ING4 and ING5 are essential for histone H3 lysine 14 acetylation and epicardial cell lineage development

The inhibitor of growth protein family comprises five members (ING1 to ING5) and is characterised by central nuclear localisation sequences (Ha et al., 2002; Scott et al., 2001), a carboxyterminal plant homeodomain (PHD) finger (Champagne et al., 2008; Palacios et al., 2006; Peña et al., 2006; Taverna et al., 2006) and a conserved domain, the ING domain (He et al., 2005), which forms a helix-loop-helix structure (Culurgioni et al., 2012). ING family proteins affect cell proliferation in normal cells as well as in cancer (Jacquet and Binda, 2021). In contrast to ING1 (Coles et al., 2007; Garkavtsev et al., 1996; Garkavtsev and Riabowol, 1997; Kichina et al., 2006) and ING2 (Nagashima et al., 2001; Pedeux et al., 2005; Saito et al., 2010), which act as tumour suppressors, ING4 and ING5 promote cell proliferation (Doyon et al., 2006; McClurg et al., 2018; Ormaza et al., 2019). Although ING4 levels were found to correlate negatively with glioma progression (Garkavtsev et al., 2004), silencing of ING4 reduces 5-bromo-2′-deoxyuridine (BrdU) incorporation, and silencing of ING5 blocks DNA synthesis in HEK293T cells (Doyon et al., 2006).

The PHD fingers of ING proteins bind to histone H3 trimethylated on lysine 4 (H3K4me3) (Champagne et al., 2008; Ormaza et al., 2019; Palacios et al., 2010; Peña et al., 2006, 2008; Shi et al., 2006; Taverna et al., 2006). The human nucleosomal acetyltransferase of histone H3 protein (NuA3) complex contains the inhibitor of growth proteins ING4 or ING5 (Doyon et al., 2006; Feng et al., 2016). The enzymatic components of the human NuA3 complex are the MYST family histone lysine acetyltransferases KAT6A (MOZ/MYST3), KAT6B (MORF/QKF/MYST4) or KAT7 (HBO1/MYST2) (Doyon et al., 2006). In isolation, these histone acetyltransferases are highly promiscuous with respect to their substrates, whereas in the context of their protein complex, they gain specificity for certain histone residues (reviewed by Voss and Thomas, 2018).

Histone acetylation weakens the interaction between histone proteins and DNA (Sung and Dixon, 1970) and is essential for gene activity (Allfrey et al., 1964). It allows the transcriptional machinery and transcription factors access to the DNA (Lee et al., 1993; Vettese-Dadey et al., 1996; Zentner and Henikoff, 2013). Histone acetylation is crucial for the establishment of new patterns of gene expression, a process which is integral to embryonic development and cell differentiation (Kueh et al., 2023, 2011).

KAT6A, KAT6B and KAT7 are essential in embryonic development (Kueh et al., 2011; Thomas et al., 2000; Voss et al., 2009, 2012). Kat6a^−/− mice die between embryonic day (E) 13.5 and birth and display anterior homeotic transformation (Voss et al., 2009), craniofacial and heart defects (Voss et al., 2012) and an absence of haematopoietic stem cells (Thomas et al., 2006). Hox, Tbx and Dlx mRNA levels are reduced, Hox gene expression is shifted posteriorly and histone H3 lysine 9 acetylation (H3K9ac) is reduced at these gene loci in Kat6a^−/− embryos (Vanyai et al., 2019; Voss et al., 2009, 2012). Kat6b genetrap mice (Kat6b^gt/gt), which retain only 10% normal Kat6b mRNA, die perinatally with craniofacial, brain and skeletal anomalies (Thomas et al., 2000), and Kat7^−/− mice fail to form a chorioallantoic placenta and arrest in development at E8.5 (Kueh et al., 2011). Kat7^−/− embryos display an absence of H3K14ac (Kueh et al., 2011) and fail to initiate the expression of several embryonic patterning genes, including cardiovascular patterning genes (Nkx2-5, Gata4, Tbx1), body segment identity genes (Hoxa2, Hoxa3) and neuronal patterning genes (Sox2, Otx2) (Kueh et al., 2011). Heterozygous mutations of one allele of the KAT6A or the KAT6B gene cause Arboleda–Tham syndrome or Say–Barber–Biesecker–Young–Simpson variant of Ohdo syndrome and genitopatellar syndrome, respectively (Arboleda et al., 2015; Campeau et al., 2012; Clayton-Smith et al., 2011; Tham et al., 2015), which are characterised by intellectual disability and distinct facial features and include, in approximately half of the cases, heart defects.

The role of ING5 in embryonic development has not been reported, but Ing4^−/− mice have been reported to be viable, fertile and largely healthy (Coles et al., 2010). ING5 most closely resembles ING4 with 79% and 92% sequence identity in their ING domain and PHD finger, respectively. Therefore, we hypothesised that ING4 and ING5 might have redundant functions in embryonic development.

To examine the unique and overlapping roles of ING4 and ING5, we generated Ing4 and Ing5 mutant mice and determined their function in embryonic development. Our results indicate shared and unique roles of ING4 and ING5 in early post-gastrulation development and in the formation of the cardiac ventricular septum, the epicardium and epicardium-derived cell types.

We generated mice with a deletion of exons 3-5 of the Ing5 locus (Ing5 null allele, Ing5⁻; Fig. S1A,B). The Ing5⁻ allele did not produce detectable mRNA (Fig. S1C). Ing5^−/− mice were present at a normal Mendelian ratio at E18.5 but were under-represented among the offspring of Ing5^+/−×Ing5^+/− matings at weaning (3 weeks old; P<10⁻⁶; Fig. 1A). Examination of serial sections of E18.5 hearts revealed that five of six Ing5^−/− foetuses had cardiac ventricular septal defects, which were not observed in ten Ing5^+/+ wild-type littermate control foetuses (Fig. 1B; Fig. S1D). No other macroscopic, histological, skeletal (Fig. S1E) or craniofacial anomalies were observed in the E18.5 Ing5^−/− foetuses (n=7 Ing5^−/−, 13 Ing5^+/− and 14 Ing5^+/+ foetuses).

ING5 occurs in the KAT6A, KAT6B and KAT7 histone acetylation complexes (Doyon et al., 2006; Feng et al., 2016). However, the nature of the anomalies observed in E18.5 Ing5^−/− foetuses here appeared much more restricted than the anomalies observed in Kat7^−/− embryos, Kat6b^gt/gt foetuses and Kat6a^−/− foetuses (Kueh et al., 2011; Thomas et al., 2000; Voss et al., 2009, 2012). Therefore, we hypothesised that another protein might compensate for the lack of ING5. Examination of the ING family protein amino acid sequences revealed that ING4 was most similar to ING5 (Fig. S1F).

Wholemount in situ hybridisation in E10.5 wild-type embryos showed that Ing4 and Ing5 mRNA was expressed widely throughout embryonic tissues (Fig. 1C). However, the heart expressed higher levels of Ing5 mRNA than Ing4 mRNA relative to the rest of the body (Fig. 1D). The loss of Ing5 mRNA in Ing5^−/− mutant embryos and hearts did not lead to an increase in Ing4 mRNA (Fig. 1E,F).

