The ECM as a driver of heart development and repair

Heart development is a highly complex process requiring tight spatiotemporal coordination of multiple chemical and physical signals to shape heart morphology and support cardiac function. Vertebrate heart morphogenesis is initiated during somitogenesis, when bilateral cell populations coalesce at the midline to form a linear tube comprising two tissue layers: an outer contractile myocardium that surrounds a specialised endothelium – the endocardium. This simple tube undergoes dramatic remodelling during development, with spatially restricted changes in morphology supporting functional regionalisation of the heart, including chamber emergence, trabeculation and valve formation (Sedmera, 2011; Staudt and Stainier, 2012). Throughout this remodelling, the myocardium and endocardium are separated by a layer of extracellular matrix (ECM, also known as the cardiac jelly), a highly specialised network of proteins and sugars that acts as a signalling centre for surrounding cells and plays multiple roles in development (Walma and Yamada, 2020).

The extracellular environment can encompass two types of ECM: (1) the basement membrane, which is a thin, densely packed network of core proteins, including laminin, collagen IV, nidogen, perlecan (Hspg2) and agrin; and (2) the interstitial matrix, which is a diverse assembly of elastic fibres, fibronectin, collagens and proteoglycans (Frantz et al., 2010). While spatiotemporal regulation of ECM production during development creates distinct ECM environments, modification and degradation also contribute to ECM diversity (Bonnans et al., 2014). ECM degradation is mediated by a suite of proteins, including matrix metalloproteinases (MMPs), adamalysins (ADAMs/ADAMTSs) and hyaluronidases, that target specific ECM components. This combination of synthesis, modification and degradation generates highly specialised ECM environments that mature during development to support organ formation. Indeed, recent findings have demonstrated the dynamic nature of the ECM during embryonic development (Keeley et al., 2020; Matsubayashi et al., 2020). Moreover, studies identifying changes in ECM composition during embryonic and neonatal heart development are emerging (Cui et al., 2019; DeLaughter et al., 2016; Liu et al., 2019), alongside evidence linking ECM composition to regenerative potential (Chen et al., 2016; DeLaughter et al., 2016; Jam et al., 2018; Notari et al., 2018).

In this Review, we discuss the diverse roles the ECM plays during cardiac development, from the formation of the linear heart tube through to heart maturation, focussing on how ECM modification and turnover are emerging as central themes in regulating distinct aspects of morphogenesis. We describe evidence for conserved ECM dynamics in promoting human heart development and provide an updated summary of congenital heart defects (CHDs) associated with mutations in ECM components, highlighting recent examples linking ECM modification or biosynthesis with specific cardiac malformations. We also review emerging evidence from comparative analyses of regenerative capacity in model organisms suggesting that ECM composition plays a key role in regulating the regenerative potential of the heart. Finally, we discuss how defining the roles of ECM dynamism during development and disease represents an important avenue of therapeutic potential and highlight some important considerations in the field moving forward.

The vertebrate heart tube is formed by the medial migration of bilateral groups of cardiac precursors (either as pools of cells or bilateral tubes) that subsequently fuse at the embryonic midline. In the case of vertebrates such as mice and humans, this fusion directly forms the heart tube, while in zebrafish, cardiac precursors initially form a disc, which asymmetrically extends to generate the heart tube (Fig. 1). The ECM protein fibronectin (Fn) is a conserved regulator of medial cardiomyocyte migration (George et al., 1997; Trinh and Stainier, 2004). Studies in zebrafish have revealed a variable defect in medial cardiomyocyte migration in fn1a/natter mutants, and whereas cardiomyocyte-derived Fn promotes medial migration through junctional organisation and epithelial maturation (Fig. 1A), comparative analyses of cardiomyocyte migration in cloche mutants (which lack endothelial cells) demonstrate that endothelial-derived Fn promotes the timeliness of medial cardiomyocyte migration (Trinh and Stainier, 2004). Medial migration is further disrupted in zebrafish lamc1/fn1a double knockdown mutants, which are deficient in both Lamc1-containing laminin trimers and Fn1a (Sakaguchi et al., 2006), suggesting that these ECM components interact to drive cardiomyocyte migration.

Alongside deposition, matrix degradation is required for heart tube formation. MMP2 (which can degrade Fn) has similar expression dynamics to Fn during chick heart tube formation. Inhibition of MMP2 between the streak and the three-somite stages results in partial cardia bifida (Linask et al., 2005), suggesting that Fn fragmentation by MMP2 is required for cardiomyocyte migration or that MMP2 degrades an alternative ECM component. In zebrafish, the ECM component hyaluronic acid (HA) is synthesised by Has2 (see Box 1) in bilateral cardiomyocyte populations, while the hyaluronidase Cemip2, which degrades HA, is more broadly expressed (Fig. 1A). Tissue-specific rescue experiments have demonstrated that myocardial-derived Cemip2 is required for the migration of both myocardial and endocardial precursors (Totong et al., 2011), suggesting that both Fn and HA turnover drives medial migration of cardiac precursors. How HA deposition and cleavage promotes cardiac fusion is unclear, although mechanistic studies of heart tube assembly provide clues. Cardiomyocytes in the heart disc display regionalised differences in migratory behaviours that are dependent upon restricted BMP activity (Smith et al., 2008; Veerkamp et al., 2013), which in turn is dependent on HA. Specifically, has2 is asymmetrically expressed on the left side of the cardiac disc (Fig. 1B), where it dampens BMP signalling, resulting in asymmetric BMP activity on the right side of the disc. This, in turn, drives the regionalised migration velocity necessary for asymmetric heart tube formation (Veerkamp et al., 2013). Together, this suggests that HA may regulate migration velocity in both contexts – medial migration and heart tube assembly.

