Differential requirement for DICER1 activity during the development of mitral and tricuspid valves

Congenital heart diseases (CHDs) occur in as many as 1–5% of newborns and remain the leading noninfectious cause of morbidity and mortality for infants in developed countries (Clark et al., 2006; Hoffman and Kaplan, 2002; Onuzo, 2006). Malformations of valves represent the major forms of CHDs, accounting for up to 30% of CHDs (Armstrong and Bischoff, 2004; Combs and Yutzey, 2009; LaHaye et al., 2014). Abnormalities during valve development can also predispose valves to late onset valvular diseases (LaHaye et al., 2014; O'Donnell and Yutzey, 2020). Therefore, understanding the molecular mechanisms underlying normal valvulogenesis has a strong translational significance.

Cardiac valves are essential for unidirectional blood flow. Valvulogenesis in mice is initiated with regional expansion of extracellular matrix (ECM) in the atrioventricular (AV) canal region and the outflow tract (OFT) region between embryonic day (E)9.0 to E10.0. Shortly after the accumulation of cardiac jelly, a group of endocardial cells in the AV and OFT conal cushions undergo epithelial-to-mesenchyme transition (EMT) to invade into the ECM. AV cushion mesenchymal cells are derived from endocardial cells and therefore endocardial deletion of a gene prior to EMT will lead to permanent deletion of the gene in cushion mesenchymal cells. Cellularized cushions undergo sophisticated remodeling processes, including condensation, elongation and ECM remodeling, and eventually become mature thin valve leaflets (Butcher and Markwald, 2007; Combs and Yutzey, 2009; de Vlaming et al., 2012; LaHaye et al., 2014; Lin et al., 2012; O'Donnell and Yutzey, 2020).

While the mitral (left) and tricuspid (right) valves are derived from similar cell sources through similar initiation and maturation processes, it is highly uncommon that both AV valves are affected simultaneously in CHD patients. Examination of the co-occurrence of different forms of CHDs in >3000 individuals revealed that the Odds Ratio between abnormal tricuspid valves and abnormal mitral valves was equal to 1, indicating that abnormalities of the two AV valves occur independently of one another (Ellesoe et al., 2018). Congenital polyvalvular heart disease mainly occurs in patients with trisomy 13 or 18, and are rarely observed in people with normal chromosomes (Bartram et al., 2001). These clinical observations cannot be easily explained by our current knowledge regarding the general mechanisms regulating AV valvulogenesis, as abnormalities in these mechanisms are expected to affect both valves. Rather, the non-association between mitral and tricuspid valve inborn dysplasia strongly suggests the presence of regulatory mechanisms that differentially regulate development of the two AV valves and defects affecting the differential regulatory mechanisms are likely the major cause of AV valvular anomalies in CHD patients.

MicroRNAs (miRNAs) are single-stranded RNA composed of 22–25 nucleotides that act predominantly on 3′ untranslated regions (3′UTRs) of target mRNAs to downregulate their stability and translation efficiency (Chen and Wang, 2012; Cordes and Srivastava, 2009; Malizia and Wang, 2011; O'Brien et al., 2018; Yan and Jiao, 2016). The DICER ribonuclease is essential for the generation of the mature miRNAs. The essential role of miRNA-mediated gene regulation on heart development has been well demonstrated by conditional inactivation of Dicer1 in cardiomyocytes (Chen et al., 2008a; Peng et al., 2014; Saxena and Tabin, 2010; Zhao et al., 2007), neural crest cells (Huang et al., 2010a,b; Nie et al., 2011; Zehir et al., 2010) and epicardial cells (Singh et al., 2011). However, the function of miRNAs in the endocardium and valves during mammalian cardiogenesis have been largely overlooked.

In this study, we tested how blocking mature miRNA biosynthesis by endocardial deletion of Dicer1 affects valvulogenesis using Nfatc1-Cre. Deletion of Dicer1 in the endocardial cells caused lethality within 3 days after birth. Furthermore, Dicer1 deletion led to mitral valve stenosis and regurgitation, whereas it only had a minor impact on other valves, including tricuspid valves. Our study therefore suggests, for the first time, that miRNA-mediated gene regulation is differentially required for mitral and tricuspid valve development. These results will help us understand the molecular basis for the clinical observations that congenital dysplasia rarely affects both AV valves simultaneously and further suggest that defects in miRNA-mediated gene regulation can be a potential etiological factor for inborn mitral valve stenosis and regurgitation.

