Abstract
Background Aging is a pleiotropic process affecting many aspects of organismal and cellular physiology. Mammalian organisms are composed of a constellation of distinct cell type and state identities residing within different tissue environments. Due to technological limitations, the study of aging has traditionally focused on changes within individual cell types, or the aggregate changes across cell types within a tissue. The influence of cell identity and tissue environment on the trajectory of aging therefore remains unclear.
Results Here, we perform single cell RNA-seq on >50,000 individual cells across three tissues in young and aged mice. These molecular profiles allow for comparison of aging phenotypes across cell types and tissue environments. We find transcriptional features of aging common across many cell types, as well as features of aging unique to each type. Leveraging matrix factorization and optimal transport methods, we compute a trajectory and magnitude of aging for each cell type. We find that cell type exerts a larger influence on these measures than tissue environment.
Conclusion In this study, we use single cell RNA-seq to dissect the influence of cell identity and tissue environment on the aging process. Single cell analysis reveals that cell identities age in unique ways, with some common features of aging shared across identities. We find that both cell identities and tissue environments exert influence on the trajectory and magnitude of aging, with cell identity influence predominating. These results suggest that aging manifests with unique directionality and magnitude across the diverse cell identities in mammals.
Introduction
Aging is a gradual process of functional and homeostatic decline in living systems. This decline results in increased mortality risk and disease prevalence, eventually resulting in death. Aging appears to be a conserved feature of eukaryotic biology, affecting organisms as phylogenetically diverse as the single celled S. cerevisiea, the eutelic nematode C. elegans, mice, and humans [45, 63, 91]. Despite the near universal nature of the aging process, the underlying causes of aging are poorly understood. Aging phenotypes have been observed and hypotheses have been proposed for more than a hundred years [97, 50, 37, 64], but we do not yet know the cellular and molecular players that cause aging or how they differ between biological contexts. Both the fundamental nature of aging and its negative effects provide motivation to enumerate these players and establish causal relationships among aging phenotypes.
Mammalian aging phenotypes manifest at the organismal, tissue, cellular, and molecular levels [100]. Extensive research has produced catalogs of aging phenotypes at the physiological level, providing functional and behavioral hallmarks of age related decline. Likewise, molecular profiling of nucleic acids, proteins, and metabolites has provided a phenotypic description of aging in individual tissues [88, 42, 4, 56, 12].
Additional lines of inquiry have worked to address a classical question of aging biology – do different tissues age in the same way? Transcriptomic analysis at the bulk tissue level have revealed common traits of aging, as well as tissue-specific features [82, 43, 12]. Proteomic analysis of brain and liver in young and old mice similarly suggests that most age related changes are tissue-specific [72]. However, the cellular origins of aging phenotypes within a tissue remain largely unknown [12, 70, 78].
Our current understanding of aging phenotypes at the cellular level is less complete than at the tissue level. Mammals contain a multitude of distinct cell identities, each exhibiting specialized functions. In the mouse alone, recent cell atlas efforts have revealed more than 100 cell types [80, 39]. These surveys have catalogued diverse murine cell identities, but the plasticity of these identities and their contributions to tissue and organism level pathology remain unknown.
At both the molecular and functional level, a host of aging phenotypes and associated mechanisms have been revealed in individual cell types [23, 18, 35, 46, 48, 84, 20, 62]. While some of these studies present unique features of aging within individual cell identities, it is difficult to compare them systematically due to differences in experimental conditions and assay methodology. Using traditional molecular biology assays, it is difficult to measure high-dimensional molecular phenotypes across multiple cell identities, making large scale comparisons of aging phenotypes across cell identities intractable. The recent development of single cell RNA-sequencing (scRNA-seq) has ameliorated this limitation, allowing for measurement of transcriptional features across all prevalent cell identities in a tissue in a single experiment.
Although the technology has only recently matured, scRNA-seq experiments in individual tissues have already revealed novel aspects of the aging process. In a pioneering single cell RNA-sequencing study of hematopoietic progenitors, the axis of aging was shown to be opposite the axis of differentiation [53]. Multiple investigations have reported that cell-cell heterogeneity [34, 4] and gene expression variance [67] increase with age. However, the specific influence of cell identity and tissue environment on the trajectory and magnitude of aging has yet to be resolved.
Here, we employ scRNA-seq to generate a set of molecular profiles in which we can compare aging phenotypes across cell identities. By profiling 50,000+ cells from three tissues in young and old mice, we identify common features of aging that span cell identities, as well as features unique to each identity. Using matrix factorization and optimal transport methods, we compute trajectories of aging for each cell identity and assess the influence of identity and environment on these trajectories.
Results
Single cell RNA-sequencing identifies a diversity of cell types and states in young and old mouse tissue
We collected transcriptional profiles of young and old cells of many identities by isolating single cells from the kidney, lung, and spleen of n = 4 young (7 months) and n =3 old (22-23 months) C57Bl/6 mice. All three tissues were collected from the same animals. Isolations were performed at the same time of day for each animal, limiting circadian variation which affects the expression of nearly half of all murine genes [101]. After single cell isolation, cells were immediately encapsulated and barcoded for library preparation using the 10X Genomics microfluidics system, followed by subsequent sequencing. Using standard techniques for the identification of cell containing microfluidic droplets, we recovered 55,293 individual cell transcriptomes (see Methods).
We determined cell type and state identity by leveraging annotations provided in the Tabula Muris compendium [80]. These annotations provide labels at the cell type level and follow the structured hierarchy of the “cell ontology” [8]. Some age-related changes may be unique to individual states within a cell type. Similarly, some changes at the level of cell types may be mediated by differences in cell state proportions (i.e. CD4 vs. CD8 T cells). To ensure that we can explicitly detect these cell state level changes, we manually annotated cell states within each cell type in the Tabula Muris (see Methods, Supp. Fig. 1). We use the term “cell identity” to refer to the combination of cell type and state labels, such that CD4 T cells and CD8 T cells are different cell identities (Fig. 1).
We trained deep neural networks to classify cell types based on these annotations, then used these networks to predict cell types in our data (Fig. 1A, B; see Methods). We found that the networks transfer labels with high fidelity. We validated classifications by inspecting marker gene expression post-hoc (Supp. Fig. 2, 3) and computing correlations between cell identities in our data and the Tabula Muris (Supp. Fig. 4). From these cell identity annotations, we identify 19 unique cell types and 28 unique cell states across the three tissues. Comparing our cell types to the Tabula Muris, we recover all but one of the cell types identified (kidney loop of Henle epithelial cells, Supp. Fig. 3, 5). The cell type proportions we recover differ from the Tabula Muris – e.g. we recover comparatively more immune cells in the kidney and lung – but are not outside expected range based on previous comparisons between single cell RNA-seq datasets [80, 74].
