Actin filament turnover removes bundles from Drosophila bristle cells

We often think of actin filaments as dynamic polymers that assemble very rapidly to provide the cytoskeleton necessary for cell extension or movement and then depolymerize almost as abruptly. Examples of this type of activity include the extension of filopodia, pseudopodia and lamellipodia in neurons,neutrophils, keratocytes and fibroblasts and the formation of the actin tail of Listeria (reviewed in Borisy and Svitkina, 2000; Pantaloni et al., 2001; Pollard et al.,2000). There are other systems, however, in which the actin filaments are much more stable. These include the thin filaments in skeletal muscle and filament bundles found in a variety of systems such as Drosophila bristles and nurse cells, microvilli of intestinal epithelial, stereocilia of the cochlea, Sertoli ectocytoplasmic specializations and Limulus and Mytilus sperm (reviewed inBartles, 2000;DeRosier and Tilney,2000).

We presume that the stability of the filaments is due to actin binding and actin crosslinking proteins, as well as capping proteins on both ends of the filaments. For example, thin filament stability in muscle is due to the presence of tropomyosin, capZ and tropomodulin (reviewed inLittlefield and Fowler, 1998). Filament stability in parallel filament bundles is due to crosslinking proteins, for example, scruin in Limulus sperm(Sanders et al., 1996),fimbrin and espin in stereocilia (Bartles,2000; Tilney et al.,1989), fascin and forked in bristles(Tilney et al., 1995;Tilney et al., 1998), villin and fascin in nurse cells (Cant et al.,1994; Mahajan-Miklos and Cooley, 1994) and fimbrin, espin and villin in gut microvilli(Bartles et al., 1998;Bretscher, 1981;Bretscher and Weber, 1980;Glenney et al., 1981;Matsudaira and Burgess,1982).

Although actin filament turnover occurs in skeletal muscle(Littlefield et al., 2001;Michele et al., 1999) and microvilli (Stidwill et al.,1984), the bundles in these systems remain intact for very long periods. An extreme case is the stereociliary bundles in hair cells that remain intact for the life of the organism, which in humans can exceed 75 years. In fact, with the exception of the Drosophila bristles, stable bundles never disappear but instead the cells die before they completely disassemble their bundles. This is also true of the intestinal epithelial cells, although the microvilli can shorten with changes in diet(Weinman et al., 1989). In bristle cells, the actin bundles break down completely(Tilney et al., 1996;Tilney et al., 2000b), but the cell remains viable.

We have been studying actin bundles in forming Drosophila bristle cells for some time now. Bristles first sprout from the surface of the thorax in 32-hour pupae and gradually elongate, reaching a mature length in 48-hour pupae (Lees and Picken, 1944;Overton, 1967;Petersen et al., 1994;Tilney et al., 1996;Wulfkuhle et al., 1998). Close examination of the actin bundles in mature-length bristles(Appel et al., 1993;Tilney et al., 1995) shows that the bundles are composed of repeating units attached end to end, units we call modules (Tilney et al.,1996). On average each module is 3 μm long. Surprisingly, the gaps between the modules of one bundle are often in transverse register with the gaps in adjacent bundles. Furthermore, the basal end of each module is flat, whereas the tip is pointed or irregular in shape. Finally, the barbed ends of all the filaments in each module are located at the tip of the module(that is at the irregularly shaped end), and the pointed ends of the filaments are located on the base or flat end of each module(Tilney et al., 1996).

During the period of bristle growth, cuticular material is gradually deposited outside the plasma membrane so that by 40-42 hours a thin cuticle surrounds the entire bristle except at its growing tip. In thin sections of 54-hour pupae, the cuticle now supports the bristle, and the modules in the actin bundles have large gaps between them(Tilney et al., 1996). By 58 hours, the bundles and all of the actin filaments, in at least some bristles,have disappeared (Tilney et al.,1996). By 60 hours all the bristles have disappeared.

