Ying Cui,1 Yun Jiao,1 Yu-Chen Chen,1 Kun Wang,2 Bo Gao,3 Song Wen,1 Shenghong Ju,1 and Gao-Jun Teng1

Altered Spontaneous BrainActivity in Type 2 Diabetes:A Resting-State Functional MRIStudy

Previous research has shown that type 2 diabetesmellitus (T2DM) is associated with an increased riskof cognitive impairment. Patients with impairedcognition often show decreased spontaneous brainactivity on resting-state functional magneticresonance imaging (rs-fMRI). This study used rs-fMRIto investigate changes in spontaneous brain activityamong patients with T2DM and to determine therelationship of these changes with cognitiveimpairment. T2DM patients (n = 29) and age-, sex-,and education-matched healthy control subjects (n =27) were included in this study. Amplitude of low-frequency fluctuation (ALFF) and regionalhomogeneity (ReHo) values were calculated torepresent spontaneous brain activity. Brain volumeand cognition were also evaluated among theseparticipants. Compared with healthy control subjects,patients with T2DM had significantly decreased ALFFand ReHo values in the occipital lobe and postcentralgyrus. Patients performed worse on several cognitivetests; this impaired cognitive performance wascorrelated with decreased activity in the cuneus andlingual gyrus in the occipital lobe. Brain volume didnot differ between the two groups. The abnormalitiesof spontaneous brain activity reflected by ALFF andReHo measurements in the absence of structuralchanges in T2DM patients may provide insights into

the neurological pathophysiology underlyingdiabetes-associated cognitive decline.Diabetes 2014;63:749–760 | DOI: 10.2337/db13-0519

Type 2 diabetes mellitus (T2DM) has been shown to beassociated with an increased risk of cognitive impairment(1,2), which primarily manifests as declining memory,information processing speed (3), attention, and execu-tive function (4). However, the pathophysiologicalmechanism of T2DM-induced cognitive impairment isstill largely unknown (5).

Neuroimaging has proven to be a useful tool for in-vestigating the diabetic brain. Cerebral atrophy and whitematter (WM) lesions, which are commonly reportedstructural abnormalities in previous studies, are believedto be modestly associated with diabetes-related cognitivedysfunction (6–8). Magnetic resonance (MR) spectroscopywas used to determine the concentration of brainmetabolites. A recent MR spectroscopy study in patientswith T2DM and major depression revealed that abnormalbrain metabolite measurements may be related to moodchanges in this population (9). However, little is knownabout the effects of diabetes on neural activity. Neuralactivity is a sensitive measurement that has been observedto be acutely altered by brain structural lesions (10).Neural abnormalities have been detected in populations at

1Jiangsu Key Laboratory of Molecular and Functional Imaging, Department ofRadiology, Zhongda Hospital, Medical School of Southeast University, Nanjing, China2Medical School of Southeast University, Nanjing, China3Department of Radiology, Yuhuangding Hospital, Yantai, China

Corresponding author: Gao-Jun Teng, gjteng@vip.sina.com.

Received 2 April 2013 and accepted 18 September 2013.

This article contains Supplementary Data online at http://diabetes.diabetesjournals.org/lookup/suppl/doi:10.2337/db13-0519/-/DC1.

© 2014 by the American Diabetes Association. See http://creativecommons.org/licenses/by-nc-nd/3.0/ for details.

See accompanying commentary, p. 396.

Diabetes Volume 63, February 2014 749

COMPLIC

ATIO

NS

risk for developing cognitive impairment (11). Based onthis evidence, measures of neural activity may be suited totrack the early effects of diabetes on brain function.

Resting-state functional MR imaging (rs-fMRI) hasbeen found to be a powerful tool for evaluating sponta-neous neural activity (12). Recently, Musen et al. (13)used rs-fMRI to investigate neural functional connectiv-ity changes in patients with T2DM and found evidence ofaltered neural networks in nondemented diabeticpatients. However, this study focused only on the func-tional connectivity changes between two distinct brainregions. As the effects of diabetes on the brain may beglobal, a whole-brain analysis of brain function inpatients with T2DM is needed.

Amplitude of low-frequency fluctuation (ALFF) andregional homogeneity (ReHo) analyses are two importantmethods for depicting the various characteristics ofglobal rs-fMRI signals (14). ALFF measures the intensityof neural activity at the single-voxel level (15), whereasReHo measures the neural synchronization of a givenvoxel with its neighboring voxels (16). A previous studyindicated that ReHo may be more sensitive than ALFFfor detecting regional abnormalities and that ALFF maybe complementary to ReHo for measuring global spon-taneous activity (17). Therefore, the combination ofthese two methods may provide more information aboutthe pathophysiological framework in the human brainthan either method alone (17).

In this study, combined ALFF and ReHo analyses wereapplied to investigate global spontaneous neural activityin patients with T2DM. We hypothesized that 1) ab-normal ALFF and ReHo values would be detected withinspecific brain regions; and 2) the spontaneous brain ac-tivity abnormalities would be related to impaired cogni-tive performance and T2DM-related biometricmeasurements.