Immune cells are more prevalent in old kidneys and lungs, while non-immune cell type proportions are preserved
One prospective way in which aging may influence tissue function is by altering the proportion of each cellular identity within the tissue. This change in cell identity compositions may occur at either the level of cell types, or by shifting the distribution of cell states within a cell type. To investigate the former possibility, we quantified the proportion of each cell type within each tissue across ages.
In both the kidney and lung, lymphocytes were significantly more abundant in old animals compared to young animals (t-test on additive log-ratio transformed proportions, q < 0.05)(Fig. 1D). In old kidneys, we found a roughly 2 fold increase in T cells, classical monocytes, and non-classical monocytes, and in old lungs a corresponding 2 fold increase in classical and non-classical monocytes and a roughly 1.3 fold increase in T cells. This may reveal increasing immune infiltration of the the non-lymphoid tissues with age, as suggested in previous studies of kidney, lung, and other non-lymphoid tissues [70, 78, 92, 6, 65]. However, we can not rule out that our recovery of specific cell types may be confounded by an interaction of aging with our isolation procedures.
Considering only non-immune cells in the kidney and lung, cell type proportions were not substantially altered by aging (Supp. Fig. 6B). Likewise in the spleen, we found minimal change in the proportion of cell types between young and old animals (Fig. 1D). While changes in non-immune cell type proportions are subtle, we cannot rule out that even these subtle changes may influence the aging process.
Shifting cell state proportions within a cell type may be an alternative mechanism by which aging phenotypes manifest. Examples of this phenomenon are present in the literature, such as the observed decrease in naive CD8 T cells relative to other T cell states [29, 38] and the shift from highly regenerative to less regenerative stem cell states in the blood and muscle [23, 14]. To address whether cell state proportions change with age, we quantify the proportion of cells in a given cell state for each cell type.
Investigating spleen T cells, we recapitulate the finding that CD8 T cells are less abundant in old animals. We observed a shift in the population of kidney collecting duct epithelial cells – old animals exhibit a decreased frequency of Cald1+ collecting duct cells relative to Slc12a3+ collecting duct cells (χ2 contingency table, q < 0.05). Aqp3+ and Slc12a3+ cells are likely principle cells of the collecting duct based on expression of Scnn1a. Cald1+ cells are also marked by Phgdh, which has a reported mosaic expression pattern in the proximal tubule, and Cryab, which was similarly identified to mark a distinct subpopulation in the Mouse Microwell Atlas [39] (Supp. Fig. 6C). However, we are not aware of a defined role for this cell population, making it difficult to speculate on the impact of this shift in cell state proportions. Other cell types with notable cell state substructure do not show shifts with age, such as lung stromal cells (Supp. Fig. 6A).
Cycling cells are similarly rare in young and old animals
Previous reports have suggested that cell cycle activity changes with age in multiple cell populations. In blood progenitors, cell cycle kinetics are accelerated with age [53], while in muscle progenitors the frequency of cycling cells increases with age [23], and in the intestinal crypt cycling cells become less frequent [69]. To investigate the possibility of changes in cell cycle frequency in our data, we evaluated cell cycle activity by scoring the expression of S-phase associated genes and G2M-associated genes [90] (see Methods). We observe only subtle changes in either of these cell cycle module scores with age across cell identities. The proportion of putatively cycling cells is also very small across cell identities (Supp. Fig. 7). These results suggest that the cell cycle rate in cell identities we observe is not dramatically changed with age. However, we cannot discount the possibility that the cell cycle scoring method we use is insufficient to detect differences.
The accumulation of non-cycling senescent cells in aging tissues has been reported in several previous studies [25]. The reported magnitude of senescent cell accumulation varies between tissues. In aging mouse kidney, the proportion of cells with senescence associated β-galactosidase activity increased from roughly 0.2% to 1.2%, whereas in epicardial cells the proportion increases from 2% to 10% (12 months to 18 months) [7]. Similar observations have been made using paired-end single cell RNA-seq in the human pancreas, where the proportion of cells expressing senescence marker gene CDKN2A increases from roughly 7% to 15+% between early- (21-22 years) and mid-adulthood (38-54 years) [34]. To investigate whether senescent cells are more prevalent in the old tissues we observe, we similarly measured expression of Cdkn2a. We find that the Cdkn2a locus (p16-Ink4a and p19-Arf) is not significantly upregulated with age in any of the cell identities we observe (Supp. Fig. 8). Due to the overlapping nature of the p16-Ink4a and p19-Arf reading frames, we note that we cannot distinguish transcripts from these two proteins using 3′-end RNA-seq alone [44]. We also scored the activity of a curated set of senescence-associated genes using the AUCell approach [1], but we do not find large differences in this score with age (Supp. Fig. 8).
Changes in cell-cell variation with age depend on cell identity
Single cell analysis allows us to measure not only the mean expression of each gene, but also the variation within a cell population. Previous work has suggested that both gene expression variance and cell-cell heterogeneity increase with age [67, 34, 5]. These two types of variation differ in subtle but important ways. Gene expression variance quantifies the mean dispersion across genes in the transcriptome, such that each gene contributes equally. Because genes are equally weighted, changes in gene expression variance are unlikely to be driven by a small number of genes. Likewise, multivariate differences in gene expression between cells are not resolved due to the focus on mean dispersion values. Increased gene expression variance may reflect a global change in transcriptional noise, perhaps due to loss of regulatory control, as suggested in previous aging studies [93].
By contrast, cell-cell heterogeneity measures the average distance in transcriptional space between cells in a population. These distances capture multivariate differences in gene expression, capturing variation in cell state that manifests across genes. They also account for gene expression level, such that a small number of more highly expressed genes can drive changes in cell-cell heterogeneity. Increased levels of cell-cell heterogeneity may reflect a diversification of cellular states within a population (Fig. 2A).
Both transcriptional variation and cell-cell heterogeneity have important implications for cell physiology and function of a cell population, as explored in seminal studies of transcriptional noise in cell fate selection [17, 89] and bet hedging [55, 2, 26, 85, 10]. To determine if age-related changes in variation and heterogeneity depend on cell identity and tissue environment, we evaluate both properties across the many combinations of cell identity and tissue environment we observe.
