The maturation of the testis during puberty involves a concerted change in the populations of germline cells and of their supporting Sertoli cells (Figure 1A). Immature rat Sertoli cells at 9 dpp grow rapidly and divide, stop dividing at 15 dpp, and are mature at 22 dpp having established the blood-testis barrier by forming tight junctions among themselves. The germ cell population at 9 dpp contains type A Spga, which show rapid proliferation and self-renewal. At 22 dpp, the germ cell fraction that we isolated still contains Spga type A, but also intermediary, type B, and preleptotene spermatocytes 31 32 33 34 .

We purified Spga and Sertoli cells from testis of pre-pubertal rats at 9 dpp and pubertal rats at 22 dpp. To demonstrate the proliferative stages of both cell types at 9 dpp, we stained sections from rat testis at 9 dpp and 22 dpp with the mitotic marker Proliferative Cell Nuclear Antigen (PCNA). PCNA was highly expressed at 9 dpp; PCNA staining was decreased at 22 dpp (Figure 1B), consistent with a lower frequency of mitosis due to the shift of the germ cell population towards differentiated spermatocytes and the maturation of Sertoli cells. These concerted changes present an opportunity to study differential gene expression of SSC and their niche by cross-comparison of purified Spga and Sertoli cells at two crucial stages of testis development.

To identify genes that are specific for SSCs and their niche, we compared gene expression profiles of spermatogonia and Sertoli cells prepared from testis of pre-pubertal and pubertal rats at 9 and 22 dpp, respectively. We established 4 expression profiles, namely for spermatogonia (G) and Sertoli (S) cells, of post-natal day 9 or 22, termed G9, S9, G22 and S22. Each profile was based on three sets of separately purified spermatogonia or Sertoli cells, which were used for mRNA preparation and microarray hybridization.

Of 31099 genes on the microarray, 6908 transcripts were expressed at a level above the threshold defining differential expression (see Methods) and at least two fold up-regulated over the other groups. Hierarchical clustering of differentially expressed transcripts of the four cell preparations showed that G9, S9, G22, and S22 formed distinct groups (Figure 2A). The clustering tree showed that the differences in expression patterns were less pronounced between the cell types of the same developmental stage (between G9 and S9 or between G22 and S22), than between different developmental stages of one cell type (between G9 and G22, and S9 and S22). Specifically, Sertoli cells and Spga shared a substantial set of similarly expressed transcripts at 9 dpp. Such overlapping gene expression pattern is consistent with the mitotic program that is common to both cell types.

We then performed pair-wise comparison of two-fold enriched transcripts of the G9, G22, S9, and S22 gene sets. As shown in Venn diagrams (Figure 2B), the cross-section of the gene lists of the pair-wise comparison (e.g. gene list of G9 versus G22 and versus S9) revealed 445, 314, 1101, and 708 transcripts that were selectively enriched in G9 and S9, G22, and S22, respectively. These four "selective" gene sets were termed G9-sel, S9-sel, G22-sel, S22-sel. As shown in Figure 2C, hierarchical re-clustering of each "selective" gene set revealed specifically enriched (minimally 2-fold) transcripts in each cell population and no overlap with the three other cell populations. Thus, all four cell fractions have their own unique, divergent gene expression profiles.

To evaluate whether the differentially expressed genes reflected specific biological functions, we assigned them to gene ontology (GO) categories. This analysis revealed the functional networks involved in the onset of spermatogenesis (Figure 3A). The G9 set contained genes involved in "cell proliferation", "progression through cell cycle" or "cell cycle". Transcripts of the same GO categories were enriched in the S9 set when compared to the S22 set. These results are consistent with the fact that both Spga A and Sertoli cells are proliferating in pre-pubertal rat testis 31 32 33 34 35 . As expected, gene transcripts involved in "meiotic cell cycle", "M phase", "DNA metabolism", and "DNA repair", were enriched in the G22 sets. These gene sets overlapped significantly with the mitotic and meiotic germ cell gene clusters defined by Chalmel et al. 21 (Additional file 1, Table S1).

