An in vitro model for the analysis of intestinal brush border assembly I. Ultrastructural analysis of cell contact-induced brush border assembly in Caco-2_BBe cells

The brush border (BB) of the intestinal epithelial cell provides an excellent opportunity for dissecting how a complex actin-membrane array is assembled. It is a highly ordered structure, consisting of organized microvilli (MV) anchored in a subjacent, filamentous terminal web (TW) (for review, see Mooseker, 1985; Bretscher, 1991). Its stability has permitted thorough ultrastructural and biochemical characterization. However, the BB is also dynamic under normal physiological conditions, such as feeding, fasting, or exposure to lectins (for review, see Heintzelman and Mooseker, 1992). The BB assembles in cells migrating from the crypt to the villus tip in the adult intestine. Similar BB assembly also occurs during embryogenesis in the epithelia of both the visceral endoderm and fetal intestine (for review, see Heintzelman and Mooseker, 1992; Louvard et al., 1993). In addition, major proteins within the BB turn over rapidly without disturbing the integrity of the structure (Stidwell et al., 1984).

The assembly of the BB cytoskeleton has been approached primarily by study of the differentiation of the intestinal epithelium during embryogenesis (reviewed by Heintzelman and Mooseker, 1992; Mamajiwalla et al., 1992). Transmission and scanning electron microscope (TEM and SEM) studies in chicken, rodent and human intestine reveal the same basic sequence of morphological events; the main difference is only the rate at which these steps occur (Chambers and Grey, 1979; Heintzelman and Mooseker, 1990a; Mathan et al., 1976; Ezzell et al., 1989; Kelley, 1973; Moxey and Trier, 1979). The basic sequence is as follows: after a polarized epithelial sheet is established, short, sparse microvilli (MV) first appear. These MV then become more numerous and organized, and the actin core bundles of the MV start to extend into the enterocyte. Next, the terminal web (TW) assembles, and the MV appear more regular. This is followed by final elongation of the MV (Heintzelman and Mooseker, 1992; Mamajiwalla et al., 1992). While these events occur gradually during chick embryogenesis, they are compressed into the last few days of rodent fetal development (Mathan et al., 1976; Colony and Neutra, 1983). Human intestinal BB assembly is almost entirely completed by the end of the second trimester (Moxey and Trier, 1979). A proximal-distal gradient is also observed in all four species; differentiation occurs first in the small intestine and later in the colon (Heintzelman and Mooseker, 1992).

BB assembly has also been examined in the adult intestine during differentiation of crypt cells. The general pattern of assembly is similar to that observed during embryogenesis, but several details differ. In the chicken, the least differentiated crypt cells have actin rootlets extending into the cytoplasm and just starting to protrude from the apical cell surface (Fath et al., 1990). Many crypt cells, however, have short MV oriented at angles to the cell surface. These MV contain actin core bundles, occasionally more than one, that extend up to 1 μm into the subjacent cytoplasm (Fath et al., 1990). The TW is almost non-existent. Further up the crypt the MV became more numerous and upright with respect to the apical cell surface, and some crosslinking filaments are first observed in the TW region. Finally, microvillar length and density increase substantially, and the TW assumes its mature, well-defined structure, including the exclusion of organelles (Fath et al., 1990). Mammalian crypt cell differentiation has not been examined as thoroughly in regards to BB ultrastructure changes (van Dongen et al., 1976; Trier, 1963; Kelley, 1973). For example, it is not known if these changes in MV core organization occur during crypt-villus differentiation. In humans, undifferentiated crypt cells already have short, wide MV covering their apical surface, but microvillar density is significantly lower than that of the villus epithelial cells (Trier, 1963). Some of the actin core filaments extend down 1-3 μm into the cytoplasm, occasionally aggregating with those of neighboring MV (Trier, 1963). No TW is present. Finally, crypt cells in fetal intestine have large stores of glycogen, unlike adult crypt cells (Moxey and Trier, 1978). Although these studies have demonstrated the value and future potential of addressing aspects of BB cytoskeletal assembly in these various tissues, each has its limitations. The time frame in which differentiation takes place along the crypt-villus axis is highly condensed as compared to that of chick embryogenesis; most of the early events of MV assembly have already occurred in the crypt cells (Heintzelman and Mooseker, 1990a; Fath et al., 1990; Trier, 1963). Assembly of the brush border during chick embryo-genesis, on the other hand, occurs more gradually, but isolation of specific, homogeneous populations of intestinal epithelial cells is technically difficult, due to the small size of the intestine.

