MousePat

Animal care and use

Twelve weeks old adult male mice of the C57Bl/6J strain, purchased from Charles River, were used. Mice were shipped at eleven weeks of age and were allowed one week adaptation before sample collection. During that time, mice were maintained 5 per cage and fed a standard chow (D03, SAFE, France) with free access to water on a 12 hour light/dark cycle (7am to 7pm light period).

Tissue harvest

In order to reduce the variation in gene expression, sacrifice of the mice and tissue harvest were carefully standardized. Influences from the age and estrus cycle were avoided by using only male C57BL/6J mice of 12 weeks of age. To minimize variation due to the nutritional status of individuals, mice were fasted for 4 hours before sample collection, which means that food was taken away at 7.00 am and samples were collected between 11.00 to 12.00 am. Mice were sacrificed by gas inhalation and whole brains were rapidly isolated. Brains for in situ hybridization (ISH) and laser micro-dissection (LMD) are embedded in OCT mounting medium (Shandon Cryomatrix™, Thermo Electron, France ), frozen on dry ice and stored at -80°C until used. For qRT-PCR, 7 sub-regions that showed clear morphological boundaries, allowing macro-dissection with reduced risk of cross-contamination, were collected: olfactory areas (OA), cerebral cortex (Cx), hippocampus (Hi), hypothalamus (Hy), colliculi (Co), cerebellum (Cb) and brain stem (BS). Resulting tissue samples were frozen in liquid nitrogen and stored at -80°C until RNA extraction . Three additional regions, that show no morphological landmarks, were obtained by laser micro-dissection (LMD-see below), i.e. caudate putamen (CP), thalamus (Th) and arcuate nucleus (Arc).

Tissue sectioning and Laser Microdissection (LMD)

The fresh frozen brains were sectioned in coronal planes using a Leica cryostat (CM 3050 S). For manual ISH, 12 sets of 18 µm serial coronal sections were cut covering the entire length of the brain, from the olfactory bulbs to the medulla. For automated ISH (see below), 8 sets of 25 µm serial coronal sections were similarly cut. The distance between sections was thus similar for both methodologies (˜ 200 µm). Sections for manual ISH are slightly thinner as no proteinase K treatment is used prior to hybridization.

For LMD, cryosections (16 µm-thick) were cut at -16°C and thaw mounted onto PEN-slides (Leica 11505158, France), stained with toluidine blue and dehydrated in 70% and 100% ethanol followed by an incubation at 37°C for 1h. The thalamus (Th), caudate putamen (CP) and arcuate nucleus (Arc) were obtained by LMD (Leica AS) at 20-40x magnification. Approximately 5000-10000 cells were collected and total RNA was purified according manufacturer’s protocol (RNeasy microkit, Qiagen, France). The typical yield of purified RNA was about 40 ng, with a 28S:18S rRNA ratio of 1.6-1.7.

Quantitative real time RT-PCR (qRT-PCR)

RNA isolation, cDNA preparation and qRT-PCR

Standardized procedures used in this study were developed and validated in the context of the EUMORPHIA European network and are available on the European Mouse Phenotyping Resource of Standardised Screens (EMPReSS). The quantitative expression of the 49 mouse NRs was assessed by qRT-PCR on a LigthCycler LC480 (Roche) using SYBR Green as the detection method (list of primers in Metadata table).

Analysis of results

The relative concentrations of the NRs were calculated by the standard curve associated to the "Fit points" method (Roche diagnostics protocol, URL). The Fit Points Method converts a sample’s exponential curve to a straight line by plotting the logarithm of fluorescence versus cycle number. Classically, uninformative background fluorescence is then discarded by setting the horizontal "noise band". In the "Fit points" method this baseline adjustment is determined manually and not automatically, allowing to minimize the differences in the PCR efficiencies that may vary for each NR primer set and to optimize difficult standard curves. To provide the most complete representation of mRNAs, we used an external "universal" cDNA standard, transcribed from a pool of RNA extracted from a large number of tissues (BAT, WAT, brain, eyes, liver, gut, muscle, testis, ovary, skin, lung, spleen, and kidney). A large amount of this standard was prepared and used repeatedly as a basis to compare the results, thus increasing the reproducibility of the measurements. Undiluted standard was given an arbitrary value of 100. In the figures, results are thus presented as relative concentration of mRNAs, calculated using the standard curve with the Fit Points Method, and normalized to 18S RNA. The resulting mean values from 5 animals were multiplied by 10⁴ for graphical representation and plotted ± standard deviation (S.D.).

