Mouse IDGenes: a reference database for genetic interactions in the developing mouse brain

Detailed search

Navigation on the Mouse IDGenes Web page by the ‘Detailed Search’ tab allows users to search for gene expression and interaction data in specific regions of the embryonic mouse CNS at mouse embryonic developmental stages E8.5, E10.5 and E12.5 (radio button: ‘display genes according to the chosen region’). CNS regions and developmental stages (according to Supplementary Table S1) can be selected with the help of combo boxes. Having set a specific developmental stage and an anatomical structure, users are further able to analyze whether a specific gene of interest is expressed during that specific developmental stage in the selected anatomical structure.

Search for interactions

On the ‘Search for Interactions’ tab, the Mouse IDGenes Web page allows users to search for specific interactions between two genes of interest and for all interactions in which one specific gene of interest is involved (by the ‘Search for specific interactions’ button). Information about the region and at which stage the specific interaction takes place in the embryonic mouse CNS can be retrieved.

Users can also search whether two genes of interest display an overlapping gene expression (by the ‘Search for overlapping gene expressions’ button).

By selecting ‘all’ in either both or only one of the gene selection boxes, users can search for all stored interactions either of the whole database or in which a specific gene of interest is involved.

The displayed interactions follow an overall scheme consisting of the attributes ‘effect’, ‘type’ and ‘name’ for a genetic interaction. The attribute ‘effect’ can be either activation, i.e. turning on gene expression, or repression, which is defined as shutting down gene expression. The attribute ‘type’ of an interaction is defined by the following six values:

• – direct: We define a direct interaction in case interaction partner 1 binds directly to the promoter of interaction partner 2.

• – direct signaling: In case of a ligand that activates a signaling cascade or any other component of a signaling cascade that does not directly interact with (or binds to) the promoter of a target gene of this signaling pathway, a direct signaling interaction refers to the activation/repression of a direct target gene of this signaling pathway.

• – indirect: Interaction partner 1 does not bind directly to the promoter of interaction partner 2, and signaling or genetic interaction cascades have to occur between the two interaction partners.

• – indirect signaling: In case of a ligand that activates a signaling cascade or any other component of a signaling cascade that does not directly interact with (or binds to) the promoter of a target gene of this signaling pathway, an indirect signaling interaction refers to the activation/repression of an indirect target gene by a direct target gene of this signaling pathway.

• – maintenance: Interaction partner 1 is not required to activate (i.e. turn on) or to repress (i.e. turn off) the promoter (or expression) of interaction partner 2, but to keep this promoter (or expression) activated (‘on’) or repressed (‘off’) over (a longer period of) time.

• – IEXP: Inferred from experiment (e.g. loss of function/gain of function) and/or expression pattern

Currently, the database contains these general interaction schemes: direct activation, direct signaling activation, IEXP activation, maintenance activation, indirect activation, indirect signaling activation, direct repression, direct signaling repression, maintenance repression, IEXP repression, indirect repression and indirect signaling repression. The attribute ‘name’ of an interaction, which is displayed on the Mouse IDGenes Web page, is composed of the official gene symbol according to MGI of interaction partner 1, followed by an arrow symbol from the third column of Table 1 and finally the gene symbol of interaction partner 2.

As pointed out before, the maintenance activation occurs over a longer period to cause a downstream effect. Time-wise, such an interaction can occur over several developmental stages as, for example, is the case of the development of midbrain dopaminergic (mDA) neurons. There, initially a Wnt1-regulated network together with a Shh-controlled genetic cascade establishes the mDA progenitor domain, by maintaining Otx2 expression in the ventral midbrain (35). Another example is Lmx1b, which is known to be necessary for the initiation of Fgf8 expression and for the maintenance of several other genes including Engrailed 1 (En1), En2 and Wnt1 (3). Lmx1b, therefore, falls into the interaction scheme ‘maintenance activation’ for En1, En2 and Wnt1. On our Web page, these interactions are depicted in the following way: Lmx1b → En1, Lmx1b → En2 and Lmx1b →Wnt1, meaning that Lmx1b keeps the expression of En1, En2 and Wnt1, respectively, on overtime.

