Partial least square regression model identified 5 clusters of proteins that correlated with distinct tauopathy
To identify possible biomarkers or mechanisms that are distinct among CTE, AD, PSP, CBD, and AGD, a 71 immune related protein multiplex ELISA was performed on anterior cingulate cortex grey matter. Using PLS regression analysis, 5 clusters of proteins were identified and correlated against each tauopathy to determine what clusters were related to specific disease (Figure 1). Cluster 1 was most correlated with CTE, Cluster 2 was most correlated with AD, Cluster 3 was most correlated with PSP, Cluster 4 was most correlated with CBD, and Cluster 5 was most correlated with AGD. Although clusters 4 and 5 demonstrated higher specificity for CBD and AGD respectively, there was minor overlap with the other tauopathies as well.
Next, each cluster was analyzed to examine which specific proteins were driving the cluster classification (Figure 2) and the top 5 proteins for each cluster were identified (Table 2). The most correlated protein for Cluster 1/CTE was 6Ckine (0.36), then followed by ENA78 (0.27), BCA1 (0.24), GMCSF (0.25), then TARC (0.23). In Cluster 2/AD, the highest correlated protein was FLT3L (0.31), then followed by IL17F (0.25), TARC (0.23), IL15 (0.23), then IL17E/IL15 (0.21). In Cluster 3/PSP, the highest correlated protein was IL13 (0.37), followed by SCF (0.25), BCA1 (0.25), MIG/CXCL19 (0.24), then IL3 (0.24). In Cluster 4/CBD, the highest correlated protein was IL1β (0.37), followed by MDC (0.37), MIG/CXCL19 (0.21), ILF (0.2), then VEGFα (0.19). Finally, in Cluster 5/AGD the highest correlated protein was VEGFα (0.33), followed by MCP1 (0.32), MIG/CXCL19 (0.31), FLT3L (0.28), then GMCSF (0.22).
The top 5 proteins in each cluster could identify distinct neurodegenerative pathologies
Using the top 5 proteins from each cluster, receiver operating curve (ROC) analyses was performed to examine how specific and sensitive each protein was for identifying distinct tauopathies (Figure 3). ROC analysis of the top 5 Cluster 1 proteins for the ability to identify CTE demonstrated that 6CKine had the highest area under the curve (AUC) (0.854, p < 0.001). ENA78, BCA1, GMCSF, and TARC all were significant as well (Figure 3A). GMCSF was the only Cluster 1 protein that had a significant AUC under 0.5 demonstrating it is found at lower levels in CTE compared to the other diseases. The top 5 proteins in Cluster 2 were used to identify AD. FLT3L had the highest AUC (0.670, p = 0.001), followed by IL15 (0.634, p = 0.017), then TARC (0.634, p = 0.015). IL17F (0.297, p < 0.001) had a significant AUC under 0.5 demonstrating decreased expression in AD compared to the other diseases (Figure 3B). Next, the top 5 proteins in Cluster 3 were used to identify PSP. IL13 was observed to have the highest significant AUC (0.777, p < 0.001), followed by SCF (0.640, p = 0.49), and IL3 (0.675, p = 0.003). MIG/CXCL9 (0.275, 0.001) had an AUC under 0.5 demonstrating less protein expression in PSP compared to other diseases (Figure 3C). Cluster 4 and Cluster 5 did not have any significant AUC values when used to identify CBD (Figure 3D) and AGD (Figure 3E) respectively.
The protein with the highest AUC values in each cluster was then selected and used to examine the ability to identify each respective tauopathy when comparing against all 127 cases together while controlling for age at death and gender. Using binary logistic regression analysis, 6CKine significantly correlated with a positive diagnosis of CTE (OR = 1.206, p < 0.001) independently of age at death (OR = 0.944, p = 0.036) or gender (OR = 0.00, p = 0.997). Next, it was observed that FLT3L, although not reaching significance, trended towards being able to predict AD (OR = 1.13, p = 0.054) independently of age at death (OR = 1.109, p < 0.001) or gender (OR = 2.679, p = 0.42). Cluster 3’s IL13 demonstrated significant ability to identify a PSP (OR = 1.313, p < 0.001), independently of age at death (OR = 0.970, p = 0.276), or gender (OR = 2.076, p = 0.183). Unlike Clusters 1-3, the protein with the highest AUC in Cluster 4 was not the top protein in the PLS analysis. LIF had the highest Cluster 4 AUC, but IL1β was the top PLS correlated protein. However, neither IL1β (OR = 0.965, p = 0.816) or LIF (OR = 1.244, p = 0.232) correlated with a diagnosis of CBD. Similarly, for Cluster 5, GMCSF had the highest AUC but VEGFα was the top PLS correlated protein. While VEGFα (OR = 1.046, p = 0.246) was not correlated with a diagnosis of AGD, GMCSF was correlated with a diagnosis of AGD (OR = 5.148, p = 0.007) independently of age (OR = 1.132, p < 0.001) and gender (OR = 0.776, p = 0.676).
6Ckine was able to identify CTE using postmortem CSF
Finally, to examine if the results taken from the brain homogenate can be extrapolated to fluids such as CSF, postmortem CSF from individuals with AD and CTE was obtained and a 6Ckine ELISA was performed. When comparing the total concentrations, individuals with CTE were observed to have significantly more 6Ckine compared to individuals with AD (p = 0.02) (Figure 4).