Figure 3. Lack of evidence for an increase in transcriptional noise of the murine aging lung and detection of an enrichment in immune cells.

(A) Bubble chart of transcriptional noise and cell type enrichment (old/young odds ratio [OR]) of 31 murine lung cell identities. The age-related change in transcriptional noise (horizontal axis) is computed by Scallop as the decrease in median membership score per cell identity between young and old cells. The enrichment of each cell type in old samples with respect to their young counterpart is represented as the old/young OR (vertical axis). The area of the bubbles represents the expected proportion of each cell type in the whole dataset according to the binomial generalized linear model fitted for that dataset. (B) Immune cell type enrichment but not age-associated increase in transcriptional noise, is consistently detected in old mice lungs. The increase in transcriptional noise associated with aging (left) and the cell type enrichment (right) are shown for several lung cell identities classified on the left as immune and non-immune. Cell identities present in at least three out of the four studied datasets are shown, and the age-related difference in transcriptional noise of missing cell identities is imputed from the remaining three measurements (mean difference across datasets).

Figure 3—figure supplement 1. Composition of the four single-cell RNA sequencing datasets of the murine aging lung used in this figure.

(A) Experimental approach. Four murine aging lung datasets were preprocessed and cell type-annotated. The cell-type labels from Angelidis were used as a reference to annotate the rest of the datasets. Differences in cell-type abundance between young and old mice were quantified using generalized linear models. From each dataset, eight subsets of related cell types were created to classify the 31 cell types into 8 categories, which were used as input for Scallop to analyze the differences in cell-to-cell variability. (b) Cell type-annotated mouse lung datasets. Uniform manifold approximation and projection plots showing the four datasets with their cell type annotations.

Figure 3—figure supplement 2. Qualitative ranking of murine aging lung cell types according to transcriptional noise and cell type enrichment.

The 31 detected lung cell types were classified in the Noise ranking (left) according to their greater age-related increase in noise. They were also classified in the Enrichment ranking (right) according to their greater enrichment in old samples. Cell categories that were represented by fewer than 100 cells were excluded from the transcriptional noise evaluation and therefore do not appear in the plot. Specific cell types are shown in the same color and with the same numbers as specified in the legend.

Figure 3—figure supplement 3. Comparison of the originally reported cell type-associated increase in transcriptional noise with the results obtained with Scallop.

The content of the first three columns was drawn from the original publications (Angelidis et al., 2019; Kimmel et al., 2019). More specifically, Angelidis_TN is the transcriptional noise per cell identity on the Angelidis dataset (from their Figure 2); Kimmel_OD is the gene overdispersion per cell type on the Kimmel dataset (from their Figure 2B); and Kimmel_DC is the cell-cell heterogeneity per cell identity measured as the Euclidean distance to the centroid of the cell identity for a particular age. Columns 4–7 summarize the results of our analysis of age-related loss of cell type identity in the murine lung. Specifically, Angelidis_S, Kimmel_S, TMS_FACS_S, and TMS_drop_S report the transcriptional noise per cell identity on the four datasets, measured as the difference in median membership score between young and old individuals. The cell identities used are those drawn from Angelidis. Since some cell identities from Kimmel dataset did not have a 1:1 correspondence to the Angelidis cell identities, they are shown using their original notation at the bottom of the table (‘Additional cell identities’). UP/DOWN: age-related increase/decrease in noise, NS: the difference in noise between young and old individuals is not statistically significant. NP: the cell identity was not present in the dataset in sufficient amounts to perform the analysis. For most cell types, it can be concluded that there is little overlap between cell identity-specific noise measurements across datasets and methods.