Identification of Receptor-Tyrosine-Kinase-Signaling Target Genes Reveals Receptor-Specific Activities and Pathway Branchpoints During Drosophila Development

Abstract

Receptor tyrosine kinases (RTKs) are an important family of signaling molecules with the unusual property that they are able to transduce their signals using the same downstream pathways. This has led to an unresolved debate as to whether individual receptors are interchangeable, or if each receptor can mediate specific downstream responses. To address this question, we have conducted a screen to identify target genes whose expression is differentially modulated by RTKs and their downstream pathway components. Using whole-mount in situ hybridization in Drosophila embryos exposed to constitutively active RTK pathway signaling, along with quantitative RT–PCR, we found that a significant fraction of target genes respond differentially in a spatial and/or quantitative manner. This includes differential responses to EGF receptor vs. fibroblast growth factor receptor signaling as well as to more downstream components such as Ras1 and pointed. We show that not only genes but also individual alternative transcripts can respond differently to signaling, and we present evidence that the differential responses can be mediated at the transcriptional level. Our results demonstrate that different RTKs can elicit distinct transcriptional responses, and the target genes obtained from our screen provide a valuable resource for further exploration of the mechanisms underlying this signaling specificity.

THE receptor tyrosine kinases (RTKs) constitute a major class of signaling molecules with important roles in development and disease. RTKs—among them the epidermal growth factor receptor (EGFR), the fibroblast growth factor receptor (FGFR), the platelet-derived growth factor (PDGF) receptor, and the vascular endothelial growth factor (VEGF) receptor—are important for regulating cell proliferation, survival, and differentiation in a wide range of tissues and organs. They are required for cell movements during gastrulation, angiogenesis and vasculogenesis, axon pathfinding, synapse formation, and cell fate specification (Hunter 1998). Remarkably, all of the known RTKs appear to function using a common set of highly related or even identical signal transduction pathways (Figure 1). This has led to speculation that RTKs might produce a generic interchangeable signal, with signaling outcomes depending perhaps more on the strength of the received signal than on the properties of a given RTK (Pawson and Saxton 1999), instead of a scenario in which individual RTKs can induce specific responses via intrinsic receptor-specific mechanisms. Which of these models is the correct one remains an unresolved question.

Figure 1.—

Common RTK-signaling pathways. The three major downstream signaling pathways common to most RTKs are shown along with some of the main proteins through which they function. Shaded boxes highlight the factors used in our study. This figure is adapted from Schlessinger (2004).

One of the most prominent RTK-signaling pathways is the Ras/mitogen-activated protein kinase (MAPK) pathway, which is highly conserved throughout the eukaryotes. In this pathway, RTK activation leads to activation of the small GTPase Ras, followed by the serial activation of the Ser/Thr kinase Raf, the MAPK kinase MEK, and the ERK MAPK. Activated MAPK is transported to the nucleus where it activates transcription factors and thus induces changes in gene expression. Additional signaling pathways common to most, if not all, of the RTKs include activation of phosphoinositol signaling via direct activation of phospholipase C-γ by the RTK as well as through activation of PI-3 kinase by way of the docking protein intermediate DOS/Gab1. Signaling via the Src nonreceptor tyrosine kinase and via the JAK/STAT pathway is also a common result of RTK activation. These downstream pathways have varied and sometimes conflicting roles in cell differentiation and survival.

Is RTK signaling merely generic, or can specific RTKs elicit distinct responses due to intrinsic differences in signaling ability? A number of studies suggest that RTKs are interchangeable (Heasley and Johnson 1992; Fambrough et al. 1999; Dossenbach et al. 2001), but others infer that the same RTKs are not (Kudla et al. 1995; Rani et al. 1997; Klinghoffer et al. 2001). Kudla et al. (1995), for example, observed different responses to PDGF and fibroblast growth factor (FGF) signaling in a skeletal-muscle cell line. However, because the cells expressed endogenous RTKs at varying levels and were grown in growth-factor-containing serum, it is difficult to say whether these differences stemmed from activation of different signaling pathways or just from different levels of signaling using the same pathway. Similarly, Fambrough et al. (1999) conclude on the basis of microarray experiments that PDGF and FGF induce identical downstream responses, at odds with Kudla et al. (1995) but in concurrence with other related cell culture experiments (Heasley and Johnson 1992). Although they see different responses with EGF signaling, they attribute this to the lower levels of EGF receptor present in the fibroblasts used for the experiments. Again, therefore, it is difficult to decide just how interchangeable the different RTKs may be.