RESEARCH DESIGN AND METHODS

Subjects

This study was approved by the local institutional reviewboard and conducted between September 2012 and June2013. Written informed consent was obtained from allsubjects. Patients were recruited from the practices ofcollaborating endocrinologists and from the local com-munity via advertisement. Patients who were between 45and 75 years of age and who had disease duration of .1year were qualified for this study. T2DM was definedaccording to the latest criteria published by the AmericanDiabetes Association (18). All patients were closely self-monitored and were routinely treated with hypoglycemicagents; none had any history of hypoglycemic episodes.Control participants were recruited through advertise-ments, and were matched with T2DM patients with re-spect to age, sex, and education. Exclusion criteria for allparticipants included a history of alcohol or substanceabuse; indication of dementia (defined as a Mini-MentalState Examination [MMSE] score of #24) (19); history

of a brain lesion such as tumor or stroke; psychiatric orneurological disorder unrelated to diabetes; and contra-indications to MRI. Control participants were excluded ifthey had a fasting blood glucose level $7.0 mmol/L; glu-cose level $7.8 mmol/L after oral glucose tolerance test(OGTT); or a Montreal Cognitive Assessment (MoCA,Beijing edition) score of ,26 (20). Participants with vas-cular risk factors were not excluded (to improve the gen-eralizability of the study results) (8). A flowchart of thestudy design is provided in Supplementary Fig. 1.

Biometric Measurements

Medical history and medication use were recordedaccording to a standardized questionnaire. Blood pres-sure levels were measured at three different time pointsduring the interview and then averaged. Hypertensionwas defined as a systolic blood pressure .160 mmHg,a diastolic blood pressure .95 mmHg, or self-reporteduse of blood pressure–lowering medication (3). An OGTT(75 g dextrose monohydrate in 250 mL water) was per-formed on all subjects except those being currently treatedfor diabetes. Plasma glucose levels at fasting and 2 h afterthe OGTT were measured, along with HbA1c and choles-terol levels. BMI and waist circumference were measuredand recorded. Insulin resistance (IR) was determined byhomeostasis model assessment (HOMA) of IR for controlsubjects and patients not being treated with insulin. Di-abetic retinopathy was assessed for all participants exceptfor four healthy control subjects who refused the exami-nation. After papillary dilation, stereoscopic color fundusphotographs were taken and graded according to theWisconsin Epidemiologic Study of Diabetic Retinopathy(21). Peripheral neuropathy, as defined by the DiabetesControl and Complications Trial criteria, was assessed forall participants using a method described previously (22).

Cognitive Assessment

All participants underwent a battery of neuro-psychological tests that covered relevant cognitivedomains. Selection of the tests was based on previousliterature describing cognitive dysfunction in T2DMpatients (3,23,24). The MMSE was used to assess possi-ble dementia (19). The MoCA was used to screen subjectswith mild cognitive impairment and to assess theirgeneral cognition (20). The Auditory Verbal LearningTest and Rey-Osterrieth Complex Figure Test (CFT) (bothof which included immediate and delayed recall tasks)were used to assess episodic memory for verbal and vi-sual information. The Trail-Making Test, part A (TMT-A)and part B (TMT-B), was primarily used to evaluate at-tention and psychomotor speed, and the Clock-DrawingTest was used to address several relevant cognitivedomains including executive function and workingmemory (24). All of the tests took ;60 min to complete.

MRI

All MR images were acquired at the Radiology De-partment of Zhongda Hospital via a Siemens (Erlangen,

750 Resting-State Brain Activity in Type 2 Diabetes Diabetes Volume 63, February 2014

Germany) 3-Tesla Trio scanner. Subjects were instructedto keep their eyes closed but to remain awake, to avoidthinking of anything in particular, and to keep theirheads still during the scanning. Functional images wereobtained using a gradient-echo planar sequence (36slices; repetition time, 2,000 ms; echo time, 25 ms; slicethickness, 4 mm; flip angle, 90°; field of view, 240 3240 mm). Structural images were acquired using a T1-weighted three-dimensional spoiled gradient-recalled se-quence (176 slices; repetition time, 1,900 ms; echo time,2.48 ms; slice thickness, 1.0 mm; flip angle, 9°; inversiontime, 900 ms; field of view, 250 3 250 mm; in-planeresolution, 256 3 256). Fluid-attenuated inversionrecovery images were also obtained with the followingparameters: repetition time, 8,500 ms; echo time,94 ms; 20 slices; slice thickness, 5 mm; voxel size,1.3 3 0.9 3 5 mm3.

Structural images (three-dimensional T1-weightedimages) were processed using the VBM8 toolbox software(http://dbm.neuro.uni-jena.de/vbm). Images were seg-mented into gray matter (GM), WM, and cerebrospinalfluid using the unified segmentation model. Statisticalparametric mapping between the patient and controlgroups was generated. The distribution of brain paren-chyma volume (sum of the GM and WM volumes) wasdisplayed as a box plot. Subjects with brain parenchymavolume values that were extreme outliers were consid-ered to have abnormal brain volume and were excludedfrom the study.

Small-Vessel Disease Assessment

WM hyperintensity (WMH) and lacunar infarcts wereassessed on fluid-attenuated inversion recovery imageswith a method described previously (25). Participantswith a rating score .1 (confluence of lesions or diffuseinvolvement of the entire region) were excluded. Twoexperienced radiologists blinded to the group allocationsperformed the ratings separately. Consensus wasobtained through discussion between the two raters.

Data Preprocessing

fMRI imaging data were preprocessed with the toolboxData Processing Assistant for Resting-State functionalMR imaging (DPARSF; http://www.restfmri.net/forum/DPARSF) through statistical parametric mapping (SPM8;http://www.fil.ion.ucl.ac.uk/spm/) and an rs-fMRI dataanalysis toolkit (REST1.8; http://www.restfmri.net). Slicetiming and realignment for head motion correction wereperformed. Any subjects with head motion .2.0-mmtranslation or.2.0° rotation in any direction were excluded.The functional images were then spatially normalized tostandard coordinates and resampled to 3 3 3 3 3 mm3.

