Mechanisms of tissue fusion during development

During embryonic development, there are many instances when two opposing tissues come into contact and fuse together to form one continuous structure. This type of tissue fusion occurs during the formation of many organs, including the optic cup, palate, heart, neural tube, eyelids and body wall. Superficially, tissue fusion in various organs appears to be similar, and mice deficient in specific transcription factors and signaling molecules exhibit defects in fusion of multiple organs (Zhang et al., 1996; Yu et al., 2010; Pyrgaki et al., 2011). However, organ formation is a complex process that often involves mechanisms highly specific to that tissue, and, accordingly, the precise mechanisms used to drive fusion in individual tissues are also context dependent.

Multiple animal models have been used to study tissue fusion and many types of tissue fusion events occur within a developing organism. Importantly, the inability of tissues to fuse correctly during development can lead to various birth defects, including cleft palate (Abbott, 2010), spina bifida (Copp et al., 1990) and heart defects (Wenink and Zevallos, 1988). The etiology of these debilitating defects in humans is likely to be complex, involving the concurrent disruption of several factors. In model organisms, the ability to ablate molecules specifically in individual tissues or at specific time points has allowed these complex developmental events to be studied and the functions of individual genes to be evaluated. Recent advances in the ability to culture and manipulate developing organs have also identified molecular pathways involved in tissue fusion events. Our understanding of how these fusion events occur has thus grown exponentially over the past 10 years, yet many questions remain.

Here, we summarize what is currently known about tissue fusion in several different organs to highlight both the similarities and differences between these fusion events. In particular, we highlight studies that largely use mouse models and have provided insights into the molecular and cellular events underlying fusion in the developing heart, palate and neural tube, disruptions of which are responsible for the largest classes of human birth defects. We also outline promising future research avenues that could increase our understanding of these events. For the purpose of this review, we refine our discussion to cover tissue fusion events during which individual cells retain their integrity, rather than cellular fusion events in which multiple cells can fuse to form multinucleated cells, as happens, for example, during skeletal myogenesis.

The best characterized developmental system, in terms of tissue fusion, is the palate, the tissue that separates the oral cavity from the nasal cavity and forms the roof of the mouth. Palate development is very complex and even small perturbations lead to craniofacial defects known as cleft lip with or without cleft palate (CLP) and cleft palate (CP) (see Box 1). There has long been interest in studying normal palatal formation to understand the etiology behind CLP and CP.

Palate development occurs over an extended period from the time of migration of neural crest cells into presumptive facial mesenchyme [before embryonic day (E) 9.0 in mice] until completion of fusion of the secondary palate (by ∼E16), which is one of the last embryonic structures to form by tissue fusion (Johnston and Bronsky, 1995). Interactions between neural crest, facial mesoderm, surrounding endoderm and ectoderm result in five facial prominences by E9.5: the singular frontonasal prominence, paired maxillary prominences and paired mandibular prominences (Fig. 1A) (Hinrichsen, 1985; Senders et al., 2003). Over the next day, the mandibular processes merge to form the lower lip and jaw. Meanwhile, the frontonasal prominence undergoes a series of morphological changes to appear as two upside down horseshoes, made up of the internal medial and external lateral nasal prominences, which together surround the nasal pits. Prior to initiation of primary palate formation, microvilli line the epithelial surface of the prominences. Just preceding fusion, microvilli disappear from cells at the lower end of the nasal pits. Fillipodia then emerge from epithelia of the lateral and medial nasal processes, cross the physical gap between these prominences and anchor in between cells of the opposing tissue, presumably to help initiate tight contact with one another (Millicovsky and Johnston, 1981). Fusion begins at E10.5, first between the maxillary and medial nasal prominences, from posterior to anterior, followed by fusion to the lateral nasal prominences starting at E11. Fusion of these three tissues continues until E12.5 and results in formation of the primary palate, which forms the upper jaw and lip.

Formation of the secondary palate occurs later in development (Farbman, 1968; Hinrichsen, 1985; Griffith and Hay, 1992) (reviewed by Bush and Jiang, 2012). The palatal shelves initially form as vertical outgrowths of the maxillary processes (Fig. 1B). Between E13 and E14, the palatal shelves elevate to lie laterally above the tongue and the medial edges of the shelves come in close proximity of one another. From E14.5 to E16, fusion between the two palatal shelves occurs first in the middle, then progresses both anteriorly and posteriorly. Successful fusion results in formation of the roof of the mouth, or secondary palate, which separates the oral and nasal cavities. Although evidence suggests that the mechanisms of tissue fusion are similar during development of both the primary and secondary palates, the majority of experiments have focused on secondary palate fusion for several reasons. First, secondary palate formation occurs later in embryogenesis; therefore, these fusion events can be disrupted by various teratogens without causing overall embryopathy. Second, secondary palates are comparatively large structures that can be cultured, which makes biochemical manipulation easier. Thus, the majority of work discussed here will center on secondary palate fusion.

