Associations Between Glucose Metabolism Measures and Amyloid-β and Tau Load on PET 14 Years Later: Findings From the Framingham Heart Study

Alzheimer disease (AD) is common in the elderly and is leading to an increasing societal burden worldwide (1). AD is characterized by aggregation of brain amyloid-β plaques up to 20 years before disease onset, and tau tangles in a later stage. These biomarkers can, in vivo, be detected by positron emission tomography (PET) scans (2). People carrying the APOE ε4 allele are known to be at greater risk for developing AD (3).

Type 2 diabetes is highly prevalent from midlife onward and is a risk factor for AD dementia (4). For both AD and type 2 diabetes, common pathways such as impaired brain glucose metabolism (5,6) and brain insulin resistance (7,8) have been suggested. In the Framingham Heart Study (FHS), type 2 diabetes and peripheral plasma glucose have also previously been associated with cognitive decline (9) and AD dementia (10). To date, research on the association between glucose metabolism and amyloid-β and tau brain pathology later in life remains scarce, even though this is highly relevant for improving diagnostics and treatments in people with type 2 diabetes and AD.

Several studies have assessed the cross-sectional relationship of glucose metabolism measures with amyloid-β and tau pathology in older populations. Most of these studies did not find any association between type 2 diabetes and glucose metabolism and amyloid-β and tau on PET (6,11,12). The association of diabetes and glucose metabolism in early adulthood or middle-aged people with future AD pathophysiology is, however, less well investigated. Typically, age of onset is associated with the impact of type 2 diabetes (13), and might have a greater impact on amyloid-β and tau accumulation when already present earlier in life. The few longitudinal studies that have been conducted thus far report associations of diabetes and elevated plasma glucose (14) as well as insulin resistance (15) with more abnormal amyloid-β PET later in life, while other studies do not find associations (16–18). Yet, previous studies did not investigate associations between type 2 diabetes and tau later in life. Neither did previous studies investigate the influence of age, sex, or APOE ε4 carriership on these associations, while these factors are known to be associated with AD. Knowledge on how impaired glucose metabolism in early adulthood and middle-aged people is related to both amyloid-β and tau pathology later in life will improve our understanding of AD etiology and can improve early diagnosis and prognosis of patients, as well as AD prevention strategies.

Our study aims to assess the association of glucose metabolism measures, that is, fasting plasma glucose, plasma insulin, and insulin resistance, with amyloid-β and tau load 14 years later on PET scans. We also assessed the potential influence of age, sex, and APOE ε4 carriership on this association. The study is performed in FHS, a relatively healthy and younger population cohort.

We included 288 participants (age range 20–70 years) from the Offspring (n = 53) and Generation 3 (n = 235) cohorts of the FHS study. Participants were selected when having at least one glucose metabolism measure (i.e., fasting plasma glucose or fasting plasma insulin) available, as well as an amyloid-β (n = 279) and/or tau (n = 240) measure on PET scans 14 years later (range 11–17 years). None of the participants had a diagnosis of dementia at the time of the PET scan. Dementia diagnosis was based on the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (19), as evaluated by an adjudication panel including at least one neuropsychologist and one neurologist (20). We excluded participants on insulin treatment, because of the direct impact on insulin and glucose metabolism (n = 2). The inclusion of participants is shown in a flowchart in Supplementary Fig. 1.

FHS is an ongoing population cohort study in the town of Framingham, MA. Since 1971 and 2002, Offspring and Generation 3 participants, respectively, have undergone health exams, on average, every 4 years to collect clinical measures from early adulthood to death. Detailed testing protocols of the cohorts are described elsewhere (21,22). Plasma glucose and insulin measures 14 years before PET corresponded to health exam 8 from the Offspring cohort and exam 1 from the Generation 3 cohort, measured from 2003 to 2005. Of all Offspring and Generation 3 participants, a randomly selected subset underwent PET scans in 2016–2019 (23,24). Hence, the selected participants showed similar characteristics as participants in the overall FHS cohorts (Supplementary Table 1). Participants with amyloid-β versus tau data largely overlapped (79.9%) and showed similar demographic characteristics. High level of education was defined as associate degree, bachelor’s degree, or higher. All participants in FHS agreed on the use of their data and signed informed consent conforming to the standard FHS protocols, as approved by the institutional review boards of Boston University Medical Center (Boston, MA) and Massachusetts General Hospital (Boston, MA).

