Adhesion-Based Self-Organization in Tissue Patterning

Abstract

Since the proposal of the Differential Adhesion Hypothesis, scientists have been fascinated by how cell adhesion mediates cellular self-organization to form spatial patterns during development. The search for molecular toolkits with homophilic binding specificity resulted in a diverse repertoire of adhesion molecules. Recent understanding of the dominant role of cortical tension over adhesion binding redirects the focus of differential adhesion studies to the signaling function of adhesion proteins to regulate actomyosin contractility. The broader framework of differential interfacial tension encompasses both adhesion and non-adhesion molecules, sharing the common function to modulate interfacial tension during cell sorting to generate diverse tissue patterns. Robust adhesion-based patterning requires close coordination between morphogen signaling, cell fate decisions, and changes in adhesion. Current advancement in bridging theoretical and experimental approaches present exciting opportunities to understand molecular, cellular, and tissue dynamics during adhesion-based tissue patterning across multiple time and length scales.

Introduction

The formation of complex tissue shape and pattern in embryos has fascinated generations of developmental biologists. Classic ideas to explain the mechanisms of pattern formation and morphogenesis have inspired novel experimental approaches that expanded our understanding of embryo development.

One of the most influential concepts is the Differential Adhesion Hypothesis, proposed more than 50 years ago in an attempt to explain how dissociated cells self-organize into reproducible spatial patterns (Steinberg, 1963). The hypothesis proposes that cell rearrangements are driven by effort to minimize the total adhesion energy. The hypothesis inspired the search of adhesion molecules that can fulfill the role of homophilic binding, leading to the discovery and characterization of cadherins (Takeichi, 1977). Following the cadherin discovery, the search for molecular toolkits to diversify the adhesion repertoire has led to identification of more than 100 members of the cadherin superfamily, as well as calcium-independent adhesion molecules. In the past decade, a more rigorous biophysical treatment of cell and tissue surface tension suggested the dominant role of cortical tension over adhesion binding, and redirected the focus of differential adhesion studies from the forces of adhesion-binding to the signaling role of adhesion proteins in modulating actomyosin contractility. The extended model of differential interfacial tension encompasses the function of adhesion molecules (e.g., cadherins) and non-adhesion molecules (e.g., Eph/ephrin) to provide a unified framework that explains how changes in homotypic and heterotypic interfacial tension result in distinct spatial patterns with a variable degree of robustness.

Besides differential adhesion, the French Flag Model is another insightful idea proposed more than 50 years ago, with long-lasting impact on the trajectory of developmental biology research (Wolpert, 1969). In the French Flag model, cells interpret biochemical signals such as morphogens to infer their positions within the tissue, and commit to distinct cell fates to form the tissue-scale pattern. Under this model, cell and tissue movement are considered detrimental to the patterning precision as they disrupt the positional information set by local morphogen concentrations. In contrast, patterning mediated by differential adhesion is more robust to cell and tissue movement as the cells possess an intrinsic ability to self-organize. Both mechanisms have ample examples in patterning of developing tissues (Dessaud et al., 2008; Schier, 2001; Schotz et al., 2008). Interplay between the two mechanisms can further improve the robustness of tissue patterning during tissue morphogenesis (Tsai et al., 2020), but the robustness depends critically on how morphogen signaling, cell fate specification, and changes of adhesion are coupled.

In this review, we aim to summarize the historical evolution of our understanding of adhesion-based patterning, and recent progress in the concept and molecular mechanisms of differential adhesion. We will discuss the diverse tissue patterns generated by adhesion-based mechanisms and the molecular toolkits to achieve the adhesion bias. We will end with how differential adhesion interacts with the morphogen signaling and cell fate specification to ensure robust pattern formation.

From differential adhesion to differential interfacial tension: the modern look of a classic idea

The observation of adhesion-based self-organization in tissue development can be dated back to more than a century ago. Henry Wilson observed that cells dissociated from sponges can coalesce to regenerate functional structures (Wilson, 1907; Wilson and Penney, 1930). Definitive evidence of adhesion-based self-organization came later, when Johannes Holtfreter reported that dissociated cells from amphibian embryos can self-assemble (Holtfreter, 1943). The size and pigment differences for distinct cell types allowed Holtfreter to trace these cells during reaggregation, and conclude that cells are sorting out, with medullary plate cells sorting to the center and epidermal cells to the periphery (Townes and Holtfreter, 1955). Aron Moscana found similar properties for dissociated cells from chick limb bud, with chondrogenic cells sorting to the center and myogenic cells to the periphery (Moscona and Moscona, 1952), suggesting the sorting behavior is preserved in well-differentiated cells from later-staged embryos with established tissue patterning.

The experiments from Holtfreter and Moscana inspired Malcolm Steinberg to propose the Differential Adhesion Hypothesis (DAH). He extended Holtfreter’s concept of tissue selective affinity with a thermodynamic framework, and proposed that each tissue has an intrinsic tissue surface tension determined by cell-cell adhesion (Steinberg, 1963). When tissues with distinct surface tension are mixed, they will sort to minimize their total surface free energy, much like immiscible fluids with different surface tension separate. The differential adhesion hypothesis predicts that cells which differ only in their expression level of a single adhesion molecule should not only sort from one another, but also adopt a spatial arrangement with cells expressing more adhesion molecules in the center of the aggregate. This hypothesis was directly confirmed by manipulating P-cadherin expression level in an in vitro cell aggregation system (Steinberg and Takeichi, 1994). Tuning cadherin levels (E-, N-, P-cadherins) in cultured L cells reveals the linear relationship between the surface tension of aggregates and the density of surface cadherins, further supporting the differential adhesion hypothesis (Foty and Steinberg, 2005).

However, quantitative measurement of tissue surface tension suggested that it is several orders of magnitude stronger than the surface tension estimated from simply breaking cadherin bonds (10⁻³ N/m vs. 10⁻⁷ N/m) (Amack and Manning, 2012; Youssef et al., 2011), raising the question of whether cadherin-based adhesion can directly determine tissue surface tension. Instead, experimental manipulation of cortical tension suggested that it is a dominant determinant of tissue surface tension (Krieg et al., 2008; Manning et al., 2010). The recent experimental evidence is aligned with the Differential Interfacial Tension Hypothesis, a modern revision of the Differential Adhesion Hypothesis, taking into account the major contribution of cortical tension (Brodland, 2002). In this model, interfacial tension is determined by two opposing factors: (1) cortical contractile machineries, working to increase interfacial tension and shrink cell-cell contact, and (2) cell-cell adhesion, working to reduce interfacial tension and extend cell-cell contact (Fig. 1A). Cell sorting is driven by cellular rearrangement to minimize overall interfacial tension of the aggregate.

Figure 1: From differential adhesion to differential interfacial tension.