We have two motivations for studying actin bundle breakdown. First, we would like to know how a cell depolymerizes crosslinked actin bundles. Second,understanding the depolymerization process may reveal more about bundle formation during bristle elongation. What we found in these studies is in many ways unexpected. The disappearance of these crosslinked modules is a slow process and takes, on average, more than three hours to complete. In contrast,the process of module formation takes 30 minutes or less(Tilney et al., 1996). Breakdown of the bundles proceeds by two processes: the longitudinal splitting of bundles into subbundles and the disassembly of individual modules. Interestingly, these processes begin well before bristle elongation is complete even though actin assembly is the driving force for bristle elongation (Tilney et al.,2000a). These observations suggest that bundle breakdown is driven by actin-filament turnover. Furthermore, breakdown begins at the barbed ends of the modules. Thus, actin filament breakdown in crosslinked bundles is different from that in other more dynamic structures like the actin network at the edge of lamellipodia whose filament turnover takes place at filament pointed ends.

The Oregon-R strain of Drosophila melanogaster was used as the wild-type in these studies. The hs-GFP-moesin strain(Edwards et al., 1997;Kiehart et al., 2000) was generously provided by D. Kiehart (Duke University) and used to visualize actin filament bundles in living bristle cells. The hgm^2-insert was maintained over a SM6a balancer chromosome. Flies were maintained on standard cornmeal-molasses-yeast food at 25°C, 60-70%relative humidity, with a 12 hour/12 hour day/night cycle. Complete descriptions of genes and symbols can be found in Lindsley and Zimm(Lindsley and Zimm, 1992) and on FlyBase (Consortium,1999).

All animals were staged from the point of puparium formation, an easily recognizable and brief stage lasting 30 minutes at the beginning of metamorphosis (Bainbridge and Bownes,1981). White prepupae were collected and placed on double-sided scotch tape in a Petri dish and returned to the 25°C incubator. At the appropriate time after puparium formation the pupae were dissected.

After removing the pupal case, we filleted the pupae as outlined in detail elsewhere (Tilney et al.,2000a). The isolated thoracic fillets were placed in 60×15 mm petri plates with 5 ml of Grace's medium (GIBCO-BRL, Life Technologies,Inc., Rockville, MD) containing inhibitor. These compounds were diluted from concentrated stocks into culture medium just before use. Control incubations using DMSO, methanol or ethanol alone (typically 0.1-0.2%) were performed in parallel. After swirling the medium in the petri dishes to dilute the compounds, we added the isolated thoraces, and the plates were returned to the 25°C incubator for the indicated times. At the end of this period thoraces were removed and fixed for light or electron microscopy. The volume of medium used was just sufficient to cover the tissue and allow for good oxygen exchange. Stock solutions of cytochalasin D (Sigma Chemical Co., St. Louis,MO; 2 mM in methanol), cycloheximide (Sigma Chemical Company; 10 mg/ml in ethanol) and jasplakinolide (Molecular Probes Inc., Eugene, OR; 1 mM in DMSO)were stored in aliquots at -20° C.

The thoraces were fixed by immersion in 2% formaldehyde in PBS for 5 minutes, then washed three times in 0.1% Triton X-100 in PBS for 5 minutes each, then placed in 0.1% Triton X-100 containing 10^-6 M phalloidin conjugated to rhodamine (Sigma Chemical Co., St. Louis, MO) or Texas Red(Molecular Probes) at 4°C overnight in the dark. The next morning, the sample was placed on a slide with the ventral side downward, the excess fluid removed with filter paper and the thoraces mounted in glycerol Citifluor (Ted Pella Inc. Redding, CA). A coverslip was applied, excess Citifluor glycerol removed and the preparation sealed with nail polish. Slides were examined with either a Leica model TCS 4D confocal microscope or an Olympus Fluoview model BX50 confocal microscope. Bristles were visualized by the fluorescence of the phalloidin-stained actin bundles. For each procedure, at least 40 bristles were measured. The methods we used for electron microscopy are detailed in Tilney et al. (Tilney et al.,1998).

Dissected dorsal thoraces from GFP-moesin pupae were prepared as described above and cultured in Grace's medium on a microscope slide under a coverslip at room temperature (22-25°C). Images were collected every 30 minutes using an Olympus Fluoview model BX50 confocal microscope equipped with a 60× oil immersion 1.4 NA objective. GFP fluorescence was easily detected using the 488-nm line of the argon laser set at its minimum intensity(6%). Since a single bristle cell was examined during each observation period(typically 2-4 hours), the stage was not moved in the X-Y plane between scans although some adjustment in the Z plane was often necessary. Each time point resulted in 3-5 optical sections, which were projected together using the Image J application (http://rsb.info.nih.gov/nih-image/). The resulting confocal images were processed using Adobe Photoshop (Adobe Systems, Inc.). In some cases, a small amount of red fluorescent, carboxylate-modified 0.02 μm microspheres (Molecular Probes) was added to the thoraces prior to culture. A fraction of the microspheres adsorbed to the bristles and could be later used as fiduciary marks in the confocal images to align the time-lapse sequence in register. Image registration was also possible by alignment of fluorescent irregularities found on some of the bundles.