Genes that were most upregulated in the S9 set comprised a significantly elevated number of genes of the GO categories "cell adhesion", "neurogenesis", "tissue development", "morphogenesis", and "regulation of cell size" (Figure 3A). Mature Sertoli cells in the adult have to support a larger number of germ cells than immature Sertoli cells in pre-pubertal rats, and accompany their differentiation 36 . Presumably, the expression of genes associated with "morphogenesis" and "cell size" in S22 reflects commitment to maturation of Sertoli cells and to the onset of spermiogenesis. "Tissue development" and "neurogenesis" genes were highly expressed in S9, but were under-expressed in S22. This suggested that during the first wave of spermatogenesis at 22 dpp, the Sertoli cells are fully mature. Further, we observed a significant enrichment of the GO "immune or defense response", "response to biotic stimulus", "reproduction", and "spermatogenesis" in S22. The up-regulated expression of genes of the immune system in S22 presumably reflects the physiological process that leads to immuno-tolerance in the testis 37 . Thus, the gene categories up-regulated in the different gene sets are consistent with the functions attributed to Spga and Sertoli cells. In the "selective" gene sets the enrichment of these GO categories was similar despite the smaller list of transcripts.

To identify the most differentially expressed genes, we sorted the "selective" gene sets by their fold change of mean expression and classified them in 14 functional categories. Listing only genes with a minimum 4 fold variation of mean expression, we defined 176 genes for G9 and 170 for S9 (Table 1). Among the "cell cycle" genes upregulated in G9 were Gadd45gamma (Gadd45g), Myc, and Polo-like-kinase 2 (Plk2). S9 cells expressed significantly more inhibitors of cell cycle, such as Cdkn2a (ARF/INK4) and Mtsg1. The most up-regulated genes in the GO category "neurogenesis" were in S9, namely Growth Associated Protein 43 (GAP43) and Prepronociceptin (Pnoc). The most prominent gene transcripts within the "cell adhesion, ligand, ECM" category, differentially enriched in S9 were pap/REG3, thrombomodulin (Thpd), and Ncam1. A complete description of the genes listed in Table 1 is shown in Additional file 1, Tables S2 and S3.

To confirm the specific differential expression of genes by other means, we quantified a small set of relevant gene transcripts by RT-PCR in total RNA extracted from the same cell preparations by the same procedures as for microarray hybridization. Figure 3B shows three typical examples: plk2 is specifically expressed in G9 cells, and expression is nearly absent in the three other cell preparations. The gap43 and pap3 genes are expressed specifically in S9 cells, and expression is completely absent in the other cell types. This supports the notion that gap43 and pap3 have stage-specific functions in pre-pubertal Sertoli cells and that plk2 is relevant for pre-pubertal spermatogonia.

Pre-pubertal Spga comprise all subtypes of undifferentiated and differentiated type A Spga, but their respective abundances are not clear 19 38 39 . To evaluate stem cell characteristics of pre-pubertal Spga, we first established a list of 63 genes, composed of selected markers of embryonic stem cells, primordial germ cells (PGCs), gonocytes, or Spga from published data (Figure 4A). Of these 63 genes, 50 were represented on the RAE230 2.0 gene chip used in this study. Most of these markers were present in the G9 and G9-sel gene sets. Not included in differentially expressed gene sets were the PGCs markers DPPA3, Thy-1, TNAP and the SSC marker Neurogenin 3 (Ngn3). The absence of Ngn3 is consistent with previous reports that Ngn3 is only expressed in adult SSC 10 . The majority of previously identified SSC genes, such as Bcl6b, Egr2, Egr3, GFRA1, and Kit Ret, showed similar expression levels in G9 and G22. Only 12 of the 50 genes were specifically up-regulated in G9 versus G22 (Figure 4A). These included Dnd1 27 , GPR125 40 , Nmyc 41 , ItgB1 and ItgA6, as well as Sox2, Myc, Klf4 recognized for stem cell maintenance in ESs or PGCs 42 . Surprisingly, KitL/SCF, coding for a protein secreted by Sertoli cells and involved in Spga differentiation, was up-regulated in G9 as compared to G22 43 . This result was confirmed by immunofluorescence microscopy on Spga and Sertoli cells isolated from pre-pubertal (9 dpp) and pubertal (22 dpp) rats (Additional file 2, Figure S1). A similar observation was made by Munsie et al. 42 .

Interestingly, the transcription repressor PLZF, a marker of SSCs, was expressed at higher level in G22 than in G9, suggesting that PLZF expression could reduce the proliferation rate of Spga. Indeed in hematopoietic cells, PLZF induces the G0/G1 arrest by repressing c-myc expression 44 . The other genes up-regulated in G22 (Sohlh1, Taf4b and Tex14) are known to be more expressed in differentiated type A Spga than in undifferentiated Spga 45 46 .

To further evaluate the "stemness" of Spga in testis at 9 dpp, we compared our microarray data with a rat SSC gene list reported by Hamra et al. 23 . In this study 255 SSC marker genes were identified based on the microarray analysis of rat Spga with in vitro defined stem cell properties. We found that genes upregulated in the G9 vs G22 or the S9 vs S22 sets were significantly enriched in the SSC gene list of Hamra et al (Figure 4B, left); namely 61 genes were common to the set of genes up-regulated in G9 versus G22 and could be classified in ten functional categories according to their GO annotations (Figure 4B, right).

