Cross-Organ Transcriptomic Comparison Reveals Universal Factors During Maturation

Abstract

Various cell types can be derived from stem cells. However, these cells are immature and do not match their adult counterparts in functional capabilities, limiting their use in disease modeling and cell therapies. Thus, it is crucial to understand the mechanisms of maturation in vivo. However, it is unknown if there are genes and pathways conserved across organs during maturation. To address this, we performed a time-series analysis of the transcriptome of the mouse heart, brain, liver, and kidney and analyzed their trajectories over time. In addition, gene regulatory networks were reconstructed to determine overlapping expression patterns. Based on these, we identified commonly upregulated and downregulated pathways across all four organs. Key upstream regulators were also predicted based on the temporal expression of downstream genes. These findings suggest the presence of universal regulators during organ maturation, which may help us develop a general strategy to mature stem cell-derived cells in vitro.

1. Introduction

Early mouse embryogenesis begins with fertilization and implantation, leading to the formation of a hollow egg cylinder by E5.0. During gastrulation, starting at E6.0, the three germ layers, endoderm, ectoderm, and mesoderm, are established (Robb et al, 2004; Takoaka and Hamada, 2012; Tam et al, 1997). Gastrulation is followed by organogenesis at E8.0, during which these three germ layers rapidly proliferate and differentiate into specialized organ systems (Robb et al, 2004; Takoaka and Hamada, 2012; Tam et al, 1997). Individual organs undergo organ-specific morphological changes, yet organogenesis is a coordinated process linked to the overall growth rate of the embryo (Kojima et al, 2014).

This coordination can be achieved by genes with similar expression profiles in multiple organs. Genes can be broadly characterized into two groups based on their spatiotemporal expression patterns across multiple tissues. Housekeeping genes, which are expressed ubiquitously, serve to maintain the basal cellular functions of a cell and are thus essential to general cell survival (Warrington et al, 2000; Eisenberg and Levanon, 2013). Selective genes, on the contrary, have enriched expression in several tissues (Liang et al, 2006). Transcriptional analysis has shown that these selective genes could mediate functional connectivity and organ cross talk among multiple organs (Qin et al, 2016). In the context of development, such cross talk would be crucial to achieving coordinated organogenesis.

In particular, transcriptional dynamics is crucial for coordinating embryonic development (Xue et al, 2011). One shift in transcription occurs during embryonic development and is characterized by increased expression of genes with early organ-specific functions and decreased expression of genes controlling cell division and general morphogenesis (Mitiku and Baker, 2007; Cardoso-Moreira et al, 2019). In mice, which are precocial, the second shift overlaps with birth and is characterized by increased expression of genes with late organ-specific function and again, decreased expression of genes controlling cell division and general morphogenesis (Cardoso-Moreira et al, 2019).

It may therefore be expected that transcriptional downregulation may be similar and coordinated across multiple organs, especially with regard to the decrease in proliferation that is generally characteristic of organogenesis and maturation. Furthermore, it can be hypothesized that transcriptional upregulation may follow certain similarities across multiple organs as well. From an evolutionary perspective, new regulatory mechanisms would develop by building upon existing ones. So as new organs evolved, their mechanisms of growth and development would likely share many similarities (Goss, 1990). Indeed, there are many common aspects of growth among different organs, such as the impact of use and disuse on promoting positive versus negative growth, hypertrophy and hyperplasia, and metabolic rate (Goss, 1990). Hormones that circulate through the bloodstream could provide a mechanism to coordinate changes in metabolism and growth of multiple organs throughout development.

We thus hypothesized that there may be common regulators of multiple organ development. To test this hypothesis, we analyzed the transcriptomes of hearts, brains, livers, and kidneys profiled at various developmental stages. The heart and kidney originate from the mesoderm, the liver originates from the endoderm, and the brain originates from the ectoderm. The four organs chosen thus represent all three germ layers. While the development of these four organs differs due to organ-specific functions, there are several important similarities. The development of these organs at early stages between E8 and E13 is characterized by proliferation and migration of cells as progenitor cells within these organs begin to form the overall structure of the organs (Takaoka and Hamada, 2012). Postnatally, cells in these organs all undergo structural, functional, and metabolic maturation. These processes are directed toward obtaining important functional characteristics of adult cells.

