Identification of downstream targets of the Bone Morphogenetic Protein pathway in the Drosophila nervous system

Authors


Abstract

Bone Morphogenetic Protein (BMP) signaling mediated by the receptor Wishful thinking (Wit) is essential for nervous system development in Drosophila. Mutants lacking wit function show defects in neuromuscular junction development and function, specification of neurosecretory phenotypes, and eclosion behavior that result in lethality. The ligand is Glass bottom boat, the Drosophila ortholog of mammalian BMP-7, which acts as a retrograde signal through the Wit receptor. In order to identify transcriptional targets of the BMP pathway in the Drosophila nervous system, we have analyzed the gene expression profile of wit mutant larval central nervous system. Genes differentially expressed identified by microarray analysis have been verified by quantitative PCR and studied by in situ hybridization. Among the genes thus identified, we find solute transporters, neuropeptides, mitochondrial proteins, and novel genes. In addition, several genes are regulated by wit in an isoform-specific manner that suggest regulation of alternative splicing by BMP signaling. Developmental Dynamics 239:2413–2425, 2010. © 2010 Wiley-Liss, Inc.

INTRODUCTION

Nervous system development and function depend on a cascade of genetically encoded combinatorial codes, direct cell-cell interactions, and intercellular communication via secreted growth factors (Takahashi and Liu, 2006). Many of these signals and their downstream effectors have been identified and characterized in model organisms such as Drosophila, due to its ease of genetic manipulation and stereotyped nervous system (Gerber and Stocker, 2006; Leyssen and Hassan, 2007; Sanchez-Soriano et al., 2007). Peptides of the Transforming Growth Factor-β (TGF-β) super family are pleiotropic intercellular signals involved in virtually every developmental process in metazoans, as reflected in the large number of diseases resulting from mutations in components of the pathway (Harradine and Akhurst, 2006). TGF-βs are required for ectodermal differentiation into epidermis and nervous system, dorsoventral patterning of the neural tube, neural crest induction, neuronal and glial differentiation, neuronal migration, axonal path-finding, and adult neurogenesis (De Robertis and Kuroda, 2004; Bovolenta, 2005; Cayuso and Marti, 2005; Salinas, 2005; Hidalgo et al., 2006; Takahashi and Liu, 2006; Levine and Brivanlou, 2007; Colak et al., 2008).

A more recently described role for the Bone Morphogenetic Protein (BMP) subgroup of the TGF-β super family is to act as enhancers of synaptic transmission, both in flies and in mammals. In Drosophila, signaling by the BMP Glass bottom boat (Gbb) through the type II receptor Wishful thinking (Wit) is required for developmental synaptic growth (Aberle et al., 2002; Marqués et al., 2002; McCabe et al., 2003). BMP signaling is also required as a permissive signal to allow fast homeostatic presynaptic plasticity in response to decreased postsynaptic responsiveness at the neuromuscular junction (NMJ) synapse (Goold and Davis, 2007). In addition to its functions at the NMJ, BMP signaling is also required for proper specification of some FMRFamidergic neurons (Allan et al., 2003; Marqués et al., 2003), synaptic transmission between interneurons and motoneurons (Baines, 2004), and eclosion behavior, likely through the regulation of genes of the neuroendocrine axis. This retrograde signaling activity of BMPs has also been described in the mammalian nervous system. Mouse BMP4 has been identified as a retrograde signal that determines the cell fate of somatosensory neurons by regulating the transcription of positional identity genes (Hodge et al., 2007). In mammals, many of the components of the BMP signaling cassette are expressed in brain areas that participate in memory formation and storage, such as the hippocampus. Furthermore, increased BMP signaling in Chordin mutant mice results in presynaptic enhancement of hippocampal synaptic transmission, and this effect can be recapitulated by bathing hippocampal slices in BMP-7, the mammalian ortholog of Gbb (Sun et al., 2007). It has also been shown that hippocampal neurons in culture express TGF-β receptors I and II, and can secrete and respond to TGF-β2 by nuclear translocation of pSmad and induction of gene transcription (Fukushima et al., 2007). It is then well established that BMPs regulate synaptic morphology and function in mammals and flies.

The canonical TGF-β signal transduction pathway consists of a phosphorylation cascade that results in the stabilization in the nucleus of the complex of the appropriate receptor-regulated Smad and the common Smad, Smad4. This Smad complex acts as a transcriptional regulator, although it has weak DNA binding and transcriptional regulatory activity of its own, and typically requires complex formation with other DNA-binding transcription factors (Ross and Hill, 2007; Schmierer and Hill, 2007). The mechanisms that underlie synaptic plasticity regulation by BMPs appear to be diverse in different biological contexts. While synaptic growth at the Drosophila larval NMJ and neurosecretory cell specification require the activity of the transcriptional regulators of the Smad family Mad and Medea (Marqués et al., 2002, 2003; Allan et al., 2003; Rawson et al., 2003; McCabe et al., 2004), synaptic plasticity at the interneuron-motoneuron synapse appears to be independent of Mad (Baines, 2004). In the mouse hippocampus, the BMP-induced enhancement of synaptic transmission appears to be transcription-independent, based on the fast time course of the response (Sun et al., 2007). The presence of Smad-independent, and possibly transcription-independent, signal transduction by TGF-βs is both well established and intriguing, with no clear molecular delineation of the non-canonical pathways used instead of Smads (Rahimi and Leof, 2007).

Our main interest is the molecular mechanisms of synaptic growth and plasticity, and we use the Drosophila larval NMJ as a model synapse. Our goal is to understand the mechanisms and effectors that translate BMP pathway activation into synaptic growth. Given the need for Mad and Medea in this process (Rawson et al., 2003; McCabe et al., 2004), we reasoned that synaptic growth requires the transcription or repression of specific target genes in response to BMP pathway activation. We decided to identify the genes that are regulated by the BMP pathway in motoneurons with a molecular approach, and compared the whole genome expression profile of wit mutant larval central nervous system (CNS) with genetically matched controls. The genes that showed differential expression in the microarray analysis were further screened by quantitative PCR, and a select group of those that were differentially expressed were analyzed by RNA in situ hybridization. We report here the results of our study, with the list of genes that resulted from our molecular analysis and a brief characterization of some of the most promising targets.

