eMouseAtlas, EMAGE, and the spatial dimension of the transcriptome

Abstract

eMouseAtlas (www.emouseatlas.org) is a comprehensive online resource to visualise mouse development and investigate gene expression in the mouse embryo. We have recently deployed a completely redesigned Mouse Anatomy Atlas website (www.emouseatlas.org/emap/ema) that allows users to view 3D embryo reconstructions, delineated anatomy, and high-resolution histological sections. A new feature of the website is the IIP3D web tool that allows a user to view arbitrary sections of 3D embryo reconstructions using a web browser. This feature provides interactive access to very high-volume 3D images via a tiled pan-and-zoom style interface and circumvents the need to download large image files for visualisation. eMouseAtlas additionally includes EMAGE (Edinburgh Mouse Atlas of Gene Expression) (www.emouseatlas.org/emage), a freely available, curated online database of in situ gene expression patterns, where gene expression domains extracted from raw data images are spatially mapped into atlas embryo models. In this way, EMAGE introduces a spatial dimension to transcriptome data and allows exploration of the spatial similarity between gene expression patterns. New features of the EMAGE interface allow complex queries to be built, and users can view and compare multiple gene expression patterns. EMAGE now includes mapping of 3D gene expression domains captured using the imaging technique optical projection tomography. 3D mapping uses WlzWarp, an open-source software tool developed by eMouseAtlas.

Fig. 1

The eMouseAtlas main page. The “film-strip” embryo selector (a) can be scrolled via click and drag directly or by using the control buttons at the top left. Selecting a single model here influences what is displayed in the bottom half of the page. Each Theiler stage may contain multiple examples of data (usually littermates) and these can be individually selected here (b). Each embryo potentially can be viewed in a number of ways: the original high-resolution sections; a 3D reconstruction from the former; movies of the embryo and its littermates; associated information such as its anatomy ontology, Theiler staging, a section browser, and download options of the embryo data (c)

Fig. 2

High-resolution sections and the IIP3D viewer. Top An example of the original sections that are digitised and could be used for creating 3D reconstructions. These are E14.5 (TS23, model EMA:83) H&E-stained high-resolution coronal histological sections. Bottom Typical IIP3D viewer window when viewing the “3D reconstruction” option from the main-stage view window on the EMA website. The IIP viewer window comprises a main viewer that is controlled by slider controls for pitch (p), yaw (y), roll (r), magnification, and distance (a). A guide to the plane being selected is provided in the view directly above. A real-time preview of the plane is provided in b, where the transparent grey box represents the viewable dimensions in the main viewer. Right-clicking (ctrl-click on Mac) in the main viewer brings up a context menu that allows, amongst other things, a measurement mode by clicking two arbitrary points in the main viewer in addition to being able to retrieve the original high-resolution image (see top panel) used to create the 3D reconstruction (c)

Fig. 3

The EMAGE combination query feature allows a user to build complex queries. In this example, the combination query is used to find all OPT data with author Dr. Paula Murphy, Trinity College, Dublin, at a single stage of embryonic development. a An expandable multi-input selector allows a user to select criteria for a complex query. In this example, these include associated with author = murphy; specimen type = opt; Theiler stage = 17. Additional options include b A list of 31 OPT entries with author Dr. Paula Murphy at Theiler stage 17 is returned from the combination query. This list can be further refined by use of the expandable multi-input selector

Fig. 4

EMAGE Pathways uses KEGG to find gene expression patterns associated with specific pathways and/or specific tissues. In this example, EMAGE Pathways is used to find components of the Notch signalling pathway that are important in somitogenesis. Two tabs are returned from a pathway query. The first tab (a) is entitled “pathway gene” and shows gene expression summaries associated with all gene components of the KEGG Notch signalling pathway. Spatially mapped EMAGE data are shown in the Gene column; if spatially mapped data are unavailable, a thumbnail will display “text annotation only” (b). The Heatmap column shows the union of all spatially mapped expression patterns for a single KEGG pathway at a single stage (c). Clicking on the second tab (d) allows access to the KEGG pathway diagram (e). The KEGG pathway diagram is marked up to show data available in EMAGE for a given gene by highlighting these genes in red. A subset of Notch signalling pathway genes can be implicated in somitogenesis by virtue of their gene expression patterns (f). These include the Notch ligands Delta-like 3 (EMAGE:152) and Delta-like 1 (EMAGE:3691); the Notch1 receptor (EMAGE:749); and the hairy-like gene Hes1 (EMAGE:3417), all of which are expressed in presomitic mesoderm. In contrast, Notch2 (EMAGE:4605) and Hes5 (EMAGE:3421) are not expressed in the presomitic mesoderm but are expressed in the somites, suggesting that these are downstream effector genes in somite formation. In this respect it is additionally noteworthy that Delta-like 1 (EMAGE:3691) shows higher levels of expression in presomitic mesoderm than in the somites. Lunatic Fringe (EMAGE:3449; EMAGE:891) is expressed in discrete domains in the presomitic mesoderm and the condensing somite and illustrates the critical importance of this gene in forming boundaries between somites by a mechanism involving oscillatory inhibition of the Notch receptor

Fig. 5

3D spatial mapping using WlzWarp. a 3D volume data are warped onto target reference models via a mesh that envelopes the target and excludes background and luminal areas. b The mesh is read into the WlzWarp interface, alongside the reference and source objects. WlzWarp allows user-defined pairwise landmarks to be placed on source and target and subsequently allows warping via the CDT algorithm. c 3D mapped data from multiple embryos can be viewed as a volumetric reconstruction or can be visualized on an arbitrary section by using a novel 3D search interface based on the IIP3D viewer. d Volumetric data can be queried in web browsers using a painted domain as the basis of a spatial query. e It is possible to refine 3D spatial mapping using advanced neuroimaging tools (ANTs) (Avants et al. 2011). These additional steps are performed after initial registration using WlzWarp. In this example, both cross-correlation (CC) and mutual information (MI)—two similarity metrics used to enhance 3D image registration—are used to refine the WlzWarp transformation of an OPT gene expression pattern. The target is the EMAP TS17 reference model, shown from its dorsal view (top left) with corresponding approximate section planes shown. The source embryo to be warped (top right) is an OPT gene expression pattern of Wnt6 expression at the equivalent Theiler stage. ANTs can use various algorithms to perform automated nonrigid image registration. In the examples shown, CC and MI methods were tested subsequent to a manual 3D warping step using WlzWarp. The results are presented as a colour overlay of warped source (see key) onto a grey-scale section plane of the target. ANTs improves the overall registration of the source (compare CC6SyN0 vs. WlzWarp and MISyN0 vs. WlzWarp)

Fig. 6

The interactive Kaufman’s atlas of mouse development. a Distance and magnification sliders control the main IIP viewer window display; the distance slider allows navigation through plates that appear in the original book. b The original sections have been digitised to generate high-resolution colour images that correspond to the plates used in the book. c These plates are annotated using Kaufman’s original terms plus the corresponding eMAP ontology terms, enabling direct links to eMouseAtlas and the EMAGE gene expression database. Anatomical components are flagged, and mousing over a flag highlights the corresponding text term and vice versa