The present study focuses on BB assembly in clones of the Caco-2 cell line. These cells, along with several other lines also derived from human colonic adenocarcinomas, have retained the ability to differentiate in culture, unlike lines derived from normal human intestinal mucosa (reviewed by Zweibaum et al., 1991). Caco-2 cells have been exploited by many groups in regard to a broad spectrum of intestinal and epithelial parameters (for review, see Zweibaum et al., 1991; Louvard et al., 1993). The clones used in this study, the Caco-2_BBe (C2_BBe) cells, were isolated based on BB expression with tight apical localization of the MV-specific protein villin (Peterson and Mooseker, 1992). The C2_BBe-1 and 2 clones both form a polarized monolayer upon reaching confluence, with an apical BB ultrastructurally indistinguishable from that of the enterocyte (Peterson and Mooseker, 1992). In addition, the expression, localization, and physical association of a number of BB proteins are comparable to that of human colonic epithelial cells. This includes the expression pattern of multiple unconventional myosins; the C2_BBe clones are also the only cell lines we have tested that express the differentiation-specific BB myosin I (Peterson and Mooseker, 1992; for more details regarding the expression of unconventional myosins, see accompanying paper).

The goal of this and the companion study (Peterson et al., 1993) is to establish whether or not the C2_BBe clones can be used as a valid model to address various aspects of BB assembly. The use of cell lines like the C2_BBe cells may circumvent some of the difficulties encountered in vivo as discussed above. Here, we have analyzed the ultrastructure of the C2_BBe cells during BB assembly. A modification of the protocol of Anderson et al. (1989) was employed to generate isolated C2_BBe cells lacking BB and junctional remnants, which were then plated at very high density on filters in low-Ca²⁺ medium. Ultrastructural changes were followed over time during the cell contact-induced BB assembly upon the addition of Ca²⁺. We observed a series of events similar to those described in vivo: after mono-layer formation, sparse MV grew more numerous and organized, and the TW assembled. Cell height increased throughout the time course. The accompanying paper addresses the expression and localization of a subset of BB proteins during the time course.

Caco-2 clones C2_BBe-1 and -2 were maintained as described by Peterson and Mooseker (1992). Low passage cells (42-49) were used in this study, although BB expression is stable until at least passage 68 (Peterson and Mooseker, 1992). Low Ca²⁺ medium consisted of suspension-modified MEM (JRH Biosciences, Lenexa, KS; no Ca²⁺ in the formulation), 10% fetal bovine serum (Hyclone, Logan, UT) dialyzed against 4 changes of saline to remove Ca²⁺, 2 mM glutamine (Sigma, St. Louis, MO), 10 μg/ml transferrin (Boehringer-Mannheim, Indianapolis, IN), and penicillin-strepto-mycin-Fungizone (JRH Biosciences). The protocol for obtaining depolarized cells used by Anderson et al. (1989) was modified for these studies. Cells for suspension culture and subsequent time course studies were plated in T75 flasks (Corning, NY). Upon reaching approximately 70% confluent density, the flasks were switched into low-Ca²⁺ medium for 24 h. The cells were then trypsinized and resuspended at 5×10⁵ cells/ml in low-Ca²⁺ medium. Cytochalasin D (Sigma; 2 μM final concentration) was added to aid the disassembly of any junctional and BB remnants, and the cells were maintained in suspension culture for 22 min. The cell suspension was then filtered through 20 μm mesh (Small Parts, Inc., Miami, FL) to remove any aggregates of dead cells and then pelleted out of the cytochalasin D-containing medium. This combination of rapid cytochalasin D treatment and filtration increased the number of viable cells plated. Finally, the cells were resuspended in fresh low-Ca²⁺ medium and plated on uncoated 24 mm Transwell filters (Costar, Cambridge, MA) at 8.65×10⁵ cells/cm²; this density is approximately 4× the initial confluent density of the C2_BBe cells. Lower densities in preliminary studies resulted in barely half the filter covered. The filters were suspended in Lexan inserts which sit in 150 mm tissue culture plates, 5 filters per plate.

