Contribution of sequence variation in Drosophila actins to their incorporation into actin-based structures in vivo

Actin is one of the most abundant and highly conserved proteins in eukaryotes (Sheterline et al., 1999). As a major cytoskeletal component it is essential for a variety of cellular functions including determination of cell shape, cell motility, cytokinesis, intracellular transport and muscle contractility. Filamentous actin, F-actin, occurs in many different forms: as cortical actin found in all cells, as bundled or branched assemblies building up filopodia and lamellipodia, as thin filaments found in muscle sarcomeres, and many more. We are interested in understanding what determines the organisation of actin into these different filamentous structures. Several studies have shown that certain accessory proteins are only associated with specific actin structures, and some studies indicate that these proteins can even dictate the type of filamentous actin-based structure that will be formed, e.g. filopodia versus lamellipodia (Mejillano et al., 2004; Revenu et al., 2004). Alternatively, the formation of particular actin structures could be dictated by the actin protein itself, since multicellular organisms contain several actin proteins. If this were the case, the difference in actin proteins could be used as a simple tool to label distinct actin structures, e.g. by expressing fluorescent protein-tagged versions of certain actins.

The genomes of mouse, human and fly contain six actin gene loci (Fig. 1A). Actin proteins differ in only a few amino acids within a species and even between distant species such as fly and mouse. The high conservation at the protein level raises the question as to why multiple actin genes have been retained through evolution. In many cases, multigene families arise through duplication of ancestral (single copy) genes and often diverge in the function of the encoded proteins or the regulation of the genes or both, making all copies essential over time (Meyer and Van De Peer, 2003). For actin proteins, it is not clear which of these options has occurred. In the mouse, two of the muscle actins are specifically expressed in striated muscles (α-skeletal and α-cardiac) and two in smooth muscles (α-vascular and γ-enteric); the other two actins are non-muscle actins, also termed cytoplasmic actins (Sheterline et al., 1999). Similarly to mouse, Drosophila has four muscle-specific actins, with actin 87E being expressed throughout life, actin 57B being expressed in embryonic and larval muscles and actins 79B and 88F being mostly expressed in pupal and adult muscles (Fig. 1C). In addition to the temporal restriction of expression, the fly actins are also restricted in their spatial expression patterns (Fyrberg et al., 1983).

As well as regulation of actins in different cells by transcriptional control, actin production at specific locations within a cell can also be regulated, by subcellular localisation of actin mRNAs. For example, the mRNA for vertebrate β-actin is selectively enriched at the motile cell periphery (Hill and Gunning, 1993), and a short cis-acting `zipcode' in the 3′ untranslated region of this mRNA is necessary for enrichment (Kislauskis et al., 1994). Similarly, in immature cortical neurons β-actin mRNA and protein are enriched in growth cones, whereas γ-actin mRNA is found in the cell body and the protein is spread uniformly throughout the cell (Bassell et al., 1998). Thus, the different actin genes are certainly regulated differently, at both transcriptional and post-transcriptional levels. Nonetheless, even small differences in primary sequence could affect the function of actins.

If the type of actin has an influence on the type of actin structure formed, in addition to the effects of actin-binding proteins, then we would expect several actins to be co-expressed within one cell. In vertebrate and mammalian tissue culture cells at least two to four appear to be present in each cell (Song et al., 2000). In Drosophila, one cytoplasmic actin, 5C, is thought to be expressed in every cell plus an additional muscle-specific actin isoform in every muscle cell (Fyrberg et al., 1983), but recent studies indicate that low levels of other actins are present in various tissues (Nongthomba et al., 2001).

Data from mammalian cells have produced conflicting results as to the influence of the actin protein sequence on its ability to incorporate into different actin structures. Amongst endogenous actins there seems to be a preference for cytoplasmic actins to be incorporated into the motile cell edges rather than stress fibres (DeNofrio et al., 1989; Hoock et al., 1991), but actin fibres isolated from smooth muscles have been shown by electron microscopic analysis to have both smooth muscle and cytoplasmic actins present within the same filament (Drew et al., 1991). Incorporation of ectopic actins produced varied results. For example, one study found that injected muscle and non-muscle actins showed identical patterns of distribution in both fibroblasts and myocytes (McKenna et al., 1985), whilst another study showed that transfected muscle and non-muscle actins are differentially distributed in cultured smooth muscle and non-muscle cells (Mounier et al., 1997). These and other studies are difficult to compare as they use very different methodologies and have only addressed subsets of the actins present in one species.

