An embryonic system to assess Wnt transcriptional targets

During animal development, complex signals determine and organize a vast number of tissues using a very small number of signal transduction pathways. These developmental signaling pathways determine cell fates through a coordinated transcriptional response that remains poorly understood. The Wnt pathway is involved in a variety of these cellular functions, and its signals are transmitted in part through a β-catenin/TCF transcriptional complex. Here we report an in vivo Drosophila assay that can be used to distinguish between activation, de-repression and repression of transcriptional responses, separating upstream and downstream pathway activation and canonical/non-canonical Wnt signals in embryos. We find specific sets of genes downstream of both β-catenin and TCF with an additional group of genes regulated by Wnt, while the non-canonical Wnt4 regulates a separate cohort of genes. We correlate transcriptional changes with phenotypic outcomes of cell differentiation and embryo size, showing our model can be used to characterize developmental signaling compartmentalization in vivo.


Introduction
Signaling pathways elicit cellular responses in part by regulating the transcription of specific cohorts of target genes. Signaling pathways that are crucial for development, homeostasis and tumorigenesis have both negative (repressive) and positive (activation and de-repression) effects on transcription 1 .
Negatively regulated targets can have different biological activities from positively regulated targets 2 .
The Wnt signaling pathway provides a striking example of this phenomenon where Wnt signals can elicit a variety of cellular responses including differentiation, growth, and polarity 3-6 . Increased Wnt signaling has apparently opposite roles on cell proliferation: excessive Wnt signaling leads to over proliferation of cancer cells, but it is also required to maintain undifferentiated, quiescent stem cells 7,8 .
As a result, therapeutic interventions that block Wnt in tumours are likely to have both beneficial and detrimental consequences; when cancer growth is inhibited, useful stem cells are likely to be lost as a side-effect 9,10 . It is probable that these opposing effects occur through the transcriptional activation of different Wnt target genes, raising the intriguing possibility that therapies targeted downstream could avoid the detrimental effects of disrupting the whole pathway.
Wnt signalling and its dysregulation has been implicated in a variety of developmental disorders and diseases, including diabetes, Robinow Syndrome, cancer and aging [11][12][13] . Wnt regulation appears to be highly context-specific, affecting different genes in different cell types at different developmental stages 14 , and the features defining how a Wnt target gene is regulated are not fully understood 15 .
Wnt signaling refers to a series of signaling pathways or networks divided into non-canonical and canonical. Non-canonical signaling is a collection of signal transduction pathways that do not use TCF/β-catenin for their transcriptional outputs 16 . These are associated with planar or apical-basal polarity affecting pathways or calcium signaling. In vertebrates, these are driven by non-canonical Wnts-4, -5a, -5b, -6, 7a, -7b, and -11. In Drosophila, the main non-canonical or Planar Cell Polarity  15,23,[25][26][27][28][29][30] . In vertebrates, the four TCF gene family members function as both transcriptional repressors and activators 31 . Drosophila has only one gene that encodes TCF, which must perform both functions, making this system simpler to manipulate genetically. This feature that was recently used to study activation and derepression of Wnt targets in Drosophila tissue culture cells lacking TCF 32 . In studies of TCF in fly embryos, the difference between the two functions of TCF becomes apparent when loss of function mutants are compared to dominant negative TCF transgenes 33 . Loss of function embryos show a loss of patterning, but the embryos remain large 33 . In contrast, expression of dominant negative TCF leads to a small embryo that lacks patterning 5 . This finding led us to propose that the two roles of TCF might be separable at the transcriptional level, and led us to develop tools to analyse transcription in vivo.
We focus on developing methods for assessing transcriptional programs downstream of Wnt signalling, and identifying the processes and mechanisms involved. To this aim, we developed a naïve embryo system in which we can activate or repress different forms of Wnt signaling at various levels in the different pathways. This system allowed us to dissect the effects of Wnt signalling at both the phenotypic level of the whole organism and at the molecular level.