In the case of cardiomyocytes and neurons, electrophysiological maturation is crucial for organ function in adults, and functional coupling of these two cell types promotes mutual maturation (He et al, 2018; Lopaschuk et al, 2010; Oh et al, 2016). Moreover, major changes in metabolism and energy handling, with increases in mitochondrial biogenesis and mitochondrial complexity, occur during the development and maturation of cardiomyocytes, hepatocytes, neurons, and podocytes (Agostini et al, 2016; Lopaschuk et al, 2010; Nishikawa et al, 2014; Yuan et al, 2020). While stem/progenitor or immature cells rely mainly on glycolysis for adenosine triphosphate (ATP) production, which is necessary to meet their proliferation needs (Shyh-Chang et al, 2017), all four cell types switch to a primary reliance on oxidative phosphorylation for their energy needs by their adult mature stages.

2. Methods

2.1. Data set compilation

Affymetrix Mouse 430 2.0 microarray data sets of whole brain, liver, kidney, and hearts were compiled and curated from Gene Expression Omnibus (GEO) as done previously (Uosaki et al, 2015). The data sets covered all major developmental stages from embryonic to adult in all four organs. Data sets corresponding to wild-type and nontreated conditions were subsequently subset and clustered according to early embryonic, midembryonic, late embryonic, neonatal, young adult, and adult time points. In total, 479 brain, 147 liver, 142 kidney, and 212 heart data sets that fit the selection criteria were analyzed (Supplementary Table S1). Data were analyzed in R using custom scripts.

2.2. Principal component analysis

Principal component analysis (PCA) was run on the full probeset expression data using the prcomp() function. Probeset expression data were converted to gene expression data by selecting the probes with the greatest interquartile range of expression. PCA was performed on the combined gene expression data for all four organs as well on the organ-specific data.

2.3. Fuzzy clustering

Fuzzy clustering on the full gene expression data was performed using Mfuzz (Futschik et al, 2005; Kumar et al, 2007). The number of clusters was determined by iteratively increasing the number of clusters until the correlation between any two clusters exceeded 0.85. Up- and downregulated clusters were identified manually. Gene set enrichment analysis was performed using the fgsea package (Korotkevich et al, 2021). Gene set enrichment analysis (GSEA) was performed for each up- and downregulated cluster using the fuzzy cluster membership score as a gene ranking score.

2.4. Gene regulatory network

Weighted gene coexpression network analysis was also performed on the gene expression data (Langfelder et al, 2008). Organ-specific correlation networks were constructed using data across all time points, using a signed network with biweight midcorrelation, a soft thresholding power of 12, and a maximum block size of 21,000. Using the developmental stage as an external trait, the modules that were most positively and negatively correlated with the development and maturation were identified based on the correlation and associated p-value.

The top 30 genes with the highest intramodular connectivity scores were selected as hub genes for each module positively and negatively associated with maturation in each organ. A consensus network was also constructed using data from all four organs to identify modules preserved between the organs. The correlation of the organ-specific consensus module eigengenes was computed to determine eigengene relationships. GSEA analysis was performed using the genes assigned to the up and down modules, with module membership score as gene ranking.

2.5. Ingenuity pathway analysis

Ingenuity pathway analysis (IPA) was performed by identifying differentially expressed genes between each stage of organ development using limma. Genes with a log fold change >2 and a p-value <0.01 were inputted to IPA. This differential gene expression information for each organ was used for network reconstruction via IPA to identify upstream regulators activated and inactivated between each stage of development for each organ. Ggplot2 R package was used to generate (Wickham, 2016).

3. Results

3.1. Global transcriptomic maturation

A total of 479 brain, 147 liver, 142 kidney, and 212 heart murine microarray expression profiles, obtained from GEO, were combined to create a full organ data set. The data for each organ represent wild-type and untreated samples, covering the developmental stages of organogenesis and maturation. Data collected from time points between E8 and E11 were classified as “Early,” E12–E15 as “Mid,” E16-birth as “Late,” P3–P10 as “Neonate,” and 2–3 months to 30 weeks as “Adult.”

We first applied PCA to the combined data set of all four organs. PCA is a dimensionality reduction algorithm that can be used to identify principal components (PCs), or eigenvectors of the data covariance matrix, that capture maximal variance of the data features. We see that by looking at the first three PCs together, all four organs cluster together early in development, after which each follows its own unidirectional, nonbranching trajectory (Fig. 1a). PCA thus reveals that the differences between organs are more similar at early embryonic stages and diverge throughout the course of development and maturation. This aligns with the current view of organogenesis during development. Furthermore, plotting the PCs, PC2 versus PC3, we see that PC2 captures a common developmental trajectory between all the four organs studied (Fig. 1b). This indicates that there are common processes underlying the development and maturation of these organs, and therefore, potentially common regulators of maturation.

