Robust differences in cortical cell type proportions across healthy human aging inferred through cross-dataset transcriptome analyses

Brain aging refers to the changes in cognitive, sensory, and motor function experienced along the natural lifespan. These deficits are associated with biological changes at the gross structural and functional, cellular, and molecular levels. Importantly, brain aging is not a homogenous phenomenon. Cognitive abilities reliant on memory encoding, working memory, and processing speed decline with age, whereas those related to knowledge and expertise remain consistent or may even improve (Glorioso and Sibille, 2011; Park and Reuter-Lorenz, 2009). Similarly, age-related grey matter loss is area-specific, with decreases consistently reported in frontal cortices (Glorioso and Sibille, 2011; Resnick et al., 2003). Transcriptome-wide gene expression profiling of the human brain has also identified differentially regulated modules of genes across the lifespan (French et al., 2017), with many age-related gene expression differences likely reflecting cell type specific changes (Erraji-Benchekroun et al., 2005; Tollervey et al., 2011). More recently, certain cell types are believed to be particularly susceptible to brain aging (Mattson and Magnus, 2006; Ouellet and de Villers-Sidani, 2014; Soreq et al., 2017), meriting further inquiry.

Cell-count studies of the brain have traditionally used histological methods (Zhang et al., 2017). Experiments labeling cells via immunohistochemistry (Ouellet and de Villers-Sidani, 2014; Soreq et al., 2017) and in situ hybridization (Rocco et al., 2017) have reported changes in neuronal and glial cell abundance with aging. While informative, these studies are limited in the variety and specificity of cell types assessed, reflecting histology's constraint to a handful of protein and mRNA markers per experiment. Furthermore, both histological and non-histological cell-counting methods can be expensive, time-consuming and technologically challenging to scale to large sample sizes (von Bartheld et al., 2016; Zhang et al., 2017). Thus, there remains a need for comprehensive and scalable assessments of changes in the abundance of various cell types with healthy aging.

In contrast to cell-counting methods, there are now several large-scale studies describing the human brain transcriptome in hundreds of individuals (Bennett et al., 2018; Gibbs et al., 2010; Hoffman et al., 2019; Seney et al., 2013). While typically used to evaluate differences in abundance of individual genes across conditions or traits, bulk tissue transcriptomes can also be leveraged for cell type estimation via deconvolution techniques (Cai et al., 2022; Dong et al., 2021; French et al., 2017; Mancarci et al., 2017; Mohan et al., 2018; Pei et al., 2021; White et al., 2022). In short, this approach allows for the concurrent abundance estimation of comprehensive sets of cell types, as defined by the expression of dozens of marker genes within a single sample. Tools for cell type deconvolution have been shown to produce accurate estimates when tested on in silico RNA-seq data (Sutton et al., 2020) and correlate with immunodetected cell count results from the same subjects (Beasley et al., 2002; Patrick et al., 2020; Toker et al., 2018). Similarly, cell type deconvolution estimates have shown validity when benchmarked against cellular proportions derived from recent single-nucleus RNA-seq (snRNA-seq) datasets (Cain et al., 2022; Consens et al., 2022). Whereas these methods have been applied to understand changes in the brain's cell type composition previously, such approaches have often been applied to study brain development in early life (Ayana et al., 2018; Pei et al., 2021; Xu et al., 2013). To summarize, cell type deconvolution of bulk gene expression data is an accessible, reliable, and comprehensive approach for understanding subject-to-subject variability in the brain's cellular composition.

To study the relationship between brain cell type proportions and healthy aging, we analyzed eight bulk gene-expression datasets collected from six cohorts of post-mortem brain samples from individuals without psychiatric conditions or known brain pathologies. We derived a novel set of cell type-specific marker genes, and employed cellular deconvolution to estimate the relative abundance of thirteen transcriptomics-defined cell types. We found robust negative age-associations in the estimated proportions of neuronal subclasses, most prominent in somatostatin (SST) and vasoactive intestinal peptide-expressing (VIP) interneurons, and positive associations in several glial subclasses, including astrocytes. We further validated these associations via neocortical cellular proportions directly estimated from two single-nucleus RNA-seq datasets, supporting our inference that our results may reflect bona fide differences in cellular proportions associated with healthy aging.

留言 (0)

沒有登入
gif