Cell Culture and Treatments

CACO-2 cells (clone C2BBe1; CRL-2102) were obtained from American Type Culture Collection and grown in flasks, 24-mm or 6-mm Transwell-Clear™ filters (for experiments), or roller bottles (Corning Costar) as described before (Salas et al. 1997). For experiments with anticytoskeletal agents, the cells, at 9 d after seeding, were incubated in DME-F12 supplemented with 33 μM nocodazole, 2 μM cytochalasin D, or 5 mM acrylamide (all from Sigma Chemical Co.) for 7 h before fixation or extraction. Metabolic labeling was performed with 0.15 mCi/ml [³⁵S]methionine and [³⁵S]cysteine (EXPRE Protein Labeling Mix; NEN) in DME with only 15 μM methionine and cysteine for 24 h. For antisense downregulation, cells were maintained in the standard tissue culture media containing 2–10 μM phosphorothioate oligonucleotides as described before (Salas et al. 1997).

Double Immunofluorescence, Confocal Microscopy, Image Analysis, and Immunoelectron Microscopy

Immunofluorescence procedures were described as before (Vega-Salas et al. 1987a,Vega-Salas et al. 1987b). The cells were fixed in 100% methanol at −20°C. We have previously demonstrated that the morphology of the apical IF cytoskeleton is similar under methanol, formaldehyde or formaldehyde/glutaraldehyde fixations (Salas et al. 1997). Likewise, Stearns et al. 1991 has shown that the method of fixation does not affect the localization of γ-tubulin. Double indirect immunofluorescence was performed using a monoclonal antibody against CK19 and a rabbit polyclonal antibody against γ-tubulin together, and two secondary antibodies raised in goats with no cross-reactivity, and labeled with FITC (anti-rabbit) or CY3 (anti-mouse). Transwell™ filters holding the cell monolayers were mounted in polyvinyl alcohol as described before (Salas et al. 1997).

Laser confocal microscopy was performed with an Odyssey XL (Noran Instruments) microscope, using an Omnichrome laser source. For colocalization experiments (FITC/CY3), a second detection channel was used separating the light with a custom made 540 nm secondary dichroic using a 565 LP emission filter in the red channel and a 535 ± 20 nm emission filter in the green channel (all from Omega Optical Inc.). To increase resolution in the z-axis, a 15-μm slit was routinely used. The images were collected using Intervision software (Noran Instruments). Each confocal section was obtained as the average of 32 frames to filter noise. The sections were collected at 0.3-μm intervals in the z-axis for low magnification images comprising the entire thickness of the monolayer or at 0.1-μm intervals for high-resolution imaging of the apical domain with 100-nm pixels (nearly cubic voxels for high resolution images) through a 63× oil immersion objective. Usually, each field comprised 60–90 confocal sections. Before three-dimensional reconstruction, the images were subjected to three-dimensional deconvolution using Intervision software for nine neighboring voxels. Convergence was usually achieved at or before 10 iterations. To experimentally adjust the parameters of the deconvolution, test the coalignment of the channels, and assess the final limit of resolution of the system after deconvolution, single or double fluorescent 1-μm beads (Molecular Probes) were used. To adjust the point spread function parameter, the same confocal stacks from preparations in the standard mounting medium were processed using different parameters until the distortion in the z-axis was abolished. Three-dimensional reconstructions were performed using Intervision software on deconvoluted stacks of confocal images. Pixel intensity histograms over single cell areas were used to assess efficiency of the antisense targeting. For z-sections (perpendicular to the plane of the monolayer), the deconvoluted images were cut in 9-pixel-thick volumes (in the x-axis) at the desired level, reconstructed, and rotated 90° to obtain a view of the apico-basal axis.

For immunolocalization at the EM level, we combined the procedures of Nanogold™ (Nanoprobes, Inc.) and immunoperoxidase for EM (Brown and Farquhar 1984). Both first antibodies (mAb anti-CK19 and rabbit anti–γ-tubulin) were added together. To avoid interactions between the peroxidase and the silver enhancer of Nanogold, we completed all the steps of the Nanogold procedure first, and then incubated the cells with the Fab anti–rabbit IgG-peroxidase and followed the diaminobenzidine reaction as described before (Vega-Salas et al. 1987b).