A number of “domain-swap” experiments, in which the ligand-binding extracellular domain of one RTK is fused to the transmembrane and cytoplasmic domains of another, have given qualitatively similar results, but have led the authors to strikingly different conclusions (e.g., Dossenbach et al. 2001; Klinghoffer et al. 2001). In an analysis of tracheal patterning in Drosophila, Dossenbach et al. (2001) come down on the side of RTK interchangeability and attribute small phenotypic differences to experimental artifact, e.g., differences in the rescue transgenes that they used. On the other hand, Klinghoffer et al. (2001), looking at PDGF receptor isoforms in mouse, conclude that, while the receptor isoforms may be interchangeable to a degree, definite differences in signaling exist. One problem is that these experiments use gross phenotype as an endpoint, making it difficult to distinguish between incomplete rescue as a result of experimental artifact or as a result of signaling differences that affect only a small number of genes. Indeed, recent work by Schmahl et al. (2007) demonstrates that phenotypic outcomes are the result of the cumulative activity of multiple signaling target genes, suggesting that looking directly at target gene responses will be necessary for pinpointing differences in RTK-signaling specificity.

We decided to address the questions of intrinsic RTK specificity and downstream pathway branching by focusing directly on the main outcome of RTK signaling: changes in gene expression. By identifying specific target genes whose expression is differentially affected by individual RTKs and RTK pathway components, we can obtain unambiguous evidence for signaling specificity while simultaneously building a collection of convenient assay targets for probing the mechanisms underlying differential signaling. Using a combination of qualitative (spatial) and quantitative measures of gene expression in an in vivo screen in Drosophila embryos, we find that a subset of RTK-responsive target genes responds differently to EGFR vs. FGFR activation in the same tissue. We also demonstrate substantial use of different transcriptional effectors downstream of the RTKs. These results provide clear evidence that different RTKs can modulate specific downstream responses in identical cellular environments and will allow for investigation into the mechanisms of differential RTK signaling from the receptor down through the target gene.

MATERIALS AND METHODS

Fly stocks and genetics:

The following Drosophila stocks were used: Oregon-R (wild type), Twi-GAL4 UAS-2EGFP (Halfon et al. 2002), UAS-λTop (constitutively activated EGFR) (Queenan et al. 1997), UAS-Dof UAS-λ-Htl (constitutively activated Heartless FGFR together with Downstream of FGFR/Heartbroken/Stumps) (Michelson et al. 1998; Vincent et al. 1998), UAS-Ras1^Act (activated Ras) (Carmena et al. 1998), UAS-Pnt^P2VP16 (activated Pointed) (Halfon et al. 2000), Twi-GAL4 mib2-lacZ (Philippakis et al. 2006), UAS-λBtl UAS-Dof (Lee et al. 1996), UAS-λPvr (Duchek et al. 2001), and UAS-Alk^act (Lee et al. 2003). Twi-GAL4 UAS-2EGFP flies were crossed to the respective UAS lines, and embryos were collected for 2.5 hr followed by aging at 25° for 5 hr so that the majority of embryos were at embryonic stage 11.

Probe generation and in situ hybridization:

The selected clones were grown in 96-well plates with all subsequent processing also performed in a 96-well format. DNA was isolated using the Qiagen DirectPrep 96 miniprep kit and transferred to PCR plates. PCR using primers to vector sequences flanking the cDNA and phage promoter (sequences available upon request) were then used to create a linear substrate suitable for in vitro transcription (IVT). PCR product (5 μl) was used directly for an IVT reaction containing digoxygenin-labeled dUTP according to standard in situ hybridization (ISH) probe labeling protocols (Tautz and Pfeifle 1989). IVT reactions were cleaned up using Qiagen MinElute 96-well PCR cleanup plates, resuspended in hybridization buffer, and stored at −20° until use.

Hybridizations were performed using Millipore MADV N65 filter plates essentially as described (Tomancak et al. 2002) with 12 different probes used in columns and the five genotypes in rows (see supplemental Figure S1). Hybridization was visualized using alkaline-phosphatase-coupled antidigoxygenin antibodies and direct observation with a dissecting microscope following transfer of each column into 48-well plates for better visualization. All embryos within a column (i.e., all genotypes using a particular probe) were stained for the same length of time to ensure that differences in staining intensity properly reflected quantitative differences in gene expression. All putative differentially expressed genes were repeated one or more times, typically with a third hybridization using fresh probes remade from the original miniprep DNA to rule out artifacts resulting from the original probe. All DNA templates were sequenced to confirm probe identity.

Quantitative reverse transcriptase–PCR assays:

Embryos of each experimental genotype that also contain the UAS-2xEGFP reporter were hand selected under a dissecting microscope equipped for epifluorescence visualization to ensure that all embryos were of the same stage. RNA was extracted in Trizol (Invitrogen), DNAse treated, and converted to cDNA using standard methods. Quantitative reverse transcriptase–PCR (qRT–PCR) was performed using SYBR green detection and a Bio-Rad iCycler. Expression values for each genotype were determined using an appropriate standard curve for each primer pair, normalized to 18s RNA levels and assessed relative to wild-type gene expression values. Significance of differential expression was determined by Student's t-test using Holm's correction for multiple testing.

