Insulin resistance is associated with reductions in specific cognitive domains and increases in CSF tau in cognitively normal adults

Growing evidence supports the hypothesis that type 2 diabetes (T2D) increases the risk of developing dementia. Experimental evidence from mouse models demonstrates that the induction of T2D/insulin resistance (IR) can promote the accumulation of Alzheimer’s disease (AD) pathological features. However, the association of T2D with pathological and clinical phenotypes in humans is unclear. Here we investigate the relationship of indices of IR (HOMA-IR) and pancreatic β-cell function (HOMA-B) with cognitive performance across several domains (Verbal/Visual Episodic Memory, Executive Function, Language and a measure of Global cognition) and AD biomarkers (CSF Aβ42, T-tau/P-tau, hippocampal volume and neocortical Aβ-amyloid burden). We reveal that HOMA-IR (p < 0.001) incrementally increases across diagnostic groups, becoming significantly elevated in the AD group compared with cognitively normal (CN) adults. In CN adults, higher HOMA-IR was associated with poorer performance on measures of verbal episodic memory (p = 0.010), executive function (p = 0.046) and global cognition (p = 0.007), as well as with higher CSF T-tau (p = 0.008) and P-tau (p = 0.014) levels. No association was observed with CSF Aβ or imaging modalities. Together our data suggest that IR may contribute to reduced cognitive performance and the accumulation of CSF tau biomarkers in cognitively normal adults.

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.

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 controls [26][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 progression 29 , 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 T2D 19-21 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 men 31 . Further, changes in hormonal status also contribute to an age-associated increased prevalence of metabolic syndrome in women 32 and may be driven by androgen/oestrogen imbalances during the peri-menopausal period 33 . For example reduction in the neuroprotective effects of oestrogen or increases in gonadotropins may act to promote AD related pathological changes (see review ref. 34). With longitudinal study, the interactions between biological markers of IR and indices of AD related biology, and its clinical expression, would provide evidence to confirm our findings that these differences are stronger in woman than men, whilst also providing an indication of the potential contributions of hormones.
Having shown that increases in IR were associated with clinical stages of AD, we wished to ascertain the associations of HOMA-IR and HOMA-B with pathological and clinical expressions of AD prior to the onset of clinical cognitive impairment. In the CN group, greater HOMA-IR was associated with reductions in global cognition and in two of the four cognitive domains assessed, namely verbal episodic memory and executive functioning. A weaker association was observed between HOMA-B and executive function and global cognition. Our findings are in agreement with previous studies showing similar associations of cognitive functioning with HOMA-IR in cognitively normal individuals from community-based volunteer 35 , population based 36 or family based 37 studies. Episodic memory is one of the earliest cognitive domains that show changes in the pre-clinical stages of AD. Previous studies have observed episodic memory changes occurring 4-8 years prior to executive function and up to 10-years prior to changes in other cognitive domains [38][39][40] . Compared to non-diabetics, verbal memory and executive functioning are also strongly affected in T2D 41,42 . The association of IR with episodic memory suggests that it plays a role at early preclinical stages of the disease process. Stratification by sex revealed these associations were again present in females only. In previous studies, there have been varied findings with respect to the effect of sex on the association between HOMA-IR and cognitive functioning. In a cross-sectional study of cognitively healthy elderly community volunteers, higher HOMA-IR  was associated with lower verbal fluency performance and reduced grey matter in the temporal lobe of both men and women 35 . However, in a large, family based Dutch study of 1898 participants, higher HOMA-IR was associated with poorer executive functioning in women only 37 . A similar association with verbal fluency in women only was shown recently in a population based study of ~6000 adults 36 . This association was