Between January 2023 and January 2024, 38 Non-CI ESRD patients (ESRD group) and 50 NC group, demographically matched for age, gender, were enrolled at the Second Affiliated Hospital of Guangzhou Medical University for participation in this study. All participants were fully capable of independently administering the MMSE, MoCA. The inclusion criteria for the ESRD group: (1) clinically diagnosed ESRD patients meeting the Kidney Disease Outcome Quality Initiative’s (KDOQI) 2003 criteria for CKD (Levey et al., 2003); (2) age between 30 and 60 years; (3) MMSE scores ≥27, MoCA scores ≥26 (Folstein et al., 1975; Nasreddine et al., 2005); and (4) right-handed.

The exclusion criteria for ESRD group: (1) patients with unstable conditions or those who had undergone renal transplantation; (2) acute renal failure or acute infectious disorders; (3) organic or functional lesions of brain such as severe traumatic brain injury, cerebral infarction, brain tumour, and cerebral hemorrhage, schizophrenia, etc.; (4) severe heart failure, liver disease, etc.; (5) a history of substance dependence or abuse including drugs, alcohol, or illicit substances; and (6) contraindications for MRI scanning such as the presence of a pacemaker, internal metallic foreign bodies, or claustrophobia.

The study was approved by the Ethics Committee of the Second Affiliated Hospital of Guangzhou Medical University. Prior to the commencement of the research, the trial procedures were thoroughly explained to the participants and their legal guardians, who provided consent through a signed informed consent form.

3D-T₁WI and associated plain scan sequence images were collected using a Philips Ingenia 3.0 T MRI device (Philips Ingenia Elition, AMS). 3D-T₁WI scans were performed using a 3D-spoiled gradient echo sequence with acquisition parameters: echo time (TE) =3.5 ms, repetition time (TR) = 7.9 ms, field of view = 250 mm × 199 mm × 170 mm, flip angle = 8 degrees, matrix = 252 × 200, slice thickness = 1 mm, number of slices = 170, interslice gap = 0 mm, and number of excitations = 1, with a scan duration of 4 min 41 s.

The raw 3D-T₁WI image data underwent voxel-based morphometry (VBM) analysis on the MATLAB R2018a data processing platform.¹ Preprocessing of all subjects’ 3D-T₁WI images was conducted using the CAT12 tool. The specific steps included:

Using the graph analysis toolbox (GAT), the 90 grey matter volumes of the cerebral cortex and subcortical regions of each subject were extracted with the AAL template and input to GAT to construct SCNs of grey matter volume. The 90 regions of interest in the AAL atlas were defined as nodes, and the edge strength of the SCNs was determined by the Pearson correlation between the grey matter volume values in corresponding brain regions across all subjects, with age, gender, and total intracranial volume (TIV) as covariates. Each entry r_ij was defined as the Pearson correlation coefficient between the grey matter volumes of regions i and j. The correlation matrix was transformed into a binary matrix, with entries being 1 or 0. The threshold of 0.1 is a standard choice in the field (Yang et al., 2021), confirmed suitable for this study via backpropagation. An entry a_ij is 1 if r_ij exceeds 0.1, otherwise it is 0. Finally, two sets of 90 × 90 binary correlation matrices were output, with the diagonal elements of the constructed matrices also set to 0 (Figure 1). When applying absolute thresholds to threshold the correlation matrices of different groups, variation in node count and degree distribution may occur, affecting network measurements and reducing the interpretability of inter-group comparison results. We therefore set the threshold for each group’s correlation matrix to achieve a binary adjacency matrix with a network density of D. The network density D is defined as the number of edges in the graph divided by the maximum possible number of edges. Comparing graph measurements requires a minimum density of graph to ensure full connectivity and non-fragmentation. For structural networks, densities exceeding 50% may have questionable biological significance. The network density is calculated as D = E/[(N*(N − 1))/2], where E is the number of edges and N represents 90 nodes (Hosseini et al., 2012). The minimum density of both networks is calculated to be 0.33, so the lower limit D_min = 0.33 is defined as the minimum density of the non-fragmented network and the upper limit of the density range D_max = 0.50 represents the maximum density of the network without exceeding the biological significance. Therefore the density range was taken from 0.33 to 0.50 and verified to be statistically significant and the optimal operating density interval of the network was 0.01, so the density range of the network was determined in increments of 0.01.

A total of 38 cases were included in the ESRD group and 50 in the NC group (Table 1). Differences in gender, age, MoCA and MMSE scores between the two groups were not statistically significant (p > 0.05).

The analysis of the VBM data using two-sample t-tests revealed that, compared with the NC group, the ESRD group exhibited decreased GMV in multiple brain regions, with significant reductions in the left temporal pole: superior temporal gyrus, right gyrus rectus, right calcarine fissure and surrounding cortex, bilateral insulae, right parahippocampal gyrus, left superior temporal gyrus, left anterior cingulate and paracingulate gyri, and left fusiform gyrus (FWE-corrected, p < 0.05, cluster size>300, t-value = 4.96). Specific brain regions, with their MNI spatial coordinates, p-values, and associated volume differences, were visualized in three-dimensional images (Table 2 and Figure 2).