Secondary palate fusion occurs progressively over time; thus, in a single palate multiple steps in fusion of opposing shelves can be visualized. These steps have been termed prefusion, fusion and fusion seam steps (Farbman, 1968). The two shelves are each surrounded by medial edge epithelium (MEE), a two- to four-cell thick epithelial layer separated from the mesenchyme by a basal lamina. During prefusion, the opposing epithelia are intact with only occasional disruptions in the basal lamina. During fusion, the two epithelia come into contact but the cellular morphology is similar to prefusion and there is no evidence of desmosomes, tight junctions or adhesive products between contacting epithelial cells. However, opposing cell layers are tightly adhered as application of physical forces causes cells from one epithelial layer to rip away from the basement membrane and instead remain with the opposing epithelial layer (Farbman, 1968). During the fusion seam step, phagocytes and dying epithelial cells appear in between healthy cells and the basal lamina degenerates (Farbman, 1968). The midline epithelium also thins, owing to convergent extension, as epithelia from both shelves intercalate and form the two-cell layer thick midline epithelial seam (MES). Some MEE cells also migrate away from the midline, where first contact occurs, towards the oral and nasal cavities along the MES to areas termed the ‘epithelial triangles’ (Martinez-Alvarez et al., 2000). As fusion continues and the epithelial layer is disrupted, mesenchymal cells from both shelves infiltrate the MES to establish continuity between the two tissues (Farbman, 1968; Hinrichsen and Stevens, 1974; Lee et al., 2008).

Early research discovered evidence of dying cells during palatal fusion, and apoptosis was, therefore, suggested to provide the driving force behind this fusion event (Farbman, 1968; Hinrichsen and Stevens, 1974). Alternatively, Fitchett and Hay proposed that the basal layer of epithelial cells lose their epithelial identity and adopt mesenchymal characteristics (Fitchett and Hay, 1989). Many studies have since attempted to determine the relative contributions of apoptosis versus epithelial-to-mesenchymal transition (EMT) to palatal fusion.

Apoptosis, as detected by terminal deoxynucleotidyl nuclear transferase dUTP nick end-labeling (TUNEL), is significantly increased within the MES and epithelial triangles of fusing mouse palates in vivo (Mori et al., 1994). Other studies used lipophilic dyes to trace the fate of MEE cells during secondary palate fusion ex utero (Griffith and Hay, 1992; Sun et al., 2000). Culture of secondary palates isolated prior to palatal fusion showed the dye was sequestered within ‘isolation bodies’ in epithelial cells, whereas, after further culture to the time of fusion, similar ‘isolation bodies’ were found within mesenchymal-like cells. In control cultures of single shelves that did not undergo fusion, mesenchymal cells did not contain dye. Thus, it was concluded that contact between palatal shelves initiates EMT in MEE cells. Additional studies have continued to provide evidence for the processes of EMT and apoptosis in palatal fusion (Martinez-Alvarez et al., 2000; Cuervo et al., 2002; Cuervo and Covarrubias, 2004).

However, other studies called into question the importance of apoptosis and/or EMT in palate fusion. In vitro studies using caspase 1 and caspase 3 inhibitors found that secondary palates can completely fuse even in the absence of apoptosis (Takahara et al., 2004). Fate-mapping studies using sonic hedgehog (Shh) and keratin 14 (K14) promoters to mark MEE cells permanently found no evidence of EMT (Vaziri Sani et al., 2005). So why are there so many studies with disparate results? This issue has been widely discussed and it is thought that variation of in vitro culture techniques and differences between in vitro and in vivo studies could contribute to the discrepancies. In addition, it appears there are differences in how fusion occurs along the anterior to posterior axis of the palatal shelves (Cuervo and Covarrubias, 2004; Takahara et al., 2004; Vaziri Sani et al., 2005). Although controversy remains, evidence points to apoptosis and migration of MEE cells to epithelial triangles as being the main contributors to MES disappearance and tissue fusion, with a possible contribution made by EMT.

Prior to palatal fusion, the facial prominences and palatal shelves must grow and become precisely positioned to bring opposing tissues in close proximity. Defects in tissue growth, morphogenesis and reciprocal signaling between epithelial and mesenchymal components can lead to CLP and CP. This is observed in mice with mutations in the transcription factors aristaless-like homeobox 3 (Alx3), Alx4, Alx1 (Cart1) and activating enhancer binding protein 2α (AP2α; Tfap2a – Mouse Genome Informatics). Disruption of these factors indirectly results in inability of opposing tissues to fuse, owing to decreased proliferation and/or increased apoptosis causing abnormal positioning of the tissues (Nottoli et al., 1998; Qu et al., 1999; Beverdam et al., 2001). So, although the events regulated by these transcription factors must occur correctly before fusion can occur, to date no transcription factor has been identified that directly drives tissue fusion in the palate.