To assess a potential functional overlap between ING4 and ING5, we generated mice with a deletion of exons 3-6 of the Ing4 locus (Ing4 null allele, Ing4⁻; Fig. S1G,H). The Ing4⁻ allele did not produce detectable mRNA (Fig. S1I). By intercrossing Ing4^+/−Ing5^+/− double-heterozygous mice, we generated all possible genotype combinations. Wild-type mice, Ing4^+/− mice and Ing4^−/− mice were not under-represented at weaning (Fig. 1G), as previously reported (Coles et al., 2010). Ing5^+/− and Ing4^+/−Ing5^+/− mice were also not under-represented at weaning (Fig. 1G). As observed above (Fig. 1A), Ing5^−/− mice were under-represented (P<10⁻⁶). Mice lacking both alleles of Ing4 and one allele of Ing5 (Ing4^−/−Ing5^+/−) were also under-represented (P=9×10⁻⁵). Importantly, mice lacking one allele of Ing4 and both alleles of Ing5 (Ing4^+/−Ing5^−/−) and mice lacking all four Ing4;Ing5 alleles (Ing4^−/−Ing5^−/−) were absent at weaning (Fig. 1G).

Macroscopic examination at E9.5 and E10.5 revealed that Ing4^−/−Ing5^−/− embryos failed to make a connection between the allantois and the chorionic plate, instead forming a bulbous-shaped allantois (Fig. 1H,I), and were arrested in development at the E8.5 stage (Fig. 1I), as reported for embryos lacking KAT7 (Kueh et al., 2011).

Gene expression analysis comparing whole E8.75 Ing4^−/−Ing5^−/− embryos to stage-matched E8.25 wild-type embryos (Fig. S2; Tables S1 and S2) revealed that 33 DNA-binding transcription factor genes (GO:0003700), which are essential for embryonic pattern specification (GO:0007389), were downregulated in Ing4^−/−Ing5^−/− compared with wild-type embryos (Fig. 1J). The downregulated genes included Sox1, Otx2 and Gata4, which are also downregulated in embryos lacking KAT7 (Kueh et al., 2011).

To investigate how histone acetylation was affected by loss of ING4 and ING5, we isolated mouse embryonic fibroblasts (MEFs) (Fig. S3A) from E9.5 embryos obtained from Ing4^+/− Ing5^+/−×Ing4^+/−Ing5^+/− matings and assessed the reported histone acetylation targets of KAT7, KAT6A and KAT6B, namely H3K14, H3K23 and H3K9 (Doyon et al., 2006; Feng et al., 2016; Klein et al., 2019; Kueh et al., 2023, 2011, 2020; Lv et al., 2017; Mishima et al., 2011; Simó-Riudalbas et al., 2015; Voss et al., 2009, 2012). Immunofluorescence staining revealed that H3K14ac in Ing4^−/− Ing5^−/− MEFs was reduced almost to the level of the negative control (Fig. 2A,B; Fig. S3B). H3K23ac was modestly reduced in Ing4^−/−Ing5^−/− MEFs compared with control MEFs (Fig. 2C-E; Fig. S3C), but H3K9ac levels were not affected (Fig. S3D,E).

Assessment of cell proliferation in MEFs isolated at E12.5 followed by Ing4;Ing5 deletion in vitro with a tamoxifen-inducible CreERT2 (Seibler et al., 2003) revealed that Ing4^fl/flIng5^+/+ CreERT2 single knockout and Ing4^+/+Ing5^fl/fl CreERT2 single knockout MEFs proliferated at a similar rate as wild-type cells (Fig. 2F). In contrast, Ing4^fl/flIng5^fl/fl CreERT2 double-knockout MEFs arrested in cell growth at day 6 of culture (Fig. 2F). Cells that retained only one functional allele of the four Ing4 and Ing5 alleles displayed a reduction in proliferation (Fig. 2F).

Immunoblotting confirmed the reduction in H4K14ac and H3K23ac levels in Ing4^−/−Ing5^−/− MEFs compared with wild-type MEFs (Fig. 2G,H). A slight reduction in H3K9ac levels was also observed with this approach (Fig. 2G,H).

These findings suggest that the combined complete loss of ING4 and ING5 in MEFs causes a loss of H3K14 acetylation, a target of KAT7 (Kueh et al., 2011), and a reduction in H3K23ac, the target of KAT6A and KAT6B (Klein et al., 2019; Lv et al., 2017; Simó-Riudalbas et al., 2015). In addition, the combined complete loss of ING4 and ING5 resulted in growth arrest in MEFs, as reported for Kat6a^−/− MEFs (Baell et al., 2018; Sheikh et al., 2015).

To investigate the absence of Ing4^+/−Ing5^−/− mice at weaning (Fig. 1E), we examined litters of Ing4^+/−Ing5^+/−×Ing4^+/−Ing5^+/− matings at developmental stages ranging from E8.5 to E16.5. Ing4^+/−Ing5^−/− embryos were under-represented from E13.5 onward (Fig. S4A) and displayed morphological anomalies from E10.5 onwards (Fig. S4B). Anomalies that were occasionally observed in Ing4^+/−Ing5^−/− embryos were a small body size (E10.5), a collapsed rhombencephalon (E11.5, E12.5), small eyes, and oedema in excess of physiological oedema (E13.5) (Fig. S4C). Notably, heart anomalies were the most frequently observed anomalies in Ing4^+/−Ing5^−/− embryos from E10.5 onward and reached full penetrance at E13.5 (Fig. 3A). The heart anomalies observed included a rounded cardiac apex, an indistinct interventricular sulcus and epicardial blebbing in excess of physiological blebbing (Fig. 3B; Fig. S4D). Serial sections of Ing4^+/−Ing5^−/− hearts from E11.5 to E14.5 revealed ventricular septal defects, an indistinct interventricular sulcus, flat apex, atrioventricular cushion remodelling defects, over-riding aorta, a thin compact myocardium layer, excessive epicardial blebbing and malformed atrioventricular and semilunar valves compared with control hearts (Fig. 3C-E; Fig. S5A-C). Atrioventricular septal defects occurred in E14.5 Ing4^+/−Ing5^−/− embryos (Fig. 3E). The more severe anomalies observed histologically (e.g. E11.5 Ing4^+/−Ing5^−/− hearts in Fig. S5A) likely corresponded to early losses observed in Ing4^+/−Ing5^−/− embryos (Fig. S4A).

Morphometric assessment confirmed that the compact myocardium layer was significantly thinner in the Ing4^+/−Ing5^−/− hearts compared with control hearts, whereas the trabecular layer had a normal depth (P<0.0001; Fig. 3F,G). In addition, the Ing4^+/−Ing5^−/− hearts were slightly smaller in the basal apical and latero-lateral axes (Fig. S5D). A number of these anomalies were also observed in Ing4^+/+Ing5^−/− hearts compared with control hearts, including the septal defects already shown in Fig. 1 and the over-riding aorta (Fig. 3D,E; Fig. S5A), as well as the thin compact myocardium (Fig. 3D,F,G; Fig. S5C). Unlike the Ing4^−/−Ing5^−/− embryos (Fig. 1F), Ing4^+/−Ing5^−/− embryos formed normal chorioallantoic placentae comparable to control placentae (Fig. S6).

These data suggest that loss of both alleles of Ing5 causes severe heart anomalies, which are exacerbated and extended by the additional loss of one allele of Ing4.

To examine the effects of the Ing4^+/−Ing5^−/− mutation on gene expression, we assessed mRNA levels in Ing4^+/−Ing5^−/−, Ing4^+/+Ing5^−/−, Ing4^+/−Ing5^+/+ and Ing4^+/+Ing5^+/+ hearts at E10.5 by RNA sequencing. At this stage in development, the majority of Ing4^+/−Ing5^−/− embryos did not display heart anomalies (Fig. 3A). Cell death and cell cycle parameters in Ing4^+/−Ing5^−/− hearts were similar to controls, as determined by terminal deoxynucleotidyl transferase dUTP nick end labelling (TUNEL) and BrdU incorporation and DNA content assessment by flow cytometry (Fig. S7A-C).