Cytoskeletal and Cytokeratin Preparations, Gradient Separation of Cytoskeletal Fragments

Cytoskeletal preparations were obtained as described before (Salas et al. 1988). In brief, the monolayers were washed in PBS, extracted in PBS containing 1% Triton X-100, 2 mM EDTA (extraction buffer, EB) and a cocktail of antiproteases (Sigma Chemical Co.) for 10 min at room temperature, and centrifuged for 9 min in an Eppendorf type centrifuge (14,000 rpm). Alternative extractions, in a more physiologic condition, were performed in 70 mM KCl, 80 mM Pipes (pH 6.5), 5 mM EGTA, 2 mM MgCl₂ supplemented with 0.1% saponin (KEB buffer), and 2 nM caliculyn, 5 μM okadaic acid, and 0.5 mM sodium orthovanadate in addition to the standard cocktail of antiproteases. Cytokeratin preparations were obtained by Triton extraction in 1.5 M KCl followed by cycles of 9 M urea solubilization and repolymerization as described by Steinert et al. 1982.

To obtain fragments of the cytoskeletal preparations, the pellets from the first centrifugation were resuspended in 1 ml EB (or KEB) in Eppendorf tubes. Then, the pellets were sonicated while immersing the tubes in ice-water for a total of 3 min (actual sonication time), with intervals of 10-s sonication (~250 watts, 5 on a scale of 10) and 15-s gaps to allow for heat dissipation. The suspensions of cytoskeletal fragments were immediately loaded on top of a preformed 10-ml continuous sucrose gradient (20–60% sucrose in EB or KEB) and centrifuged in a Beckman ultracentrifuge at 15,000 rpm for 50 min in a swinging bucket rotor (Beckman SW-40) at 4°C. The gradients were fractionated from the top by gently pipetting ten 1-ml fractions from the surface. No sedimentation standards are available for values well >100 S, except for viral particles that provide an approximate calibration for the gradient.

To determine the shape and size of the cytoskeletal fragments we used two methods: (a) 1 μl of the corresponding gradient fractions were spread on carbon-coated EM grids and stained with 2% uranyl acetate; and (b) the fractions were fixed in 2% glutaraldehyde, pelleted, embedded in Spur and sectioned. The EM samples were observed and photographed under a JEM 100-CX II (JEOL) transmission electron microscope.

Downregulation of CK19 with Antisense Oligonucleotides Dislodges Centrosomes in Targeted Cells

In a previous publication (Salas et al. 1997) we showed downregulation of CK19 by continuous incubation of expanding cell populations in media supplemented with 2–10 μM antisense phosphorothioate oligodeoxy nucleotides (A19 or A19/2) targeting two different sequences around the origin of the open reading frame of CK19 mRNA. Both antisense oligonucleotides gave similar results, clearly contrasting with their corresponding randomized sequences used as a control. The downregulation of CK19 is only partial and temporary, but, at day 9 of CACO-2 cells confluency is sufficient to show a phenotype. As reported before, fractions of the cells were not targeted at all (36–39%, Table ), some cells were partially downregulated (54–62%, arbitrarily defined as cells displaying <50% of maximum total CK19 signal in three-dimensional reconstructions, Table ) and some were fully targeted, not showing CK19 at all (2–7%, cells displaying <10% of maximum CK19 signal, Table ). Although some variability in the CK19 signal was observed in control monolayers (either under no treatment or incubated in parallel with random or random/2 oligonucleotides), no cells with these low levels of CK19 (<50% of maximum) were observed in the controls (see Fig. 2, i–j in Salas et al. 1997).