(A) Three distinct ways for cadherins to modulate interfacial tension: (1) cadherin binding in trans increase adhesion tension, (2) cadherin intracellular domains anchor to actomyosin cytoskeleton through catenins to stabilize cadherin binding in trans. (3) the intracellular domain of cadherin signals to change cytoskeleton structure and lower actomyosin contractility. (B) adhesion tension (orange arrows) acts to lower interfacial tension and stabilize the contact, while cortical tension (green arrows) acts to increase interfacial tension and shrink the contact. When interfacial tension increases, contact angle (θ) decreases and contact area shrinks. When interfacial tension decreases, contact angle (θ) increases and contact area expands.

If the direct contribution of cadherin-based adhesion to interfacial tension is much lower than the cortical tension, how do we reconcile the evidence of cell sorting based on different levels of cadherins? One critical insight comes from the recent discovery of the versatile functions of cadherins. Cadherin-based cell adhesion can reduce interfacial tension by three distinct ways: (1) adhesion tension - formation of cadherin bonds in trans to increase adhesion, (2) adhesion signaling - signaling to the cytoskeleton to reduce cortical contractility, and (3) adhesion coupling - mechanical coupling between cadherins and the contractile actomyosin cytoskeleton to stabilize cell contacts (Fig. 1B) (Maître and Heisenberg, 2013; Maître et al., 2012). Although the direct contribution of cadherin bond formation to reduce interfacial tension might be weaker, localization of cadherins to homotypic contacts can significantly reduce cortical contractility through adhesion signaling to lower interfacial tension (Anastasiadis et al., 2000). Therefore, cell types with higher levels of cadherins can generate lower interfacial tension and sort out under the differential interfacial tension framework, a conclusion that is consistent with previous experimental findings inspired by the differential adhesion hypothesis. Nevertheless, if distinct cell types exhibit different capacity for adhesion signaling and coupling, then sorting outcome might not strictly depend on cadherin abundance, as exemplified by zebrafish germ layer progenitors (Krieg et al., 2008; Maître et al., 2012).

The Differential Interfacial Tension Model can be generalized to understand cell sorting dynamics beyond cadherin-based cell adhesion, such as cell segregation with Eph/ephrin signaling. When cells expressing Eph receptor make contacts with cells expressing ephrin ligand, the ligand-receptor binding triggers bidirectional signaling to activate actomyosin contraction, increase interfacial tension, and shrink the cell contact. Under this unifying framework, sorting of two distinct cell types (A & B) is most robust when homotypic interfacial tension (γa-a & γb-b) are both much lower than the heterotypic interfacial tension (γa-b) (Canty et al., 2017).

In this review, we extend the definition of adhesion to encompass the broader framework of interfacial tension. Consequently, regulators of differential adhesion can include non-adhesion molecules that modulate interfacial tension.

Forces arising from differential adhesion (and thus differential interfacial tension) have profound implications for tissue level patterning. In the absence of additional signaling, these forces will drive cells towards the minimum energy configuration, which can take a myriad of distinct forms depending on the relative magnitudes of the homotypic and heterotypic interfacial tensions. Here we describe the biological regimes in which each type of pattern is formed, as well as their relevance to embryonic tissues.

The bullseye:

The predominant sorted pattern observed in ex vivo cell culture is the bullseye or concentric spheroids (Duguay et al., 2003; Foty and Steinberg, 2005; Foty et al., 1996; Krieg et al., 2008; Schötz et al., 2008), wherein cells with lower homotypic interfacial tension become enveloped by those with higher homotypic interfacial tension, and the heterotypic interfacial tension is intermediate (Fig. 2, γaa < γab< γbb). In the case of adhesion-based patterning, the most adherent cell population comprises the innermost compartment. This pattern has been reproduced in cellular Potts (Glazier and Graner, 1993; Krieg et al., 2008) and finite element-based simulations (Brodland, 2002).

Figure 2: Patterns mediated by adhesion-based sorting in a two-component system.

(Left) Schematic identifying interfacial tension at each possible cell-cell interface in a mixture of two different cell types labeled “A” and “B”. Example cell shapes are shown for cases where adhesions facilitate either a homotypic or heterotypic preference. (Right) The types of patterns that have been demonstrated to arise by differential adhesion, organized by the relative interfacial tension at homotypic and heterotypic interfaces.

Separate spheres and stripes:

A second configuration seen ex vivo is that of separate spheroids, which form when the homotypic interfacial tensions are lower (and adhesivity is higher) than at heterotypic interfaces (Fig. 2, γaa,γbb < γab). Adhesion-based patterning of this type has been predicted by computational modeling (Brodland, 2002; Glazier and Graner, 1993) and has been observed in vivo. However, patterning in the embryo, shaped by developmental programming of adhesion protein expression, can be much more complex. For example, in the developing chick spinal cord, differences in expression of several cadherins (which typically prefer homophilic interactions, see Classical cadherins and Protocadherins), drive the formation of at least six distinct cell clusters called motor neuron pools (Price et al. 2002). During development of the zebrafish neural tube, a Sonic Hedgehog signaling gradient drives differential expression of three different cadherins between three neural progenitor subtypes, allowing the cells to self-organize into stripes (Tsai et al., 2020).

Checkerboards and soccer ball patterns:

In contrast, when heterotypic interactions are preferred (i.e., have the highest adhesion strength and lowest interfacial tension – Fig. 2, γab < γaa,γbb) cellular patterns favor intermixing and dispersal of the cells. The classic example is the checkerboard pattern, originally predicted by computational modeling of the avian oviduct epithelium (Honda et al., 1986). Such effects have been reproduced in subsequent modeling (Glazier and Graner, 1993) and have also been observed in the embryo (Togashi, 2016). Differential expression of nectin family proteins (which prefer heterophilic interactions, see Nectin) between hair and supporting cells in the mouse auditory epithelium drives the cells to interdigitate and form a checkerboard arrangement (Togashi et al., 2011). In the mouse olfactory epithelium, differential expression of both nectins and cadherins between the olfactory and supporting cells induces a mosaic distribution (Katsunuma et al., 2016). In fact, in in vitro cell culture systems, tuning the relative strength of homotypic and heterotypic interactions between two cell populations can drive the formation of patterns ranging from cell type segregation, to an intermixed checkerboard arrangement, to a soccer ball pattern wherein each cell of one population is surrounded by cells of the opposing type (Katsunuma et al., 2016).

More complex patterns – Gradients, hitchhikers, and soap bubbles:

Adhesions are also used for positioning of individual cells in the embryo. In Drosophila, the oocyte is robustly positioned to the posterior end of the follicle by differential expression of DE-cadherin in the oocyte relative to the other germline cells inside the follicle, as well as a gradient of cadherin expression among the surrounding follicle cells (Godt and Tepass, 1998). Similarly, during gastrulation, C. elegans primordial germ cells attach to internalizing endodermal cells via E-cadherin, hitching a ride in order to undergo proper positioning inside the embryo (Chihara and Nance, 2012). Differential expression of N-cadherin has also been suggested to play a role in patterning of the Drosophila retina, wherein four cone cells are arranged in a stereotyped soap bubble-like packing within the confines of the surrounding pigment cells (Hayashi and Carthew, 2004).

Overall, a diverse range of cellular patterns mediated by adhesions have been observed, including the bullseye, stripe, checkerboard, soccer ball, and mosaic arrangements.