During their growth and elevation, the palatal shelves are surrounded by a layer of peridermal cells. Periderm acts as a barrier that prevents the palatal epithelium from inappropriately fusing with the tongue and mandible. The transcription factor interferon regulatory factor 6 (Irf6) and the Notch ligand jagged 2 (Jag2) spatiotemporally maintain the periderm so that the MEE is exposed only at the appropriate time and place for successful palate fusion (Richardson et al., 2009). Mice deficient in Irf6 or Jag2 display CP, owing to improper fusion of the palatal shelves with other oral epithelia. Irf6 and the transcription factor p63 (Tcp1 – Mouse Genome Informatics) also regulate epithelial differentiation and apoptosis, misregulation of which interferes with fusion of both the primary and secondary palate (Richardson et al., 2009; Ferretti et al., 2011; Guerrini et al., 2011).

Genetic studies of humans have found that mutations in transforming growth factor (TGF) α and TGFβ family members can cause CLP and CP (Holder et al., 1992; Lidral et al., 1997). Mice deficient in Tgfb3 have fully penetrant CP, providing an animal model with which to study TGFβ3 function in palatal fusion (Proetzel et al., 1995; Koo et al., 2001). As recently reviewed (Iwata et al., 2011), Tgfb3 is strongly expressed within the MEE at the time of fusion and has many roles, including induction of apoptosis (Martinez-Alvarez et al., 2000) and production of matrix metalloproteases (MMPs) to promote EMT (Blavier et al., 2001). In addition, signaling through the WNT pathway is implicated in palatal fusion, as mice that are homozygous null for two WNT receptors, frizzled 1 (Fzd1) and frizzled 2 (Fzd2), have completely penetrant CP (Yu et al., 2010). Wnt11 is expressed in the MEE during fusion and is required for fusion of palatal shelves, as siRNA knockdown of Wnt11 in vitro results in decreased apoptosis in MEE cells and failure of fusion (Lee et al., 2008). Retinoic acid (RA) signaling in neural-crest-derived mesenchyme also must be tightly controlled, as either too little or too much RA causes CP (Lohnes et al., 1994; Cuervo et al., 2002). Finally, the secreted ECM protein periostin is produced by mesenchymal cells directly under the MEE and induces EMT in epithelial cells within the MES, although periostin has not been shown to regulate palatal fusion directly (Kitase et al., 2011).

Ephrin signaling is also important in secondary palate fusion. Ephrins are cell surface ligands that bind to ephrin receptor tyrosine kinases (Eph) on opposing cells. Upon binding, signaling is initiated in both cells via reciprocal signaling from the ligand (reverse signaling) and receptor (forward signaling). Mice deficient in both neural kinase (Nuk; Ephb2 – Mouse Genome Informatics) and Sek4 (Ephb3 – Mouse Genome Informatics), two ephrin receptors expressed along the MES, have highly penetrant CP (Orioli et al., 1996). Forward signaling through Ephb2 and Ephb3 is required for palatal fusion; however, these signaling molecules regulate proliferation and not palatal fusion itself (Risley et al., 2009). Reverse signaling through ephrin B2, however, directly affects palatal shelf fusion, and ephrin B2 is expressed at high levels in MEE cells just prior to fusion and within the MES and neighboring mesenchyme during fusion. This reverse signaling is PI3 kinase dependent and may involve interaction with claudins, which are important epithelial adhesion proteins (Dravis and Henkemeyer, 2011; San Miguel et al., 2011).

Thus, through histological, genetic and culture experiments, there is a good understanding of the mechanisms underlying secondary palate fusion, and to a lesser extent fusion of the nasal and maxillary prominences to form the primary palate.

One of the earliest embryonic structures to form is the neural tube (NT), which gives rise to the central nervous system. Initially, ectoderm along the dorsal side of the embryo is specified to be neuroepithelium and this epithelium then thickens to form the neural plate (Fig. 2A). In response to signals between the neuropithelium and surrounding tissues, and the forces generated by tissue movements, the neuroepithelium forms hinge points and bends on both sides in a U shape to elevate the neural folds (Fig. 2B). The opposing folds approach one another (Fig. 2C) and then come into contact to undergo a tissue fusion event that results in formation of the continuous NT (Fig. 2D). During NT fusion, the neuroepithelium separates from the neighboring non-neural ectoderm and then both tissues close to form the neural tube covered by a single sheet of non-neural ectoderm. Failures in neural tube formation lead to a class of birth defects collectively referred to as neural tube defects (NTDs) (see Box 2).

There are over 240 mouse models of NTDs (Harris and Juriloff, 2010). Many are due to recessive mutations of individual genes, with some showing low penetrance NTDs when haploinsufficient. Genetic complexity is also demonstrated in models of compound mutations of two or three genes. The majority of these mouse models exhibit exencephaly, whereas spina bifida alone is less common. Although NTDs are ultimately due to failure of NT closure, very few mouse models directly relate to fusion, the final step of NT closure. For example, mice mutant for genes in the planar cell polarity (PCP) pathway, such as Vangl2 and inturned (Intu), exhibit NTDs due to lack of convergent extension, which is needed to elongate the neural plate and bring the neural folds close together for fusion to occur (Ybot-Gonzalez et al., 2007; Wansleeben et al., 2010; Zeng et al., 2010). Moreover, many NTDs disrupt proliferation, hinge point formation, primary ciliogenesis or neural patterning, but these important processes occur independently of NT fusion (Murdoch and Copp, 2010).