Heart tube formation in Drosophila is also regulated by the deposition and modification of ECM components, although, here, the main players are laminins and collagens. To date, Drosophila is the only organism in which a role for laminins in early heart morphogenesis has been described, where it maintains the integrity of migrating cardioblasts (Yarnitzky and Volk, 1995). Similar to its role in chick, Mmp2 promotes cardioblast polarisation and migration (Hughes et al., 2020; Raza et al., 2017), and loss of Laminin A or Mmp2 results in abnormal collagen localisation and improper cardioblast fusion. Collagen directly promotes cardiac precursor migration in another invertebrate model – the ascidian Ciona intestinalis (Bernadskaya et al., 2019). During Ciona development, bilateral cardiopharyngeal progenitors migrate between the epidermis and endoderm before dividing into fate-restricted progenitors, a process dependent upon the type IX collagen component Col91a. Similar to the role of the ECM in regulating asymmetric BMP signalling during cardiomyocyte migration in zebrafish (Veerkamp et al., 2013), ECM-mediated BMP signalling asymmetry is also likely required for cardiopharyngeal precursor migration in Ciona, where BMP signalling is highest in leading cells (Bernadskaya et al., 2019).

After the heart tube has formed, it undergoes morphogenesis, including looping and chamber ballooning, initiating functional regionalisation of the heart (Fig. 2). While HA synthesised by Has2 plays a conserved role in heart looping in both mouse and zebrafish (Camenisch et al., 2000; Tong et al., 2014), regionalised HA degradation is also crucial in facilitating looping morphogenesis. Zebrafish cemip2 mutants exhibit increased cardiac ECM and reduced heart looping (Hernandez et al., 2019; Smith et al., 2011; Totong et al., 2011). Similarly, expansion of has2 into the cardiac chambers results in a thicker cardiac ECM and reduced looping morphogenesis (Grassini et al., 2018; Lagendijk et al., 2011), together demonstrating that regionalised HA deposition and turnover promotes morphogenesis (Fig. 2A). Interestingly, both Has2 mutant mice and cemip2 zebrafish mutants have altered expression or organisation of other ECM components such as laminin, type I collagen, versican (Vcan) and chondroitin sulfate (Camenisch et al., 2000; Hernandez et al., 2019; Totong et al., 2011), highlighting the complex interactions that support normal ECM organisation. Vcan plays a similarly conserved role in heart looping morphogenesis in mice, medaka and zebrafish (Mittal et al., 2019; Mjaatvedt et al., 1998). Loss of hyaluronan and proteoglycan link protein 1 (Hapln1), which crosslinks HA to proteoglycans such as Vcan (see Box 1), results in cardiac malformations in mouse that are consistent with defects in looping and chamber morphogenesis (Wirrig et al., 2007), demonstrating that HA-Vcan interactions are also necessary for heart morphogenesis. Early heart looping in mice further relies upon breakdown of the dorsal mesocardium, a Fn-rich tissue connection between the heart tube and the adjacent embryo body wall. This breakdown is mediated by MMPs – potentially MMP2 – in the adjacent foregut (Le Garrec et al., 2017; Linask et al., 2005) (Fig. 2A).

After heart tube looping, each chamber undergoes ballooning morphogenesis, during which time myocardial and endocardial interactions promote regional ECM degradation by ADAMTS family members to shape the atrium (Fig. 2B). Loss of myocardial angiopoietin 1 in mice results in downregulation of several ADAMTS family members, decreased cleavage of Vcan (see Box 2) and a thicker atrial ECM, eventually leading to small dysmorphic atria and abnormal septation (Kim et al., 2018) Conversely, endocardial deletion of the cerebral cavernous malformation (CCM) family member Krit1 results in upregulation of Adamts4 and Adamts5, in increased Vcan cleavage and in reduced cardiac ECM (Zhou et al., 2015), leading to thinner myocardial walls and distended chambers. It is likely that ECM turnover plays multiple roles in atrial ballooning, regulating proliferation, tissue organisation and cell stretching (Kim et al., 2018). This link between CCM genes, ECM composition and chamber morphology is conserved in zebrafish, where ccm2 mutants also exhibit reduced cardiac jelly, thin myocardial walls and dilated hearts (Zhou et al., 2015), which are partially rescued by knockdown of adamts5. Interestingly Adamts5 mutant mice are not reported to have defects in cardiac chamber morphology (Dupuis et al., 2011; Hemmeryckx et al., 2019), suggesting redundancy between ADAMTS family members. Furthermore, HA degradation by the hyaluronidase Hyal2 is required to regulate chamber size in mice by E18.5 and at later stages to support postnatal cardiac function (Chowdhury et al., 2017). Together, this supports a major requirement for ECM remodelling or degradation, rather than synthesis, in facilitating heart looping and ballooning.

A different suite of ECM components is associated with ventricular morphogenesis. Both Col14a1 and Col11a1 mouse mutants have a rounded ventricular apex at 3 months of age (Lincoln et al., 2006; Tao et al., 2012), followed by a gradual reduction in cardiac function. Complex interactions between collagens at both transcriptional and scaffold-assembly levels likely contribute to their role in ventricular morphogenesis, as loss of Col14a1 results in upregulated expression of other fibril-forming collagens and an increase in fibril formation (Tao et al., 2012). Vcan^tm1Zim mutants (which express only V1 and V3 isoforms, see Box 2) display a similarly dampened ventricular apex (Burns et al., 2014), together demonstrating that collagens and specific Vcan isoforms promote ventricular morphogenesis. In chick embryos, ventricular ECM composition is further regionalised, with the epicardium rich in Fn, laminin and collagen, while the deeper myocardium is devoid of collagen (Jallerat and Feinberg, 2020), suggesting that regionalised ECM organisation within the chamber myocardium supports further specialised morphogenesis.