The E10.5 heart samples clustered within genotype and segregated between genotypes based on their mRNA level profiles (Fig. 4A,B). Ing4^+/−Ing5^−/− hearts had expression profiles that differed substantially from the Ing4^+/+Ing5^+/+ controls, whereas Ing4^+/+Ing5^+/+ and Ing4^+/−Ing5^+/+ hearts had similar expression profiles (Fig. 4C; Table S3; Fig. S8A). The expression profile of Ing4^+/+Ing5^−/− hearts appeared to be intermediate between Ing4^+/−Ing5^−/− and Ing4^+/+Ing5^+/+ hearts (Fig. 4C). Congruently, the changes in mRNA levels observed in Ing4^+/+Ing5^−/− hearts compared with control hearts correlated positively and highly with the changes in mRNA levels observed in Ing4^+/−Ing5^−/− hearts compared with control hearts (P=0.0003; Fig. S8B,C).

A total of 446 genes were differentially expressed between Ing4^+/−Ing5^−/− and Ing4^+/+Ing5^+/+ hearts [false discovery rate (FDR)<0.05; 164 downregulated, 282 upregulated; Table S3]. Examining marker genes for individual cardiac cell types revealed that genes expressed in epicardium were significantly downregulated Ing4^+/−Ing5^−/− compared with Ing4^+/+Ing5^+/+ hearts (Fig. 4D), whereas no obvious directional change was observed for myocardial genes (Fig. 4E) or endocardial/endothelial genes (Fig. 4F). Genes encoding regulators of heart development were generally unaffected (Fig. 4G). Other gene categories examined were also largely unaffected, including genes annotated as cardiac epithelial-to-mesenchymal transition genes, genes encoding extracellular matrix proteins, signalling pathway components important for heart development, cell cycle and apoptosis proteins, ING family proteins other than ING4 and ING5, histone acetyltransferases, other proteins in the NuA3 complexes and histone deacetylases (Fig. S8D-L).

Genes downregulated in Ing4^+/−Ing5^−/− compared with Ing4^+/+Ing5^+/+ hearts were enriched for gene ontology (GO) terms including cardiovascular development related terms (P=3×10⁻⁶ to 2×10⁻¹¹; Fig. 4H; Table S3) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, including pathways involved in cardiac pathology (P=0.046 to 4×10⁻⁴; Fig. 4I; Table S3). In contrast, upregulated genes were enriched for GO terms relating to catabolic and metabolic processes and not for cardiac-specific KEGG pathways (Table S3).

As noted above KAT6A forms a complex with ING5 (Doyon et al., 2006) and is required in heart development (Voss et al., 2012). To determine whether the changes in mRNA levels observed in Ing4^+/−Ing5^−/− compared with Ing4^+/+Ing5^+/+ E10.5 hearts could be associated with a dysfunction of the KAT6A complex, we assessed mRNA levels in Kat6a^−/− versus Kat6a^+/+ E10.5 hearts by RNA sequencing. Like E10.5 Ing4^+/−Ing5^−/− hearts, E10.5 Kat6a^−/− hearts appeared normal at this stage in development. The mRNA expression profiles of Kat6a^−/− and Kat6a^+/+ E10.5 hearts clustered within genotype and segregated between genotypes (Fig. 5A). A total of 412 genes were differentially expressed between Kat6a^−/− and Kat6a^+/+ hearts (FDR<0.05; 202 downregulated, 210 upregulated; Fig. 5B,C; Table S4). The changes in mRNA levels observed in Kat6a^−/− hearts compared with Kat6a^+/+ control hearts correlated positively and strongly with the changes in mRNA levels observed in Ing4^+/−Ing5^−/− hearts compared with Ing4^+/+Ing5^+/+ control hearts (P=0.001; Fig. 5D; Fig. S9A).

Genes downregulated in Kat6a^−/− compared with Kat6a^+/+ hearts were enriched for GO terms relating to cardiovascular development, tube morphogenesis and cell differentiation (Fig. 5E), as they were in Ing4^+/−Ing5^−/− versus Ing4^+/+Ing5^+/+ hearts (Fig. 4H), whereas genes upregulated in Kat6a^−/− compared with Kat6a^+/+ hearts were enriched for GO terms relating to heart contraction, myofilaments and striated muscle (Fig. 5F). Assessment of marker genes of cardiac cell types showed that epicardium genes were downregulated in Kat6a^−/− compared with Kat6a^+/+ hearts (Fig. 5G), myocardium genes tended to be upregulated in Kat6a^−/− hearts (Fig. 5H) and endocardium/endothelial genes were unchanged (Fig. 5I). Genes encoding factors regulating heart development were downregulated in Kat6a^−/− hearts compared with Kat6a^+/+ hearts (Fig. 5J), unlike Ing4^+/−Ing5^−/− hearts, in which this group of genes was generally unaffected (Fig. 4G). Like in Ing4^+/−Ing5^−/−hearts, other groups of genes assessed were largely unaffected (Fig. S9B-J).

Our RNA-sequencing data suggested that the epicardial lineage could be compromised in Ing4^+/−Ing5^−/− compared with Ing4^+/+Ing5^+/+ E10.5 hearts. External and histological examination revealed increased blebbing in of the epicardium of Ing4^+/+Ing5^−/− and Ing4^+/−Ing5^−/− compared with Ing4^+/+Ing5^+/+ E12.5 hearts (Fig. 6A-F; Fig. S4A). Immunofluorescence staining for the proteins WT1 (Fig. 6G-I; Fig. S10A) and PDGFRα (Fig. 6J-O; Fig. S10B), which mark epicardium and epicardium-derived cells (EPDCs) (Chong et al., 2011; Moore et al., 1999; Orr-Urtreger et al., 1992), suggested a reduction or defect in epicardial cells and epicardial derived cells on the surface of Ing4^+/+Ing5^−/− and Ing4^+/−Ing5^−/− compared with Ing4^+/+Ing5^+/+ E12.5 hearts. Enumeration of WT1-positive cells covering the surface of the heart (epicardial cells) showed an Ing4;Ing5 gene dosage-dependent decline in epicardial cells (regression analysis, R²=0.6; Fig. 6P). To determine whether the reduction in epicardial cells could simply be due to a developmental delay, we also assessed wild-type E11.5 hearts. The number of WT1-positive cells covering the surface of the wild-type E11.5 heart displayed considerable variance and did not differ significantly from that of E12.5 wild-type hearts (Fig. 6P; Fig. S10A), suggesting that individual wild-type E11.5 embryos may be variably advanced in epicardium development.

Epicardial cells undergo epithelial-to-mesenchymal transition (EMT) to populate the subepicardial space as EPDCs, from where they invade the myocardium as the primary source of cardiac fibroblasts. EPDCs also contribute to pericytes and vascular smooth muscle cells of the coronary vascular network. Contribution to other cardiac cell types remains controversial (Fig. 7A; reviewed by Cao et al., 2020; Simões and Riley, 2018). Subepicardial EPDCs express the epicardial marker proteins WT1 and PDGFRα, as well as the fibroblast marker vimentin (Orr-Urtreger et al., 1992; Moore et al., 1999; Zamora et al., 2007).