In Drosophila, all six actins are more closely related to vertebrate cytoplasmic actins than to any of the muscle-specific vertebrate actins, with the two fly cytoplasmic actins being the most similar to the two vertebrate cytoplasmic actins (Fig. 1B). Despite the higher similarity between muscle and non-muscle actins in the fly, the actin structures found in Drosophila do not appear significantly different from those observed in mammals. Even though all fly actins are closely related, functional substitution tests have shown that some can compensate for loss of another actin while others cannot. The deleterious lack of cytoplasmic actin5C can be rescued by reintroducing the coding sequence of the second cytoplasmic actin, 42A, under the regulatory control of act5C elements (Wagner et al., 2002). By contrast, lack of the indirect flight muscle-specific actin88F, which renders flies flightless, can only be rescued by expression of one out of the five other actins, with the failure to substitute correlating with the extent of sequence difference between actin88F and the actin used to rescue (Fyrberg et al., 1998). The underlying mechanistic reasons for these differences in the functional equivalence are unclear.

The goal of this work was to perform a single comprehensive study to compare the ability of all the actins from one species to incorporate into different actin structures, using the same methodology throughout and analysing a large variety of tissues. At the same time, we tested the usefulness of the transgenic fly lines we generated as tools to selectively label subsets of actin structures. To this end, we have expressed all six actin genes in Drosophila as green fluorescent protein (GFP) fusion proteins using the UAS-Gal4 system (Brand and Perrimon, 1993) to drive expression in various tissues. We found that, with only a few exceptions, each actin was efficiently incorporated into all of the different structures formed by the endogenous actins that we analysed. The most drastic difference in the site of incorporation of the actins was observed in the sarcomeres of muscles. In contrast to some of the mammalian cell culture studies we found no large difference in incorporation between cytoplasmic and muscle-specific actins, and did not observe strong dominant-negative effects of the overexpression. We hope that in addition to aiding in the understanding of the effect of actin sequence variability on the formation of actin structures, the tools presented here will prove useful for subsequent studies on actin dynamics in Drosophila.

Transgenic lines of GFPactins were crossed to the following Gal4-drivers for tissue-specific expression: ptc-Gal4 (Hinz et al., 1994), mef2-Gal4 (Ranganayakulu et al., 1996), CY2-Gal4 (Queenan et al., 1997), nanos-Gal4 (Van Doren et al., 1998) and 24B-Gal4 (Brand and Perrimon, 1993). The UASp-GFPactin lines generated in this study are available from the Bloomington Drorophila Stock Center (http://flystocks.bio.indiana.edu).

Flight ability was tested by comparing actin-expressing flies to flies carrying the GFPactin but not the Gal4-driver transgene for their ability to fly up in a transparent cylindrical cage after being shaken to the bottom: wild-type flies fly up to the top of the cage whereas flightless flies have to climb up the sides.

Actin gene sequences were amplified with specific primers in the 5′- and 3′-UTRs from genomic DNA [prepared as described previously (Huang et al., 2000)] and subcloned into pCR TOPOII. Using primers specific for the coding sequences that added NotI and XbaI sites (or a BamHI site in the case of actin79B) the actin sequences were re-amplified and cloned into a pUASp(mGFP6) vector (Röper and Brown, 2003; Rørth, 1998). The GFP was fused to the actin N-terminally, with a short linker sequence of SSSAAA in between the GFP and the first methionine of the actin protein sequence. All sequences were verified by sequencing.

