Research Article

Patterns of Gene Expression During Drosophila Mesoderm Development

See allHide authors and affiliations

Science  31 Aug 2001:
Vol. 293, Issue 5535, pp. 1629-1633
DOI: 10.1126/science.1062660


The transcription factor Twist initiates Drosophilamesoderm development, resulting in the formation of heart, somatic muscle, and other cell types. Using a Drosophila embryo sorter, we isolated enough homozygous twist mutant embryos to perform DNA microarray experiments. Transcription profiles oftwist loss-of-function embryos, embryos with ubiquitoustwist expression, and wild-type embryos were compared at different developmental stages. The results implicate hundreds of genes, many with vertebrate homologs, in stage-specific processes in mesoderm development. One such gene, gleeful, related to the vertebrate Gli genes, is essential for somatic muscle development and sufficient to cause neural cells to express a muscle marker.

Formation of muscles during embryonic development is a complex process that requires coordinate actions of many genes. Somatic, visceral, and heart muscle are all derived from mesoderm progenitor cells. The Drosophila twistgene (1), which encodes a bHLH transcription factor, is essential for multiple steps of mesoderm development: invagination of mesoderm precursors during gastrulation (2), segmentation (3), and specification of muscle types (4). The role of twist in mesoderm development has been conserved during evolution (5), perhaps because it controls conserved regulatory mesoderm genes. For example,tinman and dMef2 are regulated by Twist in flies (6, 7) (Fig. 1A) and are highly conserved in sequence and function in vertebrates (8–10).

Figure 1

Gene expression profiles in twistmutant embryos compared with wild-type embryos. (A) Twist is at the top of a transcriptional hierarchy in mesoderm development and directly regulates the expression of tinman,dMef2, and itself. (B) A SOM clustering program divided the 130 Twist-low genes into three groups with similar trends in expression (18). This analysis was done in log2 using a three by one matrix and converted back to real numbers to graph the data. The value 0.5 represents a twofold reduction in twist embryos compared with wild type. Increasing the complexity of the SOM matrix subdivides these subgroups further (e.g., three by two matrix), but the general trends remain the same. Graph lines represent average transcription changes for all genes within that group. Blue diamond, group A; red square, group B; green triangle, group C. S9-10, stage 9-10; S11, stage 11; L S11-12, late stage 11 to 12. Web table 1 (13) shows the genes within groups A through C. (C) In situ hybridization with one unknown gene from each group in (B). Scale bars in each panel of (C) are 39 μm.

In Drosophila, somatic muscle forms from progenitor cells that divide to become muscle founder cells (11). Founder cells acquire unique identities controlled by transcription factors including Krüppel, S59,vestigial, and apterous. Each of the 30 body wall muscles in an abdominal hemisegment is initiated by a single founder cell and has unique attachments and innervations (12). To further clarify mechanisms underlying founder cell specification, myoblast fusion, and muscle patterning, we have usedDrosophila mutants together with microarrays of cDNA clones.

Transcription profile of twist homozygous mutant embryos. twist mutant embryos develop no mesoderm (1) [Web fig. 1 (13)]. We compared the population of mRNA species isolated from twist homozygous embryos with that of stage-matched wild-type embryos. Drosophila lethal mutations are maintained as heterozygotes, in trans to balancer chromosomes. Atwist mutation was established in trans to a balancer chromosome carrying a transgene encoding green fluorescent protein (GFP) (14). Embryos were collected from wild-type and twist/GFP-balancer fly stocks at specific developmental stages. The twist/GFP-balancer collections contain a mixed population of embryos: one-quarter twisthomozygotes lacking GFP, half heterozygotes with one copy of GFP, and one-quarter homozygous for the balancer chromosome with two copies of GFP. Homozygoustwist mutant embryos were separated from their siblings using an embryo sorter (15) [Web fig. 2 (13)]. Putative homozygous twist embryos were assessed by immunostaining with an antibody to dMef2. More than 99% of the selected embryos had the twist phenotype.

