Fig 1.
scRNA-seq constructs a reference for CM maturation.
A. Mouse model used to generate perinatal maturation reference scRNA-seq library. In the aMHC-cre x mTmG mouse, CMs are labeled by GFP. This image was obtained and modified from “Brown Mouse Lab” by SVG-Clipart.com under a CC BY 4.0 license. B. UMAP dimensionality reduction (via Monocle 3) for the maturation reference. C. mnnCorrect-based integration of Wang and Yao et al. dataset with reference dataset. D. Our model for changes in gene distribution over CM maturation. As CMs undergo the maturation process, they transition from a broad gene distribution (characterised by high entropy) to a more narrow distribution (characterised by low entropy). E. Shannon Entropy S computed for each timepoint in the maturation reference dataset. F. Smoothed density estimates for genes expressed at 0–5000 counts per million (CPM) for each timepoint in the maturation reference dataset.
Fig 2.
Gene filtration facilitates cross-study comparisons of entropy.
A. Correction of mismapped mitochondrial reads. The use of a genomic mapping algorithm (such as zUMIs) or a “full” reference containing pseudogenes can lead to erroneous mapping of mitochondrial genes to pseudogenes, in turn inflating entropy score. We included a correction step that facilitates usage of data from a range of mapping pipelines. B. Proportion of ribosomal protein coding genes in mouse in vivo datasets, grouped by library preparation method. Given the clear protocol-dependence of these genes, we eliminated them from analysis. C. Entropy at different gene subsamplings for two studies with different sensitivities. Data from e18 and p22 from the maturation reference and p21 from Murphy et al. are shown. D. Spearman correlation between entropy and timepoint for different gene subsamplings (using median of entropy for each timepoint and study). E. Normalized variance of entropy for different gene subsamplings (using median of entropy for each timepoint). We normalized by scaling entropy at every subsampling to [0, 1].
Fig 3.
Standardized cell and study filtration enables meta-analysis of CM maturation with entropy.
A. Mitochondrial gene fractions in mouse in vivo datasets. Datasets with unusually high proportions are highlighted in red and were removed from subsequent analysis. B. Subsampling of count depth in Dueck et al. dataset (the highest depth dataset in our analysis). We subsampled to a depth where the median number of genes remained > 1000. We subsequently computed the accuracy as deviation from baseline entropy. C. Unusually low entropy cells due to high top 5 gene percentage (top) or low depth (bottom) in the Churko et al. dataset. D. SingleCellNet CM scores for mouse in vivo datasets by timepoint. Cells are labeled based on whether their highest classification was for “cardiac muscle”.
Fig 4.
Entropy score enables cross-study and cross-species comparison of CM maturation status, and recapitulates gene trends in CM maturation.
A. Workflow for computing entropy score from high quality scRNA-seq datasets. B. Entropy score for mouse and human in vivo CMs taken from publicly available datasets. HEW = human embryonic week, HPY = human postnatal year. C. Pearson correlation between entropy score and calculated pseudotimes for our maturation reference dataset for three trajectory inference methods: Monocle 2, Slingshot, and SCORPIUS. D. Venn diagram showing overlap in identified differentially expressed genes between entropy score and trajectory inference methods. Differentially expressed genes were identified by fitting generalized additive models to gene trends over the corresponding pseudotime in Monocle 2, and selecting genes with adjusted p-value < 0.05. E. Gene expression trends over entropy score for genes involved in CM maturation, including sarcomeric, cell cycle, metabolism, and calcium handling genes.
Fig 5.
Entropy score quantifies maturation status of PSC-CMs and iCMs.
A. Comparison in entropy score between human in vivo CMs and human PSC-CMs. Left side of figure reproduced from Fig 4B. B. Comparison in entropy between mouse in vivo CMs and mouse iCMs. Left side of figure reproduced from Fig 4B. C. Entropy score for three reprogramming pathways—a canonical Tnnt2+ iCM pathway and two alternative pathways (Ccnb1+ and Mmp3+).
Fig 6.
Entropy score decreases over maturation in non-CM tissue contexts.
A. Entropy score for mouse in vivo pancreatic beta cells taking from publicly available datasets. B. Entropy scores for mouse in vivo hepatocytes.