As the major cause of vascular cognitive impairment, cerebral small vessel disease (CSVD) has been progressively acknowledged as a dynamic whole brain disease because of the intricate and widespread microvascular pathology, coupled with variable rates of progression of the related lesions (Hamilton et al., 2021; Jokinen et al., 2020; Shi and Wardlaw, 2016). With the development of neuroimaging research on CSVD, several consistent imaging features, including recent small subcortical infarcts, lacunes, white matter hyperintensities (WMH), perivascular spaces, microbleeds, and brain atrophy have been identified as factors associated with cognitive decline (Wardlaw et al., 2013b). Neuroimaging is highly expected to be used as an alternative to complex neuropsychological scales such as Montreal Cognitive Assessment (MoCA), Mini-Mental State Examination (MMSE), and various cognitive domain-specific scales, to provide a more objective and patient-tolerated assessment method. Nonetheless, to date, there remains a paucity of a unified and efficacious imaging strategy for assessing the cognitive performance of CSVD patients, primarily due to the intricate nature of lesion combinations and their distribution (Jokinen et al., 2020; Wardlaw et al., 2013b). Moreover, diffusion tensor imaging (DTI) studies have suggested that CSVD induces extensive microstructural changes, which cannot be detected by conventional neuroimaging.

The brain network theory provides an additional avenue for a comprehensive assessment and understanding of the progression of cognitive impairment related to CSVD from a global perspective. Recent studies have demonstrated that brain structural network efficiency can serve as a sensitive imaging indicator for evaluating cognitive dysfunction in CSVD (Boot et al., 2020; Tuladhar et al., 2020). The brain structural network efficiency has also been proposed to mediate cognitive deficits along with various macro and microstructural brain injuries, providing a systemic perspective on brain lesions (Du et al., 2019; Lawrence et al., 2014). The global efficiency gauges the network’s capability for global information integration, which has been reported to be associated with cognitive impairment in CSVD (Tuladhar et al., 2016a). Local efficiency assesses network segregation, specialization, and fault tolerance, which has also been linked to cognitive impairment in CSVD (Latora and Marchiori, 2001). However, alterations in network efficiency may not fully account for the impact of CSVD on cognition, as observed in cases with significant CSVD lesions but maintain normal cognitive function. It is possible that brain reserve and compensatory mechanisms play pivotal roles in the pathogenesis of CSVD, suggesting that the effects of CSVD on cognitive impairment are not linearly imposed (Telgte et al., 2018).

Brain reserve has been used to reflect the neurobiological status of the brain (numbers of neurons, synapses, etc.) at any point in time (Stern et al., 2020). The redundancy is defined as “the amount/strength of alternate connections” within a network, which is considered the most likely measure related to the brain reserve (Navlakha et al., 2014; Tononi et al., 1999). Previous studies have suggested redundancy as a neuroprotective mechanism in neurodegenerative diseases such as Parkinson’s disease (PD), Alzheimer’s disease (AD), and normal aging (Arkadir et al., 2014; Langella et al., 2021; Sadiq et al., 2021). Brain redundancy is expected to maintain brain efficiency and cognitive functions. However, in the individual brains, network efficiency (shortcuts in the network) and network redundancy (multiple alternative paths, not guaranteed by shortcuts) should both be implemented under the given wiring costs (the number of connections, measured by network sparsity). Previous studies have indicated that CSVD-related damage to white matter integrity results in a reduction in the sparsity of the white matter network (Reber et al., 2021; Tuladhar et al., 2016a). Therefore, progressively decreasing wiring costs during CSVD progression should force a trade-off between brain network topologies that reflects the processes by which the brain maintains cognitive function despite of existing pathology (Ma et al., 2023). In other words, when the sparsity of brain structural network decreases due to various CSVD-related brain injuries, the efficiency of brain network may not decrease linearly. Because of brain reserve such as redundancy or compensatory mechanisms, the efficiency of brain network may decline at different rates in different stages of the disease, thus exhibiting different cognitive functional states.

So far, the relationship among network sparsity, global and local efficiency, redundancy, and cognitive functions has not been elucidated across different stages of cognitive impairment in CSVD. Analyzing the changes of these brain structural network configurations in different stages of cognitive impairment will help us to understand the robustness of the brain network in CSVD patients.

In this study, we explored the hypothesis that alterations in brain structural network across different stages of cognitive impairment in CSVD display heterogeneity, indicative of the robustness of a balanced network topology.