Although bristles do not reach their mature length until 48 hours after pupariation, changes in the actin bundles occur before maturity, even though new modules are still being assembled at the bristle tip. These changes are orderly, progressive and cumulative, and they begin in pupae of only 40 hours of age. In order to quantify these features, we examined 195 phalloidin-stained bristles from pupae 32-56 hours after pupariation and determined the fraction of bristles at each time point that had discernible characteristics involving the actin bundles.

When we examined bristles early in the process of elongation (e.g., 32-34 hours after puparium formation), we found fluorescent bundles that extend from the base of the bristle to a location near their tip. Modules were difficult to find. Instead the bundles were smooth and uninterrupted by gaps(Fig. 1a,b).

Before we found discrete gaps between modules, we found bristles in which the tip of one module overlaps with the base of the next module(Fig. 1c,d). Overlapping modules were particularly easy to detect when the bristle was artificially bent before fixation. Quantitatively, smooth bundles were the main feature of the youngest bundles (Fig. 1a,b), whereas overlapping bundles were featured in significantly older but still elongating bristles (Fig. 1c,d).

As bristles elongated further, discrete gaps appeared between adjacent modules, and the gaps in adjacent bundles were often in transverse register(Fig. 1e,f). The time when gaps first appear (43 hours) is important because the size of the gap increased with developmental age (see below). We note that there was weak rhodamine-phalloidin fluorescence in the gaps between modules, indicating the presence of a small population of actin filaments connecting adjacent modules. In addition, the lengths of the gaps in adjacent bundles often appear dissimilar.

As the bristle cell matured further, the size of the gaps increased. A detailed examination of the gap itself revealed another important feature. The basal end of each module tended to end abruptly and to be rounded, if not flat. In contrast, the apical end of each module tapered to varying degrees or was irregular (Fig. 1g,h). This tapering was prominent in 50-hour bristles(Fig. 1i,j). Since the gap between modules increased with pupal age as a result of the increasing taper of the apical end of each module, removal of subunits from filaments would logically occur from the tapered end nearest to the bristle tip from filament barbed ends.

This conclusion was strengthened by data from what appears to be the next step in module disassembly in which the apical ends of the modules lose most of their actin content, becoming tiny fluorescent threads that extend from the end of the taper toward the basal end of the module directly above it. These thin fluorescent `ghosts' become prominent in 53-hour bristles(Fig. 1i,j).

Analysis of the bristle populations that exhibit module shortening allows us to estimate the rate of module shortening. First, over 50% of the 43-55-hour bristles exhibit full-length modules defined by very short gaps on both ends (Fig. 1e,f). Thus,initiation of module breakdown occurs at an average bristle age of 49 hours. Second, over 50% of 53-56-hour bristles contain module ghosts that are one module or more in length (Fig. 1i,j). Thus, module breakdown is complete at an average bristle age of 54.5 hours. Consequently, an average module of 3 μm in length(Tilney et al., 1996) can be removed in 5.5 hours (54.5-49 hours) by barbed-end shortening at a rate of∼0.5 μm/hour (3 μm/5.5 hour). This rate is in reasonable agreement with the value determined more directly by time-lapse imaging (see below).

We examined thin longitudinal sections of bristle cells by electron microscopy to determine the details of bundle disassembly. The overlapping modules were most easily seen when the bristle was bent before fixation(Fig. 2a). At what we perceive to be the next stage in module shortening, we see gaps of increased length(Fig. 2b,c). Filament distribution appeared less uniform in the regions that would later become the gap as if there was depolymerization at one or both ends of the modules that previously overlapped. The gap then became apparent. Interestingly, the filaments on both ends of the modules showed the 12-nm transverse pattern(Fig. 2a-c) attributable to the fascin crosslink (Tilney et al.,1995), demonstrating that filaments in the bundles were maximally crosslinked right up to the gap. These images reinforce our contention, on the basis of light microscopy, that subunit loss occurred at the ragged end of each module, which is the end oriented toward the bristle tip.