FIG. 1.

Cross-organ PCA clustering of transcriptomcs. (a) 3D PCA plot demonstrating three PCs simultaneously capturing the development and maturation of the heart, liver, kidney, and brain. Color follows the same legend as (b). Heart: 213 microarray data sets. Early (E8–11, N = 17, green), mid (E12–14, N = 39, black), late (E16–18, N = 26, orange) embryonic, postnatal (P3–10, N = 16, blue), and adult (N = 114, red). Brain: 479 data sets. Early (N = 18), mid (N = 38), late (N = 12), neonate (N = 32), and adult (N = 359). Kidney: 142 data sets. Early (N = 39), mid (N = 22), late (N = 2), neonate (N = 12), and adult (N = 67). Liver: 147 data sets. Early (N = 6), mid (N = 14), late (N = 18), neonate (N = 4), and adult (N = 104). (b) PC2 versus PC3 of development and maturation of all four organs analyzed in this study. PCA revealed that differences between organs were minimal at the early embryonic stage and were more obvious at the later stages. As maturation seems to happen via a similar trajectory in all organs, we hypothesized that there could be common maturation processes and presumably common regulators of maturation through organs. (c–f) PCA: PCA plots of single organ development. Single dot represents one microarray data set. Samples were clustered and aligned through PC1 axis as maturation progress. Each individual organ shows a clear unidirectional trajectory from early stage to adult stage. PCA, principal component analysis; PCs, principal components.

PCA of the data from each individual organ captures organ-specific developmental trajectories (Fig. 1c–f), with data collected from various time points clustered by developmental stage (early embryonic, midembryonic, late embryonic, neonatal, and adult). In the kidney and heart, the developmental trajectory is primarily captured by the PC. In the liver and brain, the developmental trajectory is captured by a combination of PC1 and PC2, with the difference in gene expression between early embryonic and adult stages driving the variance explained by PC1. Altogether, these results indicate that there is a common trajectory from immature to mature among the four organs as well as organ-specific trajectories that add additional dimensions unique to each specialized tissue.

3.2. Expression trends identify conserved pathways

After identifying similar trajectories underlying the development of all four organs, we sought to identify the genes driving these similarities. We reasoned that genes following similar gene expression time-course patterns between various organs may link functional connectivity between the organs. Such similarities could reflect responses to signaling molecules coordinating organogenesis and maturation. We performed fuzzy clustering of organ-specific gene expression time series data to identify genes with similar trajectories between organs using the MFuzz R package. For simplicity, we focused on two trajectories: upregulated, where expression progressively increased at each time point, or downregulated where expression progressively decreased at each time point. After selecting the strongest genes in the cluster based on membership score, we analyzed overlap between the upregulated and downregulated genes from different organs.

We ran Gene Set Enrichment Analysis to look for functional enrichment of gene ontology (GO) terms and Hallmark Pathways (as defined by the Molecular Signatures Database MSigDB) among the genes that were up- or downregulated in multiple organs (Supplementary Tables S2 and S3).

The genes that were downregulated (Fig. 2a–d) were largely involved in cell cycle processes, enriched for GO terms such as DNA replication, mitotic cytokinesis, and centrosome duplication (Fig. 2e). This follows the understanding that as development and maturation progress, cell division slows along with an increased focus on functional maturation. The genes that were upregulated (Fig. 3a–d) were largely involved in metabolism, enriched for GO terms such as Oxidative Phosphorylation, Electron Transport Chain, and Mitochondrial Matrix (Fig. 3e). No GO terms for upregulated genes in the heart reached our adjusted p-value cutoff of 0.05, but the trend of metabolic GO terms was present. This follows the understanding that as the physiological demand on cells increases throughout development, they must alter their energy handling apparatuses.

FIG. 2.

Fuzzy clustering of downregulated genes over maturation. (a–d) Fuzzy clustering plots of gene expression trends using mfuzz. With this form of soft clustering, genes are assigned a membership score between 0 and 1 for each cluster, in contrast to hard assignment to a single cluster. Clusters for downregulated genes are shown. Genes with cluster membership scores greater than 0.5 are shown, with lighter color indicating a higher degree of cluster membership. (e) Significant enriched GO terms for each organ shown, determined using genes with high membership scores (>0.5) to downregulated clusters. Color indicates p-value, and size indicates normalized enrichment score. Those with overlap across organs include GO terms relevant to cell division. GO, gene ontology.