To understand how blocking miRNA biosynthesis affects cardiac valvulogenesis, we inactivated Dicer1 in endocardial cells using Nfatc1-Cre mice, which can efficiently inactivate target genes in endocardial cells prior to EMT (Peng et al., 2016; Wu et al., 2012) (Fig. S1A). We crossed Nfatc1-Cre/Dicer1^loxp/+ male mice with Dicer1^loxp/loxp female mice (Cobb et al., 2005) to obtain conditional knockout (cKO; Nfatc1-Cre/Dicer1^loxp/loxp) and control (Dicer1^loxp/+ or Dicer1^loxp/loxp) animals. Our quantitative real-time RT-PCR (qRT-PCR) results confirmed that expression of Dicer1 transcripts was efficiently inactivated in both mitral and tricuspid valve primordium at E11.5 (Fig. S1B). Furthermore, whereas the level of precursors of three tested miRNAs was not reduced in mutant AV cushion mesenchymal cells, the mature forms were significantly decreased (Fig. S1C), suggesting that DICER activity was efficiently inactivated in mutants. Mutant mice were recovered at the Mendelian ratio (∼25%) until postnatal day 0 (P0). Starting from P1, the percentage of live mutant mice was reduced, and no mutants survived to P3 (Fig. 1A), indicating that endocardial deletion of Dicer1 is incompatible with postnatal survival. The left atrium (LA) of mutant hearts at P1 was dilated and heavily congested with blood (Fig. 1B). The leaflets of mutant mitral valves were hyperplastic, whereas the morphology of other valves was not overtly different from that in control mice (Fig. 1C–E). The orifice of mutant mitral valves was reduced and blood cells were trapped in the LA. To examine the functional relevance of the mutant phenotypes, we examined mitral flow in P1 mice by echocardiography color Doppler (Fig. 1F,G). The mitral flow rate was significantly increased in mutant samples, consistent with the observed hyperplastic morphology and reduced orifice of mitral valves. Regurgitation was observed in all mutant samples examined.

Our above studies showed that endocardial deletion of Dicer1 caused congenital mitral valve stenosis and regurgitation with only a minor impact on other valves. This result is somewhat surprising, as mitral and tricuspid valves are formed through similar processes. Blocking miRNA biosynthesis would be expected to affect both valves. Our reporter and qRT-PCR analyses (Fig. S1) excluded the possibility that Dicer1 was only efficiently inactivated in mitral vales but not in tricuspid valves. We thus demonstrated the differential requirement of Dicer1 between mitral and tricuspid valves during their development.

To determine at which stage mutant mitral valves started to become hyperplastic, we examined valve development starting from E9.5. Hyperplastic mitral valves became overt in mutants at E18.5 (Fig. 2A,B). To reveal the cellular mechanism underlying this phenotype, we examined cell proliferation, apoptosis and remodeling. We did not observe significantly altered cell proliferation/death in mutant mitral valves at E16.5–E18.5 (Fig. 2C–E). We revealed that the cell density (number of cells normalized against the area) in the middle and distal regions of mutant mitral valves was significantly decreased (Fig. 2F,G), suggesting that deletion of Dicer1 impaired cell condensation of mitral valves. Our further collagen gel contraction assay showed that the cells isolated from mutant mitral valves were less capable of contracting collagen gels (Fig. 2H). Our data collectively suggest that Dicer1 is required for normal remodeling of mitral valves at the perinatal stage and that reduced condensation might be the major factor causing hyperplastic mitral valves in mutants.

To understand the molecular defects in mutant mitral valves, we performed single cell RNA sequencing (scRNA-Seq) using the mitral valves dissected from E17.5 embryos, a stage when the morphological defect was not overt. Using the UMAP dimension reduction technique (Becht et al., 2019), these cells could be grouped into at least 29 clusters (Fig. S2). Based on known molecular markers of cells (Hill et al., 2019; Li et al., 2016), we revealed various cardiac cell types (Fig. 3A, Fig. S3A). We then focused on mesenchymal cells, which are largely derived from AV endocardial cells and are the major cell type undergoing remodeling. Expression of 169 genes was significantly altered in mutant samples by at least 20% with adjusted P<0.01 (Table S1). Gene ontology (GO) term enrichment analysis using Metascape (Zhou et al., 2019) revealed that ‘ECM organization’ and ‘Extracellular structure organization’ are the two most significantly affected pathways (Fig. S3B, Table S2). These pathways contain >20 ECM genes and examples of ten genes are shown in the volcano chart (Fig. 3B). Therefore, our molecular examination suggests abnormal expression of ECM genes as a major molecular defect in mutant mitral valves. At the same time, we cannot exclude that there is a contribution from other pathways to the observed valve defects. For example, expression of multiple transcription factors (such as Twist1, Msx1, Scx, Sox9 and Mef2c) was dysregulated in mutant samples and these genes are known to be important for valve development (Chakraborty et al., 2010; Chen et al., 2008b; Garside et al., 2015; Levay et al., 2008; Lockhart et al., 2013). It is likely that the ultimate valvular defects are determined by the combined effects from abnormalities of multiple pathways.

