Task-Induced Brain Activity Patterns in Type 2 Diabetes: A Potential Biomarker for Cognitive Decline

Patients with type 2 diabetes are at increased risk for dementia and cognitive decline (1,2). Neuroimaging is useful for determining structural and functional biomarkers that may predict future cognitive impairment, thereby elucidating the mechanisms involved in cognitive decline and informing strategies to target vulnerable brain regions (3,4).

Recent research has focused on the default mode network (DMN), a set of medial prefrontal and temporo-parietal regions that are most active at rest and deactivated during cognitive tasks. The DMN exhibits both reduced functional connectivity (5) and impaired task-induced deactivation (6) in Alzheimer disease (AD) and mild cognitive impairment (MCI). These changes have been observed before atrophy and cognitive decline and are concurrent with other signals that are known predictors of cognitive decline (6). The DMN also has high metabolic activity, making it susceptible to the effects of type 2 diabetes and hypo- or hyperglycemia (7). Patients with type 2 diabetes show impaired functional connectivity (8) and abnormal glucose metabolism (9) in the DMN, but no study has investigated whether they show reduced ability to suppress DMN activity during a cognitive task.

This study investigates brain activity patterns during encoding and recognition tasks using functional magnetic resonance imaging (fMRI) in middle-aged type 2 diabetic patients, focusing on the DMN. These regions, in particular the prefrontal cortex, are involved in both encoding (10) and recognition (11). Since elderly patients with type 2 diabetes often demonstrate cognitive impairments (12), it is important to identify task-related functional alterations that may precede cognitive decline so that preventative therapies can be introduced in middle age.

We studied 22 patients with type 2 diabetes and 29 nondiabetic healthy control subjects between the ages of 45 and 65 years (Table 1). Patients were group matched on education, IQ, sex, and BMI, and we accounted for a group age difference in our analyses.

We gathered medical history and current medications from medical records and questionnaires. Since insulin resistance was of interest, we excluded patients treated with the insulin-sensitizing medications metformin or thiazolidinediones. Eligible patients were on stable therapy with diet, insulin, or oral medications. After approval from the institutional review boards of the Joslin Diabetes Center and Beth Israel Deaconess Medical Center (where the MRI was performed), subjects provided informed consent and information about psychiatric history, handedness, medical history, current medications, height, and weight. Exclusionary criteria were as follows: 1) stroke or myocardial infarction; 2) unstable or current episode of a DSM-IV Axis I disorder; 3) sleep, eating, or learning disorder; 4) sensorimotor handicap, central nervous system disorder, or illness affecting neurological function; 5) history of substance abuse (including alcohol and excluding nicotine); 6) left-handedness (13); 7) contraindications to magnetic resonance imaging (MRI); and 8) BMI >40 kg/m².

Participants provided a blood sample after an 8-h fast to assess creatinine, HbA_1c, lipids, serum insulin, plasma glucose (PG), and apolipoprotein E (APOE) allele status, as the ε4 isoform confers risk for AD. Medication administration occurred after baseline blood draws. Nondiabetic participants drank 75 g flavored dextrose, and blood samples were taken intravenously at 30, 60, and 120 min postingestion to measure PG and serum insulin to calculate insulin resistance using the homeostasis model assessment of insulin resistance (HOMA-IR). HOMA-IR was not assessed for insulin-treated type 2 diabetic patients because it would not reflect their natural insulin sensitivity. Participants also answered demographic and socioeconomic questionnaires, completed a fitness assessment, and underwent cognitive testing.

We measured executive function (Delis-Kaplan Executive Function System), memory (Rey Auditory Verbal Learning Test; Wechsler Memory Scale-III), intelligence (Wechsler Abbreviated Scale of Intelligence), and psychomotor speed (Grooved Pegboard Test) as described previously (8) (Table 1).

PG measures were taken prior to magnetic resonance scanning for diabetic participants. Treatment was provided for PG <4.4 mmol/L or >16.6 mmol/L (N = 2). Images were collected on a GE Healthcare Signa HDxt 3.0T system (Milwaukee, WI). Participants underwent screening for metal objects, pacemakers, pregnancy, and claustrophobia. Of the 57 eligible participants, 6 were excluded owing to head movement during scanning. Exclusion criteria included translations or rotations of >3 mm in x, y, or z directions.