Furthermore, cooperative interactions, i.e. interactions with more than two interaction partners, can be stored in the database, for example, if two transcription factors or one transcription factor and another cofactor will bind on the promoter of a target gene, i.e. interaction partner 2.

Data input

One of the core functions of Mouse IDGenes is the possibility for data input by the user. This feature allows the permanent update of data by the respective experts in the field of developmental neurosciences. To input new gene expression and interaction data into the database, the user is asked to maintain the overall CNS regionalization scheme (according to Supplementary Table S1), by choosing and subsequently storing specifications given by combo boxes, which follow our CNS model (Figure 2A). The database so far has stored data exclusively about the developmental stages E8.5 to E12.5, as these are the crucial stages in the establishment of the MHB and development of the mid-/hindbrain region, which is the authors’ main research interest. To maintain the high degree of curation in the current database, it is necessary to also introduce at least one relevant publication as well as the corresponding PubMed ID.

Users can input gene expression data as well as interaction data into the database by selecting the respective type of data on the Web page. All fields that are mandatory for data input as well as storage in the database are labeled by an asterisk (Figure 2A and B), which ensures the completeness of expression and interaction data sets. In addition, it is automatically controlled from the PubMed abstract by using the link to PubMed whether the given year of the publication and the entered PubMed ID are consistent (compare with the fields ‘Author and Year’ and ‘PubmedID’ in Figure 2A and B); otherwise, a warning message is displayed, and storage of the data is prevented. Further, gene symbols from MGI are auto-completed after typing some letters in the field ‘Gene’ (Figure 2A) or ‘First Factor’ or ‘Target Gene’ (Figure 2B), and they are internally stored via MGI identifiers. The use of predefined lists by combo boxes for the selection of e.g. a specific brain region and/or the interaction type helps to comply with our brain regionalization model as well as the model for genetic interactions and therefore ensures data consistency. In case of extending the database for an interaction, it is possible to input also cofactors and more targets of an interaction (Figure 2B). After completion of the data set, the user is asked to review the input before final submission to the database. Constraints in the PostgreSQL database prevent data from being duplicated when stored.

To control for incorrect input, the data will be regularly curated by the authors. Before curation, new data will be distinguishable on the Web page from already curated data by labeling the not yet validated gene expression or interaction data on the Web page (Figure 2C).

Confirmation of new data is performed by

• – reading the given publications,

• – comparing the indicated gene expression and/or genetic interaction,

• – comparing the developmental stage as well as the brain regions with the data entered into the Mouse IDGenes database.

A mailing list has been set up for users to discuss their plan to submit new data.

After verification of new data, the flag ‘Data not validated yet!’ is removed from the database and the Web page; otherwise, if the data could not be verified, they are deleted from the database.

Output

The output of a specific search either in one of the gene expression or the interaction menu items appears on the same page underneath the user selection (Figure 3B), organized in a table. In case of the gene expression data (radio button: ‘display genes according to the chosen region’ (Figure 3A) in the ‘Detailed Search’ menu item), the output table indicates the developmental stage the user is looking at. The output further shows the specific subdivision of the chosen anatomical area and which genes are expressed there in the column ‘Region’. Additionally, links to the public databases NCBI, MGI, UCSC and Ensembl are given in the column ‘Expression Data’, where further general information about a displayed gene can be retrieved. Most importantly, links to literature references are indicated as evidence of gene expression information. The last column ‘Interactions’ of the output shows the interaction data, which can be retrieved for the actual brain region at the actual developmental stage with the corresponding literature references.

Prediction of a new Wnt1 target based on Mouse IDGenes

To demonstrate the usefulness of the Mouse IDGenes database for other research applications, we chose an example from our own scientific interests focused on the role of the Wnt signaling pathway in the development of the mid-/hindbrain region and of neuronal populations located in this region, such as the ventral midbrain dopaminergic neurons. In this context, the Wnt signaling pathway plays a crucial role because it participates in the regulation of regional patterning, cell cycle, cell fate specification, cell differentiation and cell survival. It is also involved in various human diseases (36).

