Cognitive heterogeneity reveals molecular signatures of age-related impairment

Abstract The greatest risk factor for cognitive decline is aging. The biological mechanisms for this decline remain enigmatic due, in part, to the confounding of normal aging mechanisms and those that contribute to cognitive impairment. Importantly, many individuals exhibit impaired cognition in age, while some retain functionality despite their age. Here, we establish a behavioral testing paradigm to characterize age-related cognitive heterogeneity in inbred aged C57BL/6 mice and reliably separate animals into cognitively “intact” (resilient) and “impaired” subgroups using a high-resolution home-cage testing paradigm for spatial discrimination. RNA sequencing and subsequent pathway analyses of cognitively stratified mice revealed molecular signatures unique to cognitively impaired animals, including transcriptional down-regulation of genes involved in mitochondrial oxidative phosphorylation (OXPHOS) and sirtuin (Sirt1 and Sirt3) expression in the hippocampus. Mitochondrial function assessed using high-resolution respirometry indicated a reduced OXPHOS coupling efficiency in cognitively impaired animals with subsequent hippocampal analyses revealing an increase in the oxidative damage marker (3-nitrotyrosine) and an up-regulation of antioxidant enzymes (Sod2, Sod1, Prdx6, etc.). Aged–impaired animals also showed increased levels of IL-6 and TNF-α gene expression in the hippocampus and increased serum levels of proinflammatory cytokines, including IL-6. These results provide critical insight into the diversity of brain aging in inbred animals and reveal the unique mechanisms that separate cognitive resilience from cognitive impairment. Our data indicate the importance of cognitive stratification of aging animals to delineate the mechanisms underlying cognitive impairment and test the efficacy of therapeutic interventions.

schedule. The data were exported from EthoVision as a text file and then processed using Python scripts. Success rates for both acquisition and reversal learning were calculated and defined as the percentage of correct entries of the trailing 30 entries. The percentage of correct entries made in a moving window of the trailing 30 entries was calculated and trials to reach an 80% success rate determined. Thus, when the mouse achieved 80% correct entries within that window, the criteria was met, and the number of entries was plotted against the percentage of mice that achieved criteria for the group. The Independent Leaning Index was also calculated per hour based on correct entries minus the incorrect entries divided by the total number of entries and plotted as a cumulative learning index for acquisition and reversal learning. Cognitive flexibility was calculated as correct entries minus incorrect entries divided by the total number of entries during the first dark phase of reversal learning between hours 51 and 61 after initiation of the experiment. Extinction of learned behavior was calculated based on entries into left and middle entries (% incorrect) during the first dark phase of the reversal learning. It should be noted that animals are removed from the study if they received fewer than 10 pellet drops during the acquisition task or due to technical issues with pellet dispensers. This exclusion represented <10% of all mice in our study, irrespective of age. Following behavior testing, all mice were returned to group housing with chow food for one week before harvesting for tissue.

RNA/cDNA Preparation and qRT-PCR
Total RNA was extracted using the RNeasy mini kit (Qiagen, Germantown, MD). cDNA was prepared from equal concentrations of total RNA (1.0 μg) using the High-Capacity RNA-to-cDNA™ Kit (Applied Biosystems, Foster City, CA). qRT-PCR was performed using the following genespecific Taqman probes: Gfap (Mm01253033_m1), Aldh1l1 (Mm03048957_m1), Gapdh (Mm99999915_g1) and Hprt (Mm03024075_m1) were used as housekeeping genes for normalization. Quantitative PCR and melt-curve analyses were performed using TaqMan® Universal PCR Master Mix with UNG (Applied Biosystems) and the QuantStudio 12K Flex Real-Time PCR System (ThermoFisher Scientific). Gene Expression data were calculated from 5-6 independent samples unless otherwise stated, each with two replicates and are presented relative to the expression of the geometric mean of the house keeping genes (mean ± SEM).