We evaluate transcriptional variation using the difference from the median (DM) method [51] to estimate “overdispersion.” Overdispersion refers to the residual variation in gene expression observed for a given gene after accounting for the mean:variance relationship in gene expression data [67] (see Methods). Across the cell identity/environment combinations we observe, we find that transcriptional variation by this metric is largely unchanged (Fig. 2B). Individual cell identities can be identified that exhibit both an increase in variation with age (lung leukocytes) or a decrease in variance with age (kidney mesangial cells) (Fig. 2D).
We quantified cell heterogeneity in each cell identity/environment combination using the distance to the centroid method as previously introduced [34, 5]. Cell-cell heterogeneity appears to increase for many cell identities (Wilcoxon Rank Sums, q < 0.05), including B cells across all three tissues, and lung stromal cells. We also observe decreased heterogeneity with age in some cell identities, such as lung type II pneumocytes and kidney CD8 T cells (Fig. 2C). Taken together, these results indicate that changes in gene expression variance and cell-cell heterogeneity with age are not universal and depend on cell identity. This suggests a more nuanced view of the notion that an increase in noise is a defining hallmark of aging at the single cell level [34, 93].
Differential expression reveals aging phenotypes common across cell identities
Prior studies have revealed differentially expressed genes within whole tissues or individual cell types in aging [27, 78, 70, 12, 47, 98, 88, 43]. However, it remains unclear to what degree age-related transcriptional changes are shared or unique across cell identities. To address this outstanding question, we performed differential expression analysis within each cell identity between young and old mice.
We identified differentially expressed genes for each cell identity/tissue environment combination using the Rank Sums method (see Methods). For each differentially expressed gene, we counted how many cell identities differentially express that gene in the same direction. The majority of differentially expressed genes with age are specific to one or a few cell identities (Fig. 3A).
However, a subset of 261 genes are differentially expressed across > 5 cell identities and show consistent changes across multiple cell identity/tissue environment combinations (Fig. 3B). Here, we chose 5 cell identities as a cutoff for common differentially expressed genes to reduce the number of genes identified due to common differential expression in highly similar cell identities (i.e. monocytes and macrophages). Thus, some aspects of transcriptional aging are common to many cell identities. Using gene ontology enrichment analysis for biological processes [54], we identify SRP-dependent protein localization and protein translocation to the ER as commonly downregulated across cell identities. This observation is consistent with the observed interaction of protein translocation systems with aging [40, 87] (Fig. 3C, lower). Antigen processing and inflammatory pathways are significantly upregulated with age, a result that has likewise been observed across tissues [78, 70, 12, 72]. Performing hierarchical clustering on this common set of differentially expressed genes, we find gene clusters enriched for specific inflammatory processes (type I interferon signaling, cytokine secretion), suggesting that more than one immunological pathway is changed with age (Supp. Fig. 9).
Increases in inflammation associated gene transcription have previously been reported in multiple tissues. However, previous investigations have relied on bulk transcriptional assays, making it difficult to determine if inflammatory genes signatures were upregulated in all cells within a tissue, or if more immune cells had infiltrated the tissue [78, 70, 12]. To investigate the possibility that immune pathways are upregulated across many non-immune cell identities, we identified a set of genes that are differentially expressed in at least 3 non-immune cell identities. Gene ontology enrichment analysis on this gene set reveals that inflammatory gene sets (T cell activation, B cell activation, viral entry, response to cytokines) are upregulated, even in these non-immune cell identities. However, the enrichment is much less significant than when immune cells are included (Supp. Fig. 10). At the gene level, we find Ikgc, Cd74, and B2m are commonly upregulated with aging. B2m has previously been reported to increase in the aging systemic milieu, and reports have indicated it may play a causal role in aging brain pathology [86].
Many genes change expression uniquely in individual cell identities, even within the same tissue. For instance, Npnt lung stromal cells exhibit upregulated and downregulated genes that are unchanged in lung type II pneumocytes and vice-versa (Fig. 3D). Gene ontology enrichment analysis reveals that Npnt stromal cells downregulate fibrinolysis and muscle development gene sets, while upregulating responses to fluid shear stress that are unchanged in type II pneuomcytes. By contrast, type II pneumocytes downregulate degranulation associated gene sets that are unchanged in stromal cells (Supp. Fig. 10).
We also observe genes that are differentially expressed in a tissue-specific manner. For example, CD4 T cells in the lung demonstrate downregulated and upregulated genes that are unchanged in CD4 T cells of the spleen and vice-versa (Fig. 3E). Collectively, these results indicate that while most differentially expressed genes are specific to individual cell identities, a subset of changes appears to be common across many identities. This common subset includes previously reported increases in inflammatory gene expression, which we find in both immune and non-immune cells. Cell identity and tissue environment also influence the genes which are differentially expressed, as demonstrated in comparisons of different cell identities in the same tissue, or the same cell identity in different tissues.
Aging manifests novel B cell states in the spleen
Beyond shifting cell state proportions, aging could promote the formation of novel cell states unseen in young animals. Age-associated cell states may then contribute to aging phenotypes at the cell type and tissue scales. We observe an example of a cell state that arises with age in spleen B cells.
Spleen B cells in our experiments exhibit three distinct clusters which do not map well to a canonical subtyping scheme (Fig. 4A). We observe that one of these clusters is dominated by cells from old animals (Fig. 4A, C; p < 0.05, χ2 contingency table). We performed differential expression analysis on each of these clusters to identify marker genes, and note high levels of Apoe in the larger cluster and high levels of B2m in the other. It has been reported that C57Bl/6 mice, like those in this study, die with an unusually high incidence of lymphoma [16, 21, 75]. To ask if these clusters dominated by old cells potentially represent lymphomas, we examined the expression of several genes associated with B cell lymphomas in a previous study [58]. We identify several lymphoma associated genes significantly upregulated in each of these clusters (Fig. 4B, rows 4-9)(q < 0.05, Wilcoxon Rank Sums test). Performing gene set enrichment analysis using the MSigDB Hallmark Gene Sets [61], we observe that Myc target genes and DNA repair pathways are upregulated, while p53 pathway genes are downregulated (Fig. 4D). This pattern of differential gene expression suggests that the Apoe high population is pre-neoplastic.
At a finer level of detail, we find that the Apoe high cluster appears to have two distinct lobes. One of these two lobes seems to contain cells almost entirely derived from a single old animal (1 of 3) (Fig. 4E), while the other lobe contains cells found in all of the old animals observed. To determine what differentiates these lobes, we examined marker genes and identify differences in the expression of Zeb2, the B cell marker Cd72, Plac8, and Cd24a as discriminating features (Fig. 4F; q < 0.05, Wilcoxon Rank Sums test). Each of these genes has previously been associated with neoplasia [102, 49, 60].
