The AD-BXD mouse population is a powerful emerging model for studying AD resilience. AD-BXD mice query two different levels of genetic information: genometypes and genotypes. The genetically diverse BXD genometypes are a mosaic of the approximately 6 million segregating single nucleotide polymorphisms between C57BL/6J and DBA/2J mice. The transgenic genotype of the 5 times familial Alzheimer’s disease (5xFAD) transgene, which includes human APP carrying three pathogentic variants and human PSEN1 carrying two pathogenic, can be contrasted with non-transgenic control animals (Ntg). Reproducible BXD genometypes allow repeated measurement of the same mosaic genomes, a strategy that boosts the mappable heritability of genetic loci (cite ). This property facilitates QTL mapping of the relatively lower effect size variants that modify susceptibility to Alzheimer’s-like cognitive phenotypes in 5xFAD animals. Further, the same cognitive phenotype can be measured in both 5xFAD and Ntg mice within the same genometype. The ability to contrast a genometype’s performance with different AD genotypes facilitates the measurement of novel and interesting derived phenotypes based upon the property of trait correlation (cite ). Because of these properties, our lab has successfully leveraged AD-BXD mice to identify genetic modifiers of AD-related phenotypes in recent years. We quantified resilience using a novel regression-based metric. We utilized this measure to map resilience to a locus, then leveraged proteomics data to identify a putative molecular mechanism for the genetic resilience factor. We found potential novel targets for AD by querying elements coregulated in trans to the putative molecular mechanism. Finally, we implicated differential expression of networked synaptic proteins as the likeliest downstream consequences of differences in resilience.
The mouse hippocampus tissue was randomly balanced based on condition group ratios (genotype, sex, age and BXD strain) into batches of 14 individual hippocampal samples and 2 references. One reference was the same pool of a balanced batch to create a hippocampus specific reference that could be used as a common reference or a single-point calibrant in every batch to not only monitor the performance of the experiment workflow but also calibrate variability across different laboratories or experiments. To evaluate the run order and batch effects within the study, a second reference, which is a mixture of C57BL/6J (B6/B6) and DBA/2J (B6/D2) mouse midbrain and cerebellum tissues from all condition types, was used as a batch reference to allow for a more accurate comparison of between brain region proteomic profiles. The hippocampal tissue was processed as previously described. In summary, the tissue was briefly probe sonicated in a SDS lysis buffer and then a small volume of lysate was high pressure homogenized. In order to monitor digestion, a process control of yeast enolase was added to the protein homogenate which was then reduced, alkylated and trypsin digested on a S-trap column (Protifi). Eluates of hydrophilic and hydrophobic peptides were pooled and speed vacuumed. One μg of each digested sample, 8 ng of yeast enolase and 150 femtomole of Pierce Retention Time Calibrant (PRTC) were loaded onto a Thermo EASY nano-flow UHPLC coupled with a Thermo Orbitrap Fusion Lumos Mass Spectrometer. The yeast enolase is used as a protein process control and the PRTC is used as a peptide process control within the sample. System suitability (QC) injections of 150 fmol of PRTC and BSA were also used independently, to assess quality before and during analysis. We analyzed four of these system suitability runs prior to any sample analysis and then after every six to eight sample runs another system suitability run is analyzed. Six gas phase fractionated LC-MS/MS DIA runs of an overlapping “narrow” 4 m/z isolation window of a mass range of 100 m/z (400-500 m/z, 500-600 m/z, 600-700 m/z, 700-800 m/z, 800-900 m/z, 900-1000 m/z) for each of the six runs from a pool of all the samples from each batch are collected to be used as the on column chromatogram library. Each individual digested sample was also analyzed with a LC-MS/MS DIA run with a mass range of 400-1000 m/z and an overlapping “wide” 8 m/z isolation window. The quality control data was analyzed using Skyline and AutoQC.
Thermo instrument RAW files were converted to mzML format using Proteowizard (version 3.0.20064) using vendor peak picking and demultiplexing with the settings of “overlap_only” and Mass Error = 10.0 ppm[REF ]. EncyclopeDIA (version 2.12.30) was used to search and combine the on column chromatogram libraries of all runs from all 31 batches using the data from the six gas phase fractionated “narrow window” DIA runs of a pool from each batch[REF ]. These narrow windows were analyzed using EncyclopeDIA with the default settings (10 ppm tolerances, trypsin digestion, HCD b- and y-ions) of a Prosit predicted spectra library based on the Uniprot mouse canonical FASTA (April 2020). The results from this analysis were saved as a “Chromatogram Library'' in EncyclopeDIA’s eLib format where the predicted intensities and iRT of the Prosit library were replaced with the empirically measured intensities and RT from the gas phase fractionated LC-MS/MS data. The “wide window” DIA runs were analyzed using EncyclopeDIA (version 2.12.30) requiring a minimum of 3 quantitative ions and filtering peptides with q-value ≤ 0.01 using Percolator (version 3.01). After analyzing each file individually, EncyclopeDIA was used to generate a “Quant Report'' which stores all of the detected peptides, integration boundaries, quantitative transitions, and statistical metrics from all runs in an eLib format. The Quant Report eLib library is imported into Skyline (version 126.96.36.1991) with the mouse uniprot FASTA as the background proteome to map peptides to proteins, perform peak integration, manual evaluation, and report generation. A csv file of peptide level total area fragments (TAFs) for each replicate was exported from Skyline using the custom reporting capabilities of the document grid.
The quantitative peptide data exported from Skyline (level 2 data) was post-processed to minimize residual technical noise. The data was log2 transformed and median normalized to scale the intensities of the sample to the same median value. Using the batch factor as a predictor variable and abundance as its response, a simple linear regression model is fit to model the confounding due to the batch effect bias. The residuals from this model are assumed to represent the peptide or protein abundance without the effect of the batch covariate. Principal Component Analysis (PCA) is used to evaluate the confounding bias by projecting the data in its reduced dimensions onto the first three principal components. The normalized and batch adjusted peptide abundance is provided as level 3A data. Peptides are next constructed into an indistinguishable protein group and are summed to estimate the abundance of the peptide groups and proteins that match the same set of peptide groups merged into a single node in the graph. The normalized and batch adjusted protein abundance are provided as level 3B data.
Samples are tissue from the hippocampus of mouse brain from BXD (C57BL/6J and DBA/2J) 5xFAD (5 times familial Alzheimer’s disease) transgenic animals that express human APP carrying three pathogenic variants and human PSEN1 carrying two pathogenic variants and non-transgenic control animals. Tissue is both male and female mice and young (6 months) and old (14 months) mice. Several different strains of the BXD mice are represented.