Results:
Development of a naïve-embryo transcriptional assay system: We developed a transcriptional assay using the Drosophila embryo, where by using simple genetic manipulations we can create relatively naïve, homogeneous populations of cells, therefore minimizing the confounding effect of non-specific, secondary, and multiple signaling pathway effects that are often observed in gene expression studies 34 . In normal Drosophila development, eggs are provided with maternal patterning signals. These signals include anterior-posterior patterning molecules such as Bicoid and Nanos, terminal patterning determinants such as Torso and Torsolike (EGF pathway related), and dorsal-ventral signals such as Toll and Dpp (NFκB and TGFβ signaling pathways) [35][36][37] .
These patterning signals determine the axes of the developing embryo and activate further signals that lead to specific cellular identities for each cell in the embryo. Removal of these basic patterning signals leads to eggs that develop a simple, un-patterned epithelium "naïve embryos". For anteriorposterior patterning, we used a triple mutant (bicoid, nanos, torsolike) that eliminates anterior, posterior and terminal patterning respectively leading to highly compromised development (Supplementary Movies 1 and 2) 38,39 . A further advantage of this system lies in the fact that these are maternal effect mutations, allowing the use of homozygous females that lay 100% mutant eggs, therefore removing the difficulty of identifying mutant embryos and avoiding the use of complicated germline clone techniques 33,40,41 . This experimental setup creates a condition where all the embryonic cells are identical with respect to Wnt signalling.

Basic phenotypic analysis of Wnt signaling:
We introduced several genetic changes targeting specific components of the Wnt signalling pathway and assessed their consequences via a simple phenotypic assay (Fig. 1). This was accomplished by examining both the size of the embryos (normal/small) and their differentiation status (naked or with denticles) (Fig. 1C). Wg overexpression (Wg++) was accomplished by using the GAL4/UAS system to establish the hyper-activated pathway condition 42 . The opposite condition was the expression of a dominant-negative allele (tcf∆N-short). This form of TCF lacks the Arm binding region, the Groucho binding sequence (GBS) and the C-clamp that are required for β -catenin binding, Gro repressor binding, and additional target specificity respectively (Supplemental Figure 1)  Expression profiling reveals distinct transcriptional programs reflecting observed phenotypes: We examined changes in gene expression in all of the conditions using microarrays (see Methods).
Even in our developmentally restricted system we identified 1,360 genes whose expression was significantly altered upon perturbing either canonical or non-canonical Wnt signalling (False Discovery Rate (FDR) < 1%). Hierarchical clustering of these genes revealed distinct expression patterns associated with perturbations affecting these separate branches of the Wnt signalling pathway ( Fig.   2A, S2A). Expression of DisArmed showed a mild phenotypic change compared to WT, and this was mirrored in the similarity of their expression profiles (Fig. 1A). Overall, the transcriptional changes segregated according to whether the genetic perturbation was affecting the canonical (arm∆N, tcf∆Nlong, tcf∆N-short or Wg++), or non-canonical signalling (Wnt4++, Otk++). Full results of differential expression analyses and associated enrichments for biological processes and pathways are reported in Suppl. Table 1 and 2 respectively.
Within the canonical pathway, Arm primarily signals through TCF 19 , as such the expression profiles of tcf∆N-short-arm∆N and tcf∆N-short are highly similar ( Fig. 2A, S2A) and show highly correlated changes in gene expression compared to WT (Fig. 2B), illustrating that these are epistatic. Wnt4 has been reported as primarily signalling via PTK7/Otk 19 . In keeping with this, we found that overexpression of Wnt4 and Otk had highly correlated transcriptional responses compared to WT ( Fig. 2A, 2C), providing further evidence of their functional interaction, epistasis and location within the same branch of the Wnt signalling pathway.
Across all conditions we identified four major clusters of genes (I-IV, Fig. 2A, S2B-C), each of which was associated with distinct enrichments for biological processes, pathways and transcription factor binding sites (TFBSs) (Fig. 3). Cluster I showed reduced expression upon stimulation of canonical Wnt signalling via Wg++ or arm∆N and increased expression upon overexpression of the noncanonical branch (Fig. S2D). This cluster was enriched for processes associated with development, differentiation, cell-cell communication and morphogenesis ( Figure 3A), and contained several transcription factors (TFs). The enrichment for Trl and TCF motifs suggests that cluster I may contain direct targets of canonical Wnt-mediated repression ( Figure 3E). Genes present in cluster II are downregulated by overexpression of Wnt4 or Otk ( Fig. 2A, S2D) and hence represent a set of genes that are putatively repressed by non-canonical Wnt signalling, but that also show upregulation upon the loss of the ability of Wnt to derepress genes repressed by TCF. Intriguingly, this cluster was enriched for glutathione metabolism genes (Fig. 3B . Wg++ and Wnt4++ correspond to non-endogenous overexpression capable of driving these distinct antagonistic components of the Wnt pathway. Comparing expression of Wg++ against Wnt4++, we identified 1,798 genes as differentially expressed (Fig. 4A, absolute fold change > 1.5, FDR < 10%), with up-and down-regulated genes appearing to be associated with distinct biological processes and functions. Genes upregulated in Wg++ compared to Wnt4++ were associated with processes associated with cellular growth or the cell cycle, whilst those showing decreased expression were linked to cell adhesion, polarity and morphogenesis (Fig. 4B).
Investigation of the promoter sequences of genes upregulated by Wnt4 compared to Wg revealed enrichment for binding sites of important developmental genes (i.e., Ubx and cad) (Fig. 4C). These results support the antagonism of canonical and non-canonical Wnt signalling and show Wg and Wnt4 as regulating vastly different downstream transcriptional programs.
Despite the differences in the size of tcf∆N-short and tcf∆N-long embryos ( Fig. 1), their expression profiles appeared to be highly similar ( Fig. 2A). Surprisingly, only 222 genes were identified as differentially expressed between these conditions (Fig. 4D, absolute fold change > 1.5, FDR < 10%).
Genes upregulated in tcf∆N-short were enriched glutathione transferase activity genes (i.e., GstD4, GstD3, GstD9 and GstD6) and included genes known to be upregulated in response to severe stress (i.e., TotA, TotC and TotX) 52 , whereas the set of genes downregulated was enriched for proteolysis and chitin metabolism functions (Fig. 4E). These results suggest an upregulation of stress response and GSH depletion/redox state as a potential mechanism responsible for the differences in the sizes of tcf∆N-short and tcf∆N-long embryos.