These emergent cell states support the notion that some aging phenotypes manifest not by shifting the distribution or location of youthful states, but by creating new states altogether. Neoplastic cells are an extreme case of such a phenomenon. The observation of a cell state that is generally ubiquitous in old mice and a related state that is present in only one animal also highlights the commonalities and the stochastic animal-animal heterogeneity that are associated with studies of the aging process.
Cell identity determines the trajectory of aging
Our differential expression and heterogeneity analyses suggest that cell identities age differently, and that the same cell identity ages differently across tissues. How much do cell identity and tissue environment influence the trajectory of aging? To answer this question quantitatively, we compute a trajectory of aging for each cell identity in each tissue based on their transcriptional profiles.
We perform this analysis using an embedding space derived by non-negative matrix factorization (NMF). NMF embeddings of transcriptional data have been shown to recover relationships between genes, such that each component of the embedding may be interpreted as a gene expression program [52, 22, 83]. Likewise, values of each component in each cell can be interpreted as the activity of that program. By using an NMF embedding to compute trajectories of aging, we are able to make qualitative interpretations at a level of abstraction above individual genes.
To compute aging trajectories, we first embedded all cells observed across tissues in a 20-dimensional NMF space (Fig. 5B; Supp. Fig. 11; see Methods, dimensionality chosen based on the trade-off between interpretability and explained variance). To assign semantic meaning to the embedding dimensions, we identified genes associated with each dimension by thresholding on the dimension loadings and analyzed gene set enrichment (Supp. Fig. 12). We computed the aging trajectory for each cell identity/tissue combination as the distance between the centroid of the young cells and the centroid of the old cells in this embedding (see Methods). This procedure yields a 20-dimensional vector representing the trajectory of age-related change observed in each cell identity.
We compare these trajectories using the cosine similarity. The cosine similarity is 1 if trajectories are in the same direction, 0 if they are orthogonal, and –1 if they are in opposite directions. Clustering by these cosine similarities, we find that qualitatively similar cell identities have similar aging trajectories, while dissimilar cell identities have orthogonal to dissimilar trajectories (Fig. 5A).
Examining the clustering partition, we find that cells segregate into roughly 4 clusters: endothelial and epithelial cells (blue), myeloid cells (purple), lymphocytes (green), and a remaining cluster with both myeloid and epithelial cell types (red). Cell identities from the kidney and lung are intermixed within the endothelial/epithelial cell cluster. Likewise, within the lymphocyte cluster B cells and T cells cluster more tightly by cell identity across tissues than by tissue of origin. This suggests that cell identity has a larger influence on aging trajectories than tissue environment.
To quantify the influence of both cell identity and tissue environment, we focused on the aging trajectories of immune cell types observed in all tissues (B cells, T cells, natural killer cells). For these cell types, we construct linear models and perform an analysis of variance (ANOVA) to determine the proportion of variation explained by cell identity and the proportion explained by tissue environment [77]. Consistent with qualitative observations, we find that cell identity explains a markedly larger fraction of variation than tissue environment (Fig. 5C). We also perform this analysis using a PCA embedding with the same result, indicating that this result is robust to our choice of embedding space (Supp. Fig. 13).
Endothelial cells and lymphocytes exhibit distinct aging trajectories
We next asked how endothelial/epithelial and lymphocyte aging trajectories differed. Using the clustering assignments from (Fig. 5A), we compared aging trajectories for each cell identity in the endothelial and lymphocyte clusters. We find that the two clusters show consistent differences in the magnitude of change within multiple NMF dimensions (Fig. 5D). Based on gene ontology enrichment within these dimensions, we find that endothelial/epithelial cell types show an increase in thyroid hormone and type I interferon signaling (Dimension 2, Fig. 5E) and oxidative/fatty acid metabolic processes (Dimension 16) relative to lymphocyte cell types. By contrast, lymphocytes show a larger increase in NF-κB signaling and related immune response pathways (Dimension 13). We note that B cell activating associated pathways are increased with age in a B cell specific manner (Dimension 19). Both clusters show increases in NAD metabolism and B cell signaling pathways (Dimension 18), as well as T cell signaling and activation (Dimension 6). These results are consistent with literature observing changes in type I interferon activity [59] and thyroid hormone signaling with aging [36, 15, 94].
Optimal transport analysis indicates that cell identity determines the magnitude of aging
Do some cell identities or tissue environments age more dramatically than others? To answer these questions, we estimated the magnitude of aging using optimal transport distances in an NMF embedding, as described above. Here, we utilize an embedding with 500 latent dimensions to capture more variation within the data (Supp. Fig. 11). Optimal transport is a technique for measuring distances between equally sized samples, such as probability distributions or discrete samples with the same number of elements. It was originally designed to calculate the minimal amount of earth that needed to be moved to convert a pile of earth into a fortification. Optimal transport distances have since been applied to a wide variety of tasks, including the analysis of single cell RNA-sequencing data [81].
The discrete optimal transport distance we apply here measures the minimum amount of change needed to make one group of cells match another. This metric capture differences in the covariance structure and modality of a cell population, in addition to differences in the population means (Supp. Fig. 14). We compare the distance between two cell populations by sampling n cells from each of them a number of times, and averaging the distance across the samples (see Methods). This bootstrap sampling scheme allows us to meet the equal sample size requirement for optimal transport distances, even when we observe different numbers of young and old cells.
Here, we compare the distance between young and old cells from each cell identity and tissue environment in the NMF embedding. For each cell identity, we make three distinct comparisons. We make a heterochronic comparison between young to old cells to estimate the magnitude of aging. This process is schematized in spleen B cells to provide intuition (Fig. 6A). We also make two isochronic comparisons, comparing young cells to young cells, and old cells to old cells (Fig. 6B). These isochronic comparisons serve as a null distribution, estimating the distance we would expect to see between random samples of cells in the absence of age-related change. In each comparison, we draw 300 samples, each containing n = 300 random cells (see Methods for sampling scheme used for low abundance cell identities). We normalize the heterochronic (Young-Old) comparison values for each cell identity by dividing by the mean of the larger isochronic distance.
We interpret these normalized optimal transport distances as an estimate of the magnitude of age related change. We find a large diversity in aging magnitudes across the cell identities we observe. Some cell identities exhibit little more distance between young and old cells than between two isochronic groups of cells (normalized distance of ≈ 1), while others show 3 to 4 fold larger differences in heterochronic comparisons. Lung and kidney natural killer (NK) cells exhibit some of the highest optimal transport distances, while kidney and lung B cells and kidney capillary endothelial cells show some of the smallest distances.