Examples of cells downregulated in CK19 with A19 antisense oligonucleotide are shown in Fig. 3. The top panels (a–c) in Fig. 3 are three-dimensional projections of the whole stack of confocal sections in the XY plane, and, therefore, show the total CK19 content of the entire cell. Each panel shows at least one example of a fully targeted or partially targeted cells (Fig. 3, a–c, arrows), surrounded by neighbor nontargeted cells (Fig. 3, a–c, *) shown as an internal control. Examples of partially targeted cells are shown in Fig. 3b and Fig. c (arrows), while a fully targeted cell is shown in Fig. 3 a (arrow). To analyze the position of centrosomes in all these cells (Fig. 3, arrows and arrowheads), two XZ sections of each three-dimensional image were taken at the level of the centrosomes. The second row of panels (Fig. 3, d–f) are the XZ sections at the level of the centrosomes pointed with arrows, while the third row (Fig. 3, g–i) are XZ sections at the level of the centrosomes pointed with arrowheads. Totally or partially targeted cells are shown in the second row (Fig. 3, d–f), while their nontargeted neighbors are shown in the third row (Fig. 3, g–i, arrowheads) with the exception of the cell on the left hand side of i. The position of the apical domain and its cortical apical network of IF was hinted in targeted cells by the few remnant IF, that, in addition, were at the same level as those clearly visible in the nontargeted neighbors. In clear contrast with the control cells or the nontargeted cells in the same monolayers treated with A19, most cells downregulated in CK19 showed centrosomes 2–3 μm below the IF network (Fig. 3d, Fig. e, and Fig. i). In some cases, when the centrosomes were at the level of the apical IF network, they were always colocalizing with one of the remnant CK19 bundles (Fig. 3 f). Likewise, there was a correlation between the proportion of centrosomes not localized to the apical domain and the degree of success of the antisense treatment. In partially downregulated cells, 33–36% of the centrosomes were still apical, while none was found in totally downregulated cells (Table ). In all cases, however there was an increase in the proportion of centrosomes localized at the lateral boundaries of the cells (as identified by the vicinity to a neighbor nontargeted cell) (Table ). In the nontargeted cells, centrosomes were always colocalizing with the apical IF network (Fig. 3, g–i). Noncentrosomal γ-tubulin was also delocalized in cells downregulated in CK19 (Fig. 3d, Fig. e, and Fig. i). Since half of the noncentrosomal γ-tubulin is not in direct contact with the IF, but within a 2–3-μm range, this suggests that CK19 IF may also play an indirect role in the localization of this signal to the apical pole of epithelial cells. Both 21-mer A19 and A19/2 antisense oligonucleotides showed similar results (Table ), suggesting that the probabilities that these results are due to the downregulation of other, unrelated, mRNAs are very low. In all cases, the success of antisense downregulation was checked in parallel monolayers by immunoblot. Usually, a 50–70% decrease of CK19 signal was observed (Fig. 4), while CK18, CK8 (Fig. 4) or actin (Salas et al. 1997) were not affected. Similar results were observed with A19/2 (not shown).

Open in a separate window

Figure 4

Effect of anti-CK19 phosphorothioate antisense oligonucleotide on steady-state levels of cytokeratins in CACO-2 cells. CACO-2 cells were continuously incubated with the phosphorothioate oligodeoxy nucleotide A19, antisense against the CK19 mRNA (b, d, and f) or R, its randomized sequence (a, c, and d), as a control, for three passages. In this experiment, the cells were grown on 24-mm Transwell™ filters, and the cytoskeletal IF preparation was obtained by extraction in Triton X-100 in the presence of 1.5 M KCl. The pellets were analyzed by immunoblot using anti-CK19 (a and b), anti-CK18 (c and d), and anti-CK8 (e and f) mAbs. Molecular mass standards are expressed in kD.

Open in a separate window

Figure 3

Redistribution of centrosomes in CACO-2 cells depleted in CK19 by antisense oligonucleotide treatment. CACO-2 cells were continuously grown on Transwell™ filters in the presence of A19 oligonucleotide. The monolayers were fixed, processed for double immunofluorescence, and analyzed by confocal microscopy and deconvolution as described in Fig. 2. a–c show different examples of cells depleted in CK19 (red signal) with neighboring nontargeted cells (*), viewed in the XY plane (parallel to the monolayer). d–i are XZ sections (apical is shown up) of the same XY image above, at the level of the arrows (d–f) or the arrowheads (g–i). Arrows and arrowheads point at centrosomes. Bars, 5 μm.