In order to determine independently the polarity of module breakdown, we used time-lapse confocal microscopy. We visualized the actin bundles in living cells by decorating them with GFP fused to the C-terminal end of Drosophila moesin, a segment that contains an actin-binding domain(Edwards et al., 1997). Although this transgene is driven by the hsp70 heat shock promoter,we found that sufficient amounts of GFP-moesin accumulated in bristle cells at 25°C to allow for easy bundle detection by confocal microscopy. This transgene has no deleterious effects on the fly stock(Kiehart et al., 2000).

Thoraces from animals allowed to develop for 48-54 hours after white puparium formation were put into culture and visualized by time-lapse confocal microscopy. These culture conditions allow normal bristle development for a 6-7 hour period (Tilney et al.,2000a). We typically followed one bristle in each thorax for 2-4 hours, imaging it every 30 minutes. The resulting sequential images were aligned using either non-moving irregularities in the bundles or by including fluorescent microspheres in the preparation that adsorbed to the bristle cuticle. An example of one aligned time-lapse series is shown inFigure 3. What is obvious from this and other sequences is that there is differential breakdown of modules along the bristle length. For example, some regions showed little if any gap widening (Fig. 3, top arrows)whereas other regions exhibited distinct gap-widening activity(Fig. 3, bottom arrows) within this time window. Thus, there seems to be local control over module breakdown. This is apparent at two levels. First, laterally associated modules in local regions seem to be breaking down together as seen in the two gapped regions seen in Figure 3a. And second,those modules that constitute the ∼20 μm of bundled actin between these two gapped regions are not breaking down at all.

Careful examination of the aligned image sequence of the widening gap activity (Fig. 3, bottom arrows) shows that this widening results from barbed-end module shortening. The upper boundary of each gap corresponds to the basal end of the tip-ward module. These boundaries did not move during the 90-minute sequence,indicating that the filament pointed ends of the module were stable. The lower boundary of each bundle gap corresponds to the apical end of a module. These boundaries moved toward the base of the bristle during the time-lapse sequence, indicating that shortening of the modules occurred from the barbed ends of the actin filaments. Thus, using two independent approaches, length analysis of module populations from static images taken during the course of bundle breakdown (Fig. 1) and real-time analysis of individual modules(Fig. 3) lead to the same conclusion — modular bundles of actin filaments breakdown by subunit loss from their barbed ends.

We used these time-lapse images to directly measure the rate of module shortening. We found that the module base (filament pointed ends) that defined the upper boundary of each gap remained stationary relative to internal fiduciary marks (Fig. 3,asterisk) during the time-lapse series. Accordingly, we measured the distance from this boundary to the apical end (filament barbed ends) of the retreating module in the gap at each time point. For each gapped bundle shown inFigure 3, we plotted increasing gap length (as a measure of module shortening) as a function of time and used the least squares method to calculate the linear regression equation to fit the data. The slope of the resulting line represents the rate of module shortening. The average rate of shortening exhibited by the widening gaps shown in Figure 3 (lower portion) was 1.10±0.04 μm/hour(Fig. 3c). Module shortening was almost undetectable in non-widening gaps(Fig. 3b). Thus, the rate of barbed-end shortening determined by time-lapse microscopy of individual modules (1.1 μm/hour) is in reasonable agreement with the value (0.5μm/hour) determined by analyzing static images of module populations during the course of bundle breakdown.

Jasplakinolide is a compound capable of stabilizing actin filaments by both lowering the off-rate and increasing the onrate of actin subunits at the barbed ends (Bubb et al., 1994;Bubb et al., 2000). In fact,growing bristles cultured in the presence of jasplakinolide exhibit accelerated elongation rates, indicating that actin polymerization drives bristle-cell elongation (Tilney et al.,2000b). We reasoned that if actin depolymerization was responsible for module shortening then jasplakinolide could prevent this. To test this idea, we cultured bristles from 48 hour pupae for 5 hours in the presence or absence of jasplakinolide and evaluated actin bundle morphology by confocal microscopy. As expected, bristles incubated in the absence of jasplakinolide developed bundles that contained gaps and tapers typical of bristles of this age (Fig. 4a). In contrast,bristles cultured in the presence of jasplakinolide failed to undergo module breakdown and exhibited smooth non-gapped bundles typical of much younger bristles (Fig. 4b).

