Sixty-seven patients with T2DM were recruited for this study from May 2019 to July 2020 at Shaanxi Provincial People's Hospital, including 33 patients with MCI (DMCI group) and 34 patients without MCI (DMCN group). Thirty-four healthy controls (HCs) were recruited from the health examination center of our hospital. All participants were between 45 and 70 years of age, right-handed, and had at least 6 years of education. The diagnostic criteria of T2DM were based on the American Diabetes Association's 2014 guidelines. Patients with T2DM were on stable therapy (diet, oral medications, and/or insulin). Patients with T2DM were excluded who had a history of hypoglycemia (blood glucose concentration of < 3.9 mmol/L) or hyperglycemia (blood glucose concentration of >33.3 mmol/L). The inclusion criteria of the DMCI group were: (1) complaints of hypomnesis, which occurred after a clinical diagnosis of T2DM; (2) an MMSE score >24 and MoCA score < 26; (3) no other physical, mental, or neurological disorders that could lead to cognitive impairment. The exclusion criteria for all the participants were: (1) severe claustrophobia or contraindications for MRI; (2) alcoholism, Parkinson's disease, major depression, brain injury, epilepsy, or other neurological or psychiatric disorders; (3) any other systemic disease.

Every participant arrived at the department for MRI at 6:30–7:00 p.m. after dinner and controlled their blood glucose according to their doctor's orders on the day of the scan. MRI was performed after ~30 min of structured clinical interview and a series of psychological tests. The test procedure and scan time of the HCs were the same as those of patients with T2DM. Only one participant was scanned each day to ensure that each participant completed the examination with relatively stable blood glucose. The study was approved by the ethics committee of Shaanxi Provincial People's Hospital, and written informed consent was obtained from all the participants. All methods were performed in accordance with relevant guidelines and regulations according to the Declaration of Helsinki.

All participants underwent the following neuropsychological examinations: Mini-Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA), Clock-Drawing Test (CDT), and Trail-Making Test A (TMT-A). TMT-A was used to test information processing speed and attention, and the increased time patients spend on the TMT-A test indicates that their attention and psychomotor speed are abnormal (29). CDT is often significantly associated with visuospatial and executive function, so it is commonly used to assess these cognitive domains (30). The neuropsychological tests were performed by a psychiatrist with more than 5 years of experience. HCs data were collected from the outpatient medical examination center. The participants' medical history and clinical data, including blood pressure, height, weight, and body mass index (BMI), were obtained from medical records and questionnaires. In addition, glycated hemoglobin (HbA1c), fasting blood glucose (FBG) concentration, triglyceride concentration, cholesterol concentration, and low-density lipoprotein cholesterol concentration were measured by standard laboratory tests.

The MRI images were acquired using a 3.0T MR scanner (Ingenia, Philips Healthcare, The Netherlands) with a 16-channel phased array head coil. Routine T2-weighted and fluid-attenuated inversion recovery (FLAIR) sequence (slices = 18, thickness = 6 mm, gap = 1 mm) were used to exclude visible brain lesions. The age-related white matter change scale (31) was used to evaluate lacunar infarcts and white matter hyperintensity based on FLAIR images; participant with a score >2 was excluded. Resting-state functional MRI (rs-fMRI) images were acquired using a gradient echo planar sequence with the following parameters: repetition time (TR) = 2,000 ms, echo time (TE) = 30 ms, flip angle (FA) = 90°, field of view (FOV) = 230 mm × 230 mm, matrix = 128 × 128, slices = 34, thickness = 4 mm (no gap), axial interleaved acquisition, and 200 volumes were acquired in each scan. Sagittal 3D T1-weighted images were acquired using a fast spoiled gradient echo sequence with the following parameters: TR = 7.5 ms, TE = 3.5 ms, FA = 8°, FOV = 250 mm × 250 mm, matrix = 256 × 256, slice thickness = 0.55 mm (no gap), and 328 sagittal slices. All participants were instructed to close their eyes and stay awake throughout the scan.

The rs-fMRI data were preprocessed by the DPABI (http://rfmri.org/dpabi/) software package (32). The first 10 time points were discarded to ensure the stability of the magnetic field. The slice-timing and realignment of head motion correction were performed for the remaining 190 time points. Participants were excluded from this study whose head motion >1.5 mm and/or translation >1.5° rotation. Moreover, the scrubbing method was used to reduce the effect of high-head motion (33). We calculated the frame-wise displacement (FD) and set an FD threshold for bad volumes as 0.2 mm. The bad volumes were scrubbed which the FD values were greater than 0.2 mm, as well as one forward volume and two back volumes of the bad volumes. Then, each bad volume was modeled as a regressor in the model regression (34). Then, the images were normalized to the standard Montreal Neurological Institute (MNI) space with an echo planar imaging (EPI) template (resampling voxel size = 3 mm × 3 mm × 3 mm). In addition, we regressed out 24 head motion parameters, the linear trend signal, cerebrospinal fluid signals, and white matter signals. Finally, a temporal-band filter (0.01–0.08 Hz) was used to regress the effect of physiological noise.