While analyses of ECM requirements in cardiac development have focussed primarily on myocardial morphogenesis, defects in the cardiac ECM also impact endocardial morphogenesis. Loss of the fibrillin gene fbn2b in zebrafish results in failure of the endocardium to line the entire lumen of the atrium, suggesting that Fbn2b maintains endocardial integrity during chamber growth (Mellman et al., 2012). Overall, it is evident that each chamber has distinct pathways regulating dynamic ECM turnover and ECM patterning, which likely facilitate chamber-specific maturation.

Cardiac development requires the addition of two further cell populations to the heart tube during morphogenesis: the second heart field (SHF) and, subsequently, cardiac neural crest cells (CNCCs) (Fig. 3). The SHF contributes significantly to chamber formation as well as integrating the heart into the rest of the cardiovascular system though the outflow tract (OFT), which consists of a proximal myocardial component emerging from the ventricle and a distal smooth muscle region [termed the bulbus arteriosus (BA) in zebrafish] (Grimes et al., 2010) (Fig. 3). CNCCs contribute to diverse cardiac structures, including OFT cushions and pharyngeal arch arteries, and septate the OFT through formation of the aorticopulmonary septum (George et al., 2020). The integration of both these cell types into the heart is influenced by the cardiac ECM.

The ECM plays an important role in balancing levels of FGF signalling, which regulates the recruitment of both the SHF and CNCCs. Fn facilitates FGF signalling through binding Itga5, promoting SHF progenitor proliferation in the splanchnic mesoderm (Mittal et al., 2013), as well as CNCC maintenance and proliferation (Mittal et al., 2010) (Fig. 3A). The synthesis of heparan sulfate proteoglycan (HSPG; see Box 1) is similarly required for SHF and CNCC proliferation. Ndst1 and Ext1 knockout mice exhibit defective HSPG synthesis resulting in malformed SHF- and CNCC-derived structures via dysregulation of FGF-FGFR interactions and downstream ERK signalling (Pan et al., 2014; Zhang et al., 2015). OFT malformations arising from mesenchymal hyperplasia in a perlecan (Hspg2) mouse mutant suggest that perlecan may limit CNCC migration into the OFT (Fig. 3B) (Costell et al., 2002) and, similar to HSPG, perlecan has been implicated in promoting FGF signalling (Joseph et al., 1996).

The ECM also modulates TGFβ signalling to promote SHF and CNCC addition. Latent TGFβ-binding proteins (LTBPs) interact with the ECM to regulate bioavailability of secreted TGFβ molecules (Rifkin et al., 2018). TGFβ signalling is attenuated in the OFT of Ltbp1l mouse mutants, resulting in failure of OFT septation (Todorovic et al., 2007), suggesting a model whereby TGFβ is anchored in the ECM by Ltbp1L and is subsequently recognised by post-migratory CNCCs. Similarly, Ltbp3 mediates TGFβ signalling in zebrafish to drive SHF addition by promoting proliferation of the progenitor population (Zhou et al., 2011) (Fig. 3C). Together, this demonstrates a conserved requirement for Ltbp-mediated TGFβ signalling in vertebrate OFT development.

Vcan plays conserved roles in regulating SHF deployment into the heart (Kern et al., 2007; Mittal et al., 2019; Mjaatvedt et al., 1998; Yamamura et al., 1997) (Fig. 3A,C). Spatiotemporal cleavage of Vcan (see Box 2) occurs in a distal-proximal gradient along the OFT, overlapping with Adamts9 and fibulin 1 (Fbln1) expression (Cooley et al., 2008; Kern et al., 2007, 2010) (Fig. 3B). This cleavage precedes the loss of the distal-most myocardium as the OFT undergoes remodelling and is populated by CNCC-derived smooth muscle cells (Kern et al., 2007). These studies demonstrate a dynamic role for Vcan in promoting both SHF and CNCC addition in a timely fashion.

The expression of elastin (Eln) and aggrecan (Acan) is also conserved in the OFT (Duchemin et al., 2019; Mittal et al., 2019; Rambeau et al., 2017; Zanin et al., 1999; Zhou et al., 2011) (Fig. 3D) and, although no clear functional heart defects are present at birth in Eln mouse mutants, cardiac function rapidly declines and is ultimately lethal in these mice (Wagenseil et al., 2010). Acan appears to be required for cardiac output in zebrafish, albeit at only embryonic stages (Rambeau et al., 2017), and Acan localisation and cleavage play additional roles in development of the OFT and aortic wall. Loss of Adamts5 in mice results in a thickened aortic wall, disorganised elastic fibres and an altered balance of Acan cleavage products (Dupuis et al., 2019). Graded Adamts9 expression overlaps with Acan expression in the OFT, and Adamts9 heterozygous mutant mice develop aortic wall anomalies (Kern et al., 2010), together suggesting that degradation of Acan by multiple ADAMTS enzymes is necessary for development of the aortic wall (Dupuis et al., 2019).