FIG. 3.

Fuzzy clustering of upregulated genes over maturation. (a–d) Genes with cluster membership scores greater than 0.5 are shown, with lighter color indicating a higher degree of cluster membership. (e) Significant enriched GO terms for each organ shown, determined using genes with high membership scores (>0.5) to upregulated clusters. Color indicates p-value, and size indicates normalized enrichment score. Those with overlap across organs include terms relevant to metabolic processes.

3.3. Weighted correlation network analysis of multiorgan maturation

Given the similarities in the enrichment analysis of up- and downregulated genes during development across multiple organs, we next analyzed whether there were any similarities in the functional organization of these genes during development and maturation. Weighted gene coexpression network analysis was computed using the WGCNA R package. This identified modules of coexpressed genes in each organ (Supplementary Fig. S1) (Fig. 4). We determined which modules were positively (Fig. 4b, e, h, k) and negatively (Fig. 4a, d, g, j) correlated with time by computing the correlation of each module's eigengene with developmental time. The eigengene is the first PC of the gene expression matrix for a given module, and thus can be thought of as a weighted average representation of the gene expression profile for that module.

FIG. 4.

Weighted gene correlation network analysis of cross-organ maturation. (a–h) For each organ, the eigengene, a weighted average representation of the gene expression profile for that module, is shown for down- and upregulated genes through development. The eigenvalue for a given eigengene is plotted at each developmental stage. The top hub genes within these modules and their thresholded connections are shown using visANT. WGCNA reveals strong functional organization within the gene regulatory networks controlling development and maturation. Enriched GO terms determined from genes strongly associated with up- and downregulated modules are included in Supplementary Tables S1–S8. In agreement with the results of fuzzy clustering, common terms associated with the upregulated module across all four organs include metabolic terms, while those with the downregulated module include cell cycle terms.

The genes were compared among modules that were correlated with developmental time. We found that Cdkn2, a cell cycle repressor, was found in all the positively correlated groups for all four organs (Supplementary Table S4). GO term analysis of each module showed many cell cycle terms in the negatively correlated modules and metabolic terms in the positively correlated modules (Supplementary Tables S5 and S6). This indicates that metabolic and cell cycle suppression pathways become predominantly expressed in the adult cells. This was expected as cells slow their proliferation to maintain organ volume (Tumaneng et al, 2012). Through this analysis we confirm that we retain the temporal biological signal identified from fuzzy clustering in our coexpression network.

We then leveraged the network structure provided by WGCNA to identify hub genes within each module. For each module, the top 30 hub genes—genes with greatest intramodular connectivity—and their connections to each other were visualized with visANT (Fig. 4). Central genes within the hub network for the down modules include regulators of cell cycle progress and DNA replication, such as Cdc16 in the brain, Dbf4 in the heart, Cdk2ap1 in the kidney, and Cenpe in the liver. Central genes for the up modules included organ-specific genes such as Mog (involved in maintaining the myelin sheath of neurons) in the brain and Hnmt (histidine metabolism) in the kidney. Central genes in these hub networks also include genes involved in cellular respiration and fatty acid metabolism such as Me3 and Acacb in the heart, Pdk2 in the kidney, and genes associated with mitochondria, such as Efhd1 in the brain and Synj2bp in the liver. We also note the presence of genes involved in cell–cell communication and cell–matrix interactions such as Adam9 in the liver, Timp4 in the brain, and Jam2 in the heart, which are processes known to be important for organogenesis.

In addition to the intramodular features of these networks, the intermodular relationships in the coexpression networks were studied via eigengene correlations between modules. Consensus networks were calculated to identify shared modules among the four organs, so that intermodular relationships could be compared between organs (Supplementary Fig. S2). We then compared the consensus modules to organ-specific modules, looking for modules that maintained high preservation with all of the organ-specific up/down modules (Supplementary Figure S3). We thus identified one consensus up and one consensus down module. The module eigengenes for the consensus modules were computed with respect to each organ (the module eigengenes are still organ specific since they are based on the expression data from each organ), and the adjacency (correlation between 0 and 1) of the consensus up/down modules to the other consensus modules was plotted for each organ (Fig. 5).

