Gene expression reversal toward pre-adult levels in the aging human brain and age-related loss of cellular identity

Abstract

It was previously reported that mRNA expression levels in the prefrontal cortex at old age start to resemble pre-adult levels. Such expression reversals could imply loss of cellular identity in the aging brain, and provide a link between aging-related molecular changes and functional decline. Here we analyzed 19 brain transcriptome age-series datasets, comprising 17 diverse brain regions, to investigate the ubiquity and functional properties of expression reversal in the human brain. Across all 19 datasets, 25 genes were consistently up-regulated during postnatal development and down-regulated in aging, displaying an "up-down" pattern that was significant as determined by random permutations. In addition, 113 biological processes, including neuronal and synaptic functions, were consistently associated with genes showing an up-down tendency among all datasets. Genes up-regulated during in vitro neuronal differentiation also displayed a tendency for up-down reversal, although at levels comparable to other genes. We argue that reversals may not represent aging-related neuronal loss. Instead, expression reversals may be associated with aging-related accumulation of stochastic effects that lead to loss of functional and structural identity in neurons.

Figure 1

Expression changes during brain postnatal development and aging. (a) Correlations among expression-age correlation coefficients (in development or in aging) across subdatasets. The color of the squares changes with the magnitude of the Spearman correlation coefficient between pairs of subdatasets (across 11,563–22,713 common genes in each pairwise comparison); darker colors show stronger negative (red) or positive (blue) correlation. Row and column labels indicate the brain region and the edge color of the dendrogram shows the time period of datasets. The order of brain regions is determined by agglomerative hierarchical clustering using the Spearman correlation coefficients between datasets based on the age-related expression changes. (b) Principle components analysis of age-related expression change in brain regions. The analysis was conducted on an age-expression Spearman correlation coefficient matrix of 11,258 genes and 38 subdatasets. The x- and y-axes show the first and second principle components. The values in the parentheses show the variation explained by each component.

Figure 2

Shared expression changes with age and shared reversals. (a) The number of consistent expression changes in development and aging. Interestingly, there is a trend toward 1.7% more consistently up-regulated genes during development, and 12.3% more consistently down-regulated genes during aging. This was also in line with 16/19 datasets harboring more up-regulated significant genes during development, and 10/12 datasets harboring more down-regulated significant genes during aging (Figure S2). (b) The proportion of different trends in age-related expression change in each dataset and among the genes showing consistent change across datasets. No effect size or significance cutoff was used. Up-down: up-regulation in development and down-regulation in aging; Down-up: down-regulation in development and up-regulation in aging; Monotonic increase: up-regulation in development and up-regulation in aging; Monotonic decrease: down-regulation in development and down-regulation in aging. (c) Average expression trajectories of consistent reversal gene clusters, plotted against individual age. The x-axis shows the age of samples on the fourth root scale and the y-axis shows the scaled mean expression level for each cluster. The spline curves indicate the mean expression change with age for each dataset and brain region.

Figure 3

Shared GO Biological Process (BP) categories enriched for up-down genes. The 113 GO BP categories were chosen for showing a trend for enrichment in up-down genes (odds ratio > 1) in all 19 datasets (p = 0.017). The 113 categories were summarized by REVIGO. Different colors show different superclusters and the size of each box is determined by the uniqueness of the categories.

Figure 4

Reversal proportion of the genes up-regulated during in vitro neuronal differentiation and in postnatal development. The gray points show the reversal proportion in 1000 permutations of individual labels in each dataset, whereas the red points show the observed proportions (i.e. number of up-down genes divided by the number of up-up genes, among genes up-regulated during neuronal differentiation). Statistical significance of each dataset is indicated by asterisks, which represent nominal p-values from permutations of individual ages (without correction for multiple testing).

Figure 5

Three examples of expression-age trajectories among clusters of synapse-related and up-down-enriched GO BP categories (Group 9 in Fig. 3). The y-axis shows the scaled expression level and x-axis shows the age of samples on the fourth root scale. The spline curves indicate the mean expression change with age among cluster member genes for each dataset and brain region (for all of the clusters see Figure S11). The three clusters were chosen for showing peak expression at ages earlier or later than 20 years of age.