Introduction

Epidemiological studies indicate that Type 2 diabetes (T2D) is associated with an increased risk of dementia1,2,3,4,5. Clinical studies using cross-sectional designs support this association by showing that cognition is worse in patients with T2D as compared to matched controls without T2D6, 7. Furthermore, studies of structural magnetic resonance imaging (MRI) show that in T2D cognitive impairment is associated with greater levels of vascular lesions as well as with brain atrophy6,7,8,9,10. Prospective MRI studies also show that in T2D brain atrophy occurs at faster rates than in normal ageing11, 12, suggesting that T2D accelerates neurodegeneration.

Animal studies provide additional evidence to show that inducing T2D/insulin resistance (IR) can promote the pathological changes characteristic of Alzheimer’s disease (AD), specifically accumulation of amyloid-beta (Aβ) and tau (see review ref. 13). These studies also implicate common inflammatory or oxidative stress pathways that link these two chronic diseases of ageing (see review ref. 14). However, the association between cognition and AD pathology in human studies and the stage of AD progression where IR has greater impact remains unclear. Evidence to date using Positron emission tomography (PET) studies have so far been inconclusive.

In AD, PET studies of cerebral glucose metabolism (18F-deoxyglucose PET: FDG-PET) and Aβ deposition (e.g. C11-Pittsburg compound B-PET: PiB-PET) show that reduced neuronal glucose metabolism and increased levels of neocortical Aβ accumulation are features that occur early in the disease (see review ref. 15). A small number of cohort studies have investigated the relationship between these imaging markers of AD and T2D. A cross-sectional study in the population-based Mayo Clinic Study of Ageing assessed cerebral glucose metabolism (FDG-PET) and Aβ deposition (PiB-PET) in healthy older and cognitively normal (CN) adults and older people with T2D. The study showed that compared to the controls, those with T2D displayed cerebral hypometabolism, particularly in those regions severely affected in AD, but no differences in neocortical amyloid load16. In the Baltimore, longitudinal study of Ageing (BLSA), no association was observed between measures of peripheral IR or glucose tolerance and neocortical Aβ load (in a PiB-PET scanned cohort) or other pathological features of AD in post-mortem brain17. More recently in the Alzheimer’s Disease Neuroimaging Initiative (ADNI) study, no association was shown betweenT2D and accumulation of neocortical Aβ load (PiB-PET) or increases in CSF Aβ4210. Instead, T2D was associated with lower cortical thickness an increase in CSF total tau (T-Tau) and phosphorylated tau (P-tau). Together, these findings suggest that IR/T2D is not associated with cerebral accumulation of Aβ but with other hallmarks of the disease. However, in a recent cross-sectional study, Aβ deposition was associated with a higher Homeostatic Model Assessment of IR (HOMA-IR) in late-middle aged, normoglycaemic cognitively normal participants18. The conflicting results in the current literature may be in part due to demographic differences in populations (e.g. age, clinical staging of disease and disease progression) and study design. Further, recent meta-analyses suggest that sex can also mediate T2D associations with dementias and associated co-morbidities, such as stroke19,20,21. Thus, the relationship between IR and clinical and pathological features of the early stages of AD, and sex specific effects, requires further study, particularly in normoglycaemic individuals and prior to the onset of cognitive impairment.

In addition to IR, β-cell hyperactivity and dysfunction and subsequent hyperinsulinemia also contribute to hyperglycaemia and T2D (see review ref. 22). Further, recent evidence indicates β-cell dysfunction in AD rodent models23, 24 and that Aβ and Tau have been shown to accumulate in human post-mortem pancreatic tissue in T2D25, possibly contributing to β-cell dysfunction. Despite this evidence, there is a lack of literature investigating pancreatic β-cell activity (HOMA-B) on cognition and AD related pathology.

The overall aim of this study was to investigate if assessments of IR (HOMA-IR) or pancreatic β-cell function (HOMA-B) are altered across diagnostic groups and ascertain their associations with pathological and clinical expressions of AD in the well characterised Australian Imaging Biomarker and Lifestyle (AIBL) study. We hypothesised that HOMA-IR and HOMA-B are altered across diagnostic groups and are associated with poorer cognitive performance and higher burdens of neuroimaging/CSF biomarker load in cognitively normal participants.