We used continuous measures of plasma glucose and insulin levels, as these are more direct measures of diabetes severity and are shown to impact vascular and brain health much earlier than at the levels for a clinical diabetes diagnosis (25). One-time measures of fasting plasma glucose and fasting plasma insulin (fasted ≥8 h) were obtained at an average of 14 years (range 11–17 years) before the PET scans. Plasma measures were conducted, stored, and processed according to the standard FHS protocol for each cohort. Laboratory measure protocols are described elsewhere (9,10).

Besides using plasma glucose as continuous measure, we categorized plasma glucose into two groups: normal glucose levels, serving as controls, and elevated plasma glucose. Elevated plasma glucose was defined as having a fasting plasma glucose level ≥100 mg/dL, a clinical standard commonly used to define prediabetes (26).

As for the insulin measures, ELISA was used for measures in the FHS Generation 3 cohort, and Roche immunoassay was used for the FHS Offspring cohort. We calculated Z scores within each cohort to be able to compare and combine the insulin measures for both cohorts. Next, we calculated insulin resistance by using the homeostatic model assessment for insulin resistance (HOMA-IR) index formula fasting plasma glucose (mmol/L) × fasting plasma insulin (μU/L)/22.5.

History of diabetes diagnosis (any type) was additionally reported to characterize our cohort and was based on either medical history (by self-report), a fasting plasma glucose level ≥125 mg/dL (at the current or past measure), or the use of glucose-lowering medication.

Metabolic comorbidities were assessed, as they may impact associations between glucose metabolism measures and AD pathology. Data on total cholesterol (TC), blood pressure, and BMI were available (10). Systolic and diastolic blood pressure (SBP/DBP) were measured as the mean of two independent measures. Mean arterial blood pressure (MAP) was calculated by DBP +1/3(SBP – DBP).

The six apolipoprotein E genotypes (ε2/ε2, ε2/ε3, ε2/ε4, ε3/ε3, ε3/ε4, ε4/ε4) were determined in all participants. Participants had to give additional consent to share their genetic data for research. Participants were classified as APOE ε4 carriers when carrying one or more ε4 alleles.

Amyloid-β PET was performed in 279 participants and tau PET scans were performed in 240 participants (of which 231 participants also have an amyloid-β PET scan). Both amyloid-β and tau PET scans were completed at Massachusetts General Hospital using a GE Discovery MI PET/CT or Siemens ECAT EXACT HR+ scanner and 11C-Pittsburgh Compound B (PiB) and 18F-Flortaucipir (FTP) tracers for amyloid-β and tau, respectively. Scans were performed at the same day for the vast majority of participants (95.1%; maximum 3 months difference). Dynamic PiB PET data were collected for 60 min postinjection, and the data were reconstructed into 39 frames. Distribution volume ratio (DVR) images were created using frames corresponding to 40–60 min postinjection using cerebellum cortex as the reference region. FTP PET data were acquired 75–105 min postinjection, data were reconstructed into 5-min frames, and frame data were summed. Standardized uptake value ratios (SUVRs) were calculated for FTP data using FreeSurfer-defined cerebellum cortex as the reference region. More detailed acquisition protocols in the FHS cohort have been described earlier (23,24).

Structural T1-weighted brain MRIs at the time point closest to PET were coregistered to the PiB and FTP images using SPM12, and the resulting coregistration parameters were applied to the aparc+aseg FreeSurfer labels, processed with FreeSurfer version 7.1. Mean PiB DVR and FTP SUVRs were then extracted from each aparc+aseg region. Details per PET marker are described below.

Global composite values were calculated using the volume-weighted average across frontal, anterior cingulate cortex/posterior cingulate cortex, lateral parietal, and lateral temporal regions (27). To describe baseline study characteristics, we used a global amyloid-β DVR cutoff ≥1.0067 to define amyloid-β abnormality, as computed by Gaussian mixture modeling (27). However, continuous DVRs were used for analysis.

As the temporal cortex is described to be most sensitive to early tau changes (23,28), a mean volume-weighted temporal composite score was computed for tau SUVRs, across the entorhinal cortex, amygdala, parahippocampal gyrus, fusiform gyrus, inferior middle temporal gyrus, and middle temporal gyrus. To describe baseline characteristics, temporal tau abnormality was defined by a cutoff of mean SUVR ≥ 1.2241, as defined by 2 SDs above the mean of individuals with abnormal amyloid-β.

Differences in demographics between the cohorts and between the PET subsample and total sample of both cohorts were tested by t tests for continuous variables and χ² tests for categorical variables. As main analysis, multiple linear regression was used to test associations between each glucose metabolism measure (plasma glucose, elevated plasma glucose, plasma insulin, HOMA-IR) as predictor and global amyloid-β DVR and temporal tau PET SUVR (model 1) after 14 years (range 11–17 years) as outcome measures. We adjusted for age, sex, and interval in years to PET in all analyses. We added APOE ε4 carriership (model 2), or BMI, MAP, and TC (model 3) as covariates to our main model in subsequent analyses. The analyses were repeated, with participants with diabetes (of any type) excluded, to assess the robustness of the main findings. Main analyses were additionally performed with false discovery rate (FDR) correction for multiple comparisons.