RESULTS

A screen for differential responses to RTK activation:

To look for differential responses to RTK signaling at the level of individual target genes, we performed a gene expression pattern screen by whole-mount ISH to mRNA in Drosophila embryos. Although it is a less high-throughput approach than other methods such as microarray-based transcriptional profiling, screening by ISH has several distinct advantages, in particular, allowing for direct visual interpretation of results in unambiguously staged individual embryos. In this way, spatial as well as quantitative changes in gene expression can be assessed.

We focused our screen on gene expression in the mesoderm of stage 11 embryos, as it is well known that RTK signaling is important for cell fate specification in the mesoderm at this time (Gabay et al. 1997) and that there are different roles for various RTKs in these fate specification processes (Buff et al. 1998; Carmena et al. 1998, 2002; Halfon et al. 2000; Englund et al. 2003; Lee et al. 2003; Bruckner et al. 2004). The screen was conducted essentially as follows (see materials and methods for details): Wild-type embryos and embryos with constitutively activated RTKs or downstream signaling pathway members were plated in 96-well plates, one genotype per row (supplemental Figure S1). Each column of embryos was then hybridized with a probe directed against a different Drosophila gene. Any column in which the ISH results for one or more of the non-wild-type genotypes differed from the others was flagged for repeat hybridization; probes that gave an identical result on repetition were considered “positive.” The clones used to make all positive probes were sequenced to confirm the identity of the gene. For a subset of the positive genes, we also performed real-time qRT–PCR assays using RNA from each of the genotypes.

Our screen made use of the following five genotypes: wild type, Egfr^act, htl^act+dof (referred to hereafter as htl^act), Ras1^act, and pnt^act. Egfr and Htl are the Drosophila homologs of the EGF and FGF receptors, respectively, and are known to have distinct roles in mesodermal patterning (Buff et al. 1998; Carmena et al. 1998, 2002; Halfon et al. 2000). The scaffolding protein Dof is necessary for transduction of Htl signaling even in the presence of Htl^act (Michelson et al. 1998; Vincent et al. 1998; Imam et al. 1999), but has a restricted pattern of expression in the mesoderm (Vincent et al. 1998). Therefore, all experiments involving ectopic expression of Htl^act were performed in the presence of coexpressed Dof. Ras1 is a key downstream component of RTK signaling and is involved in all known RTK signaling in the fly mesoderm. Pnt is an ETS-domain transcription factor and is the main known positive transcriptional effector of Ras signaling in Drosophila. Our primary motivation for including Ras1^act and pnt^act in the screen was so that, when the two activated RTKs, Egfr^act and Htl^act, have different effects, we could determine which receptor is signaling through the canonical Ras/MAPK pathway (in which case it will have the same effect on the target gene as Ras1^act and Pnt^act) and which receptor is utilizing an alternative pathway (in which case its effect should be unique among the genotypes screened). However, an important additional benefit of this design is that our screen uncovers downstream branchpoints in the signaling pathways by identifying target genes that respond similarly to both RTKs but differently to either Ras1^act or Pnt^act.

We used a partially directed strategy for probe selection in that probes were randomly selected from among 1215 genes present as cDNA clones in the Drosophila Gene Collection part 1 (Stapleton et al. 2002) that were also flagged as being uniquely expressed in the Egfr^act, htl^act, or Ras1^act backgrounds on the basis of the microarray data of Estrada et al. (2006) (see supplemental Table S1). We have found that although false-negative rates in the microarray data were high, precluding an accurate assessment of which genes are uniquely responsive to one of the tested genotypes (see discussion), false-positive rates were low (Choe et al. 2005; Estrada et al. 2006). Therefore, we reasoned that this set of putatively uniquely regulated genes provided the best chances of containing truly uniquely regulated genes. In this way, we were able to bias the screen toward genes that would provide useful results in that they are mesodermally expressed and responsive to RTK signaling, and so avoid hybridizing many uninformative probes.

Twelve percent of genes respond differently to RTK/Ras/Pnt gain of function:

Results from the screen are summarized in Table 1 (for detailed qRT–PCR results, see supplemental Table S2), and representative ISH results are shown in Figure 2; Figure 3D; Figure 4, A–E; and supplemental Figure S3. Each target gene was assessed with respect to each of the four assayed gain-of-function genotypes. Of 180 genes assayed, 42 failed to give clear hybridization and were omitted from further analysis for a total of 138 tested target genes (supplemental Table S3). We observed a range of effects from purely quantitative differences in expression to spatial differences in expression in the somatic and visceral mesoderm, the anterior and posterior midgut, and the amnioserosa. Gene expression changes in the midgut are likely due to early expression of twi (and therefore of the Twi-GAL4 transgene that we used) in presumptive midgut tissues (Reuter et al. 1993). The Twi-GAL4 transgene has also been observed to drive transient ectopic expression in the amnioserosa (B. Estrada, personal communication), presumably accounting for the effects that we see there. (The expression pattern of the Twi-GAL4 transgene is shown in supplemental Figure S2.) In the majority of cases, target genes responded differently from wild type, but similarly to one another in each of the non-wild-type genotypes. However, >12% (17/138) of the target genes displayed differential expression. For a given genotype, a target gene was considered differentially expressed if it showed a clear difference in expression pattern or level with respect to each of the other non-wild-type genotypes. Note that a target gene can have more than one significant genotype; for example, CG8147 is downregulated in Egfr^act compared to the other genotypes but upregulated in pnt^act compared to each of the others as well (Figure 2, A–E, Table 1). All together, 5 genes had clear differences between Egfr^act and the other genotypes (including htl^act), 12 genes between pnt^act and the others, and 1 gene each showed differences for Ras1^act and for htl^act (Table 1). Responses to htl^act signaling were typically identical to responses to Ras1^act signaling. The identified target genes compose a diverse set of both known and novel genes with no obvious common attributes outside of their responsiveness to RTK/Ras signaling.