Isolated thoraces from 36-hour pupae were cultured for 5 hours in the presence of the protein-synthesis inhibitor cycloheximide. Bristle elongation was unaffected by cycloheximide (Fig. 5a). By confocal microscopy, we saw the expected overlapping modules and, in one case, gaps in the bundles. However, the cross-sectional area of the bundles was approximately four-fold smaller than the controls. Even so, the filaments in these small bundles were hexagonally packed(Fig. 6) and maximally crosslinked.

In contrast, pupae cultured with cycloheximide from 42-44 hours showed many examples of gap formation and premature bundle breakdown. Almost 75% of the bundles showed tapering modules and even ghosts(Fig. 5b), which were not normally seen in the controls until later in development, typically 50-56 hours (e.g., Fig. 1g-j). In thin sections, we saw bundles breaking apart into subbundles (not shown),again a stage normally seen in 48-53-hour bristles but not at this stage.

In longitudinal EM sections through modules of pupae of 54 hours or more,we found that the once solid modules began to separate longitudinally into a series of thinner subbundles (Fig. 7d,e). The size of the subbundles varied, but in many cases there were six or seven filaments in longitudinal sections, indicating that if these subbundles were round they would be composed of 25-50 filaments. The filaments in each of the subbundles displayed the 12-nm period attributable to the fascin crosslink (Fig. 7c), and in transverse section the filaments in the subbundles were hexagonally packed(Fig. 7a,b). This means that the filaments in the subbundles are maximally crosslinked together. As the pupae aged, many of the subbundles separated from those still attached to the plasma membrane (Fig. 7a,b). These were often found deep within the cytoplasm of the cell.

We also studied the stages of subbundle formation. The splits in the bundle appeared first at the frayed apical end of a module(Fig. 7c) not the flat basal end. Later these splits ran the length of the bundle proper(Fig. 7b). This invariably cut the bundles into two pieces. The smaller piece remained connected to the plasma membrane while the other piece, containing about two-thirds of the filaments, lay deeper in the cytoplasm(Fig. 7b,d). Over time, these large subbundles split further into smaller subbundles composed of 25-50 hexagonally packed crosslinked filaments.

The formation of the subbundles could also be followed in transverse sections through modules. The first step in subbundle formation seemed to be the appearance of holes or areas in the large bundles that lack filaments(Fig. 8). These bundles almost look like the end of a pallet of lumber from which several internal planks were removed. Most often, the holes first appeared five to six rows inwards from the plasma membrane.

Isolated thoraces cultured in Grace's insect medium elongate bristles at normal rates over a 6-7-hour period(Tilney et al., 2000a). Bristle elongation stops abruptly when thoraces isolated from 36-hour or 41-hour old pupae are cultured in the presence of cytochalasin D. However, the bundles that were formed prior to treatment remain(Tilney et al., 2000a). These bundles fragment into subbundles because there are many more bundles than in the controls (Fig. 9a,b). These subbundles are correspondingly thinner than the bundles observed in controls. We also found a population of randomly oriented small fluorescent units at the base of the bristle shaft and in the cytoplasm around the nucleus and Golgi of the bristle cell (Fig. 9b,c). On average, each of these units was 3 μm, the average length of the bristle modules (Tilney et al., 1996). It looks as if the bundles from the basal end of the bristle have not only separated into the subbundles but these subbundles have also moved into the cell body.

In thin sections through the bristle shaft we can view this process at higher resolution. After only two hours in cytochalasin, we found, instead of 7-11 large bundles in a microchaete, 27 small bundles located near the cell surface as well as a few internally located bundles (data not shown). Each bundle is thinner than the 7-11 seen in the wild-type at the same stage, and some of these were in the process of splitting into even smaller subbundles.

In our thin sections through the base of the bristle cell in the thorax proper, in the plane that includes the nucleus and the Golgi apparatus, we could resolve the randomly oriented ∼3 μm long bundles(Fig. 10) seen previously by confocal microscopy. Each of these bundles was composed of approximately 50 filaments rather than the 600-1000 filaments that typically make up each of the 7-11 bundles present in the shaft of fully elongated bristles. In essence,cytochalasin promotes the appearance of the subbundles that normally associate laterally to form each module. Of interest to our subsequent discussion is the observation that the 12-nm transverse period is readily apparent in longitudinal sections of these subbundles(Fig. 10b), indicating that the filaments in these subbundles remain maximally crosslinked. It is also interesting that bristle cells incubated in cytochalasin do not form gaps prematurely. Thus, cytochalasin does not effect the rate of depolymerization of the bundles from their barbed ends. However, cytochalasin induces a premature splitting of the modules into subbundles, reminiscent of the process that occurs in bristles from pupae 50 hours old or older. Also, at the base of the bristle, the subbundles from the modules become randomly oriented in the cytoplasm of the cell body.