Examination of the subepicardial space at the left and right atrioventricular junctions (AVJs) and in the interventricular sulcus (Fig. S10A) on Haematoxylin and Eosin (H&E)-stained histological sections (Fig. 7B,C; Fig. S11A) and sections stained with immunofluorescence for WT1 (Fig. S10A), PDGFRα (Fig. 7D; Fig. S11B) and vimentin (Fig. S11C) suggested that subepicardial cells were reduced in Ing4^+/+Ing5^−/− and Ing4^+/−Ing5^−/− compared with Ing4^+/+Ing5^+/+ E12.5 hearts. Enumeration of the subepicardial cells in the AVJs confirmed that the number and density of EPDCs were reduced in Ing4^+/−Ing5^−/− compared with Ing4^+/+Ing5^+/+ E12.5 hearts, in absolute numbers (P=0.0009; Fig. S11D), relative to the number of overlaying epicardial cells (P=0.004; Fig. S11E) and relative to the area of the AVJ (P=0.04 to 0.005; Fig. 7E,F). Ing4^+/−Ing5^−/− E12.5 hearts displayed a tendency for a reduction in EPDCs even when compared with wild-type E11.5 hearts (P=0.03; Fig. 7F; Figs S10A,B, S11A). The paucity of EPDCs per area (Fig. 7F) and per overlying epicardial cells (Fig. S11E) in E12.5 Ing4^+/−Ing5^−/− compared with control hearts suggests a defect in the epicardial lineage.

In addition to subepicardial cells, we assessed coronary blood vessel endothelial cells. EPDCs affect coronary vascular development by contributing vascular support cells (pericytes and smooth muscle cells), whereas coronary endothelial cells arise from the sinus venosus on the dorsal surface of the ventricles, forming a vascular network that gradually expands to cover the entire ventricle surface (Cao et al., 2020). Examination of E12.5 hearts stained by wholemount immunohistochemistry for the blood vessel endothelial cell marker PECAM1 suggested that the coronary vascular network of Ing4^+/+Ing5^−/− and Ing4^+/−Ing5^−/− hearts was underdeveloped compared with Ing4^+/+Ing5^+/+ hearts (Fig. 7G,H). The coronary vascular network in E12.5 Ing4^+/−Ing5^−/− hearts appeared underdeveloped even compared with wild-type E11.5 hearts (Fig. 7H). Morphometric assessment of the relative area of the dorsal ventricle surface covered by the PECAM1-positive coronary vascular network revealed an Ing4;Ing5 gene dosage-dependent decline in the area covered (R²=0.6; Fig. 7I). Higher magnification images (Fig. 7H) revealed that, apart from the obvious differences in the size of the vascular network, the E11.5 heart and all E12.5 Ing4;Ing5 genotypes, except Ing4^+/−Ing5^−/−, were covered with dispersed PECAM1-positive cells. In contrast, the E12.5 Ing4^+/−Ing5^−/− hearts displayed clumping of PECAM1-positive cells in restricted areas.

A group of genes that are known to be expressed in epicardial cells and are relevant to cell adhesion (GO:0007155) were downregulated in E10.5 Ing4^+/−Ing5^−/− versus Ing4^+/+Ing5^+/+ hearts (FDR=7×10⁻⁶ to 0.049; Fig. 7J; Table S3). The most significantly downregulated gene was the α4-integrin gene (Itga4). Apart from a well-known role in leukocyte-vascular endothelial cell interaction, α4-integrin and its ligand VCAM1 are important for the fusion of the allantois to the chorionic plate and the migration and spreading of epicardium cells when covering the myocardium (Kwee et al., 1995; Yang et al., 1995).

Retinoic acid signalling from the epicardium to cardiomyocytes is important for the formation of the compact myocardium layer. The epicardium expresses the retinaldehyde dehydrogenase (RALDH) enzymes required for the synthesis of retinoic acid (encoded by the Aldh1a1 and Aldh1a2 genes) and retinoic receptors. Deletion of the gene encoding the retinoic acid receptor RXRα disrupts compact myocardium formation (Sucov et al., 1994). Genes of the retinoic acid signalling pathway were downregulated in Kat6a^−/− hearts (Aldh1a1, FDR=9×10⁻⁶; Fig. 7K) and Ing4^+/−Ing5^−/− hearts (Aldh1a2, FDR=0.02; Fig. 7L). The Aldh1a2 and Itga4 genes were among the top 20 genes that were downregulated in common in the comparisons Ing4^+/−Ing5^−/− versus Ing4^+/+Ing5^+/+ E10.5 hearts and E8.75 Ing4^−/−Ing5^−/− versus Ing4^+/+Ing5^+/+ whole embryos (FDR=4×10⁻⁵ and 0.0008, respectively; Fig. 7M).

Our RNA-sequencing data suggested that cell adhesion might be affected by the combined loss of one allele of Ing4 and both alleles of Ing5. Cell adhesion is also affected by the loss of KAT7 (Kueh et al., 2020).

We assessed cell adhesion using fibronectin as a substrate that can be bound by α4-integrin (Elices et al., 1990) and gelatine as an alternative substrate. Ing4^+/flIng5^fl/fl CreERT2 MEFs adhered with high efficiency to the fibronectin substrate, similar to Ing4^+/+Ing5^fl/fl CreERT2 and Ing4^+/+Ing5^+/+ CreERT2 MEFs (Fig. S12A). Cell spreading over a 24 h period was moderately, but significantly, reduced in Ing4^+/flIng5^fl/fl CreERT2 MEFs compared with control MEFs on fibronectin (P<10⁻⁶ overall; Fig. S12B). Ing4^+/flIng5^fl/fl CreERT2 MEFs performed considerably slower in the scratch wound-healing assay on fibronectin compared with Ing4^+/+Ing5^fl/fl CreERT2 and Ing4^+/+Ing5^+/+ CreERT2 MEFs (Fig. 8A-C). The differences in MEF cell spreading and wound healing were either greatly reduced or not observed at all on gelatine (Fig. S12).

To assay proepicardium function directly, the proepicardium was dissected from E9.5 embryos and subjected to short term culture as described by Garriock et al. (2014). Proepicardium explant sizes did not differ between genotypes at the time of plating (Fig. 8D). On fibronectin substrate, Ing4^+/−Ing5^−/− proepicardium explants formed smaller outgrowths compared with Ing4^+/−Ing5^+/+ explants [P=0.04 (24 h) and 0.03 (48 h); Fig. 8E-G]. No differences in proepicardium outgrowth were observed on gelatine (Fig. S12).

Overall, our data are consistent with a deficiency in cell spreading during the formation of the epicardial cell lineage in Ing4^+/−Ing5^−/− hearts and a failure of epicardial cells to signal to the cardiomyocytes to promote myocardium compact layer formation.

