Transcriptome signature of the adult mouse choroid plexus

All experiments were conducted using 8- to 9-week old C57BL/6 male mice (Charles River, Barcelona, Spain). Animals were maintained under 12 h light/dark cycles at 22.5°C and 55% humidity and fed with regular rodent chow and tap water ad libitum. In order to reduce the stress-induced changes in the hypothalamus-pituitary axis, animals were handled for 1 week prior to the start of the experiment. Mice were anesthetized with intraperitoneal injection of ketamine hydrochloride (150 mg/Kg) plus medetomidine (0.3 mg/Kg) and transcardially perfused with cold saline. The CPs were rapidly removed, stored in RNA later (Ambion, Austin, TX, USA) and kept at -80°C. CP isolation was made under conventional light microscopy (SZX7, Olympus, Hamburg, Germany). Three pools of CP (each containing tissue removed from the 4 ventricles of 3 mice) were prepared at the time of sacrifice. While most of the cells are CP epithelial cells, the tissue sampled may also contain stroma cells (e.g. endothelial cells and pericytes). These animals correspond to those used as controls (basal transcriptome) in Marques et al. [8].

Total RNA was isolated with Trizol (Invitrogen, Carlsbad, CA, USA) following manufacturer's instructions. After quality assessment using the Agilent Bioanalyzer (Agilent Technologies, CA, USA), 100 ng of total RNA from each pool was amplified and labelled with Illumina TotalPrep RNA Amplification Kit (Illumina Inc., San Diego, CA, USA). The labelled cRNA was then hybridized in a total of two Illumina Whole-genome Mouseref-8 expression Beadchips (Illumina Inc.), following the recommended protocol, each containing 8 arrays, comprising a total of 24,049 well-annotated RefSeq transcripts per beadchip.

After scanning, raw data from BeadStudio software (Illumina Inc.) was read into R/Bioconductor and normalized using quantile normalization. A linear model was applied to the normalized data using Limma package in R/Bioconductor [13]. A contrast analysis was applied and differentially expressed genes were selected using a Bayesian approach with a false discovery rate of 5%. All data is Minimum Information About a Microarray Experiment (MIAME)-compliant and the raw data has been deposited in the GEO database (accession: GSE23714). In order to analyze and compare the CP transcriptome with transcriptomes from other tissues, the initial list containing the 24,049 CP genes was filtered: the repeated genes present in the array platform were excluded and the genes with a Coefficient of Variation (CV) ≥ 30 across the 3 replicates were also removed. The final gene list included 18,160 genes. For the analysis, normalized genes were categorized using the Gene Set Analysis Toolkit from WebGestalt (http://​bioinfo.​vanderbilt.​edu/​webgestalt/​[14]) or the Ingenuity tools (Redwood City, CA, USA). Enrichment analysis was performed using DAVID (http://​david.​abcc.​ncifcrf.​gov/​[15]) and Ingenuity (http://​www.​ingenuity.​com[16]) software. For comparison with other transcriptomes, two studies of the GEO database were used: one containing 61 mouse different tissues in physiological conditions (GDS592) [17], and another containing the transcriptome of the PECAM-1+ blood-brain barrier (BBB) microvascular endothelial cells (ppMBMECs) (GSE14375) [18]. From each gene list the repeated genes and the genes with CV ≥ 30 were removed. After these corrections, a list containing the genes common in the three studies was created, analyzed with the Limma package to normalize data between arrays and, subsequently, a cluster was constructed using the Multiexperiment Viewer program.

The GEO database has data on two other CP studies. One focuses on the differential CP expression between B10.pl WT and B10.PL RAG-/- mice (GSE11443); the other evaluates the gene expression profile of 24 neural tissues (including the CP from the 4^th ventricle) and 10 other body tissues/organs (GSE3594). We compared our transcriptome with the control transcriptome of each of these two other studies. For these comparisons, we selected the common genes excluding those presenting a CV ≥ 30. After these corrections, data between arrays was normalized with the Limma package, followed by a comparison of the expression level for each gene.