As the heart gains structural complexity, valves form between chambers and at the OFT to prevent retrograde blood flow. Initiation of valve development is highly conserved, encompassing the localised swelling of cardiac jelly at the sites of valve formation (termed the endocardial cushions), which requires HA synthesis by Has2 (Beis et al., 2005; Camenisch et al., 2000, 2002; Inai et al., 2013) (Fig. 4A). In zebrafish, this localised HA deposition induces endocardial expression of the cell-adhesion molecule alcama in the atrioventricular canal (AVC) cushions (Beis et al., 2005; Lagendijk et al., 2011) (Fig. 4A′). In addition to a requirement for HA in valve initiation, Vcan promotes AVC endocardial cushion formation, as both mouse Vcan^hdf mutants and Medaka vcan mutants exhibit a loss of cushion identity (Mittal et al., 2019; Mjaatvedt et al., 1998; Yamamura et al., 1997) (Fig. 4A). Together with a proposed requirement for chondroitin sulfate synthesis in zebrafish endocardial cushion development (Peal et al., 2009) and for HA/Vcan expression in chick endocardial cushions (Inai et al., 2013), these findings suggest that HA and Vcan deposition is highly conserved at the onset of valvulogenesis.

Alongside regionalised deposition of proteoglycans, localised degradation of HA is required to restrict the AVC programme. Loss of cemip2 in zebrafish leads to an expansion of valve markers and ectopic Wnt activity outside the AVC (Hernandez et al., 2019; Smith et al., 2011; Totong et al., 2011), suggesting that localised degradation of HA in the chambers limits Wnt activity to restrict valvulogenesis (Fig. 4A). Simultaneously, localised breakdown of ColIV in the underlying ECM by MMPs facilitates cell migration during valve endothelial-to-mesenchymal transition (EndoMT) in mouse and quail (Enciso et al., 2003; Song et al., 2000; Tao et al., 2011), with expression analyses suggesting that MMP2 and MMP15 are key candidates mediating this degradation (Rupp et al., 2008; Tao et al., 2011) (Fig. 4A). EndoMT is also promoted by LTBP-mediated TGFβ signalling, as Ltbp1l mutant mice have fewer mesenchymal cells at E10.5-E11.5 (Todorovic, 2011) (Fig. 4A).

ECM-mediated physical separation between the two tissue layers of the heart at the AVC also facilitates myocardial patterning. During looping morphogenesis in the chick, endothelin 1 (Et1) ligand and its receptors are expressed throughout the endocardium and myocardium, respectively. The expanded AVC ECM prevents the Et1 ligand from reaching the myocardium, maintaining the immature state of the non-working myocardium and defining the regional differences in electrical impulses that are necessary for organised contractions (Bressan et al., 2014).

Valve development is intimately linked with the sensation of blood flow, mediated by kruppel-like factor 2 (Klf2) – a mechanosensitive transcription factor that is expressed in both AVC and OFT endocardial cushions, and that is dependent upon sensation of oscillatory flow (Heckel et al., 2015; Vermot et al., 2009) (Fig. 4A,B). Klf2 mutant mice fail to initiate EndoMT at the AVC and display a loss of glycosaminoglycan (GAG) deposition, as UDP-glucose 6-dehydrogenase (Ugdh), which encodes a GAG-synthesising enzyme (see Box 1), is a transcriptional target of Klf2 (Chiplunkar et al., 2013). A similar mechanism may function in zebrafish, as loss of klf2a expression in non-contractile hearts correlates with a loss of alcama expression (Beis et al., 2005; Vermot et al., 2009). Downstream of zebrafish klf2a, fibronectin 1b (Fn1b) promotes invasion of cells into the cardiac jelly during early EndoMT (Steed et al., 2016b), likely acting through Itga5 (Gunawan et al., 2019). Together, this highlights that flow sensing is crucial and conserved in sculpting regionalised ECM environments (Fig. 4A).

Subsequent to the initial formation of the endocardial cushions and EndoMT, ECM remodelling and stratification lead to the formation of mature leaflets – thin trilaminar structures composed of collagen, elastin and proteoglycans. In mice, early AVC cushion remodelling includes the regionalised cleavage of Vcan by ADAMTS5 from E10.5 onwards (Dupuis et al., 2011; Kern et al., 2006). This Vcan cleavage limits mesenchymal proliferation, promotes regionalised cell packing and facilitates TGFβ signalling (Dupuis et al., 2011) (Fig. 4B). TGFβ signalling in the remodelling mouse AV valve is required to attenuate EndoMT, as Ltbp1l mutant AV valves are hyperplastic at E14.5, displaying extended expression of EndoMT markers (Todorovic et al., 2011). Loss of the ADAMTS co-factor fibulin 1 leads to hypercellular proximal OFT cushions correlating with increased EndoMT capacity (Harikrishnan et al., 2015). Adamts1 and Adamts9 are expressed in a similar domain to Vcan, and loss of a single copy of Adamts9 leads to a reduction in versikine (the cleaved form of Vcan, see Box 2) and enlarged valves (Kern et al., 2006, 2010), further highlighting potential functional redundancy between ADAMTS members. Similarly, endocardial Klf2 restricts mesenchymal proliferation and valve leaflet size, and promotes cell packing (Goddard et al., 2017), highlighting ongoing roles for flow sensing in regulating ECM dynamics during valve development (Fig. 4B).

Spatiotemporal degradation of Vcan may be controlled through cross-linking; Hapln1 expression overlaps with Vcan in both atrioventricular and outflow tract valves, and loss of Hapln1 in mice leads to reduced levels of Vcan and a spectrum of defects impacting both the AV and OFT cushions (Wirrig et al., 2007). This suggests that Hapln1-mediated crosslinking may protect Vcan from ADAMTS-mediated cleavage. HA is also subject to proteolytic cleavage during valve maturation. Hyal2 is present in the endocardial cushions at both sites of valvulogenesis at E11.5-E12.5, and Hyal2 mutant mice have persistent HA in the valves, increased mesenchymal cell number and thickened disorganized valves at 4 weeks of age (Chowdhury et al., 2017) (Fig. 4B). Overall, this suggests that, although HA and Vcan deposition initiates valve development, timely degradation of both components is required to limit valve leaflet size.