Results

Clinical and cognitive descriptive data for the diagnostic groups [cognitively normal adults (CN), mild cognitive impairment (MCI) and AD] are presented in Tables 1 and 2, respectively. A significant difference in age and frequency of the Apolipoprotein E (APOE) ε4 allele was observed across all diagnostic groups (Table 1), with AD and MCI being older and having a significantly higher APOE ε4 frequency compared to CN (p < 0.001). Significant differences between groups for pathological features and cognitive measures confirmed a clear differentiation between diagnostic groups (Table 2).

Table 1 Baseline demographic and clinical pathology data for clinical classifications across the cohort. All data is presented as means ± standard deviations or %, where indicated. CN, cognitive normal; MCI, Mild cognitively impaired; AD, Alzheimer’s disease; †HOMA indices underwent Box-Cox transformations prior to analysis.
Full size table
Table 2 Baseline AD-related phenotypic data and cognitive descriptive statistics for clinical classifications across the cohort. All data is presented as means ± standard deviations. CN, cognitive normal; MCI, Mild cognitively impaired; AD, Alzheimer’s disease; VEM, Verbal Episodic Memory; ViEM, Visual Episodic Memory; EF, Executive Function; LANG, Language; GLOBAL, Global cognitive composite; HV, Intracranial volume corrected Hippocampal Volume; NAB, Neocortical Amyloid Burden; SUVR-BeCKeT, a transformation of native SUVR into PiB-like SUVR.
Full size table

Relationships between T2D markers and clinical diagnosis

Serum glucose levels were increased significantly in AD and MCI compared with CN (p = 0.014) (AD > MCI > CN), and a non-significant trend was observed for serum insulin HOMA-IR, but not for HOMA-B (Table 1). These group differences in the HOMA indices became statistically significant after co-varying for BMI, sex, diabetes, use of diabetes medication, smoking, age and APOE genotype. HOMA-IR (Fig. 1A) was significantly increased across diagnostic groups (ANCOVA, F = 8.656, p < 0.001) with post-hoc analysis showing a significant increase in HOMA-IR in the AD group (Bonferroni p < 0.001) compared to CN. A clinical group difference was also observed for HOMA-B (Fig. 1B; F = 4.564, p = 0.011), though between the MCI and CN groups (Bonferroni p = 0.028). In both cases, significant differences were only observed in females (HOMA-IR, F = 6.989 p = 0.001; HOMA-B, F = 5.575 p = 0.004) but not males (HOMA-IR, F = 2.603 p = 0.075; HOMA-B, F = 0.911 p = 0.403). In females HOMA-IR was still only observed to be different in the AD group (Bonferroni p = 0.001) compared to CN, whilst HOMA-B, was significantly lower in CN compared to both the MCI (Bonferroni p = 0.041) and AD (Bonferroni p = 0.017) groups.

Figure 1
figure 1

HOMA-IR (A) and HOMA-B (B) at baseline within the clinical classifications of AIBL: *HOMA-IR and HOMA-B represented as Estimated Marginal Means ( ± SEM) of Box-Cox transformed raw data. Univariate analysis was performed covarying for BMI, sex, %diabetes, %diabetes medication, %hypertension, smoking, age and APOE ε4 (HOMA-IR, F = 8.656, p < 0.001; HOMA-B, F = 4.564, p = 0.011). Presented p-values are calculated using Post-hoc Bonferoni analysis.

Full size image

HOMA-IR is associated with cognitive performance and CSF biomarkers

A significant inverse relationships were observed (Table 3) between HOMA-IR and the cognitive composites, verbal episodic memory (β = −0.65, p = 0.010), executive functioning (β = −0.48, p = 0.046) and global composite (β = −0.68, p = 0.007). Significant positive relationships were also observed with CSF T-tau (β = 830.2, p = 0.008) and P-tau (β = 95.9, p = 0.014). Increases in HOMA-B were only observed to be associated with reductions in executive functioning (β = −0.095, p = 0.044) and the global composite (β = −0.011, p = 0.043). Stratification by sex (Table 4, female; Table 5

Table 3 Relationship between HOMA indices and cognitive composites and neuroimaging/CSF biomarkers in cognitively normal older adults.
Full size table
Table 4 Relationship between HOMA indices and cognitive composites and neuroimaging/CSF biomarkers in cognitively normal older adults (FEMALES).
Full size table
Table 5 Relationship between HOMA indices and cognitive composites and neuroimaging/CSF biomarkers in cognitively normal older adults (MALES).
Full size table

, male) revealed that prior significant associations held between HOMA-IR and verbal episodic memory (β = −0.63, p = 0.046), executive function (β = −0.61, p = 0.042), the global cognitive composite (β = −0.79, p = 0.014) and both CSF T-tau (β = 639.5, p = 0.048) and P-Tau (β = 93.4, p = 0.031) in females. No significant associations were observed for either HOMA-IR or HOMA-B in males.