Molecular toolkits to generate differential adhesion and interfacial tension

The differential adhesion hypothesis was proposed before any adhesion molecule was identified (Steinberg, 1963). In the past decades, scientists have discovered a wide range of adhesion and non-adhesion molecules that can generate bias of interfacial tension between homotypic and heterotypic contacts to promote tissue patterning. The identification of cadherin marks the first molecular evidence that such a homophilic adhesion molecule exists (Takeichi, 1977). Since then, the search for molecules possessing sequences similar to the extracellular domain of cadherin has identified over 100 molecules in the cadherin superfamily, all sharing the common feature of homophilic binding in trans. The classical cadherins are the most well-characterized in their homophilic binding affinity, while recent studies have begun to shed light on mechanisms involving protocadherins. Other families of adhesion molecules such as nectins have the distinct property to favor heterophilic binding. Beyond adhesion molecules, a broad range of adhesion-independent surface molecules, represented by the Eph/ephrin receptor-ligand pairs, can also modulate interfacial tension to mediate cell sorting and tissue patterning.

In this section, we will summarize current understanding of several classes of adhesion or non-adhesion interfacial tension regulators, with a focus on the molecular underpinnings of their homophilic or heterophilic binding specificities, modulation of cytoskeleton by their cytoplasmic domains to alter interfacial tension, and evidence of their roles in tissue patterning in vivo.

Classical cadherins

Classical cadherins are calcium-dependent adhesion molecules with homophilic binding activity. Their extracellular domains contain five ectodomains, also known as extracellular cadherin (EC) repeats, while the intracellular domains contain binding sites for β-catenin and p120 catenin to link adhesion to the actin cytoskeleton. They utilize the first ectodomain to dimerize in trans with cadherins from neighboring cells. Therefore, the adhesion specificity of cadherins lies in the structure of their first ectodomain, which divide classical cadherins into two types.

Type I cadherins

Type I cadherins are the earliest cadherins identified (Hatta et al., 1985; Nose and Takeichi, 1986; Yoshida and Takeichi, 1982). The family includes E-cadherin (cdh1), N-cadherin (cdh2), P-cadherin (cdh3), R-cadherin (cdh4), and M-cadherin (cdh15). Members within the type I cadherin family share common structural features in their EC1 domain, requiring a critical tryptophan residue at position 2 (W2) to form trans-dimers (Nagar et al., 1996; Shapiro et al., 1995; Tamura et al., 1998). Point mutation of the W2 residue is sufficient to abrogate the adhesion (Tamura et al., 1998). Despite evidence that cells expressing distinct type I cadherins — E- vs. P-cadherin (Nose et al., 1988) or E- vs. N-cadherin (Friedlander et al., 1989) — can sort in aggregation assays ex vivo, the high similarities of EC1 sequences among type I cadherins led structural biologists to question their adhesion specificities (Katsamba et al., 2009; Patel et al., 2003). Indeed, promiscuous heterophilic binding between type I cadherins have also been observed (Duguay et al., 2003; Niessen and Gumbiner, 2002). Quantitative measurement of the homophilic and heterophilic binding affinities between E- and N-cadherin suggested their heterophilic binding affinity is intermediate between their homophilic affinities (N-N > E-N > E-E) (Katsamba et al., 2009). Therefore, while some type I cadherins indeed exhibit stronger homophilic binding affinity, the heterophilic binding among type I cadherins is not negligible. Quantitative difference of cadherins may account for results in some of the aggregation assays involving type I cadherins. Sorting between cells expressing different type I cadherins with significant heterophilic binding can still happen if one species is more abundant than the other.

The signaling effect of cadherin to the cytoskeleton is best studied for E-cadherin. Upon E-cadherin-binding in trans, phosphatidylinositol 3-kinase and Rac1are locally activated (Perez et al., 2008; Yamada and Nelson, 2007). P120-catenin binds to engaged cadherin through its cytoplasmic domain, and inhibits RhoA locally (Anastasiadis et al., 2000). Changes of Rac and Rho GTPase activities following cadherin binding significantly alter the actomyosin cytoskeleton structure at the homotypic contact, and reduce interfacial tension to stabilize the contact (Maître and Heisenberg, 2013).

Genetic evidence suggested tissue patterning in vivo can result from differential expression of type I cadherins. During gastrulation, germ layer progenitors express distinct levels of E-cadherin, which was critical for their separation in the aggregation assay (Schötz et al., 2008). However, later studies suggested that the sorting capacity of these germ layer progenitor cells depend more critically on how E-cadherin is coupled to the cortical contractile machinery, rather than the absolute E-cadherin level, further highlighting the dominant effect of adhesion signaling over adhesion binding (Krieg et al., 2008; Maître et al., 2012). In the zebrafish somite, N-cadherin and M-cadherin are expressed in complementary compartments, sharing only a very thin zone with high N- and M-cadherin co-expression. The high N- and M-cadherin zone gradually moves from medial to lateral somite, as the M-cadherin domain expands and the N-cadherin domain shrinks. The medial-to-lateral movement of this boundary is critical to guide the migration of slow twitching muscle cells, a unique cell type enriched in both N- and M-cadherin, through differential adhesion (Cortés et al., 2003).

Type II cadherins:

Crystal structures of the EC1 domains of type II cadherin revealed two unique features distinct from type I cadherin, with one additional tryptophan residue at position 4 (W4) and larger hyprophobic regions in their binding interface, making them incompatible to bind type I cadherins (Katsamba et al., 2009; Patel et al., 2006). Indeed, in vitro sorting assays between cells expressing type I and type II cadherin yielded much clearer separation between the two cell populations, compared to the outcome of cells expressing distinct type I cadherins (Katsamba et al., 2009; Patel et al., 2006). A painstaking effort has systematically characterized homophilic and heterophilic binding affinities among 10 distinct type II cadherins to reveal three specificity groups (Brasch et al., 2018). Cadherins within the same specificity group exhibit significant heterophilic binding affinity, while cadherins across different specificity groups have minimal heterophilic affinity. Specifically, cdh6, cdh9, and cdh10 belong to the first group, cdh8 and cdh11 comprise the second group, while cdh7, cdh12, cdh18, cdh20, cdh22 define the third group. These groups also agree well with the inferred phylogeny of type II cadherins based on sequence similarities of their EC1-EC2 domains (Brasch et al., 2018).

Patterned expression of type II cadherins were long-noted in the developing nervous system, including the embryonic and postnatal mouse brain (Hertel et al., 2008; Redies and Takeichi, 1996; Suzuki et al., 1997), rat hippocampus (Bekirov et al., 2002), chick and zebrafish spinal cord (Franklin and Sargent, 1996; Price et al., 2002), and mouse retina (Duan et al., 2014). In the majority of these examples, each type II cadherin is expressed in a unique, but partially overlapping, pattern with another type II cadherin , resulting in specific combinations for distinct domains. Perturbation of this combinatorial expression disrupts the boundaries of these domains in vivo For example, eF and A motor neuron pools in the chick spinal cord differ in their expression of Cdh20. Mosaic up- and down-regulation of Cdh20 adhesion disrupts the segregation of eF and A motor neurons and result in significant mixing (Patel et al., 2006; Price et al., 2002).