RESULTS

Array Validation

In addition to the pre-hybridization RNA validation using the Agilent analyzer, quality control of mRNA isolation and cDNA preparation is provided by analysis of the ratios of 3′ to 5′ probe set signals for a set of house-keeping genes (Fig. 1A). There was no significant difference in these ratios between the control and experimental group, and the values for actin and GADPH are within the accepted cutoff of 3. The Affymetrix internal controls on the chip show the expected signal levels and the degree of hybridization is consistent between all eight samples (Fig. 1B), indicating that changes in signal levels between the samples are actual experimental variations and not artifacts of uneven hybridization. Analysis of the data sets by principal component analysis (Fig. 1C) shows that the control replicates cluster differently than the wit mutant replicates, indicating that there are consistent differences in global gene expression patterns between mutant and control samples. When we designed the experiment, a major concern was the possible loss of transcriptional targets that were only partially misregulated in wit mutants, either because the gene would be regulated by BMP signaling only in a subset of cells, or because the regulation would be partial instead of all or nothing. The only published transcriptional target of the BMP pathway in the Drosophila embryonic and larval nervous system when we first analyzed these data was the neuropeptide FMRFamide. FMRFa expression in the six Tv neurons of the apterous-positive cluster in the thoracic neuromeres depends on wit, gbb, and Mad (Allan et al., 2003; Marqués et al., 2003). FMRFa expression in another 40 or so cells in the larval brain is not wit-dependent, so we reasoned that if we could detect a decrease in FMRFa expression by microarray analysis in the wit samples, that would serve as validation of our approach. The microarray experiment did indeed detect a 6.6-fold decrease in FMRFa message (Table 1), and this decrease was statistically significant (false discovery rate (FDR) 0.007). Once we verified the quality of our results, we decided to mine the data set for genes that showed a significant (FDR<0.05) expression change and came up with 101 genes (see Supp. Table S1, which is available online; red and yellow dots above the horizontal dashed line in the volcano plot of Fig. 1D), of which 32 show more than 2-fold expression change (red dots outside the vertical dashed lines in the volcano plot). The 32 genes fulfilling those two criteria are listed in Table 1, including the level of expression in controls and wit mutants, the fold change in expression and the FDR value. In order to select genes for further characterization, we studied the differential expression of 27 of those genes by QRT-PCR of controls and wit mutant CNS. Table 2 lists the 12 genes that were thus validated, and includes the gene ontology (GO) term and information on gene expression in the CNS, either from our own in situ hybridization results, from the literature, or from Flyatlas (http://flyatlas.org) and Flybase (www.flybase.org). These genes cover several GO groups, and their characterization is ongoing to determine what role they play in synaptic growth and plasticity at the larval NMJ. In addition, we uncovered two probe sets that correspond to un-annotated transcripts, AY095522 and AY119504. It is unclear if these represent artifactual ESTs or genes that have so far escaped detection by genomic studies. The fact that QRT-PCR confirms the microarray results strongly suggests that AY095522 and AY119504 are real genes whose expression is regulated by BMP signaling. Table 2 also includes eclosion hormone and CG10207 despite their marginal FDR because they were identified in a pilot microarray experiment and QRT-PCR results were confirmatory. The gene partner of bursicon provided another validation for our experimental design (Table 2), as it had been previously found to be down-regulated in wit mutants (Michael B. O'Connor, personal communication). A subset of these genes was studied by in situ hybridization, and representative results from these experiments are shown in Figure 2. The expression patterns found varied from ubiquitous (CG3810, CG7361) to restricted (CG9335 and CG10207) to exquisitely specific (CG13565). In the sections below, we describe some of the more interesting characteristics of these genes as they relate to their regulation by BMP signaling.

Table 1. Differentially Expressed Genes Between Controls and wit Mutantsa
Probe setGeneMutantControlFold changeFDR
  • a

    Thirty-two genes show more than a 2-fold change in expression. Ordered by increasing false discovery rate (FDR).

1638295_s_atCG1129335.87133.952.510.002
1637905_s_atCG2924532.63185.752.870.002
1635715_atHis4350.90146.032.400.002
1636431_atCG876839.6083.68−2.110.003
1628309_atCG5126177.98414.10−2.330.006
1631246_atFMRFa40.15266.75−6.640.007
1628567_atCG13565199.70413.42−2.070.007
1637311_atPartner of bursicon4.8347.15−9.770.011
1639154_atCG31765293.25129.052.270.012
1637602_atCG3810/Edem1104.089.4810.980.013
1634019_atCG206460.55136.53−2.250.014
1624269_atglaikit210.08102.952.040.015
1629242_x_atU141011,547.55639.932.420.016
1628052_atCG10241/Cyp6a174.13245.55−59.530.016
1641538_atCG300173.0325.30−8.360.016
1641174_atCG7361/RFeSP10.58426.50−40.330.017
1630934_atX51967663.051,714.20−2.590.017
1626605_atCG14075491.401,123.70−2.290.018
1631605_atAY09552215.10129.25−8.560.018
1623883_atCG18661236.5280.882.920.018
1631491_atCG933556.68133.35−2.350.019
1635705_atCG1285841.33123.65−2.990.019
1640565_atEjaculatory bulb protein III42.1887.77−2.080.020
1624117_atCG18397129.6846.132.810.024
1641492_atAY119504108.2020.605.250.024
1623478_atCG16752100.4830.833.260.024
1633801_s_atCG917136.90103.67−2.810.030
1636437_atCG3163856.7827.352.080.031
1639829_atCG15382138.3363.082.190.035
1628438_atCG9044192.8083.432.310.041
1639733_s_atCG1427514.9053.78−3.610.042
1623016_atCG129937.97107.18−2.820.047
Figure 1.

Global results and quality assurance of the microarray results. A: 3′/5′ signal intensity ratio of housekeeping control genes (C: control samples, W: wit mutant samples). B: Hybridization controls. Each of the eight colored lines represents the hybridization values of internal chip controls (X axis) to one of the samples, four controls, and four wit mutants. C: Principal component analysis. Red and blue spheres represent wit mutant and control samples, respectively. D: Volcano plot. Red square: FDR ≤ 0.05, Fold Change ≥ 2; Yellow circle: FDR ≤ 0.05, Fold Change < 2; Blue circle: FDR > 0.05.

Table 2. QRT-PCR Verified Genes and Annotation of Literature Expressiona
GeneProbe setMicroarray fold changeFDRQ-PCR fold change(*)GOExpression
  • a

    QRT-PCR average results from three independent sample pairs are shown. The fold change of Cyp6a17 cannot be calculated due to no detectable amplification in wit mutants after 40 cycles of PCR. If we use 40 as a Ct for mutants, fold change can be calculated to be higher than 2,800-fold. I, Our in situ hybridization data; L, literature; F, Flyatlas (www.flyatlas.org); B, Flyexpress-BDGP (www.flyexpress.net, www.fruitfly.org).