Discussion

Our findings demonstrate modest yet significant differences, in both HOMA-IR and HOMA-B, between the clinical classifications of AD, MCI and CN within the AIBL cohort, after covarying for potential confounding variables. Within clinical classifications, compared to the CN group, the AD group had higher HOMA-IR. These findings are consistent with previous reports showing that the prevalence of IR is greater in MCI/AD patients than controls26,27,28. However, in the current study HOMA-B was increased only in the MCI group, suggesting an increase in β-cell function/insulin secretion in this group, an observation consistent with the observed trend towards increasing absolute insulin levels across groups. This may represent a response to control increasing glucose levels during disease progression29, which was also observed in this study. Overall, these initial findings suggested that increased β -cell function/insulin secretion is associated with cognitive impairment, at least in non-demented adults. However, in the absence of overt cognitive impairment (i.e. CN group), compared to HOMA-IR, HOMA-B had no or weak associations with functioning of cognitive domains. This suggests that changes in insulin sensitivity is the stronger, earlier contributor to impairments in cognition. Longitudinal analysis in the cognitively normal that do or do not show clinical disease progression may provide further support for this notion.

We also observed sex differences in levels of HOMA-IR and HOMA-B between diagnostic groups, where upon stratification by sex, the increases observed in the MCI or AD groups were only observed in females. These findings are consistent with outcomes from meta-analyses which indicate women with T2D are at higher risk of stroke and dementia compared to men with T2D19,20,21 and may be a consequence of several factors. For example, studies in different ethnic groups have suggested that age related increases in the prevalence of metabolic syndrome are greater in women then in men (see review ref. 30). Similarly, impaired glucose tolerance has been reported to be more prevalent in older women than men, although impaired fasting glycaemia more prevalent in men31. Further, changes in hormonal status also contribute to an age-associated increased prevalence of metabolic syndrome in women32 and may be driven by androgen/oestrogen imbalances during the peri-menopausal period68. As part of the clinical pathology assessment, fasting plasma insulin (FPI) and fasting plasma glucose levels (FPG) were assessed. The stated reference ranges are the ranges established in the clinical pathology laboratory in accordance with the national guidelines (http://www.nata.asn.au/, http://www.health.gov.au/npaac). The FPI and FPG were used to calculate values of Homeostatic modelling assessment (HOMA). The HOMA model is often used in cross-sectional and longitudinal studies to estimate insulin sensitivity and pancreatic β-cell functioning as alternatives to more direct assessments such as glucose clam** or acute insulin response, which are not practical in large cohort studies (see review ref. 70). Initially, comparisons of two methods of HOMA was performed; the HOMA1 (“original method”) using equations developed by Mathews and colleagues71 and the HOMA2 “the computer model”) developed by Levy and colleagues72. These models and the differences between them have been extensively discussed elsewhere70. HOMA1 was calculated using the following: HOMA-IR = (FPI x FPG)/22.5; HOMA-B (%) = (20x FPI)/(FPG = 3.5) to estimate IR and β-cell functioning respectively. HOMA2 was calculated from a Microsoft Excel macro accessed via the Oxford University website (www.dtu.ox.ac.uk/homacalculator/). Pearson’s correlation analysis of the data generated from HOMA1 and HOMA2 analysis revealed a strong significant correlation between these indices (HOMA1-IR vs HOMA2-IR, r = 0.976, p < 0.001; HOMA1-B vs HOMA2 B, r = 0.717, p < 0.0001), indicating suitability of both methods for this data set. However, HOMA1 was utilised as it is the most commonly used method in large cohort cross-sectional or longitudinal studies. All reference to HOMA in this report reflects HOMA1 calculated indices.

DNA was extracted and APOE genotype determined previously described73. Briefly, QIAamp DNA Blood Maxi Kits (Qiagen, Hilden, Germany) were used per manufacturer’s instructions to extract from whole blood. APOE genotype was determined from two separate TaqMan® (Thermo Fisher Scientific, Waltham, MA) genoty** assays for the single nucleotide polymoprhisms rs7412 (assay ID: C____904973_10) and rs429358 (assay ID: C___3084793_20). TaqMan® genoty** assays were performed on a QuantStudio 12 K Flex™ Real-Time-PCR systems (Thermo Fisher Scientific, Waltham, MA) using the TaqMan® GTXpress™ Master Mix (Thermo Fisher Scientific, Waltham, MA) methodology as per manufacturer instructions. APOE carrier status was defined by the presence (1 or 2 copies; ε4 + ) or absence (0 copies; ε4−) of the APOE ε4 allele.