TABLE 1.

Summary of results from ISH and qRT–PCR

	No. of transcripts		Genotype
Target genea			Assayb	htlact	Egfract	Ras1act	pntact	Significant genotypesc
nrv1	1	ISH	++	+	+	+	htlact
		qRT–PCR	ND	ND	ND	ND
CG8147	2	ISH	0/+	−*	0/+	+	Egfract, pntact
		qRT–PCR	−	−	0	+++	Ras1act, pntact
CG9641	2	ISH	0/−	−	0/+	0/+	Egfract
		qRT–PCR	0	0	0	0
l(2)08717	2	ISH	0	−*	0	0/+	Egfract
		qRT–PCR	0	−	0	++	Egfract, pntact
ppn	2	ISH	0/+	−	0/−	0/+	Egfract
		qRT–PCR	++	−	+	++	Egfract
RhoL	1	ISH	0	+	0	0/−	Egfract
		qRT–PCR	ND	ND	ND	ND
cher	4	ISH	0	0	+++	+	Ras1act, pntact
		qRT–PCR	NA	NA	NA	NA
CG5973	3	ISH	0/+	0/+	0/+	++*	pntact
		qRT–PCR	ND	ND	ND	ND
CG8965	1	ISH	+	+	+	++*	pntact
		qRT–PCR	ND	ND	ND	ND
edl	1	ISH	+	+	+	+++	pntact
		qRT–PCR	ND	ND	ND	ND
mib2	1	ISH	+++	++	++	0/+	pntact
		qRT–PCR	+++	+	++	+++++	htlact, Egfract, Ras1act, pntact
NtR	1	ISH	0/+	0	0/+	++	pntact
		qRT–PCR	0	0	0	+++	pntact
Nup107	1	ISH	0/+	0/+	0/+	+*	pntact
		qRT–PCR	ND	ND	ND	ND
pax	6	ISH	++	++	++	0	pntact
		qRT–PCR	NA	NA	NA	NA
RhoGAP15B	2	ISH	++	+(+)	++	0/+	pntact
		qRT–PCR	+/++	0	0	+++++	pntact
Sap-R	2	ISH	++	++	++	0	pntact
		qRT–PCR	ND	ND	ND	ND
smi35A	5	ISH	0/+	0	0/+	−	pntact
		qRT–PCR	+++	0	++	−	Egfract, pntact

^aRefers to the common portion for all transcripts of the gene.

^bISH results were scored by eye. Underlining indicates a positive change in expression relative to wild type, regular type a negative change, and italics no significant change. An asterisk (*) indicates that there is a notable change in pattern as well as in level of expression. The magnitude of change is represented by + or − (for increased and decreased expression, respectively). For qRT–PCR, magnitude is given in log₂ fold-change ranges as follows: log₂ fold change = 0–0.5, +; 0.5–1.0, ++; 1–1.5, +++; 1.5–2.0, ++++; >2.0, +++++. ND, not done; NA, not applicable due to multiple transcripts.

^c“Significant genotypes” indicates genotypes for which the target gene expression differs from all other genotypes either in a visibly obvious fashion (ISH) or with an adjusted P < 0.05 (qRT–PCR).

Figure 2.—

Representative results from the in situ hybridization screen; see Figures 3 and 4 and supplemental Figure S3 for additional results. Columns represent target genes and rows represent genotypes. All embryos are shown with anterior to the left and dorsal to the top. (A–E) CG8147 has reduced expression in the Egfr^act background (B). This is most noticeable in the amnioserosa, where expression is essentially absent (arrow). Heavier staining can also be observed in the pnt^act background, consistent with qRT–PCR results (Table 1 and supplemental Table S1). (F–J) Egfr^act is also the primary affected genotype for Ppn, with consistently reduced expression in the anterior (arrow in G). Moderately increased staining in htl^act (H) is consistent with qRT–PCR results (supplemental Table S1). (K–O) cher is the only target gene that we identified with a primary effect in the Ras1^act background, which manifested as a dramatic increase in expression (N). Somewhat increased expression in the pnt^act background can also be observed (O), consistent with qRT–PCR results (Figure 4B, Table 1, supplemental Table S1) that are complicated by the presence of multiple transcripts, not all of which were independently assayed. (P–T) A large downregulation of smi35A is observed in the pnt^act background (T).