In this study, we have elucidated the requirements for ING5 and the functional overlap of ING4 and ING5 in mammalian development. We found ING5 to be essential for heart development. The Ing5^−/− mutant heart defect was restricted to a ventricular septum defect, resembling the situation in heterozygous Kat6a^+/− mice (Vanyai et al., 2015; Voss et al., 2012) and in humans with heterozygous KAT6A mutation (Kennedy et al., 2019) (Table S5). The phenotype of the compound Ing4^−/−Ing5^−/− null embryos resembled the Kat7^−/− phenotype (failure of chorioallantoic fusion; arrest at E8.5) (Kueh et al., 2011). Combined loss of ING4 and ING5 was also associated with the same effects on histone acetylation as loss of KAT7, loss of H3K14ac (Kueh et al., 2011), congruent with biochemical studies that place ING4 and ING5 in the KAT7 histone acetylation complex (Doyon et al., 2006). ING4 and ING5 can also form part of the KAT6A or KAT6B histone acetylation complexes (Doyon et al., 2006; Feng et al., 2016), but Ing4^−/−Ing5^−/− null embryos did not survive to the developmental stages when anomalies caused by loss of KAT6A or KAT6B could have been observed, including brain defects, homeotic transformation, and craniofacial and aortic arch defects (Thomas et al., 2000; Voss et al., 2009, 2012). However, these anomalies were also not readily visible in Ing4^+/−Ing5^−/− foetuses that retained one normal copy of the Ing4 gene. Heart development was more severely compromised in Ing4^+/−Ing5^−/− foetuses than in Kat6a^−/− mice (Voss et al., 2012). This is likely due to an impairment in function of the KAT6B and KAT7 complexes in addition to the KAT6A complex. As a consequence, apart from the similarities between Ing4^−/−Ing5^−/− null embryos and Kat7^−/− (Hbo1^−/−), as well as the Ing5^−/− and the Kat6a^+/− foetuses, a simple 1:1 correlation between the functions of ING4 and ING5 and the roles of KAT6A and KAT6B in prenatal development could not be established. Nevertheless, two phenomena were observed in Ing4^−/−Ing5^−/− cells that resemble the effects of loss of KAT6A. Like Kat6a^−/− fibroblasts (Baell et al., 2018; Sheikh et al., 2015) and unlike Kat7^−/− fibroblasts (Kueh et al., 2011), Ing4^−/−Ing5^−/− fibroblasts underwent growth arrest in culture. Furthermore, Ing4^−/−Ing5^−/− null fibroblasts displayed a reduction in H3K23ac, which suggests loss of KAT6A and KAT6B function (Klein et al., 2019; Lv et al., 2017; Simó-Riudalbas et al., 2015), in addition to the loss of H3K14ac, which suggests loss of KAT7 function (Feng et al., 2016; Kueh et al., 2023., 2011, 2020; MacPherson et al., 2020; Matsunuma et al., 2016; Niida et al., 2017; Zou et al., 2013). The embryonic and fibroblast phenotypes, together with the effects on histone acetylation, suggest that the functions of KAT6A, KAT6B and KAT7 are lost in Ing4^−/−Ing5^−/− double-knockout cells (Table S5).

Appropriate histone acetylation levels appear to be crucial for heart development. The enzymatic activity of histone acetyltransferases is opposed by histone deacetylases (HDACs). Interestingly, loss of Hdac1 and Hdac2 causes a thickened ventricular septum (Montgomery et al., 2007) (and presumably a gain of histone acetylation), whereas the loss of KAT6A (Voss et al., 2012) or the loss of ING4 and ING5 causes ventricular septal defects and a reduction in histone acetylation, suggesting that HDAC1 and HDAC2 might oppose the functions of the KAT6A/ING4/ING5 complex in cardiac septum development.

Proepicardium cells at the septum transversum migrate to the heart to form the epicardium (Virágh and Challice, 1981). Migration and expansion of epicardial cells on the surface of the heart depends on epicardial cell division and epicardial-myocardial cell–cell adhesion via α4-integrin (ITGA4) expressed in the epicardial cells and VCAM1 expressed on the cardiomyocytes (Kwee et al., 1995; Sengbusch et al., 2002; Yang et al., 1995). Possible causes of the observed anomalies in the Ing4^+/−Ing5^−/− embryos could be defects occurring earlier in the epicardial lineage, such as in multipotent extra-embryonic cardiac precursors (Zhang et al., 2021), the proepicardium, the juxta-cardiac field (Tyser et al., 2021) or impaired EMT. Proepicardium marker gene expression was affected in the Ing4^+/−Ing5^−/− hearts, including Tbx18, which is not expressed in the juxta-cardiac field (Tyser et al., 2021), but EMT gene expression was not affected, suggesting that, of these possibilities, the proepicardium is a likely candidate. Indeed, Ing4^+/−Ing5^−/− proepicardium explants displayed outgrowth defects. Our data suggest that the α4-integrin–VCAM1 interaction might be compromised in Ing4^+/−Ing5^−/− embryos. Itga4 was expressed at significantly lower levels in Ing4^+/−Ing5^−/− hearts. Ing4^+/−Ing5^−/− MEF spreading and proepicardium outgrowth defects were pronounced on fibronectin as a substrate, which can be bound by α4-integrin (Elices et al., 1990), but not on the control substrate. Congruent with an epicardium defect, we also observed a thin compact layer of the myocardium in Ing4^+/−Ing5^−/− hearts. Interestingly, Itga4^−/− and Vcam1^−/− embryos share anomalies with the Ing4^−/−Ing5^−/− double-knockout embryos, which, indeed, express lower levels of Itga4 mRNA. Like all Ing4^−/−Ing5^−/− double-knockout embryos, 50% of the Vcam1^−/− embryos display a failure of chorioallantoic fusion causing early developmental arrest (Kwee et al., 1995). The surviving 50% Vcam1^−/− embryos show a paucity of epicardium cells, a failure to form a compact myocardium layer and ventricular septal defects (Kwee et al., 1995), like the Ing4^+/−Ing5^−/− embryos. Likewise, 50% of Itga4^−/− embryos show a chorioallantois fusion defect with early developmental arrest, and the remaining 50% a paucity of epicardium cells (Yang et al., 1995). In addition, Itga4^−/− display defects in coronary vessel network development (Yang et al., 1995), like Ing4^+/−Ing5^−/− embryos.

Retinoic acid production by ALDH1A2 (also known as retinaldehyde dehydrogenase, RALDH2) in the epicardium and retinoic acid signalling via the retinoic acid receptor RXRα are essential for myocardium development. Like Ing4^+/−Ing5^−/− embryos, germline and epicardium-specific Rxra mutant mice fail to form a compact layer of the myocardium (Chen et al., 2002; Merki et al., 2005; Sucov et al., 1994) and display ventricular septal defects (Sucov et al., 1994). Our RNA-sequencing data revealed a reduction in Aldh1a2 mRNA in the Ing4^+/−Ing5^−/− hearts, as has been observed in Wt1^−/− epicardial cells (Guadix et al., 2011).

Combined with other factors, insufficient retinoic acid signalling and defective α4-integrin-VCAM1-mediated cell–cell adhesion might contribute to the reduction in epicardial and subepicardial cell formation in Ing4^+/−Ing5^−/− hearts, with subsequent defects in coronary vessel network formation and compact myocardium development. In the future, it would be interesting to examine the roles of ING4 and ING5 in the myocardium and the epicardium in more detail using tissue-specific cre-recombinase transgenes.

Taken together, the morphological, cellular and biochemical consequences of loss of ING4 and ING5 combined or loss of one allele of Ing4 and both alleles of Ing5 suggest that the functions of ING4 and ING5 completely overlap with respect to their role in the KAT7 histone acetyltransferase complex during embryogenesis and in the prevention of cell growth arrest by the KAT6A histone acetyltransferase complex in MEFs. In addition, ING5 performs a non-redundant role in ventricular septum formation. Our data suggest that dominant-negative mutations in the ING5 gene ought to be considered as potential causal mutations if associated with ventricular septal defects in humans.

All procedures involving animals were conducted according to the Australian Code for the Care and Use of Animals for Scientific Purposes and with approval from the Walter and Eliza Hall Institute Animal Ethics Committee.

