Tissue surface tensions guide in vitro self-assembly of rodent pancreatic islet cells

Abstract

The organization of endocrine cells in pancreatic islets is established through a series of morphogenetic events involving cell sorting, migration, and re-aggregation processes for which intercellular adhesion is thought to play a central role. In animals, these morphogenetic events result in an islet topology in which insulin-secreting cells form the core, while glucagon, somatostatin, and pancreatic polypeptide-secreting cells segregate to the periphery. Isolated pancreatic islet cells self-assemble in vitro into pseudoislets with the same cell type organization as native islets. It is widely held that differential adhesion between cells of the pancreatic islets generates this specific topology. However, this differential adhesion has never been rigorously quantified. In this manuscript, we use tissue surface tensiometry to measure the cohesivity of spherical aggregates from three immortalized mouse pancreatic islet cell lines. We show that, as predicted by the differential adhesion hypothesis, aggregates of the internally segregating INS-1 and MIN6 beta-cell lines are substantially more cohesive than those of the externally segregating α-TC line. Furthermore, we show that forced overexpression of P-cadherin by α-TC cells significantly perturbs the sorting process. Collectively, the data indicate that differential adhesion can drive the in vitro organization of immortalized rodent pancreatic islet cells. Developmental Dynamics 236:2039–2049, 2007. © 2007 Wiley-Liss, Inc.

INTRODUCTION

The organization of endocrine cells in islets of Langerhans is established through a series of morphogenetic events involving cell sorting, cell migration, and cell re-aggregation processes for which cell adhesion is thought to be important (Esni et al.,1999). In animal islets, insulin-producing beta-cells form the core, while glucagon, somatostatin, and pancreatic polypeptide (PP) -producing cells segregate to the periphery. This topology is highly conserved among amphibians (Putti et al.,1997), rodents (Wieczorek et al.,1998), and nonhuman primates (Wieczorek et al.,1998; Sujatha et al.,2004) and is preserved in malignant islet cell tumors (Bordi et al.,1985). Its disruption is strongly associated with the onset of diabetes (Baetens et al.,1978; Gomez Dumm et al.,1990).

Various studies using isolated native pancreatic islet cells show that cells have the ability to self-assemble in vitro into clusters with exactly the same cellular arrangement as in native islets (Rouiller et al.,1990). Studies exploring the molecular mechanism underlying this self-assembly behavior have postulated that differential adhesion is the driving force underlying this process (Cirulli et al.,1994). However, this differential adhesion has never been rigorously quantified and directly correlated with self-assembly behavior. Rather, studies have explored differential expression patterns of various cell surface adhesion molecules, such as cadherins (Rouiller et al.,1991), and neural cell adhesion molecule (N-CAM; Cirulli et al.,1994), or have assessed aggregation rates (Rouiller et al.,1990) as ad hoc measurements of differential adhesion. Such methods, while demonstrating a good correlation between differential adhesion and sorting behavior, do not measure the key parameter underlying differential adhesion, namely, differences in intercellular binding energy between the different cell populations. Aggregation rates (Roth and Weston,1967; Roth,1968; Nose et al.,1988) are only reflective of the initiation of adhesions between suspended cells. Such measurements present at least three obstacles to their use to measure intercellular binding energy: (1) the rate of initial cell attachment cannot measure the strengths of mature, physiologically relevant adhesions (Moyer and Steinberg,1976); (2) rate processes are limited not by the specific free energy of reaction (in this case the strength of adhesion) but by entirely unrelated activation energies (Glasstone et al.,1941); (3) the shear force used in comparing the aggregation rates of different cell populations may itself create the functional equivalent of an activation energy barrier preventing the aggregation of one population of mutually adhesive cells while permitting that of another and thus creating a misleading impression of adhesive specificity (Duguay et al.,2003). To rigorously measure intercellular binding energy, measurements rooted in thermodynamics must be used.

Tissue surface tensiometry (TST) quantifies binding energy between cells in three-dimensional (3D) aggregates under physiological conditions. The biophysical concepts underlying TST have been previously described in detail (Foty et al.,1994,1996). The method is based on the observation that mutually cohesive cells, if maintained in shaking culture, spontaneously assemble into clusters. Over time, these clusters “round-up” to form spheres. This “rounding-up” behavior mimics the behavior characteristic of liquid systems. Aggregate cohesivity is measured by compressing spherical aggregates between parallel plates in a custom-designed tissue surface tensiometer (Supplementary Figure S1, which can be viewed at http://www.interscience.wiley.com/jpages/1058-8388/suppmat). The same equation used to measure the surface tension of a liquid droplet is used to measure intercellular binding energy of 3D tissue-like spherical aggregates. Previous studies have shown that TST measurements (1) predict how two groups of embryonic cells will interact with one another (Foty et al.,1994,1996; Davis et al.,1997); (2) correlate inversely with the degree to which aggregates spread upon biomaterials (Ryan et al.,2001); (3) correlate inversely with invasive potential of lung tumors (Foty and Steinberg,1997), fibrosarcomas (Foty et al.,1998), and brain tumors (Winters et al.,2005); and (4) can be altered not only through direct manipulation of cadherin-based intercellular cohesion (Foty et al.,1999; Duguay et al.,2003), but also by manipulation of extracellular matrix molecules such as fibronectin (Robinson et al.,2003,2004). For comprehensive reviews, see Foty and Steinberg (2004,2005).