First, binary correlation matrices were output for each group, revealing significant correlations in the brain region structures within both groups (Figure 1).

According to the values of the γ, λ, and σ parameters for both groups (Figure 3), the cortical volume SCNs of both the ESRD group and the NC group exhibited small-world network properties. Although there are no significant differences between the two groups, in the ESRD group, the normalized clustering coefficient (γ) and the small-world attribute (σ) decrease under a majority of densities, while the normalized characteristic path length (λ) increases.

The global network metrics (Figure 4) were further compared within a defined range of densities (0.33–0.5, 0.01 intervals), including the C_p, L_p, global efficiency, local efficiency, assortativity, transitivity, and modularity. Only at D = 0.42, was the transitivity significantly higher in the ESRD group compared to the NC group. No other indicators showed significant differences between the two groups. However, compared to the NC group, the C_p, L_p, and assortativity increased in the ESRD group, while global efficiency, local efficiency, and modularity decreased across most densities.

AUC analysis of the global network metrics indicated that the transitivity of the ESRD group was significantly higher than that of the NC group (p = 0.036). The differences in the remaining metrics between the two groups were not statistically significant (p > 0.05).

We studied each index of the regional network between the two groups, i.e., the node network attribute indexes, including local clustering coefficient, degree, and node betwenness, and performed AUC analysis on them separately. The standardised local clustering coefficient of the ESRD group was obtained to be significantly lower than that of the NC group in the left Heschl gyrus, and significantly higher than that of the NC group in the left middle frontal gyrus, right rolandic operculum, and right precuneus. The degree of the ESRD group was significantly smaller than that of the NC group in the right precuneus, and significantly greater than that of the NC group in the right superior frontal gyrus, medial, and left Heschl gyrus. The node betweenness of the ESRD group was significantly smaller than that of the NC group in the left middle frontal gyrus, right precuneus, and significantly greater than that of the NC group in the right temporal pole (Figure 5, p < 0.05).

Hubs were quantified by the AUC values of node betweenness centrality, with hubs defined as regions where the node betweenness centrality was greater than two times the standard deviation above the network’s average node betweenness centrality. The hubs in the ESRD group included the left middle frontal gyrus, orbital part, the left olfactory cortex, the left median cingulate and paracingulate gyri, the right lingual gyrus, the left postcentral gyrus, and the right temporal pole: superior temporal gyrus. In the NC group, the hubs included the left middle frontal gyrus, the right insula, the right median cingulate and paracingulate gyri, the right precuneus, and the left superior temporal gyrus (Figure 6).

Both populations achieved network resilience in response to random and targeted attacks (Figure 7), and the network resilience in response to targeted attacks showed some degree of diminution at certain network sparsities.

The VBM-based study findings indicate that, compared to the normal control group, the cortical volume reduction in ESRD patients involves various brain regions, including the central executive network (CEN), the limbic association system, the salience network (SN), and the dorsal attention network (DAN). These regions exert influential roles in functions such as sensation, emotional regulation, and memory (Yeo et al., 2011; Androulakis et al., 2018).

Brain networks are designed for effective information transfer and processing by striking a balance between separation and integration. Graph theoretic analysis can quantitatively analyse the separation and integration of brain networks (Rubinov and Sporns, 2010). In this paper, we found that although the SCN of Non-CI ESRD patients still retains the small-world topology property, the characteristic path lengths and clustering coefficients increase in most density distributions, and the network characteristics of Non-CI ESRD patients are gradually converging to regular networks (Latora and Marchiori, 2001). Similar alterations in the topological properties of rules are also seen in brain networks of people with AD or schizophrenia disorders (Bullmore and Sporns, 2009). This shift indicates that the structural covariance network in Non-CI ESRD patients shows reduced segregation between various brain regions and compromised segregation function. This can result in outcomes like lower efficiency in regional information processing, higher cognitive load, and diminished neural network plasticity. It suggests that brain regions that previously processed information independently may now interfere with each other, and the brain requires extra resources to differentiate and process information from different regions effectively.

Transitivity is a metric used to quantify network segregation; therefore, greater transitivity indicates that ROIs tend to form highly interconnected clusters. The presence of significantly reduced transitivity has been found in studies of brain network alterations in people with systemic lupus erythematosus (Preziosa et al., 2020), cortical dysplasia (Lee et al., 2021), and those at risk for high-risk psychiatric disorders (Prasad et al., 2023). In a study of the effect of structural connectivity on cognitive performance in multiple sclerosis, Lopez-Soley et al. found that measures of transitivity of brain structural networks were positively correlated with cognitive scores (Lopez-Soley et al., 2020), suggesting that patients with multiple sclerosis with increased transitivity have better cognitive performance, and thus suggesting that increased transitivity is an enhancement of the efficacy of the whole-brain network, and that increased transitivity may act as a structural reorganisation of the brain in ESRD patients. The increased transitivity may act as a compensatory mechanism for the reorganisation of brain structures in ESRD patients to ensure that the transmission efficiency of the whole-brain network is not significantly altered.