Three different periods of mesoderm development were analyzed: stages 9-10, 11, and 11-12 [stages according to (16)]. During stage 9-10, the earliest time GFP is detectable in the balancer embryos (14), mesoderm cells migrate dorsally and become specified as somatic, visceral, cardiac, and fat body mesoderm. twist and its direct targets tinmanand dMef2 are expressed throughout stage 9-10 mesoderm. The middle period contained stage 11 embryos and is a transition between the first period (stage 9-10) and the third period (late stage 11-12). During late stage 11 and stage 12, myoblast fusion begins and twist expression remains prominent in only a subset of the somatic muscle cells.

For each developmental period, three independent embryo collections, embryo sortings, and microarray hybridizations were conducted. The microarrays used for the analysis contained over 8500 cDNAs corresponding to 5081 unique genes plus a variety of controls [see Web fig. 3 for array details (13)]. Each embryonic RNA sample was compared with a reference sample, which contains RNA made from all stages of the Drosophila life cycle and allows direct comparisons among all the experiments. Sample and array variability was determined by calculating correlation coefficients and standard deviations for each gene for all pair-wise combinations of repeated samples. The median correlation coefficient is 0.92, and median standard deviation divided by mean is 0.246 [see Web text for validation information (13)].

To determine how transcription was affected by the twistmutation, SAM (significance analysis of microarrays) analysis was used (17). Genes that are normally highly expressed in mesoderm should have lower transcript levels in twisthomozygotes. Genes in other tissues whose expression depends on signals from the mesoderm might also have reduced expression. Transcripts of 130 genes, the “Twist-low” group, were significantly lower intwist mutants than in wild type (Fig. 2A). Conversely, cells that would have formed mesoderm may take on other fates in the absence oftwist, such as neuroectoderm; therefore, many transcript levels could increase in twist mutants. Genes whose transcription is repressed by signals from the mesoderm would also be enriched in twist mutants. One hundred fifty genes, called the “Twist-high” group, have increased levels of RNA in twistmutant embryos (Fig. 2A).

Figure 2

A comparison of gene expression profiles of mesoderm-deficient (twist mutant) and mesoderm-enriched (Toll 10B mutant) embryos. (A) Compared with wild type, 643 genes had significantly changed transcription in twist and Toll mutant embryos. Ratios of mutant/wild type were arranged with a hierarchical clustering program, and the results are shown in Web fig. 3 (13, 27). The Venn diagram shows the overlap between the twist and Toll experiments. The example clusters [(B) through (F)] are indicated by the colored bars in (A) and colored box around the corresponding cluster in (B) through (F). A white line in the diagram between the twist and Toll experiments marks the transition in the clusters between the two genotypes. Examples of non-mesoderm gene clusters are shown in (B andC). Three representative clusters for mesoderm genes are shown in (D through F). Red typeface indicates either genes known to be transcribed in the nervous system and/or ectoderm [(B) and (C)] or known mesoderm genes [(D) through (F)]. One unknown gene (blue typeface) from each cluster was selected for in situ hybridization. Scale bars in (B) through (F) are 39 μm.

In total, 280 of ∼5000 genes had significant changes in transcript levels, with 10 false positives (17) [see Web text for validation information (13)]. The genes on the array include 15 previously characterized mesoderm-specific genes, all of which were significantly reduced in twist mutant embryos (Fig. 3A). The arrays also contain genes known to be transcribed in both mesoderm and other cell types. Significant changes in expression were detected for many of these genes (Fig. 3B).

Figure 3

(A) Changes in gene expression intwist and Toll10B embryos compared with wild type for all known mesoderm-specific and muscle genes on the array. stumps was placed in this group because it is primarily expressed in the mesoderm during these stages. (B) Gene expression data for genes both in the mesoderm and other tissues.