The current study explores whether the in vitro segregation of immortalized rodent beta- and non–beta-cells can be explained by differential adhesion. In 1964, Steinberg postulated the differential adhesion hypothesis (DAH) as a physical explanation of Holtfreter's tissue affinities (Holtfreter,1944), a term describing the spontaneous liquid-like tissue segregation, mutual envelopment and sorting-out behaviors of embryonic tissues and cells (Steinberg,1964). The DAH proposes that tissue affinities arise from tissue surface tensions that in turn arise from differences in intercellular adhesiveness. The DAH has recently been directly evaluated and confirmed (Foty and Steinberg,2005). Our earlier measurements of tissue surface tensions have shown that, without exception, a tissue of lower surface tension will always envelop one of higher surface tension (Foty et al.,1994,1996; Duguay et al.,2003; Foty and Steinberg,2005). Therefore, in pancreatic islets, the DAH predicts that aggregates of the internally segregating insulin-secreting beta-cells will have higher surface tension than those of the externally segregating glucagon-secreting cells. The DAH also predicts that altering the relative cohesion of the cell populations will alter their sorting behavior.

We generated multicellular spherical aggregates of three well-characterized, immortalized cell lines of alpha- and beta-cell origin and measured aggregate cohesivity by TST. We then determined whether aggregate cohesivity correlated with sorting behavior. We also explored the expression of E-cadherin to assess whether differential expression or function correlates with differential adhesion of the different cell populations. We also altered cohesivity of the alpha-cell population and assessed effects on sorting behavior.

RESULTS

Confirmation of In Vivo Arrangement of Mouse Pancreatic Islet Cells

The organization of endocrine cells in islets of Langerhans is evolutionarily conserved. Insulin-secreting beta-cells form the core of the islet, whereas glucagon-secreting cells assume an external position. Here, we confirm that islet cell distribution in mouse pancreatic islets is indeed concentric. Figure 1A represents an islet stained for insulin and glucagon. Note that glucagon-expressing cells are localized at the periphery of the islet. In contrast, insulin-secreting beta-cells are located centrally.

Figure 1. Mouse pancreatic islet topology and sorting of immortalized pancreatic alpha- and beta-cell lines. Mouse pancreatic tissue sections were incubated in antibodies directed against glucagon or insulin. A: Note the concentric arrangement of glucagon-positive cells enveloping a core of insulin-secreting cells. B,C: The α-TC cells, here labeled red, were mixed with either INS-1 (B) or MIN6 (C) cells, labeled green. In both cases, α-TC cells adopted an external position relative to INS-1 or MIN6. This arrangement mimics that of normal pancreatic islets observed in A.

In Vitro Cell Sorting of Immortalized Pancreatic Beta and Alpha Cells

We performed sorting assays to determine whether the INS-1/α-TC or MIN6/α-TC cells sorted out in vitro in the same pattern observed in native islets. Similar to in vivo islets, mixtures of INS-1/α-TC (Fig. 1B) and MIN6/α-TC (Fig. 1C) sorted out in both instances with the beta-cells forming the core and the alpha-cells forming the periphery of the pseudoislet.

Measurement of Cohesivity and Confirmation of Liquidity of INS-1, MIN6, and α-TC Cell Aggregates

The differential adhesion hypothesis predicts that, because INS-1 or MIN6 cells form the core of the pseudoislet, aggregates of these cell lines will be more cohesive than those of the α-TC cell line. To test this hypothesis, we used TST to measure the cohesivity of aggregates of INS-1, MIN6, and α-TC cells. First, we confirmed that aggregates behaved in a liquid-like manner by demonstrating that aggregates satisfied five conditions: (1) aggregates formed spheres; (2) aggregate cohesivity (σ) was independent of the applied force; (3) σ was independent of aggregate size; (4) aggregates remained liquid within the time frame of the experiment; and (5) aggregates did not obey Hooke's Law; that is, the calculated σ remained constant in response to different degrees of compression. In such aggregates, the ratio of σ₂/σ₁ was nearly equal to 1 and was less than the ratio of the force applied at each successive compression (F₂/F₁). If aggregates obeyed Hooke's Law, the ratios of σ₂/σ₁ and F₂/F₁ would be nearly identical. Table 1 shows that for aggregates of INS-1, MIN6, and α-TC cells, the mean surface tension values measured after compressions 1 and 2 were statistically identical when compared by an unpaired t-test.

Table 1. TST Measurements and Confirmation of Aggregate Liquiditya

	σ1(dynes/cm) ± SEM n = 10	σ2(dynes/cm) ± SEM n = 10	P value σ1 vs. σ2	σ1,2(dynes/cm) ± SEM n = 20	σ2/σ1	F2/F1	P value σ2/σ1 and F2/F1
a Measurement of aggregate cohesivity by tissue surface tensiometry (TST) shows that aggregates of α-TC cells are significantly less cohesive than those of either INS-1 or MIN6. For each aggregate type, the mean surface tension values measured in the first compression (σ1) are similar to those measured after the second compression (σ2). For the three cell lines studied, the ratio of F2/F1 was significantly greater than σ2/σ1, further confirming that the aggregates behaved as liquid systems and not as elastic solids.
σ-TC	0.60 ± 0.07	0.66 ± 0.06	0.29	0.63 ± 0.04	1.11 ± 0.04	1.85 ± 0.11	<0.001
INS-1	2.50 ± 0.14	2.54 ± 0.15	0.41	2.52 ± 0.10	1.04 ± 0.07	1.87 ± 0.08	<0.001
MIN6	1.29 ± 0.11	1.41 ± 0.09	0.22	1.35 ± 0.07	1.12 ± 0.05	1.81 ± 0.06	<0.001

In this study, we applied two successive compressions, the second greater than the first, and calculated sigma after each one. For a complete data set, the measured force at equilibrium (Feq) ranged anywhere from 0.25 to 1.00 mg, or roughly fourfold. We plotted the range of Feq values against their corresponding σ values. For aggregates of MIN6 cells, here used as a representative example, the correlation coefficient (r² = 0.012) indicated little relationship between the two parameters. Furthermore, a P value of 0.594 indicated that the slope does not deviate from 0 (Fig. 2A). The demonstration that Feq is independent of σ over a fairly broad range further confirms their liquid-like properties.