To quantify the relative contributions of cell identity and tissue environment to aging magnitude, we use the same linear modeling and ANOVA approach as above. These models reveal that cell identity explains the majority of variation in aging magnitudes while tissue environment explains little variation (Fig. 6C). To confirm that these results are robust to the non-deterministic NMF embedding procedure, we compute these distances across 10 separate optimization runs of the NMF embedding and find that relative distances are preserved. We likewise compute Old-Young distances with a range of random sampling sizes and find distances are highly correlated across settings of this parameter (Supp. Fig. 15).
Perhaps surprisingly, we find that different cell states within the same cell type can have notable differences in aging magnitude. For instance, spleen CD8 macrophages exhibit a larger aging magnitude than spleen CD4 macrophages. Similarly, lung Dcn stromal cells exhibit a larger aging magnitude than other stromal cell counterparts. Across all three tissues, CD8 T cells exhibit a larger aging magnitude than CD4 T cells. These observations collectively indicate that cellular identity can have a notable impact on the magnitude of age-related change, such that even different cell states within the same cell type exhibit differences in aging magnitude. Tissue environments by contrast appear to have less influence on aging magnitude, suggesting that most difference in the magnitude of aging between tissues is driven by differences in cell identity composition.
Discussion
Aging occurs across varied mammalian species, each composed of diverse cell types and states. While aging phenotypes are well catalogued at the organismal and tissue level, the cell identities that they comprise have been less explored. To understand the causes of aging, we wish to construct a causal network of molecular players and their relationships. If these players and relationships are dependent on cell identity, as evidence suggests for tissues [82, 43, 12, 72], an accurate representation of the causal network requires resolution of the organismal building blocks – individual cells.
Here, we use single cell RNA-seq to investigate aging across a diverse set of murine cell identities in three tissues. Tissues were collected in synchrony from each animal, allowing for direct comparisons between cell identities within and across tissues. Few synchronized cross-tissue data sets investigating aging phenotypes are available, and we believe these data will be informative to the aging research community.
We find that cell identities exhibit unique differentially expressed genes with aging, consistent with previous reports of cell identity specific aging phenotypes [4]. Similar cell types (e.g. kidney capillary endothelial cells & lung endothelial cells) exhibit broadly similar aging trajectories across tissues, while distinct cell types from the same tissue (e.g. B cells & type II pneumocytes in the lung) have dissimilar trajectories. This suggests that cell identity and aging trajectory are coupled, and distinct cell types may be undergoing independent aging processes. Consistent with this notion, cell identity explains the majority of variation in these aging trajectories. Tissue environment explains a lesser amount variation, but its influence is also present in our data in small sets of unique differentially expressed genes in the same cell identity across tissues. Collectively, our results indicate that the molecular manifestations of aging differ between cell identities and tissue environments. A causal network for aging phenotypes must therefore be conditioned on both cell identity and environment to accurately reflect biology.
Although most changes are unique to tissues or cell types, we did identify a shared core of differentially expressed genes with age. This core is characterized by decreased expression of genes involved in SRP-dependent protein translation and protein targeting to the ER and increased expression of genes involved in inflammation. Decreased protein translation and targeting to the ER has previously been associated with replicative age in S. cerevisiea [40] and is consistent with associations between global protein translation and aging [87]. Likewise, increased inflammatory pathway expression has been broadly observed in studies of aging tissues [12, 70, 78, 3]. Observation of these changes across many cell identities and tissue environments suggests they are consistent molecular players in the causal network of aging, less dependent on context than others. We caution that consistency and causality are not intertwined, and functional studies modulating these pathways in multiple cell identities are necessary to establish causal links to other aging phenotypes.
We present an optimal transport metric to estimate the magnitude of age-related change between two cell populations. Using this approach, cell identities exhibit multi-fold differences in aging magnitude, and cell identity again explains the majority of variation in aging magnitudes. Our results are conceptually consistent with previous reports of dramatic transcriptional changes with age in some cell identities and more subtle changes in others [46, 47, 53, 28]. Directly comparing the magnitude of aging in this manner suggests that some cell identities may exhibit more dramatic functional decline with aging than others. However, the magnitude of transcriptional change with aging may not necessarily reflect the degree of functional decline in all cases. Future comparison between identities with high and low aging magnitudes may reveal some cellular “strategies” that moderate age-related change.
Tissue level profiling of aging tissues has suggested that cell type composition may change with age [12, 70, 78]. Increased immune gene expression in multiple old tissues has specifically suggested increased infiltration of immune cells. Our data are consistent with this notion, as we observe an increased frequency of immune cell types in both the kidney and lung. However, these changes are difficult to confirm using only cell type counts produced after single cell isolation. Age-related changes in a tissue (i.e. altered extracellular matrix composition [4]) may lead to preferential isolation of one cell type relative to another, even if the underlying cell type proportions do not change. To confirm this phenotype, immunohistology studies quantifying the number of immune cells in these aging tissues are necessary. We also observed that inflammatory pathways are commonly upregulated, even in non-immune cell identities, suggesting that changes in immune cell proportions alone do not entirely explain the increased inflammatory pathway activity in aging tissues.
Previous studies reported that multiple metrics of transcriptional variability increase with aging [67, 34]. Two measures have predominated in the literature – (1) gene expression variance measures which focus on the mean “over-dispersion” across genes and (2) cell-cell heterogeneity measures which focus on the mean distance of individual cells to the center of the cell population. While ostensibly similar, each measure provides insight into a different aspect of biology.
Gene expression variance does not weight genes by expression level, such that changes are most likely when many genes increase or decrease variance. Biologically, changes in this measure may therefore reflect global changes in transcriptional noise [33, 76], which has been reported to influence cell fate decisions [96, 24, 41, 9]. By contrast, cell-cell heterogeneity measures consider the absolute distance of each cell from the population center, such that changes in heterogeneity can be driven by a handful of moderate- or high-expression genes. These measures may better reflect the phenotypic variance across cells in a population, as associated with the classical bet-hedging phenomenon [26, 85, 10]. We find that changes in each of these measures depend on cell identity, such that neither increased or decreased variability was observed universally across identities. These results are consistent with previous reports of aging altering gene expression variance and cell-cell heterogeneity in a cell identity dependent manner [4].