The 130 Twist-low genes were divided into three groups (A, B, and C) with similar trends of expression by a self-organizing map (SOM) clustering program (Fig. 1B) (18). The 24 group A genes, which included tinman, dMef2, andbagpipe, had reduced transcript levels in twistmutants at all developmental stages assayed. Most of the Twist-low genes fall into the B and C groups. The 62 group B “early genes” encode transcripts with reduced levels of expression intwist mutants only during stages 9-10, not later. One member of group B, stumps (dof/hbr) is essential for mesoderm cell migration. stumps RNA is abundant in the mesoderm at stages 9-10 and is strongly reduced by stage 11 (Fig. 1B) (19). At stage 11, stumps RNA accumulates in trachea, which are largely unaffected in twist mutants.

The 44 group C genes have reduced transcript levels in twistmutant embryos only during late stage 11 and stage 12. These “late genes” include blown fuse, a gene essential for myoblast fusion (20); delilah, a gene required for somatic muscle attachment (21); and genes such as kettin, which is required to form contractile muscle (22). Given the predominantly early expression of twist, the early genes in groups A and B are the best candidates for direct transcription targets of Twist, though some indirectly activated genes may be present within these groups. Group C late genes are likely to be regulated by products of genes that are activated by Twist.

In situ hybridizations were done using a previously uncharacterized representative of each Twist-low group (Fig. 1C). In each case, the hybridization pattern was consistent with the predicted time of transcription. A group A gene, CG15015 (GH16741), is transcribed in somatic muscle throughout stages 9-12. A group B gene, CG12177 (GH22706), is transcribed during early mesoderm development, but not later. CG14848 (GH21860), a group C gene, is expressed in the stomodeum but not the mesoderm during stages 9-10. Its mesoderm expression initiates during stage 11, the latest period of thetwist experiment. Thus, combining loss-of-function mutant embryo analysis with staged embryo collections provides gene expression information for both tissue specificity and temporal expression.

A complementary test: The transcription profile with twist overexpression. The mis-expression of twist in the ectoderm is sufficient to convert both neuronal and epidermal tissues to a myogenic cell fate (4). RNA from embryos with ubiquitous twist expression was used to evaluate the ability of Twist to initiate mesoderm-like gene expression in cells that would normally form other tissue types. Genes whose transcript levels decrease in twist loss-of-function embryos and increase whentwist is ubiquitous are excellent candidates for regulators of mesoderm development or differentiation.

To ectopically express twist, a dominant gain-of-function mutation of the maternal gene Toll(Toll 10B) was used (23). Activated Toll induces the expression of twist and snail in early embryos and of immune response genes in older embryos (Fig. 1A) (24, 25). Thus, the effects ofToll 10B on gene expression reflect the activities of Twist as well as Snail and Dorsal, or their combined actions. Toll 10B embryos are essentially bags of mesoderm; epidermal structures are absent or greatly reduced (23), and they have been used successfully in subtractive hybridization screens to identify mesoderm genes (26). We compared the gene transcription profile ofToll 10B embryos with that of wild-type embryos during four periods of development, using the reference sample to normalize experiments. The earliest period, stage 5, is whentwist is initially expressed in presumptive mesoderm. The other three periods analyzed were those used in the twistmutant analysis: stages 9-10, 11, and 11-12.

In Toll 10B embryos, 447 genes had significant changes in RNA levels compared with stage-matched wild-type embryos (Fig. 2A), 16 of which are predicted to be false positives (17) [see Web text for validation information (13)]. Transcripts from 166 genes were reduced inToll 10B embryos compared with wild type. These genes may be involved in neuroectoderm events that are blocked when cells are turned into mesoderm (Fig. 2, clusters B and C). Transcripts of 281 “Toll-high” genes were increased inToll 10B embryos. Of the 21 previously characterized mesoderm-specific genes on the arrays, 18 have significantly increased transcript levels inToll 10B embryos (Fig. 3A). The remainder may require activators other than Toll, such as signals from the severely altered ectoderm.