ALFF and ReHo Analyses

ALFF and ReHo analyses were performed with RESTsoftware as described in previous studies (15,16).

For ALFF analysis, the resampled images were firstsmoothed with a Gaussian kernel of 4 mm. Linear trend

and band-pass filtering (0.01–0.08 Hz) were performedto remove the effects of low-frequency drift and high-frequency noise. The time series were transformed to thefrequency domain using a fast-Fourier transform. Thesquare root of the power spectrum was then calculatedand averaged across 0.01–0.08 Hz within each voxel toobtain the raw ALFF value. Subsequently, the globalmean ALFF value was calculated by extracting the rawvalues from all voxels across the whole brain and aver-aging them. Finally, the raw ALFF values for each voxelwere divided by the global mean ALFF values for stan-dardization. The resulting ALFF value in a given voxelreflects the degree of its raw ALFF value relative to theaverage ALFF value of the whole brain (26).

ReHo analysis was performed on preprocessed images.After linear trend and band-pass filtering were per-formed, ReHo maps were generated by calculating theconcordance of the Kendall coefficient of the time seriesof a given voxel with its 26 nearest neighbors (16). TheReHo value of each voxel was then standardized by di-viding the raw value by the global mean ReHo value,which was obtained with the same calculation used todetermine the global mean ALFF value. Finally, the datawere smoothed with a Gaussian kernel of 4 mm forfurther statistical analysis.

Statistical Analysis

Demographic and clinical variables and cognitive per-formance scores were compared between the two groupsusing SPSS software (version 18.0; SPSS, Inc., Chicago,IL). An independent two-sample t test was used forcontinuous variables, and a x2 test was used for pro-portions. P values ,0.05 were considered to be statisti-cally significant.

Within-Group AnalysisTo explore the within-group ALFF and ReHo patterns,one-sample t tests were performed on the individual ALFFand ReHo maps for each group using REST-StatisticalAnalysis (written by Chaogan Yan, www.restfmri.net).To display the most significant results and reflect theintrinsic nature of these two algorithms, a conservativestatistical significance was set at P , 0.005 and a clustersize of 24 voxels, which corresponded to a correctedP , 0.005 (multiple comparisons with family-wise errorusing the AFNI AlphaSim program; http://afni.nih.gov/afni/docpdf/AlphaSim.pdf).

Between-Group AnalysisTo investigate the between-group differences of ALFF andReHo values, two-sample t tests were performed with theREST software (within a GM mask). Age, sex, and educa-tion levels were imported as covariates. To exclude thepossible effects of structural differences (27), we alsostratified the modulated GM maps obtained from VBManalysis. Vascular risk factors (hyperlipidemia, waist cir-cumference, hypertension, and scores of WMH and lacunarinfarcts) were also controlled for to exclude possible

diabetes.diabetesjournals.org Cui and Associates 751

confounding effects. The statistical threshold was set atP , 0.01 and a minimum cluster size of 22 voxels, whichcorresponded to a corrected P, 0.01 (AlphaSim correction).

Correlation AnalysisTo identify the association between regional ALFF andReHo abnormalities, a bivariate correlation was per-formed between these two measurements. Briefly, theaverage ALFF and ReHo values of brain regions withsignificant differences were individually extracted andcorrelated with one another.

To investigate the relationship among ALFF/ReHovalues, cognitive performance, and diabetes-relatedparameters (fasting plasma glucose and HbA1c levels,HOMA-IR, and disease duration), Pearson correlationanalyses were performed in a voxel-wise manner with theREST software. Analyses were adjusted for the samecovariates as those controlled in the two-sample t tests.Because the results could be easily affected by noise,a conservative statistical threshold was set at P , 0.005

(after AlphaSim correction) to explore the most signifi-cant correlations among MR voxels.

To verify and extend the results of voxel-wiseanalyses, we performed a further correlation analysisbased on regions of interest (ROIs). Briefly, regionsshowing significant differences were specified within anautomated anatomical labeling template (http://neuro.imm.dtu.dk/services/brededatabase/index_roi_tzouriomazoyer.html), and the mean ALFF and ReHovalues for each ROI were extracted. Partial correlationsamong extracted values, cognition, and diabetes-relatedvariables were calculated using the same covariates asin the voxel-wise analyses.

RESULTS

A total of 58 participants (30 patients and 28 healthycontrol subjects) were recruited for this study. One pa-tient and one healthy control subject were excluded be-cause of excessive head movement. Thus, a total of56 subjects were included in the final data analysis.