independent of a marker of glucose levels (glycated haemoglobin; HbA1c), indicating that IR may have early effects on cognition and maybe detected prior to development of significant hyperglycaemia and T2D 36 . Further, in the same study HOMA-IR was associated with poorer verbal fluency in APOE ε4 non-carriers only. Several factors could account for these mixed results including age, cohort characteristics (family vs population vs community based) and the size of the cohort. For example, those studies that show associations in women only 36, 37 the analysis was undertaken in larger family or population based cohorts or were younger (~45 yrs. and 53 yrs., respectively) compared to our study and that of Benedict and colleagues 35 (71 and 73 yrs., respectively). IR was also associated with biomarkers of AD pathology within the cognitively normal group where increases in HOMA-IR were associated with increases in CSF T-tau and P-tau. No such relationships were observed with neocortical Aβ burden or CSF-Aβ42 in these groups. These findings suggest that, at earlier stages of the disease process there is a stronger relationship with tau rather than Aβ. Cross-sectional cohort studies evaluating imaging modalities (structural MRI, PET) or AD biomarkers have shown similar results. In these studies, lower cortical thickness and cerebral hypometabolism, but not changes in Aβ were associated with T2D 8,16,17 . More recently, in the ADNI study, T2D-related reduction in cortical thickness was associated with increases in CSF P-tau, but not with neocortical Aβ burden or CSF Aβ42, across diagnostic groups of CN, MCI, and AD 10 . Our study was not powered to similarly stratify for T2D as only a small proportion (~8%) of the total cohort had the condition. The association of HOMA-IR with increases in CSF tau, even after controlling for T2D, suggests that IR is an early contributor to this process. However, it is acknowledged that this is a cross-sectional analysis and a longitudinal follow-up is required to further establish IR as a determinant of cognitive decline and accumulation of AD pathological features.
Numerous studies in various animal models have shown that diet (HF diet/sucrose/fructose) or chemical (administration of the β-cell toxin streptozotocin (STZ)) induced diabetes promote, tau phosphorylation, and synaptic/neurodegeneration 23,[43][44][45][46][47] . The potential mechanisms may involve neuroinflammatory and oxidative stress mechanisms 14 . In addition, impaired cerebral insulin signalling, promoted by IR and T2D, plays a key role in this process and in triggering tau hyper phosphorylation through increased activity of GSK-3β, a major protein kinase that phosphorylates tau (see review ref. 13). These in vivo, animal studies also support a role for T2D/IR in promoting Aβ build up in the brain, where availability of Aβ degrading enzymes such as the insulin degrading enzyme is reduced, inflammatory and oxidative processes promoting Aβ accumulation and formation of toxic oligomers further exacerbating synaptic degeneration (see review ref. 13). As discussed above the findings from the current study and others 10,16,17 do not fully support this notion in human studies, but do not rule out the possibility that IR is associated with early deposition of Aβ.
Two recent studies have shown increases in HOMA-IR associated with increases in Aβ 18,48 . The study by Willette et al. 18 , showed that higher HOMA-IR was associated with higher amyloid load (as assessed by PiB-PET) in normoglycaemic participants recruited from the Wisconsin Registry for Alzheimer's disease Prevention (WRAP). The study performed analysis of HOMA-IR with PiB-PET and did not include CSF Aβ42, or CSF T-tau/P-tau levels. The more recent study by Hoscheidt et al. 48 assessed participants with a parental history of dementia due to AD recruited from a similar cohort, the Wisconsin Alzheimer's Disease Research Centre (ADRC), Investigating memory in people at risk, causes and treatments (IMPACT). The study investigated the association of HOMA-IR with CSF AD related biomarkers including Aβ42, CSF sAPPα/sAPPβ and CSF P-tau and showed that increases in HOMA-IR was associated with increased CSF levels of sAPPβ (not sAPPα), and modest association with CSF Aβ42 only. These findings were consistent with IR promoting amyloidogenic processing of APP (i.e. increased sAPPβ) and thus Aβ42 formation. Whilst associations between HOMA-IR and neocortical Aβ burden were not assessed in the latter study, the findings were suggestive of being consistent with that previously observed 18 .