CG11291638295_s_at2.510.0021.83Carbohydrate metabolismEmbryonic proventriculusB, embryonic midgutB, faint ubiquitousB
CG135651628567_at−2.070.007−4.14---BrainI, apoptotic amnioserosaB
FMRFa1631246_at−6.640.007−7.39Muscle contraction/ neuropeptide signalingBrainA, L
partner of bursicon1637311_at−9.770.011−30.16Chitin-based cuticle tanning/ neuropeptide hormone activityBrainA, L
CG3810 Edem11637602_at10.980.0133.48N-linked glycosylationBrainI, ring glandI, imaginal discsI, gonadI
CG20641634019_at−2.250.014−3.04Metabolism/oxidoreductase activitySalivary glandF, midgutF
CG10241 Cyp6a171628052_at−59.530.016<-2000Electron transport / steroid metabolismBrainF, Malphigian tubulesI, dorsal metathoracic discB, gonadB, ventral imaginal precursorB
X519671630934_at−2.590.017−9.01Transposable element geneBrainF, headF, eyeF, thoracicoabdominal ganglionF
CG7361 RFeSP1641174_at−40.330.017−263.85Electron transport / ubiquinol-cytochrome-c reductase activityBrainI, ring glandI, gutI, malphigian tubuleI, gonadI, imaginal discsI, salivary glandsI, ubiquitousB
CG140751626605_at−2.290.018−3.99---BainF, headF
CG93351631491_at−2.350.019−3.17---Larval brainI, eye discI, embryonic Bolwig's organB,I, embryonic brainB,I, ventral nerve cordB,I, lateral cord gliaB
CG12991623016_at−2.820.047−3.75proteolysisEmbryonic dorsal epidermisB embryonic/larval dorsal trunkB
AY0955221631605_at−8.560.018−33.35---BrainF
AY1195041641492_at5.250.0245.87Cell adhesionBrainF, headF, eyeF, thoracicoabdominal ganglionF
Eclosion hormone1632960_at−20.108−2.65Neuropeptide signaling / regulation of eclosionBrainA,L
CG10207 NaPi-T1623035_at−2.930.132−2.57Phosphate transport / sodium ion transportBrainI, Malphigian tubuleI, embryonic Malpighian tubuleB, faint ubiquitousB, embryonic midgutB
Figure 2.

Expression of verified target genes in the larval CNS. A–J: In situ hybridization to yw control and wit mutant CNS. Gene names are on each panel. Inset in F shows the purple AP signal filling the presumed axon (arrow) of a CG13565-expressing cell.

Cytochrome Cyp6a17 (CG10241) CNS Expression Is Eliminated in wit Mutants

The expression of Cyp6a17 (CG10241) was decreased by 59-fold in the microarray analysis, and was undetectable by QRT-PCR in wit mutant brains. This is by far the gene most changed in our experiment, and despite the clear molecular data showing expression in the larval CNS, we could not reliably reproduce in situ hybridization experiments with this gene. Some early results showed weak generalized expression, but reproducibility is poor. Some possibilities include very low level, ubiquitous expression, difficult to sort from background alkaline phosphatase signal, or conjugation of the Cyp6a17 mRNA with proteins in a complex that blocks probe hybridization to the transcript (Steward and Schuman, 2003; Haussmann et al., 2008). The lack of CNS signal is not due to technical reasons as we detect probe hybridization to Malpighian tubules. In the absence of specific spatial information on Cyp6a17 transcription, we cannot discern if Cyp6a17 is a direct target of the Wit pathway or depends on a secondary signal induced by BMP signaling. Nevertheless some of the wit mutant phenotypes so far described or still uncovered may be due to the loss of the putative activity of Cyp6a17 in the nervous system. Cyp6a17 is enriched in the adult brain, but it is also expressed in many other tissues (http://flyatlas.org/atlas.cgi?name=FBgn0015714). Cyp6a17 encodes a predicted microsomal cytochrome P450 and maps to an eight Cyp450 cluster in chromosome two. No mutant alleles have been characterized, and the closest human homolog is CYP5A1 or Thromboxane A Synthase 1, the causative gene in Ghosal Hematodiaphyseal Syndrome (Genevieve et al., 2008). It is possible that the BMP pathway is regulating lipid hormone metabolism in the Drosophila CNS through activation of Cyp6a17.

Post-Transcriptional Regulation by Wit Signaling of the Rieske Iron-Sulfur Protein Encoded by CG7361

The predicted Rieske Iron-Sulfur Protein (RFeSP) encoded by CG7361 is the second most changed gene in our array, with a 40-fold decrease in expression levels as detected by probe set 1641174_at. By QRT-PCR, RFeSP was decreased between 80- and 423-fold, barely above our level of detection. However, in situ hybridization showed diffuse general expression over the nervous system and no decrease in expression in wit mutants (Fig. 2A, B). This apparent discrepancy is resolved when looking at a different probe set for the same gene, 1631856_a_at, which shows no change in expression (Fig. 3A, B). The two probe sets represent different mRNA species of RFeSP, and semi-quantitative PCR shows that only one of them, RFeSP-RB, is expressed in wit mutants (Fig. 3C). The RNA sequence detected by probe set 1641174_at that is present in RFeSP-RA and absent in RFeSP-RB is only 55 bases long, and we have not been able to determine by in situ hybridization if RFeSP-RA is specifically present in BMP signal-receiving cells or if it is generally expressed like RFeSP-RB. The loss of one of the two mRNA species of RFeSP could be due to lack of transcription of RFeSP-RA, preferential splicing of the pre-mRNA into the RFeSP-RB isoform, or enhanced degradation of RFeSP-RA in wit mutants. Both mRNA species are abundantly represented by ESTs from a wide variety of tissues (http://flybase.org/reports/FBgn0021906.html), and the gene model prediction is that both mRNAs share a promoter and hence arise from a common pre-mRNA (Fig. 3A). This seems to rule out direct regulation of RFeSP-RA transcription by Smads, and suggests that the Wit pathway regulates splicing or mRNA stability. A recent report has shown in a mammalian system that TGF-β signaling regulates the levels of microRNA in a Smad-dependent manner, and this results in the degradation of specific mRNAs (Davis et al., 2008). It is then possible that Wit signaling is regulating RFeSP-RA stability instead of splicing of the RFeSP pre-mRNA. However, search of the Drosophila miR database (Brennecke et al., 2005) fails to identify a microRNA whose target sequence matches the 55 ribonucleotides unique to RFeSP-RA mRNA.