Alongside requirements for Vcan and HA degradation in limiting valve size, HSPG attenuates mesenchymal cell proliferation during remodelling of the AVC and OFT valves in mice at E13.5, acting together with endocardial-derived heparin-binding EGF (HB-EGF) (Iwamoto et al., 2010). This suggests that Vcan is cleaved by ADAMTS5 and/or ADAMTS9, and may act through Ltbp1l/TGFβ, together with heparan sulfate (HS)-dependent HB-EGF signalling, to terminate mesenchymal proliferation and promote sub-endocardial compaction of the mesenchyme, and that this is at least partially regulated by the sensation of blood flow (Fig. 4B).

Collagen and elastin are major constituents of mature valves, and collagen content in the endocardial cushions changes during remodelling. At E12.5 in mice, AVC valves predominantly express fibrillar collagens, with expression of non-fibrillar collagens increasing between E17.5 and neonatal stages as part of a more general increase in ECM content (Peacock et al., 2008). Collagen is crucial for correct maturation of the aortic valves in mice, as exemplified in Osteogenesis imperfecta murine (oim) mutants in which Col1a2 is disrupted. The aortic valves of oim mutants display increased distal thickening and persistent PG deposition at 5 month and 9 months (Chakraborty 2010; Cheek, 2012). Interestingly, Eln/+ neonate valves display a transient increase in mesenchymal proliferation that is lost in juveniles. However, collagen bundles in the valves are disorganised in juvenile Eln/+ mice, and valves become stiffer with late onset aortic valve prolapse (Hinton et al., 2010).

Concomitant with valve leaflet formation, the ventricle undergoes trabeculation, which in mice initiates with asymmetric myocardial cell divisions dependent on Has2 (Passer et al., 2016) (Fig. 5A). By E9.0, these laminar trabeculae become surrounded in an ECM bubble, which forms when endocardial sprouts establish contact with the underlying compact myocardium (del Monte-Nieto et al., 2018).

Initial ECM bubble formation correlates with myocardial expression of Has2, Vcan and other components (del Monte-Nieto et al., 2018) (Fig. 5B). Vcan deposition appears to play a key role, as Vcan^hdf mouse mutants exhibit a significant reduction in the number of proliferating myocardial cells and lack trabeculae (Cooley et al., 2012). Similarly, loss of nidogen 1 and nidogen 2 leads to trabeculae hypoplasia (Bader et al., 2005). Invasive endocardial behaviour correlates with expression of Adamts1, Hyal2 and Mmp2 (del Monte-Nieto et al., 2018), suggesting that regionalised degradation of the ECM is important for elaboration of the trabeculae.

Accordingly, between E9.5 and E12.5 in mice, Vcan levels decrease through cleavage, and versikine levels increase, initially only at the base of the radial trabeculae (Fig. 5B). Over time, Vcan cleavage expands apically to attenuate proliferation, such that, by E13.5, the ECM is almost totally absent and trabeculation terminates (Cooley et al., 2012; del Monte-Nieto et al., 2018). This increase in versikine is dependent on fibulin 1, the expression of which overlaps with that of Adamts1 (Cooley et al., 2012; Stankunas et al., 2008) (Fig. 5B). Supporting a functional role for ECM degradation in trabeculation, Adamts1 mutants display increased trabecular mass, suggesting that ADAMTS1-mediated ECM degradation limits trabeculation (Stankunas et al., 2008). This balanced synthesis and degradation of the cardiac jelly in the ECM bubbles is governed by the reciprocal roles of endocardial Notch1 and myocardial Nrg1/Erbb signalling (del Monte-Nieto et al., 2018).

Zebrafish display a similar change in ventricular ECM organisation during trabeculation. Between 48 and 72 h post-fertilisation, the ventricular cardiac jelly thins, increasing the contact between the myocardium and endocardium (Rasouli and Stainier, 2017). This ECM thinning appears to be chamber specific, as the atrium retains a thicker ECM layer, which may function to restrict endocardial-myocardial interactions and inhibit initiation of trabeculation. This highlights a conserved role for regionalised ECM degradation in promoting trabeculation and suggests that chamber-specific ECM degradation facilitates differential maturation of the chambers.

The epicardium comprises the outermost layer of the heart wall. It is derived from the proepicardium: a small group of cells that expands and migrates over the surface of cardiomyocytes as the heart develops. Studies in chick and Xenopus have suggested the presence of an ECM ‘bridge’ that forms between the pericardium and the heart, along which proepicardial cells migrate onto the surface of the heart (Jahr et al., 2008; Nahirney et al., 2003). This bridge comprises ECM components such as HS and Fn. After the proepicardial bridge has contacted the myocardium, HS is upregulated at the myocardial-proepicardial interface, potentially promoting proepicardial migration over the heart. It is unclear whether this ECM bridge is conserved in zebrafish, as some studies support a model in which direct contact between the myocardium and pericardial mesothelium, along with fluid forces in the pericardial cavity, transfer proepicardial cells onto the surface of the heart (Peralta et al., 2013; Plavicki et al., 2014). This is similar to proepicardial cell transfer in mice, which occurs through the extension of multicellular villi in conjunction with free-floating proepicardial cysts (Rodgers et al., 2008). However, a recent study suggests the ECM bridge is conserved in zebrafish (Lan et al., 2019) and potentially rich in ColXII fibrils, which guide the migration of epicardial cells onto the heart surface from the adjacent pericardial region (Marro et al., 2016). It is possible that an ECM bridge facilitates the initial migration of proepicardial cells onto the zebrafish heart, with fluid forces in the pericardial cavity then aiding epicardial spreading over the cardiac surface.