FIG. 5.

Heatmap of consensus module eigengene relationships between the up- and downregulated modules and all other modules in the consensus coexpression network. Consensus networks were computed from WGCNA by using all four organ expression data, and the module eigengene correlations computed for each organ. Adjacency (scaled correlation from 0 to 1) is shown.

Hierarchical clustering of these modules separates the modules into groups associated with the up module (green-yellow to pink), the down module (purple to tan), and no strong correlation to either module (dark red to red). We note that between the up and down modules, the four organs cluster differently, but in both cases, the heart and brain cluster closest together. We note that the salmon consensus module, although having low correlation to the heart, brain, kidney up/down modules and liver down module, has a very high correlation with the liver up module. Analysis of this module reveals enrichment for GO terms relevant to biosynthetic and catabolic processes and hallmark pathways such as xenobiotic metabolism and bile acid metabolism.

Central hub genes for this module include signaling molecules such as Ltbr and Fgfr4. Another gene in this module is Fgf21, which is a hepatic factor secreted by the liver that affects energy expenditure in the central nervous system (Priest and Tontonoz, 2019), providing an example of a known mediator of organ cross talk correlated with maturation.

3.4. Prediction of upstream regulators of maturation

Next, based on the similarities in coexpression networks governing the development and maturation across the four organs, we sought to determine the common transcriptional regulators of maturation that may be driving these networks. We used Ingenuity Pathway Analysis (IPA) to generate z-scores for each upstream regulator through pairwise comparisons of gene changes between two adjacent time points. Differentially expressed genes between consecutive developmental stages were inputted to IPA giving us significantly enriched upstream regulators at each developmental time point. Positively enriched (activated) and negatively enriched (inhibited) upstream regulators were plotted across all time points and organs. This led us to find transcription factors that are consistently regulated in maturation in multiple organs (Supplementary Table S7).

Positively enriched upstream regulators included cell cycle genes and regulators of proliferation and growth such as Cdkn2a, Cebpa, and Hdac1/2 as well as regulators of metabolism such as the Ppar family (Fig. 6a). These results are in accordance with the analysis using fuzzy clustering and gene regulatory network analysis and follow our understanding of rapid proliferation and metabolic changes as two critical processes of development and maturation. We would expect proliferation to be important earlier in the development, and metabolic changes to be important later in the development; however, interestingly, activation of regulators from both of these categories tended to occur at similar stages of development.

FIG. 6.

Predicted upstream regulators of global maturation. (a, b) Heatmaps of the activation Z-scores of upstream transcriptional regulators correspond to activation changes from one stage to the next. Red: Higher in later stage, Blue: Lower in later stage. Upstream regulators were identified by running ingenuity pathway analysis on differentially expressed genes between consecutive stages of development. More than 400 upstream regulators were involved in the maturation process of the organs analyzed. Among the regulators, 28 and 19 regulators were identified as commonly activated and inactivated ones.

Negatively enriched upstream regulators included genes involved with cell fate and development, such as Gata2, Sox4, and Srf among others (Fig. 6b). Regulators involved in early fate decisions would naturally be inhibited as maturation progresses. Interestingly, several of these genes are strongly associated with development and maturation of single organs. For example, Tbx2 is required for mesoderm differentiation. However, it showed strong negative enrichment in the brain (ectoderm) and kidney (endoderm) as well. Other genes, such as Esr1, Igf2bp1, and Myc, are generally critical for processes such as cell proliferation and differentiation.

Altogether, the similarities in activated and inhibited transcriptional regulators include genes that might be expected due to their ubiquitous role in regulating gene expression and homeostatic processes. In addition, several genes that are well known for organ-specific roles appear to be important in other organs and germ layers, indicating a greater degree of functional connectivity among organs than previously appreciated.

To confirm that these results reflect true similarities in the maturation of organ-specific cell types rather than signal from common cell types such as fibroblasts or vascular endothelial cells, we compared our results with single-cell analysis of individual organ development. The downregulation of cell cycle genes has been shown to be significantly regulated in the brain (Di Bella et al, 2021), heart (Kannan et al, 2021b), and liver (Su et al, 2017), representing cross-organ similarities from all three germ layers.