Differences in the brain functional hubs were obtained by comparing the DC maps of the three groups. First, a voxel-based correlation analysis of whole-brain function was performed on the preprocessed fMRI data, and Pearson's correlations were calculated for the time course from each voxel to every other voxel in the whole brain to create a correlation matrix. Second, the binarized matrix was obtained by thresholding each correlation at r > 0.25 (35, 36). Then, these voxel-wise DC values were converted into a z-score matrix to improve normality. Finally, the zDC maps were further smoothed (FWHM = 6 mm) for statistical analysis.

A seed-based approach was used for FC analyzes of the spatial smoothing (FWHM = 6 mm) data. The peak MNI coordinates of significantly altered DCs between the DMCI and DMCN groups were selected as the center of spherical regions of interest (ROIs) with a 6-mm radius. Then, the time courses were extracted for the ROIs, and correlations were calculated between each ROI and every other voxel within the brain to obtain FC maps. These FC maps were converted to z-score maps using Fisher's z transformation to improve the normality.

The chi-square (χ²) test was used to analyze the sex-based differences among the three groups. One-way analysis of variance (ANOVA) was used to analyze other demographic and clinical and cognitive scores. A two-sample t-test was used to compare disease duration between the DMCI and DMCN groups. These analyzes were performed using SPSS version 24 (IBM Corporation, Armonk, NY, United States), and P < 0.05 was considered to be statistically significant.

The statistical analyzes of DC and seed-based FC were performed in DPABI software. Analysis of covariance (ANCOVA) was performed to investigate the regions with significant differences in DC and FC among the three groups, with age, gender, and years of education as covariates (P < 0.001, uncorrected, cluster >50). The ANCOVA results were used as a mask, and post-hoc analyzes within the areas were performed to investigate pairwise between-group differences (FDR corrected, P < 0.05).

Correlation analysis of average zDC and zFC values of the brain areas with significant differences was extracted from the DMCI and DMCN groups. Then, significant partial correlations (Bonferroni corrected, P < 0.05/13) were performed between the zDC values, zFC values, and clinical scores, with age, gender, and years of education as covariates.

Two patients with DMCI were excluded from the analyzes for excessive motion; thus, 31 patients with DMCI, 34 patients with DMCN, and 34 HCs were included in the analyzes. There were no significant group differences in gender, age, education, blood pressure, BMI, triglycerides, total cholesterol, or low-density lipoprotein. Compared with HCs, the two T2DM groups showed higher FBG and HbA1c scores. Analysis of the neuropsychological data revealed the DMCI group had poorer MMSE, MoCA, and CDT scores and higher TMT-A scores compared to the DMCN group and HCs (see Table 1).

Compared to the HCs, the DMCI group showed decreased zDC values in the right parahippocampal gyrus (PHG), fusiform gyrus (FFG), and middle occipital gyrus (MOG), the bilateral calcarine (CAL), and the posterior cingulate cortex (PCC), whereas the DMCN group showed increased zDC in the left superior frontal gyrus (SFG) and decreased zDC in the bilateral CAL and right MOG. Compared with the DMCN group, the DMCI group showed decreased zDC in the right PHG, bilateral PCC, and left SFG (see Table 2, Figure 1). In addition, the zDC values in the right FFG were negatively correlated with CDT scores of patients with DMCI (r = −0.554, P = 0.001) (Figure 2A). Moreover, the zDC values in the bilateral CAL were negatively correlated with the HbA1c scores of patients with DMCI (r = −0.553, P = 0.001) (Figure 2B).

Three regions (the right PHG, bilateral PCC, and left SFG) were selected as ROIs to perform voxel-wise FC analyzes. The analyzes revealed between-group differences in the right PHG, as shown in Figure 3, Table 3. Compared to HCs, both T2DM groups showed decreased FC between the right PHG and FFG. The DMCI group also had significantly weaker FC values between the right PHG and left angular gyrus than the DMCN and HC groups. Moreover, FC analysis of the bilateral PCC showed reduced FC in the right middle temporal gyrus (MTG) and left MOG in the DMCI group, and the DMCN group showed increased FC in the left SFG compared with HCs. Compared to the DMCN group, the DMCI group showed reduced FC in the bilateral PCC and right MTG (Figure 4, Table 4). The seed-based FC analysis of the SFG found no significant differences among the three groups.