Mesoderm and non-mesoderm gene classes. Genes with altered transcription in twist and Toll10Bmutants were analyzed with a hierarchical clustering program (27) to identify similar transcription profiles. The genes were divided into putative “mesoderm” and “non-mesoderm” groups (Fig. 2A). Non-mesoderm genes were defined as having increased transcript levels in mesoderm-deficient (twist) embryos and/or decreased expression in mesoderm-enriched (Toll10B) embryos (Fig. 2, B and C). Mesoderm genes were defined as having decreased transcript levels intwist mutants and/or increased transcripts inToll10B mutants (Fig. 2, D through F). The mesoderm genes group will also contain genes expressed in other tissues in a mesoderm-dependent manner.

Non-mesoderm genes in clusters B and C are repressed inToll 10B mutants. Cluster B (Fig. 2B) genes have increased RNA levels in twist mutant embryos, whereas cluster C (Fig. 2C) genes do not change significantly. The overexpression of twist in the presumptive ectoderm inToll 10B embryos results in a conversion of ectodermal cell fate into mesoderm. snail anddorsal are ectopically expressed inToll 10B embryos (Fig. 1A) and transcriptionally repress the expression of neuroectoderm and ectoderm genes (28, 29). The conversion of ectoderm to mesoderm due to twist misexpression, and the ability of Snail and Dorsal to repress ectoderm genes, suggests that the B and C clusters should contain primarily neuroectodermal genes. Indeed, the non-mesoderm genes include 31 previously characterized neuroectodermal genes. One previously unknown cluster B gene that encodes a putative cell adhesion protein is transcribed in the ventral nerve cord (Fig. 2B). Another previously unknown gene within cluster C is transcribed within the developing brain (Fig. 2C).

The Twist-low and Toll-high genes have in common 51 genes (Fig. 2A) that are highly likely to be involved mesoderm development. For example, transcription of the genes in cluster D (Fig. 2D) is reduced in twist mutants and increased inToll 10B mutants during most or all time periods.

A complete overlap between Twist-low and Toll-high gene sets is not expected for three reasons, as follows: (i) Development of dorsal mesoderm, and of muscle founder cells marked by apterous andconnectin, requires the Decapentaplegic (a transforming growth factor-β–class signal) (30, 31) and perhaps others. Changed characteristics of cells that form ectoderm inToll 10B embryos interferes with these signaling events. We observe a significant reduction in dpp RNA levels in Toll 10B embryos (Web Fig. 3). (ii) During midgut development, endoderm cells migrate along the mesoderm. Midgut endoderm development is affected in twist mutants (32). Some Twist-low genes with unchanged expression inToll 10B embryos are transcribed in the midgut (not shown). (iii) Ectopic Twist inhibits visceral mesoderm and heart development and promotes excess somatic muscle development (4). Toll 10B embryos produce high levels of Twist throughout the embryo, so genes that have reduced RNA levels in both twist and Toll 10Bmutant embryos are likely to be visceral muscle and heart genes. Indeed, bagpipe and connectin, genes expressed in visceral mesoderm, are among the 79 Twist-low genes not induced by ectopic Twist (Fig. 3A).

Of the 281 Toll-high genes, 230 were unaffected in twistmutants. Some of the 230 are normally expressed late in embryogenesis in wild-type embryos but are expressed prematurely inToll 10B embryos due to ectopic Twist. These include Myo61F, MSP-300, andParamyosin, genes normally active in terminally differentiated muscle (stage 16) (Fig. 2F and Fig. 3A). Ectopic Snail and Dorsal in Toll 10B embryos may activate genes that are unaffected in twist mutants. Snail can repress neuroectodermal genes and may also activate mesoderm genes (26). Dorsal activates immune response genes later in development (25). relish,drosomycin, and metchnikowin genes—all immune response genes—have higher transcript levels inToll 10B embryos.

