Mice bearing Sox9 ^flox/flox 20 and AMH-Cre transgenes 21 were crossed to obtain AMH-Cre Sox9 ^flox/flox mice which were genotyped as published 18 .

Total testis samples were prepared from C57BL/6 Sox9 ^flox/flox and AMH-Cre Sox9 ^flox/flox mutant animals at 3 (90 dpp) and 5.5 (165 dpp) month after birth. Testes of at least three animals for each time points were decapsulated and combined into duplicate pools using standard laboratory practice.

Total RNA preparation, cRNA target synthesis and raw data production using Mouse Genome 430 2.0 GeneChips (Affymetrix) were essentially done as previously published 22 .

The microarray data were pre-processed and analyzed using the AMEN (Annotation, Mapping, Expression and Network analysis) suite of tools 23 . The data quality was verified by plotting the surface intensity distribution, 3'-5' RNA degradation and log2 signal distribution across samples. Data were normalized using the Robust Multi-Array Average (RMA) method as previously published and the global signal intensities from Sox9 ^flox/flox and mutant AMH-Cre Sox9 ^flox/flox samples were visualized using a Distance Matrix in combination with a dendrogram as published 22 24 .

Probe sets yielding a signal higher than the detection threshold (median of the normalized dataset, cutoff 5.3) and a fold-change ≥2.0 between Sox9 ^flox/flox and AMH-Cre Sox9^flox/flox at 90 and 165 dpp were selected. A LIMMA statistical test (F-value adjusted with the False Discovery Rate method: p ≤ 0.01) was employed to identify significantly differentially expressed probe sets which were subsequently classified into two groups using the k-means algorithm (k = 2).

Enrichment of GO terms and predicted TFBSs were estimated with the Fisher exact probability using a Gaussian Hypergeometric test as previously published 22 . A GO term or a TFBS matrix was considered to be significantly enriched in a group of genes when the FDR-corrected p-value was ≤0.01 and the number of genes bearing this annotation or a TFBS was ≥5.

The network representation was drawn using AMEN 23 . The protein-gene regulation data were downloaded from the TRANSFAC Professional Database release 2010.1 25 and from Transcription Factor Encyclopedia (TFe, accessed May 1^st, 2010).

Transcriptional Start Sites (TSS) extracted from all_mrna.txt and refseqAli.txt UCSC mapping files were localized by associating annotated mouse protein-coding genes (mm9 genome) with their corresponding transcripts as defined in the gene2accession file provided by the NCBI 26 . TFBSs matrices from the TRANSFAC Professional Database release 2010.1 were predicted using the MATCH software 27 with the minSUM_good profile to minimize false negative and false positive predictions. Motif predictions were limited to a region of 1 kb upstream of the Transcriptional Start Site (TSS). Motifs displaying a core similarity score (CSS) and a matrix similarity score (MSS) of ≥0.8 were selected. To further reduce the number of false positives potential motifs had to be conserved 28 . For each prediction a cross-species conservation score was computed by averaging the base-by-base phastCons scores calculated between 30 vertebrates as provided by the UCSC genome browser 29 . Predicted motifs with a conservation score ≥0.8 were selected. Mouse genes whose promoters contained at least one predicted motif were used to calculate its enrichment as compared to the promoters of all annotated genes.

Raw data CEL files corresponding to 8 total testicular samples collected in duplicate at the age of 90 and 165 dpp from C57BL/6 Sox9 ^flox/flox and AMH-Cre Sox9 ^flox/flox mice, respectively, are available via the EBI's ArrayExpress under the accession number E-TABM-528 30 . Normalized data are available for viewing at GermOnline 31 .

It was found earlier that Sertoli-cell specific deletion of Sox9 from E14 onwards had no effect on the fertility of young adults but caused late spermatogenic failure due to progressive degeneration of the seminiferous tubules 18 . To gain insight into which Sox9-dependent processes may contribute to this phenotype we compared the testicular mRNA profiles of phenotypically normal Sox9 ^flox/flox mice and mutant AMH-Cre Sox9 ^flox/flox mice. As a control, samples from both backgrounds were first taken at 90 dpp prior to the onset of the phenotype where testes from Sox9 ^flox/flox and AMH-Cre Sox9 ^flox/flox mice are morphologically and histologically indistinguishable. A second set of samples was prepared at 165 dpp which is approximately two weeks after the point where pathological changes first become histologically apparent, but well before testicular architecture breaks down completely in the AMH-Cre Sox9 ^flox/flox background 18 . Affymetrix Mouse Genome 430 2.0 high-density oligonucleotide GeneChips are extremely robust tools that yield highly reproducible data 32 33 and complex mammalian samples are amenable to reliable analysis when examined in duplicate 22 24 34 . We therefore used two independent pooled samples of 90 dpp and 165 dpp testes from Sox9 ^flox/flox and AMH-Cre Sox9 ^flox/flox mice to prepare total- and cRNA of uniformly high quality (Additional file 1 Figure S1A and B). GeneChip hybridization patterns were normal in all cases (Additional file 1 Figure S1C) and the signals displayed the expected RNA degradation and intensity distribution profiles (Additional file 1 Figure S1D and E).