In addition to functional analyses of ECM composition, recent single-cell analyses have highlighted the importance of the ECM in cardiac development. Transcriptomic analyses of cells from dissected regions of the mouse heart between E9.5 and postnatal day 21 (P21) identified a non-fibroblast cardiomyocyte subpopulation expressing high levels of ECM components, including collagens, decorin and periostin (DeLaughter et al., 2016). Further meta-analyses of single-cell transcriptomic data highlighted specialised ECM synthesis and modulation as a central pathway regulated by key developmental cardiac transcription factors, with dynamic regulation of genes encoding ECM components (including Has2, Chsy1 and Hapln1) particularly prevalent in atrial cells (Liu et al., 2019). Importantly, changes in the cardiac ECM have also been suggested to occur during human heart development. A single-cell transcriptomic analysis of human foetal hearts between 5 weeks and 25 weeks of gestation identified nine cell clusters that exhibited distinct expression profiles over development, including an ECM-rich fibroblast-like cell population that increased proportionally over time, and a cardiomyocyte cluster characterised by immediate and gradually increasing levels of ECM components such as lumican (Cui et al., 2019). Co-expression network analysis of cardiomyocytes also uncovered an ECM-related sub-network of genes that is active not only at early stages of development but also at later stages, with only a few genes being conserved between stages. ECM components conserved in both early and late stage cardiomyocytes include several collagen subunits, decorin, and Fn1, a similar profile to the ECM components expressed in mouse embryonic non-fibroblast cardiomyocytes (DeLaughter et al., 2016). This suggests that, during human heart development, cardiomyocytes have a dynamic ECM expression profile that is linked to maturation.

The importance of ECM synthesis, remodelling and turnover in human heart development is further supported by the identification of mutations in genes encoding ECM components linked to CHDs. Alongside this, large-scale studies have identified loci associated with CHDs that encode ECM components but for which a causative link has not yet been established. The links between ECM components and CHDs identified to date are summarised in Table 1. Below, we discuss three examples that further highlight the conserved role of ECM turnover in developmental heart morphogenesis.

Marfan syndrome (MFS) is caused by mutations affecting fibrillin 1 (Fbn1) and is associated with altered TGFβ signalling and elastolysis, leading to disruption of the aortic wall (thoracic aortic aneurysm and dissection) (Sakai et al., 2016). The MFS mouse model Fbn1C1039G/+ and MFS patient aortas display a significant reduction in ADAMTS1 levels (Oller et al., 2017). Additionally, the Fbn1mgR/mgR MFS mouse model exhibits reduced aggrecan cleavage (Cikach et al., 2018). Supporting a role for reduced ADAMTS1 activity in driving MFS cardiac abnormalities, Adamts1 heterozygous mice display similar medial degeneration of the aorta and increased TGFβ signalling (Oller et al., 2017). Thus, loss of a single ECM component can have complex effects on the cardiac ECM environment during development.

Homozygous mutations in ADAMTS19 have recently been associated with a non-syndromic CHD leading to progressive valve disease (Wünnemann et al., 2020). Homozygous Adamts19 mutant mice have a low penetrance of aortic valve (AoV) disease at 3 months, with a disorganised and thickened ECM. Levels of Klf2 are dramatically increased in Adamts19 mutant AoV endocardium at 3 weeks, suggesting that a loss of Adamts19 results in alterations to shear stress prior to AoV defects. Interestingly, no known proteolytic targets of ADAMTS19 have been identified, suggesting a distinct mechanism of regulating ECM composition compared with other ADAMTS enzymes (Wünnemann et al., 2020).

Proteoglycan (PG) synthesis and turnover are crucial for valve development, and a number of rare mutations in key enzymes that control this turnover and, hence, affect valve function in humans have been identified. Mitral valve (MV) disease has been linked to mutations in UGDH, a key synthetic enzyme related to PG synthesis (see Box 1) and previously associated with heart valve defects in zebrafish (Hyde et al., 2012; Walsh and Stainier, 2001). Downstream of UGDH activity, GAG attachment to the core protein is dependent on B3GAT3 [the gene encoding glucuronosyl-transferase I (GlcAT-I; see Box 1)]. The R227Q mutation in human GlcAT-I is linked to familial CHDs with a degree of variability, but four out of five patients present with MV prolapse and patient-derived fibroblasts display reduced formation of chondroitin sulfate proteoglycan (CSPG) and HSPG chains (Baasanjav et al., 2011). Supporting a conserved requirement for HA turnover in heart development, a single nucleotide polymorphism (SNP) in HYAL2 has been linked to Cor triatriatum, a particularly rare CHD, with similar structural defects also observed in some Hyal2 mutant mice (Muggenthaler et al., 2017). Together, these transcriptomic analyses and links between CHDs and ECM components demonstrate the importance of the cardiac jelly during human heart development and also highlight the power of animal models in elucidating the molecular mechanisms through which the ECM drives cardiac morphogenesis.

Whereas most adult vertebrates have limited cardiac regenerative potential, fish, salamanders and neonatal mice retain regenerative capacity. This regenerative capacity involves coordinated clearance of dead tissue, repopulation of cardiomyocytes, revascularisation and scar removal.