Studies of NT fusion in the head regions of mice using transmission electron microscopy (TEM) revealed that the tissue layer that makes initial contact differs depending on the region of closure (Geelen and Langman, 1979). At the rhombencephalon and mesencephalon, non-neural epithelial cells first come into contact followed by the neural epithelium. In the anterior neuropore, the neural epithelium appears to make the first contact. In the prosencephalon region, both cell layers appear to make contact simultaneously. However, no matter which tissue layer first comes into contact, the cells exhibit similar behaviors at the time of fusion along the rostrocaudal axis; finger-like projections extend from cells of both neural folds and intercalate with each other as the two folds come together. No epithelial junctional complexes are observed except in the mesencephalon, where junctions are observed between cells of the non-neural ectoderm. Recent advances in mouse embryo ex utero culture techniques have allowed for live imaging of NT closure and hence a more dynamic evaluation of cellular behaviors during this process. In such a study of cranial NT closure, cellular projections were also seen in the gap between the neural folds (Pyrgaki et al., 2010).

Recent genetic studies provide important information on the molecular mechanisms that directly regulate fusion of the neural folds. Two members of the Grainyhead-like (Grhl) transcription factor family, Grhl2 and Grhl3, are expressed in the non-neural ectoderm during NT fusion in discrete, as well as overlapping, patterns; loss of expression of one or both genes leads to NTD and developmental defects of many tissues (Rifat et al., 2010). Here, we concentrate on NTDs; however, Grhl2 mutants, for example, also show failure of face and body wall closure, suggesting that Grhl2 plays a more general role in multiple fusion processes, although this has not been directly tested. Different alleles of Grhl2 show some phenotypic differences, possibly owing to differences in genetic background and the specific mutation, yet all exhibit highly penetrant exencephaly and cleft face, and some alleles show spina bifida (Rifat et al., 2010; Werth et al., 2010; Pyrgaki et al., 2011). Histological analysis of Grhl2 mutants at E9.5 showed proper elevation and apposition of the neural folds in the head region but fusion itself failed to occur (Pyrgaki et al., 2011). Grhl2 directly regulates expression of E-cadherin and claudin 4, both of which are important for formation of epithelial junctional complexes, and Grhl2 mutant mice show decreased expression of these two proteins in non-neural ectoderm with a concurrent increase of N-cadherin (Werth et al., 2010). E-cadherin has been implicated in NT closure; knockdown of E-cadherin with antisense oligonucleotides during NT closure in rats resulted in NTD (Chen and Hales, 1995). Additional direct and indirect Grhl2 targets include molecules involved in adhesion, such as desmoglein 2, desmocolin 2, desmoplakin and epithelial cell adhesion molecule, and an increase in the matrix metalloproteases Adamts1 and Adamts3 (Pyrgaki et al., 2011). The identification of a number of adhesion genes regulated by Grhl2, and perhaps Grhl3, presents an opportunity for further study of the role of adhesion in NT fusion.

Loss of Grhl3 function also results in fully penetrant NTDs, characterized by spina bifida or curled tail, and infrequent exencephaly (Ting et al., 2003). Grhl3 can activate transcription and several epithelial-specific genes are direct targets of Grhl3, including keratin 5, keratin 6 and keratin 10, filaggrin and involucrin (Boglev et al., 2011). Grhl2 and Grhl3 have overlapping expression patterns and embryos with compound heterozygous and homozygous null mutations of both Grhl2 and Grhl3 have demonstrated their relative contribution to NT closure: closure point 3 appears to depend on Grhl2, whereas closure point 2 is regulated by both transcription factors with the overall gene dose being the most important factor in successful closure. However, closure point 1 occurs in the absence of both Grhl2 and Grhl3 (Rifat et al., 2010). A role for Grhl3 in hindgut endoderm has been suggested in spinal NT closure, as loss of Grhl3 expression in the hindgut endoderm leads to decreased proliferation, which causes increased ventral curvature and hence pulls the neural folds apart so that fusion cannot take place (Ting et al., 2003; Gustavsson et al., 2007).

Two well-characterized mouse NTD models that arose from spontaneous mutations are axial defects (Axd) and curly tail (Ct). In both cases, heterozygous embryos exhibit a ‘curly tail’ due to delayed closure of the posterior neuropore (PNP), whereas homozygotes display spina bifida. The causative mutations are not known but both result in misregulation of Grhl expression. Axd mutants exhibit increased Grhl2 expression, and NT closure can be restored by decreasing the level of Grhl2 (Brouns et al., 2011). Ct mutants show decreased expression of Grhl3, and NT closure can be rescued by Grhl3 overexpression (Gustavsson et al., 2007). Thus, these studies suggest that Grhl activity must be tightly controlled during NT closure.