The manually curated, and thus, highly reliable data set of Mouse IDGenes provides an ideal basis to further analyze, for example, the complex gene regulatory network at the MHB and in the ventral midbrain in which Wnt1/β-catenin signaling has so far been implicated (37, 38). We used the Mouse IDGenes database to predict novel direct or indirect targets of the Wnt1/β-catenin signaling pathway that might be involved in the development of the mid-/hindbrain region and of associated neuronal populations and validated our results experimentally.

We obtained known targets with the interaction type ‘direct signaling’ and ‘indirect signaling’ of the Wnt1/β-catenin pathway from the Mouse IDGenes Web page by choosing Wnt1 as search term in the ‘Search for Interactions’ menu as well as from literature searches (Figure 4A) (35, 39–52). One hallmark of the Wnt1/β-catenin signaling pathway is the stabilization and nuclear translocation of cytoplasmic β-catenin (Figure 5). In the absence of a Wnt1 signal, the Lef1/Tcf transcription factors are bound to the promoter regions of the direct Wnt1 target genes together with other co-repressors, thereby inhibiting the activation of these genes. In the presence of a Wnt1 signal, the replacement of these co-repressors and binding of β-catenin to the Lef1/Tcf transcription factors activates the transcription of the direct Wnt1 target genes. The frequent presence of, in particular, evolutionary conserved Lef1/Tcf TFBSs in the promoter region of a gene is therefore indicative that this gene might be a direct target gene of the Wnt1/β-catenin signaling pathway. An indirect target gene of the Wnt1/β-catenin pathway was defined as a gene that is upregulated on Wnt1/β-catenin signaling activity but is not directly bound by β-catenin and Lef1/Tcf transcription factors and thus requires another mediator, i.e. an intermediate gene regulatory step.

We performed an in silico promoter analysis for Wnt1 target genes of the interaction type ‘direct signaling’ and ‘indirect signaling’ in the training data set (Figure 4A) as well as in the test set (Figure 4B) consisting of interesting target genes for which the interaction type is not yet known in the CNS (53–59). The MatInspector (Genomatix) program was used to identify Lef1/Tcf binding sites in the promoter sequences of these target genes with help of predefined position weight matrices (PWMs) (60). For classification of the interaction type, we assigned ‘direct signaling’ target genes of Wnt1/β-catenin pathway to class 1 and ‘indirect signaling’ targets to class 2. We computed two parameters for each gene: (i) the number of evolutionary conserved Lef1/Tcf TFBSs (V$LEFF). Orthologous promoter sequences from human, chimp, mouse, rat, dog, horse, cow and opossum were taken into account, and only binding sites present in at least two species were considered. (ii) The average matrix similarity of all Lef1/Tcf binding sites in the promoter, a score indicating the similarity of a predicted binding site with the consensus matrix of the TF (60, 61). It lies in [0, 1] and reaches 1 only if the predicted sequence corresponds to the most conserved nucleotide at each position. While the matrix similarity allows the assessment of the structural quality of a binding site, the number of conserved binding sites supports the possibility that these binding sites might be functionally relevant (62). It is assumed that in case of direct interactions the matrix similarity is higher and conserved binding sites are more frequent than in case of indirect interactions.

We trained a support vector machine (SVM) with ‘direct signaling’ targets (class 1) and ‘indirect signaling’ targets (class 2) using both parameters (Figure 4C). Classification via SVM has been successfully used not only for feature selection of microarray data (63–65) but also to integrate expression as well as genomic data, e.g. evolutionary conservation or binding site clusters for the improvement of TFBS prediction (21, 22). The statistical model was calculated with help of the ksvm function from the kernlab package in R statistical software by using a polynomial kernel matrix similarity (degree = 2), a cost parameter C = 1 and a 13-fold cross validation (leave-ne-out cross validation). Class 1 consists of Otx2, Lmx1a, En1, Ccnd1, Dkk1 and Sp5 and class 2 consists of Msx1, Pitx3, Gbx2, Shh, Pou4f1 (Brn3a) and Six3 (Figure 4A). Additionally, Fgf20 was included in class 1. Interaction data for this gene cannot be retrieved from Mouse IDGenes, as expression of Fgf20 in the mouse neural tube starts only after E12.5 (66). We assigned En1 to class 1 because it was shown that En1 expression under the control of the Wnt1 enhancer in mice rescues the Wnt1-/- mid-/hindbrain phenotype (41, 42), thus indicating that En1 is the downstream target of Wnt1 signaling in mid-/hindbrain development, and because a direct interaction of the promoter of the homologous gene engrailed-2 (En2) with the Lef1/Tcf transcription factor was observed in the frog (43). For our statistical model, we obtained a training error of 7.69% and a cross-validation error of 46.15%.