We first identified 14,106 and 14,438 genes, respectively, for which signals above the threshold level of detection (5.3 log2 units corresponding to the median intensity) were obtained at 90 dpp and 165 dpp (Additional File 2 Table S1). We next selected genes significantly differentially expressed (2-fold change, LIMMA test with FDR adjusted p-value < 0.01) and used the k-means clustering algorithm (n = 2) to group them into 16 genes showing stronger signals and 7 genes showing weaker signals in the mutant samples at 90 dpp (Figure 1A). The extremely low number of gene for which variations were found is in keeping with the lack of any detectable phenotype at the 90 dpp control time point as compared to the Sox9 ^flox/flox control. We then repeated the procedure with the 165 dpp samples and identified 1119 genes for which we measured increased signals and 113 genes associated with decreased signals (Figure 2A). This result is consistent with the morphological changes observed in testicular tissue after the onset of dysgenesis at the age of 5 months 18 .

Among the loci for which signals are stronger in the mutant, only two were found exclusively at 90 dpp, 14 were observed in both samples and 1105 were detected only at 165 dpp. Similarly, three, four and 109 genes were found to be weaker only at 90 dpp, in both samples or only at 165 dpp, respectively (Figure 1B and Additional File 2 Table S1). These results show, as expected, that a substantial effect on RNA concentrations is detectable only at 165 dpp at the onset of testicular failure.

We integrated the expression signals obtained with total testis samples from Sox9 ^flox/flox and AMH-Cre Sox9 ^flox/flox mice with our data from enriched wild-type Sertoli cells, spermatogonia, spermatocytes, spermatids and total testis samples 22 as well as ovary samples obtained from the GEO public array data repository (Methods) 35 . We note that the experimental design allows for determining overall mRNA concentration changes between replicate samples which may or may not be due to transcriptional effects. We refer to statistically significant changes in expression values between wild-type and mutant testes alternatively as stronger (increasing) and weaker (decreasing) signals which implies that more or less mRNA, respectively, is present in the given sample.

Figure 2A shows a heatmap summarizing expression signals for two genes showing stronger signals only at 90 dpp while no increase was found at 165 dpp (Class 1; Fam181b, Igf1r), 14 genes showing stronger signals at both time points (Class 2; Aqp5, C3, Ccdc80, Ildr2, Hoxd10, Lrg1, Serpina3n, Spon1, Sult1e1, Thrsp, Timp1, Tnfrsf12a, Tspan8, Wwtr1) and 1105 loci which pass our selection criteria (>2-fold) for a significant signal increase only at 165 dpp (Class 3; see Additional File 2 Table S1 which contains find and filtering functions to call up the complete expression data set for individual genes or groups of loci; see also GermOnline for expression data freely available for all genes represented on the Mouse Genome 430 2.0 GeneChip; 36 ). For genes in Class 3 we observe the strongest signals in purified somatic wild-type Sertoli cells and spermatogonia as well as ovary (Figure 2A).

Figure 2B depicts the signals for three genes showing weaker signals only at 90 dpp (Class 4; Kctd14, Schip1, LOC100044139), four cases where signals were decreased in both samples (Class 5; Clic6, Diras2, Sox6, Sox9) and 109 loci whose transcript concentrations decrease in the mutant at 165 dpp (Class 6; Additional File 2 Table S1). As opposed to the genes shown in panel A we typically find the loci to display the strongest signals in purified pachytene spermatocytes and round spermatids; these results are confirmed by strong signals in testes and mostly weak ones in ovary (Figure 2B).

We next explored the biological processes associated with groups of genes showing signal changes among the sample set by determining if these groups were statistically significantly enriched for genes bearing specific Gene Ontology (GO) terms (Methods) (Figure 2C). While no GO terms were found to be enriched in the group showing decreased signals, we identified several relevant processes among the genes for which stronger signals are observed in the mutant. First, as anticipated, we found Reproduction (GO:0000003, 62 observed/35 expected by chance, p-value: 3.6 × 10^-5), Male sex differentiation (0046661, 12/4, 4.5 × 10^-4), and Female sex differentiation (0046660, 11/4, 3.8 × 10^-3). A large group of genes was annotated as being involved in Organ development (0048513, 169/90, 1.3 × 10^-15); consistently, we find enrichment of terms relevant for bone formation and the reproductive tract such as Skeletal system development (0001501, 37/16, 4.8 × 10^-6) and development of the Urogenital system (0001655, 25/9, 2.3 × 10^-5), Prostate gland (0030850, 8/2, 1.5 × 10^-3), and Gonad (0008406, 16/6, 3.7 × 10^-4).

