Synopsis

Keywords: Alzheimer's Disease, fMRI (resting state)

The default mode network (DMN) could be further divided into three subsystems, and each subsystem of DMN serves different cognitive functions. In this study, we revealed that cognitive decline is driven by weakened functional connectivity within the core subsystems, but enhanced functional connectivity between the core and MTL subsystems as well as the core and DMPFC subsystems. This dissociation may reflect abnormal core-centered connection abnormalities, along with the disruption of the MTL subsystem and the DMPFC subsystem to some degree. Our findings provided a promising imaging biomarker for the early diagnosis of AD.

Background

Previous studies have revealed the functional network alterations of the default mode network(DMN), suggesting that the DMN could be the critical brain network underlying cognitive impairment severity or progression. DMN could be further divided into three subsystems: the core subsystem, the dorsal medial prefrontal cortex (DMPFC) subsystem, and the medial temporal lobe (MTL) subsystem. Each subsystem of DMN serves different cognitive functions. In the present study, we sought to investigate the difference of functional connectivity (FCs) within DMN on subsystem-level between MCI and HC, and developed several disease classification models with clinical information and the FCs features.

Material and methods

In this study, 117 MCI patients, 110 age and sex-matched elderly healthy controls were prospectively recruited. All participants were scanned on a 3 Tesla MR scanner (Discovery 750, GE Healthcare, US) using a standard eight-channel head coil. The high-resolution 3D T1-weighted images were acquired by a fast spoiled gradient-echo (FSPGR) sequence. The resting-fMRI images were obtained using a single shot gradient-recalled echo-planar imaging (EPI) sequence. The resting-state fMRI Data preprocessing was conducted in the toolbox for Data Processing & Analysis of Brain Imaging (DPABI, Yan et al., 2016, http://rfmri.org/DPABI).
Twenty-four anatomical regions of interest (ROIs) of DMN were defined in the present study based on Yeo's 17-network parcellation. These 24 ROIs can be further divided into 3 DMN subsystems following previous publications. We calculated Tikhonov Partial Correlation as FC instead of conventional Pearson’s correlation in the present study. We extracted the average time series from each ROI and calculated Tikhonov Partial Correlation between each ROI-pair as FC, forming a 24 × 24 FC matrix. We then Z-transformed all FCs with a Fisher's r-to-z transformation. For the entire DMN, we simply averaged across all FCs in the matrix. For FCs at the subsystem level, we averaged FCs among ROIs of the same subsystem as FCs within a subsystem and all FCs reflecting pair-wise connections between subsystems as FCs between subsystems. To give a comprehensive view of the alterations of FCs, we subsequently conducted an ROI-level analysis on FC. Similarly, FCs of all ROI pairs in the two groups were calculated the same way as in the DMN subsystems. After the subsystem level and ROI level analysis, FCs with significant between-group differences in MCI group and HC group were selected as the significant features. Then, combined with clinical characteristics, we used multivariate logistic regression to find the independent discriminative features and the receiver operating characteristic (ROC) analysis to assess the disease discriminative ability.

Results

Demographic data and cognitive performance
No significant differences in age, gender, and education level were found between the MCI and HC-elder groups (p＞0.05). MMSE and MoCA scores were significantly higher in the HC group when compared to MCI group (both p＜0.0001).
FCs within and between subsystems of DMN
The overall FC strengths within the DMN were compared between the MCI and HCs, and no significant differences were found between the two groups (t= -0.7015, p=0.4837).FC within the core subsystem of MCI group was significantly decreased compared with the HC group (t=-5.5592, p＜0.0001). FC between the core and MTL subsystems was significantly enhanced in MCIs as compared to the HCs(t=4.1741, p＜0.0001). Besides, FC between core and DMPFC subsystem were significantly increased in the MCI group compared to the HCs (t=2.3376, p=0.0203).No significant differences were found in FC between the DMPFC and MTL subsystems or within the DMPFC or MTL subsystem. (Figure 1)
ROI levels
We found eighteen ROI connections were significantly different between the MCI and HC individuals, including seven decreased and eleven enhanced connections (MCI versus HC). ROI-wise results that were generally in correspondence with the network and subsystem level findings.
Disease discrimination and evaluation
Five models were constructed using the above independent discriminative features to discriminate MCI from HC. In Model I, FCs of within and between DMN subsystems features were set as independent discriminative features, which achieved the AUC of 0.7149. In Model II, features were identified as nine FCs of ROI connections with significant between-group differences. The performance AUC of model II was 0.8070. In Model III, independent discriminative features were set as the combination of significant features from the model I and model II, which was the joint model of significant FCs of subsystems and ROIs, and the AUC were enhanced to 0.8190. In Model IV, we regarded the demographic features and MMSE scores as independent discriminative features, which achieved the AUC of 0.8731. In Model V, features were set as the combination of significant features in model III and model IV, which reached the highest AUC of 0.9253.

Conclusion

In conclusion, this present study offers direct evidence on the pivotal role of the DMN in the neural mechanisms of AD. This dissociation may reflect abnormal core-centered connection abnormalities, along with the disruption of the MTL subsystem and the DMPFC subsystem to some degree. Our findings provided a primary assurance of its potential clinical utility for providing novel insight into the potential pathogenesis and offering a promising imaging biomarker for the early diagnosis of AD.

Acknowledgements

Authors thank all the staff and participants in the study.