Embryos were collected on apple-juice agar plates at stage 14-15 of development, and processed for immunofluorescence using standard procedures. Ovaries were dissected from well-fed females and processed for fluorescence using standard procedures. Third instar larvae were either imaged live, or opened, fixed in 4% formaldehyde and stained with phalloidin in PBT (PBS plus 0.5% bovine serum albumin and 0.3% Triton X-100). Indirect flight muscles were dissected from well-fed 1- to 3-day-old adult flies, fixed in 4% formaldehyde and stained with phalloidin in PBT. Confocal images were obtained using a Radiance 2000 confocal microscope (Bio-Rad, Hemel Hempstead, UK). Confocal laser, iris and amplification settings in experiments comparing intensities of labelling were set to identical values. Confocal pictures were assembled in Adobe Photoshop.

The coding region of each of the six actin genes in Drosophila was amplified by PCR and cloned (see Materials and Methods for details). Constructs were then generated containing the modified UASp promoter (Brand and Perrimon, 1993; Rørth, 1998), the GFP coding sequence and actin protein coding sequence. GFP was fused to the N terminus of each actin, as tagging at this end has been shown to be less disruptive for polymerization in a study using His-tagged actin88F (Brault et al., 1999a). Also, the function of the widely used UAS-GPFactin5C (Verkhusha et al., 1999) confirms that tagging at the N terminus allows tagged monomers to be incorporated into actin filaments. For each GFPactin construct, several independent transgenic lines were generated.

We then tested the ability of each GFPactin to be incorporated into a variety of the actin based structures formed by the endogenous actins in Drosophila: (1) branched networks and parallel bundles of actin filaments such as lamellipodia and filopodia, which are easily seen in the embryonic epidermis and amnioserosa; (2) cortical actin and stress fibre-like structures, found for instance in the ovarian follicle epithelium; (3) specialized actin structures found in the female germline such as ring canals and actin cages; (4) highly ordered actin-myosin assemblies found in the sarcomeric structures of embryonic, larval and adult muscles. To do this, we expressed each GFPactin in the relevant tissue with the appropriate Gal4 driver, and visualized GFP expression in fixed and live samples. All GFPactins appeared to be expressed at similar levels, judging by the brightness of the fluorescence and western blot analysis (data not shown). We also examined whether the expression of each GFPactin perturbed the endogenous actin structures by double-labelling with rhodamine-phalloidin to reveal the overall distribution of filamentous actin. Actin overexpression did not detectably impair normal filamentous actin assembly, because in all tissues analysed, the phalloidin-labelling in wild-type cells and those expressing GFPactin was indistinguishable. In most cases GFPactin overexpression did not interfere with cellular function, but the few exceptions that we noted are described below.

Lamellipodia and filopodia are actin-based cellular extensions important for cell-cell contact and cell migration (Jacinto and Wolpert, 2001; Small et al., 2002; Svitkina et al., 2003). Actin filaments are arranged in parallel bundles in filopodia and branched networks in lamellipodia. A variety of accessory proteins have been discovered that can determine which of the two structures is formed (Mejillano et al., 2004; Revenu et al., 2004), but the contribution of actins has not been examined. In the fly embryo, elaborate filopodia and lamellipodia are found at the interface between the embryonic epidermis and the amnioserosa (Fig. 2A). The amnioserosa is an extraembryonic tissue covering the dorsal region of the embryo, which is eventually displaced by the embryonic epidermis (Jacinto et al., 2002). The amnioserosa cells also have numerous lamellipodia and filopodia.

Expression of the GFPactins was driven in the embryonic epidermis using the patched-Gal4 driver (Hinz et al., 1994), leading to expression in segmentally repeated stripes within the epidermis and `patchy' expression within the amnioserosa. The two cytoplasmic GFPactins, actin 5C and 42A, and the muscle-specific actin88F highlighted the leading edge filopodia very strongly compared with the overall cytoplasmic labelling (Fig. 2B,C,G), with GFPactin88F showing the most striking enrichment in the filopodia (Fig. 2G). The other three GFPactins, actin57B, 79B and 87E, displayed a higher level of cytoplasmic staining (Fig. 2D-F) compared to the amount in the filopodia. GFPactin79B was the only actin that did not appear to be incorporated into lamellipodia and filopodia within the amnioserosa (Fig. 2E), whereas all other GFPactins labeled lamellipodia to varying degrees, with 42A and 88F being the strongest (Fig. 2C,G). Of the two cytoplasmic actins, GFPactin42A appeared to be the one that was better incorporated into filopodia and lamellipodia. Comparison of the phalloidin labelling of GFPactin-expressing and non-expressing cells showed that the overall distribution of filamentous actin was identical and that cell morphology was not altered.