Figure 3.

Differential regulation of RFeSP expression in wit mutants. A: Genomic and mRNA structure of RFeSP. Lines, introns; bars, exons; grey bars, non-coding exons. White arrowheads, PCR primer pairs for transcript A used for QRT-PCR in Table 2; black arrowheads, primer pairs for transcripts A and B used for semi-quantitative PCR in C; white bar, Affymetrix probe set region for transcript A; black bars, Affymetrix probe set regions common for both transcripts. B: Microarray results: 1641174_at probe set (white background, specific for transcript A), 1631856_a_at probe set (black background, common for both transcripts). Test microarray: unpublished pilot microarray data (see methods). C: Semi-quantitative PCR results: cont, yw; wit: wit mutant. A: transcript A band, 421 bp; B: transcript B band, 366 bp. D: Predicted protein domain structures of RFeSP isoforms. UCR-TM: Ubiquinol cytochrome reductase transmembrane region; Rieske_Cytochrome_bc1: a [2Fe-2S] cluster binding domain found in mitochondrial cytochrome bc(1) complexes. RFeSP-PA has a truncated Rieske domain.

Isoforms of the Edem1 Encoded by CG3810 Are Differentially Regulated by BMP Signaling

When we noticed that different transcripts of RFeSP were differentially regulated by Wit, and that this was likely due to post-transcriptional regulation, we reanalyzed our dataset to search for genes in which only some of the probe sets for a gene were differentially expressed in wit mutants. We found several genes in which some probe sets were affected while others were not, or when different probe sets were affected in opposite directions (Table 3). Among those that showed the most significant difference in expression (lower FDR), we focused on CG3810, which encodes an endoplasmic reticulum degradation-enhancing alpha-mannosidase-like protein (Edem1) and is expressed in three different mRNAs (Fig. 4A). Of the two Edem1 probe sets in the microarray, only the probe set specific for the non-coding exon at the 3′ end of isoform Edem1-RC showed a significant change in expression, an 11-fold increase in message levels in wit mutants (Table 3, Fig. 4B). The probe set that detects all Edem1 isoforms shows no significant change of expression (FC 1.88, FDR 0.239). The microarray-detected increase in Edem1-RC expression was verified by QRT-PCR (4-fold increase over controls, Table 2) and the results of our pilot microarray experiment (Fig. 4B). Contrary to RFeSP, Edem1 isoforms are predicted to be transcribed from different promoters (Fig. 4A), so the effect of wit on Edem1-RC expression could be due to either transcriptional repression or post-transcriptional effects. In situ hybridization showed general staining of the CNS (Fig. 2C). Unsurprisingly, no difference is observed in wit mutants, as the riboprobe detects all three Edem1 isoforms (Fig. 2D). This generalized expression of Edem1 in the nervous system makes it an unlikely direct transcriptional target of the Wit pathway, and like Cyp6a17 it may be regulated by a secondary signal that depends on BMP signaling. Alternatively, it is possible that Edem1-RC is a direct target specifically expressed in a subset of CNS cells that receive BMP signaling, such as motoneurons.

Table 3. Candidate Genes for Isoform-Specific Regulationa
GeneProbe setMutant signalControl signalFold changeFDRDetectable transcript
  • a

    Manually selected probe sets corresponding to the same gene that show different change in expression levels in wit mutants.

lola1624729_at242.23346.55−1.430.030L
lola1635096_at406.90211.031.930.071K
CG16371623285_at464.53375.401.240.767C
CG16371624797_at126.2555.782.260.053A
CG16371633563_at115.3285.751.340.694B
oaf1626218_s_at143.8378.051.840.175A,B,D
oaf1627494_s_at206.48170.401.210.919A,D
dlg11624021_a_at106.8368.951.550.391K,L,D,A,H,G,B,E
dlg11627882_at8.7518.95−2.170.173C,J,I
dlg11635382_at48.8347.271.030.635F
CG171241633795_a_at685.60354.651.930.069A,B
CG171241640465_at1832.881195.631.530.315B
CG3810/Edem11632200_s_at120.5864.071.880.239A,B,C
CG3810/Edem11637602_at104.089.4810.980.013C
CG10207/RFeSP1631856_a_at893.60708.701.260.355A,B
CG10207/RFeSP1641174_at10.58426.50−40.330.017A
Figure 4.

Differential regulation of Edem1 expression in wit mutants. A: Genomic and mRNA structure of Edem1. Lines, introns; bars, exons; grey bars, non-coding exons. White arrowheads, PCR primer pairs for transcript C used for QRT-PCR in Table 2; white bar, Affymetrix probe set region for transcript C; black bar, Affymetrix probe set region common for all transcripts. B: Microarray results: 1637602_at probe set (white background, specific for transcript C); 1632200_s_at probe set (black background, common for all transcripts). Test microarray: unpublished pilot microarray data (see Experimental Procedures section).

The Na/Pi Co-Transporter CG10207 Is Regulated by Wit Signaling

CG10207 was identified as decreased 2.9-fold in the microarray analysis, and pursued despite its marginal FDR (0.132) because it had also been detected in a prior pilot experiment (see Experimental Procedures section). QRT-PCR shows a 2.5-fold decrease in message level and in situ hybridization to larval tissues shows expression restricted to a subset of cells in the CNS (Fig. 2E) and the Malpighian tubules (not shown). This pattern of expression is compatible with CG10207 being a direct transcriptional target for the Wit canonical pathway, which is largely restricted to the ventral ganglion (as determined by wit-dependent nuclear accumulation of pMad; Marqués et al., 2003). CG10207 encodes a transmembrane protein of the major facilitator super family. The closest human homologs are members 5–8 of the solute carrier family 17A (SLC17A). SLC17A6, SLC17A7, and SLC17A8 have been described as vesicular glutamate transporters (VGLUT) with specific punctate localization in neurites, presumably at presynaptic active zones (Herzog et al., 2006). It is not clear though if the cells where CG10207 is expressed are glutamatergic motoneurons or other cell types. In situ hybridization to wit mutant CNS shows partial down-regulation of CG10207 (Fig. 2F), as expected from the relatively low down-regulation of the transcript measured by molecular methods (2.9–2.5 FC).