A large proportion of these 61 genes encode receptors, transporters, transcription factors and elements of signal transduction. Among the latter, five protein tyrosine phosphatases (PTP), namely Dusp6, Ptpn14, Ptprg, Ptprm, and Ptprd, are known to regulate adhesion, cell growth, differentiation and cell migration. The dual specific phosphatase 6 (dusp6 or MKP-3) encodes a negative feedback regulator of ERK signaling and regulates FGFR signaling during mouse development 47 . Ptpn14 and PTPrm are associated with adherent junctions and dephosphorylate key signaling substrates like beta-catenin 48 and cadherin adhesion molecules 49 . PTPrd encodes the receptor tyrosine phosphatase that regulates actin stress fibers 50 . Ptprg is transiently expressed during ES-derived embryoid body differentiation and is required for HSC lineage commitment 51 . In addition to PTPs genes, we found other genes, such as the thrombin receptor gene (F2r) or the fatty acid elongase (Elovl6), which were previously found to be differentially expressed in neuronal, embryonic and hematopoietic stem cells 52 . Thus, the G9 versus G22 gene set presented here is largely similar to the previously defined set of SSC markers. However, the SSC markers established by Hamra et al. also overlap significantly with the Sertoli-specific gene set at 9 dpp, defined by our more comprehensive approach.

Stem cell niche functions are inherently linked to the communication between the niche and the stem cells. To define stem cell niche elements, we extracted the expression profiles of receptors, ligands, adhesion and extracellular proteins that are specifically up-regulated in Spga and immature Sertoli cells (Figure 5A). Some molecules are already defined as cell surface markers of SSCs, in particular integrin α6 and β1 and cadherin 1 (Cdh1), and have been associated with niche functions in pre-pubertal spermatogonia 53 54 . We found integrins α6 and β1 up-regulated in G9, but Cdh1 was not (Figure 5A). The basement membrane of the seminiferous tubule is composed of laminins, collagens, and fibronectins, ECM components that interact with integrins α6 and β1. The S9 gene set from immature Sertoli cells comprised a number of genes involved in the formation of the ECM of the basement membrane (Table 1 and Figure 5A). Interestingly, the most upregulated genes in S9 were two members of the C-type lectin gene family, pap and pap3 (pancreatitis-associated protein), which bind collagens, fibronectins and laminins. The pap/REG III family promotes cell proliferation and regeneration of pancreatic islets and may protect against apoptosis and oxidative stress 55 .

The complex including the GDNF receptor GFRA1 and the receptor tyrosine kinase c-RET plays an important role in the maintenance of mouse SSC 56 . GFRA1 and GDNF were expressed both in rat Spga and in Sertoli cells, and no significant change was observed between 9 and 22 dpp (Figure 4A). This result is consistent with previous findings that GFRA1 and GDNF are expressed both in germ cells and Sertoli cells of rat testis 39 . However, although these genes are not differentially expressed in pre-pupertal versus pubertal rats, the GDNF signaling pathway is likely to be more active in pre-pubertal testis, since the downstream effector N-myc is strongly up-regulated in G9 (Table 1 and 42 ).

Genes encoding the osteoblast type (Ob-Cdh, cdh11), neuronal type (N-cdh, Cdh2) or PB type (cdh22) cadherins, appeared up-regulated in G9 and S9 (Figure 5A), and Ncam1 and Madcam1 were strongly induced in S9. PB-Cdh (Cdh22) and NCAM1 (Table 1 and Figure 5A) were reported to be highly expressed in neonatal pups and down-regulated during early stages of spermatogenesis 57 58 . Both factors have been shown to interact with gonocytes in promoting SSC cell survival 57 58 .

Pathways driving the development of somatic tissues (e.g. neurogenesis), may also play a role in the SSC niche (see the S9 gene sets in Figure 3). Such pathways control cell migration, proliferation or differentiation. In particular, we found the chemokine receptor CXCR4 up-regulated in G9 and its ligand SDF1/CXCL12 in S9. The SDF1/CXCL12-CXCR4 pathway is important for stem cell homing and mobilization in hematopoiesis 59 . Somatic reticular cells close to vascular endothelial cells secrete a high amount of SDF1/CXCL12 creating a vascular HSCs niche in the bone marrow different from the osteoblast niche 60 .