The follicular epithelial cells, which surround the maturing oocyte and nurse cells, provided good examples of two other actin structures: the cortical actin cytoskeleton and stress fibres (Fig. 3A). Cortical actin is the actin associated with the plasma membrane and is thought to perform diverse functions, such as anchoring of transmembrane and peripheral membrane proteins, cell adhesion and endocytosis (Arpin et al., 1994; Bretscher, 1991; Schafer, 2002; Vasioukhin and Fuchs, 2001). Stress fibres are particularly prominent in cells in culture that are well spread on an extracellular matrix. They consist of bundles of actin filaments spanning the cell that are interspersed with myosin II and are thus thought to be contractile (Kolega et al., 1991). One of the few examples of similar structures that can be found within Drosophila are the parallel arrays of actin fibres at the basal surface of the follicle cells. Like stress fibres, these fibres terminate at a membrane structure containing integrins (Bateman et al., 2001).

When we expressed GFPactins in the follicle epithelium using the CY2-Gal4 driver (Queenan et al., 1997), each actin was incorporated into both structures, but with differing efficiency (Fig. 3). Actin87E showed especially strong cortical actin labelling (Fig. 3F), whereas actin79B was mostly cytoplasmic and only weakly labelled the cortex (Fig. 3E). The four other actins labelled the cell cortex at intermediate levels (Fig. 3B-D,G). Incorporation into the basal stress fibres was strongest for the cytoplasmic GFPactin42A (Fig. 3I), strong for 5C, 87E and 88F (Fig. 3H,M,N), and only weak for 57B (Fig. 3K). GFPactin79B very poorly incorporated into the basal actin fibres (Fig. 3L). For both actin structures present in follicle cells, the cortical actin and the stress fibre-like arrangements, the cytoplasmic actin42A appeared a better marker than the widely used actin5C.

We next examined two actin structures within the female germline that are specializations of this particular tissue, but nonetheless related to actin-based structures in other tissues. Ring canals are stable cytoplasmic bridges between neighbouring germ cells that are formed from an arrested cleavage furrow by addition of several actin-binding proteins (Mahowald, 1971; Robinson et al., 1997; Sokol and Cooley, 1999). At stage 10-11 of oogenesis, actin struts form `cages' around the nurse cell nuclei and are thought to hold these nuclei in place during the pumping of cytoplasmic content of the nurse cells into the oocyte (Guild et al., 1997). The struts are built up of smaller overlapping bundles of actin filaments, each of which resembles actin bundles in microvilli (DeRosier and Tilney, 2000).

When expressed in the germline using the nanos-Gal4-VP16 driver (Van Doren et al., 1998), all GFPactins strongly labelled the ring canals within the germarium and early egg chambers (Fig. 4A-F) and also at late stages of oogenesis (some ring canals at these stages are visible in Fig. 4I-O). All actins except actin79B were incorporated into the actin struts around nurse cell nuclei (compare Fig. 4M with I,K,L,N,O).

Overexpression of two of the GFPactins interfered with normal germline development. Expression of either GFPactin5C or 57B caused a failure in formation of germline cysts with the correct number of germ cells in a fraction of ovarioles (Fig. 4C,G,H), indicating that the mitotic divisions generating the germline cysts were perturbed. Germ cells were multinucleate and ring canals appeared to contain excess actin (Fig. 4G). Nonetheless, late stage egg chambers could be found that contained a morphologically identifiable oocyte despite having too few germ cells (Fig. 4H, asterisk).

