For each aging-related tissue-cell-gene tuple (defined as the gene significantly related to aging in a tissue-cell type), we computed its aging trajectory by first computing the mean expression of each time point and then z-scored the expressions over the time points (making mean 0 and std 1). Next, in order to learn the typical aging trajectories, we used the k-means clustering algorithm to cluster the aging trajectories of all those 118,693 aging-related tissue-cell-gene tuples. As shown in panel A, according to the elbow method there are k = 4 clusters. We further visualized those four cluster centers in panel B. As indicated by the figure legend, most tuples are associated with cluster centers 0 and 1, corresponding to monotonically decreasing gene expressions from 3 m to 24 m. Interestingly, 15,499 tuples are associated with cluster center 2 whose gene expressions first increased from 3 m to 18 m and then decreased from 18 m to 24 m. Also, 13,788 tuples are associated with cluster center 3, corresponding to a monotonically increasing aging trajectory. In panel C, we further counted those tuples by tissue-cell types. While most tissue-cell types have aging-related genes belonging to cluster centers 0 and 1 corresponding to monotonically decreasing trajectories, B cells in the brown adipose tissue (BAT), epithelial cells in the large intestine, and mesenchymal stem cells of adipose in the mesenchymal adipose tissue (MAT) have more aging-related genes in cluster 2, where the gene expressions increased from 3 m to 18 m and then decreased from 18 m to 24 m. In addition, cell types in the MAT, large intestine, and spleen have more upregulated aging-related genes that correspond to clusters 2 and 3, consistent with Figure 1B.