We found other ligand-receptor couples that have a physiological role in the vascular niche specifically up-regulated in G9 and S9, such as VEGFC growth factor and receptor KDR and Endothelin 1 (End-1) and its receptor Endrb, and BDNF and NTRK or NTF5 (Figure 5A) 61 . Importantly, the receptors were upregulated in the stem cell niche S9, and their corresponding ligands were specifically up-regulated in the stem cell, G9. VEGFC and End-1 are also known to drive angiogenesis in epithelial cancers 62 . For three pairs of ligand/receptor communication we performed real time RT-PCR to quantify the relative cell type and stage-specific expression of each factor. While the ligands vegf, end1, and bdnf were significantly enriched in G9 cells, as compared to Sertoli cells or G22 cells, transcripts coding for the corresponding receptors were significantly and selectively enriched in S9 cells, as compared to all others gene sets (Figure 5B). In summary, the examples of coordinated enrichement of gene transcripts involved in the communication between SSCs and their niche corroborates the relevance of the SSC and niche-specific gene sets defined by this study.

Testicular germ cell tumors (TGCTs) originate from a precursor lesion, known as carcinoma in situ (CIS). CIS can be considered the neoplastic counterpart of PGCs, as they share similar morphological features, gene expression, pattern of genomic imprinting, and markers of pluripotency 28 29 30 .

How PGCs are transformed into CIS and converted to invasive TGCT is unknown. Genomic instability of CIS might be one, but not the only contributing factor 28 63 . It has been argued that changes in the Spga microenvironment or stem cell niche could favor neoplastic transformation, and thus may lead to CIS 30 63 . Several lines of evidence support this hypothesis: first, during the formation of CIS gap junctions are disrupted, the blood-testis barrier is lost and Sertoli cells become de-differentiated 64 65 66 67 ; second, the activity of the canonical SSCs signaling pathway, GDNF/c-ret is increased in CIS 68 .

Our data show that the gene expression profile of pre-pubertal Sertoli cells is consistent with their presumed niche function. Consequently, de-differentiation of Sertoli cells in adult testis may reactivate this niche potential and promote SSC proliferation and neoplastic transformation towards CIS and seminoma.

To test this hypothesis, we compared the Spga (G9; G22) and the Sertoli cell (S9; S22) gene sets with the preferentially expressed gene orthologs of type II TGCTs extracted from the recent literature (Additional file 1, Table S4; 69 70 71 72 73 74 75 76 ).

We compared 1436 ortholog genes, specifically upregulated in testicular cancers, with the gene sets corresponding to two-fold up-regulated transcripts of the comparisons G9 vs G22 or S9 vs S22. Significantly more testis cancer related genes, were present in those gene sets than expected by random distribution (Figure 6A). In contrast, testis cancer related genes were significantly depleted in the set of two-fold up-regulated genes in Spga at 22 dpp versus 9 dpp. These up-regulated genes were then classified, based on their redundancy in the different TGCT studies. Respectively, 27 genes of G9 vs G22 and 30 genes of S9 vs S22 were listed in two or more TGCT studies, 14 of which were up-regulated in G9 and in S9 (Figure 6B).

This latter group included CyclinD2, an early marker of CIS and important for transformation of germ cell tumors 77 78 , and Mycn, a proto-oncogene effector downstream of GNDF/GRFa1 favoring proliferation of SSCs 41 . Interestingly, the overlap of the TGCT-specific genes with S9 comprised a large number of genes encoding secreted proteins (Ccl2, MMP9, MMP12, Plat) or ECM proteins (collagen type I, decorin, fibronectin 1, lumican), which are considered important for the somatic mammalian stem cell niches 79 . Other factors, like the matrix Gla protein or Glypican 3, are known to repress differentiation pathways such as BMP2/4 80 and Hedgehog 81 .

We confirmed the specific and significant enrichment of gene transcripts common to G9 and testicular cancer or to S9 and tumor environment by real time RT-PCR. Both candidates tested, protein phosphatase 1 regulatory 1A (Ppp1r1a) and midkine (mdk), were more expressed in G9 cells than in S9 cells and showed very little expression in G22 and S22 cells (Figure 6B).

These results suggest that the gene expression programs of SSC and their niches, may be reactivated and amplified in testicular cancer. Further studies in TGCTs might validate the candidate genes emerging from this study as potential prognostic markers and/or targets for treatment of early TGCT.