Functional images were collected in the axial plane using a gradient echo planar imaging sequence sensitive to blood oxygen level dependent (BOLD) contrast (repetition time/echo time = 2,000/25 ms; flip angle = 90°; slice thickness = 4 mm; two-dimensional in-plane: field of view = 24 × 24 cm², matrix size = 64 × 64; voxel size = 3.75 × 3.75 × 4 mm³).

fMRI scans were collected during both the encoding and recognition tasks (5 min each). Stimuli were presented using Presentation software (Neurobehavioral Systems, Albany, CA), projected on a screen visible via mirror attached to head coil. Trials were synchronized with the acquisition sequence. Responses were collected using a four-button Lumina fiberoptic response pad (http://www.cedrus.com/lumina), connected to a personal computer via optical cable interface.

During encoding, subjects saw green and red images and concentrated on the pairing of the object and its color (14). During recognition, subjects indicated via button press the original color of the same images now presented in black and white. Control and rest blocks were used during both tasks. Block order was randomized, and images were presented for 3,000 ms followed by a fixation cross for 1,000 ms.

fMRI image analyses and processing were completed using the Functional MRI of the Brain Software Library (FSL) software package version 5.0.4. Images were registered to the standard MNI152 brain (Montreal Neurological Institute). The first two volumes were deleted for T1 equilibration. Motion and slice time correction, brain extraction, spatial smoothing with Gaussian filter of 6 mm full width at half maximum, linear trend removal, and a temporal high-pass filter with cutoff of 120 s were performed. All image analysts were blind to group status.

Demographic, clinical, and cognitive data were compared between groups using Mann-Whitney U tests for all continuous variables and Fisher exact tests for all categorical variables except sex, for which a Pearson χ² test was used.

For examination of regional brain activation in response to the encoding and recognition tasks, a general linear model was used to compute statistical parametric maps by random-effects multiple regression analysis for each task. Model predictors for BOLD signal time courses were constructed using the boxcar time courses for the task stimulus paradigm convolved with a γ function to account for hemodynamic response. The model for the baseline rest condition predictor was a constant function. Cerebral BOLD activations and deactivations were determined by applying relevant contrast F tests to β-weights for each voxel position (Fig. 1). Activations were computed by contrasting the encoding and recognition blocks to their respective control blocks, and deactivations were compared with their respective rest periods.

To verify hippocampal activation, we created binary masks of the left and right hippocampus using the Harvard-Oxford Subcortical Structural Atlas thresholded at the 50% probability level. We extracted average- and maximal-activation z scores from the hippocampal regions of interest (ROI) using FSL’s fslstats tool.

We investigated the effects of the AD risk allele APOE4 (15,16), current and lifetime average HbA_1c (calculated by averaging 4-year means), and HOMA-IR on BOLD activation and deactivation within each group by performing separate mixed-effects general linear model group analyses in FSL. We also compared BOLD activation and deactivation between diabetic patients with prescan PG ≥11 mmol/L and <11 mmol/L to determine the effects of acute hyperglycemia (Fig. 1). We then performed anatomical ROI analyses using the Harvard-Oxford atlas and FSL’s Featquery tool to examine Spearman correlations between prescan PG and deactivation in DMN regions.

All data are presented as mean ± SD, and statistical tests were conducted using a two-sided α-level of 0.05 with SPSS Statistical Software, version 19.

Table 1 shows demographic and clinical data for diabetic patients and healthy control subjects.

There were no significant differences between groups on any cognitive tests, including vocabulary (a measure of premorbid IQ [17]) and the fMRI task (Table 1).

For both tasks, Fig. 1 shows activated and deactivated regions and Table 2 lists local maxima. Table 3 lists the effects of covariates. APOE4 was evaluated in each group separately, HOMA-IR only in control subjects, and current and lifetime average HbA_1c only in diabetic patients.

During both encoding and recognition, we observed bilateral activation of the hippocampus in diabetic patients (mean z score ± SD, encoding, left 3.89 ± 0.62, right 3.06 ± 1.13; recognition, left 3.48 ± 0.48, right 3.82 ± 0.93) and control subjects (encoding, left 4.44 ± 0.85, right 4.09 ± 1.16; recognition, left 4.42 ± 0.60 right 4.11 ± 0.68), with no group differences. Local maxima for the hippocampus are listed in Table 2.

We observed negative correlations between PG and deactivation in anatomically defined ROIs, including the cuneus (ρ = −0.49, P = 0.04) and precuneus (ρ = −0.49, P = 0.04) during encoding and the medial frontal gyrus (ρ = −0.52, P = 0.03) during recognition.