Table 1—Demographic, clinical, and cognitive characteristics of study patients and control subjects

Characteristics T2DM patients (n = 29) Control subjects (n = 27) P value

Age (years) 58.3 6 7.3 57.8 6 5.9 0.78

Sex (male/female)* 14/15 11/16 0.60

Education (years) 10.4 6 4.0 10.2 6 2.5 0.80

Disease duration (years) 9.3 6 3.8 — —

Fasting glucose (mmol/L)† 7.8 6 2.2 5.4 6 0.4 ,0.01

HbA1c (%) (mmol/mol)† 7.9 6 1.7 (63 6 18.6) 5.6 6 0.4 (38 6 4.4) ,0.01

HOMA-IR 3.3 6 1.2 2.5 6 1.0 ,0.01

Insulin treatments 6 (21) — —

Retinopathy (background) 8 (28) 0 —

Diabetic peripheral neuropathy 8 (28) 0 —

Vascular risk factorsSystolic BP (mmHg) 133 6 14 130 6 12 0.33Diastolic BP (mmHg) 84 6 12 83 6 9.5 0.74Antihypertensive medications 8 (28) 5 (19) —

Waist circumference (cm)† 91.7 6 9.8 85.4 6 9.2 0.03Total cholesterol (mmol/L) 5.4 6 1.1 5.2 6 1.0 0.31Cholesterol-lowering medications 3 (10) 1 (4) —

Cerebral vessel diseaseWMH 1 (0–6) 0 (0–7) 0.82Lacunar infarcts 6 (21) 3 (11) 0.47

Cognitive performanceMMSE 28.3 6 1.4 29.0 6 1.1 0.07MoCA† 23.6 6 2.9 27.3 6 1.1 ,0.01TMT-A (s)† 72.8 6 24.3 58.9 6 14.6 0.01TMT-B (s)† 174.8 6 59.1 144.4 6 51.3 0.01AVLT 6.2 6 1.3 6.3 6 1.6 0.81AVLT-delayed recall (20 min) 6.1 6 2.5 6.3 6 2.5 0.82CDT 3.5 6 0.6 3.7 6 0.5 0.12CFT-delayed recall (20 min)† 13.3 6 6.2 19.7 6 5.5 ,0.01

Data are mean6 SD, n (%), or median (range) unless otherwise stated. AVLT, Auditory Verbal Learning Test; BP, blood pressure; CDT,Clock-Drawing Test. *The P value for proportions was obtained by x2 test. †P , 0.05.

752 Resting-State Brain Activity in Type 2 Diabetes Diabetes Volume 63, February 2014

Demographic and Cognitive Characteristics

Insulin-treated patients comprised 21% of the diabetesgroup (HbA1c, 8.6 6 1.5%); the remaining patientswere treated with oral antidiabetic agents (58%;HbA1c, 7.9 6 1.9%) or dietary restriction only (21%;HbA1c, 7.2 6 1.3%). The patient and control groupsdid not differ in terms of age, sex, or education (Table1). No significant differences were observed in totalcholesterol level, blood pressure, presence of cerebralsmall-vessel disease, or MMSE scores between the twogroups. Patients with T2DM had significantly higherfasting plasma glucose levels, HOMA-IR, HbA1c levels,and waist circumference than the control group (allP , 0.01). Retinopathy was diagnosed in eightpatients, but all had background nonproliferativechanges only. Peripheral neuropathy was diagnosed ineight patients. In terms of cognitive assessment, patientshad a significantly lower MoCA score than control sub-jects, suggesting that their general cognition was impaired.

Patients also performed significantly worse than controlsubjects on the TMT-A, TMT-B, and CFT-delayed recalltests, indicating that cognitive decrements in thesepatients are more likely to involve the domains ofinformation-processing speed, attention, and visualmemory.

Structural Results

No participants were excluded because of severe atrophy.No significant difference was observed in GM or WMvolume between the patients and control subjects;overall, however, the GM and WM volumes in the di-abetes group were lower than those in the control group(see details in Supplementary Table 1).

ALFF and ReHo Analyses

In both groups, standardized ALFF and ReHo values inthe posterior cingulate cortex (PCC), precuneus (PCu),and medial prefrontal cortex were significantly higherthan the global mean values (Fig. 1).

Figure 1—Representative one-sample t test results of ALFF and ReHo maps (P < 0.005, AlphaSim corrected). Within each group,standardized ALFF and ReHo values in the PCC, the adjacent PCu, and the medial prefrontal cortex were significantly higher than theglobal mean values in both groups. Other regions, such as the inferior parietal lobe and bilateral occipital lobes, also had higher spon-taneous activity. Note that the brain regions were mainly part of the default-mode network, which demonstrated the correctness of ourdata analysis. R, right; L, left; Ctrl, control.

diabetes.diabetesjournals.org Cui and Associates 753

In T2DM patients, the ALFF and ReHo values weresignificantly decreased in the occipital lobe (bilaterallingualgyrus, right fusiform, left cuneus, and right calcarinecortex) and parietal regions (left postcentral gyrus [PoCG])(Fig. 2, Table 2). The ReHo values were also decreasedsignificantly in the thalamus/caudate (Fig. 2B, Table 2).

ALFF and ReHo values in the posterior lobe of cere-bellum (PLC) were higher in diabetic patients than incontrol subjects (Table 2). Increased ALFF and ReHovalues were also found in the anterior cingulate cortexand frontal lobe, respectively (Fig. 2).

Correlation Analysis

Bivariate correlation analyses indicated that the ALFF andReHo values extracted from the occipital lobe and leftPoCG were significantly correlated with each other (Fig. 3).

Voxel-wise correlation analyses revealed that therewere significant correlations between decreased neuralactivity in the occipital lobe and impaired neurocognitiveperformance (i.e., CFT-delayed recall test and TMT-B)(corrected P , 0.005) (Fig. 4). For example, ALFF andReHo values in the cuneus and PCu at the occipital lobewere positively correlated with CFT-delayed score andnegatively correlated with time spent on TMT-B. HOMA-IR in the diabetes group was found to be negativelycorrelated with neural activity in the parietal, frontal,and temporal lobes, and the lingual gyrus. However, nosuch correlations were detected in the control group.Detailed coordinates of these correlations are provided inSupplementary Table 2.