Ing4 and Ing5 conditional mutant mice were generated by OZgene Pty Ltd, Bently, Western Australia, using Bruce 4 embryonic stem cells, which are C57BL/6 derived. The neomycin phosphotransferase cassette was removed by crossing the mice to FLPe-recombinase transgenic mice (Farley et al., 2000). Exons 3-6 of the Ing4 gene were flanked with loxP sites (Ing4^fl) and exons 3-5 of the Ing5 gene were flanked with loxP sites (Ing5^fl; Fig. S1A). Removal of Ing4 exons 3-6 and Ing5 exons 3-5 was achieved by crossing the Ing4^fl and Ing5^fl mice to a Cre-deleter mouse strain (Schwenk et al., 1995), which deletes in the germline and in both genes produced a frame shift and a premature stop codon, generating the Ing4^– and Ing5^– alleles used in this study. The region of deletion (Fig. S1A) encodes part of the N-terminal ING domain and the nuclear localisation signal in the Ing5 locus and, in addition, part of the PHD finger in the Ing4 locus (Fig. S1F). Any protein product of splicing around the deleted exons would be out of frame. Mice carrying the Ing4^– or the Ing5^– allele were backcrossed to wild-type C57BL/6J mice for more than eight generations. Experiments were conducted in mice on the C57BL/6 genetic background, except proepicardium explants, which were performed on mice that had been backcrossed for eight generations to an FVBxBALB/c hybrid background. Mice were routinely genotyped by three-way PCR using a common oligonucleotide (Ing4: 5′-ATTTCCCTCGAGGTTTGGTT-3′; Ing5: 5′-TGCTGGGACTGTTTACAAATTAGA-3′) that together with oligonucleotide 1 (Ing4: 5′-ATTGCCTTTGTCACCAGGTC-3′; Ing5: 5′-AAAGGAGTGAACAATACAGCATGA-3′) detected the wild type (Ing4: 352 bp; Ing5: 322 bp products) or with oligonucleotide 2 (Ing4: 5′-CCTGGGCAAGACTCAAAGAG-3′; Ing5: 5′-ATGTACCGAATGTGGGAACTAAAT-3′) detected the null allele (Ing4: 510 bp; Ing5: 528 bp products).

Ing4^+/− and Ing5^+/− mice were crossed to generate compound heterozygous Ing4^+/−Ing5^+/− mice. Compound Ing4;Ing5 mutant embryos of heterozygous and homozygous mutant genotypes indicated in the figures and tables were recovered from Ing4^+/−Ing5^+/−×Ing4^+/−Ing5^+/− matings and, on occasion, from Ing4^+/−Ing5^+/− (mother)×Ing4^−/−Ing5^+/− (father) matings.

For BrdU labelling, pregnant dams were injected intraperitoneally with 100 µg/g bodyweight BrdU (APC BrdU Flow Kit, BD Biosciences, 552598) 1 h before the embryos were recovered.

Ing4 and Ing5 single and double conditional mutant mice (Ing4^fl/+, Ing5^fl/+, Ing4^fl/+Ing5^fl/+) were crossed to Rosa26-CreERT2 mice (Seibler et al., 2003) and used to isolate Ing4^+/+Ing5^+/+ CreERT2, Ing4^+/flIng5^+/+ CreERT2, Ing4^fl/flIng5^+/+ CreERT2, Ing4^+/+Ing5^fl/fl CreERT2, Ing4^fl/flIng5^+/fl CreERT2, Ing4^+/flIng5^fl/fl CreERT2 and Ing4^fl/flIng5^fl/fl CreERT2 primary MEFs.

We reported our Kat6a^+/− mutant mice previously (Moz^+/−; Voss et al., 2009). These mice carry a deletion of exons 4-7 of the Kat6a gene. Kat6a^+/− mice were kept on a C57BL/6 genetic background.

Ing4 and Ing5 compound mutant embryos and controls were dissected and examined under a dissection microscope (Zeiss Stemi 2000-C). Embryos with an observable heartbeat only were used for experiments and are displayed. Embryos were developmentally staged as described (Kaufman, 1995). Embryos were stage-matched in some experiments to avoid variation owing to differences in developmental stages, i.e. E10.5 Ing4^+/−Ing5^−/−, Ing4^+/+Ing5^−/−, Ing4^+/−Ing5^+/+, Ing4^+/+Ing5^+/+ embryos for heart dissection for RNA sequencing (Fig. S8A) and E8.75 Ing4^−/−Ing5^−/− and E8.25 Ing4^+/+Ing5^+/+ embryos for RNA sequencing. As an example of the assessment of developmental stages, E10.5 (Kaufman, 1995) is described here. The forelimb was examined for the presence of a prominent vein subjacent to a well-defined apical ectodermal ridge. The rostral and caudal extents of the apical ectodermal ridges were yet to become well defined. The marked eversion of the frontal nasal processes and prominent olfactory pits were noted. The maxillary and mandibular portion of the first pharyngeal arch were discernible. The extremely thin roof and dilation of the fourth ventricle characteristic for this stage was seen in the controls, but in Ing4 and Ing5 compound mutant embryos the fourth ventricle was occasionally collapsed. The otocysts were still spherical, but already had a small posteromedial endolymphatic appendage characteristic of this stage.

qRT-PCR was performed using pairs of oligonucleotides amplifying cDNA 3′ of the deleted exons (Ing4: forward 5′-AGTATGGGATGCCCTCAGTG-3′ and reverse 5′-GACCTGGTGACAAAGGCAAT-3′; orIng4: forward 5′-CGTTTTGAGGCTGATCTGAA-3′ and reverse 5′-GGGGGCTTCTTCATCTGAGT-3′; Ing5: forward 5′-CCAGAAGCCTGAGTGTCTCC-3′ and reverse 5′-TGCCAGTCTGTTGATGAAGC-3′; orIng5: forward 5′-GCCATGCAGACCTACGAGAT-3′ and reverse 5′-TTCCATCCATCCTGTCCTTC-3′) on RNA isolated from E10.5 embryos using QIAGEN RNeasy Mini Kit (QIAGEN, 217004), followed by cDNA synthesis using Super Script III Reverse Transcriptase (Invitrogen, 18080085) and qPCR amplification using SYBR Hi-ROX (Bioline, QT605-05). Values for qRT-PCR are displayed relative to the mRNA levels of the housekeeping gene Pgk1 or compared with the housekeeping gene Hsp90ab1.

H&E-stained serial paraffin sections of embryos and hearts were generated using standard histology techniques. Images were taken with a compound microscope and a digital camera (Axioplan 2 and Axiocam HR, Zeiss).

Wholemount in situ hybridisation was carried out using standard techniques (Thomas et al., 2007). The antisense probes used were NM_133345 bases 102-1281 for Ing4 and NM_025454 bases 539-1731 for Ing5.

Skeletal preparations were performed using standard techniques (Thomas et al., 2000).

MEFs from E9.5 Ing4 and Ing5 germline null mutant embryos or E12.5 Ing4 and Ing5 conditional mutant embryos were isolated and cultured using standard techniques (Voss et al., 2003). To induce gene deletion, Ing4^+/+Ing5^+/+ CreERT2, Ing4^fl/flIng5^+/+ CreERT2, Ing4^+/+Ing5^fl/fl CreERT2, Ing4^+/flIng5^fl/fl CreERT2, Ing4^fl/flIng5^+/fl CreERT2 and Ing4^fl/flIng5^fl/fl CreERT2 MEFs were treated with 500 nM 4-OH-tamoxifen (H7904, Sigma-Aldrich) for 3 days in culture. Successful gene deletion was confirmed at this point by PCR before further experiments were conducted. For the assessment of cell growth, 60,000 cells per 3.5 cm well in 6-well plates were plated and, at each passage (every 3 days) counted using an automated cell counter (Countess, C10227, Invitrogen) and re-plated at 60,000 cells per well. For the analysis of cell adhesion, cell spreading and scratch wound closure, 40,000 cells were plated into 6-well plates coated with fibronectin or gelatine and analysed in an automated live-cell analysis instrument (IncuCyte SX5; Sartorius).