ECM plays critical roles in regulating valve formation and homeostasis (Kodigepalli et al., 2020; O'Donnell and Yutzey, 2020). Mutations in multiple ECM genes, such as COL1A1, COL3A1, FLNA and TNXB, can lead to valve diseases (Brady et al., 2017). A mutation in a ciliogenesis regulatory gene, DZIP1, was recently found to cause mitral valve prolapse and it was further determined that primary cilia acts through regulating ECM distribution and/or organization to promote mitral formation and/or homeostasis (Toomer et al., 2019). Our in vivo mouse genetic evidence clearly showed that Dicer1 regulates expression of >20 ECM genes, which cover many valve ECM components, including collagen, the core proteins of proteoglycans, elastic fibers, fibronectin and laminins.

We next performed trajectory and pseudotime analysis on valve mesenchymal cells using Monocle 3.0. The mesenchymal cells (control+mutant) could be grouped into five subclusters, and, based on pseudotime ordering, the trajectory started with subcluster 1, which sequentially gave rise to cells in subclusters 2–5 (Fig. 3C; Fig. S4A). Cells in subcluster 1 are closely adjacent to endocardial cells in the UMAP chart and are least matured cells (referring to cells that have undergone the fewest maturation processes; Fig. 3A). An increased percentage of cells in subcluster 1 was observed in the mutant sample compared to wild type (14% versus 5%) (Fig. 3D), suggesting that mutant mitral valves contain more immature mesenchymal cells. Interestingly, when performing trajectory analysis separately on control and mutant cells, the trajectory between subcluster 1 and other subclusters was broken in control samples (Fig. S4B). This result suggests that in wild-type mitral valves at E17.5, differentiation of cells in subcluster 1 into cells in other subclusters has already stopped, whereas such a trajectory still exists in mutant valves.

Among the ECM genes that were upregulated by Dicer1 deletion, we focused on Col1a1, Col3a1 and Tnxb. Mutations in these three genes are associated with mitral valve defects (Brady et al., 2017). We first verified increased expression of these genes at the protein level in mutant mitral valves (Fig. 4A). We then performed sequence analysis using miRNA target prediction programs (TargetScan and miR Walk) and identified multiple potential miRNA targets in their 3′UTRs (Fig. S4C). To test whether their 3′ UTRs are targeted by relevant miRNAs, we generated reporter constructs and performed reporter analysis using tsA58-AVM cells, which are temperature sensitive immortal AV cushion mesenchymal cells (Peng et al., 2016). Mimics of miR98-5p and miR29a reduced the activity of Col1a1 and Col3a1 reporters and the mimic of miR30c significantly reduced the activity of the Tnxb reporter (Fig. 4B). On the other hand, mimics of miR148a and miR196b had no effect on Col3a1 and Tnxb reporters, arguing against their direct role in regulating transcripts of Col3a1 and Tnxb (Fig. S4D). To further test whether mitral valves express corresponding miRNAs that target transcripts of Col1a1, Col3a1 and Tnxb, we performed qRT-PCR analysis, and confirmed that AV valve mesenchymal cells express multiple family members of the miR98 (also known as let7), miR29 and miR30 families (Fig. 4C). Interestingly, the expression level of multiple miRNAs of these families was significantly lower in tricuspid valves than in mitral valves. The reduced expression of miRNAs in tricuspid valves suggests that miRNA-mediated gene regulation plays a less important role in regulating ECM genes compared to mitral valves, helping explain the absence of overt defects in cKO tricuspid valves. We speculate that in mutant tricuspid valves, the loss of DICER1 can be sufficiently compensated for by other regulatory mechanisms. By contrast, in mitral valves, miRNA-mediated regulation of ECM gene expression plays a more substantial role, and compensatory mechanisms cannot sufficiently restore the normal level of ECM proteins in mutants. We predict that mutations in miRNA genes or genes involved in miRNA biosynthesis preferentially affect mitral valves in CHD patients. Therefore, miRNA-mediated gene regulation might serve as a specific therapeutic target for congenital mitral valve defects.

