Subjective cognitive decline (SCD) is often one of the earliest symptoms of Alzheimer’s disease (AD) [1, 2], in which individuals report experiencing cognitive decline but still achieve normal results in objective neuropsychological assessments [3]. Increasingly recognized as a crucial early sign of the pathological processes underlying AD, this condition underscores the importance of early detection and intervention strategies to potentially alter the disease’s trajectory.

In the context of genetic influences on AD, Apolipoprotein E4 (APOE$\epsilon$4) has been identified as a significant risk factor [4]. Its impact, especially on individuals with SCD, has become a focal point of recent research [5]. Studies indicate that the APOE$\epsilon$4 allele might accelerate cognitive decline in these individuals [6], yet the intricate mechanisms underlying these effects, particularly in the context of structural brain networks, remain unclear. This study aims to delve into the influence of APOE$\epsilon$4 on SCD, building on the findings of Lee et al. [7], who observed specific changes in the fractional anisotropy (FA) values in certain brain regions of SCD individuals carrying the APOE$\epsilon$4 allele.

Diffusion-tensor imaging (DTI) is instrumental in this research, offering a window into the microstructural integrity of brain tissues by analyzing water molecule diffusion patterns. DTI’s role in AD studies has been pivotal, particularly in detecting early microstructural neuronal changes that precede more significant atrophic developments [8, 9]. Extensive quantitative DTI studies have reported various findings in individuals with SCD, ranging from significant alterations in FA and mean diffusivity (MD) in critical brain regions like the hippocampus and entorhinal cortex to contradictory results showing no statistical differences in dispersion parameters [10, 11, 12, 13, 14, 15, 16].

However, these studies primarily focused on quantifying DTI parameter differences without a broader investigation into the structural network. Building on the work of Shu et al. [17], who reported changes in global and local efficiency in the structural networks of SCD individuals, and Yan et al. [18], who found reduced efficiency and increased shortest path length in certain brain regions, our study aims to further investigate these aspects. Specifically, we seek to understand the potential impact of the APOE$\epsilon$4 allele on the structural network of individuals with SCD.

To achieve this, we analyzed data from 36 individuals without SCD and APOE$\epsilon$4 (healthy control, HC group) and 54 individuals with SCD from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database, including 21 individuals with SCD carrying APOE$\epsilon$4 (APOE$\epsilon$4+ group) and 33 individuals with SCD but without APOE$\epsilon$4 (APOE$\epsilon$4- group). The study employs an FA-based deterministic fiber tracking method and graph theory analysis to explore the white matter structural network and its characteristics across these groups, focusing on discerning the role of the APOE$\epsilon$4 allele in modulating structural network attributes in SCD.

In this study, we used graph theory to study whole-brain network characteristics and intracerebral node characteristics in the structural network in APOE$\epsilon$4+, APOE$\epsilon$4-, and HC groups. Our results showed that the three groups presented with small worldness and had statistically significant differences in Cp, Eloc, and local node properties in some brain regions, including the default mode network (DMN), occipital lobe, temporal lobe, and subcortical regions. Moreover, different intracerebral node characteristics were observed between the APOE$\epsilon$4+ and APOE$\epsilon$4- groups.

In all three groups, the brain structural network had high Cp ($\gamma$ $>$ 1) and close Lp ($\lambda$ $\approx$ 1), with the characteristics of a small-world network. The Cp and Eloc in APOE$\epsilon$4+ and APOE$\epsilon$4- groups were significantly lower than those in the HC group. However, no statistical difference was observed between the APOE$\epsilon$4+ and APOE$\epsilon$4- groups. Cp measures the local information integration capability of the network and reflects local efficiency. Brier et al. [28] found that the Cp changed with the progression of the disease and was associated with the known cognitive impairment in AD patients. Daianu et al. [29] and Lo et al. [30] demonstrated higher Lp and lower Eg in the structural network of patients with AD. This study showed the reduction of only Cp and Eloc, with no changes in Lp or Eg, which further indicated that SCD included mild neuron injury. However, there existed compensation in the structural network; thus, the global efficiency of the whole structural network was not significantly changed.