The Ly-6 Peptide CG9355 Is Regulated by Wit Signaling

Another tissue-restricted target we have validated is CG9335, which shows a modest decrease in expression in wit mutants by microarray analysis (2.3-fold, FDR 0.02) and QRT-PCR (3-fold). In the third instar larval nervous system, CG9335 is expressed in what looks like the inner proliferating center of the optic lobes and in discrete cells of the ventral ganglion that look like motoneurons, based on stereotypic pattern and location and co-staining with other markers (Kim and Marqués, unpublished data). CG9335 is not expressed in the ring gland (Fig. 2G). This pattern of expression makes CG9335 an excellent candidate for a direct transcriptional target of the Wit pathway, which is active in motoneurons in the ventral ganglion. The relatively low fold change in CG9335 expression in wit mutants is explained by the expression in the optic lobes, which remains unaffected in the absence of BMP signaling, while expression in the ventral ganglion is nearly abolished (Fig 2H). CG9335 encodes a putative secreted peptide of 166 amino acids of the Ly-6 superfamily of GPI-anchored molecules that have been described in several systems, including Drosophila. Members of this family act as modulators of the nAChR (Miwa et al., 2006), septate junction proteins (Hijazi et al., 2009), and regulate traffic of ion channels (Wu et al., 2010).

CG13565 Encodes a Putative Neuromodulator Regulated by BMP Signaling

A final class of wit-regulated genes found in our screen is represented by CG13565, with an exquisitely specific pattern of expression (Fig. 2I). CG13565 is expressed in only six pairs of bilaterally symmetrical cells in the larval CNS, one pair in the central brain, and 10 cells in the ventral ganglion, possibly the most anterior abdominal segments. The ventral ganglion cells appear neurons, based on the axon-like processes that appear filled by the strong alkaline phosphatase signal in the mRNA in situ hybridization (Fig. 2I, inset). In agreement with the modest decrease in CG13565 expression in wit mutants measured by molecular methods (2–4-fold change) and the really strong signal in control CNS, in situ hybridization to wit mutant CNS (Fig. 2J) shows a pattern similar to controls. These neurons look like neurosecretory cells, and this limited expression pattern makes it likely that CG13565 is a direct target of Wit. In addition to motoneurons, other cells in the larval CNS receive BMP signaling, based on their staining with anti-phosphorylated Mad (Marqués et al., 2003). Among them are the Tv neurons that secrete FMRFamide (Allan et al., 2003; Marqués et al., 2003), the peptidergic lateral cluster and Va neurons (Miguel-Aliaga et al., 2004), and the cells that produce the insulin-like neuropeptide Ilp7 (Miguel-Aliaga et al., 2008). Based on our microarray data and results previously obtained by Michael O'Connor (personal communication), the subset of crustacean cardioactive peptide (CCAP)-expressing neurons that produce the hormone partner of bursicon (pburs) (Luan et al., 2006) is also the target of BMP signaling. Other neuropeptides, such as eclosion hormone (EH), Insulin-related peptide 2 (CG8167), and adipokinetic hormone (Akh, CG1171) are slightly down-regulated in wit mutants in our microarray (Suppl. Table S1), although outside of the stringent selection criteria used in Table 1. However, EH does not appear to be down-regulated at the protein level in wit mutants (not shown). CG13565 encodes a 133 amino acid peptide with a secretion signal peptide. CG13565 has no clear mammalian homologs, and it probably represents a novel hormone or neuromodulator. Although there are related proteins in C. elegans (F46C5.1 and F46C5.10, Wormbase, www.wormbase.org), no information is available on their role in worm development or physiology.

DISCUSSION

Gene Expression Profiling Identifies BMP Pathway Target Genes in the Fly Nervous System

Differential expression profiling has become a powerful technique to study the effect of signaling pathways on responding tissues. This technique is particularly useful when little is known of the mechanistic details of the signal transduction pathway and its primary transcriptional effectors, or when these effectors have an unknown or degenerate DNA sequence as a response element. It is well established that BMP signaling through the Wit type II receptor is required for synaptic growth and function of Drosophila motoneurons (Marqués and Zhang, 2006), but no transcriptional targets have been identified so far that can link lack of BMP signaling with the phenotypic outcome at the synaptic terminal. The primary transcriptional effectors of the BMP pathway in Drosophila, Mad and Medea, are also required for synaptic terminal growth and function. The simplest possible model posits that transcriptional regulation of genes essential for synaptic growth and function in response to retrograde BMP signaling is a key aspect of developmental synaptic plasticity in flies. Given the short and poorly conserved DNA sequence of the Smad response elements identified so far, a genome-wide bioinformatic analysis in search of transcriptional targets of the Wit pathway is unlikely to be fruitful.

We decided to search for BMP transcriptional targets by comparing the genome-wide transcriptome of wit mutants and genetically matched controls. The neuronal BMP pathway is active from late embryo until late larval stages (Marqués et al., 2002, 2003). Identification of early response genes would require differential expression profiling in late embryonic or early larval stages. However, the number of cells responding to the Wit pathway is a small fraction of the nervous system, let alone the complete embryo, and we feared that transcriptional differences in motoneurons would be diluted out in the total embryonic RNA. Dissection of the embryonic CNS is a difficult procedure and prohibitive when hundreds of them are required; so we decided to perform the analysis in late third-instar larvae. We reasoned that as the pathway is active throughout larval development, the early response genes would be picked alongside the genes regulated by secondary transcriptional events. In situ hybridization of the identified targets in embryos and other analyses would allow us to distinguish primary from secondary targets. Furthermore, disregulation of secondary targets may be responsible for some of the phenotypes of wit mutants. This choice of tissue was a compromise between ease of material procurement and sensitivity of detection. Ideally, we would have selected embryonic motoneurons for the analysis, or performed tissue-specific mRNA capture (Roy et al., 2002), but technical difficulties prevented these approaches. As such, we fully acknowledge that we may have missed targets whose expression changes in response to Wit signaling is moderate, particularly if those genes are abundantly expressed in areas of the nervous system that do not respond to the BMP pathway. On the flip side, by looking at the whole CNS we can identify targets in interneurons and neurosecretory cells, and can use FMRFa as an internal control for our experiment.

While the overwhelming evidence indicates that all Wit responses in motoneurons are mediated through Smads, it is also possible that some non-canonical pathways contribute to BMP neuronal signaling (Baines, 2004; Eaton and Davis, 2005; Sun et al., 2007). To the extent that these Smad-independent pathways operate through transcriptional regulation (Lee et al., 2007), their targets would also have been identified in this screen.