Figure 3.—

Alternative transcripts are independently and differentially regulated. (A–C) Results from qRT–PCR for individual transcripts of l(2)08717 (A), cher (B), and Pax (C). Fold changes are reported as log₂ values. Results for l(2)08717-RA are estimated by subtracting the values for l(2)08717-RB from the values for the common portion of the transcripts detected by primer l(2)08717-AB. Note that for each gene, all PCR reactions were performed on a common pool of cDNA. These results thus serve as an internal control for differences in expression due merely to differences in transgene strength; some primers demonstrate differential expression whereas others show identical expression (compare Egfr^act for cher-RB and cher-RD in B). (D) ISH results for Pax. Note how the pnt^act embryo and wild-type (WT) embryo are similar, while the Egfr^act and Ras1^act embryos are different from these but similar to one another (and to htl^act; data not shown). (E) Genomic organization of the Pax gene based on release 5.4 annotation of the Drosophila genome. Transcripts that were assayed in this study are solid and others are shaded.

Figure 4.—

A mib2-lacZ transgene responds to signaling identical to the endogenous gene. Embryos are oriented as in Figure 3. (A–E) ISH results using a mib2 probe in each of the five genotypes. (F–J) β-Galactosidase expression from a mib2-lacZ transgene (Philippakis et al. 2006). Expression patterns are similar in the ISH and reporter gene embryos, including a reduced, more wild-type expression in the pnt^act background (E and J: compare to A and F).

Surprisingly, for many of the target genes that we identified, qRT–PCR revealed additional differential expression for genotypes other than those identified by ISH (Table 1, supplemental Table S2). Most often, this additional differential expression was correlated between Egfr and pnt gain-of-function genotypes. That is, genes with an ISH-based designation for Egfr^act or pnt^act tended to also have differential expression in the pnt^act or Egfr^act genotypes, respectively, when tested by qRT–PCR. Interestingly, and consistent with such a correlation, we have previously observed that muscle progenitor cells requiring Egfr signaling for their proper specification are more sensitive to pnt loss of function than those dependent only on Htl signaling (Halfon et al. 2000). However, other combinations were also observed, including significantly different levels of expression among all four genotypes for mib2. In addition to differential expression observed by qRT–PCR but not by ISH, there was also a minority of cases in which a change seen by ISH was not reflected in the PCR results. These discrepancies are not necessarily contradictory, as the ISH and qRT–PCR methods in part assay different properties. For example, ISH can assess spatial differences in gene expression whereas qRT–PCR cannot, but qRT–PCR is much more sensitive to quantitative differences than ISH. However, we place the highest confidence on the ISH results (see discussion).

We were concerned that the differences that we observed might be merely an artifact of different levels of transgene expression rather than a true differential response to signaling. The fact that we do not see a consistent pattern of up- or downregulation in a single genotype argues against this, although we do note a trend toward decreased expression in the Egfr^act background. To rule out any effects due to the strength of transgene insertion, we measured the expression levels of the Egfr^act and htl^act transgenes by qRT–PCR using primers specific to a λ-phage sequence fragment found in both transgenes but not in the endogenous genes. We found that the Egfr^act transgene is expressed ∼1.5× more strongly than the htl^act transgene (data not shown). To control for this, we reared our Egfr^act embryos at 29° instead of 25°, which led to a reduction in the amount of transgene expression to a level comparable to that of the htl^act transgene at 25° (P = 0.3 by t-test). Western blots measuring levels of activated (diphospho) MAPK gave results consistent with those seen for levels of transcription for the activated receptors and also confirmed that Ras1^act gave a level of pathway activation equivalent to that seen for htl^act at 25° and Egfr^act at 29° (data not shown). The temperature-sensitive result is unusual, given that elevated temperature typically results in an increase of Gal4-mediated gene expression, and we speculate that it may be the result of negative feedback on the twi promoter at high signaling levels. Regardless of cause, however, the observed reduction in transgene expression is fortuitous as it allowed us to repeat the ISH for all of the affected genes under conditions of equivalent transgene expression. These ISH results were identical to those obtained previously (data not shown), indicating that the observed differences in target gene expression were not due to different transgene expression levels but rather to the effects of RTK signaling. Additional evidence for true differential expression, vs. transgene artifacts, is provided by our results with transcript-specific qRT–PCR (below).