This study identified significant changes in node Cp and node Eloc in some brain regions of the structural network in both APOE$\epsilon$4+ and APOE$\epsilon$4- groups, mainly including the bilateral cuneate lobes, the left supplementary motor area, and the left paracentral lobule. The bilateral cuneiform lobes are located on the medial side of the occipital lobe and form a primary visual area between the spur sulcus and the parietal-occipital sulcus and the surrounding cortex [24]. The left supplementary motor area is located on the medial side of the left frontal lobe. Cona and Semenza [31] found that the supplementary motor area involves perceptual and productive tasks (e.g., time/language/music perception and production) but has no association with the information to be processed (e.g., spatial working memory and linguistic working memory). The left paracenter lobule is located in the parietal lobe and belongs to the DMN. The DMN includes a wide range of the frontal lobe and the posterior midline and inferior parietal lobule. It is not a single network but rather consists of multiple parallel intersecting networks [32]. Furthermore, it performs functional activities in the human brain at rest, including those related to emotional processing, self-introspection, and episodic memory extraction [33]. Hahn et al. [34] found that the DMN of patients with AD was subject to selective and gradual damage. Previous studies on DMN changes of SCD mainly focused on the resting state functional MRI, and only a few studies have been performed on the white matter structural network. However, Horn et al. [35] observed a high similarity between functional and structural connectivity in the DMN region. In studies on resting state functional MRI, Verfaillie et al. [36] and Hafkemeijer et al. [37] showed that the functional connectivity in the DMN of individuals with SCD was increased; however, Wang et al. [38] and Dillen et al. [39] observed less functional connectivity in the DMN of individuals with SCD. The heterogeneity of functional connectivity in the DMN may be associated with compensation and its degree in individuals with SCD. In a study on the white matter structural network, Shu et al. [40] noted lower local node efficiency in the paracentral lobule. Moreover, Ye et al. [41] studied the abnormal connection mode in the structural network using multivariable distance matrix regression and found that the supplementary motor area was abnormal in patients with mild cognitive impairment (MCI), whereas the DMN was abnormal in patients with AD. The results of the present study indicated that individuals with SCD had similar changes to those with MCI and AD, and there were differences between the APOE$\epsilon$4+ and APOE$\epsilon$4- groups. These findings suggest that functional integration and separation are damaged in the structural network at the SCD stage [42], and the structural network is affected by APOE$\epsilon$4.

Degree centrality was significantly changed in the DMN, occipital lobe, and temporal lobe in both the APOE$\epsilon$4+ and APOE$\epsilon$4- groups. Node centrality is the number of edges shared by a node and other nodes in the network, and it used to measure the importance of a single node in the network. Brain regions with any change in DMN centrality included the right gyrus rectus, the right intraorbital superior frontal gyrus, the left paracentral lobule, and the right supramarginal gyrus, in agreement with previous studies [41, 43]. Moreover, the brain regions with temporal lobe changes included the right superior temporal gyrus and left middle temporal gyrus. The right superior temporal gyrus involves language processing, connects auditory and visual pathways [44], and is sensitive to emotional information [45]. The left middle temporal gyrus plays a role in the process of memory. Brascamp et al. [46] stimulated the middle temporal gyrus cortex using transcranial magnetic stimulation and found that it was associated with the disruption of memory trace, suggesting that abnormalities of the middle temporal gyrus may result in lower memory confidence. The present study also found that changes in the internal structural network in the DMN and temporal lobe were different between the APOE$\epsilon$4+ group and the APOE$\epsilon$4- group, which may be associated with the different compensatory pathways of the two groups after structural network destruction.

Betweenness centrality was significantly changed in the occipital lobe, DMN, and basal ganglia between the APOE$\epsilon$4+ group and the APOE$\epsilon$4- group. Betweenness centrality is the number of shortest paths cutting through this node among all other pairs in the whole network, and it reflects the node’s importance in the network. The DMN region includes the right gyrus rectus. The betweenness centrality in the DMN region was increased in the APOE$\epsilon$4+ group and reduced in the APOE$\epsilon$4- group. These different results observed in the two SCD groups explain the findings of previous studies [47, 48, 49]. The reasons for the heterogeneity in the DMN region are associated with the division of study subjects and their inclusion criteria. The occipital lobe includes the bilateral cuneate lobes, the right middle occipital gyrus, the right superior occipital gyrus, and the right lingual gyrus. The betweenness centrality in the right superior occipital gyrus was increased in the APOE$\epsilon$4+ group and reduced in the APOE$\epsilon$4- group. The betweenness centrality in the right middle occipital gyrus was reduced in both the APOE$\epsilon$4+ and APOE$\epsilon$4- groups, with a more significant reduction in the APOE$\epsilon$4- group. Moreover, the betweenness centrality in the right lingual gyrus was increased in both the APOE$\epsilon$4+ and APOE$\epsilon$4- groups, with a more significant increase in the APOE$\epsilon$4- group. Hayes et al. [50] identified fewer occipital cortex activities in memory coding. However, Viviano et al. [15] found no significant difference in the occipital lobe. The caudate nucleus is closely related to repetitive stereotyped behaviors and is a component of the cortical–striatal–thalamocortical neural circuit [51]. In the present study, the betweenness centrality in the right caudate nucleus increased in both the APOE$\epsilon$4+ and APOE$\epsilon$4- groups, with a more significant increase in the APOE$\epsilon$4- group. The different node characteristics of the two groups of individuals with SCD may be related to different degrees of compensation generated to achieve network balance after an early injury.