To further elucidate the Wnt1-controlled gene regulatory network at the MHB, we applied SVM prediction by using the number of conserved Lef1/Tcf binding sites and the average matrix similarity in the promoters of four interesting genes, Fgf8, Lef1, Islet1 and Dkk3, representing the test set (from Figure 4B) to predict whether they are directly or indirectly activated by the Wnt1/β-catenin pathway. For these four genes, it is not known whether they are direct or indirect targets of Wnt1 in neural tissues, but it was observed that Lef1 and Dkk3 are co-expressed with Wnt1 in the midbrain (Götz, S. et al., unpublished data), whereas Fgf8 and Islet1 are not co-expressed with Wnt1 but depend on Wnt1 expression in the mid-/hindbrain region (53, 54, 58, 59). Using the SVM on the test set (Figure 4B), our analysis predicts that Lef1 and Dkk3 are direct targets of Lef1/Tcf-mediated Wnt1/β-catenin signaling, whereas Fgf8 and Islet1 are not direct target genes of this signaling pathway and interact indirectly with the Wnt1 signaling cascade. Fgf8 and Islet1 therefore would represent Wnt1/β-catenin target genes that are most likely activated by other genes that in turn are activated by Wnt1/β-catenin signaling. Direct binding of Lef1/Tcf transcription factors to the Lef1 promoter was shown in colon cancer/lymphocytes (55) as well as in HEK 293 cells (56), which is in accordance with our prediction. A direct interaction of Lef1/Tcf transcription factors with a larger promoter region of the Islet1 gene was shown in embryonic heart tissue but not in neural tissues (57). However, direct activation of Fgf8 and Islet1 in the CNS by Lef1/Tcf binding sites was up to now never observed although it was inferred from mutant mouse embryo analyses (53, 54, 58, 59).

Experimental validation of the predicted Wnt1 target gene Dkk3

To show the predictive power of our approach, we experimentally validated the so far unknown Wnt1 target gene Dkk3 as a direct target gene of the Lef1/Tcf-mediated Wnt1/β-catenin signaling cascade in vitro.

To identify conserved Lef1/Tcf binding sites in ∼700 bp extended promoter region (as used for the SVM prediction) of the Dkk3 gene of five different mammalian species (mouse, rat, cow, pig and opossum), we applied the DiAlign TF program in the Genomatix software suite GEMS Launcher to evaluate the overall promoter similarity and to identify conserved Lef1/Tcf binding sites in these regions. For the alignment, we chose the five most conserved Dkk3 promoter sequences among 14 organisms from Genomatix homology group Hg3927. Four Lef1/Tcf binding sites were predicted in the putative mouse Dkk3 promoter, of which one [binding site ‘c’, the most proximal Lef1/Tcf binding site to the transcription start site (TSS)] was highly conserved among all five species (Figure 6A). To determine whether these predicted Lef1/Tcf binding sites control the Wnt1/β-catenin and Lef1/Tcf-mediated activation of the murine Dkk3 promoter, we cloned a 744-bp-long fragment of this promoter containing three of the four predicted Lef1/Tcf binding sites into a promoter-less luciferase reporter vector (Figure 6E). Co-transfection of increasing amounts of rat Lef1 complementary DNA (67) or a constitutively active β-catenin [ΔN-β-catenin, which mimics the activation of Wnt1 signaling, (68)] into ‘Wnt-responsive’ HEK293T cells [exhibiting a basal level of Wnt1/β-catenin signaling activity, Prakash, N. et al., unpublished data, (69)] led to a dose-dependent activation of luciferase expression mediated by this mouse Dkk3 (mDkk3) promoter fragment (Figure 6F and G), indicating that the promoter of the mDkk3 gene is a direct target of Lef1-mediated Wnt1/β-catenin signaling in this in vitro context. Additional in vivo evidence indicates that the murine Dkk3 gene is also a direct target of Lef1-mediated Wnt1/β-catenin signaling in the mouse ventral midbrain (Zhang, J. and Prakash, N., unpublished data).

