Studies on the molecular determinants of surface tension have revealed a role for both direct cell-cell cohesion, as mediated by cadherins [2], [3], and indirect cell-ECM adhesion, as mediated by the interaction of integrins and fibronectin [4]. Understanding with certainty how these adhesion systems combine to give rise not only to surface tension, but also to other tissue mechanical properties, has been impeded by the multiplicity of factors involved in regulating adhesion. Cells can interact through direct cell-cell effects alone, a mixture of cell-cell and cell-ECM effects, or strictly by cell-ECM effects. Furthermore, these modes of interaction depend on factors such as the level of expression of surface receptors and their cytoplasmic regulators, and the amount and type of extracellular matrix in the microenvironment. Whereas several recent studies have addressed the role of cadherins as determinants of tissue surface tension, few have directly explored the influence of varying both α5β1 receptor density and soluble fibronectin (sFn) concentration on tissue mechanical properties.

Measurements of the surface tension of aggregates of mouse fibroblast L-cells genetically engineered to express cadherin at various levels, and in which those cadherins are effectively the only source of cohesion, describe a linear relationship between surface tension and cadherin expression [2]. However, tissue cohesion is not exclusively mediated by cadherins. Other adhesion systems also contribute to mechanical properties. In a 3D tissue-like context, the ECM may act as a cellular cross-linker, indirectly gluing cells together through integrin-ECM bonds. This concept is supported by studies in which integrin α5-null Chinese hamster ovary cells transfected to express high levels of α5 integrin formed spherical aggregates only in the presence of exogenous fibronectin (Fn) [4], [5], [6]. Indeed, on a molecule per molecule basis, such aggregates were more cohesive than CHO aggregates expressing comparable levels of cadherin. In order to fully support aggregate formation and cohesion, adequate Fn matrix (FnMA) assembly is required. Accordingly, the relationship between the expression of α5β1, sFN concentration, and tissue surface tension may not be as simple as that described for cadherin-mediated cohesion and is a question of investigation in the current study.

Various methods have been developed to measure the surface tension of an aggregate. A common method used, tissue surface tensiometry (TST), applies an external force to an aggregate cultured under physiological conditions, and measures the resistance by the aggregate to the compressive force and changes in aggregate geometry. TST studies over the past 10 years have revealed that aggregates, upon compression, exhibit both visco-elastic solid and liquid-like behaviors, depending on the time-frame over which the measurements are made. In the short-term, compressed aggregates behave as elastic solids: cells in the interior of such aggregates deform upon application of a compressive force and remain deformed until the force is removed, whereupon, analogous to a spring, they rapidly assume their original shape. In contrast, cells in aggregates with viscoelastic-liquid properties, under comparable compressive force, take much longer to regain their original shape as cells slide past one another and overcome the arising friction [7], [8], [9]. These earlier studies revealed that embryonic tissues can not only change their mechanical properties, but can also exhibit a shift in behavior from liquid-like to elastic solid as a consequence of the amount of time in culture and their state of differentiation. For instance, fragments of embryonic chick myocardium cultured for up to 5 days exhibit purely liquid-like behavior. After 5 days in culture, however, the apparent surface tension markedly increases and aggregates then behave almost completely as elastic solids [9]. This change in apparent surface tension and shift in behavior has also been demonstrated during amphibian gastrulation and neurulation. For example, the surface tension of frog deep ectoderm just underlain by the archenteron roof is significantly greater than that of not-yet-underlain deep ectoderm [10]. This suggests that a differentiation-associated shift in surface tension may be responsible for the envelopment of the neural plate.

In a 3D aggregate, the ECM is not only an important source of cell-substratum adhesion for cell motility and shape change, but may also act as a cellular cross-linker, indirectly “gluing” cells together through integrin-ECM bonds. Previous studies showed that Chinese hamster ovary (CHO) cells, transfected to express high levels of α5 integrin, formed spherical aggregates only in the presence of soluble Fn [4]. Aggregate formation was also shown to be dependent on the ability of cells to assemble FN into an insoluble matrix [5]. Interestingly, integrin-based cohesion was found to be, on a molecule per molecule basis, more effective at mediating strong tissue cohesion than N-cadherin, a more classic adhesion system that is more commonly identified as a significant source of cohesion of cells in tissues. In those studies, measurement of aggregate cohesion was performed under conditions in which α5 integrin expression and the amount of sFn were kept constant. Accordingly, nearly all aggregates in any particular data set displayed liquid-like behavior. Here we varied both α5 integrin expression levels and the amount of available sFn. In so doing, we observed differences in FNMA, aggregate cohesion, and shifts in mechanical properties from liquid to elastic solid, such that in some data sets, nearly all aggregates displayed elastic solid-like behavior. We assessed this shift in material properties by determining whether aggregates, when compressed, obeyed Hooke's Law.