A previous report has described comparative expression analysis of forced activation of the two branches of the TGF-β pathway, BMP, and activin, in the Drosophila larval CNS (Yang et al., 2004). That study found 91 genes that were differentially expressed in the BMP pathway over-activated sample, and 216 differentially expressed genes in the activin over-active sample when compared to controls. There is no overlap between the data set of Yang et al. and our results, and this is not particularly surprising. Two main important aspects differentiate their experimental approach from ours. First, it is unclear that driving the expression of the constitutively active form of the BMP receptor Tkv (TkvCA) mimics BMP/Gbb signaling in Drosophila motoneurons. BMP signaling requires both type I receptors, Tkv and Sax. TkvCA cannot induce synaptic growth in a wild type background (McCabe et al., 2004), rescue wit mutations, nor induce phosphorylated Mad accumulation in neurons when over-expressed (M.B. O'Connor, personal communication). Second, Yang et al. (2004) over-expressed TkvCA in all cells of the nervous system (glia and neurons) for a brief period of time in late larval development. In this report, we are looking at the loss of BMP signaling only in the cells that actually receive the signal in vivo, and the loss of BMP signaling is maintained through development. The absence of overlap between the two data sets is then not surprising, as the experimental set-ups are very different: transient, ubiquitous gain of function in Yang et al. (2004), versus maintained loss of function in cells that normally receive a BMP signal in this report.

One disadvantage of our approach is that some of the identified targets may not be directly regulated by Mad/Medea, but instead regulated by some of the early genes that are direct targets of the Wit pathway. Co-expression of Wit and the target gene in embryonic motoneurons, as in the case of CG9335 (Kim and Marqués, unpublished data), is a good indication of direct regulation. Computational prediction of Smad response elements in the regulatory regions of the target genes is also suggestive of direct regulation. However, the poorly defined Mad and Medea DNA binding site consensus makes it difficult to identify BMP-responsive enhancers. Parsing of the literature for experimentally determined Mad-binding sites and excluding those defined only as GNCN, we built a position-weight matrix from which we derived the consensus sequence YGSCGNSG. A search for this sequence in the vicinity of the six genes described in this report was inconclusive, due to the high number of hits caused by the high degeneracy of the consensus. The search was negative for the Mad-Shn-Medea transcription inhibitory signature, GRCGNCNNNNNGTCTG (Pyrowolakis et al., 2004). The BMP-responsive enhancers in our targets will have to be identified by DNA footprinting or computational methods that do not require knowledge of the motif sequence, and then experimentally validated for Mad binding and responsiveness to BMP signaling.

BMP Signaling Regulates mRNA Splicing or Stability

We have shown that the Wit pathway individually regulates different mRNA species of the Edem-1 encoded by CG3810 and the RFeSP encoded by CG7361. Although we have not studied in detail the other genes in Table 3, it seems likely that several of them will also be subject to isoform-specific regulation by Wit signaling. These genes were hand-picked from the raw primary data, and systematic bioinformatics analysis of the microarray primary dataset would be needed to identify more genes that show differential isoform regulation. The Affymetrix GeneChip® Drosophila Genome 2.0 Array chips used in these experiments are not designed for isoform detection, and only about 900 out of roughly 14,000 Drosophila genes are represented by multiple probe sets that could potentially detect different splicing isoforms. With the data we have now, it appears that two different mechanisms could be at play in wit-dependent isoform-specific regulation. RFeSP isoforms share the same promoter, and the two mRNA species are predicted to be derived from a common pre-mRNA. In this case, the effect of BMP signaling has to be exerted through splicing regulation to promote RFeSP-RA formation, or by decreasing the degradation of RFeSP-RA. This could be accomplished by down-regulation of a microRNA that may target isoform RFeSP-RA for degradation. In this respect, it must be noted that the Affymetrix microarrays do not contain miRNA probe sets, so our data cannot address the possible regulation of miRNA expression by the Wit signaling pathway.

Transcriptional regulation of an alternative splicing regulator could result in the specific expression of particular isoforms in cells subject to BMP signaling. Several of these regulators have been identified (Wang and Cooper, 2007; David and Manley, 2008), and their role in developmental regulation is perhaps best exemplified by the role of Drosophila Sxl in the sex determination cascade (Penalva and Sanchez, 2003). Given the restricted expression and function of wit, the alternative splicing regulator could also be a ubiquitous one, such as members of the SR proteins and hnRNP families. These events would be difficult to detect in our microarray analysis due to the noise contributed by cells that do not receive a Wit signal. In the case of Edem1, in addition to the previous mechanisms of mRNA splicing and differential degradation, direct transcriptional regulation of Edem1-RC message by the Wit pathway is a possibility, as it is transcribed from a different promoter than the unaffected transcripts Edem1-RA and Edem1-RB. Finally, it is also possible that the Wit pathway directly regulates splicing through modulation of splicing regulators activity (as opposed to their transcription), for example by phosphorylation through a non-canonical pathway.

A final consideration regards the biological relevance of isoform-specific regulation of Wit target genes. It is possible that the regulated isoforms have a different pattern of temporal and spatial regulation than the unregulated ones, although we have not been able to obtain in situ hybridization data for either Edem1-RC or RFeSP-RA. While the different isoforms of Edem1 are predicted to encode identical proteins, the two splicing isoforms of RFeSP translate into different proteins with potentially different cellular functions.

Do These Targets Explain the wit Phenotypes?

wit mutants have a complex phenotype that includes defects in motoneuron synaptic growth, with morphological, electrophysiological, and ultrastructural phenotypes, in addition to loss of FMRFa expression in Tv neurons (Aberle et al., 2002; Marqués et al., 200, 2003; Allan et al., 2003). We have identified and confirmed 11 genes that are differentially expressed in wit mutants, plus isoform-specific regulation of another two genes, and potentially several others. In addition, we have also found two un-annotated genes that are subject to BMP regulation. It is likely that several of these genes contribute to the different phenotypes of wit mutants, and quite possible that mutations in any single one of the wit targets fail to recapitulate a wit loss of function phenotype. The gene most changed by loss of wit signaling is Cyp6a17. Cytochrome P450 family members catalyze oxidative reactions (oxidation and hydroxylation) in the biosynthesis of the insect hormone ecdysone, and also participate in the degradation and detoxification of drugs and metabolites (Chung et al., 2009). How a cytochrome P450 might be related to the wit mutant phenotype is unclear, although considering that wit mutants fail to eclose it is tempting to speculate on faulty hormonal synthesis or degradation (ecdysone, juvenile hormone) that may influence some of the secondary changes required upon metamorphosis to accomplish eclosion behavior, a process that is disrupted in wit mutants. Similar issues may result from loss of the putative neuromodulator CG13565. Some of these neuromodulators and peptide hormones are only partially down-regulated in wit mutants, but it is possible that the cumulative effect of partial loss of several signals has a severe effect on eclosion.