The actin bundle in developing Drosophila bristle cells is a useful model for studying the in vivo biology of a functional cytoskeletal unit composed of many proteins acting in concert. In earlier studies, we showed that although actin assembly is essential for bristle elongation(Tilney et al., 2000a), once the bristle elongates to its mature length the actin filament bundles disappear and the cell shape is maintained by the chitinous exoskeleton(Tilney et al., 1996). We show here that modules breakdown in two ways. The filaments shorten from their barbed ends, and the modules split longitudinally into smaller subbundles(Fig. 11). This process literally represents bundle assembly in reverse, in that actin is now lost from the barbed end of each filament and the large maximally crosslinked bundles now split longitudinally into smaller subbundles. Interestingly, we find that inhibitors of both actin filament assembly and protein synthesis cause bundles to breakdown prematurely. Further, breakdown is prevented if the filaments are stabilized by jasplakinolide, a membrane-permeant phalloidin-like compound. These results indicate that bundle breakdown is driven by actin filament turnover.

We find that even as actin polymerization and bundle formation at the bristle tip drives bristle elongation(Tilney et al., 2000a;Tilney et al., 2000b),transverse gaps develop in the older portions of these same bundles near the base of the bristle. Once the bristle reaches its mature length (48-hour pupae), these gaps increase in size until all the F-actin disappears (60-hour pupae). Thus, actin filament loss occurs over many hours.

We conclude that the subunits are being selectively removed from the apical ends of the modules for three reasons. First, the basal end of each module is nearly flat and maintains its shape at all stages. In contrast, by confocal microscopy, the apical end is ragged with tiny threads of F actin extending distally. Second, the basal ends seem to hold their relative positions in the bristle during development, as evidenced by the tendency of these ends to remain in transverse register. In contrast, the apical ends of adjacent modules can be highly variable in length. These apical ends shrink toward the flat end of the module, leaving behind their subbundle ghost still attached to the plasma membrane. Since the apical end corresponds to the barbed end of the actin filaments (Tilney et al.,1996), subunits must be removed from this end and not from filament pointed ends at the module base. Third, we have directly monitored module breakdown in living cells by time-lapse microscopy. It is clear from these images (Fig. 3) that actin subunits are lost from the filament barbed ends. This conclusion agrees with our interpretation of our confocal `snapshots' of modules in increasingly older bristles (Fig. 1). Furthermore, when jasplakinolide is used to prevent depolymerization, module shrinkage does not take place (Fig. 4).

Module breakdown is much slower than module synthesis. Although we know that the rate of bristle elongation increases as bristles grow longer(Tilney et al., 2000a), we can estimate the average rate of elongation of 70 μm microchaetes during their 16 hour growth period to be at least 4.4 μm/hour and of 250 μm macrochaetes to be at least 15.6 μm/hour. Since bristle growth is driven by actin polymerization (Tilney et al.,2000a), these are reasonable estimates for module growth as well. Careful analysis of module breakdown using time-lapse microscopy(Fig. 3c) indicates that module barbed-end shortening occurs at a rate of ∼1 μm/hour. This depolymerization rate is two orders of magnitude slower than pointed-end depolymerization and actin turnover at the leading edge of lamellipodia, which must match the rate of membrane protrusion - up to 5-10 μm/min(Cassimeris et al., 1990;Svitkina and Borisy, 1999). This relatively slow rate of filament breakdown in bristle bundles could result from the actin filaments being maximally crosslinked, or it could be because the ends of the filaments are capped or both. Interestingly, the actin bundles in microvilli of intestinal epithelial cells also shorten slowly in vivo, at a rate of ∼0.2 μm/hour, when treated with lectins(Weinman et al., 1989), a rate comparable to what we observe in bristle modules.