Figure 2. Relationship between aggregate surface tension and force at equilibrium, volume, or time in culture for immortalized cell lines. A: For MIN6 cells, surface tension remained constant over a large range of force. Aggregates of α-TC and INS-1 cells behaved in a similar manner (data not shown). B: For all three-cell lines, the surface tension of aggregates remained relatively constant over a sixfold range in volume. C: Surface tension of aggregates of the cell lines remained constant over 2–6 days in culture.

We compared the ratios of σ₂/σ₁ and F₂/F₁to ensure that these aggregates did not obey Hooke's law, as they would if they behaved as elastic solids. As demonstrated in Table 1, the ratio of σ₂/σ₁ does indeed approach 1. Moreover, the ratio of F₂/F₁ was significantly greater than σ₂/σ₁ (unpaired t-test; P < 0.05), further confirming that these aggregates do not obey Hooke's law and in fact behave as liquids. We also explored the relationship of σ and aggregate volume. The surface tension of aggregates remained relatively constant over a volume range of 10 to 60 μ³ (Fig. 2B). Linear regression analysis generated correlation coefficient values well below those required to establish a relationship between these two parameters (r² = 0.013, 0.0028, and 0.046 for INS-1, α-TC, and MIN6, respectively). The slope of the regression lines was also not significantly different from zero (P = 0.75, 0.88, and 0.61 for INS-1, α-TC, and MIN6, respectively). We also demonstrated no significant difference in cohesivity for aggregates cultured for 2–6 days, typically the time course required for spheroid formation and acquisition of TST measurements from each batch of aggregates (Fig. 2C). Here again, regression analysis generated low correlation coefficients (r² = 0.010, 0.001, 0.0003 for INS-1, MIN6, and α-TC, respectively), and P values for the slope of the regression lines that did not differ from zero (P = 0.78, 0.94, and 0.97 for INS-1, MIN6, and α-TC, respectively). These data clearly demonstrate that aggregates of the three cell lines, when placed under long-term compression, behave as liquid systems and confirm the validity of the cohesivity measurements for the cell lines. Our previous study using various chick embryonic tissues more fully characterized other biophysical properties, including viscosity and elasticity. Neither of these properties, however, have been shown to contribute directly to sorting behavior (Foty et al.,1996; Forgacs et al.,1998). As predicted by the DAH, aggregates of the internally segregating INS-1 and MIN6 cell lines are of greater cohesivity (2.52 ± 0.10 dynes/cm, 1.35 ± 0.07 dynes/cm), respectively, than those of externally segregating α-TC cell line (0.63 ± 0.04 dynes/cm). These data confirm that TST measurements correlate perfectly with the in vitro sorting pattern of these pancreatic islet cell lines.

E-cadherin Expression Correlates With Aggregate Cohesivity

Previous studies have shown that in native islets, E-cadherin expression is uniform (Cirulli et al.,1994). We explored whether this was the case in immortalized cell lines. Figure 3A shows that, whereas α-TC cells express barely detectable levels of E-cadherin mRNA, MIN6 and INS-1 cells express relatively high levels. We then assessed E-cadherin protein expression by immunofluorescence microscopy. Figure 3B shows that α-TC cells do not express any detectable levels of E-cadherin, whereas surface expression of the protein at sites of cell–cell contact appears to be at higher levels in INS-1 (Fig. 3D) than in MIN6 cells (Fig. 3C). We quantified E-cadherin expression by immunoblot analysis. Figure 4A is an immunoblot showing a similar E-cadherin expression pattern as in Figure 3. We quantified expression levels by assaying five separate lysates and expressing the data as a function of actin expression, here used as a loading control. Optical density was compared by analysis of variance (ANOVA) and Neuman–Keul's multiple comparisons test. Figure 4B confirms that α-TC cells express barely detectable levels of E-cadherin protein, whereas MIN-6 and INS-1 express E-cadherin at significantly different levels, INS-1 greater than MIN6. When E-cadherin expression is plotted as a function of aggregate surface tension (σ), the relationship is linear with a correlation coefficient r² of 0.9874, indicating a strong relationship between E-cadherin expression and aggregate cohesivity (Fig. 4C).

Figure 3. Reverse transcriptase-polymerase chain reaction and immunofluorescence analysis of E-cadherin expression by immortalized pancreatic islet cells. A: E-cadherin mRNA expression is barely detectable in α-TC cells as compared with MIN6 and INS-1. A similar pattern was observed for surface expression of E-cadherin protein by immunofluorescence microscopy. B: α-TC cells do not express E-cadherin. C,D: Both MIN 6 and INS-1 cells express E-cadherin, MIN6 (C) at lower levels than INS-1 (D).

Figure 4. A: Immunoblot analysis of E-cadherin expression by immortalized cells. The α-TC cells do not express detectable levels of E-cadherin protein. MIN6 and INS-1 cells express different levels, MIN6 lower than INS-1. B: Semiquantitative densitometric analysis of E-cadherin expression by immortalized islet cells can be seen. We first determined that actin expression does not differ between these cell lines (analysis of variance [ANOVA], P = 0.1448) and could, therefore, be used as a loading control. E-cadherin expression was found to be absent in α-TC cells and significantly higher in INS-1 than in MIN6 (ANOVA, Neuman–Keul's multiple comparison test, P < 0.01). C: A direct relationship exists between E-cadherin expression and aggregate cohesivity as measured by tissue surface tensiometry (TST).