After 24 h in low-Ca²⁺ medium, the first filter was processed. This was considered the zero time point, and the cells are referred to as depolarized cells. The remaining filters were switched to DMEM supplemented as described by Peterson and Mooseker (1992). Subsequent time points are referred to by the amount of time elapsed since return to the regular maintenance medium. These include 12 h, 1 d, 2 d, 3 d, 5 d, 9 d, 14 d (TEM only) and 19 d. The data shown here are from one of four studies with this, or a similar protocol, on the C2_BBe-1 clone; comparable results were obtained with clone C2_BBe-2.

Preparation for both TEM and SEM was as described by Heintzel-man and Mooseker (1990b) with modifications as specified by Peterson and Mooseker (1992). One filter per time point was processed. During dehydration in ethanol, the filter was cut in half; one half was embedded for thin sectioning, while the other was prepared for SEM.

Microvillar density was determined by counting MV in SEM micrographs. A minimum of 300 μm² per time point was examined, with the exception of the 19 d time point. In this case, 223 μm² were scanned, but these represented six distinct areas of the filter. Microvillar length was obtained by measuring the MV in TEM micrographs on a minimum of 20 cells from the 3 d time point onward. Only those MV whose full length was in the plane of section were measured. Consequently, fewer MV were measured at the 1 d and 2 d time points, as many MV traversed in and out of the plane of section. Cell height was also calculated from TEM micrographs; a minimum of 15 cells per time point were measured. Again, only those cells whose full length was in the plane of section and were approximately 90° to the filter were measured; this includes the MV. Cells in large aggregates were not measured.

As a starting point for examining BB cytoskeletal assembly, ‘depolarized’ C2_BBe cells were obtained by a modification of the Ca²⁺ jump protocol (Gonzalez-Mariscal et al., 1985; Anderson et al., 1989). With this protocol in MDCK cells, densely plated ‘contact-naive cells’ are induced to undergo differentiation simultaneously upon restoring the Ca²⁺ concentration to physiological levels, with a minimum of cell division (Nelson and Veshnock, 1987). Suspension culture was employed as a means of generating ‘contactnaive’ C2_BBe cells, as a previous study indicated that, unlike the MDCK model, removal of Ca²⁺ from very subconfluent cultures was not sufficient for full junctional disassembly (Anderson et al., 1989). The zero time point was taken 24 h after the C2_BBe cells had been plated at high density in low-Ca²⁺ medium. At this time, small, round cells, as well as bits of cell debris, covered the surface of the filter, although complete coverage of the filter was not obtained. Some heterogeneity in the size of cells was also noted (Fig. 1A). The free surfaces of these cells were frequently covered with numerous, irregular protrusions or blebs (Fig. 1B). Twelve hours after the cells were returned to regular maintenance medium, a variety of morphologies were observed. Small, round cells were still present, while other cells had started to spread; many cells, however, formed complex aggregates (Fig. 1C). A striking change in the appearance of the cell surface was noted; the irregular protrusions had given way to small, sparse MV (Fig. 1D). By 1 d, many more cells had spread, and the monolayer was virtually complete (data not shown). The cell aggregates, now positioned on top of the monolayer, were shed between 2 d and 3 d.

We examined the topographical changes accompanying cell contact-induced BB assembly in the C2_BBe cells by SEM, particularly focusing on changes in microvillar density and distribution (Fig. 2). This higher magnification view highlights the difference between the membrane blebs of the depolarized cells and the small, cylindrical MV present in 1 d cells (Fig. 2A and B). These blebs appeared to resemble small folds of membrane (Fig. 2A). At the 1 d time point, the MV were relatively evenly scattered over all the apical cell surface (Fig. 2B). The MV had increased in number by 2 d. Heterogeneity in MV distribution was observed on single cells (Fig. 2C), with dense tufts of MV splaying outward immediately adjacent to a more sparsely populated area. Patchiness in the microvillar density was also noticeable throughout the monolayer (data not shown).