At the cellular level we found two terms particularly pertinent for Sertoli cell- germ cell interactions as spermatogenesis progresses: Cell migration (0016477, 37/14, 3.9 × 10^-7), and Cell adhesion (0007155, 56/29, 8 × 10^-6) while Phagocytosis (0006909, 11/3, 5 × 10^-5) and Apoptosis (0006915, 42/26, 3.9 × 10^-3) reveal extensive absorption of cellular material and programmed cell death typical for Sertoli cells and meiotic germ cells, respectively. As far as biochemical processes important for reproduction are concerned we identified metabolism of Prostaglandin (0006693, 7/1, 7.6 × 10^-6), and Retinoid (0001523, 9/2, 1.7 × 10^-5).

Interestingly, the group of genes showing increased signals includes some involved in Signaling (0023033, 102/68, 1.6 × 10^-4) and notably Cell surface receptor linked signaling (0007166, 90/64, 2.9 × 10^-3) such as the MAPKKK cascade (0000165, 16/7, 7.5 × 10^-3). These signal transduction pathways play roles in similarly enriched processes such as Response to stress (0006950, 154/72, 1.6 × 10^-19) and Inflammatory-, (0006954, 38/13, 6.2 × 10^-2) as well as Immune response (0006955, 53/23, 1.1 × 10^-7).

These results indicate that reproducible mRNA concentration changes occur in AMH-Cre Sox9 ^flox/flox samples at the onset of testicular breakdown, and that the group of genes for which we find a signal increase contains more genes involved in male reproductive functions, female sex differentiation, stress response and inflammatory processes than would be expected to occur by chance. Lists of genes bearing specific GO annotations as described are available via Additional File 2 Table S1.

To gain further insight into the regulatory processes that contribute to the RNA signature triggered by the absence of Sox9 in Sertoli cells, we first identified 202 genes for which mRNA was detectable only in mutant testes but not the normal controls. We found that the promoters of these genes - for which we find strong signals in purified Sertoli cells, spermatogonia and ovary but not in testis - were statistically significantly enriched for conserved target sites of Sox9 (Figure 3A-C, Methods). Moreover, we identified the motifs of regulators known or thought to be involved in Wnt signaling (Tcf7l2, Zbtb33/Kaiso), embryonic and post-natal development (Maf, Nfia, Pou5f1, Runx2, Tead1, Tef), steroid hormone signaling (Esr1, Nr2f2), stress response (Nfe2l2, Ppara) and immune function (Bhlhe40, Cebpa, Cebpb, Elf1, Irf8, Jun, Stat6) (Figure 3E).

We next selected 23 genes for which mRNA was detected in Sertoli cells and that showed weaker signals in AMH-Cre Sox9 ^flox/flox samples to determine if they were potential direct Sox9 targets. The majority of them showed the strongest signals in purified meiotic and post-meiotic germ-cells and among those for which the highest signals were observed in purified Sertoli cells only five (Mpb, Tcf7l2, Acaca, Ank2 and Apbb1ip) were below the threshold level of detection in the mutant and had at least one Sox9 binding site in their upstream promoter regions (Figure 3D). The only enriched motif identified in the group showing decreased signals in the mutant was the target site bound by the Androgen Receptor (Ar) which is involved in testosterone signaling (Figure 3E).

To better understand the level of interconnectivity between transcription factors and target genes that responded to the Sox9 deletion we integrated the output of our analysis with published transcriptional regulatory data from TRANSFAC and TFe (Methods). We identified one large network including 14 transcription factors showing stronger signals in the AMH-Cre Sox9 ^flox/flox testes and five showing weaker ones (including Sox9 itself for which, as expected, no transcript was detected) (Figure 4A). Furthermore, we found a small complex (Irf8 and B2m/Cybb/H2-D1 targets) and two binary interactions (Zbtb33/S100a4 and Hoxd9/Hoxd10) (Figure 4B). The interactions typically reveal concordant patterns for transcription factors and their target genes in the group of genes showing increased signals in the mutant. Factors such as Cebpb, Epas1, Irf8, and Jun are involved in controlling immunological processes often via regulating cytokine gene expression. The network data also suggest a certain level of coordination between immune-, and stress response, development (Cebpa, Fosl2, Hoxd9, Hoxd10, Maf, Nfe2l2, Nfia, Ppara) and gene expression controlled by steroid hormone receptors (Ar, Esr1, Nr2f2 and Rxra).

In addition to Sox9, we identified four transcription factors which displayed weaker signals in AMH-Cre Sox9 ^flox/flox testes in spite of the fact that their target genes show a signal increase: Sox6 (development of the central nervous system, chondrogenesis and maintenance of cardiac and skeletal muscle cells), Pou5fl (embryonic stem cell pluripotency), Pou2f2 (B-cell maturation) and Runx2 (chondrocyte-, and osteoblast differentiation).