Comparative analyses of injured zebrafish and mouse hearts have revealed that ‘extracellular matrix’ is the most upregulated gene ontology (GO) term in zebrafish, compared with ‘inflammatory response’ in mouse (Mercer et al., 2013). Proteomic analyses of decellularised zebrafish hearts post-ventricular amputation have provided a more focused analysis of the changes in ECM composition (Garcia-Puig et al., 2019). In both studies, fibronectin 1b is elevated post-injury, while fibrous ECM components, including fibrillin 2b and multiple collagen-encoding genes are reduced. Supporting an important role for the ECM in facilitating cardiac regeneration, application of zebrafish ECM (zECM) to human cardiac progenitor cells (CPCs) improves migration and proliferative capacity (Chen et al., 2016). Moreover, administration of zECM to mouse hearts in a myocardial infarction (MI) model improves regenerative response (Chen et al., 2016).

The decrease in cardiac regenerative capacity in neonatal mice soon after birth is accompanied by reduced cardiomyocyte proliferation, with transcriptomic and protein analyses demonstrating increased collagen and laminin content (Notari et al., 2018). A similar transcriptomic analysis of mouse hearts suggests a shift in ECM content as the heart matures, concomitant with loss of regenerative potential (Jam et al., 2018). This is supported by studies in rats, where juvenile-associated ECM composition changes as the rodent ages, with Fn the most prevalent component in foetal and neonatal hearts, and type I collagen the most abundant in adults (Williams et al., 2014). These changes in ECM composition in mature hearts correlate with decreased proliferative capacity, as only ECM derived from foetal rat hearts or neonatal mouse hearts has a pro-proliferative effect when applied to cardiomyocytes in vitro, similar to the effect of zECM on CPCs (Bassat et al., 2017; Chen et al., 2016; Williams et al., 2014). Upregulation of MMPs post-injury is a conserved phenomenon in both zebrafish and newt models of regeneration (Lien et al., 2006; Mercer et al., 2013), consistent with a role for MMPs in degrading cell debris and matrix. However, MMPs may also regulate the bioavailability of ECM components that promote early cardiomyocyte proliferative potential, as MMP inhibition reduces the ability of neonatal mouse ECM to promote cardiomyocyte proliferation in vitro (Bassat et al., 2017).

Changes in ECM composition in regenerating zebrafish hearts are accompanied by a decrease in tissue stiffness (Garcia-Puig et al., 2019), whereas altered ECM composition in postnatal mice is linked to increased ECM stiffness (Notari et al., 2018). Changes in cardiac stiffness during development and into adulthood appear to be conserved, as chick heart tissue stiffens around tenfold from early development (Majkut et al., 2013). Pharmacological disruption of collagen and elastin crosslinking decreases tissue stiffness in P3 mice, reducing fibrosis and scarring after ventricular resection, suggesting an inverse correlation between ECM-modulated tissue stiffness and regenerative capacity (Notari et al., 2018).

The role of ECM modulation in driving cardiac regeneration likely extends beyond simply regulating tissue stiffness. The epicardium facilitates multiple aspects of the regenerative response (Quijada et al., 2020). In zebrafish, Fn is upregulated in the epicardium soon after injury, potentially promoting cardiomyocyte migration to the wound (Wang et al., 2013). This cardiomyocyte-centred response contrasts with epicardial migration in explanted hearts, where Fn density regulates epicardial leader cell behaviour (Uroz et al., 2019), suggesting Fn is required in multiple cardiac subtypes post-injury. Temporal changes in ECM composition during zebrafish heart regeneration drive a transient fibrotic response, which may promote initial wound healing and progression through subsequent phases of regeneration (Chablais and Jazwinska, 2012; Sánchez-Iranzo et al., 2018). A proteomic analysis of heart regeneration in a zebrafish resection model (Missinato et al., 2015) revealed that multiple HA pathway components (hmmr, cd44, has1 and has2) are also upregulated post-amputation, suggesting that HA synthesis and signalling are important for the regenerative response. Supporting this, pharmacological inhibition of HA synthesis or loss of the HA receptor hmmr results in a failure in clot resolution. HA pathway genes are also upregulated in rodent MI models, suggesting some conservation of the initial injury response (Petz et al., 2019; Shi et al., 2019). Pharmacological and genetic inhibition of HA synthesis in the mouse MI model results in loss of HA synthesis post-injury, and a reduction in the already limited regenerative potential. This initial, subacute phase of HA synthesis may support myofibroblast differentiation and macrophage survival to promote healing post-injury (Petz et al., 2019). However, it is unclear why, despite a similar initial HA response in zebrafish and murine injury models, overall regenerative capacity is different.

Understanding how ECM composition promotes regeneration presents multiple therapeutic possibilities. For example, laminin 221 improves the generation of cardiovascular precursors from embryonic stem cells (ESCs), which can mature into cardiomyocytes after implantation into the heart of a mouse MI model and induce modest improvements to cardiac function (Yap et al., 2019). More direct approaches, such as the administration of agrin (a HSPG) to mouse MI models, result in reduced scarring, increased cardiomyocyte proliferation and improved cardiac function (Bassat et al., 2017). Similarly, co-injection of HSPG and FGF into rat hearts post-MI improves cardiomyocyte survival and proliferation, reduces infarct size and improves cardiac function (Shi et al., 2019). More generally, promoting regenerative potential by reducing tissue stiffness through modulation of the ECM is an interesting concept, although ECM-driven increases in tissue stiffness postnatally may help support cardiac function in vertebrates with more complex cardiovascular systems, and thus it is unclear whether long-term modulation of tissue stiffness would result in additional pathologies. Nevertheless, the specific and temporal modulation of ECM composition may be key to improving multiple aspects of regenerative potential.

