Transcriptional changes of the aging lung

Abstract

Aging is a natural process associated with declined organ function and higher susceptibility to developing chronic diseases. A systemic single-cell type-based study provides a unique opportunity to understand the mechanisms behind age-related pathologies. Here, we use single-cell gene expression analysis comparing healthy young and aged human lungs from nonsmoker donors to investigate age-related transcriptional changes. Our data suggest that aging has a heterogenous effect on lung cells, as some populations are more transcriptionally dynamic while others remain stable in aged individuals. We found that monocytes and alveolar macrophages were the most transcriptionally affected populations. These changes were related to inflammation and regulation of the immune response. Additionally, we calculated the LungAge score, which reveals the diversity of lung cell types during aging. Changes in DNA damage repair, fatty acid metabolism, and inflammation are essential for age prediction. Finally, we quantified the senescence score in aged lungs and found that the more biased cells toward senescence are immune and progenitor cells. Our study provides a comprehensive and systemic analysis of the molecular signatures of lung aging. Our LungAge signature can be used to predict molecular signatures of physiological aging and to detect common signatures of age-related lung diseases.

Keywords: aging; inflammation; lung; senescence; single-cell RNA-seq.

FIGURE 1

Single‐cell RNA‐Seq profiles human lung heterogeneity. (a) Uniform Manifold Approximation and Projection (UMAP) representation of the 22 identified cell populations (106,969 cells) (b) Cells on the UMAP plot of all 29 samples were colored by age group (Young: 19–23 years old, Middle‐Aged: 29–49 years old, Aged: 55–78 years old) (c) Cell counts of the identified cell types. AT1, alveolar type 1 cells; AT2, alveolar type 2 progenitor cells; lympho, lymphocytes; macro, macrophage; NK, natural killer cell.

FIGURE 2

Differential expression analysis of single‐cell RNA‐Seq data on young and aged lungs identifies aging genes across cell types. (a) Differentially expressed genes (DEGs) with darker color indicating statistical significance (adjusted p < 0.1). Volcano plots and heatmaps (top 20 most significant DEGs) are shown for monocyte‐like (b, c) and FABP4 macrophages (e, f). Enriched canonical pathways of monocyte‐like (d) and FABP4 macrophages (g) were detected using IPA. GeneRatio is the fraction of DEGs found in the specific gene set/pathway.

FIGURE 3

Identification of Lung Aging gene signatures. Upset plots to visualize significant upregulated (a) and downregulated (d) differentially expressed genes (DEGs). In upset graphs, each column corresponds to an intersection set. Bar charts (top) show the size of the set. Colors indicate the number of cell types in the intersection set from red (7 cell types in d) to yellow (one cell type). Each row represents which intersection sets each cell type participates. (b) Module aging scores of all cell types and cohorts were inferred based on upregulated DEG set (Wilcoxon test). (c) Breakdown by cohort of the results in (b). (e) Module aging scores of all cell types and cohorts were inferred based on the downregulated DEG set. (f) Breakdown by the cohort of the results in (e). (g) Enrichr detected enriched transcription factors of downregulated gene signature.

FIGURE 4

Random forest (RF) classifier for age state prediction. (a) The Lung Aging Signature gene features of the PGH 5P Frozen dataset with age group annotation (label) were randomly stratified into training data and testing data by sample, cell type and age group. A RF classifier was trained with the training data to predict the aging state (Young/Aged). (b) The area under the receiver operating characteristic (AUROC) evaluated prediction power of the RF classifier on the testing data using upregulated differentially expressed genes (DEGs) only (Red), downregulated genes only (Blue), and all DEGs (Grey). (c) Heatmap of the top 20 most important variables in the classification task colored as either upregulated in red or downregulated in blue across cell types.

FIGURE 5

Specific cell populations exhibit a higher score distribution of consensus senescence gene signature. Boxplots of module score distribution of Consensus Senescence gene signature in cell populations. Only cell types with significant differences are presented (Wilcoxon test).

FIGURE 6

Specific cell populations exhibit a higher score distribution of CSGene signature. Boxplots of module score distribution of CSGene gene signature in cell populations. Only cell types with significant differences are presented (Wilcoxon test).