Regulatory pathway analysis by high-throughput in situ hybridization

Abstract

Automated in situ hybridization enables the construction of comprehensive atlases of gene expression patterns in mammals. Such atlases can become Web-searchable digital expression maps of individual genes and thus offer an entryway to elucidate genetic interactions and signaling pathways. Towards this end, an atlas housing approximately 1,000 spatial gene expression patterns of the midgestation mouse embryo was generated. Patterns were textually annotated using a controlled vocabulary comprising >90 anatomical features. Hierarchical clustering of annotations was carried out using distance scores calculated from the similarity between pairs of patterns across all anatomical structures. This process ordered hundreds of complex expression patterns into a matrix that reflects the embryonic architecture and the relatedness of patterns of expression. Clustering yielded 12 distinct groups of expression patterns. Because of the similarity of expression patterns within a group, members of each group may be components of regulatory cascades. We focused on the group containing Pax6, an evolutionary conserved transcriptional master mediator of development. Seventeen of the 82 genes in this group showed a change of expression in the developing neocortex of Pax6-deficient embryos. Electromobility shift assays were used to test for the presence of Pax6-paired domain binding sites. This led to the identification of 12 genes not previously known as potential targets of Pax6 regulation. These findings suggest that cluster analysis of annotated gene expression patterns obtained by automated in situ hybridization is a novel approach for identifying components of signaling cascades.

Figure 1. Building and Mining the Genepaint.org Expression Pattern Atlas

(A) Robotic ISH uses digoxigenin-tagged riboprobes that are hybridized to serial sections of wild-type mouse embryos. Images of expression patterns are captured by automated microscopy, annotated, and deposited on the genepaint.org database. Data can be queried using standard web browsers. (B) In the present study, expression pattern atlas data were subjected to hierarchical clustering and genes that grouped with Pax6 were passed through several “filters” for validation purposes including ISH with Pax6^sey loss-of-function embryos and EMSAs to detect PAX6-paired domain binding sites.

Figure 2. Examples of Gene Expression Patterns in E14.5 Mouse Embryos

(A) Pie chart shows the distribution of the ten most abundant molecular function classes of the 887 expressed genes of the 1K dataset. (B–P) Sample expression patterns of genes with highly regionalized brain expression at E14.5. Shown are sagittal sections through the head. Typical examples of regional expression in the cortex (ctx) are seen in (B, E–G, I, N, and O), where transcripts are either found in the ventricular layer (E, G, O) or in the MZ (B, F, I, N; see Figure 7A for layer definitions). Scattered expression is illustrated by the insets of (C) where single A330102H22Rik-positive cells can be seen. Diagrams on the left illustrate predicted open reading frames and their putative domain structures. Abbreviations: BS, brain stem; CB, cerebellum; CTX, cortex; DI, diencephalon; GE, ganglionic eminence; PS, pons; SC, spinal cord; TC, tectum; TG, tegmentum; TH, thalamus.

Figure 3. Tree of Anatomical Structures That Were Annotated for Gene Expression

For details of annotation methods and terminology, see “Annotation of Expression Patterns” in Results.

Figure 4. Cluster Analysis of Expression Pattern Annotations

The annotation of the expression pattern for each gene at every anatomical location is represented by color (yellow for regional or ubiquitous, blue for scattered) and intensity (dark for weak expression, midtone for medium expression, and bright for strong expression). See also color key on figure. Dendrograms composed using Ward's clustering represent 70 unique anatomical structures (horizontal) and 652 genes (vertical). A total of 12 clusters/groups of genes are labeled.

Figure 5. Detailed View of Cluster 7 with Pattern and Annotation Icons Used in Genepaint.org

Notice the predominance in this cluster of genes with “strong regional” expression in the central nervous system. Icons are defined in the inset on the top left.

Figure 6. Hierarchical Clustering Is Consistent with Image Data

(A–N) Selected expression patterns of genes from cluster 7 showing that genes located at the same branch of the tree have similar expression patterns.

Figure 7. Expression Pattern of 30 Genes That Are in Cluster 7 but Are also Coexpressed with Pax6 in the Cortex of the E14.5 Mouse Embryo

(A–AE) Pax6 is strongly expressed in the VZ, and for the majority of the other genes this is also the case. Note however, genes such as Sst or Trim9 are expressed in a subset of VZ cells and additionally are also expressed in the MZ.

Figure 8. Expression Pattern of Pax6 Candidate Targets in the Cortex of E15.5 Pax6^sey/sey Embryos

(A–Q) The majority of genes are downregulated in mutant cortex. For details, see “Altered Expression of Putative Pax6 Targets in Pax6^sey/sey Embryos” in Results. Abbreviations: CPL, cortical plate; IZ, intermediate zone; MZ, marginal zone; SPL, subplate; SVZ, subventricular zone; VZ, ventricular zone; WT, wild type; -/-, Pax6^sey/sey mutant.

Figure 9. Location of Conserved Pax6-Paired Domain Binding Sites

Arrowheads indicate the location of human/mouse conserved Pax6 binding sites relative to exons of the corresponding genes. Conservation profiles indicate sequence conservation and were generated using the UCSC genome browser [23,24]. Arrow heads indicate conserved Pax6-paired domain binding sites that were examined by EMSA. Black arrowheads and asterisks indicate band shifting and supershifting, black arrowheads indicate band shifting only, white arrowheads indicate absence of bandshifting.

Figure 10. EMSA Results for Pax6 on Our Candidate Genes

Autoradiograms with results of EMSA experiments for genes containing a conserved Pax6-paired domain site. Each experiment—except for panel 2—has four lanes, which are (from left to right): (1) probe only; (2) probe incubated together with Pax6-PD-GST protein; (3) probe incubated with Pax6-PD-GST protein first in the presence of 50-fold excess of unlabeled fusion probe, and then with labeled probe; (4) probe incubated with protein plus anti-GST antibody; the supershifted band is indicated by an arrowhead. In case of multiple binding sites in a single gene, the left-most autoradiograph corresponds to the most 5' binding site.

Figure 11. Estimated Binding Affinity Based on EMSA Results

Estimates of binding affinity of Pax6-PD-GST to the binding sites found in each of the candidate genes relative to the consensus sequence. Ordinate: percent binding relative to consensus sequence; abscissa: gene name and site identity.