We additionally explored interactions of the glucose metabolism measures with age, sex, and APOE ε4 carriership when associations between a glucose metabolism measure and AD biomarker measure were significant (after FDR correction). To assess whether the association of glucose metabolism with tau was dependent on amyloid-β, we tested interactions of all glucose metabolism measures × global amyloid-β DVR on temporal tau SUVR.

To assess whether similar results were found with a shorter time interval to PET, we repeated the main analyses with glucose metabolism measures 8 years (range 5–10 years) before PET. As sensitivity analyses, we assessed amyloid-β and tau associations for plasma glucose quartiles and explored results for consecutive elevated plasma glucose (at 14 + 8 years and 14 + 8 + 0 years before PET). Methods for the glucose metabolism measures at 8 and 0 years before PET are described in Supplementary Material. We also explored whether exclusion of individuals using glucose-lowering medication influenced the results. As the inclusion of subsamples of two separate cohorts could have resulted in selection bias, we additionally repeated the main analyses in a propensity-weighted mixed model (in which weights are calculated based on the existing differences in age and years before PET between the two cohorts) with cohort as random factor. All analyses were performed using R3.3.0+ (RStudio, Inc.) with P < 0.05 as level of significance.

We included 288 participants with a mean age of 43.2 ± 10.7 (mean ± SD) at time of glucose metabolism assessment. Almost half were female (49%), most of them were highly educated (69%), and 23% were carriers of the APOE ε4 allele. Mean amyloid-β DVR (mean ± SD) was 0.94 ± 0.08, and mean tau SUVR was 1.11 ± 0.07. At 14 years before the PET scan, 8 participants (2.8%) had a diagnosed case of diabetes (any type), while 69 participants (24%) showed elevated plasma glucose levels (≥100 mg/dL). Demographics are further described in Table 1. The demographics for each cohort separately are described in Supplementary Table 2.

Associations between glucose metabolism measures and global amyloid-β and temporal tau measures on PET 14 years later (range 11–17 years) are reported in Table 2. We found that lower plasma insulin levels and lower HOMA-IR index were associated with increased amyloid-β load 14 years later (B [95% CI] = −0.01 [−0.02 to −0.00], P = 0.035, and B [95% CI] = −0.01 [−0.02 to −0.00], P = 0.034, respectively) (Supplementary Fig. 2). However, after FDR correction for multiple comparisons, these associations did not remain statistically significant (both P = 0.07). In addition, we found that elevated plasma glucose was associated with higher tau load 14 years later (B [95% CI] = 0.03 [0.01–0.05], P = 0.006). This association remained statistically significant after FDR correction (P = 0.02) and when adjusting for amyloid-β load (B [95% CI] = 0.03 [0.00–0.05], P = 0.02). Glucose metabolism measures and tau were not moderated by amyloid-β load (Supplementary Table 3). No associations between plasma glucose and amyloid-β load or insulin/HOMA-IR and tau load were found.

Next, we adjusted for other factors that may influence the association between glucose metabolism and AD biomarkers. Results largely remained similar after adjusting for APOE ε4 (model 2) or BMI, TC, and MAP (model 3), as summarized in Supplementary Table 4. The association of plasma insulin and HOMA-IR and amyloid-β 14 years later (range 11–17 years) did not remain significant when correcting for APOE ε4 carriership (P = 0.057 and P = 0.052, respectively).

Only an association between elevated plasma glucose and tau load was found after FDR correction. For this association, we explored the impact of age, sex, and APOE ε4 carriership on the relationship of glucose metabolism measures with global amyloid-β and temporal tau load 14 years later. The identified association between elevated plasma glucose and higher tau load showed an interaction with APOE ε4 (B [95% CI] = −0.08 [−0.12 to −0.03], P = 0.001) (Fig. 1), indicating that the association was observed in non-carriers only. The association was not dependent on age (B [95% CI] = 0.001 [−0.001 to 0.004], P = 0.34) or sex (B [95% CI] = 0.004 [−0.04 to 0.05], P = 0.87).

To assess whether glucose metabolism measures were associated with PET measures after a shorter time interval, we repeated the main analyses for glucose measures taken 8 years before PET. Demographics at 8 years before PET are presented in Supplementary Table 5. No associations were found between any of the glucose metabolism measures and amyloid-β or tau load 8 years later (Supplementary Table 6). This might indicate that associations of glucose metabolism are only observed in the long term.