Changes in metabolic state, such as fatty acid and lipid metabolism, accompany the maturation of all neurons (Zheng et al, 2016), cardiomyocytes (Kannan et al, 2021b), and hepatocytes (Mu et al, 2020). Furthermore, mitochondrial biogenesis has been shown to accompany maturation of neurons, cardiomyocytes, and hepatocytes, with the master regulator of mitochondrial biogenesis Ppara/Pgc1a upregulated throughout maturation in all three cell types (Agostini et al, 2016; Murphy et al, 2021b; Yang et al, 2017).

4. Discussion

In this study, we performed a meta-analysis of mouse microarrays to track gene expression during multiorgan maturation. A global trajectory of transcriptomic maturation was generated from PCA. We sought to capture the similarities in organ maturation through differential expression and gene regulatory network reconstruction. Although no single pathway appears to be a master regulator of maturation, genes involved in metabolism and cell cycle were consistently regulated among organs. These trends were expected as cell cycle rate decreases when organ size reaches homeostasis. In addition, the transition from glycolysis to fatty acid oxidation after birth has been well described in the heart (Murphy et al, 2021a).

We took multiple approaches to study similarities in pathways and genes associated with maturation to infer key upstream regulators. Briefly, we used fuzzy clustering to select genes that trended up or down in expression from early embryonic to adult time points, and to select genes that become more expressed or repressed during maturation. We see that the overlapping genes in the downregulated clusters were related to mRNA splicing, possibly indicating that a spliceosome construct associated with immature tissues is downregulated. Coexpression network analysis identifies similarities in network structure and genes associated with development as well as organ-specific differences. Finally, IPA identifies upstream regulators of differentially expressed genes that are commonly activated or deactivated between the four organs throughout maturation.

Other groups have studied comparisons between species such as human and mouse (Anzai et al, 2020), mouse and zebrafish (Irie et al, 2011), and human and macaque (Cardoso-Moreira et al, 2019). We decided to focus on the mouse as this built the most complete developmental time series. Since our meta-analysis used microarray data sets, we compared our results with those from single-cell RNA-seq analysis of individual organ development. Further validation with single-cell RNA-seq data could allow analysis with increased granularity and overcome confounders including the heterogeneity of maturation as shown earlier (Murphy et al, 2021b).

Further biological validation can also elucidate the specific mechanisms through which organ cross talk and coordinated maturation can be achieved. During homeostasis, for example, circulating factors such as hormones, chemokines, and cytokines coordinate organ responses to blood glucose levels and metabolic state (Guay and Regazzi, 2017; Priest and Tontonoz, 2019). Based on our analysis, this metabolic cross talk provides an intriguing possibility for coordinated organogenesis. Biological experiments can be used, to validate the importance of ligand–receptor pairs, among the identified factors in this study, that mediate cell–cell communication or circulating factors that mediate organ cross talk to achieve a coordinated program of development and maturation.

The approaches used here can also be used to study the relationship between in vivo maturation and in vitro maturation of pluripotent stem cell-derived cell types. For example, increasing expression of the positively enriched upstream regulators in stem cell-derived cardiomyocytes, hepatocytes, or neurons may promote their functional maturation. We note that activating the expression of PGC1α/PPAR, a regulator of metabolism and one of the positively enriched upstream regulators identified by IPA, indeed enhances cardiomyocyte maturation (Murphy et al, 2021b). It remains to be seen if activating such metabolic regulators can enhance the maturation of other stem cell-derived cell types. Their maturation status can also be quantified by use of transcriptomic analysis (Kannan et al, 2021a; Uosaki et al, 2015) to aid this evaluation. This may help alleviate the maturation issue in the stem cell field and hasten the translation to stem cell-based modeling and therapies (Murphy et al, 2021a).

Data Availability

The list of all compiled microarray data sets for the heart, brain, liver, and kidney and metadata annotations are included in Supplementary Table S1.

Authors' Contributions

S.K., S.M., and H.U. carried out the experiments. All authors designed the experiments. S.K. and S.M. wrote the article.

Author Disclosure Statement

The authors declare they have no conflicting financial interests.

Funding Information

This work was supported by the NIH (R01HL156947, T32HL007227), MSCRF (2019-MSCRFD-5044), and AHA (18EIA33890038).

Supplementary Material

Supplementary Table S1

Supplementary Table S2

Supplementary Table S3

Supplementary Table S4

Supplementary Table S5

Supplementary Table S6

Supplementary Table S7

Supplementary Table S8

Supplementary Figure S1

Supplementary Figure S2

Supplementary Figure S3