Although early EM studies indicated a lack of epithelial junctional complexes at the NT fusion seam (Geelen and Langman, 1979), the studies above highlight a role for cell adhesion in NT fusion. Dynamic regulation of cell-cell adhesion during NT fusion is further implicated by studies of the tumor suppressor gene neurofibromatosis type 2 (Nf2) (McLaughlin et al., 2007). Nf2 (also known as merlin) regulates assembly of apicolateral junctional complexes. Nf2 expression decreases at the tips of the dorsal neural folds just prior to fusion, and then sharply increases after completion of fusion. Knock out of Nf2 in the neuroepithelium at E8.5 does not affect the initial fusion process but, after E9.5, the NT reopens, resulting in NTDs. Neural and non-neural ectoderm cells are healthy in Nf2 mutants but they detach from the apical surface due to a lack of epithelial junctions. Interestingly, Nf2 mutants have other developmental defects, including cleft palate, eye and body wall defects and cardiac ventricle septal defects, although whether these represent fusion defects remains to be tested (McLaughlin et al., 2007). Thus, Nf2 may act to differentially regulate cell-cell adhesion during multiple tissue fusion events.

The transcription factor AP-2α is also required for NT closure and proper palate formation (Schorle et al., 1996; Zhang et al., 1996). In AP-2α mutant embryos, the neural folds elevate but fusion does not occur. However, the failure of neural fold fusion may result from overproliferation of the neuroepithelium such that the folds cannot physically meet. Loss of the transcriptional activators CBP (Crebbp – Mouse Genome Informatics) and p300 (Ep300 – Mouse Genome Informatics), as well as their co-activators Cited2 and Cart1 (Alx1 – Mouse Genome Informatics), all individually result in many developmental defects, including NTDs. All of these proteins are found specifically in the dorsal neural folds at the time of NT closure, suggesting they may regulate neural fold fusion (Bhattacherjee et al., 2009).

Similar to palate fusion, signaling via ephrins and their receptors is required for NT fusion. Ephrin A5 (Efna5 – Mouse Genome Informatics) and the Epha7 receptor are expressed at the tips of the cranial neural folds. Mice deficient in Efna5 or Epha7 exhibit exencephaly at low penetrance, apparently owing to an inability of the neural folds to fuse (Holmberg et al., 2000). In addition, Efna1 and Efna3 are expressed in the caudal neural folds and PNP, as are the receptors Epha1, Epha2, Epha4 and Epha5. Both Epha2 and Epha4 are strongly expressed at the tips of the neural folds during fusion with Epha2 expressed in non-neural ectoderm. Blocking EphA activity, with an EphA fusion protein to disrupt ligand-receptor interactions, in the caudal region resulted in increased PNP size, which was attributed to defective tissue fusion and not to a problem in NT morphogenesis (Abdul-Aziz et al., 2009).

Two recent studies discovered roles for G-protein-coupled receptor (GPCR) signaling in NT fusion. The spontaneous mouse mutation vacuolated lens (vl), which is considered a model of failure of neural fold fusion, results from mutation of an orphan GPCR, Gpr161, which is expressed in the neural folds (Matteson et al., 2008). Another class of GPCRs is the protease-activated receptors (PARs), which respond to proteases in the environment. Compound null mutations for Par1 and Par2 (Gpr172b – Mouse Genome Informatics) result in exencephaly, and occasional spina bifida or curly tail. Par2 is expressed in non-neural ectoderm at the time of fusion and disruption of downstream signaling (G_i or Rac1) in the non-neural ectoderm also causes exencephaly (Camerer et al., 2010).

In contrast to the requirement for apoptosis in primary and secondary palate fusion, a role for apoptosis during NT fusion is unclear. Scanning electron microscopy (SEM) showed scattered apoptotic cells in neural tissue undergoing fusion (Geelen and Langman, 1979). TUNEL staining showed a correlation between the presence of apoptotic cells and the bending and fusing of opposing neural folds (Massa et al., 2009). To more directly investigate the role of apoptosis in NT closure, in vivo and in vitro models have been used. Casp3^–/– or Apaf1^–/– embryos fail to close the caudal midbrain and hindbrain regions (Cecconi et al., 1998; Leonard et al., 2002; Massa et al., 2009). Although the absence of apoptosis in these mice leads to cranial NTDs, it is difficult to draw conclusions about the direct role of apoptosis in fusion owing to the general effects that a lack of apoptosis may have on other cellular processes. In embryo cultures in which pharmacological inhibitors of apoptosis were added just at the time of NT closure, it was found that apoptosis is not required for NT closure (Massa et al., 2009). A recent study used live imaging techniques to further explore a role for apoptosis in NT closure (Yamaguchi et al., 2011). In Casp3^–/– or Apaf1^–/– mice, there was reduced bending of the neural plate and reduced flipping of dorsal ridges in the midbrain-hindbrain region. Additionally, these mutant mice and wild-type embryos treated with caspase inhibitor showed reduced speed of closure at closure points 1 and 2, although complete closure could still occur. This work suggests a model in which apoptosis helps NT closure proceed within a necessary time frame before additional forces prohibit successful fusion.

