plasma insulin levels were measured using radioimmunoassay in a subset of the OGTT. The lower limit of detection for this assay is 15 pmol/mL.22 Measures of fasting and 120-minute postload insulin resistance were calculated using the homeostasis model assessment (HOMA) technique.23
Neuritic plaques were scored in 4 brain regions (the superior/middle temporal gyrus, medial frontal lobe, inferior parietal cortex, and orbitofrontal cortex) as described.18 Pathological features indicative of AD were examined on silver stains performed according to Hirano’s modification of the Bielschowsky method and graded according to the Braak criteria24 and the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD).25 For CERAD scoring, we determined the maximum neuritic plaque score seen in all cortical regions examined (peak CERAD score) and the mean of the CERAD scores for all cortical regions examined (mean CERAD score). We translated CERAD scores (0, A, B, and C) into a linear numeric scale (0, 1, 2, and 3) to allow statistical analysis. In addition, we generated a composite AD pathology score by summing the CERAD and Braak scores in equal measure. For this latter analysis, CERAD scores were divided into the following 3 groups: 1 indicates 0 or mild neuritic plaques; 2, moderate neuritic plaques; and 3, frequent neuritic plaques. Braak scores were divided into the following 3 groups: 1 indicates Braak stages 0, I, and II; 2, Braak stages III and IV; and 3, Braak stages V and VI. The sum of the modified Braak and CERAD scores yielded a composite score ranging from 2 to 6. A previous study18 has shown this scale to be a very useful method to quantitate the combined pathological effects of Aβ and tau on cognition.
We compared AD pathology, 11C-PiB retention in the brain, and serum glucose, insulin, and HOMA values using grouped analyses (analysis of variance) and a continuous analysis with linear mixed models26 to accommodate the longitudinal nature of the data. For the grouped analyses, we divided the pathological cohort into 3 different categories of glucose and insulin homeostasis values to maximize differences between the groups with the most normal and most abnormal findings and still retain a sufficient group size for statistical purposes. For the 11C-PiB cohort, we only used 2 groups owing to the small number of subjects. For the continuous analyses, we used AD pathology or brain 11C-PiB retention as independent variables and age at death (or age at 11C-PiB scan) and sex as covariates. An analysis of the effect of the rate of change of insulin, glucose, and HOMA values during a lifetime on AD pathology at death was performed by dividing the 197 participants into groups with low (score of 2 or 3 [n = 75]), medium (score of 4 [n= 58]), and high (score of 5 or 6 [n = 64]) composite AD pathology scores at death. Rates of change of glucose, insulin, and HOMA values across multiple tests were compared in the 3 groups using mixed-models analyses, controlling for sex and age at death. We used commercially available software (SAS, version 9.1; SAS Institute, Inc) for all analyses.
A total of 197 BLSA autopsy program participants had more than 1 OGTT. Of these, 186 had insulin levels measured at least once during the OGTT. The characteristics of these participants are given in Table 1. Participants were divided into 3 equal-sized groups based on mean lifetime fasting or 120-minute glucose, insulin, or insulin resistance values (Table 2). Insulin resistance was calculated using the HOMA technique. None of the characteristics detailed in Table 1 differ significantly among any of these groups, including the number of OGTTs, the years covered, or the rate of dementia (not shown). No significant differences in any measure of AD pathology was seen between groups stratified on the basis of glucose, insulin, or insulin resistance values, even when the mean fasting and 120-minute glucose, insulin, and HOMA values taken from multiple OGTTs covering a mean (SD) period of 22.1 (8.0) years before death were markedly different among the groups (Table 2).
The continuous mixed-models analyses (rather than a grouped analysis) of the relationship between glucose, insulin, or insulin resistance values and AD pathology and controlling for sex and age at death showed no significant association between AD pathology and any measure of glucose or insulin homeostasis (Supplement [eTable 1]). An analysis based on the rates of change of fasting and 120-minute postload glucose, insulin, and HOMA values during the lifetime of the participants and using linear mixed models also showed no differences in participants with low, medium, or high AD pathology scores (data not shown).
When we divided the 197 participants into those with dementia (n = 101) and those without dementia (n = 96), we also detected no significant differences (by analysis of variance) in the mean nondementia/dementia values for fasting glucose (100/98 mg/dL), fasting insulin (9.6/9.0 μIU/mL), fasting HOMA (2.4/2.2), 120-minute glucose (156/149 mg/dL), 120-minute insulin (63/57 μIU/mL), and 120-minute HOMA (25/22). (To convert glucose to millimoles per liter, multiply by 0.0555; to convert insulin to picomoles per liter, multiply by 6.945.)
Using a second cohort of 53 subjects undergoing periodic 11C-PiB scanning (7 of whom were also in the autopsy series), we evaluated the relationship between glucose, insulin, or insulin resistance values and brain Aβ accumulation as assessed by mean cortical 11C-PiB DVR. As given in Table 3, when the 53 subjects were divided into 2 equally sized groups based on mean lifetime fasting or 120-minute glucose, insulin, or insulin resistance values, no significant difference in mean cortical 11C-PiB retention was seen. Moreover, when we compared the top third of subjects, based on mean cortical 11C-PiB scores (mean 11C-PiB DVR, 1.49), with those in the lowest third (mean 11C-PiB DVR, 0.93), no significant difference in any measure of fasting or 120-minute postload glucose metabolism or insulin resistance was seen (data not shown). When we analyzed the relationship between lifetime fasting or 120-minute glucose, insulin, or insulin resistance values and 11C-PiB retention using a continuous mixed-model analysis (rather than a grouped analysis) and controlling for age at 11C-PiB scan and sex, no significant association was seen using mean cortical 11C-PiB DVR or using 11C-PiB DVR scores for the posterior cingulate/precuneus or medial temporal lobe, 2 areas where fibrillar Aβ is deposited early in the course of AD (Supplement [eTable 2]).