TCF occupancy confirmed at potential targets by HA-ChIP-qPCR:
For those genes whose expression changed in response to perturbations in canonical Wnt-signalling, we investigated publicly available TCF ChIP-seq data from Drosophila embryos 53 . As canonical Wntsignalling primarily signals via TCF/pan, we expected to see an enrichment of genes bound by TCF in several sets of differentially expressed genes (i.e., Wg++, tcf∆N-short or tcf∆N-long). However, we observed that only those genes upregulated by Wg++ were enriched for TCF binding. Both technical (e.g., antibody specificity) and biological factors (e.g. differences between whole embryos and our naïve system) could potentially explain this lack of enrichment for TCF binding.
The simple embryonic system we have developed allows HA-tagged isoforms of factors of interest to be introduced into the system, making it possible to perform ChIP experiments against chromatin binding factors that either lack or only have low quality antibodies. We randomly selected 23 genes whose expression profile mirrored the size changes observed for tcf∆N-short, tcf∆N-long and Wg++ ( Fig.5 A,B) (cluster IV). We performed ChIP-qPCR on their promoters from embryos expressing HAtagged tcf∆N-long or HA-tagged tcf∆N-short constructs (Fig. 5C), to investigate whether these genes are direct or indirect target of Wnt signalling in this system. In addition, we investigated whether H3K27me3, a histone mark associated with repressed and bivalent genes 54 , was present at this set 1 0 that these genes are direct targets of Wnt/β-catenin repression in our system. Peritrophin-15A was found to lack tcf∆N-short-HA at its promoter but showed high H3K27me3 signal, suggesting that this gene is an indirect target of Wnt/β-catenin repression, whose expression is regulated, at least in part, via repressive histone modifications.

Blocking Apoptosis restores cells to epidermis:
The differences in size between the tcf∆N-long and tcf∆N-short can be explained at least in part by two processes, either the embryo is growing less or cells are undergoing more apoptosis. Our transcriptional analysis identified an upregulation of stress response in tcf∆N-short embryos (Fig. 4C, D), which could implicate either of the processes. To evaluate how expression of tcf∆N-long or tcf∆Nshort influences cell division and apoptosis in our system, we performed immunostaining using antiphospho-H3 antibody and anti-cleaved caspase 3, respectively. Immunostaining was performed on the embryos collected from flies overexpressing Wg and used as control in the experiment. As apoptosis begins at stage 11-12 during Drosophila embryogenesis 55 , we chose stage 14 embryos for immunostaining. We could not detect much apoptosis at these stages, but we did observe a large number of cell divisions occurring in wild type and Wg++ embryos ( Fig. 6 C, D, G, H). There was a small increase in apoptosis in tcf∆N-short as quantified in Fig. 6I.
These differences might indicate increased apoptosis as a potential mechanism to explain the smaller embryos in tcf∆N-short. Therefore, we tested whether blocking apoptosis could rescue the embryo size phenotype. To block apoptosis in small embryos, we used a deletion (Df(3L)H99) which removes three apoptosis genes grim, reaper, and hid 56 . We combined H99 with a wg mutation and observed a restoration of cells between the segment boundaries ( Fig. 6K, note the small denticles, bracket) as compared to wg mutant alone (Fig. 6J) and WT (Fig. 6L). This was similar to previous finding using arm mutants along with H99 57 . Although some of the embryo length is restored in wg; H99 double mutant embryos, the embryos do not resemble the large embryos of tcf∆N-long ( Fig. 1), suggesting that apoptosis alone cannot completely explain the effect of tcf∆N-short.