One of the most prominent actin-based structures in animals is the sarcomeric assembly found in striated muscles. In a nearly crystalline array, actin-based thin filaments and myosin-based thick filaments are arranged in a hexagonal lattice. The actin filaments originate and overlap in the Z-lines, where multiple actin binding and capping proteins help to anchor the barbed ends of actin filaments (Clark et al., 2002). The Z-lines are also anchored to the plasma membrane where the ends of muscles are linked to the tendon matrix on the outside of the muscle via integrin junctions forming an enlarged `modified' Z-line (Reedy and Beall, 1993b). Phalloidin staining is strongly enriched in Z-lines, indicating a concentration of filamentous actin.

GFPactins were expressed in muscles using the mef2-Gal4 driver (Hinz et al., 1994). In the larval visceral (Fig. 5A) and body wall muscles (Fig. 5B,C) all actins were incorporated into the striated sarcomeric structure (Fig. 5A, compare GFPactin to phalloidin labelling). No difference in the overall efficiency of incorporation was detected between cytoplasmic and muscle-specific actins (Fig. 5A-C, compare a,b with c-f). Also, at the light-microscopic level, muscle architecture or function in larvae was not disrupted (see phalloidin labelling in Fig. 5A). This is in strong contrast to mammalian muscle cells in culture, in which overexpression of cytoplasmic actins led to strong dominant negative effects (von Arx et al., 1995). The distribution of both fly cytoplasmic actins (5C and 42A) was indistinguishable from the distribution of F-actin (Fig. 5A,a and b), indicating that the majority of GFPactin was incorporated into the sarcomeres.

We next analysed more closely how the different actins were incorporated into striated muscles (Fig. 5B). All but one actin isoform appeared to be incorporated into the thin filaments along their entire length; actin79B was slightly more enriched in the Z-lines compared to the other actins (Fig. 5B,d). GFPactin88F expressed in larval muscles appeared to be specifically excluded from Z-lines, but strongly present throughout the rest of the thin filaments (Fig. 5B,f).

The muscles that move the wings of adult flies and enable flight, the indirect flight muscles, reside within the thorax. Each indirect flight muscle is a single multinucleate cell containing many myofibrils with especially regular sarcomeres (Reedy and Beall, 1993a). This highly ordered structure allows for the high-frequency contractions that are necessary to power the beating wing and thus enable flight (Bate, 1993). The major actin isoform expressed in the indirect flight muscles is actin88F (Fyrberg et al., 1983). Indirect flight muscles appear especially sensitive to interference with the sarcomeric components, and most manipulations result in an inability to fly (Brault et al., 1999a; Brault et al., 1999b; Fyrberg et al., 1998).

Expression of GFPactins in the adult indirect flight muscle showed the most striking differences in incorporation, and also revealed unexpected patterns of actin incorporation (Fig. 6). Independent of the actin isoform expressed, all flight muscles showed a core of elevated levels of GFPactin (Fig. 6A). As seen in Fig. 6, the size of the core varied between individual flies analysed, but none of the specimens showed homogeneous labelling of the thin filaments. The core staining must represent stronger incorporation into the forming sarcomeres during early pupal stages of indirect flight muscle assembly. The individual sarcomeres grow continuously in length and diameter during pupal development, with thin filaments growing in length and layers of thin filaments being added around the periphery (Mardahl-Dumesnil and Fowler, 2001) (see diagram in Fig. 6C). The stronger incorporation into the core was not specific to the level of GFPactin production driven by mef2-Gal4, as two other drivers, CY2 and 24B, produced similar patterns but are expressed at different levels (Fig. 6B and data not shown). A likely explanation for the labelling of the core is that there is a temporal change in the ratio of the GFPactin to endogenous actin. For example, a large increase in the expression of endogenous actin88F at the end of flight muscle assembly could dilute the GFPactin. Regardless of the cause of the core labelling, our results highlight a difference between the Z-lines, which homogeneously incorporated the GFPactin, and the thin filaments, which did not, even though the Z-lines and thin filaments are assembled at the same time (Reedy and Beall, 1993a).

