In this study, we propose a methodology to estimate the brain structural asymmetries, namely the CAI, for neurodegenerative diseases, especially FTD and AD. We used an information theory method to calculate each individual's CAI. CAI mean values differentiate healthy controls from dementia participants, and FTD from AD, and between FTD phenotypes. We studied the brain asymmetries over time and found an increase in asymmetry for FTD. Cluster analysis using the CAI values for both FTD and AD enabled us to identify disease subgroups and study the relationship between brain asymmetry and the levels of fluid biomarkers.
Our first finding, the increased asymmetry for FTD, aligns with what has been described in the literature using a mix of non-standardized measures and visual inspection. The svPPA is FTD’s most asymmetrical clinical phenotype, usually showing greater left than right atrophy of the temporal lobes [6, 8, 10, 12, 32–36]. Also, in agreement with our findings, the FTD brain asymmetry has been reported to increase over time [10, 36]. In contrast, AD is commonly described as symmetric dementia: the atrophy presented in one hemisphere of the brain, usually, is similar in the other. However, some studies reported asymmetric patterns in AD [34, 37–40]. In our study, AD patients present lower CAI when compared with FTD patients, meaning less asymmetry. However, if compared to healthy controls, they present an asymmetrical brain.
Previous works studying brain asymmetries in FTD and AD usually report these asymmetries at the visual level. Still, further studies are needed to present a methodology to evaluate and quantify systematically. Although some studies have been done to quantify asymmetries [32, 35, 38, 41, 42], we chose a more complex methodology based on information theory to estimate an asymmetric index. For this reason, we proposed to estimate the asymmetric index with the JSD measure. This novel measure based on entropy has recently been used to study similarities in different biological and clinical areas. For example, it has been used to help in the detection of ischemic stroke, to study long-term surgical outcomes in the brain for epilepsy patients, to study cancer cells to find link cells with near-identical gene expression, or to study the individual metabolic network in patients with type 2 diabetes [22, 43–45]. We demonstrated that the CAI defined here could be of help for the differentiation of the clinical expressions or dementias, for studying the progression of the disease, or for identifying subgroups.
Other studies have explored quantification strategies for brain symmetries to differentiate between different dementias successfully [32, 46]. However, there is currently no standard and accurate methodology for this, and the CAI presented here showed promising results. Also, it helped identify FTD clinical phenotypes. In this sense, svPPA presented differences compared to bvFTD. We found that the svPPA is the most asymmetric clinical expression of FTD, which is in accordance with previous literature using visual scales [8, 9]. We replicated the previous results in this study and quantified these differences using the CAI.
Our cluster analysis yielded FTD and AD participants into two subgroups according to their asymmetry indexes. As cluster analysis is an unsupervised statistical approach, we aimed to find an explanation of the identified subgroups using clinical and biomarker data. Among the two subgroups for FTD patients, the more asymmetric cluster was enriched with svPPA participants and the other cluster with nfvPPA and bvFTD. The cluster analysis was in accordance with the differences studied between FTD clinical phenotypes using ANOVAs, where we found a significant difference between svPPA and bvFTD. Overall, we observed that svPPA presented the highest CAI compared to nfvPPA and bvFTD. AD participants were also subdivided into two subgroups; however, we could not find an explanation based on the clinical profile of the subjects. We then analyzed the association between clusters and the fluid biomarkers to study further implications of cortical brain asymmetry. We found that in the FTD Cluster 1, CAI values were associated with higher levels of NfL in both CSF and plasma with respect to the bvFTD and nfvPPA and that higher CAI predicted higher NfL (CSF and plasma) in FTD patients. This suggests that NfL (CSF and plasma) is directly associated with brain asymmetry. Previous studies reported that NfL levels in CSF and plasma were associated with brain atrophy [47–49]. However, the association between NfL and brain asymmetries has not been investigated. When studying AD, we found that different CAI groups were associated with plasma GFAP levels. This biomarker has been previously associated with brain atrophy due to aging or disease severity [50, 51]. However, again, its association with brain asymmetry has not been studied before. Then, we examined the correlation between GFAP levels and CAI in AD patients and obtained a positive correlation. Overall, these associations between brain asymmetries and fluid biomarkers suggest that both contribute to defining AD subgroups.
Finally, FTD participants presented higher levels of brain asymmetry over time, suggesting that the CAI could indicate FTD progression. Previous studies have shown that FTD's different genetic or clinical expressions behave differently in becoming more asymmetric over follow-up visits [52–54]. Regarding AD, no differences in CAI over time were found for AD patients.