ROI-based correlation analyses indicated that spon-taneous neural activity in the cuneus and lingual gyrus

Figure 2—ALFF (A) and ReHo (B) differences between T2DM patients and healthy control subjects (P < 0.01, AlphaSim corrected).A: Compared with healthy subjects, patients with T2DM showed significantly decreased ALFF in the PoCG, calcarine cortex, and bilaterallingual gyrus (blue), and increased ALFF in the anterior cingulate cortex (red) and PLC (data not shown). B: Compared with healthysubjects, patients with T2DM showed significantly decreased ReHo in the PoCG, bilateral thalamus/caudate, calcarine cortex, middletemporal gyrus, and bilateral lingual gyrus (blue), and increased ReHo in the medial frontal gyrus (red) and PLC (data not shown). Colorscale denotes the t value. x, z, Montreal Neurological Institute coordinates; R, right.

754 Resting-State Brain Activity in Type 2 Diabetes Diabetes Volume 63, February 2014

had significant correlations with cognitive performance(Fig. 5). Higher ALFF and ReHo values in the cuneus wererelated to less time on the TMT-B (ALFF, R =20.451, P =0.035; ReHo, R = 20.535, P = 0.010). Both values in thelingual gyrus were also negatively correlated with TMT-B(ALFF, R = 20.435, P = 0.043; ReHo, R = 20.500, P =0.018). CFT-delay scores were correlated with ReHovalues in the cuneus (R = 0.403, P = 0.041). Amongdiabetic patients, HOMA-IR was found to be significantlycorrelated with neural activity in the cuneus (R = 20.469,P = 0.037).

Effects of Retinopathy and Neuropathy

We performed an analysis to explore the possible effectsof retinopathy on neural activity in the visual cortex andthe effects of neuropathy on neural activity in the PoCG.First, we divided the patients into the following twogroups: patients with and patients without retinopathy.ALFF and ReHo values were extracted from three brainregions in the visual cortex and compared betweenthese two groups and the control subjects. Sub-sequently, patients were divided into another twogroups: patients with and patients without neuropathy.ALFF and ReHo values were extracted from the PoCGand compared between these two groups and the con-trol subjects. Results showed that the mean ALFF andReHo values at the occipital lobe in the control groupwere significantly higher than the patient values.However, no such difference was observed between thepatients with and without retinopathy (Fig. 6A and B).

ALFF and ReHo values extracted from the PoCG also didnot differ between patients with and without neurop-athy (Fig. 6C and D).

DISCUSSION

This study demonstrated decreased neural activity inseveral specific brain regions in T2DM patients versuscontrol subjects. These neural abnormalities were pri-marily found in the occipital lobe and PoCG, and wererelated to the impaired cognitive performance seen inpatients.

ALFF and ReHo analyses have been used to in-vestigate the intrinsic neuropathology of various mentaldisorders (17,26,28,29). These two methods are basedon different neurophysiology mechanisms, with ALFFanalysis demonstrating neural intensity and ReHoanalysis demonstrating neural coherence. In this study,abnormal neural activity was detected by both methodsin several brain regions. The coexisting functional in-tensity and coherence abnormalities in these regionsmay represent more severe functional changes thanthose reflected by a single method. Therefore, we be-lieve our results represent reliable information that isnecessary for understanding cognitive decline in T2DMpatients.

Significantly decreased ALFF and ReHo values in theoccipital lobe and the PoCG in T2DM patients are themajor findings in this study. Previous fMRI studies in-dicated that decreased neural activity in the occipital areaand PoCG were related to visual impairment (30) and

Table 2—Differences in ALFF and ReHo values between the patient and control groups (P , 0.01, AlphaSim corrected)

Brain region BA

MNI coordinates

Voxels Maximal t value*x y z

ALFF differencesR lingual gyrus/fusiform 18 24 275 26 138 23.81L lingual gyrus 18 26 275 23 27 23.43L middle occipital gyrus/cuneus 18/19 18 284 215 28 24.14R calcarine cortex 30 18 263 12 25 23.53L PoCG 3/1 218 230 78 34 24.01R ACC — 9 9 24 41 3.95R PLC — 18 281 251 40 3.93

ReHo differencesR lingual gyrus/fusiform 18 30 281 212 74 23.59L lingual gyrus/cuneus 18/30 23 272 0 31 23.57R calcarine cortex 30 21 257 6 25 23.24L PoCG 3/4 215 230 75 35 24.25R MTG 37 48 266 0 28 23.40B thalamus/caudate 79 9 23 12 79 24.81R PLC — 30 257 230 91 3.94L MFG — 212 33 42 32 4.19

Comparisons were performed at P , 0.01, corrected for multiple comparisons. ACC, anterior cingulate cortex; BA, Brodmann area;MFG, medial frontal gyrus; MNI, Montreal Neurological Institute; MTG, middle temporal gyrus; x, y, z, coordinates of primary peaklocations in the MNI space. *Negative t values: patients with T2DM , control subjects; positive t values: patients with T2DM . controlsubjects.

diabetes.diabetesjournals.org Cui and Associates 755

sensory loss (31), respectively. Diabetes is known to beassociated with retinopathy and neuropathy that canlead to visual and sensory impairment. In this study,most of the patients did not have clinical visual or sen-sory changes, suggesting that decreased neural activitymay be an early change that occurs before the appearanceof clinically measurable symptoms. It will be interestingto follow these patients over time to determine theclinical significance of these findings.