Microvillar density further increased at 3 d, and some regions exhibited dense packing and lateral aggregation of MV (Fig. 2D). MV density was still heterogeneous from cell to cell (data not shown). By 9 d, however, cells were more evenly covered with a thick carpet of MV (Fig. 2E), although maximal packing had not been reached (compare Fig. 2E and F). MV achieved their highest density by the end of the time course; individual MV were difficult to distinguish in some fields (Fig. 2F). Quantification of the number of MV per μm² over time confirmed the visual impression of steadily increasing microvillar density (Fig. 3). Average microvillar density increased almost nine-fold throughout BB assembly.

TEM analysis revealed considerable alteration of overall cell morphology throughout BB assembly in C2_BBe cells. The initial cell population consisted of fairly small, generally round cells averaging 10.9 μm in height. The cells are dominated by a large nucleus which is usually positioned in the upper part of the cell, with a thin rim of cytoplasm between it and the membrane (Fig. 4). Although an occasional mitochondrion or vacuole was spotted above the nucleus, the majority of organelles, including mitochondria and vacuoles, occupy the space directly underneath the nucleus (Fig. 4). Cellular processes extended down into the pores of the filter, and the free cell surface appeared uneven, with no evidence of BB or junctional remnants (Fig. 4). This is in stark contrast to the cellular appearance after 12 h in regular maintenance medium (Fig. 5A). By this time the C2_BBe cells had formed multicellular aggregates; within some aggregates, intercellular cysts with MV pointing into the lumen were noted (Fig. 5A). Some cells within the aggregate made significant contact with the filter, and had established junctions with neighboring cells.

Adjacent serial sections revealed that other cells have extensive lateral contact with one or more others and with junctional complexes, but no contact with the filter (e.g. Fig. 5A, upper two cells and data not shown). Still others are closely apposed, but no junctional complex is present (data not shown). By 1 d, a monolayer had formed, and apical junctional complexes were present between neighboring cells (Fig. 5B). These short, spreading cells had MV of variable densities localized to the apical surface (Fig. 5B). Large cell aggregates upon the top of the monolayer had complex connections with one another, including apical junctions, but it is not clear how they were associated with the monolayer underneath (data not shown). Still, these aggregates survived processing for EM and for immunofluorescence (see accompanying paper). At 2 d, the monolayer was composed of relatively cuboidal cells (Fig. 5C), although they are taller than those at 1 d. MV were more numerous, and the beginnings of the complex plicae were present along the lateral sides of C2_BBe cells (Fig. 5C), characteristic of cells in the mature monolayer (Peterson and Mooseker, 1992).

Cell height continued to increase; by 3 d the cells had almost doubled their initial height (Fig. 6A; see below for quantification of changes in cell height). The MV appeared more upright, and small amounts of glycogen were present, a characteristic of these clones and the parent line (Pinto et al., 1983; Peterson and Mooseker, 1992) (Fig. 6A). A more organized appearance was notable in the 5 d monolayer (Fig. 6B). The distinctly columnar cells now contained a significant amount of glycogen. Such an accumulation of glycogen both above and below the nucleus is also characteristic of human fetal intestine (Moxey and Trier, 1979). An apical zone of organelle exclusion, signifying TW assembly, became evident, regardless of the MV density on a given cell (Fig. 6B). However, the monolayer did not appear fully differentiated until the 9 d time point (Fig. 7), although only a small height increase was noted. Some of the cells in the monolayer appeared twisted or contorted around each other, making it difficult to follow a single cell from top to bottom in some sections. Other than a continued increase in cell height, no major ultrastructural changes were observed throughout the remainder of the time course (Fig. 8). As noted previously (Peterson and Mooseker, 1992), the C2_BBe cells are heterogeneous in height despite being a clonal population. This heterogeneity is illustrated clearly in a comparison of the monolayers at 14 d versus 19 d (Fig. 8); a shorter area of the monolayer was sectioned at the 19 d time point. The gradual change in cell height is summarized in Fig. 9; the C2_BBe cells underwent over a three-fold increase in height, from an average of 10.2 μm to 35.3 μm; some cells reached 49 μm. The cells continue to increase in height beyond the 19 d time point (Peterson and Mooseker, 1992; and data not shown. Note that all micrographs in Figs 4-8 are at the same magnification).