There are number of factors that may account for the differences in the findings in our cohort and the Wisconsin cohorts. Age may have a major role in determining where in the progression of AD pathology, IR has the greatest impact. The participants in the Wisconsin studies were younger (57.7 yrs 49 , and 60 yrs 18 ) compared to our study cohort (CN average age 70 yrs.) and other cohorts where no associations with Aβ-amyloid was reported (~75 yrs. 10 and ~79 yrs. 17 ). Therefore, it is possible that an earlier relationship occurs between IR and Aβ-amyloid burden, but in older and presymptomatic individuals a relationship with Tau exists. Longitudinal studies that incorporate Aβ-amyloid imaging, tau imaging and CSF Aβ/tau biomarkers are required to map out where IR fits into the progression of the disease. However, both relationships may be useful in the possible use of IR as a factor that contributes to predicting the conversion to cognitive impairment from pre-symptomatic stages.
It is also worth noting that the Wisconsin cohorts had a high percentage of participants with a family history of dementia due to AD, and in the Hoescheidt et al. study 49 , analysis was only performed on those with a family history. In addition, 47% of the cohort were APOE ε4 carriers compared to 26.8% in this study and CSF sAPPβ and P-tau/Aβ42 were higher in APOE ε4 carriers. The Willette et al. 18 study also had a higher percentage of APOE ε4 carriers (38.7%). The carriage of APOE ε4 is associated with greater CSF Aβ42 and Aβ-amyloid plaque load 50-53 , a strong predictor of AD pathology and cognitive decline 54,55 . The apoE4 isoform is also thought to have a greater impact on Aβ accumulation, through impairing clearance/degradation [56][57][58][59][60] , but also has been associated with the dysregulation of APP processing to promote Aβ production 61,62 . These characteristics of the Wisconsin cohorts may also explain the greater association of HOMA-IR with Aβ-amyloid pathology compared to what we observe in the AIBL cohort.
Our findings, in CN adults, indicate that increases in IR are associated with reductions in cognition function and elevations in CSF tau. This and other studies suggest that IR may contribute to AD progression, although it is possible that it is AD pathology that gives rise to metabolic dysfunction. For example, evidence from AD post-mortem studies shows brain regions involved in regulating brain and systemic energy metabolism, such as the hypothalamus contain Aβ plaques and Tau tangles (see review ref. 63). Further, FDG-PET studies show lowered hypothalamic glucose metabolism in MCI or AD than in age matched controls. Additionally, animal studies suggest that peripheral Aβ may drive IR and glucose dysregulation and that tau may be important in maintaining glucose homeostasis (see review ref. 64). Our findings that IR is associated with preclinical reductions in cognitive performance and increased CSF tau levels contribute to this understanding of the relationship between IR and AD. These data are also consistent with those from a recent cross-sectional study from the ADNI cohort which observed that T2D-related brain atrophy was associated with increases in CSF P-tau in CN, MCI and AD groups and which led to the conclusion that T2D may promote neurodegeneration independent of AD dementia diagnosis 10 . Our observed associations of increasing IR with reductions in cognition and increases in CSF T-tau and P-tau in CN adults suggests IR may also lead directly to changes in cognition and tau accumulation in early AD. Longitudinal analysis, particularly in controls/MCI participants that convert may provide some insight into the direction of this association. Ideally, these studies would be complemented with PET imaging of relevant brain regions including the hypothalamus. Although, FDG-PET studies have shown reduced glucose utilisation in the AD/MCI hypothalamus 65, 66 studies with Aβ or tau tracers have not been forthcoming, potentially due to the small size of the hypothalamus, limiting spatial resolution, and off-target binding, which may increase the difficulty in assessing this brain region.