Proepicardium was dissected from E9.5 embryos and explant cultures were established as described (Garriock et al., 2014) with modifications. In brief, explants were plated in 24-well plates coated with fibronectin or gelatine and cultured in DMEM supplemented with 1% (v/v) foetal bovine serum, L-glutamine, penicillin and streptomycin. Explants were visually examined and photographed at plating (0 h), 24 h and 48 h of culture using an inverted imaging microscope (Diaphot 300; Nikon) and a digital camera (AxioCam ICm1, Zeiss). The extent of monolayer outgrowth of cells with a cobblestone morphology was assessed using imaging analysis software (ImageJ version 2.9.0/1.54f).

MEFs were seeded at 15,000 cells/cm² in 0.1% gelatine-coated chamber slides (Nunc, C7182) and, after overnight culture, were processed through fixative [4% (v/v) paraformaldehyde in PBS] for 10 min at room temperature (RT), followed by two 5 min washes in PBS at RT, permeabilisation [0.2% (v/v) Triton X-100 in PBS] for 15 min at RT and blocking solution for 15 min at RT, primary antibody for 1 h at RT or overnight at 4°C, three 5 min washes in PBS at RT, secondary antibody for 1 h at RT or overnight at 4°C, washes in PBS, nuclear counterstain (0.1 μg/ml DAPI; Invitrogen, D1306) for 10 min at RT, washes in PBS and coverslipping. Antibodies and blocking solution are listed in Table S6. ‘No primary antibody’ control stains were performed on wild-type samples. Cells were imaged using a compound fluorescence microscope and a digital camera (Axioplan 2 and Axiocam HR, Zeiss) and were analysed using ImageJ (version 2.0.0-rc-69/1.52u).

Adherent cells were washed with PBS, nuclei were extracted on the plate (5 min on ice) in nuclear isolation wash buffer [15 mM Tris-HCl, 60 mM KCl, 15 mM NaCl, 5 mM MgCl₂, 1 mM CaCl₂ and 250 mM sucrose, pH 7.5, HCl-adjusted, with protease inhibitors (PIC, Roche, 11697498001), 1 mM PMSF (Roche, 10837091001), 1 mM DTT (Supelco, 646563) and 5 mM sodium butyrate (Sigma-Aldrich, 303410)] supplemented with 0.2% (v/v) NP-40 Alternative (Millipore, 492016), scraped from the plate, collected, centrifuged (5 min, 2500 g, 4°C), washed in nuclear isolation wash buffer (without NP-40 Alternative), lysed on ice in 0.4 N H₂SO₄ (Sigma-Aldrich, 339741) and centrifuged (10 min,10,000 g). Nuclear protein lysate supernatants were transferred to new tubes with 0.5 volumes of 100% trichloroacetic acid (TCA; Sigma-Aldrich, T0699) and precipitated for 2 h or overnight on ice. Precipitated acid extracted proteins were collected by centrifugation (10 min, 10,000 g), washed in 0.1% (v/v) HCl in acetone, centrifuged (10 min, 17,300 g), washed in 100% acetone, air-dried (5-15 min), resuspended in 100 µl sterile MQ-H₂O, centrifuged (2 min, 10,000 g), aliquoted and snap-frozen at −80°C. Protein concentrations were quantified using a Bradford assay according to the manufacturer's instructions (Bio-Rad, 5000204).

Acid-extracted protein samples were diluted in 2× SDS loading buffer heated (95°C for 5 min), separated on a 4-12% Bis-Tris agarose gel, transferred to PVDF membrane (Immobilon FL, IPFL00010, pore size 0.45 µm), blocked (1 h at 4°C) in Odyssey Blocking Buffer (Li-Cor), incubated in primary antibody diluted in Odyssey Blocking Buffer with 0.1% Tween-20 (1.5 h at room temperature or overnight at 4°C). Antibodies used are listed in Table S7. Membranes were rinsed in 0.1% PBS with 0.1% Tween 20 (PBST), washed in PBST, incubated with secondary antibody for 1 h at room temperature in Odyssey Blocking Buffer with 0.1% Tween 20, washed in PBST, rinsed in PBS, and then assessed using the Odyssey CLx Imaging System (Li-Cor) and quantified using Image Studio Lite.

Embryonic hearts were dissected and arrested in diastole by treatment with 0.6 mM KCl, then immediately immersed in 2% paraformaldehyde for 2 h at 4°C with gentle agitation. Fixed hearts were washed twice, 10 min each wash, with PBS at room temperature and then aligned in specimen moulds (Tissue Tek, 4557) filled with 1% low melting point agarose. Agarose-embedded samples were submerged in PBS at room temperature for 10 min and transferred to 70% (v/v) ethanol in H₂O before being embedded in paraffin and sectioned at 4 μm. This method was used for the detection of histone H3 phosphorylated on serine 10 and for TUNEL assays.

For the detection of all other proteins, dissected and diastole-arrested embryonic hearts were washed once with PBS, and submerged in 30% (w/v) sucrose solution, followed by two 5-min washes with PBS at room temperature. Samples were aligned in specimen moulds (Tissue Tek, 4557) filled with O. C. T. compound (Sakura, 4583). Samples in moulds were snap-frozen on dry ice and sectioned on a cryostat (HM525, HM550, Microm) at 6 μm thickness.

Marker proteins of mitosis or cell types in the developing heart were detected by immunofluorescence on serial paraffin (deparaffinised) or frozen sections on gelatine-coated Superfrost glass slides (Fisher Scientific). Sections were permeabilised using 0.2% (v/v) Triton X-100, blocked in 0.25% (w/v) gelatine in PBS, incubated with primary antibodies diluted in blocking solution, washed in PBS, incubated with secondary antibodies in blocking solution, rinsed, incubated in 0.1 μg/ml DAPI (Invitrogen, D1306) in PBS, washed and mounted in Dako fluorescence mounting medium (Agilent, S3023). Sections were imaged using either a compound fluorescence microscope (Axioplan 2, Zeiss), a digital camera (AxioCam, Zeiss) and AxioVision software, version 3.1.2.1, Zeiss), or a laser-scanning confocal microscope (Zeiss 780 LSM or Zeiss 880 with AiryScan) and Zen Software (Zeiss, 2012 Black edition). Images were analysed using ImageJ (version 2.0.0-rc-69/1.52u).

TUNEL assays were performed using a TUNEL kit (Roche, 12156792910) following the manufacturer's instructions.

Wholemount immunohistochemistry of hearts was performed using standard wholemount immunohistochemistry techniques (Voss et al., 2003).

Morphometric analyses and cell counting were performed on images taken with a compound fluorescence microscope (Axioplan 2, Zeiss; AxioCam, Zeiss; and AxioVision software, version 3.1.2.1, Zeiss). Measurements were conducted in ImageJ (version 2.0.0-rc-69/1.52u). Data were analysed using statistical analysis software (Prism 9.5.0, GraphPad Software).

The latero-lateral and apical-basal diameters of the heart, as well as the thickness of the compact myocardium and the trabecular myocardium were measured on H&E-stained paraffin sections. The mean of 11-18 sections per animal was calculated.