Conclusions
Single cell RNA-seq of young and old cells from three separate mouse tissues has revealed common and cell identity specific aspects of aging. We find that multiple metrics of transcriptional variation change with age, and that these changes are cell identity dependent. Cell identities exhibit unique gene expression changes with age, but we also identify a common set of protein translation and inflammatory pathway changes across many cell identities. Leveraging matrix factorization techniques, we define trajectories of aging and attribute the majority of variation in these trajectories to cell identity. Using optimal transport, we compute a magnitude of aging and find that observe multi-fold differences in this magnitude between cell identities. We again find that cell identity rather than tissue environment explains the majority this variation in magnitude. Collectively, these results highlight the influence of cell identity on the aging process and the importance of measuring aging phenotypes with cellular resolution.
Methods
Animals
Young (29-34 weeks old) and old (88-93 weeks old) male C57Bl/6 mice were used for all experiments. Mice were housed communally with a standard 12 hour dark cycle and fed ad libitum. Euthanization was performed by administration of carbon dioxide at a controlled flow rate. Animals were euthanized at the same time each day to minimize circadian variation, shortly after the beginning of the light cycle. Kidneys, lungs, and spleen were collected from each experimental animal and weighed. Downstream cell isolation for each tissue proceeded immediately.
Cell Isolation
We removed each tissue and washed each in HBSS then dissected the tissue into small pieces using a razor blade.
Kidney and Lung
We incubated dissected kidney tissue in 1.48 U/mL Liberase DL enzyme mixture (Roche) in a total volume of 7 mL DMEM at 37°C for 20 minutes in a 50 mL conical tube shaking at 200 rpm. We incubated dissected lung tissue in 1.48 U/mL Liberase TM (Roche) and 200 U/mL DNase I (Roche) in a total volume of 7 mL DMEM at 37°C for 30 minutes in a 50 mL conical tube shaking at 200 rpm. Tissue was further mixed using a 10 mL pipette tip and 40 mL of DMEM (2% FBS) was added to stop digestion. We sequentially pipetted cells through 100 μm, 70 μm, and 40 μm filters then peletted cell suspensions by centrifugation at 500 × g for 10 minutes.
After centrifugation, we treated cells with ACK Lysing Buffer (ThermoFischer) for 5 minutes at room temperature. We centrifuged ACK treated cells, discared the supernatant, and repeated ACK treatment. We subsequently washed cells in DMEM (2% FBS) and peletted by centrifugation. We removed debris from samples using the Miltenyi Debris Removal Solution. We added 6.2 mL cold PBS and 1.8 mL Debris Removal Solution to cell pelettes and resuspended by pipetting. We added 4 mL cold PBS to the top of the mixed solution and centrifuged samples at 4°C, 3000 × g for 10 minutes. We washed samples in 10 mL cold PBS and centrifuged at 4°C, 1000 × g for 10 minutes, then resuspended in PBS with 2% FBS. We counted cells using a TC20 Cell Counter (Bio-Rad) and diluted cells to a concentration of 106 cells/mL. We then proceeded to single cell library preparation.
Spleen
We used a syringe plunger to further fragment the spleen tissue after razor blade dissection. We mechanically dissociated tissue fragments by pipetting in RPMI cell culture media (2% FBS). We subsequently forced tissue fragments through 100 and 40μm filters. After centrifugation at 500 × g for 10 minutes, we treated cells with ACK Lysing Buffer (ThermoFischer) for 5 minutes at room temperature. We washed cells in RPMI (2% FBS) and pelleted by centrifugation. We counted cells using a TC20 Cell Counter (Bio-Rad) and diluted cells to a concentration of 106 cells/mL then proceeded to single cell library preparation.
Single Cell RNA Sequencing Experiments
We prepared libraries (individual lanes on the 10X Chromium) with the 10X Single Cell 3’ v2 kit using 6,000 cells per lane on the 10X Chromium microfluidics device. We sequenced libraries at a target depth of 50 million reads/sample on an Illumina HiSeq 4000. For each sample, we performed two technical replicates by preparing two separate libraries from the same cell suspension across two channels of the Chromium microfluidic device.
Read Alignment and Gene Expression Quantification
We aligned reads to the mm10 reference genome obtained from ENSEMBL. We used gene annotations from the GENCODE vM20 release with slight modification. We have replaced annotations for lincRNAs Gm42418 and AY036118 with a single contiguous gene annotation for the rRNA element Rn45s. This locus harbors an Rn45s repeat as reflected in RefSeq, such that contaminating 18S rRNA in our library preparations may lead to inflated expression counts for these lincRNAs.
We performed alignment to this amended reference using 10X cellranger 3.0.2, which employs the STAR sequence aligner [30]. We determined gene expression counts using unique molecular identifiers (UMIs) for each cell barcode-gene combination. Following alignment, we filtered cell barcodes to identify those which contain cells using the approach implemented in cellranger 3.0.2, and only these barcodes were considered for downstream analysis. The output of this analysis is a Cells × Genes matrix, where each element i, j represents the number of UMIs mapping to gene j in cell i.
Quality Control
We removed libraries which contained a low sequencing depth (< 10 million reads) or very few cells detected (< 500 cells) from subsequent analyses as likely experimental errors. Based on this metric, only one technical replicate failed QC, so all tissues from all animals in the experimental design are represented in our downstream analysis. We leveraged the scanpy toolkit [99] in subsequent analyses. We quantified the total number of UMIs, total number of genes detected, fraction of UMIs mapping to the mitochondrial genome, and fraction of reads mapping to the Rn45s repeat annotation for each cell. We filtered cells with fewer than 1000 UMIs or fewer than 500 unique genes detected, as described for the Tabula Muris [80]. Additionally, we filtered out cells with a high fraction of reads mapping to the mitochondrial genome (> 10%) or the Rn45s repeat (> 0.5%) as likely dead cells.
Dimensionality Reduction and Embedding
We normalized the Cells × Genes matrix by the total number of UMIs in each cell and scaled by 106 to yield Counts Per Million (CPM). We natural log transformed this matrix after the addition of a pseudocount of 1 to avoid undefined values.
To enable unsupervised clustering and cell type identification, we perform dimensionality reduction with principal component analysis (PCA) to the combined set of samples for each tissue. First, we identify a set of highly variable genes within the tissue based on overdispersion as described previously [79]. Briefly, we computed a “normalized dispersion” score for each gene by binning genes with similar mean expression levels, and subtracting the mean dispersion (variance/mean) within a bin from the dispersion score for each gene in the bin. The residual values represent the dispersion after accounting for the mean:variance relationship using this binning scheme. We set the following minimum criteria for the selection of genes as highly variable: minimum overdispersion of 0.5, minimum mean expression of 0.5 ln(CPM + 1), maximum mean expression of 7 ln(CPM + 1). After subsetting to this set of highly variable genes, we centered expression values to a mean of 0 and unit variance. We used these scaled genes as input features for PCA. Once embedded in this PCA space, we construct a nearest neighbor graph identifying the k = 15 nearest neighbors for each cell. We derived UMAP embeddings presented for visualization from this nearest neighbor graph using a minimum distance of 0.5 and a spread of 1.0 [11].