In summary, we unexpectedly observed that Dicer1 is differentially required for mitral and tricuspid valve development. Dicer1 is essential for proper expression of multiple ECM genes in mitral valves, and endocardial deletion of Dicer1 impaired maturation and remodeling of mitral valves. Our results help explain the clinical observation that co-occurrence of congenital defects affecting both AV valves is uncommon. These data will ultimately contribute to development of novel clinical applications toward diagnosis and treatment of AV valve defects.

This study was carried out in accordance with the Guide for the Care and Use of Laboratory Animals published by the US National Institutes of Health (NIH Publication no. 85–23, revised 2011). All protocols were approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Alabama at Birmingham. Mice were euthanized by inhalation of CO₂ followed by cervical dislocation. The Nfatc1-Cre and Dicer1^loxp/loxp mouse lines were described previously (Cobb et al., 2005; Wu et al., 2012). Mice were maintained on the C57BL/6 background. Nfatc1-Cre;Dicer1^loxp/+ male mice were crossed with Dicer1^loxp/loxp females to obtain control and cKO mice at different developmental stages. Staged littermates of embryos were collected, with stage E0.5 defined by the presence of a copulation plug. Primers and PCR conditions for genotyping were described previously (Peng et al., 2014; Yan et al., 2020).

All experiments were performed as described in previous studies (Liu et al., 2014; Peng et al., 2016; Yan et al., 2020). For hematoxylin and eosin (H&E) staining, mouse embryos and perinatal hearts were fixed in 4% paraformaldehyde overnight at 4°C. The next day, tissues were washed in PBS, dehydrated, and embedded in paraffin. 10 µm thick paraffin sections were stained with HE for morphological examination. For immunofluorescence, paraffin sections were dewaxed, rehydrated, and treated with citrate buffer for antigen retrieval. After being blocked with 10% serum in TBST at room temperature for 1–3 h, slides were incubated with a primary antibody (see below) overnight at 4°C, and then incubated with an Alexa-Fluor-conjugated secondary antibody (Thermo Fisher Scientific) at room temperature for 1 h. Finally, slides were briefly stained with 4′,6-diamidino-2-phenylindole (DAPI, Thermo Fisher Scientific) to reveal the nuclei. When needed, the intensity of the immunostaining signal was amplified by the TSA Plus Cyanine 3.5 System (PerkinElmer). Immunostained samples were observed using a Zeiss Axio fluorescent microscope. The primary antibodies used for immunostaining were purchased from Iowa Hybridoma Bank (cardiac myosin heavy chain, MF20, 1:500), Cell Signaling (cleaved caspase 3, #9661, 1:500), Abcam (Ki67, #ab15580, 1:500; COL1A1, #ab34710, 1:100), Novus Biologicals (COL3A1, #NB600-594, 1:100), and R&D systems (TNXB, #AF6999, 1:100).

Embryonic hearts at various stages were frontally sectioned to a 10 µm thickness. The area of each leaflet of all sections was measured using ImageJ, and then added to calculate the total area for each valve leaflet. The sections were stained with an antibody against cardiac myosin heavy chain (MF20) to avoid inclusion of cardiomyocytes in the measurement. Data were averaged from at least five independent hearts of each genotype and the total area for controls was set at 100%. To measure cell density in mitral valve leaflets, cardiac sections (E18.5) were stained with an MF20 antibody (against cardiac myosin heavy chain) to view cardiomyocytes and with DAPI to view total nuclei. Staining with the MF20 antibody helped us to exclude cardiomyocytes in our counting. The cell density of the base, middle and distal regions of anterior and posterior leaflets of mitral valves were calculated as total number of nuclei normalized against the area examined. At least five sections of each heart from four hearts of each genotype were quantified. No difference between anterior and posterior leaflets was observed and thus data from both leaflets of each mitral valve were combined.

Mitral flow was identified in control and mutant neonatal mice (P1) by echocardiography color Doppler and velocity was measured by pulsed wave Doppler using the Vevo 3100 ultra-high-resolution ultrasound system (Fuji, Visual Sonics, Canada) with an MX400 linear array mouse transducer. Pups were lightly anesthetized under inhaled isoflurane anesthesia (1.5–2.0% isoflurane in 100% O₂) and maintained on a heated platform (37°C) during echocardiographic analysis. Mitral valve inflow was localized in parasternal long axis 2 chamber view in the basal aspect posterior to the aortic valve and velocity measured after optimizing sample volume placement by pulsed wave Doppler. Maximal mitral flow velocity was determined in each pup and used in group analysis.

