The individuals with and without type 2 diabetes in this study were recruited from the endocrinology department of Tianjin Medical University General Hospital and from community recruitment respectively between 2018 and 2019. A total of 109 individuals with type 2 diabetes and 119 individuals without type 2 diabetes were initially recruited. Four individuals with type 2 diabetes and five individuals without type 2 diabetes were excluded due to poor blood sample quality. Two individuals with type 2 diabetes were excluded due to poor image quality. Finally, 103 individuals with type 2 diabetes and 114 well-matched individuals without type 2 diabetes were enrolled. Type 2 diabetes were diagnosed according to the 2010 criteria of the American Diabetes Association (ADA)(Association, 2010). Individuals with type 2 diabetes related complications, including peripheral neuropathy, retinopathy, and nephropathy were excluded from the study. None of the individuals had experienced severe hypoglycemia during the past two years. All the participants were right-handed. Exclusion criteria were: (1) previous history of brain disease, including stroke, epilepsy, trauma or hemorrhage; (2) Mini-Mental State Examination (MMSE)(Folstein, Folstein, & McHugh, 1975) score < 27; (3) psychiatric or neurologic disorders that may affect cognition; (4) alcohol or drug abuse; and (5) contraindications for MRI scan. All subjects underwent a series of standardized clinical evaluations.
All the participants were in Chinese Hans. Height, weight, and body mass index (BMI) were measured for each participant. The criteria for hypertension were systolic blood pressure >140 mmHg or diastolic blood pressure >90 mmHg or taking antihypertensive medications. Fasting blood glucose (FBG) and glycosylated hemoglobin (HbA1c) were measured via standard laboratory testing.
A battery of cognitive assessments was performed before MRI. General cognition was assessed with the MMSE(Folstein et al., 1975). Anxiety and depression were evaluated with the Self-Rating Anxiety Scale(Zung, 1971) and the Self-Rating Depressive Scale(Zung, 1965), respectively. Episodic memory was assessed by two tests: (1) The Rey-Osterrieth Complex Figure Test (ROCF)(Shin, Park, Park, Seol, & Kwon, 2006) which consists of three conditions: copy, immediate recall, and delayed recall. (2) The Chinese version of the Auditory Verbal Learning Test (AVLT)(Rosenberg, Ryan, & Prifitera, 1984) including short- and long-term memory.
Image data acquisition
All imaging data were obtained on a 3.0-T MR system (Discovery MR750; General Electric, Milwaukee, WI, USA), equipped with an 8-channel phase-array head coil. DTI images were acquired by a single-shot echo planar imaging (EPI) sequence with the following parameters: repetition time (TR) = 7000ms; echo time (TE) = 95ms; flip angle (FA) = 12°; field of view (FOV) = 256mm × 256mm; matrix = 128 × 128; slice thickness = 3mm, no gap; 48 axial slices; 64 encoding diffusion directions with b = 1,000 s/mm2, and 12 non-diffusion b = 0 s/mm2 images. The Sagittal T1-weighted images were obtained using a brain volume sequence, as follows: TR = 8.2ms, TE = 3.2ms, TI = 450ms, FA = 12°, matrix = 256 × 256, and 188 continued sagittal sections with section thickness of 1 mm.
Imaging preprocess and network construction
The DTI data preprocessing steps were performed using a PANDA toolbox(Cui, Zhong, Xu, He, & Gong, 2013) based on FMRIB Software Library v5.0(Jenkinson, Beckmann, Behrens, Woolrich, & Smith, 2012), briefly including brain extraction, realignment, eddy current and motion artifact correction, fractional anisotropy calculation, and diffusion tensor tractography. An FA threshold of 0.2 and a turning angle threshold of 45°in the Fiber Assignment by Continuous Tracking (FACT) algorithm were set when tracking WM fibers.