CSF collection and Aβ42, T-tau, and P-tau181P quantitation were performed as previously described74. Briefly, approximately 10–14 ml of CSF was collected in the morning by routine lumbar puncture after overnight fasting directly into one 15 ml polypropylene tube (Greiner Bio-One188271), employing a protocol like that recommended by the Alzheimer’s Biomarkers Standardization Initiative75. All CSF samples for analysis were taken from aliquots prepared and stored as previously described74 and thawed at time of assay. All CSF samples were analyzed in duplicate using the enzyme-linked immunosorbent assay (ELISA): INNOTEST Aβ-amyloid (1–42; Aβ42), INNOTEST hTAU Ag (T-tau), and INNOTEST Phospho-tau (P-tau; 181 P; P-tau181P) (Innogenetics, Ghent, Belgium) per published standard methods. Mean intra-assay coefficients of variation for these assays are as previously published74. This study reports on data from 66 study participants from whom CSF was taken at the baseline time point of the AIBL study.

Brain Imaging

Data was available at baseline from a total of 379 AIBL participants (262 CN, 69 MCI, 48 AD) who underwent Aβ-amyloid imaging with positron emission tomography using either 11C-Pittsburgh Compound B (PiB), 18F-florebetapir or 18F-Flutemetamol as described elsewhere76,77,78. PET standardized uptake value ratios (SUVR) were determined for all tracers using CapAIBL, a web based freely availably MR-less methodology79. Briefly, SUVs were summed and normalized to either the cerebellar cortex SUV (PiB), whole cerebellum SUV (florbetapir, FBP) or pons SUV (flutemetamol, FLUTE) to yield the target-region to reference-region SUVR. To allow for the analysis of tracer specific SUVRs as a single continuous variable, a linear regression transformation, termed the “Before the Centiloid Kernel Transformation” (BeCKeT) scale, was applied to FBP and FLUTE SUVR to generate PiB-like SUVR units80.

Hippocampal Volume data was available for 319 (229 CN, 52 MCI, 38 AD) participants at baseline. Hippocampal volumes were determined through MRI, parameters of which have been previously described81. Briefly, participants underwent T1 weighted MRI using the ADNI 3-dimensional (3D) Magnetization Prepared Rapid Gradient Echo (MPRAGE) sequence on 1.5 T or 3 T scanners. Hippocampal volume was calculated after correcting for age in years and intracranial volume (sum of grey matter, white matter and cerebrospinal fluid volumes), as previously described82.

Statistical analysis

All statistical analysis was conducted using IBM SPSS Statistics (version 23; IBM Corp, Armonk, NY, USA) with the level of significance set to α = 0.05 (two-tailed). All variables were assessed for conformation to a normal distribution. Box-Cox transformations was used to correct variables departing from a normal distribution83. For all variables, except HOMA indices, the calculated lambda (λ) equated to no transformation. HOMA indices underwent transformations (T(Y)) prior to analysis, specifically: T(Y) = ((Y + 1)−0.079−1)/−0.079 and T(Y) = (Y0.04−1)/0.04 for HOMA-IR and HOMA-B, respectively (with Y representing the HOMA index). Analysis of demographic variables was undertaken using a One-way Analysis of Variance (ANOVA) to determine differences in continuous variables across clinical classifications differences in categorical variables determined using the χ2-test. Differences in HOMA indices between clinical classifications were assessed using an analysis of covariance (ANCOVA) via a General Linear Model (GLM) with Bonferroni correction. Associations between HOMA indices and cognitive composites and pathological brain changes in CN were assessed using linear regression analysis. Both ANCOVA and linear regression analyses, for all dependent variables, covaried for body mass index (BMI), diabetes (yes/no), diabetes medication (yes/no), hypertension (yes/no), smoking (yes/no), and APOE genotype (ε4+/ε4−), with sex and age only covaried for with biomarker dependent variables.

Data availability

All data and samples used in this study are derived from the Australian Imaging, Biomarkers and Lifestyle (AIBL) Study of Ageing. All AIBL data, and that specific to this study, is publically accessible to all interested parties through an Expression of Interest procedure and is governed by the AIBL Data Use Agreement, for more information please see https://aibl.csiro.au/awd/.