Collagen analysis was performed using the CytoSelect Cell Contraction Assay Kit (Cell Biolabs, Inc). Mitral valves were dissected from control and mutant hearts at E18.5 and were digested with 1 mg/ml collagenase and dispase (Sigma-Aldrich, #10269638001) at 37°C for 4 min followed by 0.25% trypsin at 37°C for 10 min to acquire single cell suspensions. Cells from three or four mitral valves of the same genotype were pooled to acquire 10,000 cells for one well of a 96-well plate. Cells were mixed with the collagen solution following instructions provided by the vendor. Plates were then incubated in a cell culture incubator (37°C with 5% CO₂) for 48 h. After incubation, the area of gels was measured using ImageJ. The area of gels without a mixture of cells was set at 100%.

The potential miRNA targets in the 3′ UTRs of Col1a1, Col3a1 and Tnxb were predicted by TargetScan (http://www.targetscan.org/vert_80/) and miRWalk (http://mirwalk.umm.uni-heidelberg.de/) programs. The 3′ UTRs of Col1a1, Col3a1 and Tnxb were PCR amplified using mouse genomic DNA and cloned into the pMIR-REPORT-LUC reporter vector (Ambion). The primers for PCR amplification were: forward 5′-ATCGACTAGTTTTGGAGCCAGGCAGGGTCAC-3′ and reverse 5′-AGCTAAGCTTTGGTCTAGGGAGCATCTCAGC-3′ for Col1a1, forward 5′-ATCGACTAGTCACCCAATACAGGTCAAATGC-3′ and reverse 5′-AGCTAAGCTTTATGGCTTGAATGAAGGTACC-3′ for Col3a1, and forward 5′-ATCGACTAGTTGCGACCCAGAAACTTCCAGG-3′ and reverse 5′-AGCTAAGCTTCAGTTTCTCCTTTATTGCTCC-3′ for Tnxb. miRNA mimics were purchased from Sigma (miR-98, #HMI0982; miR-30c, #HMI0458; miR-29a, #HMI0434; miR-196b, #HMI0325; miR-148a, #HMI0237; miRNA negative control, #HMC0002). The reporter constructs with different miRNA mimics (including a negative control miRNA) were co-transfected into tsA58-AVM cells (generated in house) using Lipofectamine 2000 (Invitrogen) as described previously (Peng et al., 2014). For transfection, 5, 10 and 20 pmol of miRNA mimics were added to each well of a 24-well plate containing 500 µl of medium (DMEM with 10% FBS, from Thermo Fisher Scientific). To normalize the transfection efficiency, a plasmid expressing a lacZ reporter driven by a constant promoter was included in the co-transfection. Luciferase activities were measured using the Luciferase Assay System (Promega, #E1500) and the Agilent Bio Tek Synergy 2 multi-mode microplate reader. β-galactosidase assays were performed with the same cell extract using the β-Galactosidase Enzyme Assay System with Reporter Lysis Buffer (Promega, #E2000). The luciferase activity was normalized against the lacZ activity of each culture. To examine expression of mature miRNAs through qRT-PCR, miRNA was purified from GFP+ cells of mitral leaflets or tricuspid leaflets from E18.5 Nfatc1-Cre;mTmG embryos using mirVana miRNA isolation kit (Ambion, #AM1560). Leaflets were digested with 0.25% trpsin, and GFP+ cells (endocardial+mesenchymal cells) were isolated using a FACS Aria II sorter (BD BioSciences, performed by the UAB Comprehensive Flow Cytometry Core facility). A total of four samples for each genotype were analyzed. For each sample, 10–15 leaflets were pooled together for RNA extraction. cDNA synthesis and qRT-PCR were performed using miRCURY LNA miRNA PCR Starter kit (Qiagen, #339320) according to the manufacturer's instructions. miRNA-specific PCR primers were purchased from Qiagen (miR-98-5p, #YP00204640; Let-7a-5p, #YP00205727; Let-7b-5p, #YP00204750; Let-7c-5p, #YP00204767; miR-29a-3p, #YP00204698; miR-29b-3p, #YP00204679; miR-29c-3p, #YP00204729; miR-30a-5p, #YP00205695; miR-30b-5p, #YP00204765; miR-30c-5p, #YP00204783). The level of mature miRNA expression was normalized against Snord65 (Qiagen, #YP00203910) expression. To measure pre-miRNA levels, GFP+ AV valve cells were isolated from E13.5 control and mutant hearts through cell sorting. A total of four samples for each genotype were analyzed. Total RNA was reverse transcribed using SuperScript VILO cDNA Synthesis Kit (Invitrogen) and qRT-PCR was performed using TaqMan pre-miRNA Gene Expression Assays (Applied Biosystems). Mirlet2c-2, Mir3c-2 and Mir98 were examined with assay IDs Mm04238179_s1, Mm04238167_s1 and Mm04238198_s1, respectively. Rn18s (Mm03928990_g1) was used as the input control.