An Automated Anatomical Labeling (AAL) atlas was used to parcellate each brain into 90 anatomical regions. For detailed parcellation process please see the Online Resource. Each AAL region was taken as a node, and interconnections between brain regions were taken as the edges of the structural network. Interconnections between brain regions were considered present if at least three WM fibers were present between the regions, as in previous studies(Shu et al., 2012). The WM fiber number (FN) threshold was set as three. Finally, a binary 90 × 90 matrix for each subject was obtained.
The nodal topological properties including the nodal global efficiency (NGe), nodal local efficiency (NLe), nodal clustering coefficient (NCp), nodal shortest path length (NLp), and nodal degree (Nd) were calculated with the graph-theoretical network analysis toolbox (GRETNA; http://www.nitrc.org/projects/gretna). The explanation of these properties was documented in the Online Resource. Detailed calculation methods have been documented in a previous article(Rubinov & Sporns, 2010).
All the participants were genotyped using the Illumina Infinium Asian Screening Array (ASA) (https://support.illumina.com.cn/array/array_kits/infinium-asian-global-screening-array.html), a high-throughput genotyping chip designed for the Asian population (Illumina Asian screening array chip) with 700,000 sampling SNPs.
Polygenic risk score
After the strict QC and imputation process (for detailed steps please see Online Resource), PRS was calculated. A recent meta-analyzed GWAS of type 2 diabetes in East Asian individuals (including 77418 individuals with type 2 diabetes and 356,122 individuals without type 2 diabetes)(Spracklen et al., 2020) has identified 111 type 2 diabetes risk SNPs (adjusted for BMI), 97 of which were found in our imputed genotype dataset. The 97 risk SNPs were used for the calculation of PRS. The score was calculated on SNPs reaching genome-wide significance (P = 5 × 10−8). For each participant, the PRS was calculated by multiplying the number of risk alleles for each SNP by the weight for that SNP, and then taking the sum across the 97 SNPs, according to the following formula(De Jager et al., 2009):
where i was the SNP, was the weight for SNP i, and Gi was the number of risk alleles. The weight was the natural log of the odds ratio (OR) value of each SNP, which was obtained from the above-mentioned type 2 diabetes GWAS research(Spracklen et al., 2020).
Demographic, clinical, behavioral data, and PRS
Statistical analyses for demographic, clinical and behavioral data were performed using Statistical Package for Social Sciences (SPSS, v. 23.0, IBM SPSS Statistics, IBM Corporation). The two samples T-test was used for continuous variables, and the Chi-square (χ 2) test was used for categorical variables. The significant level was set at P < 0.05.
PRS-by-disease interaction analysis
The PRS-by-disease interactions on the bilateral hippocampal topological properties (NGe, NLe, NCp, NLp and Nd) were evaluated by analysis of variance (ANOVA) with the hippocampal topological property was treated as a dependent variable, diagnosis (individuals with type 2 diabetes vs. without type 2 diabetes), PRS, and their interaction as interesting independent variables, and age, gender, education years, and BMI as confounding variables. When ANOVA showed a statistically significant interaction effect after the Bonferroni correction, the post hoc correlation analyses were performed to test the relationship between the hippocampal topological properties and the PRS in the individuals with and without type 2 diabetes respectively.
Correlation analysis between hippocampal topological properties and episodic memory
Partial correlation analyses were performed to test the correlation between the hippocampal topological properties and episodic memory tests after controlling for the age, gender, educational level, and BMI in the individuals with type 2 diabetes and the individuals without type 2 diabetes. Statistical significance was set at P < 0.05.
The PROCESS macro version 3.4 implemented in IBM SPSS was adopted to perform the mediation analyses to examine whether the hippocampal topological properties mediated the association between the PRS and the episodic memory, with the PRS as the independent variable, the hippocampal topological property as the mediation variable, and the episodic memory as the dependent variables. The age, gender, education years, and BMI were controlled. During the mediation analysis, ordinary least squares regression was used to calculate statistics for specific paths, and 5000 bias-corrected bootstrap resamples were conducted to generate a CI for the mediation effect(Preacher & Hayes, 2008). When the 95% CI does not contain zero, it is considered a significant mediation effect, whereas the main effect for mediation analysis is unnecessary.