In order to validate the array data, 20 genes within the lower range of expression (6.1-7.0) were analyzed in an independent set of CP samples. The selected genes were: inducible T-cell co-stimulator, myotubularin-related protein 7, retinal pigment epithelium 65, dual oxidase 2, arginine vasopressin receptor 1B, glutathione peroxidase 5, nestin, epithelial mitogen, interleukin 6, neuropeptide Y, matrix metalloproteinase 9, glucagon receptor, claudin 15, ceruloplasmin, lipocalin 2, CD14 antigen, vav 1 oncogene, retinoic acid receptor gamma, interferon regulatory factor 1 and toll-like receptor 4. Gene abbreviations and gene names are specified in accordance with the HUGO Gene Nomenclature Committee at http://​www.​genenames.​org/​[19]. Total RNA was isolated from five pools of CP using Trizol reagent (Invitrogen), and 500 ng amplified using the Superscript RNA amplification system (Invitrogen) according to the manufacturer's instructions. CP RNA was reverse transcribed using random primers of the Superscript First-strand Synthesis system for RT-PCR (Invitrogen).

The oligonucleotide primers (sequences available upon request) were designed using the Primer3 software on the basis of the GenBank sequences. Real-time PCR reactions, using equal amounts of total RNA from each sample, were performed on a CFX 96™ real-time system instrument (Bio-Rad Laboratories, Hercules, CA, USA) using QuantiTect SYBR Green RT-PCR reagent kit (Qiagen, Hamburg, Germany). Product fluorescence was detected at the end of the elongation cycle. All melting curves exhibited a single sharp peak at a temperature characteristic of the primers used.

In order to identify major functions of the CP, networks based on the genes displaying higher expression (threshold 13 in the array) were constructed. Analysis yielded 59 genes that were subsequently analysed in the FunNet (http://​www.​funnet.​info[20]) for GO Molecular Function and Kyoto Encyclopedia of Genes and Genomes (KEGG) functional networks. This program clusters the genes depending on their biological function, creating nodes corresponding to those that are most enriched. The relation between every pair of genes is calculated using the signal expression of the genes on the corresponding biological functions.

From the complete array analysis, we found that the majority of genes in the CP transcriptome are expressed at very low levels. From a range of expression levels that varied from 6.1-15.0, (log2-transformed absolute expression levels), 81% of the genes were expressed between the 6.1-8.3 level; 15% were expressed at levels between 8.3-10.5; 4% were expressed with the level between 10.5-12.8 and only 0.4% of the genes were highly expressed and presented an expression level between 12.8-15.0 (Figure 1). All the genes detected and their relative expression levels are presented as Additional File 1 (Expression levels of the genes transcribed in the basal choroid plexus) and deposited in the GEO database GSE23714.

Twenty genes, with expression levels ranging from 6.1 to 7.0, were further analyzed by RT-qPCR in CP collected from an independent group of animals. This analysis confirmed the relative expression of these genes (data not shown), and indicated that genes with an expression level of 6.1, the lowest level in the array, are detectable by RT-qPCR. Altogether, it was inferred from the analysis that the array data represented genes expressed under basal physiological conditions.

Table 1 lists the fifty-nine most highly expressed genes (threshold 13.0 in the array). Analysis of the genes displaying expression levels above 13 were clustered in particular functional networks (Figure 2). Analysis of gene function networks identified 3 modules of interaction for the 59 mostly expressed genes. One included genes participating in pathways such as regulation of the tissue remodelling, fibril organization, negative regulation of processes such as extracellular matrix disassembly and blood vessel remodelling. Another, included genes of iron homeostasis, carbohydrate and lipid metabolism. A third module included genes involved in free radical homeostasis, regulation of the inflammatory response, mitochondrial metabolism and RNA processing. A table containing the various genes in each module is provided as Additional File 2 (Functional networks of the most highly-expressed genes in the choroid plexus).

Figure 3 identifies the molecular pathways encompassing genes with an expression level between 6.1-12.8. Of note, is the number of genes clustered in classes related to oxidative phosphorylation and ribosomal function, which is in agreement with the high secretory capacity of the CP. Interestingly, adhesion and regulation of actin cytoskeleton are also highly expressed, probably reflecting the barrier functions. In addition, it highlights genes belonging to various signalling pathways.