APOE is the main lipid carrier of the central nervous system, and APOE$\epsilon$4 is the most potent genetic risk factor for sporadic AD [52]. APOE$\epsilon$4 affects not only the clearance and aggregation of A$\beta$ but also neuron growth, membrane repair and remodeling, and synaptogenesis and neuroinflammation [53]. The present study found that the structural network was different between the APOE$\epsilon$4+ and APOE$\epsilon$4- groups, indicating that APOE$\epsilon$4 had an impact on the structural network, although the specific mechanism requires further research.

Based on the preliminary nature of this study, the following limitations should be considered. First, the sample size was small. Although the ADNI database contains a large amount of data, there were few individuals with SCD with T1-weighted images, DTI, and APOE$\epsilon$4, which may weaken the statistical power of this study. Second, this study is a cross-sectional study without clinical follow-up. By expanding the sample size and using a more diverse range of cognitive assessment tools in longitudinal follow-up studies, future research may reveal meaningful associations that we have not been able to observe at present. Third, to study the mechanism of the effect of APOE$\epsilon$4 on the white matter network, future studies should be performed with live animal models, supplemented by molecular imaging and molecular biological data for confirmation. Fourth, the original version of the AAL atlas [17] was selected. Nevertheless, the latest version of AAL3 [54] adds many previously undefined areas of the brain that are interested in many neuroimaging studies, such as a new version of anterior cingulate, thalamus, and brain nuclei (nucleus accumbens, substantia nigra, ventral tegmental area, red nucleus, locus coeruleus, and raphe nuclei). AAL3 can be used as a toolkit for statistical parametric mapping (SPM) and with MRIcron. AAL3 has an isotropic voxel size of 2 $\times$ 2 $\times$ 2 mm. In addition, the voxel sampling size of 1 $\times$ 1 $\times$ 1 mm is provided. The AAL3 atlas should be selected for future studies. Additionally, while we have implemented measures to alleviate the impact of connection density, such as empirically selecting a threshold for fiber connections and retaining only those observed in at least 80% of participants, we have not directly analyzed the implications of variations in connection density on the results. Future research could delve deeper into investigating the impact of connection density on topological property analysis and employ a more comprehensive approach to address this factor. Finally, based on the DTI, we constructed a structural network instead of a resting state functional network. Future studies should further combine structural and functional networks for multimodal MRI studies and further explore the effects of APOE$\epsilon$4 on the brain functional networks of individuals with SCD in multimodal MRI.

In this study, the DTI of individuals with SCD showed structural connection changes in the DMN, the occipital lobe, the temporal lobe, and subcortical regions. APOE$\epsilon$4 affects the structural network, but its mechanism should be further studied. In practice, clinicians can use DTI as a detection indicator to identify individuals with SCD.

All raw data used in this study were from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). Data used in preparation of this article were obtained from the ADNI database (http://adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf.

These should be presented as follows: SD and WL designed the research study. SD and WC performed the research. SF and HL provided help on analyzing the data. SD and WL wrote the manuscript. All authors contributed to editorial changes in the manuscript. All authors read and approved the final manuscript. All authors have participated sufficiently in the work and agreed to be accountable for all aspects of the work.

The ADNI passed a review by the Institutional Review Board (IRB), and all subjects provided written informed consent and unified inclusion criteria (https://adni.loni.usc.edu/). We obtained the approval of our local ethics review committee (Dongguan Eighth People’s Hospital medical ethics committee AF/SC-15/01.0) before the analysis began.

We thank LetPub (https://www.letpub.com) for linguistic assistance.

This research was funded by Guangdong Medical Science and Technology Research Fund, grant number 20231129323888.

The authors declare no conflict of interest.