Apart from the unexpected enrichment of ectopic GFPactin in the core structure, incorporation of different actins differed most within the Z-line. The most extreme cases were actin79B, which was highly enriched in the Z-line, and actin88F, which was excluded from Z-lines. The exclusion of actin88F from Z-lines was the more surprising, as actin88F is the primary actin isoform expressed in these indirect flight muscles (Fyrberg et al., 1983). The exclusion of GFPactin88F from Z-lines was found in all muscles analysed: the larval visceral muscles (Fig. 5), the larval body wall muscles (Fig. 5), the adult indirect flight muscles (Fig. 6), as well as the adult direct flight muscles and adult leg muscles (data not shown).

The indirect flight muscle was the only muscle in which the function was affected by overexpression of GFPactins. Expression of any of the muscle-specific or cytoplasmic actins in this tissue led to flightless adults that were otherwise healthy and fertile (data not shown). This is most probably due to an imbalance in the relative amounts of actin and myosin (Beall et al., 1989). In some cases the disruption of the flight muscle structure could be observed in dissected flight muscles fibres (Fig. 6D for GFPactin5C, arrows). Nonetheless, the overall assembly of thin filaments into sarcomeres appeared indistinguishable from wild-type as judged by phalloidin-labelling (Fig. 6A, phalloidin).

In the process of generating the GFPactin-transgenic flies, we realized that one expression construct of actin79B harboured a point mutation in a highly conserved residue: R291 was changed into a histidine (Fig. 7A). Expressed in fly muscles, this mutant, actin79B^R291H, was almost exclusively incorporated into the Z-lines of sarcomeres, with very low levels in the rest of the thin filaments. This Z-line-specific incorporation was found in all muscle types analysed: larval visceral muscles (Fig. 7B), larval body wall muscles (Fig. 7C and D), adult indirect flight muscles (Fig. 7E,F), adult direct flight muscles and adult leg muscles (data not shown). Similarly to wild-type actin79B, actin79B^R291H was not incorporated into filopodia and lamellipodia in the amnioserosa (Fig. 7G), and did not label the cortex of follicular epithelial cells, though it labelled the apical microvilli (Fig. 7H). However, in contrast to the wild-type actin79B, the mutant strongly labelled the basal actin stress fibres in the follicle cells (Fig. 7I).

To address the molecular basis of the specific exclusion of actin88F from Z-lines, and the preferential incorporation of actin79B^R291H into Z-lines, we constructed chimeric GFP-fusion proteins. Comparing the position of the mutation in actin79B^R291H with the crystal structure of actin (Kabsch et al., 1990), it appeared to be in a position that might affect packing of monomers into the filament. Overall, actin79B^R291H and actin88F differ from each other in 12 positions at the amino acid level (see Fig. 1A). Four of the changes are clustered in the very N terminus, the most divergent region for all actin proteins (see Fig. 1A). The other differences are spread over the length of the protein. We therefore exchanged the first 149 amino acids of actin79B^R291H with those of actin88F and vice versa and fused them to GFP as done for the others (Fig. 8). Expression analysis in larval and adult muscles revealed that localization within the sarcomeric structure was determined by the C-terminal half. This half contains the R291H mutation and there are only two residues in this half that are unique to actin88F: F170, which is a Y in all other actins, and S298, which is variably I, N, or T in the others.