Our conclusion that subunit loss from the barbed end of the filaments in the bundles is particularly interesting as it conflicts with prevailing views of subunit loss in other systems, where treadmilling occurs by pointed-end loss dependent on `depolymerizers' like ADF/cofilin. Two things should be kept in mind. First, in bundles found in bristles, microvilli and stereocilia, the actin filaments are maximally crosslinked together by two or more protein crosslinkers. These crosslinks could presumably inhibit actin turnover. Second, the rate of subunit loss from these crosslinked bundles is orders of magnitude slower than the rate of subunit loss from actin meshworks found in lamellipodia or in Listeria tails (e.g.,Cassimeris et al., 1990;Loisel et al., 1999;Svitkina and Borisy, 1999;Theriot and Mitchison, 1991). Thus, how crosslinked actin filaments depolymerize when they exist in bundles may be very different from how dendritic arrays of actin filaments and/or free actin filaments disassemble. In short, there is no conflict, and instead there are real differences.

When gaps appear between modules, individual modules begin to split longitudinally into subbundles containing 25-50 filaments each. This splitting begins at the apical (barbed) end of each module where actin depolymerization is occurring. After splitting, some of the subbundles move to the center of the bristle. The filaments in the resulting subbundles are still hexagonally packed and display the 12-nm period in longitudinal section attributable to fascin, indicating that they remain maximally crosslinked together.

The normal process of module breakdown into subbundles, a process that is accelerated by cytochalasin (an inhibitor of actin polymerization) and cycloheximide (an inhibitor of protein synthesis), seems to represent a reversal of module assembly. For example, from studying the developmental formation of modules, we know that they are formed from subbundles(Tilney et al., 1996) and that the forked crosslinker plays a key role in assembling the subbundles into modules (Tilney et al., 1998). Thus, these subbundles are present in both pupae, whose bristles are almost the mature length, and in younger, still elongating bristles. Therefore, the subbundles are not only a product of bundle breakdown but also represent a stage in the formation of large bundles.

There seems to be two different populations of filaments in these actin bundles: the hexagonally packed filaments that constitute the subbundles plus the cytochalasin-sensitive filaments that `glue' the subbundles together. The subbundle filaments seem to be very stable and only shorten late in development, whereas the glue filaments are more dynamic because they are lost soon after cytochalasin treatment. It is interesting to note that even the glue filaments contribute substantially to bundle cross-sectional area. Modeling the construction of a 1000-filament bundle requires approximately 750 subbundle filaments (30×25-filament subbundles) and approximately 250 glue filaments to hold everything together.

Our inhibitor studies on growing bristles show that actin bundles are not static structures but require constant maintenance to maintain their structure. For example, treatment with cytochalasin D blocked bristle elongation and led to the splitting of the already-assembled bundles into smaller subbundles. However, the bundles still extended continuously from the base to the tip of each bristle, and no gaps were seen. Thus, the aggregation of the many smaller subbundles into the 7-11 larger bundles requires continual actin filament polymerization and presumably filament crosslinking. Perhaps the lateral assembly of subbundles into the larger bundles is imprecise,leading to subbundle interfaces that are not maximally crosslinked with fascin and are subject to breakdown. The resulting splits may be repaired by new actin filament synthesis within the split and subsequent crosslinking. In fact, such imperfections in the hexagonal lattice can be seen normally during module development. This is presumably due to the failure to zipper subbundles together perfectly during the early phases of module formation. In contrast,inhibiting protein synthesis with cycloheximide does not block bristle elongation but does cause the appearance of gaps within the bundles and the breakdown of the modules from their apical barbed ends.

Although the precise details and molecular mechanisms of subunit removal remain to be determined, bundle breakdown seems to be the inevitable consequence of a failure to balance filament breakdown with filament replacement that normally maintains module and bundle integrity. A relevant analogy for this is the `classic' math problem of filling a bath without the drain plug in place. As long as water enters the bath faster than it can escape down the drain, the bath will fill. In the case of actin polymerization, if filament bundling and the maintenance processes exceed any breakdown processes, the bristle will elongate by actin polymerization. Once the volume of water filling the bath falls below the volume escaping, the bath will inevitably empty, the rate (and time) of emptying being determined by the difference between the two processes. We imagine that once the bristle becomes fully grown, the processes of actin polymerization, filament bundling and bundle maintenance are downregulated or stopped while filament breakdown continues or speeds up, resulting in the disassembly of the crosslinked actin bundles.

We would like to express our thanks to Dan Kiehart for generously making available the hs-GFP-moesin stock. This work was supported by grants from the National Science Foundation (MCB-0077839) and the University of Pennsylvania Research Foundation to G.M.G. and the National Institute of Health (GM-52857) to L.G.T.