Clustering and Cell Type Identification
Louvain community detection [79, 19] was applied to the nearest neighbor graph constructed in PCA space to define a cluster partition. To infer cell types, we trained a neural network classifier to predict cell ontology classes given single cell RNA-seq expression information.
This classifier is a fully connected neural network with four hidden layers each paired to a rectified linear unit activation, dropout layer, and batch normalization. Each hidden layer has 1024 units and the dropout probability on each layer is set to p = 0.3. We apply a softmax activation to the final layer and utilize cross entropy as an objective function for training. During training, we perform class balancing with a mixture of over- and under-sampling. Classes (cell types) with fewer than 128 examples are oversampled, while classes with more examples are undersampled. Optimization is performed using the Adagrad optimizer [31].
As a training set, we utilized the recently published Tabula Muris compendium which provides expert cell type annotations in the mouse [80]. In addition to these annotations, we manually added cell state annotations to the Tabula Muris data to provide a level of granularity below cell ontology classes (Supp. Fig. 1). Example cell states include categorizing T cells into CD4 T cell and CD8 T cell subgroups, as well as the addition of subgroup labels to heterogeneous cell types such as lung stromal cells in the Tabula Muris. We name these cell states which do not have canonical names based on the expression of a prominent marker gene. For instance, we refer to a Gucy1a3+ subset of lung stromal cells as Gucy1a3 lung stromal cells.
We first trained a classifier for each tissue individually, yielding a set of tissue-specific classifiers that have no knowledge of cell types outside the training tissue. We also trained a tissue independent classifier by training on all cell types present in the Tabula Muris simultaneously.
We first inferred cell types for the lung and spleen using a tissue-specific classification model trained to predict cell ontology classes. We classified kidney cells with a tissue independent model to provide further resolution of immune cell types not annotated in kidney cells of the training set. We subsequently inferred subtypes using a model trained to predict our added subtype annotations. We chose subtypes as the most likely subtype allowed within our defined hierarchy. For instance, we chose the subtype of a T cell as the most likely subtype among the “CD4”, “CD8”, and “memory” subtypes. We used subtype classification models trained on the spleen to predict T cell and macrophage subtypes in all tissues, as the training set contained insufficient cell numbers to perform subtype annotation in other tissues. After direct inference of types and subtypes, we refined cell type information by using a k-nearest neighbors smoothing approach. Here, we chose k ∈ [30, 100] empirically depending on the tissue context.
Differential Cell Type Proportion Analysis
To determine if cell type proportions differed between old and young animals, we performed an additive log ratio (ALR) transform on the observed cell type frequencies and assessed significant changes for each cell type using a t-test. Within a given cell type, we performed a χ2-test of the Age × Cell State contingency table to determine if the proportions of cell states change with age.
Differential Variability Analysis
We measured differences in transcriptional variation between young and old animals in two distinct ways.
The first method evaluates changes in the variability of each gene between young and old animals, and attempts to identify a shift in the distribution of gene-wise variation. We assessed gene-specific variability by measuring the “overdispersion” of each gene. We defined overdispersion as the residual between a gene’s observed dispersion and the expected dispersion based on the gene’s mean expression value. We computed overdispersion values using the “difference from the median” (DM) method, as introduced previously [53]. We restricted DM calculation to genes with a mean expression value greater than 35 CPM to reduce the influence of poorly measured genes. To determine if the distribution of overdispersion values is significantly changed across ages, we employ the Wilcoxon Rank Sums test for a difference in means. We controlled the False Discovery Rate (FDR) to α = 0.05 with the Benjamini-Hochberg procedure. We performed DM analysis for each cell state in each tissue in our data set separately.
The second method we employ evaluates cell-cell heterogeneity based on the Euclidean distance between cells in expression space, as introduced previously [34]. For each cell state in each tissue, we compute the centroid of the cell state in gene expression space. We computed the distance from each cell to this centroid as a metric of cell-cell heterogeneity within each cell state. We employed the Wilcoxon Rank Sums test with the Benjamini-Hochberg procedure [13] as before to determine if this cell-cell heterogeneity metric is significantly different across ages.
Differential Expression Analysis
We computed differentially expressed genes between two groups of cells A and B using the Wilcoxon Rank Sums test with the Benjamini-Hochberg procedure for FDR control. We performed differential expression between young and old cells within each cell state in each tissue independently. In addition to comparing mean expression values with the Rank Sums test, we also computed the proportion of cells expressing each gene in each group. Cells with >= 1 UMIs for a given gene are considered to be expressing the gene, whereas cells with 0 UMIs are considered to not be expressing the gene.
Identification of Common and Unique Differentially Expressed Genes
To identify common differentially expressed genes with age across cell identities, we computed for each gene the number of cell identities in which that gene is significantly differentially expressed in the same direction. Note that we counted each cell identity as a single entry in this score, regardless of how many tissues it appears in. We then selected genes that are differentially expressed in the same direction in k >= 5 cell identities and consider this gene set to be “commonly differentially expressed” with age.
To identify genes which are uniquely differentially expressed between two cell identities A and B, we computed genes which are significantly differentially expressed in A and are (1) not-significantly differentially expressed in B and (2) have a log2 fold-change < 0.1 in B.
Gene Ontology Enrichment Analysis
We used Enrichr [54] to perform gene set enrichment analysis against the Gene Ontology Biological Process (2018) gene set collection. We also used the MSigDB Hallmark Gene Sets [61], for which we computed enrichment scores using Fisher’s exact test. In both cases, we corrected for multiple hypothesis testing using the Benjamini-Hochberg procedure.
Cell Cycle Scoring
Cell cycle activity was estimated by scoring the expression of a set of S phase associated and G2/M phase associated genes, as shown previously [90] and as implement in Seurat [79]. Briefly, a set of genes associated with each of these phases was derived from single cell RNA-sequencing in 293T and 3T3 cell lines. For each cell, the sum of the expression of these phase-associated genes is computed. As a null distribution, a set of genes are selected from the set difference of the observed genes and the phase-associated genes. These null genes are selected by binning the sample genes into 24 equally sized bins by mean expression, then randomly sampling 100 null genes per bin per phase-associated gene with replacement. The difference between the mean of the phase-associated genes and the selected null genes is considered the module score.