Acrylamide Decreases the Colocalization of Centrosomes and Noncentrosomal γ-Tubulin with IF

To dissect the molecular mechanisms connecting γ-tubulin–containing structures to apical IF, I reanalyzed the distribution of γ-tubulin in CACO-2 cells treated with anticytoskeletal agents. Nocodazole and cytochalasin D were chosen because of their well known effects on microtubules and microfilaments, respectively. Unfortunately, there is no equivalent anti-IF drug available. A toxic effect reported for IF is that of acrylamide, that does not necessarily depolymerize IF, but may rather detach them from their anchors, for example, the desmosomes (Shabana et al. 1994). In all cases, long incubations (7 h) with these agents were used to allow for redistribution of the relatively large MTOC that may have exceedingly slow diffusion rates in the cytoplasm (Luby-Phelps et al. 1987). Neither nocodazole or cytochalasin D treatments resulted in a noticeable delocalization of apical centrosomes. In fact, none was observed detached from IF in our samples. A fraction of the centrosomes (33%), on the other hand, was observed within the cytoplasm away from CK19 IF after incubation of the cells in acrylamide (Table ). Interestingly, this particular effect of acrylamide was almost completely reversible after 24 h (Table ). Likewise, a count of 1,309 noncentrosomal γ-tubulin spots in the apical-most 2.4 μm of the cytoplasm in cells subjected to treatment with anticytoskeletal agents showed no differences between control cells, and cells treated with nocodazole or cytochalasin D. In all three cases ~40% of the noncentrosomal γ-tubulin colocalized with IF (Fig. 5, distance range 1), and the treatments did not cause major changes in the distribution profile (Fig. 5, distance ranges 2–9). In cells treated with acrylamide, on the other hand, the percent of γ-tubulin spots decorating apical CK19 IF fell to <10% (Fig. 5, arrow). It is suggestive that the fraction of γ-tubulin colocalizing with IF is similar to the percent of insoluble γ-tubulin.

A similar study was attempted using colocalization experiments of γ-tubulin and CK18. Unfortunately, the tight codistribution of CK19 and CK18 IF in the apical cortical cytoskeleton made this analysis very difficult, as we cannot determine whether a spot of γ-tubulin signal in that region is attached to a bundle of CK19 IF or to a neighboring bundle of CK18 IF, or both. However, an analysis conducted on the relatively less common γ-tubulin signal attached to the extensions of CK18 IF under the lateral domain (Fig. 2 c) yielded an intriguing observation. In cells treated with acrylamide, the number of sparse γ-tubulin spots colocalizing with lateral CK18 IF did not change. Instead, only in cells treated with cytochalasin D we observed a dramatic decrease in the γ-tubulin decorating lateral CK18 IF. This result, a preliminary suggestion that attachment to both types of IF may be mediated by different mechanisms, will be further analyzed in the next section using coimmunoprecipitation procedures.