Mitral leaflets were dissected from E17.5 control and mutant embryos and digested with 0.05% trypsin to obtain single cells. For each genotype, samples were pooled from three or four hearts. To collect viable single cells, 7-amino-actinomycin (7-AAD, Thermo Fisher Scientific, for labeling nonviable cells)-stained cells were sorted using a FACS Aria II sorter (BD BioSciences). Preparation of the single-cell transcriptome libraries was performed using 10x Genomics single cell 3′ v3 reagent kits, following the manufacturer's instructions. The prepared libraries were sequenced using Hiseq 4000 (Genewiz) with the expectation of at least 2000 total reads per cell. The 10x Genomics Cellranger software (version 6.0.2), mkfastq, was used to create the fastq files from the sequencer. After fastq file generation, Cellranger count was used to align the raw sequence reads to the mouse reference genome. The matrix tables created by count were then loaded into the R package Seurat (version 4.0.5), which allows for selection and filtration of cells based on quality control metrics, data normalization and scaling, and detection of highly variable genes (Butler et al., 2018). We followed the Seurat vignette (https://satijalab.org/seurat/articles/pbmc3k_tutorial.html) to create the Seurat data matrix object. In brief, we kept all genes expressed in more than three cells and cells with at least 200 detected genes. Cells with mitochondrial gene percentages >10% and unique gene counts of >5000 or <200 were discarded. The data were normalized using Seurat's NormalizeData function, which uses a global-scaling normalization method, LogNormalize, to normalize the gene expression measurements for each cell to the total gene expression. The result is multiplied by a scale factor of 10⁴ and the result is log-transformed. Highly variable genes were then identified using the function FindVariableFeatures in Seurat. We also regressed out the variation arising from library size and percentage of mitochondrial genes using the Seurat function ScaleData. We performed principal component analysis (PCA) of the variable genes as input and determined significant principal components on the basis of the Seurat JackStraw function. The first 50 principal components were selected as input for UMAP using FindNeighbors, FindClusters (resolution=0.8) and RunUMAP in Seurat. Cell types were identified by Seurat FindConservedMarkers for each cluster. These cell types were then used as labels in all downstream Seurat analyses and plots. Seurat DimPLot was used to create the UMAP plot with the updated cluster labels. To identify differentially expressed genes (DEGs) in each cell cluster, we used FindMarkers function in Seurat on the normalized gene expression data. Seurat DotPlot was used to plot the genes of interest to generate dot plot. The R package, EnhancedVolcano, was used to create the volcano plot and label the genes of interest. To identify single-cell trajectory, the Seurat object was converted to a Monocle3 (Trapnell et al., 2014) object using the ‘as.cell_data_set’ function in Seurat. The cells were clustered to determine partitions using Monocle3 ‘cluster_cells’ function. The next step was to generate the trajectory graph using ‘learn_graph’. The cells were then ordered in pseudotime using ‘order_cells’. The scRNA-Seq data have been deposited in the NCBI Gene Expression Omnibus (GEO) repository with the accession number GSE207226.

All data are shown as mean±s.e.m. An unpaired two-tailed Student's t-test was used to compare between control and mutant groups with P<0.05 considered statistically significant. P-values for each dataset are provided in the corresponding figures or figure legends. For detection of differentially expressed genes between control and mutant cells in scRNA-Seq analysis, the Benjamini–Hochberg adjusted P-values were calculated.

We thank Drs Bin Zhou (Elbert Einstein Medical School, New York, USA) and Matthias Merkenschlager (Imperial College London, UK) for providing Nfatc1-Cre and Dicer1-loxP mice, respectively. We thank the UAB Genomics Core Facility for performing Sanger sequencing.