Systematic image-driven analysis of the spatial Drosophila embryonic expression landscape

Abstract

Discovery of temporal and spatial patterns of gene expression is essential for understanding the regulatory networks and development in multicellular organisms. We analyzed the images from our large-scale spatial expression data set of early Drosophila embryonic development and present a comprehensive computational image analysis of the expression landscape. For this study, we created an innovative virtual representation of embryonic expression patterns using an elliptically shaped mesh grid that allows us to make quantitative comparisons of gene expression using a common frame of reference. Demonstrating the power of our approach, we used gene co-expression to identify distinct expression domains in the early embryo; the result is surprisingly similar to the fate map determined using laser ablation. We also used a clustering strategy to find genes with similar patterns and developed new analysis tools to detect variation within consensus patterns, adjacent non-overlapping patterns, and anti-correlated patterns. Of the 1800 genes investigated, only half had previously assigned functions. The known genes suggest developmental roles for the clusters, and identification of related patterns predicts requirements for co-occurring biological functions.

Figure 1

Processflow for converting expression patterns into TIs. (A) Digital photograph of CG10033. (B) Segmentation of the embryo shown in (A) with the boundary indicated by the red line. (C) On the basis of the boundary coordinates in (B), the best fitting ellipse was superimposed on the embryo. (D) Green dots show anchor points. (E) Grayscale image of the embryo in (A) with shadows introduced by DIC microscopy removed. (F) Ellipse in the ratio 4:2 subdivided into equilateral triangles. (G) Alignment of the ellipse from (F) to the embryo using the anchor points. The remaining corner points of the triangles are adjusted with a thin spline deformation algorithm. (H) Virtual representation of the staining intensities in (E) as a TI. Raw intensities of the triangles are enhanced (color saturation increased to 60%) for visualization. (I) Quality assessment of TIs and (J) genes at different developmental stages. Accurate and inaccurate fractions are shown as the percent of samples evaluated. Shown are the results of a random selection of 518 TIs/110 genes (S4–6), 269 TIs/100 genes (S7–8), 266 TIs/109 genes (S9–10), 590 TIs/109 genes (S11–12), and 705 TIs/100 genes (S13–16). (K) Three rows, the top shows four digital photographs, the middle row shows the converted TIs with their superimposed mesh, and the bottom row shows the controlled vocabulary (CV) used to annotate the expression patterns. TIs reveal subtle differences in the patterns not captured by the CV.

Figure 2

Comparing images from the literature to our data set. (A) Image of sna in situ hybridization (top) (Stathopoulos et al, 2002), converted to a TI with the fully automatic pipeline (middle). The TI is nearly identical to the TI from the BDGP collection (bottom). (B) Image of sage in situ hybridization at stage 11 (Abrams et al, 2006), showing expression in the salivary gland (top) and the corresponding TI after manual segmentation (bottom). (C) Similarity search using the TI in (B) showing top hits with other genes expressed in the salivary gland.

Figure 3

Mapping co-expressed genes on the blastoderm embryo. (A) Structure of the hierarchical clustering of the 311 triangles displayed as dendrogram. Triangles 1–311 are shown on the horizontal axis, the distance scores on the vertical axis. The red arrows denote the cut-off values chosen for B, D, E, and F. (B, D, E, F) Clusters at different cut-off values (0.30, 0.20, 0.15, and 0.10, respectively) visualized as TI. Each cluster is labeled with a different color and given a numerical identifier displayed in each triangle. (C) Fate map of the blastoderm after (Hartenstein, 1993). amg, anterior midgut rudiment (endoderm); as, amnioserosa; dEpi, dorsal epidermis; eph, epipharynx; es, esophagus; hg, hindgut; hy, hypopharynx; lb, labium; md, mandible; ms, mesoderm; mx, maxilla; pmg, posterior midgut rudiment (endoderm); pNR, procephalic neurogenic region; pv, proventriculus; vNR, ventral neurogenic region; T1, thoracic segment 1; tp, tracheal placodes.

Figure 4

Clustering genes with similar expression patterns. (A) Normalized consensus patterns for each of the Clusters #1 to #39 are displayed. The consensus patterns were computed from the most distinct 553 TIs representing 336 genes. (B) Distribution of genes and relationships among clusters after each pattern was classified. The size of the pie chart on the top of each cluster is proportional to the number of genes in each cluster. The total number is shown in green above each pie chart. The blue area of each pie is proportional to the fraction of characterized named genes and the red area to the fraction of marginally studied genes (CG identifiers only). The blue lines between clusters denote the occurrence of genes that were classified into multiple clusters. (C) Distribution of patterns. D=dorsal, V=ventral, A=anterior, P=posterior. The height of the bars corresponds to the percentage of patterns in the direction D-V, A-P, predominantly D, V, A, or P or combinations of D and V (D&V) or A and P (A&P).

Figure 5

Targeted mining for differences from the cluster consensus. (A) Single TI shows the consensus Cluster #14 expression pattern, enlarged relative to Figure 4A and in two color schemes. On the left, blue indicates uniform expression, white no expression, and the normalized standard deviation of triangles of expression patterns in the cluster in yellow. Triangles with greater variation are brighter yellow. On the right, for emphasis of the standard deviation, blue and yellow are switched. Darker blue indicates greater variation. The color-coding emphasizes variations in the bandwidth and also reveals three major regions of discontinuous expression (green arrows). (B) First 30 TIs from Cluster #14. Boxed in green is the expression pattern of the gene, tin, highlighted because the pattern is discontinuous at the anterior end. (C) Using the MRF algorithm and the cluster consensus expression pattern, we extracted the tin regions of discontinuity. Both the consensus pattern (as determined by the median of all patterns in the cluster) and the pattern of tin are shown at the top. Shown below the curly bracket are the two extracted regions, the first in black at the anterior and the second in gray at the posterior. The black region was used as bait for a systematic search of the data set. The top 10 results of this search are shown below the arrow. Boxed in green is zfh1, a known interactor of tin. Two distinct patterns were returned for CG33099 as a result of the query.

Figure 6

Anti-correlation mining identifies known interacting genes. (A) Single TI showing the sna expression pattern. (B) Using the modified MRF algorithm and filtering for vertically oriented patterns, we used sna, as bait to identify genes with expression patterns that share similar boundaries and do not overlap. One of the top hits of the search was hkb (boxed in green), previously shown to restrict sna expression at the anterior and posterior poles.

Figure 7

Analysis of clusters for gene ontology (GO) term enrichment in clusters. (A) Enrichment of GO terms in the cellular process (GO:0009987) category (vertical axis) for each of the 39 clusters (horizontal axis). The level of significance is displayed as color intensity between white (90% and below) and red (100%) as indicated by the color bar on the top. (B) Enriched GO terms in each cluster. Clusters are on the horizontal axis, individual GO terms (614 in total) are shown as blue dots on the vertical axis and associated with one or more of the 23 parent terms. Alternating gray and white backgrounds are used to separate parent terms. Identical GO terms are placed at the same vertical height within a parent section.