Analysis of Variance in Transcriptional Space
To determine the proportion of variance in transcriptional space (gene-wise UMI counts, NMF embedding dimensions) explained by experimental factors in our data, we use the linear modeling approach of Robinson et. al. [77]. Briefly, we fit a linear model for each dimension of the relevant transcriptional space (i.e. gene, NMF dimension) of the form where Y is a Feature × Cell matrix of observed transcriptional features, X is a Samples × Parameters design matrix containing p experimental parameters, β is a Feature × Parameters coefficients matrix, and E is a Feature × Cell matrix of residuals. We calculated the proportion of variance explained by each experimental parameter for each gene by ANOVA. To determine the total proportion of variance explained by a factor, we simply sum the sum of squares for each parameter p across features for each parameter and divide by the summed total sum of squares: where f is a feature in the set of features .
Non-negative Matrix Factorization Embedding
We perform non-negative matrix factorization using a standard multiplicative update optimization and random initialization [57], as implemented in the nimfa package [103]. We performed NMF optimization using all cells observed across all three tissues after ln(CPM + 1) normalization. The NMF embedding was fit to a set of highly variable genes, identified as described above. We chose a rank k = 20 for aging vector by selecting the “knee” in a plot of Rank vs. Explained Variance. We also utilize an NMF embedding of rank k = 500 for optimal transport distance calculation to capture more variation in the data, also chosen based on a later “knee‘ in the Rank vs. Explained Variance curve.
NMF Embedding Interpretation
To assign semantic meaning to each dimension of the embedding, we first identify genes associated with each dimension by performing Otsu thresholding [73] on log-transformed gene loadings. From the set of associated genes (loadings above the threshold), we perform gene ontology enrichment analysis with the Biological Process Gene Ontology database. We empirically name a consensus “program” for each dimension based on the enriched gene sets.
Aging Trajectory Calculation
We compute aging trajectories for each cell identity/tissue environment combination individually. For each cell identity, we compute the centroid c of the young cells and old cells in the NMF embedding and compute the vector from the young cells to the old cells .
Optimal Transport Estimation of Aging Magnitude
We use a discrete optimal transport (OT) distance to estimate the magnitude of difference between two cell populations A and B. Due to the conservation of mass assumption in the optimal transport formulation, we take random samples of the same size n from each population for comparison. We compute the OT distance as minimum cost solution to the linear sum assignment problem, solved using the Munkres algorithm [68]. For each comparison of populations A and B, we perform 300 random samples of size n = 300 cells and compute the average to reduce the variability inherent in this stochastic sampling approach. In the data set examined here, multiple populations have < 300 cells for sampling. In this circumstance, we set n = 0.8min(∥A∥, ∥B∥) and use the same repeated sampling approach as above.
To estimate the “magnitude of aging,” we compute OT distances for three distinct comparisons. We make a heterochronic comparison of young cells to old cells as a measure of the difference between these populations (Old-Young comparison). As negative controls, we compute both isochronic comparisons, measuring distances between random samples from pool of young cells (Young-Young comparison), and measuring distances between samples from the pool of old cells (Old-Old comparison). The isochronic comparisons serve as an estimate of the distance we would expect between random samples simply due to heterogeneity within the population.
We normalize the heterochronic Old-Young comparison by dividing these measurements by the mean of the largest isochronic distance (i.e. if Young-Young distance is larger than Old-Old, we divide by the Young-Young mean, and vice-versa). This normalization scheme is conservative, using the upper bound estimate of differences caused by cell-cell heterogeneity as our baseline for noting an effect due to aging. Following normalization, we therefore interpret Old-Young OT distances > 1 as indicative of differences caused by aging and interpret a larger value of this normalized distance as reflecting a larger magnitude of age-related change. To ensure these distance estimates are robust to different optimizations of the NMF embedding, we perform NMF optimization using 10 distinct random initializations and compute OT distances in each of these embedding spaces.
Software Tools
We leveraged GNU parallel [66], the scipy computing environment [71], and the Seaborn plotting package for several analyses [95].
Data Availability
Experimental data have been submitted to the Gene Expression Omnibus [32]. Additionally, we provide access to relevant data and analysis code on our website: https://mca.research.calicolabs.com/.
Competing interests
The authors are current and former employees of Calico Life Sciences.
Author’s contributions
JCK performed analysis and wrote the paper. LP conceived the study and performed experiments. NDR performed analysis. DGH performed analysis and provided experimental support. DRK supervised research and wrote the paper. AZR conceived the study, supervised research, and performed experiments.
Acknowledgements
We thank the Calico Physiology group, including Chunlian Zhang, Adam Kennedy, and Ganesh Kolumam for animal experiment support. We thank the Calico Genomics group led by Margaret Roy for assistance with library preparation and sequencing. We thank Allon Klein and Daniel Gottschling for helpful discussions and support.
Footnotes
References
- [1].↵
- [2].↵
- [3].↵
- [4].↵
- [5].↵
- [6].↵
- [7].↵
- [8].↵
- [9].↵
- [10].↵
- [11].↵
- [12].↵
- [13].↵
- [14].↵
- [15].↵
- [16].↵
- [17].↵
- [18].↵
- [19].↵
- [20].↵
- [21].↵
- [22].↵
- [23].↵
- [24].↵
- [25].↵
- [26].↵
- [27].↵
- [28].↵
- [29].↵
- [30].↵
- [31].↵
- [32].↵
- [33].↵
- [34].↵
- [35].↵
- [36].↵
- [37].↵
- [38].↵
- [39].↵
- [40].↵
- [41].↵
- [42].↵
- [43].↵
- [44].↵
- [45].↵
- [46].↵
- [47].↵
- [48].↵
- [49].↵
- [50].↵
- [51].↵
- [52].↵
- [53].↵
- [54].↵
- [55].↵
- [56].↵
- [57].↵
- [58].↵
- [59].↵
- [60].↵
- [61].↵
- [62].↵
- [63].↵
- [64].↵
- [65].↵
- [66].↵
- [67].↵
- [68].↵
- [69].↵
- [70].↵
- [71].↵
- [72].↵
- [73].↵
- [74].↵
- [75].↵
- [76].↵
- [77].↵
- [78].↵
- [79].↵
- [80].↵
- [81].↵
- [82].↵
- [83].↵
- [84].↵
- [85].↵
- [86].↵
- [87].↵
- [88].↵
- [89].↵
- [90].↵
- [91].↵
- [92].↵
- [93].↵
- [94].↵
- [95].↵
- [96].↵
- [97].↵
- [98].↵
- [99].↵
- [100].↵
- [101].↵
- [102].↵
- [103].↵