There are some limitations to our findings. Firstly, there is the potential that observed findings are currently sample dependent for several reasons; participants were volunteers and not randomly selected from the community, therefore the findings of this study may only be applicable to similar cohorts and the study used cognitive composites derived from specific cognitive assessments, and the use of similar composites derived from different cognitive assessments may result in different findings. Second, the analysis is cross-sectional and informs that there is an association of HOMA indices with cognitive functioning. The findings merit longitudinal analysis to determine if; (1) insulin resistance is associated with the progression of AD pathology and at what stages this relationship is observed (i.e. amyloid at early stages/ tau at later stages) and (2) the contribution of IR/T2D to predicting cognitive impairment. The HOMA indices are calculated from fasting glucose and insulin levels. Although commonly used in association studies, particularly large cohort studies, a closer measure of insulin sensitivity is utilising the euglycaemic-hyperinsulinaemic clamp to quantify insulin secretion and assess metabolism of glucose. Given that this method requires to be done in a clinical setting it would not be feasible in large cohort studies, but once confirmation of a role for IR in the progression of the disease/conversion to cognitive impairment/ AD pathology, studies could be undertaken in a small subset to tease out potential mechanisms. As the focus was associations with IR, our study variables did not include cardiovascular or atherosclerotic factors (i.e. cholesterol, triglycerides), however recent studies investigating the relationship of IR with atherosclerotic factors in AD/MCI patients show that increased insulin/decreased response to insulin are independent predictors for AD and MCI 67 .
We have demonstrated that in the AIBL cohort increases in HOMA-IR are associated with reductions in cognitive domains that show early changes in AD with concomitant increases in CSF tau levels. Longitudinal analysis in this and additional cohorts that incorporate tau imaging would be requried to confirm this association with accumnulation of pathological changes. A similar analysis in individuals at risk will also determine if there are earlier associations with Aβ burden and if T2D contributes to the conversion of pre-sympotmatic to early cogntive impairment (MCI) or MCI to AD. As early IR, prior to significant hyperglycaemia, is modifiable this work has implications in identifying and assessing strategies to reduce/prevent the onset cognitive decline and pathology associated with T2D.

Materials and Methods
Participants. The study reports on data collected from the AIBL study, a prospective longitudinal study of ageing. This cross-sectional study presents baseline data from 1264 participants (905 cognitively normal (CN), 156 mild cognitive impaired (MCI), 203 Alzheimer's disease (AD)). Information regarding AIBL study design, enrolment process, neuropsychological assessments and diagnostic criteria has been previously described 68 . Ethics approval for the AIBL study and all experimental protocols was provided by the ethics committees of Austin Health, St Vincent's Health, Hollywood Private Hospital and Edith Cowan University. All experiments and methods were carried out in accordance with the approved guidelines and regulations and all volunteers gave written informed consent before participating in the study. for age, sex, years of education, premorbid IQ (WAIS-III Full Scale Intelligence Quotient (FSIQ) and depressive symptoms (Geriatric Depression Scale (GDS)) 69 .
Biochemistry, Genotyping and HOMA estimations. Baseline fasted blood samples were taken and fractionated, with one aliquot sent to clinical pathology laboratories in Perth and Melbourne, as described previously 68 . 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 clamping 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 colleagues 71 and the HOMA2 "the computer model") developed by Levy and colleagues 72 . These models and the differences between them have been extensively discussed elsewhere 70 . 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 described 73 . Briefly, QIAamp DNA Blood Maxi Kits (Qiagen, Hilden, Germany) were used per manufacturer's instructions to extract from whole blood.
CSF collection and Aβ42, T-tau, and P-tau 181P quantitation were performed as previously described 74 . 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 Initiative 75 . All CSF samples for analysis were taken from aliquots prepared and stored as previously described 74 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 published 74 . 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 elsewhere [76][77][78] . PET standardized uptake value ratios (SUVR) were determined for all tracers using CapAIBL, a web based freely availably MR-less methodology 79 . 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 units 80 .
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 described 81 . 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 described 82 .
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 distribution 83 . 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) = (Y 0.04 −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 χ 2 -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/.