We explored associations of plasma glucose quartiles with amyloid-β and tau. In line with our main models, we did not find any association of plasma glucose quartiles with amyloid-β. Regarding tau, only the highest quartile was associated with tau, and showed higher temporal tau than the third (B [95% CI] = 0.03 [0.00–0.05], P = 0.02) and second (B [95% CI] = 0.03 [0.00–0.05], P = 0.02) quartiles, but did not reach statistical difference compared with the lowest quartile (B [95% CI] = 0.02 [0.00–0.05], P = 0.10). Other quartiles did not differ from each other (Supplementary Fig. 3).

We also assessed whether the association between elevated plasma glucose and tau remained consistent for consecutive measures of elevated plasma glucose. The association between elevated plasma glucose (≥100 mg/dL) and temporal tau load remained significant for participants with elevated plasma glucose consecutively at 14 and 8 years before PET (B [95% CI] = 0.03 [0.002–0.05], P = 0.035, n = 47) as well as at 14, 8, and 0 years before PET (B [95% CI] = 0.03 [0.01–0.06], P = 0.016, n = 38). Associations between consecutive elevated plasma glucose and amyloid-β remained nonsignificant.

When excluding participants using glucose-lowering medication (n = 10), the association of higher insulin (B [95% CI] = −0.01 [−0.02 to −0.001], P = 0.03) and HOMA-IR (B [95% CI] = −0.01 [−0.02 to −0.001], P = 0.02) with decreased global amyloid-β load remained similar. Associations between elevated plasma glucose and temporal tau load also remained similar (both B [95% CI] = 0.03 [0.01–0.05], P = 0.007).

To adjust for potential differences between the two cohorts, we repeated the main models using mixed models with propensity weighting. We found similar results as compared with our main analyses (Supplementary Table 7).

The aim of our study was to assess associations between glucose metabolism measures and amyloid-β and tau pathology on PET 14 years later. We found that elevated plasma glucose was associated with greater tau load in APOE ε4 non-carriers but not with amyloid-β load. This suggests that glucose metabolism in early adulthood and middle-aged people is associated with tau neuropathology later in life, which is possibly unrelated to AD. Plasma insulin and insulin resistance were not associated with amyloid-β or tau load after correction for multiple comparisons.

Our study demonstrates an association of elevated plasma glucose with future tau uptake but not with future amyloid-β uptake on PET. Previous population studies similarly did not find an association between glucose metabolism and future amyloid-β on PET (16,29). To our knowledge, this is the first study to assess the association between glucose metabolism and future tau on PET. We found that elevated plasma glucose in early adulthood and middle-aged people is associated with increased tau load on PET 14 years later, independent from amyloid-β load. These results remained stable when considering consecutive measures of elevated plasma glucose. Previous studies only assessed the association of diabetes status at the time of PET with tau and did not find associations (18,30,31).

The current findings are in line with our recent meta-analysis showing that glucose metabolism measures and diabetes status are associated with increased tau biomarkers but not amyloid-β biomarkers (32). As amyloid-β typically aggregates before tau in AD, one would not expect to find only associations with tau if glucose metabolism were related to AD. Neurodegenerative effects of diabetes may be independent and possibly additive to those of AD and driven by pathways that promote neuronal tau more than amyloid-β (33). Therefore, our observed association with increased tau may not be related to AD but could, for instance, reflect primary age-related tau (33,34). We also found that the association between elevated plasma glucose and tau was only present in APOE ε4 non-carriers. As carriers of the APOE ε4 allele are known to be at greater risk for AD, this finding strengthens the hypothesis that elevated plasma glucose might be an alternative pathway to tau pathology independent from AD.

However, we did not find associations of insulin or HOMA-IR with increased AD pathology. Earlier studies have indicated differential associations of plasma glucose and insulin with AD pathology (35). They even suggest that lower plasma insulin may be associated with insulin deficiency in the brain, contributing to amyloid-β pathology (35,36). This hypothesis, however, requires further exploration in longitudinal studies with larger sample sizes.

It also remains contradictive that we did not find an association between plasma glucose and tau when measured continuously. Our quartile analyses suggest that plasma glucose is only associated with increased tau when ∼≥99 mg/dL. As this entails only one-fourth of our tau sample (n = 57), this may explain why we do not see an association for the continuous measure in the overall group.