In the correlation analysis, neural abnormalities in thecuneus and lingual gyrus were found to be related toimpaired cognitive performance on TMT-B and CFT-delayed tests in the diabetes group. The cuneus is thecenter of inhibitory control (32), whereas the lingualgyrus is responsible for visual memory (33). Decreasedneuronal activity in the cuneus and lingual gyrus maytherefore have contributed to patients’ poor performanceon related cognitive tests such as the TMT-B and CFT-delayed tests. These region-specific neural cognitionrelationships support our hypothesis that neural activityabnormalities play an important role in diabetes-relatedcognitive dysfunction.

In this study, HOMA-IR was negatively correlatedwith neural activity in the frontal, temporal, andparietal regions in T2DM patients, but not in thecontrol group. These results are compatible with those

from a positron emission tomography study in which IRwas associated with reduced brain metabolism in T2DMpatients (34). Interestingly, a similar pattern of hypo-metabolism has been observed in patients with earlyAlzheimer’s disease (AD) (35). These results suggest thatincreased IR may be a risk for the development of AD indiabetic subjects. Our results showed no correlation be-tween blood glucose and neural activity changes, sug-gesting that blood glucose was not a main contributor tothe neural activity difference, at least during the periodof the current study.

A previous rs-fMRI study suggested that reducedneural activity in AD patients was predominantly locatedin the PCC, medial temporal lobe, and several otherregions (26). In our study with T2DM patients, signifi-cant hypoactivity was found mainly in the occipital lobeand PoCG. Although T2DM is a known risk factor fordementia (36), the relationship between T2DM and AD isstill under debate. The Rotterdam study suggested thatthere is an increased risk for the development of AD inpatients with T2DM (37). However, a population-basedstudy failed to find such a risk in elderly patients withT2DM; this study did demonstrate a twofold higher riskof vascular dementia among diabetic patients (38).A recent systematic review of 14 longitudinal studies in-dicated that the risks for both AD and vascular dementia

Figure 3—Correlations between the ALFF and ReHo values extracted from significantly different regions (bivariate correlation). The meanALFF values had a significant positive correlation with the mean ReHo values in representative regions, including the left PoCG (A; R =0.384, P = 0.040), the right occipital lobe (B; R = 0.544, P = 0.002), the left occipital lobe (C; R = 0.462, P = 0.012), and the calcarine cortex(D; R = 0.813, P < 0.01). A.U., arbitrary units.

756 Resting-State Brain Activity in Type 2 Diabetes Diabetes Volume 63, February 2014

were increased in diabetic patients (39). Our study sug-gests that increased IR might be a potential risk factorfor the development of AD; however, further studies areneeded to test this relationship.

The current study has several limitations. First, thediabetic subjects were receiving various medications thatmay have effects on neural activities. Further studiesshould include medication-naive subjects to rule out thispossible bias. Second, the small sample size in the currentstudy may have reduced our ability to detect more neuralactivity changes. Third, there are no diagnostic criteriafor diabetes-related cognitive dysfunction, and this lackof objective and specific neurocognitive assessment lim-ited our interpretation of the results. Should such criteria

become available, they would help to identify patientswho are at risk for the development of dementia andwould allow researchers to explore their brain patterns.Finally, our study did not measure cerebrovascular re-activity because of technical limitations. Combining CO2

stimulation with blood oxygen level–dependent MRmapping is a proven potential tool for detecting cere-brovascular reactivity changes (40) and may further as-sist our understanding of the cognitive changes thatoccur in diabetic patients.

In conclusion, our combined ALFF and ReHo analysesdemonstrated a significant decrease in spontaneous brainactivity in various brain regions prior to structuralchanges in patients with T2DM. These decreased neural

Figure 4—Representative results of correlation analysis in T2DM patients (P < 0.005, AlphaSim corrected). A and B: Correlation mapsof ALFF/ReHo values and CFT-delayed recall test scores. Significant positive correlations were found in the left cuneus, the left PCu,and the calcarine cortex (red). C and D: Correlation maps of ALFF/ReHo values and performance on TMT-B. Significant negativecorrelations were found in bilateral cuneus and PCu (blue). E and F: Correlation maps of ALFF/ReHo values and HOMA-IR. Neuralactivity in the parietal lobe, frontal lobe, middle temporal gyrus, and lingual gyrus was negatively correlated with HOMA-IR (blue). x, z,Montreal Neurological Institute coordinates; R, right. Color scale denotes R value.

diabetes.diabetesjournals.org Cui and Associates 757

activities were mainly located in the occipital lobe andPoCG, and were correlated with impaired cognitivefunctioning. This study provides a new approach to in-vestigating brain function abnormalities in T2DMpatients and enhances our understanding of the re-lationship between neural abnormalities and cognitivedysfunction in diabetic patients.

Acknowledgments. The authors thank Wen-Qing Xia, Department ofEndocrinology, Zhongda Hospital, Nanjing, China, for assisting with the data

collection. The authors also thank Matthew Tam, Department of Radiology,Southend University Hospital, Essex, England, and Han-Yue Xu, MacalesterCollege, Saint Paul, MN, for editing the manuscript. The authors give specialthanks to Weiping Wang, Imaging Institute of The Cleveland Clinic, Cleveland,OH, for critical revision of the manuscript for intellectual content.

Funding. This work was supported by grants from the Major State BasicResearch Development Program of China (973 Program) (2013CB733800 and2013CB733803) and the National Natural Science Foundation of China GeneralProjects (81271739).