This study demonstrates impaired task-induced deactivation in the DMN and aberrant activation in task-relevant regions during a memory task in type 2 diabetic patients. During encoding, we observed reduced activation of task-relevant regions and less extensive deactivation of the DMN in type 2 diabetic patients relative to control subjects. During recognition, we observed impaired deactivation of the DMN in type 2 diabetic patients.

Among diabetic patients, prescan PG levels ≥11 mmol/L were associated with reduced deactivation of DMN regions during both encoding and recognition. There was some overlap between these regions and the regions that differed between diabetic patients and control subjects (e.g., anterior cingulate). We also observed that current HbA_1c, lifetime average HbA_1c, and HOMA-IR influence brain activity in the DMN. These observations suggest that several aspects of diabetes, including acute hyperglycemia, short- and long-term glycemic control, and insulin sensitivity, influence DMN activity. Additionally, diabetic patients differed significantly from control subjects on triglycerides and blood pressure. These variables, often comorbid with diabetes, may also contribute to brain differences. Thus, it is possible that the observed abnormalities are related to both metabolic and vascular dysfunction, which are often associated with type 2 diabetes.

The observed brain activity patterns are similar to those found in patients with AD (6) prior to evident cognitive problems. Recent research has focused on the putative relationship between type 2 diabetes and AD, which may stem from the crucial role that insulin plays in both disorders (16,18). Glucose metabolism may also be involved (19); however, the results are not clear (20). We observed that control subjects who carry the AD risk allele APOE4 showed reduced deactivation in the cerebellum and fusiform gyrus during encoding. Others have found that these regions are recruited during episodic memory tasks (21), and the cerebellum has previously been shown to be less active in E4 carriers as they age (22). Differences between our finding and those of others who have observed effects of APOE4 in the medial temporal lobe (23) may stem from particulars about the memory task and ages of the participants.

This study is cross-sectional; thus, we cannot assert that the observed pattern of functional abnormalities confers increased risk for AD or MCI, although similar abnormalities are seen in populations at risk for AD and in patients with current AD or MCI (6). There was no evidence of cognitive impairment in this sample; however, significant impairment has often been shown in elderly type 2 diabetic patients (12), suggesting that the observed fMRI abnormalities may precede cognitive decline. Future studies should use longitudinal designs and evaluate the impact of previous hypoglycemia, exogenous insulin administration, white matter hyperintensities, vascular reactivity, and task vigilance. Exogenous insulin, in particular, has been shown to reduce the BOLD response in a visual task (24). By incorporating these variables into the design of future studies, we can begin to isolate the etiology of DMN abnormalities in diabetes.

In summary, this study demonstrates that type 2 diabetic patients show hypoactivation of task-relevant regions during encoding and impaired deactivation in the DMN during recognition despite normal cognitive performance, suggesting a reduced ability to modulate task-related brain activity. Acute hyperglycemia, short- and long-term glycemic control, and insulin resistance likely contribute to this abnormality. Through a better understanding of the functional brain abnormalities observed in type 2 diabetes, diagnostic tests and therapies can be developed to detect abnormalities and ameliorate function in vulnerable brain regions, networks, and pathways. Beginning therapies in middle age can help prevent development of clinically significant cognitive impairment.

Acknowledgments. The authors acknowledge the Specialized Assay Core at the Joslin Diabetes Center. The authors thank the study participants for their contribution to the study and Fotini Kourtelidis and Michael Amico for their help with scanning patients at Beth Israel Deaconess Medical Center throughout this research study.

Funding. This study was supported in part by National Institutes of Health grants 5R01-AG-034165-A2 (G.M.) and P30-DK-036836 (to the Joslin Diabetes Research Center, Genetics Core) and by the Herbert Graetz Fund.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. T.J.M., V.L.F., and G.M. researched data, contributed to discussion, and wrote, reviewed, and edited the manuscript. N.R.B., W.S.H., D.C.S., A.M.J., S.E.F., M.E.S., and R.A.S. researched data, contributed to discussion, and reviewed and edited the manuscript. G.M. 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 Harvard Medical School Department of Psychiatry 2013 Research Day and Mysell Lecture, Boston, MA, 10 April 2013, and at the 73rd Scientific Sessions of the American Diabetes Association, Chicago, IL, 21–25 June 2013.