In this study we set out to address whether the small variation in amino acid sequence found in the six different actins of Drosophila affected their ability to associate with certain actin structures formed by endogenous actins. Actins were expressed as GFP fusion proteins and expression was driven in various tissues throughout development. Previous studies analysing the equivalence of actin paralogues in either mammalian tissue culture or Drosophila produced conflicting results. In our study we found that in most actin structures analysed, the differences in the efficiency of incorporation were smaller than we expected (see Table 1). Nonetheless, the different actins behaved far from identically. Three out of four muscle-specific actins, 57B, 87E and 88F, were incorporated very efficiently into all cytoplasmic structures analysed and the fourth, 79B, was incorporated into some cytoplasmic structures. Also, both cytoplasmic actins were incorporated into sarcomeres of embryonic, larval and adult muscles, without causing any apparent deleterious effects. This is in contrast to studies on mammalian cells in culture, where overexpression of cytoplasmic actins in muscle cells led to strong dominant-negative effects, disruption of the thin filament system of the sarcomere and induction of excess filopodia around the cell circumference (von Arx et al., 1995). In these studies, cytoplasmic actins were not incorporated into sarcomeres or contractile fibres (Mounier et al., 1997). This difference could be due to the closeness of fly actin sequences to each other compared to the larger sequence difference between mammalian cytoplasmic and muscle actins (Fig. 1A,B). Although fly muscle-specific actins show a series of conserved muscle-specific residues when compared to the fly cytoplasmic actins, these residues differ from muscle-specific residues found in vertebrate muscle actins (Fig. 1A, compare the green and orange boxes). Nonetheless, fly muscle actins form sarcomeric structures that appear indistinguishable from mammalian sarcomeres, and contain the same sets of accessory proteins (Clark et al., 2002), thus the higher conservation of actin sequence in the fly does not change the ability of these actins to form sarcomeric structures. Overall there appears to be no general difference in the structures that can be formed by vertebrate and fly actins.

Taken together, this study compares, for the first time, the ability of all actin paralogues found in one species to be incorporated into a large variety of actin structures in vivo. Our aim was to better understand the extent to which the six actin proteins found in Drosophila are functionally equivalent. The ability of fly actins to integrate into most actin structures and their sequence conservation suggest that the regulation of expression rather than a divergence in function of individual proteins is key to understanding why this multigene family maintained that many members through evolution. But this explanation is likely to be too simplistic, as it does not account for the following factors. Firstly, different fly actin protein sequences cannot always substitute for another actin (Fyrberg et al., 1998). Secondly, in this study we observed differences in the efficiency of different actins to be incorporated into different structures and in a few examples inability of certain actins to be incorporated into an actin structure: actin79B was not incorporated into filopodia or lamellipodia (Fig. 2), and actin88F was excluded from Z-lines in muscles (Figs 5 and 6). Thirdly, our experiments do not address whether each actin is able to form each of the actin structures by itself, because we have just tested their ability to become incorporated into structures formed by the actins normally present in each cell type. These results suggest that the main difference between actin paralogues in flies could lie in the expression control, but that in addition, tissue- or actin structure-specific adaptation of individual actins evolved, so that the present actin paralogues cannot unequivocally substitute for each other.

How could the primary actin protein sequence lead to the preferable incorporation of one actin over another for a single filament within an actin structure? It could be explained by the inability of two or more actins to co-polymerize. However, in vivo and in vitro studies in mammals have shown that even the more divergent muscle and non-muscle actins can be found copolymerized within one filament (Drew et al., 1991; Mounier et al., 1997), therefore, it appears very likely that this would also be the case in Drosophila and thus preclude `immiscibility' of actins as a cause of segregation of paralogues. Alternatively, actin-binding proteins that support the formation of certain structures could show a preference for certain actins. An example of such a preference has been described in the protein beta cap73 that preferentially binds to β-actin but not α-actin and probably links β-actin to ezrin, with which it forms a complex (Shuster et al., 1996). Such preferential binding of accessory proteins could explain the striking difference of incorporation into Z-lines that we observed for actin88F and the mutant actin79B^R291H. The analysis in various tissues showed that actin79B^R291H was not efficiently incorporated into filamentous structures. The strong incorporation into Z-lines could be due to binding to a Z-line-specific actin-binding protein such as α-actinin. This could also explain the incorporation of the mutant actin into the basal stress fibres, as they also contain α-actinin (Otey and Carpen, 2004). Conversely, the exclusion of GFPactin88F could be due to its inability to associate with certain Z-line-specific actin-binding proteins.