Discussion
We show that our naïve embryo system is amenable to quantitative analysis of both whole-body phenotype and associated transcriptional responses to perturbations targeting specific components and branches of the Wnt signalling pathway. The Wnt signalling pathway is of particular interest for this type of analysis as it consists of a canonical pathway with a well-defined mechanism of signal transduction, and a series of cell polarity pathways regulating a variety of cellular behaviours 13,58 .
These pathways can be thought of as a signalling network 59,60 , where upon signal activation a poorly defined mechanism selects the pathway and outcome. Our system allowed the observation of specific transcriptional profiles that were clustered depending on which branch and which component was perturbed, illustrating clear differences in the sets of regulated genes and the involved biological processes.
For the canonical pathway, much but not all of the cellular response is mediated through β -catenin and TCF transcriptional activation and repression 48 . For a strong activation of canonical Wnt signaling, we used overexpression of Wg, which resulted in full embryo growth (aside from a head involution defect 55 ). Phenotypically embryos generated by overexpression of Wg appear the same as those with a gain of function Arm allele. However, we found a marked difference between the two conditions with a large number of genes activated by Wg++ but not by arm∆N (cluster III, Fig. 2A).
These genes were associated with cell proliferation and the cell cycle. The results from our transcriptional analysis therefore suggest that Arm independent transcriptional activation occurs downstream of Wg, perhaps through the proposed Wnt/Stop mechanism 47,48,53 , a pathway not previously documented in Drosophila.
For the strong loss of signaling condition, we performed two experiments using tcf∆N-short alone and tcf∆N-short along with arm∆N. We observed that the transcriptional profile was very similar illustrating that all Arm dependent transcription requires a form of TCF that can interact with Arm ( Fig. 2B). This re-establishes the Arm/TCF interaction as the main source of transcriptional activity due to Arm transactivation 5 . For the intermediate loss of signaling condition, we expressed a tcf∆N-long construct, a condition where we observe a loss of patterning, with most epidermal cells producing denticles but 1 2 without a strong effect on embryo size. The identical phenotype was produced by the dominant negative DisArmed allele 23 . As this is a highly-expressed form of the Arm protein that is immune to standard ubiquitin-mediated degradation and fails to act in transcriptional activation, the most likely explanation is that DisArmed binds to TCF, either sequestering it or preventing TCF from taking part in transcription. Either way, it perfectly phenocopies the absence of TCF (Fig. 1), and shows a very similar transcriptional profile to tcf∆N-long especially in clusters III and IV ( Fig. 2A).
The fourth condition was the use of a non-canonical Wnt4 molecule that signals through a different receptor (PTK7/Otk) opposing Wg. We found that in all three conditions Otk expression, Wnt4 expression, and Wnt4/Otk expression a similar cohort of genes was regulated ( Fig. 2A), and that the highly-correlated expression profiles of Otk and Wnt4 compared to WT support that they reside within the same section of the Wnt signalling pathway ( Recently, Wang and colleagues found that redox state in germ line stem cells was regulated by Wnt signaling 72 . Different cellular states (i.e. proliferation, apoptosis, differentiation) have been associated with different redox states 73,74 . Our transcriptional analysis indicated that glutathione metabolism, a metabolic pathway associated with regulation of redox potential in the cell, was regulated by Wnt signaling in our system. tcf∆N-short embryos showed an increase in the expression of genes associated with both glutathione metabolism and response to stress, whereas such a similarly strong change in expression was not observed in tcf∆N-long and Wg++. This finding suggests that embryonic Wnt signaling is required to modulate redox metabolism and its dysregulation in tcf∆Nshort might result in increased stress and apoptosis 75 . However, inhibition of apoptosis did not completely rescue the size differences between tcf∆N-long and tcf∆N-short suggesting the involvement of additional pathways.
Our gene expression and ChIP data do not support a simple explanation for which genes are activated and which are repressed. Previous studies attempting transcriptional profiling of genes downstream of Wnt have found a wide range of results and thousands of genes 14,34,76 . For example, an early study looking at developmentally important transcription factors in Drosophila embryonic development found more than 1,000 sites where TCF was bound by ChIP-Chip 34 . Since these genes are not expressed in the same way in different cells, it is likely that a complex combinatorial system with multiple transcription factors or epigenetic regulation is in place.
Overall, we present a useful in vivo Drosophila system that allowed us to characterize and bring together several aspects of Wnt signaling. We have looked at transcriptional repression and activation, moderate and strong canonical signaling conditions, and at the effects of opposing Wnt ligands. Showing the utility of our experimental system, our transcriptional analysis led us to identify a novel, Drosophila β -catenin independent set of genes activated by overexpression of Wg and completely different gene cohorts downstream of Wnt4 and Wg. We envision that detailed cellular and molecular studies in this naïve embryo system will allow to identify and test specific transcription factors and binding sites, and to delineate different signaling outcomes from different perturbations of Wnt and other signaling pathways.