We found that when α5 expression is increased from low to moderate levels, surface tension increases from 5.5 dynes/cm to 10.4 dynes/cm, but further increase in α5 receptor expression paradoxically caused a drop in surface tension back to those measured for low levels of α5 expression. This was associated with a biphasic transition from viscoelastic-liquid to viscoelastic-solid behavior such that aggregates behaved as viscoelastic-solids for moderate α5 expression levels and liquid-like for either higher or lower expression values. These data are in sharp contrast to those observed for cadherin-based cell-cell cohesion, where studies have demonstrated a near-perfect correlation between receptor expression and surface tension and no associated transitions in tissue mechanical properties. Interestingly, the biphasic nature of the relationship in 3D tissues between receptor expression and surface tension appears to mirror the relationship in 2D between integrin receptor expression and cell migration [15]. That is, an optimal level of integrin expression is required in order to maximize cell motility. In the current study, aggregates expressing either low or high levels of α5 integrin exhibited liquid-like behavior, indicative of an actively motile cell population, whereas cells expressing intermediate levels of integrin became essentially locked in place, behaving as elastic solids. For cells to become locked in place, they must lose the ability to detach from one-another or from the matrix. Accordingly, cells must no longer be able to detach from the matrix by release of integrin-matrix bonds [16] or by fibril breakage [17]. It is likely that in aggregates displaying elastic solid behavior, the fibronectin matrix is stiffer than that of aggregates displaying liquid-like behavior. Matrix stiffness has been shown to significantly influence cell movement in 3D in vitro [18] and in vivo [19].

The biphasic nature of the interaction between α5 integrin expression and sFn levels can be explained by considering the role of sFn in providing the means by which cells become cross-linked through integrin-ECM interaction. Previous studies exploring integrin-mediated tissue cohesion have demonstrated a critical role for fibronectin matrix assembly in generating strong tissue cohesion [4], [5]. In this study, we varied the amount of sFn and showed that at 30 µg/ml, irrespective of α5 expression levels, aggregates behaved predominantly as liquids, whereas increasing the fibronectin concentration 10-fold rendered aggregates, even those expressing low levels of α5 integrin, almost completely elastic. Moreover, we showed that when aggregates composed of 25,000 or 50,000 cells are incubated in hanging drop culture, those containing 25,000 cells tend to compact over a 5-day period irrespective of the level of integrin expression. In contrast, those containing 50,000 cells initially compact over 3-days, but then undergo a significant decompaction as sFn becomes depleted from the microenvironment. This effect was more pronounced for cells expressing higher levels of α5 integrin, and corresponded closely to the time-frame of sFn depletion from the medium in the hanging drops.

During embryonic development, both receptor expression and ligand levels change. Studies have shown that embryonic tissues isolated at different stages of development exhibit significantly different mechanical properties, and that tissues isolated from earlier stages of development can undergo similar changes if extirpated and grown in culture [9]. For example, the cohesivity of chick embryonic heart aggregates was shown to be nearly constant for 4 days in culture, whereupon it nearly tripled when aggregates were incubated for another 3 days [9]. This increase in cohesion was predominantly due to a transition from liquid to elastic solid behavior by aggregates. Changes in mechanical properties can result in phase reversals between tissues, as well as in the elimination of boundaries between different tissue types, and are thought to occur primarily due to changes in the level of fibronectin in the microenvironment [9], [20], [21], [22], [23]. Earlier studies showed that when heart myocyte aggregates, one pre-incubated in normal medium, the other in a urea extract of a fibroblast-conditioned medium, were fused in hanging drop culture, the one pre-incubated in fibroblast-conditioned medium became enveloped by the control aggregate [22], and was blocked by inclusion of GRGDSP peptide [24]. This type of segregation behavior typically occurs between tissues of differing cohesivity [9]. In heart myocytes, envelopment of one tissue by the other was plausibly mediated by the presence of fibronectin in the urea extract of the fibroblast-conditioned medium [25]. By cross-linking cells together, fibronectin would effectively increase the cohesion of the treated myocytes, causing them to internalize relative to their untreated counterparts. Other studies reveal that fibronectin deposition can also stabilize the interface between tissues and prevents invasion of one tissue type by another [24]. More recent studies have shown that ECM formation can become restricted to tissue surfaces and interfaces by a mechanism utilizing surface de-repression of integrin-dependent inhibition of matrix assembly and that this enables self-organization and maintenance of tissue boundaries [26].