RFeSP is predicted to be an ubiquinol-cytochrome-c reductase involved in mitochondrial electron transport. Characterization of the number of synaptic mitochondria and their functional status by Mitotracker staining did not evidence any difference between wit and controls (not shown). In addition, the mRNA that is not detected in wit mutants, RFeSP-RA, is predicted to encode a protein with a different activity, as it lacks a functional iron-sulfur domain (Fig. 3D). What the function of this isoform may be, if any, is unclear. Recent results indicate that splicing regulation can shift the ratio between full-length, coding isoforms to non-coding versions of the transcript. This mechanism is developmentally regulated and modulates the activity of basic helix-loop-helix (bHLH) transcription factors during retinal development (Kanadia and Cepko, 2010). It is possible that BMP signaling is regulating RFeSP activity by modulating the relative levels of functional (RFeSP-PB) versus non-functional protein (RFeSP-PA).

CG10207 encodes a transmembrane protein of the major facilitator superfamily. The closest human homologs are members 5–8 of the solute carrier family 17A (SLC17A). SLC17A6, SLC17A7, and SLC17A8 have been described as vesicular glutamate transporters (VGLUT) with specific punctate localization in neurites (Herzog et al., 2006). In Drosophila, DVGLUT1 is required for proper synaptic transmission at the NMJ, as loss of the transporter results in empty synaptic vesicles (SV) and a decrease in the frequency of spontaneous release (mEPSP) (Daniels et al., 2006). The frequency of mEPSP is decreased in wit mutants (Aberle et al., 2002; Marqués et al., 2002), and this is compatible with a decreased expression of a VGLUT. DVGLUT1 mutants have normal amplitude of mEPSP, while wit mutants have 80% of wild type (Aberle et al., 2002). It is possible that DVGLUT1 and CG10207 are both required to completely fill a SV with glutamate, although in this case DVGLUT1 mutants should have mEPSPs of different amplitudes. However, based on their location within the CNS, cells expressing CG10207 are unlikely to be motoneurons, and then CG10207 could be a secondary target of the BMP pathway, like Cyp6a17, and the phenotypic effect of loss of CG10207 in these cells in wit mutants has not been uncovered yet.

Based on the molecular functions of Ly-6 GPI-anchored molecules, CG9335 may be regulating interneuron cholinergic input to motoneurons (Miwa et al., 1999) or acting as a component of intercellular junctions (Hijazi et al., 2009). In the second case, loss of cell-cell adhesion at the NMJ synaptic terminal may explain the detachment of pre- and postsynaptic membrane seen in wit (Aberle et al., 2002; Marqués et al., 2002) and other BMP pathway mutants (McCabe et al., 2004). Alternatively, like its homolog Quiver (Wu et al., 2010), CG9335 could be regulating the subcellular traffic of ion channels and thus influencing synaptic transmission at the NMJ.

Lastly, the molecular identities of the verified targets do not shed a lot of light on the link between BMP signaling and synaptic growth. These targets will have to be investigated by standard genetic and molecular methods to ascertain if they are indeed direct transcriptional targets of the BMP pathway and to show a phenotypic connection between their function and synaptic terminal growth and function.

In summary, by studying the expression profile of wit mutants we have found just over a 100 genes that were significantly and differentially expressed, and of those only a third showed over a 2-fold change in transcript levels. This probably reflects the mRNA dilution problems mentioned above. Over half of the genes tested were significantly changed by QRT-PCR, and those are the ones we concentrated on, uncovering a potential role of BMP signaling in splicing regulation and identifying genes that may be directly involved in synaptic growth and plasticity. We must stress that this is a first pass characterization of our data, as a proof of concept and a way to share our results. More powerful analysis of our primary microarray data is indeed possible, as we used a conservative 2-fold change in expression as our cut-off, and motoneurons represent a small fraction of the total cells in the larval CNS. The BMP-regulated genes found in this study can also be used as internal controls when repeating the experiment in embryo or early larvae using tissue-specific mRNA selection techniques or more stringent tissue selection procedures.

EXPERIMENTAL PROCEDURES

Fly Stocks

Trans-heterozygote third instar larvae of the genotype bw; witA12, st/witB11, st were obtained by selecting Tb+ larvae from a cross of bw; witA12, st/TM6B to bw; witB11, st/TM6B. Genetically matched control animals of the genotype bw; l(3)64AkF8, st/ l(3)64AlA9, st were Tb+ larvae from a cross of bw; l(3) 64AkF8, st/TM6B to bw; l(3) 64AlA9, st/TM6B. These mutants came from the same mutagenesis screen as the wit alleles (Harrison et al., 1995), and are then in the identical genetic background. For in situ hybridization, yw larvae were used in most cases. Flies were reared in standard yeast/cornmeal/sugar medium and kept at 25°C in a 12-hr day/night cycle.

Preparation of Samples for Microarray Analysis

Six independent samples were prepared for each genotype with 50 central nervous systems (CNS) each. Tb+ third instar larvae from each cross were collected and dissected in dissecting solution (20 mM Na2HPO4, 145 mM NaCl, 1 mM Na2EGTA, 2 mM MgCl2, pH 7.2) between 9 am and 12 pm to reduce circadian rhythm effects on gene expression, and the dissected brains (including the ring gland but with the eye-antennal disks removed) placed in RNAlater (Ambion, Austin, TX). Tissues were stored in RNAlater at −80°C until all samples were harvested. Total RNA was obtained from all samples at the same time. Briefly, RNAlater was exchanged with DEPC-treated PBS and total RNA purified with TRIZOL® LS Reagent (Invitrogen, Carlsbad, CA) following the manufacturer's instruction. Purified total RNA was treated with 10 U RQ1-DNase (Promega, Madison, WI) for 30 min at 37°C and then DNase was removed by TRIZOL® LS Reagent purification. Typically, 10 μg total RNA was obtained for each sample. Total RNA was quantified by OD260, and quality tested using an Agilent bioanalyzer and quantitative real-time PCR of control genes, such as actin5C, and target genes identified in a previous pilot screen (see below). The best four pairs of samples were used for the microarray analysis.