P-Cadherin Expression by α-TC Cells Alters Sorting Behavior

Neither α-TC nor MIN6 cells express P-cadherin (Fig. 5A). According to the DAH, transfection of P-cadherin into either one of the populations should increase the cohesivity of that population. As can be seen in Figure 5B, untransfected α-TC cells completely envelop MIN6 cells. Transfection of α-TC cells with P-cadherin resulted in an equilibrium configuration of incomplete envelopment of MIN6 cells by α-TC-Pcad (5C). Incubating the α-TC.Pcad/MIN6 cells in the presence of the P-cadherin function-blocking antibody pCD-1, resulted in normal sorting out with the α-TC.Pcad cells adopting an external position relative to MIN6 (5D), suggesting that the altered sorting behavior observed in Figure 5C is due to increased P-cadherin expression.

Figure 5. P-cadherin transfection of α-TC cells disrupts sorting behavior. Neither α-TC nor MIN6 cells express P-cadherin. A: Transfecting α-TC cells with a P-cadherin cDNA results in a marked expression of P-cadherin as demonstrated by immunoblot analysis. B: Sorting assays between untransfected α-TC and MIN6 cells result in complete envelopment of MIN6 by α-TC cells. C: Transfecting α-TC cells with P-cadherin resulted in an equilibrium configuration of incomplete envelopment of MIN6 cells by α-TC-Pcad. D: Incubating α-TC.Pcad/MIN6 in the presence of a P-cadherin function-blocking antibody results in complete envelopment of MIN6 by α-TC.Pcad.

DISCUSSION

Many studies have invoked the concept of differential adhesion to explain cell sorting and segregation behavior in various model systems. For example, a cell sorting process is thought to be involved in early neurogenesis in Xenopus. Neural tube formation and patterning in Xenopus is associated with differential expression of adhesion molecules, including R-cadherin and cadherin 6. Within the ventricular cells of the telencephalon, cell-labeling studies have demonstrated the existence of two neighboring cell-lineage–restricted compartments, the lateral ganglionic eminence (lge) and the presumptive cerebral cortex (ctx). Cells belonging to the same compartment mix freely with each other, whereas cells belonging to adjacent compartments do not mix and a smooth boundary forms between them (Inoue et al.,2001). Such cell lineage restricted compartments maintain embryonic brain organization by preventing cells, once patterned, from randomly intermingling during development. In early chick limb development, anterior and posterior cells sort out from one another to form alternating anterior and posterior stripes of cells that extend distally along the proximal–distal axis. Anterior and posterior limb bud cells also sort out from one another in monolayer culture, providing evidence for the existence of position-specific differences in cell–cell affinity in the anterior/posterior (A/P) axis (Yajima et al.,1999). Later in limb bud development, cells destined to become chondrocytes sort out from undifferentiated cells. The establishment of cell–cell contacts is thought to play an active role in the induction of mesenchymal cell condensation (Tavella et al.,1994). DE-cadherin is thought to mediate oocyte positioning during follicle biogenesis in Drosophila. The position of the oocyte is determined by the position of DE-cadherin–expressing follicle cells to which the oocyte attaches itself selectively. The oocyte preferentially contacts those cells that express higher levels of DE-cadherin (Godt and Tepass,1998). When DE-cadherin is removed from either the germline or the posterior follicle cells, the oocyte becomes misplaced (Gonzalez-Reyes and St Johnston,1998). Such studies demonstrate a clear relationship between differential adhesion and cell sorting.

During development of the endocrine pancreas, differentiation of the various cell types occurs before their re-arrangement into histologically correct islets (Gittes and Rutter,1992). Initially, the dorsal and ventral buds of the vertebrate pancreas develop as an evagination from the primitive gut epithelium (Slack,1995). The epithelial sheet of cells spreads out into the overlying mesenchyme. These epithelial cells expand and give rise to the mature pancreatic cell types that eventually sort out from the duct epithelium and begin to aggregate into islets of Langerhans. In rodents and other animals, islet cells are arranged in a nonrandom manner; insulin-producing beta-cells form the core, while glucagon, somatostatin, and PP-producing cells segregate to the periphery (Slack,1995). Perturbation of this architecture in certain forms of diabetes suggests that this organization is essential for the function of this microorgan (Baetens et al.,1978; Unger and Orci,1981; Gomez Dumm et al.,1990). Studies in which rat pancreatic islets were stained with antibodies against E-cadherin showed a uniform expression pattern between alpha- and beta-cells (Rouiller et al.,1991). Interfering with endogenous E-cadherin activity in beta-cells during pancreatic development, however, has been shown to markedly influence islet topology. Dahl et al. introduced a dominant-negative mutant of mouse E-cadherin in pancreatic beta-cells in transgenic mice. Expression of the truncated E-cadherin receptor displaced both E- and N-cadherin from pancreatic beta-cells and perturbed the clustering of endocrine cells into islets. Instead of clustering together, transgenic beta-cells were found dispersed in the pancreas as individual cells, while alpha-cells selectively aggregated into islet-like clusters (Dahl et al.,1996). Clearly, E-cadherin must make some contribution to the overall cohesive balance between alpha- and beta-cells, because its disruption perturbed the in vivo sorting process. These in vivo data were recapitulated in vitro using native cultured rat islet cells. The expression levels of two key mediators of cell–cell cohesion, namely E-cadherin and N-CAM, were examined (Rouiller et al.,1990,1991). Here too, it appears that E-cadherin mediates most of the Ca²⁺-dependent cell adhesion between all cell-types in the islets. Ca²⁺-independent CAMs, such as N-CAM, are thought to promote the segregation of non–beta-cells from beta-cells (Rouiller et al.,1991). However, the picture is complicated because additional cadherins, R- and N-cadherin, are also expressed in islets (Moller et al.,1992; Hutton et al.,1993).