For a subset of the positive genes, we also assessed their responsiveness to constitutive mesodermal activation of three other Drosophila RTKs known to signal via the Ras/MAPK pathway: Btl, a second FGF-receptor homolog (along with coexpressed Dof); Pvr, a member of the PDGF/VEGF family of RTKs; and Alk, homologous to the human anaplastic lymphoma kinase (ALK). Target gene expression in these backgrounds was always identical to that seen with Htl^act (data not shown), consistent with the closer relationship of these RTKs to FGFR as compared to the more diverged EGFR (Robinson et al. 2000).

Differential response of individual transcripts:

Nine of the differentially regulated target genes identified in our screen are annotated as having two or more distinct transcripts (Table 1). For five of these, we designed transcript-specific RT–PCR primers and performed qRT–PCR on all of the experimental genotypes with each set of primers. Four genes—cher, Pax, CG9641, and l(2)08717—showed different patterns of regulation of individual transcripts (Figure 3, A–C; Table 2; supplemental Table S2).

TABLE 2.

Differential response of individual transcripts

Target gene (transcript)a	Genotypeb				Significant genotypesc	Total no. of transcripts	Alternative promoters?
		Htlact	Egfract	Ras1act				Pntact
CG9641	0	0	0	0	None	2	Uncertain
(CG9641-RA)	0	−−	0	+++	Egfract, Pntact
cher	NA	NA	NA	NA	NA	4	Yes
(cher-RA/RD)	0	−−−	++++	++++	Htlact, Egfract
(cher-RB)	+++	+++	+++	+++++	Pntact
(cher-RC)	+++	+	+++++	+++	Egfract, Ras1act
(cher-RD)	++	0	+++	+++++	Egfract, Pntact
l(2)08717	0	−/−−	0	++	Egfract, Pntact	2	Yes
(l(2)08717-RB)	++	−−	+/++	−−/−−−	Noned
pax	NA	NA	NA	NA	NA	6	Yes
(pax-RA)	+++	++	+++	+++	Egfract
(pax-RB)	+++++	++	+++/++++	++	Htlact, Ras1act
(pax-RC)	++++	0	+++	+++	Egfract, Htlact
(pax-RD)	0	−	0	+++	Egfract, Pntact
smi35A	++/+++	0	++	−−/−−−	Egfract, Pntact	5	No
(smi35A-RC/RD)	+++	0	++	−−−−	All different

^a“Target gene” refers to the common portion for all transcripts of the gene; “transcript” refers to results for the indicated transcript only.

^bResults are from from quantitative real-time PCR with RNA from the indicated genotype. + indicates a positive change in expression relative to wild type, − a negative change, and 0 no significant change. Change is represented by + or − (for increased and decreased expression, respectively); magnitude is given in log₂ fold-change ranges as follows: log₂ fold change = 0–0.5, +; 0.5–1.0, ++; 1–1.5, +++; 1.5–2.0, ++++; >2.0, +++++. NA, qRT–PCR not performed for a portion of mRNA common to all transcripts.

^c“Significant genotypes” indicates genotypes for which the target gene expression is significantly different from all other genotypes (adjusted P < 0.05).

^dEgfr^act and Pnt^act are similar; Htl^act and Ras1^act are similar.

These results help to explain the observed discrepancies between ISH and qRT–PCR data, i.e., differential signaling based only on qRT–PCR evidence or ISH results that are not supported by qRT–PCR. The ISH probes are typically common to multiple alternative transcripts and thus provide a gene-wide, but not transcript-specific, readout of transcription. This means that, while an individual transcript might show a markedly different response in two different genotypes, the overall level of expression for that gene may not vary sufficiently to register as a quantitative difference at the level of ISH. The identified target gene Pax provides a good example of this. Pax has six annotated transcripts transcribed from five different promoters. Our ISH probe contains sequences common to all six transcripts and thus should provide an overall picture of Pax gene expression. Pax was called positive in our screen with its primary affected genotype pnt^act, which had an expression pattern closer to that of the wild type than to that of the other tested genotypes (Figure 3D). We performed qRT–PCR specific to four of the transcripts (Pax-RA–Pax-RD; Table 2). The PCR results from all four transcripts suggest that Pax is upregulated in the pnt^act background, contrary to the ISH results, although the two remaining transcripts, only recently identified, remain to be tested. Looking closely at the individual transcripts, however, reveals additional complexity. For example, all four transcripts are significantly downregulated in response to Egfr signaling as compared to either Ras or Htl signaling (Figure 3C, Table 2, supplemental Table S2). Thus Pax-RA, Pax-RC, and Pax-RD are considered differentially expressed in the Egfr^act genotype. This is seen most clearly for Pax-RC and Pax-RD, neither of which is upregulated in Egfr^act compared to wild type. Pax-RB is not considered an Egfr^act target as its response in the Egfr^act and pnt^act backgrounds is identical. However, Pax-RB does respond differentially in htl^act vs. Ras1^act, which are distinct from one another as well as from Egfr^act and pnt^act.