Type II cadherins also collaborate with type I cadherins during tissue patterning in vivo. In the mouse telencephalon, cdh4, a type I cadherin, is expressed in future cortex, while cdh6, a type II cadherin, is expressed in the neighboring striatum. Ectopic expression of either cdh4 or cdh6 results in mis-localization of striatal cells or cortical cells into the neighboring compartment, respectively, and therefore disrupts the boundary (Suzuki et al., 1997). In the zebrafish spinal cord, cdh2, a type I cadherin, and cdh11, a type II cadherin, are enriched in distinct neural progenitor types (p0 and pMN). Loss of cdh2 or cdh11 reduces the strength of the p0-p0 or pMN-pMN homotypic adhesion, respectively, leading to errors in patterning of the p0 or pMN progenitors in vivo (Tsai et al., 2020).

Protocadherins:

Protocadherins were initially identified as a family of proteins with high homology with the extracellular domains of cadherins (Sano et al., 1993). They contain six or seven subdomains that are similar to the five ectodomains of classical cadherins, while the intracellular domains of protocadherins are more distinct from those of cadherins, and do not possess binding sites for catenins or any actin-binding proteins. There are two major families of protocadherins. Non-clustered protocadherins are distributed throughout the genome, and are divided into δ-1 and δ-2 families, distinguished by the number of extracellular cadherin repeats (seven for δ-1 and six for δ-2). Clustered protocadherin includes protocadherin α, β, and γ, each of which is clustered at a single genomic loci highly conserved across the vertebrate species.

Non-clustered protocadherins

Within the non-clustered protocadherin (ncPcdh) families, δ-1 protocadherins contain seven cadherin repeats at their extracellular domain, and include 4 members, pcdh1, pcdh7, pcdh9, and pcdh11. δ-2 protocadherins contain six cadherin repeats at their extracellular domain, and include 5 members, pcdh8, pcdh10, pcdh17, pcdh18, and pcdh19. There has been controversial evidence of whether these δ -protocadherins mediate homophilic adhesion with relevant strength. Earlier studies have reported homophilic adhesion activities for pcdh1 (1993), Pcdh7 (Yoshida, 2003), Pcdh8 (Yamagata et al., 1999), and Pcdh10 (Hirano et al., 1999). However, when compared with classical cadherins in the same assay, ncPcdhs exhibit weaker adhesion, as exemplified by the comparison between L cells expressing Pcdh10 or R-cadherin (Hirano et al., 1999), or beads coated with extracellular domains of ncPcdhs or N-cadherins (Blevins et al., 2011; Cooper et al., 2016; Emond et al., 2011). Despite the weaker adhesion than classical cadherins, recent efforts over-expressing the extracellular and transmembrane domains of ncPcdhs in cultured K562 cells suggest all ncPcdhs, with the exception of Pcdh10, can mediate calcium-dependent homophilic adhesion to promote cell aggregation (Bisogni et al., 2018). Cells over-expressing distinct ncPcdh subtypes cannot aggregate, suggesting strict homophilic specificity (Bisogni et al., 2018). Structural study reveals that δ2 -protocadherins (e.g., pcdh19) form homophilic adhesion in trans through anti-parallel interaction of fully overlapped EC1 to EC4 domains, a mechanism distinct from those in classical cadherins (Cooper et al., 2016). Together, these works suggest distinct protocadherin genes can mediate calcium-dependent homophilic adhesion in cultured cells in vitro, while the adhesion strength may be weaker than classical cadherins in comparable conditions.

The weaker ncPcdh adhesion can be partly attributed to its cytoplasmic domain. As opposed to the strong coupling between cytoplasmic domains of classical cadherins and the actin cytoskeleton to stabilize adhesion, the cytoplasmic domains of ncPcdhs have no known binding partners that anchor to the actin cytoskeleton. Instead, they appear to suppress adhesion. When the cytoplasmic domains of pcdh17 or pcdh19 are deleted, cells exhibit greater aggregation (Hayashi et al., 2014; Tai et al., 2010). The cytoplasmic domains of ncPcdhs have versatile signaling roles, including crosstalk with Wnt and Wave signaling (Biswas et al., 2021; Pancho et al., 2020). The cytoplasmic domain of Pcdh17 binds to the WAVE complex, and recruits it to the cell-cell contact to convert the local cytoskeleton structure to more leading-edge like (Hayashi et al., 2014). With the diverse signaling functions of the cytoplasmic domains of ncPcdhs, it remains to be determined how they modulate interfacial tension at homotypic contacts.

Does the weaker ncPcdh adhesion imply its role in sorting of cadherin-expressing cells is less relevant? A critical in vitro study suggests ncPcdh (e.g., pcdh19) can form a complex with N-cadherin in cis to generate a novel mode of homophilic adhesion with stronger adhesion than ncPcdh alone (Emond et al., 2011). This complex depends on the extracellular domains of Pcdh19, but not Cdh2, to form adhesion in trans. Importantly, beads coated with both Pcdh19 and Cdh2 can segregate from beads coated with only Cdh2 or beads coated with pcdh17 and Cdh2 (Emond et al., 2011), suggesting the Pcdh19-Cdh2 complex can mediate an orthogonal adhesion mode from Cdh2 alone or co-expression of Cdh2 with other δ2-protocadherins. The in vivo relevance of this novel adhesion mode is supported by experiments of endogenous neural progenitor cells isolated from the zebrafish spinal cord. Manipulation of Pcdh19 level in endogenous neural progenitors suggest cells co-expressing Pcdh19 and Cdh2 exhibit similar strength of adhesion as cells only-expressing Cdh2 but not Pcdh19. But a cell co-expressing Pcdh19 and Cdh2 has minimal adhesion with another cell only expressing Cdh2 (Tsai et al., 2020).

Extensive literature has reported patterned expression of ncPcdhs from embryos to adults in different model organisms including zebrafish (Blevins et al., 2011; Kubota et al., 2004; Liu et al., 2009), Xenopus (Kim et al., 1998), chick (Lin et al., 2012; Müller et al., 2004) and mouse (Hertel et al., 2008; Vanhalst et al., 2005). Genetic perturbations have revealed their roles in cell sorting and tissue patterning. In the gastrulating Xenopus laevis embryo, axial protocadherin (pcdh1) is expressed in axial mesoderm while paraxial protocadherin (PAPC or pcdh8) is expressed in paraxial mesoderm (Kim et al., 1998). pcdh1 is necessary and sufficient to mediate sorting of axial mesoderm cells from non-axial mesoderm cells in the cell aggregation assay (Kuroda et al., 2002). In addition, Xenopus cells over-expressing pcdh1 or Pcdh8 can sort into separate compartments in the re-aggregation assay (Kim et al., 1998). Both gain- and loss-of-function perturbations of pcdh1 led to abnormal axial mesoderm patterning, exhibiting improper boundaries with paraxial mesoderm (Kuroda et al., 2002), while loss of Pcdh8 disrupted convergent extension movement of paraxial mesoderm (Kim et al., 1998). Subsequent studies have shown that Pcdh8 mediates cell sorting by down-regulating the activity, but not the level, of C-cadherin (Chen and Gumbiner, 2006). In addition to C-cadherin, Pcdh8 also down-regulates N-cadherin by enhancing its endocytosis. In hippocampal neurons, Pcdh8 reduces N-cadherin at dendritic spines to reduce synapse formation (Yasuda et al., 2007). In the chicken paraxial mesoderm, Pcdh8 reduces N-cadherin at the rostral part of each somite to form proper somite boundaries (Chal et al., 2017).