The exclusion of actin88F from Z-lines in the indirect flight muscles was particularly surprising, because actin88F is the primary actin isoform expressed in these muscles (Fyrberg et al., 1983). Therefore, this result suggests that the Z-line actin in the indirect flight muscles is not composed of actin88F, but instead must normally be made of one or more of the other actins. This could also explain why the other GFPactins labelled the whole of the Z-line but only the core of the thin filaments: a late burst of actin88F expression may dilute out GFPactin incorporation into thin filaments, but not the Z-line. The alternative explanation is that actin88F is normally found in the Z-lines, but the GFP-tag blocks its localisation to Z-lines. This would indicate that the recruitment of actin88F to the Z-lines occurs differently than all the other actins, since the GFP-tag did not interfere with their Z-line association. The effect of the GFP on actin88F would also have to be a long-range conformational change, since the inability of GFPactin88F to go to Z-lines mapped to the C terminus, rather than the N terminus, which contains the GFP tag.

A fortuitous mutation in actin79B, R291H, converted it into a Z-line-specific actin. Comparing this mutant with 88F, we mapped the residues responsible for the exclusion of actin88F to one or both of two residues in the C-terminal half, F170 and S298. Both of these residues are within subdomain 3 of the actin monomer, which is part of the barbed end. One of the actin88F-specific residues, S298 (T297 in human β-actin), is less likely to cause Z-line exclusion, because it is variable between the different actins and is buried in the structure of the actin monomer (Kabsch et al., 1990). The other residue, F170, is usually a tyrosine (Y169 in human β-actin), and, like R291, is exposed on the surface of the structure. F170 and R291 are both in regions that form contacts with other actin subunits during filament formation (Holmes et al., 1990). The lysine adjacent to R291 has been shown to be less reactive upon actin polymerization, indicating that it is buried in the filament (Holmes et al., 1990), thus residues in this region potentially form stabilizing contacts between actin monomers. The difference in localisation between actin88F and the mutant actin79B^R291H might be explained if we hypothesize that some F-actin in the Z-lines is present as very short filaments. The mutant actin79B could cap the ends of these filaments without being too detrimental, and, conversely, perhaps actin88F cannot associate with a barbed end capping protein at the Z-lines, so it does not accumulate.

Both residues, F170 and R291, are also part of the binding site for profilin (Schutt et al., 1993). The equivalent residue to R291 in bovine β-actin (R290), is part of the primary contact site between profilin and bovine β-actin (Schutt et al., 1993). Profilin binds actin monomers, catalyses the nucleotide exchange on actin monomers and delivers them for polymerization to free barbed ends of actin filaments (Paavilainen et al., 2004). Binding sites for other actin-binding proteins are distinct from these specific residues. For example, the interaction of actin-binding calponin-homology domains, like those in the Z-line-specific protein α-actinin, with actin filaments involves the outside of the filaments and subdomain 1 of the actin monomer (Moores et al., 2000; Sutherland-Smith et al., 2003). Thus, either a reduction in the ability to form filaments or reduced binding to profilin interaction may explain why actin79B^R291H was not easily incorporated into cytoplasmic actin structures such as filopodia, lamellipodia and cortical actin.

As well as allowing an insight into the basis of actin diversity, the GFPactins are valuable tools to study or highlight actin-based structures in a live organism. We would recommend using the following GFPactins to label specific structures (see also Table 1): actin 42A or 88F for cell cortices, filopodia and lamellipodia in the embryonic epidermis and amnioserosa; actin87E for cell outlines and actin42A for basal fibres in the follicular epithelium; actin 42A or 87E for ring canals or actin struts in the germline; actin 57B or 87E for sarcomeres in muscles. It is worth mentioning that a number of the newly generated fusion proteins, and in particular the second cytoplasmic actin42A, were in most cases a better marker for actin structures, i.e. better signal to noise ratio, in many cell types than the commonly used GFPactin5C (Verkhusha et al., 1999).

The authors would like to thank John Overton for the injection of constructs into fly embryos. We would also like to thank the following people for comments and critical reading of the manuscript: Isabelle Delon, Danelle Devenport, Sean Munro, Maithreyi Narasimha, Guy Tanentzapf. This work was supported by grants from the Wellcome Trust to N.H.B. (31315 and 69943), and K.R. was supported by a Long Term Fellowship from the Human Frontier Science Programme Organisation and a BBSRC David Phillips Fellowship.