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

Figure 5—Correlations between the mean ALFF and ReHo values in the cuneus and lingual gyrus and HOMA-IR/neurological perfor-mance (partial correlation with age, sex, education, and vascular risk factors controlled). Images in the middle are the cuneus andlingual gyrus masks extracted from the automated anatomical labeling template. The mean ReHo values in the cuneus (black circles)were found to be significantly correlated with HOMA-IR (P = 0.037), the CFT-delayed score (P = 0.041), and TMT-B (P = 0.010). Themean ReHo values in the lingual gyrus (white circles) were also correlated with TMT-B (P = 0.018). ALFF values in both the cuneus(black triangles) and lingual gyrus (white triangles) were negatively correlated with TMT-B (P = 0.035 for cuneus, P = 0.043 for lingualgyrus). A.U., arbitrary units.

758 Resting-State Brain Activity in Type 2 Diabetes Diabetes Volume 63, February 2014

Author Contributions. Y.C. collected the data, performed the analysis,and wrote the manuscript. Y.J. and G.-J.T. made contributions to the design ofthe study and revised the manuscript for intellectual content. Y.-C.C. and K.W.collected the data and contributed to the discussion. B.G., S.W., and S.J.contributed to the discussion and manuscript revision. G.-J.T. is the guarantorof this work and, as such, had full access to all the data in the study and takesresponsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract format RSNA 2013, the annual meeting of the Radiological Society of NorthAmerica, Chicago, IL, 1–6 December 2013.

References1. Jacobson AM, Musen G, Ryan CM, et al.; Diabetes Control and Compli-

cations Trial/Epidemiology of Diabetes Interventions and ComplicationsStudy Research Group. Long-term effect of diabetes and its treatment oncognitive function. N Engl J Med 2007;356:1842–1852 [erratum in N EnglJ Med 2009;361:1914]

2. Cukierman T, Gerstein HC, Williamson JD. Cognitive decline and dementiain diabetes—systematic overview of prospective observational studies.Diabetologia 2005;48:2460–2469

3. van den Berg E, Reijmer YD, de Bresser J, Kessels RP, Kappelle LJ,Biessels GJ; Utrecht Diabetic Encephalopathy Study Group. A 4 year follow-up study of cognitive functioning in patients with type 2 diabetes mellitus.Diabetologia 2010;53:58–65

4. Mogi M, Horiuchi M. Neurovascular coupling in cognitive impairment as-sociated with diabetes mellitus. Circ J 2011;75:1042–1048

5. Euser SM, Sattar N, Witteman JC, et al.; PROSPER and Rotterdam Study.A prospective analysis of elevated fasting glucose levels and cognitivefunction in older people: results from PROSPER and the Rotterdam Study.Diabetes 2010;59:1601–1607

6. Chen Z, Li L, Sun J, Ma L. Mapping the brain in type II diabetes: voxel-based morphometry using DARTEL. Eur J Radiol 2012;81:1870–1876

7. Brundel M, van den Heuvel M, de Bresser J, Kappelle LJ, Biessels GJ;Utrecht Diabetic Encephalopathy Study Group. Cerebral cortical thicknessin patients with type 2 diabetes. J Neurol Sci 2010;299:126–130

8. Reijmer YD, Brundel M, de Bresser J, Kappelle LJ, Leemans A, BiesselsGJ; Utrecht Vascular Cognitive Impairment Study Group. Microstructuralwhite matter abnormalities and cognitive functioning in type 2 diabetes:a diffusion tensor imaging study. Diabetes Care 2013;36:137–144

Figure 6—Differences in ALFF and ReHo values among the three groups (i.e., patients with retinopathy/neuropathy, patients withoutretinopathy/neuropathy, and control group). Low mean ALFF and ReHo values indicate decreased neural activity. A and B: The meanALFF and ReHo values extracted from three brain regions of the occipital lobe among control subjects were all significantly higher thanvalues in both of the diabetes groups. However, the two diabetes groups (with and without retinopathy) did not differ from each other.C and D: The mean ALFF and ReHo values extracted from the left PoCG also did not differ between patients with and patients withoutperipheral neuropathy. *P < 0.05. A.U., arbitrary units.

diabetes.diabetesjournals.org Cui and Associates 759

9. Ajilore O, Haroon E, Kumaran S, et al. Measurement of brain metabolites inpatients with type 2 diabetes and major depression using proton magneticresonance spectroscopy. Neuropsychopharmacology 2007;32:1224–1231

10. Jones DT. Neural networks, cognition, and diabetes: what is the connec-tion? Diabetes 2012;61:1653–1655

11. Machulda MM, Jones DT, Vemuri P, et al. Effect of APOE ´4 status onintrinsic network connectivity in cognitively normal elderly subjects. ArchNeurol 2011;68:1131–1136

12. Mantini D, Perrucci MG, Del Gratta C, Romani GL, Corbetta M. Electro-physiological signatures of resting state networks in the human brain. ProcNatl Acad Sci USA 2007;104:13170–13175

13. Musen G, Jacobson AM, Bolo NR, et al. Resting-state brain functionalconnectivity is altered in type 2 diabetes. Diabetes 2012;61:2375–2379

14. Margulies DS, Böttger J, Long X, et al. Resting developments: a review offMRI post-processing methodologies for spontaneous brain activity.MAGMA 2010;23:289–307

15. Zang YF, He Y, Zhu CZ, et al. Altered baseline brain activity in children withADHD revealed by resting-state functional MRI. Brain Dev 2007;29:83–91

