Analysing the developing brain transcriptome with the GenePaint platform

Abstract

We discuss technical means by which the complexity of gene and protein signalling cascades can be projected onto the complex structure of the mammalian brain. We argue that this requires both robotics and novel computational tools to register images of gene expression, annotate expression patterns and quantify gene expression. When sufficiently enriched and detailed, such gene expression/neuroanatomical atlases are hypothesis-generating tools and contain in themselves much of the information needed to investigate function in normal and genetically or otherwise modified brains. To be successful and useful, data-rich and comprehensive gene expression/neuroanatomical atlases have to be web accessible and structured in a way that allows the application of data exploration and mining tools.

Figure 1. Major challenge in merging cellular and molecular data

A typical neuronal circuit involves multiple neurons (Ramón y Cajal, 1892). Shown is an idealized case where three different types of neurons each express a particular gene (A, B and C). While it is easy to determine which gene marker colocalizes with a particular neuron in a two-dimensional section using double-labelling strategies, this assignment becomes a formidable challenge when extended to the complete genome and to the entire three-dimensional brain. Moreover, it is unlikely that genetic markers exists that tag each and every type of neuron in the brain. Thus, many neurons probably can only be identified by a combination of gene markers requiring high quality image registration.

Figure 2. GenePaint robot and results

A, a solvent delivery robot whose cluster of pipettes (p) aspirate solutions from containers located on the platform and deliver them into flow-through hybridization chambers located in temperature-controlled chamber racks (c). B, a flow-through hybridization chamber in which a 5 mm thick glass plate is clamped together with a slide that carries sections. The thickness of this chamber is 80 μm which is achieved by placing two spacers (s) between the glass plate and the slide. A solvent reservoir (r) accommodates the solutions delivered by the pipettes to the chamber. C, expression of arginine vasopressin (Avp) in the paraventricular nucleus of the hypothalamus. Shown is a 25 μm fresh frozen mouse brain section. Note that despite this thickness the data provide single cell resolution of the staining as long as Avp-expressing cells are clearly spatially separated. D, a pseudo-darkfield of C generated by applying the ‘invert’ command of Adobe Photoshop. This operation generates an image similar to one that would be generated using a radioactive riboprobe and emulsion autoradiography. E, the expression of a solute carrier (Slc12a1) in an 8 μm thin paraffin section of a paraformaldehyde-fixed adult mouse kidney. Frozen sections (C and D) and paraffin section (E) both give expression data with strong signal and low background. Scale bars: C and D, 100 μm; E, 650 μm.

Figure 3. Neurog2 expression in wild-type and Pax6^Sey/Sey cortex at E15.5

Genepaint-ISH detection of Neurog2 transcripts on sagittal sections of wild-type (A) and Pax6^Sey/Sey (B) cortex at E15.5. Arrows indicate the region of expression in the germinal layers of the cortex. Neocortical expression is completely absent in the mutant (B). Abbreviations: BG, basal ganglia; CTX, neocortex; HIPP, hippocampus. Scale bars 500 μm.