Intergroup analysis revealed significant differences in network characteristics among HC and different CSVD groups. The NCI group exhibited a significant reduction in sparsity and redundancy compared to the HC group, indicating that CSVD-related damage primarily affected redundant connections during the early stages of the disease. Eloc and Eg showed no statistically significant changes in the NCI group compared to the HC group, implying that network integration and segregation remained relatively unaltered during the early stages of CSVD. Consequently, the cognitive function appeared to be maintained within normal parameters. The alterations observed in the brain structural network topology within the NCI group align with the principles of the brain reserve theory. Brain reserve is typically conceptualized as neurobiological capital that enables individuals to effectively withstand brain damage before clinical or cognitive manifestations become evident (Stern et al., 2020). Network redundancy appeared to function as a brain reserve factor in the early stage of CSVD, exhibiting resilience against cognitive impairment. The reduction in redundant connections could arise from either compensatory mechanism responding to impaired effective connections or potentially result from random attacks inflicted upon the brain structural network by CSVD-related brain injury. In either scenario, it is theoretically conceivable that redundant connectivity may offer a degree of delay in the onset of cognitive impairment associated with CSVD-related brain damage. Further investigations, such as longitudinal studies, are needed for substantiating this hypothesis. In the MCI group, Eg exhibited a significant decrease when compared to the NCI group, indicating a disruption in network integration that potentially contributes to cognitive impairment. The association between the decline of brain structural network Eg and CSVD related cognitive impairment has been consistently reported in previous studies (Boot et al., 2020; Lawrence et al., 2014; Tuladhar et al., 2016b). For example, a cross-sectional study found that both Eg and Eloc were significantly disrupted in CSVD and associated with cognitive function. The association of Eg with cognition were stronger than other MRI measures including conventional DTI measures (Lawrence et al., 2014). Longitudinal study showed that patients developed dementia had lower Eg and Eloc at baseline as compared with the group without dementia. Lower Eg increased the risk of incident all-cause dementia (Tuladhar et al., 2016b). In alignment with prior research, our study highlights a significant reduction in Eg within the MCI group when compared to the NCI group. However, marginal changes were observed in redundancy and Eloc. These findings might suggest the preservation of reserve connections and localized information exchange within the brain network in MCI group. These factors may contribute to the relative preservation of cognitive function in individuals with MCI, potentially delaying the onset of dementia. The observation that Eg and Eloc exhibit distinct impacts on cognitive change diverges from the outcomes of prior studies, which considered Eg and Eloc as homogenous network characteristics (Boot et al., 2020; Lawrence et al., 2014; Tuladhar et al., 2016a). Drawing from topological theory, Eloc is indicative of the network’s resilience against random attacks, suggesting its potential protective role in cognitive function. Prior studies did not thoroughly consider the interplay among different network attributes within the context of wiring costs. Further exploration of the relationships among these attributes can provide insights into the hypothesis regarding the protective influence of Eloc and redundancy. In our present study, we regulated network sparsity in consideration of wiring costs. The results showed that individuals with lower Eg exhibited relatively higher Eloc within each CSVD group and higher redundancy within MCI group, which substantiated our supposition. These observations signify the presence of brain reserve or compensation mechanisms, characterized by alternative connections and enhanced local information exchange to mitigate potential disruptions. This hypothesis needs to be confirmed by more compelling and direct evidence, such as by observing the dynamic changes in these network characteristics by conducting longitudinal studies in the future. All of the network characteristics were decreased in the VaD group compared to the MCI group, which indicated severe destruction of the network. This result posited that a reduction in Eloc and redundancy could also potentially play a pivotal influence on the pathogenesis of dementia in CSVD, thus extending the findings of prior research that emphasized the sustained decrease in Eg as a significant contributing factor to dementia development in CSVD (Tuladhar et al., 2016b). The collective reduction in sparsity, redundancy, Eg, and Eloc could suggest the depletion of brain reserve, potentially signifying the irreversibility of VaD.