Data from loss- and gain-of-function experiments, combined with careful staging, yield a useful picture of genes that are likely to be required for mesoderm specification and muscle differentiation. Of 360 identified mesoderm genes, 273 have not been the focus of developmental studies. The predicted proteins encode transcription factors, signal transduction molecules, kinases, and pioneer proteins. The stage at which each gene is active is one criterion for assigning possible functions. Another key criterion will be finding a mutant phenotype. As a pilot, we have taken this additional step for the gene CG4677 (LD47926). Changes in CG4677 transcript levels were also observed in aToll 10B subtractive hybridization screen (26).

gleeful, a gene required for somatic muscle development.CG4677 is transcribed in the visceral mesoderm at stages 10-13 and the somatic mesoderm during stages 11-13 (Fig. 4, A through C). This gene encodes a C2H2 zinc finger transcription factor with high sequence similarity to vertebrate Gli proteins, so we have named the gene gleeful (gfl). Mammalian Gli proteins act downstream of Hedgehog signaling proteins to control target gene transcription (33).

Figure 4

Gleeful (Gfl), is essential for somatic muscle development. (A through C) In situ hybridization of gfl (CG4677, LD47926). gfl is expressed transiently in the visceral mesoderm (red arrow) and somatic mesoderm (green arrows) from stages 10-13. (D) Embryos injected with a control double-stranded RNA and immunostained with α-dMef2 antibody. (E) Embryos injected with double-stranded RNA for gfl (from LD47926 cDNA) and immunostained with α-dMef2. There is a severe loss of somatic muscle cells [green arrows in (D) and (E)]. (F) Same embryo shown in (E) focusing on visceral mesoderm (blue arrow). The heart [arrowheads in (D) and (E)] and visceral mesoderm appear normal. (G) Ventral view of a wild-type stage 16 embryo stained with α-dMef2. (H) Embryo containing UAS-gfl anden-GAL4 stained with α-dMef2. Ectopic expression ofgfl induces dMef2 expression in the ventral nerve cord (black arrow). dMef2-expressing cells are embedded within the ventral nerve cord [inset in (H), 63× magnification]. Scale bars in (A) through (H)are 39 μm.

The role of gfl in mesoderm development was assessed by disrupting its function using RNA interference (34). Injection of a double-stranded RNA (dsRNA) control sequence had no effect on mesoderm development (Fig. 4D). In contrast, gfl dsRNA injection caused severe loss and disorganization of somatic muscle cells (Fig. 4E), whereas heart and visceral muscle were unaffected (Fig. 4, E and F). A similar phenotype was seen in Df(3R)hh homozygous embryos (not shown), the deficiency removes gfl but not the nearbyhedgehog gene

To determine whether gfl can induce muscle cell development, a UAS-gfl transgenic fly strain was generated. Ectopic expression of gfl using anen-GAL4 driver results in lethality and induction of ectopic dMEF2 expression in the ventral nerve cord (Fig. 4H). Remarkably, Gfl is sufficient to induce expression of a muscle gene in neuronal cells. Previous studies have shown an essential role for Sonic hedgehog signaling in the formation of slow muscle in avian and zebrafish embryos (35, 36).gfl may be performing a similar role inDrosophila somatic muscle development.

This study has combined mutant embryo analyses with DNA microarrays to identify genes that are downstream of twistin mesoderm development. These efforts should be helpful in gaining a comprehensive view of cell fate determination, organogenesis, cell proliferation, and pattern formation in the mesoderm.

  • * Present address: Mail Stop 68-425 Department of Biology, Massachusetts Avenue, Cambridge, MA 02139, USA

  • Present address: Department of Genetics, Sterling Hall of Medicine, 333 Cedar Street, Yale University School of Medicine, New Haven, CT 06520, USA.

  • To whom correspondence should be addressed. E-mail: scott{at}


View Abstract

Navigate This Article