Microarray Hybridization and Data Analysis

Microarray experiments using Affymetrix GeneChip® Drosophila Genome 2.0 Array chips (http://www.affymetrix.com), including cDNA synthesis, amplification, labeling, hybridization, and signal detection were done in the Comprehensive Cancer Center of UAB following the manufacturer's instructions. Raw data were extracted using Affymetrix (Santa Clara, CA) Genechip Operating Software and analyzed by HDBstat (http://www.ssg.uab.edu/hdbstat/index.html) after filtering out probe sets that have more than 50% absent calls (i.e., more than four absent calls in the eight samples). After normalizing data by Quantile-Quantile normalization, we calculated the equal-variance t-tests and false discovery rate (FDR). Fold changes (FC) were calculated using the means of unadjusted intensity values. The Pearson correlation coefficient and a deleted residuals approach were used to examine outliers (Trivedi et al., 2005). There is no extreme outlier among the samples analyzed (not shown). Principal component analysis (PCA) was done with Genespring GX (Agilent). In addition to the microarray analysis performed at UAB, we did a pilot experiment through a commercial supplier (Ambion) using three replicates, and used some of those results to confirm the data of our microarrays. For the analysis of independent regulation of different isoforms, genes represented in the Affymetrix GeneChip® Drosophila Genome 2.0 Array chips by multiple probe sets were analyzed for changes in expression between mutant and control, which were significantly different (P < 0.01). Those probe sets were then inspected to see if they were specific for distinct gene isoforms.

Quantitative Real-Time PCR

Three independent sample pairs (wit and control) of total RNA were prepared for Quantitative Real-time PCR (QRT-PCR) as above. Three micrograms of total RNA of each sample were used to prepare cDNA. Reverse transcriptions were done with the AffinityScript™ QPCR cDNA synthesis kit (Stratagene, La Jolla, CA) according to the manufacturer's instruction. For real time PCR, 1/100 of a cDNA reaction was used in a 25-μl QRT-PCR reaction. A Cepheid Smartcycler was used for SYBR-based QRT-PCR reaction and melting curve generation. Triplicate reactions were performed for each gene. Fold changes for each gene were calculated by 2-ΔΔCt method (Livak and Schmittgen, 2001) using actin5C as a control. The primer sequence list is presented as Supp. Table S2.

Semi-Quantitative PCR

Semi-quantitative PCR for two transcripts of RFeSP was done using Cepheid Smarcycler with primer pairs amplifying two transcripts simultaneously (Forward primer: 5′-GCAAGTCGGTTACTTTCAAGTG-3′, Reverse primer: 5′-GGGGCTTAAAC GATAACAGCTC-3′). Linear amplification regions were determined using Real-time SYBR signal curves. For actin5C control and RFeSP transcripts, 23 cycles and 33 cycles of PCR reaction were done, respectively.

In Situ Hybridization

cDNA clones for each target gene were provided by the Drosophila Genomics Resource Center (DGRC, https://dgrc.cgb.indiana.edu). The reverse transcription reactions for some pOT2-based clones were not successful in our hands, and those genes were sub-cloned into pBluescript II KS(+) and used for RNA probe generation (Supp. Table S3). After appropriate 5′ or 3′ restriction enzyme digestion, digoxigenin (Dig) labeled anti-sense or sense probes were generated in a 10-μl reaction containing 1 μg linearized plasmid, 1 μl 10× Dig RNA labeling mix (Roche, Nutley, NJ), 5 mM DTT, 40 U RNAsin (Promega), and 1 μl (17–18 U) of the appropriate reverse transcriptase (Promega). One of ten of the RNA probes was hydrolyzed, precipitated, and re-suspended in 100 μl hybridization solution (50% formamide, 5×SSC, 50 μg/ml heparin, 0.1 % Tween-20, 100 μg/ml salmon sperm DNA). Whole third-instar larvae carcasses flipped inside out or dissected brains were fixed in 4% formaldehyde for 30 min at room temperature. After several washes with DEPC-treated PBS (20 mM Na2HPO4, 145 mM NaCl, pH 7.2) and treatment with Proteinase K (Promega) for 5 min at room temperature, pre-hybridization was carried out in hybridization solution for 2 hr at 55°C. RNA probe hybridizations were performed for 30–40 hr at 55°C. The hybridized samples were washed several times over 2–3 days with hybridization solution at 55°C. After serial washings with 80, 60, 40, and 20% hybridization solution diluted with DEPC treated PBT (PBS 0.1% Tween-20) and five PBT washings, the samples were incubated for 1 hr at room temperature with pre-absorbed anti-Digoxigenin-AP Fab fragments (Roche) diluted 1:1,000. After several washing steps in PBT, samples were washed in AP staining buffer (100 mM NaCl, 50 mM MgCl2, 100 mM Tris, pH 9.5, 0.1% Tween-20). Color development was done in AP staining buffer containing 338 μg/ml NBT (Promega) and 175 μg/ml BCIP (Promega).

Bioinformatic Analysis of Candidate Gene Regulatory Regions

Mad and Medea binding sites in the Drosophila genome have been experimentally determined for both activating (Kim et al., 1997; Xu et al., 1998; Halfon et al., 2000; Knirr and Frasch, 2001; Rushlow et al., 2001; Grienenberger et al., 2003; Wharton et al., 2004; Lee and Frasch, 2005; Xu et al., 2005; Lin et al., 2006) and inhibiting (Muller et al., 2003; Pyrowolakis et al., 2004; Gao et al., 2005; Yao et al., 2006; Gao and Laughon, 2007; Walsh and Carroll, 2007) cis regulatory elements (CRM). We collated the different sites from the literature and used the program eCis-analyst (Berman et al., 2002) to generate consensus sequences that were used with the program Genome Enhancer (Markstein et al., 2002) to search for individual or clustered BMP response elements in the vicinity of differentially expressed genes.

Acknowledgements

Dr. Lihong Teng performed the microarray hybridization at the Comprehensive Cancer Center of UAB, and Dr. Xiangqin Cui, Genetics and Translational Medicine, UAB, provided help with the statistical analysis. We are grateful to Dr. Rosa Serra for use of her Cepheid Smartcycler QRT-PCR machine and Dr. Douglas Ruden for the gift of Affymetrix chips. Dr. Michael O'Connor is gratefully acknowledged for sharing results ahead of publication and comments on the manuscript. We thank the anonymous manuscript reviewers for their helpful suggestions. Research was supported by funds from the State of Alabama and a Pew Scholars Award in the Biomedical Sciences to G.M.

Ancillary