Recent studies have largely focused on identifying the qualitative and quantitative features of SCD that are specifically related to the underlying AD pathology [
26]; in contrast, previous contradictory results of FDG-PET in this group have been somewhat neglected. ROI-based glucose metabolic biomarkers are easily extracted and quantified and may have clinical meaning, and this method has also been frequently used in AD research [
27‐
29]. Several research groups tried to explore the ROI-based metabolic pattern of SCD, and the results showed that its metabolism was similar to that of NC [
30,
31] and not influenced by Aβ deposition [
32]. Other studies performed voxel-based analyses, suggesting that individuals with SCD had a significant reduction in glucose metabolism in the periventricular regions [
33] or other scattered areas [
13] or had no metabolic changes [
34,
35] when compared with NCs. Importantly, one study performed both ROI-based and voxel-based analyses, and the results showed that hypometabolism of the right precuneus is a typical feature of SCD [
12]. The large intra-group differences among SCD samples and mismatches between groups may explain some of the inconsistent results; for example, the age span of SCD subjects in one study reached 31 years, from 53 to 84 years old [
34], and in another study, the presence of the APOE ε4 allele in the SCD subjects was 0%, while it was 52% in the NCs [
30]. In addition, it should be noted that the ROIs selected in previous studies were all based on prior knowledge, which may be affected by different samples, and cannot reflect the characteristics of the preclinical state; a whole brain study without hypotheses can avoid this problem to some extent and help detect early limited functional changes. In our study, we have included data from two centers, and all data strictly followed the inclusion criteria of our project. We performed methodological optimizations to increase the robustness of the results and to settle previous disputes. First, we combined undifferentiated cortical ROI-based analysis and voxel-based analysis. Second, multiple permutation tests and test-retest methods of data from the two centers were performed. Third, the selected hypometabolic region was further verified by repeated cross-validations. Therefore, we thought the glucose hypometabolism of RMTG reflected by FDG-PET is likely to be a reliable biomarker of SCD in methodology and be a good additional index for the inclusion of SCD since it can reduce the individual errors caused by subjective descriptions to a certain extent.
Previous studies have not reported the metabolic changes of RMTG in individuals with SCD. This novel finding may provide a new perspective for the disease changes of SCD or the spectrum of AD. According to previous reports, the MTG region has close functional connectivity with the hippocampus [
36], is primarily involved in verbal or semantic cognition, and is also associated with oral memory [
37]; furthermore, it represents a signature area of cortical atrophy in patients with symptomatic stages of AD [
38‐
41]. These results indicate the important roles of MTG in the Alzheimer’s continuum. Furthermore, the RMTG in the AAL template is an important part of the default mode network [
42], and a study performed by Lim et al. previously showed that the RMTG has already developed slight atrophy as early as the SCD stage [
43], providing a structural basis for our results. Specifically, we supposed that structural atrophy may be due to the death of neurons and then lead to a decrease in metabolism. Thus, it seems reasonable that the hypometabolic region of SCD is located in the RMTG. In our study, the metabolic difference of MTG was only observed on the right side, which is consistent with the atrophy side of MTG in the SCD subjects [
43] and is also consistent with the side of the MTG in the default network [
42]; however, the specific reason for this is still unclear currently, which may be related to the laterality, and there may be some unexplained disease-related mechanisms in the right hemisphere. Previous studies have found that the function and atrophy patterns of the bilateral temporal lobes were asymmetric in patients with neurodegenerative diseases [
44‐
48], and cerebral glucose metabolism in the bilateral hemispheres was also significantly different in healthy individuals [
49]. To further verify the clinical rationality, individuals with other stages on the cognitive continuum, other than NC and SCD, were also enrolled to make cross-sectional comparisons. Although the average age of patients in the ADD group was lower than that in the other groups, the metabolism of RMTG was still gradually decreased across the cognitive continuum, suggesting that the decrease in metabolism might be due to cognitive changes instead of the influences of aging, and the RMTG was damaged as early as the SCD stage and the damage gradually progressed, accompanied by cognitive deterioration. Importantly, compared with the PCC and PCUN, which are thought to be characteristic hypometabolic regions of AD [
8‐
10,
50], the ROC analysis showed that the metabolism of RMTG was better in distinguishing NC from SCD, as well as NC from symptomatic patients. These results support our hypothesis to some extent that the hypometabolic abnormalities in the dementia stage may start from a local area and then gradually spread into signature regions. In other words, the RMTG may be the seed region. Considering the heterogeneity of SCD, we also included patients with DLB; since ADD and DLB are similar in metabolic patterns [
51], it is understandable that there was no metabolic difference in the RMTG between them. We noted that the ability of RMTG to distinguish NC from SCD was not outstanding (AUC = 0.638–0.717), but we thought it was within a reasonable range when compared with previous studies of monoparametric MRI [
52‐
54]. For example, Peter et al. proposed a multivariate pattern recognition framework integrating the gray matter atrophy pattern in the differentiation of SCD from NC, and the AUC was 0.67 [
52]. The studies performed by our group suggested that the classification performance was approximately 60% for diffusion tensor imaging [
54] and approximately 70% for functional MRI [
53,
54]. From another perspective, we found that the glucose metabolism of RMTG was correlated with the abilities of delayed memory, which was consistent with previous reports [
35,
55,
56]. Other studies have suggested relationships between the structure and function of this area and emotions [
57,
58], which also confirms our results that the metabolism of RMTG is related to depression. We also observed that the degree of complaints was negatively correlated with metabolism, which was supported by a recent study where the degree of self-reported SCD was negatively correlated with glucose metabolism in the temporal and parietal regions [
59]. These correlation results showed that the RMTG is involved in a variety of cognitive processes and further explained the rationale for the involvement of this area. Although the correlation between RMTG metabolism and Aβ deposition did not reach a significant level, it echoes previous reports showing that the reduction in glucose metabolism in AD-sensitive areas is not directly related to Aβ deposition [
60,
61], and other evidences suggesting that the elevated brain Aβ deposition alone is probably insufficient to produce neuronal damage and cognitive changes [
62,
63]; the correlation between brain Aβ deposition and metabolism is likely to be mediated by neurofibrillary tangles with a temporal delay [
60]; however, this was not proven in this study. Based on the above, we thought that the hypometabolism of RMTG in SCD is also reliable in the practical sense, and it may indicate the initiation of nerve injury, and the deterioration of cognitive function in the future.
Several limitations of this study should be addressed. First, SCD is a heterogeneous state and it is affected by the cultural background [
14,
15]. Previous studies on its metabolism were inconsistent [
12,
13,
30,
31,
33‐
35], and our results led to new conclusions. Therefore, it should be noted that the participants in our research were all Chinese community-sourced, female-dominated, and comparatively young, whether the hypometabolism of RMTG is a common or unique feature of this population needs to be further confirmed. Second, the amyloid information was not available for the full dataset, and not all the SCD subjects were AD sourced. Third, the small sample number limited the statistical power of our data. We tried to overcome this issue by enrolling participants from another subcenter, but the requirement for FDG-PET data greatly limited the quantity of potential participants in the SILCODE project. Fourth, our study is cross-sectional, and long-term longitudinal follow-up data can further support our conjecture, which is our future research direction. Fifth, we calculated the metabolism with the whole cerebral cortex as reference regions, and choosing different reference areas may affect the results. Sixth, the metabolic difference of MTG was only on the right side, but the current evidence is insufficient to provide a clear explanation for this, and we will continue to explore cerebral functional laterality in future studies. Considering the shortcomings of our research and the limitations in this field, multicenter collaboration to include more confounding factor-matched and pathology-identified participants is needed in the future.