High magnification TEM of the apical domain of the cells was performed to examine in detail the cytoskeletal changes during BB assembly. Initially, no MV or organized BB remnants were present in the depolarized cells; instead, the cell surface was uneven (Fig. 10A), or, more commonly, covered with many large blebs (Fig. 1B). These blebs contained abundant ribosomes and other cytoplasmic elements (e.g. see Fig. 10C), indicating that they are not artifacts of glutaraldehyde fixation (Hay and Hasty, 1979). No evidence of junction formation was observed at points where adjacent cells were in contact (Fig. 10B). Microtubules coursed through the cells, and numerous ribosomes and rough endoplasmic reticulum were directly adjacent to the plasma membrane (Fig. 10A,B,C). Also notable was a dense mesh of filaments at the base of the cell just above the basal membrane (Fig. 10C). Twelve hours after return to normal maintenance medium, the blebs have been replaced with short (approximately 0.5 μm), sparse MV on free cell margins (Fig. 10D); the MV appear to contain actin filaments. No obvious change in the distribution of filaments or organelles in relation to the cell surface was noted.

The apical surface of 1 d cells had MV of mixed sizes, ranging from 0.4 to 1.6 μm, and many were at an angle to the apical surface of the cell (Figs 2B, 11A). Again, actin core filaments could be discerned in the MV, as well as associated with the junctional complex (Fig. 11A). The 2 d MV appeared to have increased in length (av. 1.25 μm), although this was difficult to quantitate, as the MV traversed in and out of the plane of section (Fig. 11B). This impression was consistent with the high magnification SEM studies (compare Fig. 2B with 2C). MV were generally perpendicular to the apical cell surface and considerably more dense by 3 d (Fig. 12A), corresponding to the topographical change seen previously (Fig. 2D). Crossbridges linking the core bundles to the overlying membrane could be detected in optimal MV cross-sections (data not shown). A few actin MV core filaments were visible extending into the subjacent cytoplasm. While various filament types were readily visible near the apical membrane, no organized meshwork or evidence of TW assembly was seen (Fig. 12A).

A defined TW region first became distinct at 5 d, with actin core bundle rootlets clearly visible (Fig. 12B). Although some region of organelle exclusion was evident throughout the monolayer, the extent of TW formation varied from cell to cell (Fig. 12B). The organization and extent of organelle exclusion in the TW correlated with well-ordered, densely packed MV. Some variability in microvillar density was still apparent from cell to cell (Fig. 12B). By 9 d, all cells within the monolayer had a well-ordered BB comparable to that of human intestinal epithelial cells. Densely packed, upright, uniform MV were anchored by their rootlets in a well-stratified TW (Fig. 13A). Few changes in the BB were subsequently observed, although a further increase in MV density is evident at 19 d (Fig. 13B; also Fig. 2F). Two types of BBs were observed, morphologically similar to human small intestinal and colonic BBs (Fig. 13B). Long MV with short actin rootlets in a relatively shallow TW are characteristic of the small intestine, while colonic BBs have fairly short MV rooted into a deep TW (Carboni et al., 1987). This has been noted previously in highly confluent C2_BBe cells (Peterson and Mooseker, 1992). The two distinct types were reliably detected at 9 d, with MV heights averaging 0.73 μm for the colonic-like BBs and 1.35 μm for small intestinal type BBs.