The CP has traditionally been viewed as a tissue that provides the brain with secreted CSF, which through bulk flow, supplies the brain with some nutrients such as vitamin C, amino acids, nucleosides and transport proteins such as transthyretin. The CP is also known to facilitate the removal of products of brain metabolism such as free iodine, glutamate and amyloid beta peptide. On the other hand, the CP provides a barrier that restricts the passage of molecules and cells into and out of the brain. The present study provides solid evidence to support this view. In agreement we found that the CP transcriptome includes genes encoding for the tight junctions proteins (Figure 4A), for transporters namely for monocarboxylic acid (Slc16a9), glucose (Slc2a1), vitamin C (Slc23a2), thiamine (Slc19a2), zinc (such as Slc39a1) (Figure 4B) and for receptors of various classes of neurotransmitters such as dopamine, serotonin, opioids, cannabinoids, glutamate, histamine and GABA and also receptors for folate, benzodiazepine, estrogen and growth hormone (Figure 4C). Of notice, with respect to transporters, the CP highly expressed both the ABC (all 7 distinct sub-families: ABC1, MDR/TAP, MRP, ALD, OABP, GCN20, White) and the Slc families (Figure 4B). In addition to the MDR/TAP and MRP, the basal transcriptome also revealed the presence of the solute carrier organic anion families (OAT and OATP) of efflux transporters. Finally, the sodium-dependent system for the excitatory acidic amino acids, e.g., glutamate and aspartate, was also expressed in the CP and suggests a mechanism for net removal of potentially neurotoxic amino acids from the brain. The CP also expressed different genes that encode transporters for glucose, fatty acid, monocarboxylic acid, nucleoside, glycerol, cationic amino acid transporter y⁺ system, iodide and neurotransmitter transporters (Figure 4B).

We chose to further highlight genes encoding for axonal guidance (Figure 5A), and for growth factors (Figure 5B) given their relevance for brain development and also for adult neurogenesis, as will be discussed later. These include insulin-like growth factor 2, transforming growth factor beta 2, vascular endothelial growth factor A and B, brain derived neurotrophic factor, ciliary neurotrophic factor transcript variant 2, fibroblast growth factors, betacellulin epidermal growth factor family member and its receptor ErbB-4, neuregulin 1, 3 and 4, amphiregulin, epiregulin and several members of the Wnt family (Additional File 1). With respect to axonal guidance, the CP basal transcriptome includes the four conserved families of guidance cues shown to have prominent developmental effects: netrins, semaphorins, SLIT and ephrins, as well as their respective receptors. Not surprisingly, among the most highly expressed genes are those encoding for major CSF proteins, such as transthyretin, prostaglandin D2 synthase, transferrin, insulin-like growth factor 2, clusterin (apolipoprotein J) and cystatin C (Table 1).

From the 62 tissues analysed, the CP transcriptome correlated with the transcriptome of the BBB endothelial cells more closely than with tissues from other brain regions or other peripheral tissues (data not shown).

We compared our CP array data with two others that are deposited in the GEO database, the GSE11443 and GSE3594. We observed that the correlation between our CP data was approximately 56% when compared to GSE11443 and 57% when compared to GSE3594. To test if the correlation we found between our study and the two others was acceptable, we also compared the correlation between the two other studies and found a correlation of 63%. Interestingly a similar correlation was found when the liver transcriptome from different studies was compared, which in this case had a correlation of 53%. Since the mostly expressed genes are those that best characterize a tissue, we also analysed if the fifty most highly-expressed genes in our study corresponded to the same fifty most highly-expressed genes in the two other CP studies. We observed that 34 and 20, from the GSE11443 and GSE3594 respectively, are common between the studies. Among the mostly expressed genes in these 3 studies are the genes encoding for Arl6ip1, Ckb, Cox6a1, Cst3, Igf2, Ptgds, Rbp1, Rpl41, Slc4a2, Ttr and Ubc.