This study showed that E-cadherin expression by immortalized pancreatic islet cells, unlike that of native islets, is nonuniform. The α-TC cells do not appear to express E-cadherin in any measurable quantity, whereas MIN6 and INS-1 cells express E-cadherin at levels that directly correlate with aggregate cohesivity (Fig. 5C). That our results do not agree with those demonstrated for the in vivo expression patterns in native islets suggests that, from an adhesion perspective, the immortalized cells may not resemble the cells from which they were originally derived. Indeed, these immortalized cells at one point underwent malignant transformation and should not be expected to completely recapitulate their in vivo adhesion molecule expression pattern. Of interest, however, the immortalized cells retained their ability to self-assemble in vitro into a configuration similar to that adopted in vivo and in a manner similar to freshly isolated islet cells, suggesting that a common adhesion-based mechanism may be operating to drive the sorting process, both in vitro and in vivo. The current study rigorously quantifies the difference in intercellular adhesion that drives the sorting process, by measuring the cohesive properties of discreet phases of each cell type. We show that aggregate cohesivity correlates perfectly with whether cells adopt an internal or external position when co-mixed in hanging drop culture.

To confirm that differential adhesion is the driving force underlying the sorting process, we first attempted to generate an E-cadherin–expressing α-TC cell line. According to the DAH, increasing the cohesivity of α-TC cells to a level higher than that of either of the beta-cell lines would result in phase reversal, α-TC cells now adopting an internal position relative to INS-1 or MIN6. Unfortunately, despite generating a stable cell line with high levels of E-cadherin expression, we could never achieve aggregate cohesivities great enough to exceed those of either the INS-1 or MIN6 cells. We therefore opted to increase the cohesivity of the α-TC cells by transfection of P-cadherin, a cell surface adhesion molecule that neither cell line expresses. We showed that P-cadherin transfected α-TC cells (α-TC.Pcad), when mixed with MIN6, tended to “pull away,” resulting in a configuration of partial envelopment. This result is interesting in view of the observation by Duguay et al., that in L-cells, E- and P-cadherin interact with equal molar binding/unbinding energies (Duguay et al.,2003). Providing that this is also the case in our system, we should expect that α-TC.Pcad cells would not only increase their self-cohesion, but would also increase the degree of cross-adhesion, P-cadherin on α-TC.Pcad cells cross-adhering to E-cadherin on MIN6 cells.

In 1978, Steinberg provided a conceptual framework of how the reversible works of cohesion (W_a-a and W_b-b) and adhesion (W_a-b) determine the most stable configuration of a liquid system (see Fig. 6 and Steinberg,1978). A configuration of complete envelopment of MIN6 by α-TC cells requires that the degree of cross-adhesion (W_a-b) between this pair be less than the average of the combined self-adhesions but greater than the self-adhesion of the α-TC cell line (Fig. 6B). Moreover, the self-adhesion of the MIN6 must be greater than that of α-TC. To shift toward a configuration of partial envelopment (Fig. 6C), the degree of cross-adhesion between α-TC.Pcad and MIN6 must be less than the degree of self-adhesion between α-TC.Pcad cells. This would happen if P-cadherin expression by α-TC.Pcad cells was substantially higher than that of E-cadherin expression by MIN6 cells. This is indeed the case, since flow cytometric analysis of cadherin expression showed that P-cadherin expression by α-TC.Pcad cells is approximately sixfold higher than is E-cad expression by MIN6 cells (as measured by comparison of mean peak channel fluorescence, data not shown). In this scenario, excess E-cad on MIN6 cells would be titered out by P-cad interaction, leaving excess P-cadherin on α-TC.Pcad for self-association. This could effectively define a situation in which W_a-b is weaker than either W_a-a or W_b-b and result in partial envelopment of MIN6 by α-TC.Pcad. The altered sorting behavior observed was P-cadherin–mediated, because incubating a combination of α-TC.Pcad/MIN6 cells in P-cadherin function-blocking antibody restored the normal sorting process, further confirming that differential adhesion guides in vitro self-assembly.

Figure 6. Illustration of how the reversible works of cohesion (W_a-a) and (W_b-b) and adhesion (W_a-b) determine the most stable configuration of a liquid system. These relationships should apply to any multisubunit system that adopts liquid-like equilibrium shapes, whether the subunits are molecules or cells. The figure depicts, for a two-phase liquid system, the equilibrium configurations determined by different sets of Ws. By convention, when the two phases differ in cohesiveness, the more cohesive phase is designated a and the less cohesive phase is designated b. In the figure, W_a-a and W_b-b are represented as having higher and lower arbitrary values represented by horizontal solid lines. The figure is divided into four vertical areas, the shading representing a range of values of W_a-b from infinity (column A) to zero (column D). A high value of W_a-b (higher than the mean of W_a-a and W_b-b) causes the two phases to mix preferentially (column A). Zero adhesion between a and b subunits (column D) causes phases a and b to round up into separate, isolated spheres. Low but positive values of W_a-b (between 0 and W_b-b) lead to a partial envelopment of phase a by phase b (column C), whereas higher values of W_a-b (above W_b-b but below the average of W_a-a and W_b-b) lead to complete envelopment of phase a by phase b (column B). After Steinberg (1978).