To examine whether our negative results were skewed by glucose and insulin values obtained late in life, we analyzed the relationship of glucose, insulin, and insulin resistance with AD pathology and Aβ 11C-PiB binding, limiting the analysis to glucose and insulin values obtained at the baseline OGTT (mean [SD] age, 66.3 [11.1] years; 197 subjects) or performed before age 70 years (60.3 [8.4] years; 105 subjects) using a mixed-model analysis adjusting for sex and age at death (Supplement [eTable 3]). Again, no significant relationship was seen. Furthermore, the cohort with 11C-PiB scans had their initial OGTT at a relatively young mean age of 53.3 years (Table 1).
Finally, we analyzed data from participants in the BLSA autopsy program who were taking medications to treat diabetes (n = 30). This group included 16 participants from the 197 in this study who were eventually prescribed medications to lower glucose levels and another 14 subjects in the BLSA autopsy program who never had an OGTT because they were already taking the medications. Of these 30 participants, 9 were taking insulin therapy as part of their regimen. As given in Table 4, we found no significant difference in any measure of AD pathology, whether or not the participant was taking the medication .
Our study found no association between lifetime measures of glucose homeostasis and standard measures of AD pathology or cortical fibrillar Aβ deposition measured with 11C-PiB. Similarly, we did not find an association between a clinical diagnosis of dementia and hyperglycemia or hyperinsulinemia. Our results concur with other studies that found no association between diabetes mellitus and AD pathology,27,28 and we extend these observations more broadly to hyperglycemia and insulin resistance. The strengths of our study are its prospective nature, the large number of well-characterized participants, and the multiple OGTTs obtained more than 20 years before death. These serial assessments allow us to determine the effect of prolonged burdens of hyperglycemia and insulin resistance on pathological findings in the brain. Although glucose and insulin levels measured even earlier in life than those in the present study might better predict AD pathology or dementia, our analyses of OGTTs performed at younger ages also revealed no association with AD pathology or 11C-PiB retention, suggesting that glucose intolerance does not affect AD pathology even at its earliest stages.
Limitations of the present study include the method for determining insulin resistance, which is calculated rather than determined by an insulin clamp procedure,29 and our pathological assessments of AD, which are semiquantitative rather than quantitative and do not include immunostaining for Aβ and tau. In addition, participants in our study are a sample of convenience, not an epidemiologically representative group, and have access to high-quality health care. Thus, they are unlikely to have experienced prolonged periods of severe hyperglycemia. In view of a recent study suggesting that cognitive changes in diabetes may be related to the severity of prolonged hyperglycemia,5 we may have underrepresented that population. Our results disagree in part with the study of Matsuzaki and colleagues,30 who found differences in 2-hour postload glucose and HOMA values in subjects with no AD pathology at autopsy compared with those who had any degree of AD pathology at death. Differences in cohort size, composition, and intensity of testing may account for these disparate findings.
Our data certainly do not preclude insulin therapy (or endogenous insulin) from having a beneficial effect on cognition independent of its effects on AD pathology. Moreover, the effects of insulin may vary in subjects with and without dementia.31 In our cohort, the number of participants taking exogenous insulin was small, limiting any direct test of the hypothesis. Second, the effect of insulin on the brain is complex and not confined to Aβ precursor protein processing and Aβ production. Such effects include growth factor regulation, gene transcription, and protection against oxidation.32,33 Moreover, insulin may have effects downstream of Aβ deposition. Finally, insulin resistance within the brain itself may not correlate with peripheral measures of insulin resistance.34 Given that AD is likely more than just Aβ accumulation,35 long-term therapeutic trials are important to elucidate this issue.
Corresponding Author: Richard J. O’Brien, MD, PhD, Department of Neurology, Johns Hopkins Bayview Medical Center, Mason Lord Center Tower, Ste 5100, Baltimore, MD 21224 (firstname.lastname@example.org).
Accepted for Publication: February 1, 2013.
Published Online: July 29, 2013. doi:10.1001/jamaneurol.2013.284.
Author Contributions:Study concept and design: Thambisetty, Yang, Zonderman, Resnick, O’Brien.
Acquisition of data: Metter, Yang, Dolan, Zonderman, Troncoso, Wong, Egan, Resnick, O’Brien.
Analysis and interpretation of data: Thambisetty, Metter, Yang, Marano, Zonderman, Zhou, Wong, Ferrucci, Resnick, O’Brien.
Drafting of the manuscript: Thambisetty, Yang, Dolan, Marano, Zhou, O’Brien.
Critical revision of the manuscript for important intellectual content: Thambisetty, Metter, Zonderman, Troncoso, Wong, Ferrucci, Egan, Resnick, O’Brien.
Statistical analysis: Thambisetty, Yang, Zonderman, Zhou, Ferrucci, O’Brien.
Obtained funding: Troncoso, Egan, Resnick, O’Brien.
Administrative, technical, and material support: Metter, Dolan, Marano, Zonderman, Troncoso, Wong, Egan, Resnick, O’Brien.
Study supervision: Thambisetty, Egan, Resnick, O’Brien.
Conflict of Interest Disclosures: None reported.
Funding/Support: This study was supported by grants P50 AG05146 and U01 AG033655 from the National Institute on Aging (NIA); by the Burroughs Wellcome Fund for Translational Research; and by the Intramural Research Program, NIA, National Institutes of Health.