16. Zang Y, Jiang T, Lu Y, He Y, Tian L. Regional homogeneity approach tofMRI data analysis. Neuroimage 2004;22:394–400

17. An L, Cao QJ, Sui MQ, et al. Local synchronization and amplitude of thefluctuation of spontaneous brain activity in attention-deficit/hyperactivitydisorder: a resting-state fMRI study. Neurosci Bull 2013;29:603–613

18. American Diabetes Association. Diagnosis and classification of diabetesmellitus. Diabetes Care 2013;36(Suppl. 1):S67–S74

19. Galea M, Woodward M. Mini-Mental State Examination (MMSE). Aust JPhysiother 2005;51:198

20. Nasreddine ZS, Phillips NA, Bédirian V, et al. The Montreal CognitiveAssessment, MoCA: a brief screening tool for mild cognitive impairment.J Am Geriatr Soc 2005;53:695–699

21. Klein R, Klein BE, Magli YL, et al. An alternative method of grading diabeticretinopathy. Ophthalmology 1986;93:1183–1187

22. Ryan CM, Geckle MO. Circumscribed cognitive dysfunction in middle-agedadults with type 2 diabetes. Diabetes Care 2000;23:1486–1493

23. Reijmer YD, van den Berg E, Ruis C, Kappelle LJ, Biessels GJ. Cognitivedysfunction in patients with type 2 diabetes. Diabetes Metab Res Rev2010;26:507–519

24. Zhou H, Lu W, Shi Y, et al. Impairments in cognition and resting-stateconnectivity of the hippocampus in elderly subjects with type 2 diabetes.Neurosci Lett 2010;473:5–10

25. Wahlund LO, Barkhof F, Fazekas F, et al.; European Task Force on Age-Related White Matter Changes. A new rating scale for age-related whitematter changes applicable to MRI and CT. Stroke 2001;32:1318–1322

26. Wang Z, Yan C, Zhao C, et al. Spatial patterns of intrinsic brain activity inmild cognitive impairment and Alzheimer’s disease: a resting-state func-tional MRI study. Hum Brain Mapp 2011;32:1720–1740

27. Oakes TR, Fox AS, Johnstone T, Chung MK, Kalin N, Davidson RJ. In-tegrating VBM into the General Linear Model with voxelwise anatomicalcovariates. Neuroimage 2007;34:500–508

28. Liu CH, Ma X, Wu X, et al. Regional homogeneity of resting-state brainabnormalities in bipolar and unipolar depression. Prog Neuropsychophar-macol Biol Psychiatry 2013;41:52–59

29. Chen HJ, Zhu XQ, Jiao Y, Li PC, Wang Y, Teng GJ. Abnormal baseline brainactivity in low-grade hepatic encephalopathy: a resting-state fMRI study.J Neurol Sci 2012;318:140–145

30. Liu Y, Liang P, Duan Y, et al. Abnormal baseline brain activity in patientswith neuromyelitis optica: a resting-state fMRI study. Eur J Radiol 2011;80:407–411

31. Luo C, Chen Q, Huang R, et al. Patterns of spontaneous brain activity inamyotrophic lateral sclerosis: a resting-state FMRI study. PLoS One 2012;7:e45470

32. Haldane M, Cunningham G, Androutsos C, Frangou S. Structural braincorrelates of response inhibition in Bipolar Disorder I. J Psychopharmacol2008;22:138–143

33. Bogousslavsky J, Miklossy J, Deruaz JP, Assal G, Regli F. Lingual andfusiform gyri in visual processing: a clinico-pathologic study of superioraltitudinal hemianopia. J Neurol Neurosurg Psychiatry 1987;50:607–614

34. Baker LD, Cross DJ, Minoshima S, Belongia D, Watson GS, Craft S. Insulinresistance and Alzheimer-like reductions in regional cerebral glucosemetabolism for cognitively normal adults with prediabetes or early type 2diabetes. Arch Neurol 2011;68:51–57

35. Mosconi L, Sorbi S, de Leon MJ, et al. Hypometabolism exceeds atrophy inpresymptomatic early-onset familial Alzheimer’s disease. J Nucl Med2006;47:1778–1786

36. van den Berg E, Kessels RP, Kappelle LJ, de Haan EH, Biessels GJ; UtrechtDiabetic Encephalopathy Study Group. Type 2 diabetes, cognitive functionand dementia: vascular and metabolic determinants. Drugs Today (Barc)2006;42:741–754

37. Ott A, Stolk RP, van Harskamp F, Pols HA, Hofman A, Breteler MM. Di-abetes mellitus and the risk of dementia: the Rotterdam Study. Neurology1999;53:1937–1942

38. Hassing LB, Johansson B, Nilsson SE, et al. Diabetes mellitus is a riskfactor for vascular dementia, but not for Alzheimer’s disease:a population-based study of the oldest old. Int Psychogeriatr 2002;14:239–248

39. Biessels GJ, Staekenborg S, Brunner E, Brayne C, Scheltens P. Risk ofdementia in diabetes mellitus: a systematic review. Lancet Neurol2006;5:64–74

40. Spano VR, Mandell DM, Poublanc J, et al. CO2 blood oxygen level-dependent MR mapping of cerebrovascular reserve in a clinical popu-lation: safety, tolerability, and technical feasibility. Radiology 2013;266:592–598

760 Resting-State Brain Activity in Type 2 Diabetes Diabetes Volume 63, February 2014