To analyze the codistribution of CK19 IF and γ-tubulin at the ultrastructural level, the former was localized with Nanogold, extremely small colloidal gold particles coupled to Fab affinity-purified anti–mouse IgG, that have extensive accessibility to antigens in cells fixed in toto. The size of gold particles was then increased with a silver enhancer that deposits around preexisting gold. To make them visible at low magnifications, the silver was allowed to deposit for 20 min, rendering particles of 200–400 nm. The apical distribution of CK19, with some extensions to the lateral domain that we reported before using immunoperoxidase (Salas et al., 1987) was fully confirmed. In addition, it was verified that, according to specifications, Nanogold particles can diffuse within fixed/permeabilized cytoplasms (Fig. 6 a, an extension of the signal along the lateral domain). For colocalization, γ-tubulin was localized using an immunoperoxidase reaction in cells embedded in cross-linked albumin, that restricts the diffusion of the diaminobenzidine product. Centrosome images were found in ~1 out of 30 cells in nonserial sections, always in the apical pole of the cells. The immunoperoxidase reaction for γ-tubulin was found, as expected, in the pericentriolar material, observed at high magnification (Fig. 6 b, diffuse stain). The diameters of the peroxidase-positive images were always <400 nm, indicating that the peroxidase reaction product had not significantly diffused beyond the boundaries of centrosomes. For colocalization, the silver enhancing of Nanogold was kept to 5 min, rendering particles in the 15–20-nm range. This CK19 signal was observed on filamentous material in the apical region around the centrosomes. The minimum gold particle-to-peroxidase signal distances was found to be ~10 nm (Fig. 6 b, arrows). Gold particles were never observed within the pericentriolar material nor peroxidase reaction product decorating filaments.

Open in a separate window

Figure 6

Colocalization of γ-tubulin and CK19 IF at EM level. CACO-2 cell monolayers were grown on filters as described above and fixed in methanol. CK19 was localized with Nanogold™ gold particles followed by silver developing, while γ-tubulin was localized using a standard immunoperoxidase procedure. (a) CK19-Nanogold™ signal was developed with silver for 20 min to attain 200–400 nm particles visible at low magnification. (b) The CK19-Nanogold™ signal was developed with silver for only 5 min, so that the particles were in the range 15–20 nm, and colocalized with γ-tubulin detected by immunoperoxidase (diffuse stain) reaction confined within a matrix of cross-linked albumin. Note that the minimum distance from gold particles to peroxidase stain is 10–20 nm (arrows). Bar: (a) 1.6 μm; (b) 0.1 μm.

Colocalization of γ-Tubulin with Intermediate Filaments

The largest of the γ-tubulin structures were easily identified as centrosomes. All the noncentrosomal γ-tubulin signal was observed in discrete spot-like images. It should be noted that, because these images are the result of indirect immunofluorescence, they represent a slight overestimation of the real size. In fact, given the Stokes radius of the IgG (~7 nm) in the two layers of antibody covering a structure, one should add a total of ~28 nm to the original diameter of the structure. The smaller noncentrosomal spots, therefore, after subtracting the contribution of the IgG must be ~140 nm in diameter and are difficult to identify with previously described structures. This diameter represents nearly five times the diameter of the ring-shaped soluble γ-tubulin complexes described by Zheng et al. 1995. However, because these structures are at the limit of resolution of our instrument, they must be considered as points without any further estimation of their real size. Furthermore, neither of our anti–γ-tubulin antibodies recognized the epitope in aldehyde fixed cells, and were amenable to use only under methanol fixation, so that I cannot be sure if this discrete γ-tubulin signal is a fixation artefact. Therefore, I am not drawing any conclusion about the structure of noncentrosomal γ-tubulin signal, except that its localization in the apical pole of epithelial cells is consistent with the localization of noncentrosomal MTOC described by others (Meads and Schroer 1995).

Resolution of scanning laser confocal microscopy coupled to appropriate deconvolution analysis of three-dimensional images has been proven to yield resolutions at or below the 100-nm level in the XY plane (Rizzuto et al., 1998). Deconvolution processes can yield sub-pixel resolution, even for standard (nonlaser) epi-illumination (Carrington et al., 1995). We experimentally verified this fact in our system, together with the coalignment of the channels using fluorescent beads. The resolution in the z-axis, on the other hand, was slightly less reliable, and we experimentally verified it at ~300 nm. It must be noted that the colocalization of centrosomes and noncentrosomal γ-tubulin with IF at this level of resolution is more accurate than most colocalizations reported in the literature with standard epifluorescence microscopy, where resolution is 250 nm at best. It can be argued that IF in the apical network are very crowded and that any structure in the same region will randomly touch an IF bundle. While this may be true for centrosomes due to their size, this argument is unsustainable for noncentrosomal γ-tubulin. The surface occupied by IF signal at any XY confocal section is <50% of the image, and the gaps are significantly larger than the noncentrosomal γ-tubulin spots (Fig. 2d and Fig. e, small arrows). Yet, <10% of the γ-tubulin signal was not in direct contact with CK signal, indicating a high degree of colocalization. In the case of centrosomes, there was also a perfect colocalization in the z-axis within the ~1 μm thickness of the apical IF network. Considering that differentiated CACO-2 cells grown on filters are 10–20 μm tall, the position of centrosomes exactly at the level of the apical IF network can be considered a bona fide colocalization. The colocalization with F-actin with γ-tubulin, on the other hand, was good in the lateral domain, but poor in the terminal web, a result consistent with the possibility that different mechanisms of anchoring operate at different domains.