Tissue fusion plays an integral role during development of the mammalian heart, as it transforms a simple tube-like structure in to a complex four-chambered organ. Fusion occurs in several different areas of the heart and is involved in separation of the atria and ventricles, and in division of the initial singular outflow tract into the aorta and pulmonary trunk. There is considerable variation in the processes of fusion within the heart, possibly owing in part to the contribution of neural crest cells in the developing outflow tract (Poelmann et al., 1998; Waldo et al., 1998). Early understanding of the morphological and fusion events in the developing heart came from histological and SEM studies. Moreover, direct comparisons of this morphogenetic process have been made between mouse, chicken and human embryonic hearts (Pexieder, 1978; Thompson et al., 1985; Vuillemin and Pexieder, 1989). Unfortunately, the advances in in vitro organ culture methods that spurred a mechanistic understanding of palate and neural tube fusion have lagged behind in the field of heart development. Thus, of the three organs highlighted in this review, we know the least about what drives fusion in the heart.

By E9.5 in the mouse, the unseptated tubular heart has already looped to the right (Webb et al., 2003). Between E9.5 and E10.5, two sets of endocardial cushions (ECs) begin to grow towards each other across the open internal space of the tube (Fig. 3A). The first set, the conotruncal ECs, form across the common outflow tract (the conotruncus; CT), which leads outwards from the primitive right ventricle to the aortic sac. The conotruncal ECs are further divided into proximal conotruncal and distal conotruncal EC pairs (PCEC and DCEC). Eventual fusion of the DCECs creates the conotruncal septum, which separates the aortic and pulmonary trunks between E12.5 and E13.5. The PCECs fuse later and contribute to the membranous ventricular septum. The second set of cushions, the atrioventricular ECs, develop across the bend of the looped heart within the atrioventricular canal and eventually fuse to separate the atrial and ventricular spaces (Webb et al., 1998). At the same time, the primary atrial septum and the ventricular septum begin to grow inwards from the roof of the common atrium and floor of the ventricular chamber, respectively. These two septa will fuse with the atrioventricular ECs to create the four chambers of the heart (Fig. 3B). These fusion events occur in a precisely coordinated pattern between E12.5 and E13.5. Once fusion occurs, myocardiocytes invade the ECs and tissue remodeling creates the major valves of the heart. Thus, defects in fusion result in a multitude of cardiac defects (see Box 3).

The primary heart tube consists of an external layer of myocardium and an internal endocardial layer. These layers are separated by extracellular matrix called cardiac jelly produced by the myocardium (Henderson and Copp, 1998). As the endocardial cushions develop, endocardial cells proliferate, undergo EMT and migrate into the cardiac jelly (Fig. 3A, inset). As the two cushions come into contact, the endocardial cell barrier breaks down, and mesenchymal cells form a bridge between the two cushions to stabilize fusion of the two tissues (Fig. 3B; bottom inset). If bridge formation is perturbed, then fusion is incomplete and the cushions are pulled apart (Ghyselinck et al., 1998). The atrial and ventricular septa, by contrast, both consist of growing myocardium with leading edges of mesenchymal cells. These mesenchymal ‘caps’ then fuse with mesenchymal cells of the endocardial cushions (Fig. 3B, insets).

Signaling between endothelial and mesenchymal components is important for regulating growth of the ECs, and perturbation of several signaling molecules can lead to fusion defects. Maternal deficiencies in retinoic acid (RA) cause many congenital abnormalities, which are together termed vitamin A deficiency syndrome, that affect several tissues, including the heart. RA signals through heterodimers of RA receptors and retinoic X receptors (RXRs). Mice deficient in both Rxra and Rxrb exhibit defects in conotruncal EC fusion, leading to ventricular septal defects (Ghyselinck et al., 1998). In these mutants, EC contact occurs but the endocardium does not break down and mesenchymal bridge formation is also perturbed owing to increased apoptosis and decreased mitosis of EC mesenchyme. This fusion defect is specific to the conotruncus, whereas atrioventricular EC fusion proceeds normally.

Conversely, deficiencies in ephrin signaling lead to fusion defects specifically in the atrioventricular ECs and not in the CT (Stephen et al., 2007). Epha3 is expressed in the mesenchyme of both the atrioventricular and conotruncal ECs, and in the mesenchymal cap of the atrial septum, whereas Epha1 is expressed in the adjacent endothelial cells. Epha3^–/– mutants exhibit defects in EMT and mesenchymal cell migration, and show increased apoptosis in both sets of ECs, with delayed fusion of the atrioventricular ECs, as well as defective atrial septum formation.