Our studies reveal that it may not simply be a matter of ECM deposition, but a quantitative interplay between fibronectin and integrins that collectively regulate tissue mechanical properties. The exact mechanism underlying how this interaction can directly influence tissue properties likely involves integrin activation [27] and the assembly of sFn into an insoluble matrix [28]. Indeed, previous studies have shown that fibronectin matrix assembly (FnMA) is necessary for imparting strong tissue cohesion to aggregates whose primary cohesion mechanism is integrin-based, since CHO cells transfected to express a chimeric integrin which does not support FNMA resulted in flat sheets of cells that failed to round up into spheres [5]. Additionally, fibronectin deposition by these aggregates was punctate. Fibers, when present, were short and tended to extend only locally. We showed that FNMA by aggregates expressing mid levels of α5β1 integrin generated a rich matrix, with fibers extending from cell-cell throughout the aggregate. FNMA by aggregates of cells expressing high levels of α5 integrin was, in contrast, reduced and tended to generate pockets of matrix assembly and shorter fibers. This may explain why such aggregates tend to be of lower cohesion and more liquid-like than aggregates in which the matrix is better developed. A matrix composed of shorter fibers may facilitate cell migration by effectively reducing the apparent viscosity of the tissue as cells may be less globally interconnected.

The mechanism by which a liquid-solid transition emerges appears to be related to that previously described in other problems, for example in polymer pinning [29], and in granular jamming transitions [30]. In these systems, as lengths of polymers or granular stress chains grow, connections become entangled and the scale over which rearrangements can occur becomes vanishingly small. We believe that the same mechanism is likely at work in our cellular system: when the density and length of the fibronectin network exceeds a critical value, cells can no longer move with respect to one another. Since cellular rearrangements are intrinsic to the liquid state, this is manifested as a transition from liquid-like to solid-like tissue rheology. Our study demonstrates that α5β1 interaction can promote phase transition from liquid-like to elastic solid-like states. Increased cell-matrix engagement to a 3D matrix has been demonstrated to significantly influence viscoelastic properties of prostate cancer cells embedded within 3D matrices [31]. The observation that α5β1 integrin is often down-regulated in metastatic cancers, and that re-expression has been shown to rescue a transformed phenotype [32], is consistent with our view that matrix-based transitions between liquid and solid states may play an important role in specifying whether normal or cancer cells remain locked in place or are free to move and interact with other cell types.

Differences in tissue cohesion have been demonstrated to play a critical role in not only establishing compartments and boundaries between tissues, but also in specifying the spatial relationships that give rise to tissue architecture and to the overall body plan. The Differential Adhesion Hypothesis (DAH) was formulated to explain how the cell rearrangement and reorganization observed during the process of tissue self-assembly can be driven by tissue mechanical properties, i.e. cohesion, adhesion and surface tension [33], [34]. While the process is inherently a physical one, it has molecular roots traceable largely to the activities of adhesion systems mediating both cell-cell and cell-ECM interactions [23]. The contribution of cadherins to tissue cohesion, cell rearrangement, and boundary formation during embryonic development is well known. From a molecular perspective, differences in the level of expression of a cell adhesion molecule (cadherin) have been experimentally shown to cause cell sorting [2]. Cells interacting directly through cadherin-cadherin bonds re-arrange to reduce the adhesive free energy of the system. It is as yet unclear how differences in integrin expression levels could drive the cell-sorting process. Several physical models have been advanced to address this question. Matrix-driven translocation [35], mesenchymal condensation [36], and percolation networks [37], have been proposed to explain ECM-based sorting, particularly between mesenchymal tissues (reviewed in [38]). We show that both CHO-X5C5 M and HH sort out from CHO-P3 to assume an internal position. This behavior is in general agreement with the DAH, which would predict that CHO-X5C5 M and HH must be more cohesive than CHO-P3. However, since CHO-P3 cells do not form spherical aggregates, it was not possible to confirm this by directly measuring aggregate cohesion. Accordingly, it is possible that sorting between the α5-null and α5-expressing CHO cells could be mediated by one or more of these mechanisms.

As we demonstrate in Table 2, CHO-X5C5 M and HH have different surface tensions, HH higher than M. When mixed with CHO-P3 cells, CHO-X5C5 HH sort out more efficiently than CHO-X5C5 M cells and in a sFn-dependent manner. These data are relevant since, at least in some cancer settings, down-regulation of α5β1 integrin is a hallmark of malignancy [39]. Accordingly, cancer cells in which α5β1 is down-regulated may be excluded from those expressing the receptor and become effectively a different cell population that may become squeezed away from the tumor mass, assuming a more peripheral position. This could, in principle, place such cells in contact with stromal cells into which they can then invade.