Additional evidence was provided by colocalization experiments at the EM level. While there is consensus in the precision of the localization with gold particles, some concerns may arise in regards to the peroxidase/diaminobenzidine reaction product that, potentially, may diffuse away from the site of the antigen. This diffusion can be efficiently confined within ~10 nm of the site of the antigen by performing the peroxidase reaction within a matrix of glutaraldehyde cross-linked albumin as described by Brown and Farquhar 1984, and also shown for membrane proteins (Vega-Salas et al. 1987b). Given this possible margin of error and the size of the gold particles themselves, the minimum distance between CK19 and γ-tubulin may be in the range of 10–20 nm. Because no overlapping of signals was observed, the binding interface between centrosomes and IF may be located at the outer boundary of the pericentriolar material (as labeled by γ-tubulin). In addition, it is possible that one or more intermediary proteins may be intercalated between CKs and γ-tubulin.

To our knowledge, this is the first report of colocalization of γ-tubulin with cytokeratin IF. Trevor et al. 1995, however, have reported colocalization of centrosomes with vimentin IF in cells that normally do not express IF, transfected with a vimentin-expressing vector, suggesting that this may be a more general function of IF.

Coimmunoprecipitation of γ-Tubulin–containing Structures with CK Intermediate Filaments

In general, coimmunoprecipitation is accepted as evidence of binding between two proteins. In this case, however, we attempted the difficult task of analyzing the protein interactions of the ~50% of γ-tubulin that is insoluble (Stearns and Kirschner 1994) and has not been previously characterized. To avoid denaturing conditions, we resorted, as in a previous publication (Rodriguez et al. 1994), to a fine homogenization of the Triton insoluble cytoskeletal preparation by sonication that yields multi-protein fragments. To apply this method for immunoprecipitation purposes we faced the limitation that the largest fragments (fractions 6–10 in our gradients) are heavy enough to pellet together with the agarose beads. Our original approach to this problem was to separate the beads from the unbound fragments by filtration through a 15-μm nylon mesh. Although this method successfully washes the beads, it has been our experience that, during the backwash of the filter, to recover the specific signal, sometimes there is loss of material, presumably beads retained in the filter or the filter holder. To avoid this and to ensure that the results would enable us to compare gradients from monolayers under different treatments, we decided to study only those fractions (1–5) that were amenable for standard immunoprecipitation, and to do so with careful parallel controls for nonspecific precipitation (Fig. 8 and Fig. 9, − lanes). The results indicate that we were able to precipitate multi-protein complexes in a specific antibody-dependent fashion, and which contained insoluble γ-tubulin and also either CK19 or CK18.

Interestingly, the results indicate that the complexes of γ-tubulin with the two types of IF were different in their migration in the gradient and their stability after treatments with anticytoskeletal agents. The CK19–γ-tubulin complexes displayed a broader range of sizes and were sensitive to pretreatment of the cells with acrylamide, while the CK18–γ-tubulin complexes were smaller, and sensitive only to pretreatment with cytochalasin D. These data are consistent with the morphological results (Fig. 5 and Table ) and suggest the possibility that MTOC may be bound to CK18 and CK19 via different mechanisms.