To evaluate whether the predicted Lef1/Tcf binding sites in the murine Dkk3 promoter are functional, we mutagenized each of these binding sites either individually (mutation of a single Lef1/Tcf binding site) or altogether (mutation of all three Lef1/Tcf binding sites) within the mDkk3 promoter fragment such that they cannot be recognized by Lef1/Tcf transcription factors anymore (Figure 6B–E) (46). Site-directed mutagenesis of single or all three Lef1/Tcf binding sites in the mDkk3 promoter/reporter constructs revealed that

• – luciferace activity was significantly decreased relative to the wild-type mDkk3 promoter after co-transfection of Lef1 cDNA (Figure 6H),

• – the activation of the mDkk3 promoter/reporter construct carrying a mutagenized Lef1/Tcf binding site ‘c’ by Lef1 was completely abolished relative to the pcDNA3.1 control, in contrast to a still significant activation of the mDkk3 promoter/reporter constructs carrying a mutagenized Lef1/Tcf binding site ‘a’ or Lef1/Tcf binding site ‘b’ (Figure 6I).

This result strongly suggests that the most conserved (by position and sequence similarity among five mammalian species) and proximal (relative to the TSS) Lef1/Tcf binding site ‘c’ is the functionally most important of the three Lef1/Tcf binding sites for Lef1-mediated activation of the murine Dkk3 gene by Wnt1/β-catenin signaling.

Lef1/Tcf transcription factors were predicted to directly activate the mouse Dkk3 gene in the context of Wnt1/β-catenin signaling by our SVM analyses. Therefore, the experimental validation of Lef1/Tcf binding site ‘a’ and Lef1/Tcf binding site ‘c’ with a matrix similarity of 0.867 and 0.961, respectively, confirm the result of our SVM classification, indicating that direct Lef1/Tcf targets have in general more conserved binding sites than indirect targets (Figure 4B). Altogether, our experimental results indicated that the activation of the mouse Dkk3 gene is mediated at least in part by the predicted Lef1/Tcf binding sites in its promoter region and therefore highlight the importance and predictive power of a database combining both expression and interaction data.

Bioinformatics prediction of Lef1/Tcf binding sites in the promoter regions of Wnt/β-catenin target genes

To discriminate interactions of the type ‘direct signaling’, i.e. genes bound by Lef1/Tcf transcription factors activated by Wnt1 signaling from correlations of the type ‘indirect signaling’ and to predict the interaction type of new target genes, a support vector machine (SVM) was applied. The data matrix of the SVM is composed of two variables (columns) for each object (row), i.e. the frequency as well as the averaged matrix similarity of conserved TFBSs for each promoter sequence of a specific gene. The classification of the interaction type in training data set for each object (gene), meaning whether Lef1/Tcf transcription factor binds directly or indirectly to the promoter sequence, was derived from the Mouse IDGenes database. Promoter sequences for Wnt/β-catenin target genes were derived from the ElDorado genome database (Genomatix/Germany), versions 12-2010 and 08-2011. Orthologous promoter sequences from different mammalian species of the Genomatix homology group (human, chimp, mouse, rat, dog, cow, pig and opossum) were analyzed using the MatInspector program (with the Matrix Family Library Version 8.4) from Genomatix to identify potential Lef1/Tcf binding sites and to extract matrix similarities. Conserved binding sites were determined by using the DiAlign TF program (Genomatix) and by searching for common Lef1/Tcf sites occurring at the same position in (aligned) orthologous promoter sequences of each Wnt/β-catenin target gene. The length of the promoter regions used for the detection of Lef1/Tcf sites to derive the total number and the average matrix similarity of binding sites both included in the SVM algorithm were generally in the range of 600–1400 bp. In addition, a longer promoter region of mouse Dkk3 mRNA (with the Genbank identifier AK013054, firstly detected in whole body of 10- and 11-day-old mouse embryos) was defined as 1800 bp upstream, including the proximal region, and 200 bp downstream of the TSS. Dkk3 promoter sequences of 2000 bp length from five different mammalian species (mouse, rat, cow, pig and opossum) were analyzed with the MatInspector to predict Lef1/Tcf binding sites.