Versican (also known as PG-M) is a chondroitin sulfate proteoglycan expressed in areas of EMT in mouse embryos, although it is considered non-permissive for migration and instead is implicated in regulating cell-cell or cell-substratum adhesion. In the heart, versican is dynamically expressed in the developing ventricles, conotruncus and trabeculated myocardium, and probably plays several different roles. Moreover, versican is highly expressed in the mesenchymal cap of both the atrial and ventricular septa and its expression sharply declines following EC fusion. Versican is also highly expressed within the growing ECs, and in versican knockout mice, the ECs do not grow and endocardial cells show defective migration in tissue explants (Henderson and Copp, 1998). A similar phenotype is seen in mice deficient for hyaluronin, another ECM component of the cardiac jelly (Camenisch et al., 2001). Thus, these ECM components are required for development of the ECs and septa; however, a specific involvement in fusion of these tissue components has yet to be investigated.

Another ECM component expressed in the developing heart is the secreted matricellular protein cysteine-rich angiogenic inducer 61 (Ccn1; Cyr61 – Mouse Genome Informatics). Ccn1 interacts with integrin receptors to promote adhesion, migration, proliferation, differentiation, survival or death, depending on the cell type. Ccn1 is expressed in the space between the septa and cushions prior to fusion, and Ccn1 knockout mice exhibit fully penetrant atrioventricular or valvular septal defects owing to lack of tissue fusion. Ccn1 may promote fusion through induction of matrix metallopeptidase 2 (Mmp2), which contributes to tissue remodeling, and by promoting cell survival (Mo and Lau, 2006).

Neural crest cells, which contribute to the outflow tract of the developing heart, are also required for conotruncal EC fusion; however, the mechanisms by which they influence fusion may be different between the distal and proximal cushions. These cells migrate through the third pharyngeal arch to the conotruncal ECs between E9.5 and E12.5. Mice deficient in neural crest-derived meltrin β, an ADAM family metalloprotease, exhibit normal fusion of the distal ECs, but are unable to fuse the proximal ECs, even though they come in close proximity (Komatsu et al., 2007).

As with the palate and neural tube, WNT signaling appears to be involved, as evidenced by ventricular septal defects in Fzd1/Fzd2 knockout mice. However, this could result from general deficiencies in neural crest migration and direct involvement with heart tissue fusion remains to be determined (Yu et al., 2010). A number of transcription factors are also needed for correct formation of the heart (Takeuchi et al., 2003); however, none of these has yet been implicated specifically in fusion.

Conserved patterns of cell death have been observed in chick, rat and human embryonic hearts within the fusion seams of both the outflow tract and atrioventricular ECs, and in the ventricular septum (Pexieder, 1975). These observations, along with teratogenic studies using chick hearts, led Pexieder to hypothesize that cell death plays an important role in heart morphogenesis. TUNEL assays also demonstrated dynamic patterns of apoptosis in the heart from E11.5 onwards (Zhao and Rivkees, 2000). At E12.5, when the atrioventricular ECs start to fuse, apoptosis occurs within the fusion seam, then decreases after fusion is complete. Fusion of these cushions occurs in a zippering manner from anterior to posterior, and the dynamic increase in apoptosis closely follows this pattern. However, in the conotruncal ECs, increased apoptosis is seen not only in the fusion seam, but throughout the cushions, and therefore may play a more general role in tissue remodeling (Zhao and Rivkees, 2000). A subset of neural crest cells within the conotruncal ECs is highly susceptible to apoptosis, and this may be reflected in the differential pattern of apoptosis observed (Poelmann et al., 1998). These studies suggest a role for apoptosis in fusion events of the heart; however, additional research is needed to determine its necessity, as has been carried out in the palate and NT.

So far, we have focused on tissue fusion events in the murine palate, neural tube and heart. There are, however, many more fusion events that occur during development of mammals and other organisms. As shown in Fig. 4, proper development of the eyes, diaphragm and urethra all require tissue fusion. Failures of these fusion events result in birth defects such as coloboma, diaphragmatic hernia and hypospadias, respectively. Very little is known about the mechanisms of fusion in these tissues. During urethral development, disruptions in Shh and Fgf signaling lead to decreased outgrowth of the urethra and hypospadias, although fusion itself is not disrupted (Cohn, 2011). Signaling through several Ephs is also involved in urethral closure; however, there have been no direct mechanisms identified that drive this tissue fusion event (Shaut et al., 2007; Cohn, 2011). It remains to be seen whether there are highly conserved mechanisms among some or all of the tissue fusion events that occur during development.