Non radioactive In situ hybridisation (ISH)

Probes

Gene expression was detected with digoxigenin-labeled RNA probes that were generated from DNA templates using classic in vitro transcription procedures with SP6, T3 or T7 polymerases (Chotteau-Lelievre et al., 2006). Two different types of templates were used: plasmid vectors containing cDNA fragments of the genes of interest and templates synthesized by RT-PCR, prepared according to the GenePaint procedures as described online at http://www.genepaint.org/. Details about the probes and their origin are provided in the Metadata table.

In situ hybridization

To map gene expression within coronal sections of the brain, two different methodologies of ISH were used : manual and automated. The first method, manual non radioactive ISH, allows to detect specific complementary mRNA sequences at cellular level using digoxigenin-labeled probes in a 5 step procedure: hybridization of the probe to pretreated tissue at 65°C; post-hybridization stringent washes; blocking steps to prepare immuno-detection; primary antibody anti-DIG-AP incubation; and colorimetric enzymatic detection. Classically detection is performed for a period of 2 to 3 days. Details of this procedure have been previously described (Chotteau-Lelievre et al., 2006). The second methodology is automated and uses a Tecan robot with GenePaint technology (Herzig et al., 2001; Visel et al., 2004) available at the ICS in the context of the Eurexpress network. Although the basic steps of hybridization are similar in both methodologies, a Tyramide Signal Amplification (TSA) step is utilized in the automated procedure to maximize sensitivity. This step leads to a specific granular appearance of the signal as blue particulate precipitates, in opposition to the diffuse blue aspect of signal in the manual procedure. TSA also greatly reduces the length of detection step, down to approximately 30 min. The manual methodology used here produces very low level of background, which allows a long staining reaction and a very sensitive detection. On the other hand, this long staining reaction leads to a saturation of the labeling and thus reduces the difference in levels of intensity between structures. Due to the amplification step and the short staining reaction, the automated method allows a more quantitative annotation of signal. The combination of both ISH strategies increases the robustness and the sensitivity of the analysis (Figure S2A). We focused our analysis only on results that are reproducible and consistent between both ISH methods, otherwise ISH was repeated (Figure S2B).


Figure S2A


Figure S2B
All procedures have been standardized and validated in the context of the EUMORPHIA and the Eurexpress European networks and are available online at: http://empress.har.mrc.ac.uk/ and http://www.genepaint.org/. One set of sections of each brain is hybridized with a sense probe to evaluate background and non-specific labeling.

Analysis of results

Annotation is carried out for over 100 brain anatomical regions, identified on the basis of the "Comparative Cytoarchitectonic Atlas of the C57Bl/6 and 129Sv Mouse Brain" (Hof et al., 2000). Each region/nucleus is assigned a level of expression, and for some regions, a pattern of expression. Levels and patterns are adapted from the GenePaint annotation procedures to fit both manual and automated ISH signal (Figure S2C). Five different levels of expression were defined :

• 0: no color precipitate detected
• 0.5: very weak expression that was difficult to discriminate from background
• 1 : weak expression, a few small particles of color precipitate per cell
• 2: medium expression, colored precipitate filling approximately half of the cell
• 3: strong expression, color precipitate completely filling the cell

For some structures, such as the cerebral cortex or cerebellum, the expression was not uniform, for those we distinguished three different types of expression patterns:

• u: ubiquitous expression, all cells within the structure expressing the gene
• r: regional expression, signal restricted to certain subregions of the structure
• s: scattered expression, signal restricted to a subpopulation of cells within the structure (http://www.genepaint.org/)

Figure S2C
A sequential analysis of the hybridization signal was performed in order to obtain the final expression data (Figure S2B). As the first step, all sections generated by both manual and automated methods were analyzed in detail using a Leica microscope, at 5 to 40x magnification, and levels and patterns were recorded for each gene in all regions. A minimum of one brain per methodology is analyzed. In a second step, data are pooled to produce a unique expression annotation for each gene, taking into account the characteristics of both methodologies.

Image acquisition

For illustration, the slides issued from only one hybridization that was of the best quality and was most representative were scanned. For this reason, there may be some slight discrepancies between the expression annotation table and the images on the website. Whole slides were automatically scanned using the NanoZoomer Digital Pathology (NDP) C9600 ( Hamamatsu Photonics, France). Highly contrasted structures were easily focused and could be scanned at the highest definition using the 40x objective, while weakly stained slides had to be scanned at 20x. For slides scanned at 40x, a generic image generated by the scanner with a median resolution was used as such. For the slides scanned at 20x, resolution of this generic images was lower, and we thus elaborate a general flow scheme for the treatment of raw images acquired by the scanner. These raw images were about 300 Go per slide and were cut arbitrarily in 2 parts by the scanner. A first "script" was developed to first resize the images subsections (25%), and to paste them together to regenerate the full slide. The resulting files were opened with Adobe Photoshop and individual sections were cropped and saved as JPEG file with a 150 or 120 pixels/inch resolution. From the ˜60 sections that were hybridized per gene, about 30 have been selected and saved. Another dedicated script was developed to automatically load the final images files into the database, according to naming conventions, and to generatie of thumbnail images visible on our website. Finally, a third dedicated script was developed to automate the generation of zooming files, based on the Zoomify file format specifications. A Zoomifiy Flash plug-in is freely available at http://www.zoomify.com/ and allows embedding of zooming capabilities directly within HTML pages. The three image manipulation scripts developed in python rely heavily on the Python Imaging Library (PIL: http://www.pythonware.com/products/pil/).

To provide information on where the selected coronal section lies in the brain, we have labelled each image in regards of its position along the antero-posterior (AP) axis of the brain. As detailed above, for manual ISH, 12 sets of 18 µm serial coronal sections were cut covering the entire length of the brain, starting from the olfactory bulbs towards the medulla, while for automated ISH 8 sets of 25 µm serial coronal sections were similarly cut. The distance between sections is thus 216 µm for manual ISH and 200 µm for automated ISH. Considering the olfactory bulbs to be at the zero point, each section can thus be located along the AP axis according to its distance in micron from the anterior extremity (see Figure S3). The precision of this value can vary slightly, taking into consideration the loss of sections during the sectioning procedures. This localisation value has been calculated and associated to each ISH images. This value is then used

1. to link the selected section to a schematic drawing of the brain, allowing the visualisation of where the section lies in the brain;
2. to provide information on the brain structures that are present on the section (two first levels of hierarchy from the annotation table). This information appears on the top of each images when the “cursor” is positioned over the thumbnails, and is included with the image when the zoomify tool is used.

Figure S3 : schematic representation of the adult mouse brain, illustrating the localisation of coronal sections along the antero-posterior axis. In this example, the brain was sectioned for manual ISH, so sections collected for one gene were separated from each other by 216 µm (see text for details).