The number of epicardial cells overlying the subepicardial space in the atrioventricular junctions was counted on H&E-stained paraffin sections. The number of subepicardial cells within the subepicardial space at the atrioventricular junctions was counted. The area of the region was measured. The area of blood vessels was measured and excluded. The number of subepicardial cells per epicardial cell and per area excluding blood vessels was calculated. The mean of three to nine sections per animal was calculated. The number of subepicardial cells per epicardial cell and per area were calculated and presented in consideration of the fact that a perfectly proportional dwarf mouse would have fewer subepicardial cells, but not fewer subepicardial cells per epicardial cell or per area. In contrast, a mouse with a defect in the production of subepicardial cells would have a paucity per area and/or per epicardial cell. The analysis of subepicardial cells was conducted at E12.5, which is on the cusp of EPDC emergence in the subepicardial space. A marker for EPDCs, including intramyocardial fibroblasts, is PDGFRα. Although we saw occasional PDGFRα⁺ cells in the myocardium, there were too few (zero to five per heart) to obtain data that could be assessed statistically. Therefore, we did not score intramyocardial EPDCs.

The number of WT1-positive cells, cells in mitosis (phospho-histone H3 serine 10 positive) or cells displaying cell death with DNA fragmentation (TUNEL positive) were counted on immuno- or TUNEL-stained sections at four or five histological levels 24 µm apart counterstained with DAPI. The total numbers of positively labelled cells were enumerated, and the linear length of the surface of the heart (WT1) or the areas of heart tissue labelled with DAPI (phospho-histone H3 and TUNEL) were measured. The number of positive cells were expressed as cells per mm ventricular wall (WT1) or DAPI-positive area (phospho-histone H3 and TUNEL). The mean of three or four sections per animal (WT1), 14-25 sections per animal (phospho-histone H3) or 25-35 sections per animal (TUNEL) was calculated.

The area of the dorsal surface of the heart covered by a network of PECAM1-positive cells and the total dorsal surface of the heart were measured on images of hearts stained by wholemount immunohistochemistry for PECAM1. Three to four hearts per genotype were analysed.

Embryos were recovered from dams injected intraperitoneally with 100 μg/g body weight BrdU 1 h earlier.

Embryonic hearts were immediately dissected, and single cells were generated as described previously (Chong et al., 2011). Briefly, hearts were excised from mouse embryos in warmed PBS and incubated with collagenase II (Worthington Biochemicals, LS0004174). Collagenase II was inhibited using 7% (v/v) foetal calf serum in ice-cold PBS. Tissues were dissociated mechanically and passed through a 40 μm cell strainer (Corning, 352340). Cells were stained using a BrdU staining kit (BD Biosciences, 552598), as instructed by the manufacturer, and analysed using an LSRIIC flow cytometry analyser (BD Biosciences). Data were analysed using FlowJo v10.4 (Tree Star).

Embryos were dissected and stored in DNA/RNA Shield (Zymo Research) until genotyping was completed. Total RNA was isolated from E8.75 and E8.25 embryos or E10.5 hearts using an RNA isolation kit (QIAGEN RNeasy Micro Kit; QIAGEN, 74004) according to the manufacturer's instructions. RNA-sequencing libraries from 100 ng of RNA per sample were generated using a kit (TruSeq Stranded RNA library Prep Kit; Illumina, 20020594) and indices (Illumina, 20020492) following the manufacturer's instructions. Libraries were sequenced using a NextSeq 150 high output (v2) kit (Illumina, FC-131-2001) to generate 66 bp (embryos) or 75 bp (hearts) paired end reads. Two independent datasets were generated – one for the E8.25/8.75 samples and another for the E10.5 samples.

For both datasets, FASTQ files were generated and de-multiplexed for the samples using the bcl2fastq conversion software (v2.15.0.4). Reads were aligned to the mouse reference genome mm10 (E10.5 samples) or mm39 (E8.25/8.75 samples) using Rsubread v1.28.1 (Liao et al., 2013; Liao et al., 2019). In all cases, over 97% of reads mapped to the reference genome for each sample. Successfully mapped reads were summarised into gene-level counts using Rsubread's featureCounts function. Genes were identified using NCBI RefSeq annotation.

In both datasets genes on the Y chromosome as well as the Xist gene were excluded to remove gender bias. For the E10.5 samples, non-protein-coding immunoglobulin genes, ribosomal RNA, and genes with obsolete Entrez IDs were also removed to reduce noise. In the E8.25/8.75 samples, all non-protein-coding genes, Riken and haemoglobin genes were removed together with genes with obsolete Entrez IDs. Furthermore, lowly expressed genes were filtered out using the filterByExpr function with default settings in edgeR v3.20.9 (Robinson et al., 2010), leaving 15,575 (E10.5) and 14,755 (E8.25/8.75) genes for downstream analysis. Compositional differences between the libraries were normalised using the weighted trimmed mean of M-values method (Robinson and Oshlack, 2010).

In both cases, differential expression analyses were performed using edgeR and limma v3.34.9 software packages (Ritchie et al., 2015). For the E10.5 samples, counts were transformed to log2 counts per million (CPM) and differential expression between genotypes was assessed using linear models and robust empirical Bayes moderated t-statistics with a trended prior variance. To increase precision, the linear models incorporated sample quality weights, a batch correction for litter effect and two surrogate variables (Phipson et al., 2016). For the analysis of the E8.25/8.75 samples, counts were first transformed to log2 CPM with associated precision weights using voom (Law et al., 2014). Differential expression between genotypes was then assessed using linear models and robust empirical Bayes moderated t-statistics.

For both analyses, P-values were adjusted to control the FDR at below 5% using the Benjamini and Hochberg method. Pathway analyses were carried out on differentially expressed genes to test for over-representation of biological pathways as defined by GO terms (Ashburner et al., 2000) and KEGG (Kanehisa et al., 2017, 2016; Kanehisa and Goto, 2000) pathways using the goana and kegga functions from limma. Rotation gene set tests were performed and barcode plots generated using limma's roast and barcodeplot functions, respectively, to compare differentially expressed genes identified from different comparisons (Wu et al., 2010).

The methods for statistical analyses, tests, and sample sizes are described in the figure legends. The RNA-sequencing data were analysed as described in the ‘RNA-sequencing data analysis’ section. Other statistical analyses were performed using GraphPad Prism 8, except Fisher's exact tests and two-way ANOVA, which were performed on Stata/SE 16.1, and the assessment of Ing4;Ing5 genotype distributions, which were analysed computing the cumulative binomial probability of being less than or equal to the expected value (pbinom) using R 4.0.0 GUI 1.71 Catalina build (7827). No samples were excluded. Randomisation was achieved by examining embryos and foetuses in order of recovery. The embryos and foetuses were examined externally at recovery before the genotype was known, therefore the operator was unaware of the genotype at the time of analysis. Analyses of other experimental parameters was automated (e.g. RNA sequencing, ImageJ, flow cytometry, Odyssey CLx Imaging System) ensuring unbiased assessments.

We thank Faye Dabrowski, Leanne Johnson, Heather Miller and Jaclyn Gilbert from WEHI Bioservices for expert animal care; Niall Geoghegan and Lachlan Whitehead from the WEHI Centre for Dynamic Imaging for help with the automation of image analysis; and Ellen Tsui, Emma Pan, Alex Johnston and Yuyin Hoang from the WEHI Histology Facility for processing of histological samples.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.202617.reviewer-comments.pdf