To isolate Spga and Sertoli cells, testes of 45 rats at 9 dpp and 20 rats at 22 dpp were excised and decapsulated. The procedure of cell purification is depicted in Additional file 2, Figure S2 and was adapted from previous publications 84 85 86 . Seminiferous tubules were first isolated from surrounding interstitial cells by collagenase dispase treatement. After three sedimentation steps, the tubules were separated in two fractions. To purify Spga, a fraction of seminiferous tubules was treated with trypsin, then subjected to centrifugal elutriation to collect the diploid cells. The diploid cells were then resuspended in a minimal medium (15 mM Hepes buffered F12/DMEM supplemented with 20 μg/ml gentallin, 20 U/ml nystatin, 1.2 g/L sodium bicarbonate, 10 μg/ml insulin, 10 μg/ml human transferrin, 0.2% serum) and subjected to differential plating for 15 hours. Floating Spga cells were collected by centrifugation and directly frozen in liquid nitrogen. To purify the somatic supporting cells, half the fraction of tubules was treated with collagenase dispase, then, differential plating was directly performed to collect the corresponding adherent Sertoli cells. In the germ cell fractions, 78 ± 2% of the cells were vimentin negative and their viability was 93 ± 1%. In the Sertoli cell fraction, 76.5 ± 3.0% of the cells were stained positive with an antibody against vimentin 87 88 89 ; their viability was 82 ± 5%.

Total RNA was isolated from frozen cells using Trizol (Invitrogen), then, a subsequent purification step was performed using the RNAeasy kit (Qiagen; Hombrechtikon, CH). The purity of the total RNA was analyzed using an Agilent Bioanalyzer RNA Chip (Agilent Inc. Paolo Alto, CA).

Immuno-histochemistry was performed following established procedures 90 91 on sections of rat testis from 9 dpp and 22 dpp using antibodies to PCNA (DAKO).

A small-scale protocol from Affymetrix (High Wycombe, UK) was used to reproducibly amplify and label total RNA. 100 ng total RNA were converted into double-stranded cDNA using a cDNA synthesis kit (Superscript; Invitrogen Corp., Carlsbad, CA) with a special oligo(dT)_24 primer containing a T7 RNA promoter site. After the first cRNA amplification by in vitro transcription using the Ambion MEGAscript T7 kit (Ambion, Austin, TX), 400 ng cRNA were once more reverse transcribed, and biotinylated cRNAs were generated from double-strand cDNAs using an in vitro transcription labeling kit from Affymetrix. For each probe, 20 μg of the second amplification biotinylated cRNA were fragmented and hybridized to Affimetrix rat expression array 230 2.0 (Affymetrix; Santa Clara, CA) following standard protocols. Three independent sets of total RNA were extracted from purified cells prepared on different days on distinct pools of animals. For each condition, the three independent sets of total RNA were purified and used as template for probe generation. These triplicates preparations were performed to define biological variability between the samples. GeneChips were incubated at 45°C for 16 h with biotin-labeled cRNAs probes, and then washed and stained using a streptavidin- phycoerythrin conjugate with antibody amplification as described in the protocol from Affymetrix, using Affymetrix GeneChip Fluidics Station 450. GeneChips were scanned on a GCS3000 scanner (Affymetrix, Santa Clara, CA).

To identify differentially expressed transcripts, pairwise comparison analyses were carried out with Affymetrix GCOS 1.2. Each of the experimental samples (n = 3) was compared with each of the reference samples (n = 3), resulting in nine pairwise comparisons. This approach, which is based on the Mann-Whitney pairwise comparison test, allows the ranking of results by concordance, as well as the calculation of significance (P value) of each identified change in gene expression 92 . Genes for which the concordance in the pairwise comparisons exceeded a threshold (e.g., 60%) were considered to be statistically significant. A 77% cutoff in consistency of change (at least 7 of 9 comparisons were either increased or decreased) was then applied to identify potential dimorphic-regulated genes. Only genes that satisfied the pairwise comparison test and displayed two-fold change in expression were selected for further study. This conservative analytical approach was used to limit the number of false-positives. Regulated genes were organized and visualized using the GeneSpring software (Agilent Inc., Paolo Alto, CA).

Category enrichment or depletion was performed by collecting the observed number of transcripts and respective hypergeometric P-values for each category using Genespring. The expected number of transcripts was calculated based on the total number of annotated RAE230 transcripts in the Gene ontology consortium (5727), the number of annotated transcripts in the corresponding gene ontology category, and the fraction of transcripts in the cluster present in the category. GO category enrichment or depletion were displayed as in Chalmel et al 21 for each respective cluster. A GO category was considered as enriched in a group of transcripts if the P value was <0.001 and the number of observed transcripts in the cluster was >3. P values close to 0 indicate significant enrichment whereas P values close to 1 represent significant depletion.

All gene expression array data are avaible at Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/projects/geo/).