Identification of specific cell types

To identify neuronal and glial subtype, we hybridized and annotated brain sections with probes for specific molecular markers. Glutamatergic excitatory neurons were identified by their content in vesicular glutamate transporters (vGlut 1 and 2 or Slc17a7 and Slc17a6, respectively), which pack the transmitter into the synaptic vesicles for release. GABAergic inhibitory neurons were identified by the presence of the enzyme glutamate decarboxylase (GAD) that catalyses the decarboxylation of Glu to GABA and CO2. In rodents, it exists in two isoforms: Gad1 or Gad67, and Gad2 or Gad65. Cholinergic excitatory neurons were identified by the presence of the Ach synthesizing enzyme, choline acetyltransferase (ChAT). ISH with antisense probes for vGlut1, vGlut2, Gad1, Gad2 and ChAT have thus been performed to identify these neuronal cell types. The oligodendrocytes, a glial cell type, were identified by the expression of the proteolipid protein 1 (Plp), one of the main components of myelin.

Clustering

Unsupervised hierarchical clustering was performed on either normalized gene expression data for the qRT-PCR results, or on raw data for the ISH results, using the tools available at the GEPAS web site (http://www.gepas.org) (Montaner et al., 2006) NAR 34 (Web Server issue): W486-W491). Prior to clustering, the qRT-PCR data was Log2 transformed, and further Z-score transformed using the global mean and SD to facilitate easy visualization. Clustering was performed using Complete linkage with Euclidean distance for both tissue and gene axes, and visualized using the Treeview tool in the GEPAS tool box. The original ISH scores were subtracted by 0.5 for better visualization, and clustered as above without any further transformation.

Informatics resources

The programming underlying the MousePat web resource follows a classical three-tier architecture (MVC, model-view-controller), mostly based on open-source technologies. An Apache web server (http://httpd.apache.org/) serves directly the static part of the site (images, css, zoom files) and proxies dynamic requests to the application server. The server-side part was developed using the python programming language. This choice was motivated by the scripting capabilities of the language, the richness of libraries available, and the general agility of the framework. MousePat uses Pylons, a modern application server (http://pylonshq.com/). The templates for HTML pages were developed using Myghty (http://www.myghty.org/). Finally, the database is accessed through an object relational mapper (SQLAlchemy: http://www.sqlalchemy.org/). A thin layer was developed above the latter to automate the generation of standards as well as customized HTML form widgets. Data are stored in the Postgresql relational database engine, which has all modern features required (views, stored procedures, triggers). An export of the schema is available for download on MousePat. The graphs generated for qRT-PCR are made with a dedicated graphing library (Matplotlib: http://matplotlib.sourceforge.net/)
At the programmatic level, the organ structure in the database is managed as an in-memory hierarchical tree of objects representing the different ranks, which are at the number of three for the moment, namely the brain major structures (telencephalon, diencephalon, brain stem, cerebellum), the brain regions (olfactory areas, cerebral cortex, subcortical structures, septum, hippocampus, hypothalamus, thalamus, sub/epithalamus, mesencephalon, pons, medulla, deep cerebellar nuclei, cerebellar cortex), and the areas/nuclei of different regions (around 100).

References

• Chotteau-Lelievre, A., Dolle, P., and Gofflot, F. (2006). Expression analysis of murine genes using in situ hybridization with radioactive and nonradioactively labeled RNA probes. Methods Mol Biol 326, 61-87.
• Herzig, U., Cadenas, C., Sieckmann, F., Sierralta, W., Thaller, C., Visel, A., and Eichele, G. (2001). Development of high-throughput tools to unravel the complexity of gene expression patterns in the mammalian brain. Novartis Found Symp 239, 129-146; discussion 146-159.
• Hof, P. R., Young, W. G., Bloom, F. E., Belichenko, P. V., and Celio, M. R. (2000). Comparative cytoarchitectonic atlas of C57BL/6 and 129/SV Mouse Brains: Elsevier).
• Montaner, D., Tarraga, J., Huerta-Cepas, J., Burguet, J., Vaquerizas, J. M., Conde, L., Minguez, P., Vera, J., Mukherjee, S., Valls, J., et al. (2006). Next station in microarray data analysis: GEPAS. Nucleic Acids Res 34, W486-491.
• Visel, A., Thaller, C., and Eichele, G. (2004). GenePaint.org: an atlas of gene expression patterns in the mouse embryo. Nucleic Acids Res 32, D552-556.