We found significant correlations between network characteristics and cognitive function at different stages of CSVD. In the NCI group, network sparsity and redundancy exhibited a significant positive correlation with MoCA. This suggests that, even in the early stages of CSVD when patients have not yet exhibited significant cognitive impairment, the overall cognitive performance of individuals remains contingent on the structural organization of the brain network. The findings in the NCI group are of particular significance, as prior studies predominantly concentrated on CSVD cases characterized by pronounced clinical cognitive impairment. The concept of a preclinical stage of cognitive impairment has been introduced to describe CSVD patients without evident cognitive impairment symptoms despite the substantial burden of pathological lesions, who remain at a high risk of transitioning to MCI (Wardlaw et al., 2013a). The findings in the NCI group within this study offer valuable insights into this concept. No significant correlations were observed between Eg and MoCA, as well as between Eloc and MoCA within NCI group. This could be attributed to the initial loss of redundant connections in the early stages of CSVD, which, in turn, delays the reduction in network efficiency. It is not until the disease progresses to the MCI stage that the decline in Eg starts to make a noteworthy contribution to cognitive impairment. In the MCI stage, the influence of brain network sparsity and redundancy on cognitive function diminishes, while network redundancy exhibits a degree of counteraction to the decline of Eg. The phenomenon of changes in brain network redundancy not being synchronized with the decline in Eg in both NCI and MCI groups may be attributed to the brain’s capacity to adapt its internal network structure in response to varying degrees of injury. However, during the VaD stage, there was no discernible linear correlation between the sparsity, redundancy, and Eg of the brain network and cognitive function. This observation suggests that the structure of the brain network becomes increasingly disorganized in VaD stage. Notably, Eloc displayed a significant negative correlation with cognitive function, signifying that worse cognitive function was associated with higher local efficiency. This observation may suggest that, with the exacerbation of cognitive impairment in the VaD stage, the brain network structure tends to become increasingly segregated. Theoretically, the capability of segregated information processing not only allows for flexible and rapid reconfiguration of the brain network in response to various task demands, but also benefits network robustness (Sporns and Betzel, 2016; Wig, 2017). The seemingly paradoxical negative correlation between Eloc and cognitive function in this study can be elucidated within the framework of the trade-off theory, which is postulated as a fundamental principle of the human connectome. This theory suggests that there are trade-offs among wiring costs (represented by network sparsity), integration (represented by Eg), and segregation (represented by Eloc) within the human brain network (Bullmore and Sporns, 2012; Ma et al., 2023). Effective communication within the human brain necessitates not only efficient global information integration but also the ability for segregated information processing, which is a key element of the small-worldness concept (Sporns, 2013). For a given wiring cost, a competition may arise between Eloc and Eg. Hence, in the context of a substantial reduction in wiring costs, the significant negative correlation between Eloc and cognitive function observed within the VaD group suggested that the brain is counterbalancing Eg by enhancing Eloc. The more pronounced this trade-off, the more severe the cognitive impairment. Through partial correlation analysis (while controlling for wiring cost), we detected a trade-off relationship between Eg and Eloc at different stages of cognitive impairment in CSVD. This finding indicates that the disequilibrium between brain network integration and segregation may persist throughout the entire course of the disease.

Our findings, as derived from intergroup comparisons and correlation analyses, corroborate our initial hypothesis regarding the heterogeneous alterations in the brain structural network across different stages of cognitive impairment in CSVD. By meticulously examining the wiring cost, this investigation delves into the intricate relationships among the attributes of the brain structural network. These discoveries provide empirical support for the theories related to brain reserve and the concept of a brain network trade-off. The robust network system in CSVD reaffirms the notion that the analysis of brain structural networks is a promising method for evaluating cognitive impairment caused by CSVD, surpassing traditional macroscopic imaging assessments. Although previous studies have demonstrated the advantages of structural network efficiency in assessing cognitive impairment due to CSVD, and the complete mediating effect of network efficiency between traditional MRI features and cognitive impairment, they have overlooked the roles of reserve and compensation in neuroprotection, as well as the complex relationships among network characteristics (Boot et al., 2020; Du et al., 2019; Lawrence et al., 2014). Our present study filled this gap in CSVD research.

This study presents several limitations. First and foremost, it is essential to acknowledge that this research was cross-sectional in nature and featured relatively modest sample sizes in certain CSVD groups. Consequently, the conclusions reached herein warrant further validation through longitudinal studies involving more extensive participant cohorts. Second, it is worth noting that the absence of a universally agreed-upon methodology for defining nodes and connection weights when constructing brain structural networks based on DTI introduces the potential for methodological bias, as distinct approaches to network construction can yield varying research outcomes. Third, despite employing comprehensive clinical histories, imaging analyses, and neuropsychological assessments to mitigate the influence of AD, it is impossible to entirely rule out the impact of mixed dementia in this study. Lastly, given the diverse presentations of cognitive dysfunction arising from CSVD, there is a need for subsequent studies to concentrate on specific cognitive domains through subnetwork analysis. This necessitates the inclusion of larger sample sizes encompassing CSVD-related cognitive impairment across different cognitive domains.

In summary, this study furnishes evidence of heterogeneous alterations within the brain structural network across distinct stages of cognitive impairment in CSVD. At the NCI stage, the disconnectivity of the brain structural network predominantly manifests as the loss of redundant connections. In contrast, at the MCI stage, the decline in Eg takes on paramount significance in its contribution to cognitive impairment, while network redundancy, regarded as a form of brain reserve, presents a certain degree of counteraction to the decline in Eg. The prominent manifestation in the VaD stage encompasses a decline in all network characteristics. Notably, there are trade-offs among brain network wiring cost, integration, and segregation across the various stages of cognitive impairment in CSVD.