In conclusion, this study demonstrates that the SCNs pattern of overall grey matter volume in ESRD patients is shifting from a small-world to a regular network. Certain genetic and immunological factors, such as apolipoprotein E (APOE), a complex protein crucial for neuronal repair and plasticity following injury (Samatovicz, 2000), may significantly influence brain structural network alterations. A study by Bijkerk et al. (2022) has confirmed that circulating angiopoietin-2 and decreased levels of specific microRNAs (particularly miR-132) may serve as important biomarkers in elderly ESRD patients. They are correlated with quantitative changes in white matter hyperintensities (WMH) volume and cognitive decline in the brain. These factors could lead to changes in the connectivity of the SCNs in ESRD patients, causing a reorganization of the network structure.

Analysis of node betweenness and degree measures can assess differences in brain centrality interactions between two groups. This study found significant decreases of RE in the right precuneus and the left middle frontal gyrus of the ESRD group, wherein the right precuneus showed a more pronounced decrease, resulting in a severe network loss and a significant reduction in network bridges. Many neurofunctional connectivity studies have also reported that ESRD patients have significantly reduced node mediation in the right precuneus and middle frontal gyrus, and that these regions may also be associated with neurovascular coupled NVC dysfunction, which correlates with cognitive impairment in ESRD patients (Yue et al., 2021; Hu et al., 2024). In in patients with Non-CI ESRD, the precuneus serves multiple roles, including involvement in the core of the DMN and its situational memory network as well as the paracentral gyrus network, which are critical for complex cognitive functions (Dadario and Sughrue, 2023).

The middle frontal gyrus, a highly interconnected cortical area implicated in attentional processing, working memory, and language generation and comprehension, plays a key role in the frontoparietal network (FPN) (Briggs et al., 2021). In conclusion, our study offers a macroscopic perspective on the markedly reduced covariation network capacity in the right precuneus and the left middle frontal gyrus of ESRD patients, providing a basis for the earlier detection of cognitive changes and macroscopic alterations in brain structure and network.

Non-CI ESRD patients also exhibit regions of significantly increased RE, such as the right superior medial frontal gyrus, left transverse temporal gyrus, and right temporal pole. The mechanisms behind these high centrality regions remain unclear but may be associated with protective factors identified in existing research. Aerobic exercise during dialysis sessions has been shown to reduce the incidence of cardiovascular diseases in ESRD patients, and high-volume hemodiafiltration enhances hemodynamic stability and better preserves cerebral perfusion, potentially mitigating the accelerated cognitive decline and progression of white matter lesions in ESRD patients (Isnard-Rouchon and Coutard, 2017; Kim, 2018). Further research into the relationship between brain structure or functional connectivity and clinical indicators in ESRD could elucidate the contributing causes. The current findings also indicate that potential secondary neuroendocrine factors may impact the connectivity of large-scale brain structural networks (Hosseini et al., 2012).

In the NC group, hubs were predominantly partly overlapped with the hubs observed in the brain networks of healthy adults (Ni et al., 2014).

The presence of brain reorganizational plasticity predisposes the right temporal pole and right superior temporal gyrus to serve as focal points for altered compensatory networks in ESRD patients. A recent diffusion tensor imaging DTI investigation into ESRD patients revealed a reduction in average white matter fiber tracts across the board. Notably, however, there was a significant increase in white matter fiber tracts within the bilateral superior temporal gyri (Ma et al., 2022), indicating a compensatory mechanism for white matter fibers in these regions among ESRD patients. Previous research on the diffusion tensor imaging DTI network in long-term hemodialysis patients with ESRD revealed nodes exhibiting elevated clustering coefficients (Chou et al., 2019). This increased connectivity among local nodes and enhanced efficiency of information transfer imply the existence of compensatory mechanisms in ESRD patients. This phenomenon was further supported by the analysis of functional brain networks in ESRD patients via fMRI, which identified areas of enhanced connectivity within the DMN and the SN. These findings suggest that specific connectivity changes and neuropathological alterations may be linked to cognitive compensation in ESRD patients (Chang et al., 2021; Huang et al., 2018; Wu et al., 2020).

The dynamic behaviour of a network may be closely related to its underlying topology (Kaiser and Hilgetag, 2004). Recent studies have indicated that a modular architecture is associated with the maintenance of the brain’s robustness (Chen et al., 2021). Thus, variations in network parameters reflect disruptions to the network’s general performance. This study demonstrates that compared to the control group, brain networks of Non-CI ESRD patients showed a more pronounced decrease in topological stability in the face of target attacks, which may be related to histopathological changes in the brain cortex of Non-CI ESRD patients.

This study is accompanied by certain limitations. Initially, it was a cross-sectional design, precluding the assessment of the direct dissociated effects of ESRD on network measures. Future research should consider longitudinal assessments of network metrics in patients with ESRD to mitigate this limitation. Furthermore the brain structural network changes observed in ESRD patients at the individual level merit further investigation. Concurrently, correlation analyses should be performed in tandem with pertinent clinical markers to delve into the impact of these markers on brain network alterations.