Cloning of a mouse Dkk3 (mDkk3) promoter/reporter vector

A 744-bp fragment of the putative mDkk3 promoter (Entrez Gene ID: 50781; chromosome 7, strand: −, position: 112 158 266 to 112 159 009 bp, NCBI build 38) was amplified from C57BL/6 mouse genomic DNA by PCR using the forward primer 5′-ctcgag TGACCA G A T C C A G C T TGCA-3′ and reverse primer 5′-aagctt CCTC CTGA GG GTAG TTGAGA-3′ that included an XhoI and HindIII restriction site (underlined sequences in italics), respectively. The amplified fragment was cloned into the pCR®II TOPO TA vector (TOPO® TA Cloning® Kit, Life Technologies/Germany) and sequenced throughout its entire length (Sequiserve/Germany). The mDkk3 promoter fragment was excised from the pCR®II TOPO TA vector by XhoI/HindIII digestion and subcloned into an XhoI/HindIII-digested pGL3-Basic Vector (Promega/USA).

Site-directed mutagenesis of the mDkk3 promoter fragment

Site-directed mutagenesis of the most conserved and proximal (relative to the TSS) Lef1/Tcf binding sites predicted in the 744-bp mDkk3 promoter fragment was done using the QuickChange Lightning Multi Site-Directed Mutagenesis Kit (Agilent Technologies/USA) according to the manufacturer’s instructions. Mutagenic primers were the following: Dkk3-mA: 5′-ccagcttgcagctcagctgcgttcattc gaa ttgggtg-3′; Dkk3-mB: 5′-gtccaagagatcccagtaatagaacggatgg ggaaa tagta aaggaa-3′; Dkk3-mC: 5′-ggtggtcctgcagtctgcga tagctttccgggac-3′ (mutagenized nucleotides in bold; core sequence of the corresponding Lef1/Tcf binding site in italics). Mutated promoter fragments were confirmed by sequencing (Sequiserve).

Cell culture, transfections and luciferase reporter assays

HEK-293T cells were kept at 37°C and 5% CO₂ in DMEM medium + 10% fetal calf serum/glutamine (Life Technologies). HEK-293T cells (1.25 × 10⁵ cells/well of a 24-well plate) were co-transfected with 300 ng/well pGL3-mDkk3 promoter/reporter vectors (wild-type and mutagenized sequences), 30 ng/well pRL-SV40 (as internal transfection control, Promega) and 150–225 ng/well pcDNA3.1 (Life Technologies) ‘empty’ vector, alone or together with 75 ng/well, 150 ng/well or 300 ng/well constitutively active ΔN-β-catenin (68) or rat Lef1 cDNA (67) using Lipofectamine 2000 (Life Technologies). The total amount of plasmid DNA transfected in each well was 630 ng. Cells were lysed in Passive Lysis Buffer (Promega) after 24 h, and Firefly and Renilla Luciferase luminescence were measured in a Centro LB 960 luminometer (Berthold Technologies/Germany) using the Dual-Luciferase® Reporter Assay System (Promega) according to the manufacturer’s instructions. Firefly luminescence was normalized against Renilla luminescence for each well. Assays were performed in triplicates, and data are derived from three independent experiments.

Statistical analyses

All values shown are mean ± SEM. Statistical significance between groups was assessed by two-tailed independent-samples t tests using the SPSS 18.0 software (SPSS Inc./USA). A value of P < 0.05 was considered significant.