The adhesion function of Pcdh19 is highly relevant in human disease, such as pcdh19 girls clustering epilepsy (PCDH19-GCE). The Pcdh19 gene is located on the X chromosome of humans and mice. The disease typically arises in heterozygote females but spares hemizygote males, and is caused by random X inactivation, creating mosaic individuals with cells carrying either wildtype or mutant PCDH19 alleles (Dibbens et al., 2008). The mosaic expression of wildtype and mutated Pcdh19 caused abnormal cell sorting in the developing cortex, resulting in abnormal brain network activity (Pederick et al., 2018). pcdh19 mediates several key patterning processes in neural development. In the zebrafish spinal cord, Pcdh19 is expressed in the p3 progenitors to avoid mixing with the neighboring Pcdh19-low pMN progenitors. (Tsai et al., 2020). In the telencephalon of mouse embryos, progenitor cells in medial entorhinal cortex express the transcription factor COUP-TFI, which activates Pcdh19 expression to maintain sharp boundary with the neighboring Pcdh19-negative neocortex (Feng et al. 2021).

Clustered protocadherins

Clustered protocadherins (cPcdhs) represent the biggest subfamily of cadherin-related genes, with 58 genes in mouse and 53 in human. In human, mouse, and chicken, clustered protocadherins are composed of three closely linked gene clusters: Pcdh-α, Pcdh-β, and Pcdh-γ (Sugino et al., 2000, 2004; Wu and Maniatis, 1999). In the mouse locus, the Pcdh-α, Pcdh-β, and Pcdh-γ cluster each encodes 14, 22, and 22 distinct genes, respectively (Wu and Maniatis, 1999). Each of the gene clusters contains a variable region with a tandem array of variable exons, which includes six extracellular cadherin domains (EC1-EC6), the transmembrane domain, and a short intracellular domain. The Pcdh-α and Pcdh-γ clusters also have 3 constant exons, encoding the intracellular domain that is common to all genes in the same cluster (Wu and Maniatis, 1999). The sequences of the last two variable exons of the Pcdh-α cluster and the last three variable exons of the Pcdh-γ cluster diverged from the other variable exons in the same cluster, and are also referred to as the C-type Pcdhs (Wu and Maniatis, 1999). Each of the variable exons is preceded by a promoter, and promoter choice determines which variable exon is expressed (Tasic et al., 2002)

When over-expressed in K562 cells in vitro, all three subfamilies of cPcdhs can mediate highly specific trans homophilic adhesion, forming aggregates only with cells expressing identical subtype of cPcdhs (Thu et al., 2014). cPcdhs depend on their EC2 and EC3 domains to determine their homophilic binding specificity, which exhibit greatest amino acid sequence diversity among all EC domains. (Schreiner and Weiner, 2010; Wu, 2005). Remarkably, cPcdhs exhibit combinatorial homophilic specificity when multiple subtypes are co-expressed: cells only aggregate when all subtypes of cPcdhs expressed are matching, and one mismatched subtype is sufficient to disrupt aggregation despite other matching subtypes (Thu et al., 2014). The combinatorial homophilic specificity is attributed to the cis-interaction among different cPcdhs through EC5-EC6, resulting in a multimeric trans recognition unit (Goodman et al., 2016; Rubinstein et al., 2015). The EC1-EC4-dependent homophilic-specific trans-dimerization and EC5-EC6-dependent promiscuous cis-dimerization can generate a zipper-like structure only when all subtypes of cPcdhs are matched, and form the molecular basis of combinatorial homophilic specificity (Rubinstein et al., 2017).

Despite the ability to mediate homophilic adhesion, there is no clear evidence that cPcdhs participate in cell sorting during tissue patterning. Instead, the most notable function of clustered protocadherin is for neurons to discriminate self from non-self to avoid crossovers of neurites from the same neuron. Deletion of all 22 genes in the mouse Pcdh-γ cluster disrupts self-avoidance in neurons during neural development (Lefebvre et al., 2012). The function of dendritic self-avoidance for cPcdhs mimics the role of Dscam1 in Drosophila (Hughes et al., 2007; Matthews et al., 2007), where alternative splicing generates Dscam isoforms with 19,008 distinct extracellular domain sequences with homophilic specificity (Schmucker et al., 2000). Mechanistically, the cytoplasmic domains of cPcdh bind to two tyrosine kinases, Pyk2 and focal adhesion kinase (FAK), inhibiting their functions (Chen et al., 2009). The inhibition of Pyk2 and FAK result in activation of Rho GTPase (Suo et al., 2012), which in turn activate myosin contraction to increase cortical tension. While the matched extracellular domains of cPcdhs form adhesion in trans to lower interfacial tension, it is the larger increase of interfacial tension triggered by signaling of the cPcdh cytoplasmic domains that determine the repulsive outcome of cell-cell contact.

Other homophilic adhesion molecules

Beyond the cadherin superfamily, other homophilic adhesion molecules can also mediate tissue patterning, despite scarcer examples. Members of the cell adhesion molecule (CAM) family mediates calcium-independent homophilic adhesion, and are shown to mediate retinotectal map formation in frog, dermal condensation patterning in chick skin explant, and sensory domain patterning of cochlea epithelium (Fraser et al., 1984; Gallin et al., 1986; Harley et al., 2018).

Nectin

Nectin is a family of immunoglobulin-like adhesion molecules. It is comprised of four members (nectin-1, −2, −3, and −4) and their spliced variants (Takahashi et al., 1999; Takai and Nakanishi, 2003). The majority of the nectin genes consist of three immunoglobulin-like domains in the extracellular region, a single-pass transmembrane domain, and a cytoplasmic domain. Unlike the cadherin family genes, adhesion between nectin genes is calcium-independent, and nectins exhibit stronger affinity for heterophilic binding, with adhesion affinity nectin 1-3 > nectin 2-3 > nectin 1-1, 2-2 & 3-3 (Satoh-Horikawa et al., 2000). The cytoplasmic domain of nectins are connected to F-actin through the nectin- and F-actin-binding molecule afadin (Takahashi et al., 1999). In Drosophila ovaries and embryos, boundaries between domains with and without Echinoid, a nectin ortholog, exhibit contractile actomyosin cables to promote epithelial morphogenesis (Chang et al., 2011; Laplante and Nilson, 2006), suggesting nectins are also involved in modulating interfacial tension between homotypic and heterotypic contacts.