The dynamic nature of the ECM during heart development raises further questions around how spatiotemporal coordination of ECM production, modification and turnover supports ongoing tissue morphogenesis. One consideration is that the cardiac ECM is synthesised at early stages of heart morphogenesis to support tissue growth, and a subsequent temporal switch to modification or degradation promotes tissue remodelling to generate form. Supporting this idea, Vcan deposition drives proliferation to build tissue during valvulogenesis and trabeculation, but later in development its cleavage acts as a switch to attenuate cell proliferation as the valves and trabeculae remodel (Cooley et al., 2012; Dupuis et al., 2011; Kern et al., 2006; del Monte-Nieto et al., 2018; Stankunas et al., 2008). Similarly, HA deposition is initially required for heart tube formation, but subsequently undergoes regional degradation to facilitate tube looping (Smith et al., 2011; Totong et al., 2011). This temporal switch in ECM composition, which may also underlie changes in tissue stiffness, could be key to understanding how to use the ECM as a tool to promote specific aspects of heart regeneration. Localised changes in ECM degradation also highlight roles for ECM turnover in facilitating the myocardial-endocardial interactions that promote functional regionalisation of the heart. During zebrafish development, the ventricular ECM is degraded prior to the onset of trabeculation, while the atrial ECM remains thicker, and it is likely that these differences in chamber ECM thickness facilitate or suppress the myocardial-endocardial interactions that support trabeculation (Rasouli and Stainier, 2017). Endocardial-myocardial crosstalk can also promote the ECM degradation that drives ballooning morphogenesis of the atrium (Kim et al., 2018; Zhou et al., 2015), and the ECM may play further roles in relaying signals between the two tissue layers to ensure their coordinated growth (Bornhorst et al., 2019). The combination of timely ECM degradation along with additional chamber-specific signalling pathways may therefore promote different maturation events for individual chambers.

Myocardial and endocardial contributions to ECM modulation are also likely to shift during development. For example, during zebrafish cardiogenesis, HA is synthesised first by the myocardium and subsequently by the endocardium, with the hyaluronidase cemip2 following similar temporal functional dynamics (Smith et al., 2008, 2011; Totong et al., 2011). A better understanding of the tissue-specific contributions of ECM synthesis and ECM-mediated signal transduction will help define the complex relationship between myocardial-endocardial interactions during heart development. Furthermore, how these chamber-specific differences in ECM turnover are established and maintained during early development remains unclear. Mechanisms underlying the regionalisation of genes such as Has2 place emphasis on the role of the chamber/non-chamber genetic programme. However, blood flow and flow-sensing also appear to play conserved roles in promoting regionalised ECM synthesis (Steed et al., 2016a) and modification (Dupuis et al., 2011; Goddard et al., 2017). For example, loss of endocardially expressed Adamts5 or endothelial deletion of Klf2 results in enlarged AV valves, suggesting that onset of cardiac function is linked to ECM modification and morphogenesis, in part, through flow sensing (Dupuis et al., 2011; Goddard et al., 2017). Whether fluid dynamics impact further on regional gene expression associated with ECM modification in other aspects of cardiac morphogenesis requires further interrogation, such as whether the defects in myocardial chamber wall integrity in zebrafish klf2a/b mutants result from abnormal cardiac ECM composition (Rasouli et al., 2018).

Advances in our understanding of the ECM in cardiac development are increasingly arising from studies of zebrafish embryos. In particular, the ability to image heart development in live embryos at high resolution provides fantastic opportunities to investigate the links between blood flow, tissue morphology and ECM composition. However, to better define how regionalised and dynamic ECM changes drive cardiac morphogenesis, one important challenge remains: the ability to visualise ECM synthesis and turnover in real time in a developing heart. Tools are now being developed to perform live in vivo ECM imaging. For example, transgenic zebrafish expressing the HA sensor ssNcan-GFP (in which the HA binding domain of Neurocan is fused to GFP) can be used to facilitate visualization of the HA-rich cardiac jelly (Grassini et al., 2018). Developing other such biosensors by using well-characterised binding domains of ECM proteins may prove useful in visualising changes to ECM composition during development. However, it is unclear whether such biosensors could outcompete endogenous ECM-binding proteins, potentially disrupting ECM homeostasis and impacting development or ongoing cardiac function. A recent highly promising technique in invertebrate embryos involves tagging ECM proteins with fast-folding fluorophores to visualise and quantify ECM dynamics live (Keeley et al., 2020; Matsubayashi et al., 2020). Importantly, tagging should be undertaken at the endogenous locus rather than through overexpression constructs, as copy number variations in ECM proteins have been linked with CHDs (Silversides et al., 2012; Soemedi et al., 2012). Finally, although zebrafish provide fantastic opportunities to correlate ECM dynamics with heart morphogenesis in live embryos, recent studies detailing live imaging in toto of developing mouse hearts at cellular resolution (Yue et al., 2020) highlight exciting possibilities for complementary studies defining the conserved roles for ECM turnover in cardiac morphology, maturation and function. Together, focusing on the development and application of novel techniques to capture spatiotemporal changes in the ECM, in combination with candidate-driven functional analyses, will help move the field forward to define how ECM dynamism drives cardiac development and repair.

We thank Juliana Sánchez-Posada and Emma Armitage for helpful comments on earlier versions of the manuscript.