Library preparation and sequencing
Total RNA extracted from the hippocampi was suspended in RNase-free water and RNA purity and concentration was measured on an Agilent Bioanalyzer (Agilent Technologies., Palo Alto, CA) and HiSeq libraries were sequenced. Each read generated was mapped to the mouse genome (mm10) by gSNAP. In addition to the genome sequence, gSNAP also accounts for splice variations, and SNP variants from dbSNP, and uses these ancillary databases to assist in mapping highly polymorphic sample data to the monomorphic reference genome, thereby increasing the mapping accuracy. Subsequently, Cufflinks calculates the prevalence of transcripts from each known gene based on normalized read counts. For each gene, Cufflinks quantifies transcript levels in fragments per kilobase of exon per million mapped reads (FPKM). The FPKM reflects the molar concentration of a transcript in the starting sample by normalizing for gene length and for the total read number in the sample. This allows for comparison of transcript levels both within and between experiments.
From this information, we then determined significant gene expression using ANOVA in R. Once a gene list had been formed, we utilized pathway and gene set enrichment analyses to identify pathways of interest that may be modified. Genes significant at an FDR < 0.05 were submitted to pathway analysis using Ingenuity Pathway Analysis (Qiagen, Germantown, MD) to identify pathways of interest that were modified by age and cognitive status.

Targeted Quantitative Proteomics
Lysates (100 µg protein; n = 6/group) were run on an SDS gel for analysis as previously described (4). Each gel lane was cut as a complete sample, then divided into smaller pieces and washed/destained. The proteins were reduced with DTT and alkylated with iodoacetamide.
Samples were then washed with ethanol and bicarbonate and digested with 1 µg of trypsin overnight at room temperature. The peptides produced were extracted from the gel, the extract evaporated to dryness, and reconstituted in 1% acetic acid for analysis. The digest samples were injected in 5µl aliquots and quantified using an QEx orbitrap and TSQ systems. A BSA internal standard was added for quantification, and the mass spectrometer was operated in selected reaction monitoring (SRM) mode to analyze two peptides per protein. Data were analyzed using the program SkyLine, and the response for each protein was calculated as the geometric mean of the two-peptide area normalized to the total ion count (TIC) (5). The principal component analysis (PCA) plot was generated using ClustVis with default settings (Row scaling = unit variance scaling, PCA method = SVD with imputation, clustering distance for rows = correlation, clustering method for rows = average, tree ordering for rows = tightest cluster first) (6). Normalized values of each individual protein were analyzed via one-way ANOVA across young/aged-intact and agedintact/aged-impaired groups. (a) Violin plots depicting the number of entries to achieve 80% criterion during the acquisition phase of the PhenoTyper were comparable among young cohorts across a 3-year period.
(b-c) Hours required to achieve 80% criterion during the acquisition (b) and reversal (c) phases of the PhenoTyper were comparable among young cohorts across years tested.
(b) The total distance moved between cohorts were comparable during the light and dark phases of the PhenoTyper, respectively.
(c) Average maximum segment velocity in the PhenoTyper was comparable among groups.
(d) Violin plots depicting the hours required to achieve 80% criterion during the acquisition (left panel) and reversal (right panel) phases of the PhenoTyper.
(e) Plotted independent learning index during initial learning (acquisition; 0-49 h) and reversal (50-89 h) phases show a significant increase in performance by the aged-intact group relative to young (p = 0.0279) and aged-impaired (p = 0.0008) animals during the acquisition phase. Agedimpaired animals had a significant decrease (p < 0.0001) in performance relative to aged-intact animals during the reversal phase.
(f) The independent learning index indicates decreased performance in the aged-impaired group relative to aged-intact animals during both the light (p = 0.0002) and dark (p < 0.0001) phases of the L:D cycle.
(g) Plot depicting the percent of left entries were comparable among groups during the extinction portion of the reversal phase.
(h) Bar plot depicting the distribution of pellets per hour of the aged-impaired group was decreased compared to young and aged-intact animals (p = 0.0067) during the dark portion of the reversal phase.

a b
Young Oxygen consumption rate in hippocampal tissue extracted from young (black, n = 11), aged-intact (blue, n = 7), and aged-impaired (red, n = 6) animals in response to increasing concentrations of ADP. Age-related declines in OCR were detected at 1250 µM (p = 0.0197) and 2500 µM (p = 0.0138) concentrations.
Error bars depict the mean ± SEM. Significance was tested using two-way ANOVA (*p < 0.05).  Significant differences between aged intact and aged impaired groups are represented with an asterisk (*p < 0.05, **p < 0.01). Significant differences between aged intact and aged impaired groups are represented with an asterisk (*p < 0.05, **p < 0.01). Significant differences between aged intact and aged impaired groups are represented with an asterisk (*p < 0.05).  Significant differences between aged intact and aged impaired groups are represented with an asterisk (*p < 0.05).