Superficially, many of the tissue fusion events discussed above appear similar. Prior to fusion, tissues must proliferate and position themselves such that they are in apposition at the right time and place for fusion to occur. During fusion, two separate tissues must come together to form one continuous tissue. However, detailed studies have shown that while some mechanisms are conserved among multiple tissues, there are also mechanisms that are highly specific to individual tissues. The issue remains of whether knowledge gained from studies in the palate, heart and neural tube can be used to understand the mechanisms behind fusion in less studied tissues. There are many human syndromes that are associated with multiple fusion defects, and genetic studies may help to uncover common mechanisms that control tissue fusion. Most of what we currently know about these syndromes, however, cannot be directly attributed to tissue fusion but rather to more global changes in key developmental signaling pathways or changes in gene dose. For example, Pallister-Killian syndrome causes cleft palate and diaphragmatic hernia, owing to an abnormal extra isochromosome (12p) (Holder et al., 2007). It may be more useful to infer similarities in fusion mechanisms based on the germ layers involved. For example, NT and urethral fusion occurs between opposing ectodermal cells, whereas fusion in the conotruncus, primary palate and diaphragm involves neural crest derivatives. Thus, there may be germ layer-specific mechanisms. There are also similarities in tissue fusion events among different species. Neural tube fusion also occurs in frog, chick and zebrafish embryos, whereas a similar event known as dorsal closure occurs in Drosophila. These organisms can provide distinct experimental and genetic advantages for studying tissue fusion. Indeed, insights derived from one organism have later been confirmed in mice. For example, Irf6 function in palate fusion is similar in chick and mice (Knight et al., 2006). Nonetheless, a common function is not always the case and it is important to not draw conclusions without careful analysis between animal models. In the case of NT closure, studies in chick found that apoptosis is indispensable for fusion, whereas, in mice, apoptosis plays a minimal role in NT fusion (Weil et al., 1997; Massa et al., 2009; Yamaguchi et al., 2011). Drosophila dorsal closure differs from NT closure in that the dorsal epithelia move along the amnioserosa rather than meeting across a physical gap (Jankovics and Brunner, 2006). Yet in both species fillipodia are extended towards each other before fusion occurs; in Drosophila this is required for zippering the epithelial sheets together. Therefore, although tissue fusion events may use somewhat different strategies or genes, information gained from the study of one fusion event could be extremely useful in learning about other fusion events.

This review of tissue fusion events in the developing mammalian palate, neural tube and heart highlights several key processes that they have in common, as well as processes that are different or yet unexplored (summarized in Table 1). For two separate tissues to become one cohesive entity, cells in each tissue must decrease their tight associations with one another in order to allow new contacts to form with cells from the opposing tissue. Although this general principle holds true in the tissues discussed here, the process by which it is achieved varies from intercalation of opposing epithelial cells, to EMT, to apoptosis or a combination of these processes. Apoptosis is observed during fusion of all three tissues but its importance in neural tube and heart fusion remains unclear. Changes in cell-cell adhesion will underlie fusion, but a molecular understanding of the regulation of cell adhesion during fusion is still limited. One clear similarity is the requirement for ephrin/Eph signaling. Ephrin/Eph signaling is usually associated with repulsion but in the three organs discussed above, ephrin/Eph signaling elicits an adhesive response, the loss of which results in failure of tissue fusion. Grhl family members transcriptionally regulate adhesion genes and are required for fusion in the neural tube and face, and probably the body wall and eye; however, a possible role in heart fusion is unexplored.

Although we are beginning to understand the mechanisms that drive tissue fusion in embryogenesis, many questions remain. Communication between opposing tissues is likely to be crucially important but has as yet been little studied. During neural tube fusion, PARs are expressed specifically within the non-neural ectoderm, suggesting that proteases in the paracellular environment could activate these receptors. But from where would such secreted factors arise? Do cells from the opposing tissues secrete factors that aid the two tissues in finding each other across open space? Tight coordination of cell proliferation is also a key factor in correct alignment of opposing tissues such that fusion can occur. However, what triggers the cellular responses that arise before the tissues come into contact (such as the fillopodia that epithelial cells extend towards each other)?

All of the fusion events described here are tightly regulated such that even small perturbations can disrupt tissue fusion. However, a great challenge when studying tissue fusion using an animal model is that the processes preceding tissue fusion are also highly complex and disruptions to these earlier events can result in disease phenotypes that resemble fusion defects, even though fusion may not be directly affected. The ability to culture palatal shelves in vitro provides a way to uncouple the final tissue fusion from earlier events; however, there are no similar protocols for specifically analyzing tissue fusion in the heart and NT. Historically, most studies of these tissues have relied on static images obtained from embryos at various stages, leading sometimes to ambiguous results. Recent advances in in vitro culture techniques of whole embryos and live imaging techniques, along with the use of tissue-specific fluorescent reporters and gene disruptions, can provide a dynamic analysis of tissue morphogenesis. This will enable more detailed evaluation of specific mechanisms of tissue fusion and address how specific genetic changes affect the cell behaviors necessary for tissue fusion.

It is unclear whether knowledge gained from study of palate, neural tube and heart fusion can be extrapolated to other tissue fusion events, such as during development of the body wall and eye. A few genes regulate fusion in more than one tissue, but the large number of cellular and molecular differences outlined here suggests there may not be a unifying mechanism. Finally, our understanding of the mechanisms involved in embryonic tissue fusion events largely comes from single gene mutations in animal models. Although individual mutations have been identified in humans with developmental defects, the cause of most birth defects is likely to be more complex. The biggest challenge then remaining is to elucidate and link the complex interplay between genetic factors, as well as environmental influences, to the etiology of tissue fusion defects in the developing human embryo.

We are grateful to LeAnn Fenton for the figures and to Trevor Williams for critical reading of the manuscript.