The heterophilic preference of the nectin-afadin adhesion system makes them suited to form checker-board-like patterns. Indeed, nectins have been found to mediate the checkboard-like patterns between nectin-1-expressing hair cells and nectin-3-expressing support cells in the auditory epithelium of mouse cochlea. Loss of either nectin-1 or nectin-3 disrupted the checkerboard pattern (Togashi et al., 2011). Although not directly linked to a patterning function, nectins also mediate heterotypic adhesion between distinct cell types in tissue development, including Sertoli cells (nectin-2 rich) and spermatids (nectin-3 rich) in mouse testes (Ozaki-Kuroda et al., 2002), and pigment cells (nectin-1 rich) and non-pigment cells (nectin-1 low) in eye ciliary epithelium (Inagaki et al., 2005).

Nectins also interact with cadherins to form adherens junction. Formation of nectin-based adhesion in trans can initialize adherens junction and recruit E-cadherin even before cadherin-binding in trans. The nectin-E-cadherin association is mediated by their corresponding binding partners, β-catenin, α-catenin, and afadin (Tachibana et al., 2000). Antagonizing nectin-based adhesion can disrupt E-cadherin adhesion in cultured cells in vitro (Honda et al., 2003). However, whether the heterophilic-prone nectin and homophilic-prone cadherin adhesion systems interact during cell sorting and patterning in vivo remains unanswered.

Eph/ephrin signaling

Eph receptors and their ephrin ligands are classic examples of how non-adhesion molecules can mediate cell sorting and tissue patterning. The Eph receptors represent the largest subfamily of receptor tyrosine kinases, with 16 receptors divided into two classes (Nguyen et al., 2016; Pasquale, 1997). EphA receptors primarily bind to ephrin-A ligands, attached to the cell surface through a glycosylphosphatidylinositol (GPI) anchor. EphB receptors bind to ephrin-B ligands, containing a transmembrane domain and a short cytoplasmic tail. In human, there are nine EphA receptors, five ephrin-A ligands, five EphB receptors, and three ephrin-B ligands. The receptor-ligand binding exhibits high degree of promiscuity within the same class, and some Eph receptors (e.g., EphA4 and EphB2) can also bind ligands of a different class (Pasquale, 2004).

When an Eph receptor-expressing cell encounters a ephrin ligand-expressing cell, the receptor-ligand binding triggers bi-directional signaling. Forward signaling in the Eph-expressing cell activates the kinase activity of the Eph receptor, and auto-phosphorylates the Tyr residue at the juxtamembrane region of the Eph receptor (Kullander et al., 2001; Zisch et al., 1998). In reverse signaling, Src family kinases phosphorylate the intracellular domain of ephrin-B ligands (Palmer et al., 2002). Both the forward and reverse signaling activate actomyosin contractility and increase interfacial tension along the heterotypic contact to lower cell adhesion or repel the cells (Calzolari et al., 2014; Kindberg et al., 2021). Eph receptor or ephrin ligand lacking the intracellular domain can still bind, but cannot faithfully mediate bidirectional signaling to mediate cell sorting and tissue separation (Mellitzer et al., 1999; Xu et al., 1999).

The Eph/ephrin signaling has a well-established function in generation of tissue boundaries in vivo (Cayuso et al., 2015). In zebrafish embryos, Eph receptors (EphA4, EphB2, EphB3) are expressed in odd-numbered rhombomere (r3 and r5), while ephrin receptors (e.g., ephrin B1, B2, B3) are expressed in even-numbered rhombomeres (r2, r4, r6). Blocking Eph receptor results in disruption of rhombomere boundaries (Xu et al., 1995). Mosaic over-expression of Eph receptors or ephrin ligands is sufficient to drive cell sorting to the odd- or even-numbered rhombomeres, respectively.(Xu et al., 1999). Similar mechanism also mediates somite boundary formation (Nakajima et al., 2006; Watanabe et al., 2009). In Xenopus embryos, Eph/ephrin signaling acts to separate early ectoderm and mesoderm, as well as axial and paraxial mesoderm (Fagotto et al., 2013; Rohani et al., 2014). In both contexts, cell types on both sides of the boundary express a combination of Eph receptors and ephrin ligands. When all ligand-receptor interactions are taken into account, only the interactions at the tissue boundary, but not those within the tissue, generate strong enough activity to repel the two cell types from each other (Rohani et al., 2014).

Additional molecules to generate heterotypic bias of myosin contractility

The example of Eph/ephrin signaling suggests molecular interactions in trans, if properly coupled with actomyosin cytoskeleton to modulate cortical tension at the interacting surface, can in principle generate bias in interfacial tension to mediate cell sorting and tissue patterning. In the Drosophila embryo, stripe patterns of pair-rule transcription factors (e.g., Eve and Runt) organize the tissue into distinct compartments with unique cell identities along the anterior-posterior axis. Eve and Runt instruct similar stripe patterns of the expression of leucine-rich repeat (LRR) receptors, Toll-2, Toll-6, and Toll-8. Heterophilic interaction between Toll-2 and Toll-6 or Toll-8 localize contractile myosin to the heterotypic contact between Toll-2 and Toll-6 or Toll-8 expressing cells. The high heterotypic interfacial tension, together with patterned expression of Toll receptors, drive convergent extension movements to extend Drosophila embryos along the anterior-posterior axis (Paré et al., 2014). In addition, another LRR receptor Tartan and its heterophilic binding partner Ten-m, are also expressed in distinct compartments and their interaction also creates high heterotypic interfacial tension to organize compartment boundaries (Paré et al., 2019). Together, the Toll receptors and the Tartan-Ten-m system provide high resolution spatial cues to direct polarized myosin contractility to guide tissue elongation (Paré et al., 2014, 2019).

Benefits and drawbacks of different molecular mechanisms to create adhesion bias across cell types

Adhesion-based sorting can be accomplished by variations in both the amount and identities of expressed adhesion molecules – often referred to as quantitative and qualitative differences in adhesivity, respectively (Duguay et al., 2003). Both types of adhesion bias play a role in sorting in vivo, and each conveys distinct features to the sorted patterns. Here we discuss these effects in a context where homophilic interactions are preferred over heterophilic ones, as is the case with cadherins.

With quantitative differences in the amount of adhesion molecule expression, used for example in positioning the Drosophila oocyte (Godt and Tepass, 1998), the strongest bonds (and lowest surface tension) will be between highly expressing cells, followed by the bonds between high and low expressing cells, and the weakest interactions will be between the cells with the lowest expression. Therefore, heterotypic interfacial tension is “intermediate” (Fig. 2, γaa < γab< γbb). There are several possible consequences of this effect. First, inadequate homotypic binding within the lowest-expressing layers could lead to poor structural integrity (Garcia et al., 2018; Harris et al., 2014). Second, stronger interactions at heterotypic interfaces will incur more extreme mechanical coupling between the separate compartments, as they do with homotypic interactions (Chihara and Nance, 2012). Finally, the intermediate strength of heterotypic interactions could have the effect of increased mixing of cells at the boundaries (Canty et al., 2017).

