Targeted brain proteomics uncover multiple pathways to Alzheimer's dementia

Abstract

Objective: Previous gene expression analysis identified a network of coexpressed genes that is associated with β-amyloid neuropathology and cognitive decline in older adults. The current work targeted influential genes in this network with quantitative proteomics to identify potential novel therapeutic targets.

Methods: Data came from 834 community-based older persons who were followed annually, died, and underwent brain autopsy. Uniform structured postmortem evaluations assessed the burden of β-amyloid and other common age-related neuropathologies. Selected reaction monitoring quantified cortical protein abundance of 12 genes prioritized from a molecular network of aging human brain that is implicated in Alzheimer's dementia. Regression and linear mixed models examined the protein associations with β-amyloid load and other neuropathological indices as well as cognitive decline over multiple years preceding death.

Results: Average age at death was 88.6 years. Overall, 349 participants (41.9%) had Alzheimer's dementia at death. A higher level of PLXNB1 abundance was associated with more β-amyloid load (p = 1.0 × 10^-7 ) and higher PHFtau tangle density (p = 2.3 × 10^-7 ), and the association of PLXNB1 with cognitive decline is mediated by these known Alzheimer's disease pathologies. On the other hand, higher IGFBP5, HSPB2, and AK4 and lower ITPK1 levels were associated with faster cognitive decline, and, unlike PLXNB1, these associations were not fully explained by common neuropathological indices, suggesting novel mechanisms leading to cognitive decline.

Interpretation: Using targeted proteomics, this work identified cortical proteins involved in Alzheimer's dementia and begins to dissect two different molecular pathways: one affecting β-amyloid deposition and another affecting resilience without a known pathological footprint. Ann Neurol 2018;83:78-88.

Figure 1

In-network proteins in module m109. Figure 1 shows selected proteins in m109 network. Nodes with thicker border are proteins (N=18) that are detected in our SRM experiment and shaded nodes (N=12) are those that are quantifiable. Arrows represent proteins that are associated with cognitive decline (N=5), and squares represent those that did not reach the cut-off for a significant association.

Figure 2

PLXNB1 protein with cognitive decline. Figure 2 illustrates the association of PLXNB1 protein abundance with annual rate of cognitive decline before and after the adjustment for measures of AD pathology. On the x-axis, “Model A” refers to the protein association from the model without adjusting for AD pathology; “Model B” refers to the association from the model adjusted for amyloid pathology; “Model C” refers to the association from the model adjusted for both amyloid and tangle pathologies. For each model, the point estimate +/- 1.96 standard error of the effect size is shown on the y-axis.

Figure 3

Associations of IGFBP5, HSPB2, AK4 and ITPK1 proteins with common neuropathologies. Figure 3 illustrates associations of AK4, HSPB2, IGFBP5 and ITPK1 with common age related neuropathologies. Results are presented as -log₁₀(p) such that higher the value, more significant the association. Each bar was derived from a separate model with corresponding pathologic index as the outcome and protein abundance the predictor. Black color represents positive association and white negative association. Vertical dotted line is the reference cut-off representing p=0.005.

Figure 4

IGFBP5 and HSPB2 proteins with cognitive decline before and after the adjustment for neuropathologies. Figure 4 illustrates the associations of the IGFBP5 (Panel A) and HSPB2 (Panel B) proteins with cognitive decline. Dark gray circles are person-specific rates of decline estimated from a linear mixed model adjusted for demographics, plotted against protein abundance. The black line represents the regression line between protein abundance and cognitive decline without controlling for neuropathologies. White circles are person-specific rates of decline estimated from a separate linear mixed model adjusted for demographics and neuropathologies, plotted against protein abundance. The gray line represents the regression line between protein abundance and cognitive decline after controlling for neuropathologies. Notably, the gray line is flatter than the black line, but it still deviates from the horizontal reference line that represents the null.

Figure 5

AK4 and ITPK1 proteins with cognitive decline. Figure 5 illustrates the associations of the AK4 (Panel A) and ITPK1 (Panel B) proteins with cognitive decline. For each panel, the line types show cognitive decline for representative females with mean age and education but different levels of protein abundance (solid lines: 10th percentile; dash lines: 50th percentile; dotted lines: 90th percentile).