Supplemental Data.
Table S1: Excel sheet with results of the differential expression analysis.  .

Materials and Methods
Fly strains and transgenics: Microarray: We collected approximately 50 embryos per microarray experiment, with the control embryos being non-expressing naïve embryos. Extracted mRNA from the various genetic conditions was then analysed on Affymetrix Drosophila 2 microarrays by standard procedures. For each condition, we performed two biological replicates. Microarrays were normalised using GC-RMA. Prior to differential expression analysis, probesets were filtered by 1) removing probesets not mapping to a gene 2) removing probesets which mapped to multiple genes 3) if a gene had multiple probesets assigned to it the probeset with the largest IQR was used 4) probesets not showing expression greater than 2.5 in at least two samples and those mapping to non-canonical chromosomes were removed. Following these preprocessing steps, differential expression analysis was performed using LIMMA 48 . Clustering of gene expression profiles was performed by converting gene expression to zscores and clustering them using (1 -cor) / 2 as a dissimilarity measure. To assess the stability of sample level clustering we used pvclust with 10000 iterations to calculate approximately unbiased (AU) p-values ( Figure S2A). All of the major expected clusterings remained stable. The set of samples relating to perturbation of the non-canonical pathway (Wnt4++, Otk++, Wnt4Otk++) did not form a stable cluster but were highly unstable between each other. The optimum number of clusters to cut the gene-associated dendrogram was determined by calculating the mean silhouette width over a number of different cluster sizes (Fig. S2B,C).
Enrichment for Gene ontology was performed using a hypergeometric test from the GOStats package 84 . Results from GO enrichments were simplified for presentation purposes by filtering terms with a high semantic similarity 85 , all significant results (adjusted p-value < 0.05) from the enrichment analyses are available in Supplemental Table S2. Enrichment for pathways was performed using ReactomePA 86 , all significant results (adjusted p-value < 0.05) are available in Supplemental Table   S2. For pairwise comparisons, enrichments performed using the set of genes used in the differential expression analysis as the background, whereas for the enrichments based on the clustering ( Fig.   2A) the background was the set of genes identified as differentially expressed over all conditions. Enrichment for TFBS motifs was performed using AME (from the MEME suite 87 ) against the JASPAR magnification with Zeiss Axiocam. Quantification of fluorescence was done using ImageJ software tools 89 . FFT band pass filter was applied to the images for correction of any uneven illumination and horizontal scan lines acquired by phase contrast microscope followed by conversion to 40 pixels. For fluorescence quantification in the cells, small structure default pixels were optimized to 3 pixels and tolerance threshold was set at 5% using binary process function. Intensity density was obtained by using the particle analyser tool. Standard error was calculated using data from n=3 for each condition and error bars were plotted.

Antibodies:
The following antibodies were used in the study: polyclonal Anti-phosphorylated histone

Data availability
All microarray data from this study is available from GEO under accession number GSE97873.   z-score