In rodents (and other animals), the concentric organization of pancreatic islet cells represents the geometry best suited for normal islet function. This topology effectively maximizes beta-cell–beta-cell contact. In rodent islets, beta-cell oscillations in membrane potential in response to high glucose concentrations are coordinated. As a result, the whole islet displays a synchronous oscillatory response that may constitute the molecular basis for the typical pulsatility in insulin release (Valdeolmillos et al.,1989). This synchrony could be facilitated by maximizing beta-cell contact. Recent studies using human islets, however, show that this specific cellular topology is not conserved. Human beta- and non–beta-cell populations appear to be intermixed (Cabrera et al.,2006). An arrangement in which most cells contact other endocrine cells predisposes human islets to stronger paracrine interactions than those observed for rodent islets. Indeed, this random cellular arrangement does not allow cell oscillatory activity to be coordinated (as in animal islets) throughout the human islet (Cabrera et al.,2006). Accordingly, this intermixed geometry could reflect an absence of differential adhesion. This, according to the DAH, would predispose human islet cells to intermixing. Many studies are currently attempting to identify an alternative source of cells that can produce insulin and other islet-specific hormones as a potential method for controlling diabetes. Irrespective of the cell source, it will be important that the cells be arranged in a manner that will optimize their normal function. The information generated by this study has the potential to impact the emerging field of islet cell engineering by providing the groundwork for engineering the adhesive properties of islet cell substitutes in a manner that will program them to self-assemble, in vitro, into 3D structures resembling native pancreatic islets.

EXPERIMENTAL PROCEDURES

Cell Lines

The α-TC cells were obtained from the American Type Culture Collection (ATCC). The line was derived from an adenoma created in transgenic mice expressing the SV40 large T antigen under the control of the rat preproglucagon promoter (Powers et al.,1990), is terminally differentiated, and produces glucagon, but not insulin or somatostatin (Hamaguchi and Leiter,1990). The α-TC cells were maintained in DMEM supplemented with 10% fetal calf serum (FCS), 15 mM HEPES buffer, 0.1 mM nonessential amino acids, and 0.02% bovine serum albumin (Invitrogen, CA). The immortalized beta-cell line MIN-6 was obtained from the laboratory of Dr. Manami Hara (University of Chicago). MIN6 cells produce insulin and have morphological characteristics typical of pancreatic beta-cells. MIN6 cells exhibit glucose-inducible insulin secretion comparable with cultured normal mouse islet cells (Miyazaki et al.,1990). MIN6 cells were maintained in high glucose DMEM and 10% FCS. The INS-1 line was obtained from the laboratory of Dr. Christopher Rhodes (Pacific Northwest Research Institute, Seattle, WA). The INS-1 line was established from cells isolated from an X-ray–induced rat transplantable insulinoma (Asfari et al.,1992). INS-1 cells retain a high degree of differentiation and have been extensively used to study various aspects of beta-cell function. Cells were maintained in RPMI 1640 supplemented with 10% FCS, 0.1 mM nonessential amino acids, 1 mM sodium pyruvate, and 50 μM β-mercaptoethanol. A mixture containing antibiotics and antimycotics was added to all complete media (Invitrogen).

Localization of Insulin- and Glucagon-Expressing Cells in the Mouse Pancreas

Mouse pancreatic tissue was removed directly after killing and fixed in 4% buffered formalin for 12 hr. Tissue was then dehydrated in an alcohol series then cleared in xylene and embedded in Paraffin. Six-micron sections were cut from trimmed paraffin blocks and deposited onto glass slides. Slides were deparafinized and rehydrated through another alcohol series and rinsed in distilled water. Immunohistochemistry was performed using prediluted antibodies specific for insulin (mouse anti-insulin Clone Z006, Zymed, South San Francisco, CA) and rabbit anti-glucagon, (Zymed). Slides were blocked in CAS Bloc (Zymed) for 30 min then incubated overnight at 4°C in primary antibodies. Slides were then rinsed several times in PBS and incubated with the appropriate secondary antibody conjugated to either Alexa Fluor 488 or 568 (Molecular Probes, Eugene, OR). After several more rinses in PBS, slides were mounted in FluorSave mounting medium (Calbiochem, San Diego, CA) and sealed. Images were captured with an Olympus IX81 microscope equipped with a DSU spinning disc confocal imaging system and merged either in Microsuite Five (Soft Imaging System Corp., Lakewood, CO), or Slidebook 4 (Intelligent Imaging Innovations Corp., Denver, CO) imaging software.

In Vitro Cell Sorting of Immortalized Pancreatic Beta- and Alpha-Cells

Cells were detached by trypsin/ethylenediaminetetraacetic acid (TE) treatment and differentially stained using two membrane-intercalating dyes. INS-1 and MIN6 cells were stained with PKH-2 (Sigma, St. Louis, MO) for green fluorescence and α-TC cells were stained with PKH-26 (Sigma) for red fluorescence. Cell concentration was then adjusted to 5 × 10⁶ cells/ml and cells were mixed 1:1 in the following combinations: α-TC_(red)/INS-1_(green) or α-TC _(red)/MIN6_(green). Ten microliters hanging drops were deposited on the underside of the lid of a 10-cm tissue culture dish, and the lid was inverted over 10 ml of PBS for hydration. Hanging drops were incubated until cell sheets formed. Micro-scalpels were used to cut the sheets into approximately 1 mm² fragments. These fragments were then transferred to 10-ml shaker flasks (Belco Glass, NJ) in 3 ml of complete medium and incubated in a gyratory waterbath shaker at 110 rpm and 37°C for 2–3 days or until they became spherical. Aggregates were mounted on glass slides and imaged as previously described (Foty and Steinberg,2005).

Measurement of Aggregate Cohesivity by TST

Aggregate formation.

The 3D multicellular spheroids were generated as previously described (Robinson et al.,2004). Isolated cells were resuspended at a concentration of 5 × 10⁶ cells/ml in complete medium supplemented with 2 mM CaCl₂. Twenty-microliter aliquots of this suspension were used to generate hanging drops. Drops were incubated for 24 hr, allowing the cells to coalesce at the base of the droplets and form sheets. Sheets were transferred to 10-ml shaker flasks (Bellco, NJ) and incubated in 3 ml of complete medium on an orbital shaker at 110 rpm until spheroids formed.