Because we were immunoprecipitating fragments of the cytoskeletal preparation or naturally occurring discrete particles, and not soluble proteins, careful thought has to be given to the possibility that Triton insoluble γ-tubulin–containing structures may be physically trapped within IF networks. An obvious prediction of a caging model is that structures of similar sizes, from the same domain must be caged together. If the CK19 cages can physically trap 300-nm centrosomes, they should also trap bundles of CK18-8 IF, especially since both sets of bundles are closely intermingled in the apical network. This was not the case in our experiments. If the γ-tubulin containing structures were physically trapped in fractions 2–5 of the gradients, it is inconceivable that no CK18 IF were trapped in the same cages (Fig. 8), even in the relatively large fragments of fractions 4 and 5. It must be remembered that CK18 was present in fractions 4 and 5 of the gradient before the immunoprecipitation (Fig. 7). In fact, the different behavior of CK18 and CK19 coimmunoprecipitation with γ-tubulin should lead us to conclude that two perfectly distinct sets of cages were generated by sonication: a small one (fractions 1 and 2) made of CK18 IF that can be opened with cytochalasin D but not with acrylamide, and another one, of a wider range of sizes made only of CK19 IF, sensitive to acrylamide but not to cytochalasin D. Such an exquisite specificity seems extremely unlikely for randomly generated three-dimensional lattices trapping structures inside.

A second line of experimental evidence against physical trapping comes from the visualization of the cytoskeletal fragments separated by the gradient under EM (Fig. 8, a and b). No evidence of three-dimensional lattices capable of trapping centrosomes was found. The structure of insoluble γ-tubulin is unknown and it has been tentatively equated to that associated with centrosomes (Stearns and Kirschner 1994). There is a possibility, however, that a fraction of the rather abundant noncentrosomal γ-tubulin may also be found in the insoluble cytoskeletal preparation. One can only speculate that the structure of this putative noncentrosomal insoluble γ-tubulin may be similar to that of the ring-like γ-tubulin particles (Zheng et al. 1995) in the soluble fraction, or, perhaps a multimer of them. In this regard, it is unlikely that the noncentrosomal γ-tubulin exists in pieces smaller than 28 nm. In this scenario, therefore, it is also unlikely that 40–50 nm fragments of the cytoskeleton in fraction 1 can sustain physical trapping of 28 nm (or bigger) γ-tubulin complexes or vice versa. Finally, because coimmunoprecipitation was also observed under totally different, more physiological, extraction conditions in the presence of phosphatase blockers (Fig. 8, KEB), and resistant to high salt (Fig. 9), we believe that an artefactual attachment of IF to MTOC due to detergent extraction is also very unlikely. On this basis, the coimmunoprecipitation results indicate that there is binding of γ-tubulin and cytokeratins to the same multiprotein complexes.

These complexes comprised γ-tubulin, the CKs, and about six other unidentified proteins. γ-tubulin was clearly more abundant than CK8 (Fig. 9), a result that suggests that some of the MTOC fragments, not associated with IF, are also being immunoprecipitated. However, it must be noted that CKs display turnovers of 100 h or more (Denk et al. 1987). Therefore, the 24 h labeling may be tagging 30% or less of the CKs. Two of the other proteins disassembled from the complexes in high salt buffer while the association with CKs was maintained. Because of the complex nature of the MTOC (Archer and Solomon 1994), this suggests that the sonication fragments contain only part of the MTOC, perhaps corresponding to the outer boundary of the pericentriolar region. The remaining five proteins were resistant to high salt extraction. Any of these proteins may be intermediaries in the attachment of γ-tubulin to IF, or alternatively the binding between γ-tubulin and CKs may be direct. Further research is necessary to establish the architecture of the protein-protein interactions at the centrosome–IF interface.

I am deeply thankful to Drs. Fulvia Verde (Department of Biochemistry and Molecular Biology) and Richard Rotundo for critically reading the manuscript. I would like to thank Ms. Yolanda Figueroa and Susan Decker for technical assistance.

This work was supported by a grant from National Institutes of Health (K14HL03397-01).