The sequence of ultrastructural events during cell contact-induced BB assembly described above are strikingly similar to those observed in vivo and are summarized diagrammatically in Fig. 14. Caco-2 cell differentiation has been suggested to resemble most closely that of the fetal colon (Pinto et al., 1983; Blais et al., 1987). This notion is generally reinforced by the ultrastructural data presented here, with the exception of the events at the very beginning (see Fig. 14 for summary). When the C2_BBe cells were returned to Ca²⁺-containing medium, small, sparse MV were present before a polarized monolayer was formed (Figs 1D, 5A), a phenomenon not observed in vivo. Sub-sequent ultrastructural features, however, were similar. As in avian intestinal embryogenesis (reviewed by Heintzelman and Mooseker, 1992), small, sparse MV were at an angle to the cell surface initially (Fig. 11A). They became straighter and more upright over the next 3-4 days (Figs 11B, 12). It is not clear if the increasing straightness of the MV reflects a change in the organization of the actin core bundle or a differential response of the unorganized actin filaments during fixation. The actin core rootlets just began to extend at 3 d, while TW assembly was underway at 5 d (Fig. 12). MV density increased steadily over time (e.g. see Fig. 2). While the initial BB assembly appeared complete by 9 d (e.g. see Fig. 13A), both MV density and cell height continued to increase (Figs 2 and 13). Unlike chicken and rat small intestine (reviewed by Heintzelman and Mooseker, 1992), we did not observe a final dramatic elongation of the MV after TW assembly. However, human colonic MV are shorter than those of the small intestine (Carboni et al., 1987), so perhaps significant MV elongation does not take place. To our knowledge, the ultrastructure of colonic BB assembly has not been examined in detail in any species.

The C2_BBe model appears to parallel fetal intestinal differentiation more closely than that of the adult crypt cell. As outlined earlier, chicken and human crypt cell MV have distinctive rootlets extending 1 μm into the cytoplasm early in their differentiation, well before the TW assembles (Fath et al., 1990; Trier, 1963). This is simply not observed in the C2_BBe model. Instead, increased microvillar organization is underway before short rootlets extend. This occurs just prior to TW formation, resembling the embryonic sequence of events.

The Ca²⁺ jump model reported here differs in several respects from the response of MDCK cells under similar conditions (Nelson and Veshnock, 1987; reviewed by Cereijido, 1992). Unlike the studies on MDCK cells, we did not achieve a >90% plating efficiency, even with plating densities as high as 8× the initial confluent density. The cells adhered to one another after return to normal medium and formed aggregates above the filter surface, although ample surface area was available. One avenue to pursue would be the testing of the efficacy of different extracellular matrix components in promoting cell attachment to the filters. Increased plating efficiency might resolve the problems at the early time points noted above. Preliminary studies of confluent C2_BBe cells on laminincoated plastic dishes revealed better overall monolayer organization than on plastic alone (M. D. Peterson, M. Basson, J. Madri, and M. S. Mooseker, unpublished observations). However, this has yet to be tested for C2_BBe cells on laminin-coated filters. Characterization of such cells will be necessary as culture conditions have been shown to affect both cell morphology and function (Riley et al., 1991; M. D. Peterson and M. S. Mooseker, unpublished observations).

When employing the C2 _BBe system, certain caveats must be kept in mind. Differential cell-cell contact leads to some asynchrony in the initial steps of BB assembly from cell to cell. The large cell aggregates observed at 12 h persist until 1 d, and the developing monolayer is still not quite organized properly with respect to the polarity of some cells. Therefore, the C2_BBe cells will be of limited use for bio-chemical studies regarding the earliest events in MV formation. However, this model provides a valuable system for examination of other events in BB assembly. These include formation of a dense, well-ordered array of MV, TW assembly, regulation of MV length, and differentiation of the MV membrane. In particular, as described in the accompanying paper, this system should be well-suited to analysis of those changes in protein expression which underlie the observed changes in C2_BBe ultrastructure.

The authors thank Barry Piekos for thin sectioning and help with the SEM. We also thank Corey Thompson for help in producing the summary Figure and Son Do for assistance with the morphometry. Many thanks also to all of the members of the Mooseker laboratory for support, advice, and encouragement. This work was supported by NIH grants DK 25837 and GM 37756 to M.S.M.; M.D.P. was supported in part by an NSF predoctoral fellowship.