TST.

Aggregate cohesivity was measured by TST as previously described (Foty et al.,1994,1996). Briefly, spherical aggregates ranging in size from 50–150 μ in diameter were transferred to the inner chamber of the tissue surface tensiometer (Supplementary Figure S1) and positioned on the lower compression plate (LCP). The inner chamber contained prewarmed, de-gassed CO₂-independent medium (Gibco-BRL, NY) supplemented with 10% FCS and antibiotics. The upper compression plate (UCP), attached to a nickel–chromium wire, was then positioned above the aggregate and connected to a Cahn electrobalance. The weight of the UCP was zeroed to establish a precompression UCP weight baseline. To minimize adhesion of cell aggregates to the compression plates, both the lower and upper plates were precoated with poly-2-hydroxyethylmethacrylate (poly-HEMA, Sigma, MO), a polymeric material to which cells do not adhere (Folkman and Moscona,1978). Compression was initiated by raising the LCP until the aggregate became compressed against the UCP. Adjusting the height of the LCP controlled different degrees of compression. The force with which the aggregate resisted compression was monitored by the Cahn recording electrobalance. Aggregate geometry was monitored using a Nikon dissecting microscope equipped with a CCD video camera and connected to a Macintosh Power PC. Images of aggregates were captured, digitized, and their geometries were analyzed using NIH Image software (Bethesda, MD). Each aggregate was subjected to two different degrees of compression, the second greater than the first. Measurements of aggregate geometry and the force of resistance to the compressive force were then used to solve the Young–Laplace equation (Davies and Rideal,1963), producing numerical values of apparent tissue surface tension (σ). For more details, see Supplementary Figure S2.

Confirmation of aggregate liquidity.

The two likely material states to be considered as they apply to tissue aggregates are liquidity and elasticity. We have previously shown that, when subjected to compression, aggregates behave as elastics on short time scales and as liquids on longer time scales. The calculated surface tension of a liquid aggregate, when subjected to two different compressions, the second greater than the first, will remain constant. In such aggregates, the ratio of σ₂/σ₁ will be equal to 1 and will be less than the ratio of the force applied at each successive compression (F₂/F₁). In contrast, the calculated surface tension of an elastic aggregate will obey Hooke's law and increase proportionately to the applied force. For elastic aggregates the ratio of σ₂/σ₁ will not be equal to 1, but will instead approach the ratio of F₂/F₁. The surface tension of liquid aggregates will also be independent of aggregate size (Foty et al.,1994,1996). Only measurements in which surface tension is independent of the applied force and size were used to calculate average σ for each cell line. Measurements that did not meet these criteria were not included because they do not represent a true liquid surface tension. There are several reasons why, in practice, aggregates sometime fail to exhibit liquid behavior. Aggregates are subjected to compression manually. Consequently, in some experiments, the elastic regime, of the aggregate is exceeded and the aggregate is injured. In other cases, aggregates were decompressed before they had reached shape or force equilibrium. This is shown by the aggregate springing up upon decompression. In this case, the apparent σ measured after either the first or second compression would not truly reflect the actual σ at equilibrium. Because coating of the compression plates with polyHEMA is sometimes incomplete, some aggregates adhere to and spread upon either the UCP or LCP. This finding could cause an overestimate of the measured apparent σ. Such aggregates are eliminated from consideration. Overall, we included two thirds of all aggregates that were subjected to compression. Aggregates were prepared in batches and maintained in shaker flasks for up to 7 days in culture. Aggregate surface tension was determined throughout the course of the investigation to confirm that they remained liquid-like.

Reverse Transcriptase-Polymerase Chain Reaction Assessment of E-cadherin Expression by α-TC, INS-1, and MIN6 Cells

Total RNA was prepared using the RNAeasy Micro Kit (Qiagen, MD). Five micrograms of total RNA was used to generate cDNA using the Advantage RT for PCR kit (BD Biosciences, Palo Alto, CA). Five microliters of this reaction were used for reverse transcriptase-polymerase chain reaction (RT-PCR) for amplification of E-cadherin mRNA. A single primer set based on both INS-1 rat (AB017696) and α-TC, MIN6 mouse (X06115) E-cadherin sequences was designed. The primers (F) 5′-ATG GGA GCC CGG TGC CGC AGC TTT C-3′ and (R) 5′-GAT GGG AGG GAT GAC CCA GTC TCG T-3′ were used to amplify a 497-bp fragment from rat cDNA (corresponding to nucleotides 151–597) and a 491-bp fragment from mouse cDNA (corresponding to nucleotides 93–533). The primers were very similar in their primability and stability. Another set of primers was designed for 18S rRNA (F) 5′-CCT CCA ATG GAT CCT CGT TA-3′ and (R) 5′-AAA CGG CTA CCA CAT CCA AG-3′ and used to amplify a 157-bp fragment corresponding to nucleotides 468–584 and 467–583 for rat and mouse 18S rRNA, respectively. This primer set had identical primability and stability of match parameters for both rat and mouse 18S rRNA sequences. For amplification of E-cadherin mRNA, the PCR conditions were 94°C for 4 min, 40 cycles of 94°C for 30 sec denaturation, 58°C for 30 sec annealing, and 72°C for 1 min extension steps, then 72°C for a 4-min final extension step. For 18S rRNA, conditions were identical except for the annealing temperature which was 52°C. Fragments were separated on a 0.8% agarose gel and the gel was stained in 0.5% ethidium bromide, briefly destained in distilled water, and photographed under ultraviolet light. Images were digitized with an Agfa Duoscan T1200 scanner and captured in Adobe Photoshop imaging software.