We found decreased DC values in the bilateral CAL and right MOG in both T2DM groups. These two regions are thought to be involved in processing visual information and encoding visual memories (37). Visual function impairment exists in the early stage of patients with T2DM (38, 39) and long-term hyperglycemia is the main factor leading to visual cognitive impairment (40, 41). The correlation results of the present study also suggested that the functional abnormalities of bilateral CAL may be affected by HbA1c in patients with DMCI. In addition, we found that patients with DMCI exhibited decreased DC values in right FFG and correlated with worse CDT scores. Those evidence suggest that visuospatial impairment is more widespread in patients with DMCI. The right FFG is involved in visuospatial function (42), and alterations in the FFG can cause visual cognitive impairment in patients with MCI, especially facial cognition (43, 44).

The brain regions of PHG, PCC, and SFG are the core nodes in declarative memory (45), default mode network (46), and executive control functions (47), respectively. These brain regions play an important role in maintaining higher cognitive functions. Chronic hyperglycemia can affect hub nodes and disrupt the topological integration of the brain (48), which may be the reason for the decreased DC values of these nodes in the DMCI group compared with the DMCN group. The results of this study suggest that patients with DMCI have more abnormal higher cognitive function brain regions. Interestingly, compared with HC, the DC value of SFG in the DMCN group increased, which may suggest that patients with T2DM had abnormal changes in brain function before MCI. Multiple previous studies have found increased FC in the SFG in patients with subjective cognitive impairment (49, 50). The different patterns of changes in the SFG in the two groups of patients with T2DM seem unusual, but this may suggest that the changes in the SFG under different cognitive states may be a complex process. Multiple studies have demonstrated that the impairment of brain function may not be a linear process (51, 52), which may be the reason why the SFG was not significantly different among groups in subsequent FC analysis. In addition, the present study also found that patients with DMCI showed worse psychomotor speed and attention, which is consistent with previous findings (53, 54) that patients with MCI are often accompanied by the impairment of multiple cognitive domains.

In our study, both T2DM groups showed reduced connectivity between the right PHG and right FFG. The FFG connects with the medial temporal lobe, including the PHG, which is related to visual working memory and visual-spatial abilities (55). A neuropathological study suggested that the Tau protein levels in the FFG and PHG were associated with longitudinal visuospatial cognitive decline in patients with AD (56). The pathological mechanism of cognitive impairment in T2DM is similar to that of AD (57), which may suggest that the disordered functional connectivity of PHG and right FFG is related to the abnormal visuospatial function of patients.

The angular gyrus is connected to both the PHG and hippocampus via the inferior occipitofrontal fasciculus (58, 59), and it plays a critical role in memory retrieval (60, 61). The angular gyrus, hippocampus/parahippocampal gyrus, and dorsomedial prefrontal cortex are activated during episodic autobiographical memory (62). Moreover, the parahippocampal region exhibits strong connectivity with the angular gyrus during episodic memory retrieval processing (63). Studies (2, 64) suggest that episodic memory dysfunction exists in patients with T2DM. The current study found that the DMCI group showed lower FC values in the right PHG and left angular gyrus than the DMCN and HCs groups, suggesting that patients with DMCI may have a dysfunction in episodic memory retrieval.

Research has confirmed that the spontaneous activity of the MTG is strongly correlated with semantic processing efficiency (65). Patients with MCI showed less effective connectivity between the MTG and PCC, and this impaired connectivity is correlated with cognitive performance during auditory verbal learning (66). Our study found the PCC showed reduced FC with the right MTG in the DMCI group. This is consistent with previous findings (67, 68), which may suggest abnormal semantic memory functioning in patients with DMCI.

The MOG is involved in visual processing, and patients with MCI show reduced connectivity between the PCC and MOG during visual cognitive tasks, which is associated with impaired visual processing (69). Consistent with the previous DC results, FC analysis further confirmed that visual impairment was more severe in patients with DMCI. In addition, similar to previous research (67), we found increased FC values between the PCC and left SFG in the DMCN group, which may be interpreted as a compensatory increase in patients' cognitive control of food. The SFG and PCC/precuneus play important roles in the cognitive control circuit, supporting cognitive control of appetite upon encountering visual food cues. It was found that neuropeptide oxytocin (OXT) was able to reduce food craving trends under cognitive control conditions and that the effects of OXT paralleled increased activity in the SFG, precuneus, and cingulate cortex (70).

This study has several limitations. First, this experiment is a cross-sectional study with a small sample size, and a larger sample could increase the credibility of our results. Second, although we adopted a relatively loose correction method in the ANCOVA analysis, the post-hoc analysis adopted strict FDR correction, which could minimize the false positive of between-group differences. Third, many of the patients in this study received various medications that may have biased the results, but that would be hard to avoid. Finally, relatively few behavioral scales were used in this study, which leads to the lack of behavioral support for some results. We will employ more abundant and comprehensive behavioral scales in future studies.