The complexity of the patterns that could be sustained by quantitative differences in adhesion molecule expression is limited by the number of “levels” of adhesion expression that could be robustly distinguished in the system. Sorting ability in this case is both (1) limited by the dynamic range of copy number and (2) susceptible to stochastic noise in gene expression (Raj and van Oudenaarden, 2008) and biochemical binding interactions (Fenz et al., 2017). Further, Cellular Potts models suggest that the rate of sorting is strongly dependent on the magnitude of the interfacial tension, such that aggregates with many levels of adhesion molecule expression (with each level having relatively low interfacial tension with its neighboring layers) will sort more slowly than those with just two levels (Zhang et al., 2011).

When patterning is mediated solely by qualitative differences in adhesion molecule expression, the different cell populations express incompatible adhesion molecules, and thus only undergo homotypic interactions. In this case, heterotypic interfacial tension is prohibitively high to promote interaction between the compartments (Fig. 2, γaa,γbb << γab). Sorting will be robust, but there is no mechanical coupling between the cell layers. Such a tissue may exhibit significant gaps between sorted compartments, as generally seen when cells are less adherent (Barua et al., 2017; Fagotto, 2014), leading to poor tissue integrity. Furthermore, while the number of different compartments can be equal to the number of different types of cell adhesion molecules existing in the genome, different cell populations cannot interact or spread over each other – limiting patterning complexity to that of separable blobs (Duguay et al., 2003). This approach can be advantageous in cases where clear boundaries between compartments are desired for the tissue’s physiological function, for example in the notochord and paraxial mesoderm, or for hindbrain rhombomeres.

An alternative strategy observed in developing embryos is sorting based on a combinatorial adhesion code, in which patterning is enabled by cell-type-specific combinations of several adhesion molecules. This approach is used in the developing zebrafish neural tube to stratify neural progenitor subtypes into stacked stripes (Tsai et al., 2020), as well as to establish distinct motor neuron pools in the chick spinal cord (Price et al., 2002). Unlike in the cases of patterning by solely quantitative or solely qualitative adhesion, a combinatorial adhesion code can attain equally preferable homotypic interfacial tension among various cell populations while maintaining lower but non-negligible heterotypic binding between distinct layers. (Fig. 2, γaa,γbb < γab). A combinatorial sorting mechanism thus maximizes both the potential for pattern complexity as well as the structural integrity of the tissue.

Patterning adhesion – how cell fate decisions and morphogen signaling shape the adhesion landscape

A prerequisite of adhesion-based patterning is existing differences in adhesion or interfacial tension within the cell population, creating the initial condition for adhesion-based self-organization to act upon. This is often accomplished by instructive cues such as morphogen signaling that trigger transcriptional responses, leading to cell fate specification and changes in expression level of interfacial tension regulators. In addition, morphogen signaling can also modulate cell adhesion through transcription-independent mechanisms. Understanding how morphogen signaling and cell fate decisions are coordinated with changes in cell adhesion is critical to reveal mechanisms of robustness in adhesion-based patterning.

Coordinating cell fate decisions with expression of interfacial tension regulators

Adhesion-based patterning requires two critical cellular decisions to be closely coupled. The first cell fate decision is defined by expression of one or multiple transcription factors that control a transcriptional program, leading to stable fate choice over alternative fates. The second adhesion decision is the repertoire of adhesion molecules that a cell chooses to express. Close coordination between the cell fate and adhesion decisions allows each cell type to utilize specific adhesion program to organize into spatial patterns. When the two decisions are well-coupled, adhesion-based self-organization can help refine the imprecise initial pattern generated by noisy morphogen signaling (Tsai et al., 2020), while mismatch of the two decisions will likely create errors in patterning. The two decisions can be coordinated by at least three different mechanisms. In one scenario, cells first commit to specific fates, and the cell fate regulating transcription factors then up- or down-regulate relevant adhesion molecules to create adhesion differences. In the second scenario, cells first express different amount of adhesion molecules independent from any cell fate decisions. Differential adhesion drives cell sorting, which expose cells to distinct signaling environments, eventually contributing to different cell types. In the third scenario, a common upstream signal (e.g., morphogens) triggers cell fate decisions and changes of adhesion expression through parallel programs. The precision of patterning relies on how well the two parallel programs are coupled in the same cell.

Several examples in neural development suggest cell fate decisions precede and regulate downstream adhesion programs. In the zebrafish spinal cord, olig2, a master regulator of the motor neuron progenitor (pMN) fate (Novitch et al., 2001; Park et al., 2002), is responsible for repressing Pcdh19 expression. The repression leads to lower heterotypic adhesion between pMN and other neighboring neural progenitor types with high Pcdh19 expression, and facilitate robust patterning (Tsai et al., 2020). In the chick spinal cord, Pax3 and Pax7, two dorsal transcription factors acting redundantly to repress ventral cell fates (Mansouri and Gruss, 1998; Moore et al., 2013), also repress Cdh7 to limit its expression at the middle portion of the spinal cord (Lin et al., 2016), although it is not established whether Cdh7 directly regulates cell sorting during tissue patterning. During hindbrain development, transcription factor Krox-20 specifies cell identity in rhombomeres r3 and r5 and also directly activates EphA4 expression, suggesting the sorting capacity of rhombomere cells is directly linked to their cell fate decisions (Giudicelli et al., 2001; Nonchev et al., 1996; Sham et al., 1993; Theil et al., 1998).

Two recent examples demonstrated how regulation of cortical contractility directs cell fate decisions. The zebrafish myocardium is composed of an inner trabecular layer and outer compact layer. Proliferation-induced crowding of the compact layer triggers cells with higher contractility to delaminate into the trabecular layer. The delamination event is sufficient to induce notch signaling between cells in the compact layer and the delaminated trabecular cell, which ultimately drives fate specification (Priya et al., 2020). In early mouse embryos, asymmetric division generates blastomeres with variable contractility. Cells with higher contractility sort internally, change subcellular YAP localization and specify into inner cell mass-like fate. (Maître et al., 2016)

An example of the parallel decisions between cell fate and adhesion comes from the balance between neural progenitor (NPCs) and neurons in the neural tube. NPCs form N-cadherin-based adherens junctions with neighboring NPCs to attach to the luminal surface of the ventricle. As the NPCs differentiate into neurons, the adherens junctions are disassembled, and neurons migrate away from the ventricular zone. Two transcription factors, Foxp2 and Foxp4, are expressed as motor neuron progenitors differentiate into motor neurons, and are necessary and sufficient to repress N-cadherin expression (Rousso et al., 2012). Loss of Foxp2 and Foxp4 blocks neuronal detachment, but does not prevent neuronal differentiation, resulting in differentiated neurons that remain in the ventricular zone and disrupt the overall tissue pattern.

Adhesion modulation by morphogen signaling

Cells exposed to morphogen signals can change their adhesion dynamics (Barone et al., 2017; Durdu et al., 2014; von der Hardt et al., 2007; Kim et al., 1998). This can be mediated by changes in transcriptional programs that are dependent or independent on cell fate decisions, as discussed in the previous section. In addition, morphogen signaling can directly modulate cell adhesiveness through non-transcriptional mechanisms.