Importantly, the transcript-specific results also provide an internal control for the effects of strength of transgene expression discussed above as well as for possible artifacts due to isolation of RNA for qRT–PCR. Because the qRT–PCR assays were conducted on the same pool of RNA, these potential artifacts cannot explain the observed results. For example, transcript cher-RB has increased expression in both the Egfr^act and htl^act backgrounds, whereas transcript cher-RD is upregulated in the htl^act background but unchanged in the Egfr^act background (Figure 3B and supplemental Table S2). These data thus provide additional, internally controlled evidence for distinct differential responses to RTK pathway signaling.

Differential responses are mediated at the transcriptional level:

In many of the cases of transcript-specific differential expression, the transcripts appear to be initiated at separate promoters (Table 2 and Figure 3E). This suggests the possibility that the differences in gene expression that we observe are transcriptional in nature, rather than, for example, due to differences in alternative splicing or effects on RNA stability. We made use of a previously defined transcriptional regulatory module for mib2 expression to test this idea (Philippakis et al. 2006). A mib2-lacZ reporter construct was crossed into each of the activated signaling pathway genotypes used in our screen, and embryos were stained for β-galactosidase expression. mib2 was identified as a pnt^act-responsive gene in which expression in the pnt^act background was more similar (but not identical) to wild-type expression than that seen in the other genotypes (Figure 4, A–E). The expression of the mib2-lacZ reporter exactly paralleled the results from the ISH screen (Figure 4, F–J), demonstrating that the mib2 enhancer element is sufficient to mediate the differential expression that we observed.

DISCUSSION

We have provided here a clear demonstration that the expression of individual target genes can be differentially regulated by activated EGF and FGF receptors, as well as by activation of Ras1 and the ETS-1 homolog Pnt, in an otherwise identical in vivo cellular milieu. On the basis of our data, >7% of genes are differentially regulated by EGF vs. FGF receptors. However, the true number of differentially regulated genes is likely to be higher, as we noted qRT–PCR-only expression changes for a number of genes in which consistent differences at the level of ISH could not be observed (data not shown). Because we have greater confidence in the hybridization results compared to the PCR results (see below), we have taken a conservative approach in considering primary responses to be only those that were visually observable by ISH.

Our results are consistent with a growing amount of evidence that suggests that considerable diversity exists in the downstream responses to individual members of the receptor tyrosine kinase family, in addition to the well-known ability of RTKs to activate a set of common signaling pathways. For example, a genetic screen for modifiers of RTK signaling in the Drosophila eye uncovered a number of genes specific to FGF receptor signaling, or common to both FGF receptor and PVR signaling but not to EGF receptor activation (Zhu et al. 2005). During Drosophila oogenesis, EGFR and PVR appear to have overlapping but not identical functions in follicle cell migration (Duchek et al. 2001), while cell proliferation and survival in the developing Drosophila larval dorsal thoracic air sac requires the EGF receptor but not FGF receptor signaling (Cabernard and Affolter 2005). Finally, proteomic analysis (Kratchmarova et al. 2005) found an overlapping but distinct set of proteins to be tyrosine-phosphorylated following stimulation of cells with either EGF or PDGF, with the unique set comprising ∼10% of the total; this number fits well with the >7% differentially regulated target genes that we observed. However, since the proteomic analysis took note only of the phosphorylation profiles of intermediate pathway members, a clear demonstration of different cellular outcomes as a result of this divergent RTK behavior could not be made. Unfortunately, in none of these cases have the mechanisms responsible for the differential signaling been determined; nor, in many of them, do convenient screens and assays exist for use in discovering these mechanisms.

Our finding that defined transcriptional regulatory elements, such as the mib2 mesoderm enhancer, can mediate the observed differential responses to RTK signaling provides an important set of tools that will allow us to focus the full power of Drosophila molecular genetics on investigating the mechanisms behind this phenomenon. Analysis and mutagenesis of the enhancer will allow us to trace back the pathways from transcriptional effector to RTK signal to find the components responsible for each of the individual signaling responses, and the reporter constructs will provide a robust and convenient assay for visualizing the effects of mutations in candidate genes functioning downstream of the receptors. We have also established that the mib2 enhancer mediates responses to Ras1 and Pnt consistent with those that we observed in embryos in cultured Drosophila S2 cells (M. Boutros and M. S. Halfon, unpublished observations). This raises the exciting possibility of complementing lower-throughput in vivo methods with genome-scale RNA interference screens (Kuttenkeuler and Boutros 2004; Friedman and Perrimon 2006) designed to discover the relevant downstream effectors of the differential responses.