Assessment of E-cadherin Expression by Indirect Immunofluorescence

Cells were plated into 24-well dishes at a concentration of 2 × 10⁵ cells/ml and incubated for 48 hr, whereupon they were fixed in 4% paraformaldehyde. Fixed cells were rinsed in Hanks; balanced salt solution (HBSS) and blocked in CAS-Block buffer (Zymed, CA) for 3 hr. Cells were first incubated with 10 μg/ml unconjugated E-cadherin antibody (Clone 36, BD Biosciences) in CAS-Block buffer overnight at 4°C then rinsed in HBSS and re-incubated at room temperature for 1 hr in an Alexa Fluor 488-conjugated secondary antibody diluted to a concentration of 2 μg/ml in CAS-Block buffer. Cells were rinsed three rimes in HBSS and imaged by epifluorescence microscopy.

Immunoblot Analysis of Differential E-Cadherin Protein Expression by α-TC, INS-1, and MIN6 Cells

Cell lysates were prepared as follows: near-confluent 10-cm tissue culture plates of each cell line were washed twice in ice-cold Tris-buffered saline containing 5 mM CaCl₂ and 1 μM phenylmethyl sulfonyl fluoride (TBS+ Ca²⁺). Cell monolayers were lysed by the addition of 500 μl of RIPA lysis buffer (150 mM NaCl, 50 mM TRIS pH 7.5, 1% NP40, 0.25% DOC) containing a protease inhibitor cocktail (Calbiochem, CA). The lysates were transferred to microcentrifuge tubes, rotated at 4°C for 1 hr, then passed through a Qia-shredder (Qiagen, CA) and centrifuged at 14 × g for 15 min at 4°C. Protein concentration was determined by the BCA assay (Pierce, IL). Twenty micrograms of protein was separated on 7% sodium dodecyl sulfate (SDS) -polyacrylamide gel electrophoresis gels and blotted to polyvinylidene fluoride membranes using standard protocols. Blots were blocked in Membrane Blocking Buffer (Zymed, CA) for 1 hr, then incubated at 4°C overnight in 1 μg/ml mouse monoclonal E-cadherin antibody (BD Biosciences, clone 36). Blots were rinsed three times in TBS-0.2% Tween 20, then incubated at room temperature for 1 hr in the appropriate secondary antibody conjugated to horseradish peroxidase. After three more rinses, blots were developed using enhanced chemiluminescence (Amersham Biosciences, Piscataway, NJ). All blots were then stripped in 62 mM Tris HCl pH 6.8, 2% SDS, and 100 mM 2-mercaptoethanol for 30 min at 50°C, washed in TBS-T, and re-probed with an anti-actin antibody (0.1 μg/ml, Sigma) to confirm equal lane loading. X-ray films were digitized using an Agfa Duoscan T1200 digital scanner and quantified by NIH Image gel scanning software (Bethesda, MD). For quantification of the Western blot analysis, means and standard errors of densitometry data collected from five separate experiments were calculated and compared by one-way ANOVA and Neuman–Keul's multiple comparisons test.

P-Cadherin Transfection of α-TC Cells

Cells were transfected with 1 μg of a plasmid encoding the expression of mouse P-cadherin (pβactPcad; Nose et al.,1988) along with 1 μg of pCDNA3 (Invitrogen, Carlsbad, CA) for G418 selection using an Amaxa Nucleofector II Device (Amaxa Biosystems, Gaithersburg, MD). Three-million cells were suspended in 100 μl of Nucleofection Solution V and transfected using program G-24. These conditions were empirically determined to yield high transfection efficiency and high viability. Transfected cells were diluted 1/100 and plated into medium containing 800 μg/ml G418. Resistant cells were grown to confluence, detached by 0.05% trypsin/5 mM Ca²⁺ (TC) treatment, and stained with an anti–P-cadherin antibody (PCD-1, Zymed, CA) on ice for 45 min. After several washes in HBSS, cells were mixed with an Alexa Fluor-488–conjugated secondary antibody and incubated on ice for 30 min. P-cadherin–expressing cells were fluorescence-activated cell sorted, positive cells were expanded, re-analyzed by flow cytometry, and re-sorted until pure populations of P-cadherin–expressing cells were isolated. Once expanded, P-cadherin expression was assessed by Western blot analysis. The α-TC cell lines exhibiting increased P-cadherin expression were used for in vitro sorting assays.

In Vitro Sorting of α-TC.pCad and MIN6 Cells

P-cadherin–transfected α-TC cells (α-TC.Pcad) were stained with the red fluorescence dye PKH-26 and mixed with MIN6 cells that were stained with the green fluorescence dye PKH-2. Mixtures were placed in hanging drops, incubated for 24 hr, and transferred to culture flasks. Cultures were placed in an orbital shaker and incubated for another 48 hr until spherical aggregates formed. Studies were also performed in which mixtures of α-TC.Pcad/MIN6 were incubated in the presence of 200 μg/ml pCD-1, a P-cadherin function-blocking antibody (Zymed, CA). In this case, hanging drops were incubated for 24 hr before imaging and were not transferred to shaker flasks. Aggregates were imaged as previously described.

Acknowledgements

The authors thank Dr. Manami Hara (University of Chicago) and Dr. Christopher Rhodes (Pacific Northwest Research Institute, Seattle, WA) for kindly providing cell lines. We also thank Dr. Margaret Schwarz and Dr. Siobhan Corbett for use of the DSU spinning disc confocal microscope and Nikon epifluorescence microscope, respectively. We thank Mr. Atila Entersz for constructing and maintaining the tissue surface tensiometers. R.A.F. was funded by the Department of Defense.