Wnt11 can modulate cell adhesion through transcriptional independent protein-protein interactions. In the zebrafish embryo, cells exposed to Wnt11 recruit Frizzled7 and the intracellular mediator Dishevelled to the plasma membrane, which increases the cell-cell contact persistence through interactions with atypical cadherin Flamingo (Witzel et al., 2006). Interestingly, an opposite effect of Wnt11 on cell adhesion was observed during gastrulation of Xenopus embryos. Wnt11/Frz7 reduce cell adhesion during convergent extension by forming two separate complexes with C-cadherin and Pcdh8 through distinct Frz7 interacting domains to prevent lateral clustering of C-cadherin (Kraft et al., 2012).

During zebrafish gastrulation, a positive feedback between nodal signaling and cell adhesiveness regulates the specification of anterior axial mesoderm fate (Barone et al., 2017). Cells exposed to nodal signaling increases their adhesiveness and contact duration with neighboring cells. At the same time, cells with stable cell-cell contact are more competent to respond to nodal signaling and specify into anterior axial mesoderm cells. The positive feedback determines whether cells specify into anterior axial mesoderm cells or endoderm cells (Barone et al., 2017). In Xenopus embryos, activin, a nodal-like signaling protein, activates pcdh8 expression to reduce C-cadherin-dependent cell adhesion (Chen and Gumbiner, 2006).

Open questions

Armed with a good understanding of (1) how a broad array of cell patterns observed in vivo can arise simply from variations in interfacial tension between cells, (2) the diverse molecular strategies used to alter interfacial tension and drive patterning, and (3) how developmental signaling can initiate these sorting programs, we suggest that the field stands poised to answer a myriad of exciting questions.

Decoding the combinatorial logic of adhesion molecules

Much of the mechanistic work on adhesion-based sorting has taken reductive approaches in simplified systems – in particular, over-expressing one or two adhesion molecules at a time in ex vivo cultures of L cells, which lack E-cadherin and other adhesion proteins. In vivo, adhesion protein expression profiles are often much more complex, but the rules for sorting in the context of a combinatorial adhesion code remain unclear. Given the growing evidence for crosstalk and promiscuity among adhesion molecules, there is no reason to assume effects on sorting from individual molecular species will always add linearly when combined. To decode this combinatorial logic, it will be necessary to both (1) over-express increasingly complex mixtures of adhesion molecules in reductive in vitro systems, and (2) to directly measure homotypic and heterotypic interactions in a wide range of endogenous cells with various adhesion molecule expression profiles.

Limits of differential adhesion: Where biophysics meets developmental biology

One fundamental remaining question is: What are the design limits of differential adhesion? For example, how many different types of adhesion molecules are necessary to pattern an organism? To what extent is adhesion-mediated sorting sufficient for embryonic patterning? How sophisticated and robust can adhesion-based patterns be? Can an arbitrarily complex design be formed, given a sufficient diversity of adhesion molecule expression? With many organisms having hundreds of cell adhesion molecule species available, it seems as though patterns of any level of complexity could be achieved. It remains to be seen what diminishing returns there might be on the rate and accuracy of patterning for increasingly intricate designs. Engineering-based methods with programmable molecular toolkits may pave the way for this new frontier in a well-established field (Toda et al., 2019).

A second area of inquiry relates to the timescales of sorting. In vitro, aggregates typically take 1-2 days to sort completely, whereas many patterning events in vivo can occur within hours. Is the sorting rate observed in vitro sufficient to pattern the fast-developing embryo? What are the biophysical limitations on the rate of sorting, and how might the embryo overcome them? One possibility is that faster developmental patterning events are driven by greater differences in interfacial tension. One could also imagine that alternative developmental pathways position cells near their final configuration, and adhesion-mediated sorting then fine-tunes the pattern. Alternatively, differential adhesion could merely be a way to stabilize the patterned configuration once additional programming has positioned the cells. To distinguish between the possible mechanisms in different developmental contexts will require close interplay between theory and experiment.

Finally, it is becoming increasingly clear that two fundamental assumptions underlying the Differential Interfacial Tension Hypothesis – (1) that the tissue is fluid-like and (2) that cells are passively pulled by interfacial tension – may not hold in all tissues. For example, phase transitions between a fluid-like to a solid-like tissue state have been reported in the zebrafish blastoderm during epiboly (Petridou et al., 2021) and in the presomitic mesoderm during somitogenesis (Mongera et al., 2018). In addition, there are many contexts in development where cells actively migrate and deform the tissue, for example during convergent extension. Recent work on cell lines undergoing the epithelial-mesenchymal transition (EMT) suggests directional motility, friction, and jamming could potentially overcome forces of interfacial tension to alter the final cellular configuration (Pawlizak et al., 2015). More work is needed to establish how physical tissue properties impact the ability of cells to undergo adhesion-based sorting.

Integrating the final scale: How do the molecular kinetics of adhesion formation affect sorting?

In addition to tuning the diversity of expressed adhesion molecules, cells also exert tight control over the assembly of adhesions at the molecular level. Adhesion molecule expression levels alone convey only a fraction of the information needed to understand the connectivity between cells. What proportion of adhesion molecules are engaged to their partners and to the cytoskeleton at any one point in time? How do the kinetics of cadherin clustering, maturation, and turnover affect sorting? Fluorescent probes for the kinetics of adhesion assembly, in addition to the adhesion protein expression, will be crucial for directly linking molecular activity to tissue patterning in vivo.

Conclusion

Over the past century, incredible progress has been made to bring what began as a fascinating but mysterious observation that a mixture of unlike cells will spontaneously self-segregate, to what is now the Differential Interfacial Tension Hypothesis – allowing us to understand, explain, and predict increasingly complex adhesion-based patterning both in vitro and in vivo. Along the way, a vast collection of adhesion molecules has been discovered, and their diversity of mechanisms delineated. The great challenge that lies ahead is to understand how this orchestra of adhesion molecules work in concert with each other and with developmental programming to pattern the embryo.

Figure 3: Molecular toolkits to modulate interfacial tension.

Cadherin superfamily genes form homophilic in trans to modulate stability of homotypic contacts. Among them, classical cadherins can signal to actomyosin cytoskeleton to reduce interfacial tension, while clustered protocadherins can signal to increase interfacial tension at homotypic contact. Nectins form stronger heterophilic adhesion than homophilic adhesion, and preferentially stabilize heterotypic contact. Upon binding of Eph receptors and ephrin ligands, they signal to actomyosin cytoskeleton to increase interfacial tension and shrink the heterotypic contact.

Glossary

Homophilic adhesion

binding of the same type of adhesion molecules in trans

Heterophilic adhesion

binding of distinct types of adhesion molecules in trans

Homotypic adhesion

adhesion between two cells of the same type (i.e., express same type of adhesion molecules or same cell fate identity)

Heterotypic adhesion

adhesion between two cells of different types (i.e., express different types of adhesion molecules or distinct cell fate identity)

Interfacial tension

the force acting to contract the area of cell-cell contact between two neighboring cells