We found associations of glucose metabolism measures with amyloid-β and tau at 14 years but not 8 years. This could indicate a greater impact of impaired glucose metabolism when longer present, with small amounts of brain pathology slowly aggregating over time. This could require a longer time interval than 8 years to be detected by PET. Moreover, larger heterogeneity and comorbidities at older ages make it more challenging to detect smaller associations.

FHS is a unique and extensively characterized longitudinal population cohort. To our knowledge, this is the first population study to assess glucose metabolism in relation to both amyloid-β and tau PET later in life. Despite these strengths, some limitations need to be addressed. The FHS cohort is highly educated, mostly White, and relatively healthy, with better health than the average U.S. population. The selected subset that underwent PET scanning was relatively young and had better vascular health as compared with the full FHS sample, possibly decreasing the generalizability of our study. This could have contributed to the relatively small effect sizes in this study. Associations might be more evident in older populations with higher prevalence of (pre)diabetes and should be studied further. While our PET study had a sample size comparable to other PET studies in the field, our sample size might have led to reduced statistical power. However, we were still able to identify associations that remained stable across sensitivity analyses. Moreover, we could not account for changes in (vascular) health conditions during the 14 years of follow-up and were not able to investigate amyloid-β and tau accumulation at baseline. However, given the relatively young and healthy population, we do not expect much pathology to be present 14 years before PET. In addition, one-time glucose metabolism measures may be subject to the variation of the day the plasma was drawn. To validate our findings, future research should employ larger sample sizes, repeated glucose metabolism measures over time, and a higher prevalence of vascular conditions and AD pathology.

Another limitation of the study is that we used two cohorts of different generations, with significant differences in age and years before PET, different insulin assays, and, potentially, other factors. In our main analyses, we could not adjust for cohort due to its multicollinearity with age, by which potential cultural or general differences between the cohorts might have been neglected. However, propensity weight mixed models that considered cohort as random factor suggested similar results.

Our study suggests that elevated plasma glucose in early adulthood and middle-aged people is associated with greater future tau but not amyloid-β pathology. This knowledge helps in understanding how impaired glucose metabolism relates to brain pathologies later in life and thereby provides relevant knowledge for prognostics and prevention strategies to improve future health care. The link of diabetes and glucose metabolism with tau, independent from amyloid-β, as well as how the APOE ε4 allele might play a role in this association, needs further exploration. Future research in larger populations with higher prevalence of (pre)diabetes should explore associations with other neurodegenerative pathways to further examine how diabetes and glucose metabolism are related to cognitive decline.

Acknowledgments. The authors thank Boston University FHS team members for the opportunity to work with the FHS data and the great collaboration between Maastricht University and Boston University within this project.

Funding. This study was funded by Alzheimer Nederland, under grant number WE.08-2022-12. The study was additionally partly supported by the European Union’s Horizon 2020 research and innovation program under grant agreement no. 847879 (Prevention and Remediation of Insulin Multimorbidity in Europe) and by Stichting Adriana van Rinsum-Ponsen. FHS was funded by multiple sources: the National Heart, Lung, and Blood Institute (NHLBI)-FHS contract (75N92019D00031), and other National Institutes of Health grants, among others, including NHLBI contracts (N01HC25195, HHSN268201500001I), and grants from the National Institute on Aging (NIA) (AG008122, AG016495, AG068753, R01AG033040, R01AG049607, R01AG016495, RF1AG062109, RF1AG072654, U19AG068753). Other funding grants included NIA K99AG071837 and Alzheimer’s Association AARFD-21-849349.

Duality of Interest. W.J.J. receives research support from Biogen. S.J.B.V. receives research support from the European Platform for Neurodegenerative Diseases (EPND) project, which received funding from the European Commission, Innovative Medicines Initiative (IMI) 2 Joint Undertaking under grant agreement 101034344. The IMI Joint Undertaking receives support from the European Union’s Horizon 2020 research and innovation program and European Federation of Pharmaceutical Industries and Associations (EFPIA). R.A. serves as a scientific advisor to Signant Health and Novo Nordisk and is a consultant to the Davos Alzheimer’s Collaborative, a Swiss Foundation and a U.S.-based 501c3 organization. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. V.v.G. drafted the manuscript and performed all statistical analyses. Q.T. and T.F.A.A. helped prepare the data set and supported in methodological decisions and statistical analysis. C.B.Y. prepared and preprocessed the PET data. R.A. was responsible for data acquisition and overall study supervision. The current project was executed under supervision of P.J.V., R.A., W.J.J., and S.B.J.V. All co-authors critically reviewed and approved the manuscript. V.v.G. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 2024 Alzheimer's Association International Conference, Philadelphia, PA, 28 July–1 August 2024.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Cheryl A.M. Anderson and Vanita R. Aroda.