The role of adaptor proteins:

One mechanism accounting for specific differences among the RTKs lies in their requirements for different docking proteins that couple the receptor to downstream components such as Ras. This is clearest with respect to FGFR signaling. In vertebrates, FGFR requires the docking protein FRS2 to carry out most, if not all, of its signaling functions (Hadari et al. 2001); in Drosophila, the structurally unrelated but functionally similar Dof carries out the same role as an obligate adaptor for all FGFR-mediated signaling (Michelson et al. 1998; Vincent et al. 1998; Imam et al. 1999). Other examples also show that adaptor proteins can confer specificity on RTK signaling. The adaptor Crk can activate MAPK in response to signaling by nerve growth factor (NGF) but not by EGF (Tanaka et al. 1995). Interestingly, this activation occurs through stimulation of the Ras-related GTPase Rap1 instead of Ras itself (York et al. 1998) and may be responsible for the well-known differences in the effects of EGF vs. NGF stimulation on PC12 cells (Marshall 1995). The Drosophila Shc protein (Dshc) is involved in signal transduction via the EGFR and Torso RTKs, but not via the RTK Sevenless (Luschnig et al. 2000). However, Dshc is not an obligate member of the EGFR and Torso signaling pathways; rather, it acts in parallel with other signaling components (such as Dos and Drk) to activate the Ras/MAPK pathway. Although this suggests a mechanism by which differential coupling of Dshc to RTKs could lead to specific downstream responses, it is not yet known whether Dshc causes activation of additional pathways or only acts redundantly with other downstream components. Neither is it known which other RTKs do or do not require Dshc for proper signal transduction. Shc may also contribute to branching of the signal transduction pathway downstream of the receptor; mutation of different sets of tyrosine residues in mammalian Shc has been shown to differentially affect MAPK and non-MAPK-mediated aspects of EGF growth factor signaling (Gotoh et al. 1997). Significantly, although our study took into account the dependence of FGFR signaling on the Dof adaptor by including expression of dof along with expression of htl^act, we still saw differential target gene regulation following EGFR and FGFR activation. Thus, the requirement for an FGFR-specific adaptor cannot be the sole mechanism responsible for different responses to these RTKs.

Transcriptional effectors:

Roughly half of the responses that we observed had as their primary genotype pnt^act. Similarly, Cabernard and Affolter (2005) have demonstrated previously that pnt is required for FGF-receptor- but not EGF-receptor-specific functions in the developing air sac of the Drosophila dorsal thorax. These results are unsurprising, given that pnt is at the base of the RTK signaling cascade, allowing many opportunities for other transcriptional effectors to be brought into play. Nevertheless, they highlight the need for the identification of Ras1 transcriptional effectors in Drosophila other than pnt and the related Ets domain factor aop (yan), of which few are known. We have shown previously that mutation of Ets protein-binding sites leads to a stronger phenotype than pnt loss of function in a transcriptional regulatory element for even skipped, suggesting that other Ets domain transcription factors may act redundantly to or in parallel with Pnt in response to RTK signaling (Halfon et al. 2000). A similar situation may be the case for mib2, whose regulatory sequences were identified in part on the basis of the presence of so-called Pnt-binding sites (Philippakis et al. 2006), but which does not respond strongly to Pnt activation as compared to EGFR, FGFR, or Ras1 activation (Figure 4).

A versatile method for studying signaling specificity:

Although large-scale ISH has been performed in Drosophila for purposes of classifying gene expression patterns (Simin et al. 2002; Tomancak et al. 2002), it has not been used as a method to screen for genotype-specific transcriptional responses in the manner employed here, which has more commonly been the domain of higher-throughput and less labor-intensive methods such as microarray screens. However, we believe that our approach has several distinct advantages. For one, ISH provides a spatial as well as a quantitative readout, so that even in situations where the overall levels of gene expression are not very different, it is possible to observe changed patterns of expression (e.g., CG8147, Figure 2B). Another consideration is the demonstrated poor sensitivity of microarrays with respect to small changes in gene expression (Choe et al. 2005). We explored this with respect to the data of Estrada et al. (2006), which includes Affymetrix profiles of mesodermal cells of the same genotypes that we used here. qRT–PCR analysis of a subset of genes scored as differentially expressed between Egfr^act and htl^act revealed that most were in fact similarly expressed, but were not scored accurately in the microarray data from one of the two genotypes (M. S. Halfon, unpublished observations). Of the target genes found in our screen, only CG9641 was correctly identified as being differentially regulated by Egfr^act vs. htl^act in the microarray data, but mistakenly described as being reduced in the htl^act rather than in the Egfr^act background (supplemental Table S1). A further advantage to the ISH approach is that it is easy to visually isolate and score specific developmental stages with high accuracy. This is not possible with microarray and qRT–PCR approaches, which require large pools of embryos to generate sufficient RNA. It is virtually impossible to ensure precise and narrow staging of all embryos in such pools; however, gene expression patterns can be highly dynamic, and temporally specific changes in expression are easily masked when analyzing RNA from pools of embryos. For this reason, in general we place higher confidence on our ISH results than on our qRT–PCR results. Our work here demonstrates that large-scale ISH screens can serve as a valuable alternative or complement to microarrays and other high-throughput quantitative approaches to studying the important issue of signaling specificity not only for RTKs, but also for any other pathways and especially for combinations of pathways as well.