Identification of a gene signature of a pre-transformation process by senescence evasion in normal human epidermal keratinocytes

To establish a gene signature specific of the earliest steps of tumorigenesis, we performed a transcriptomic analysis of post-senescent emergent NHEKs in comparison with senescent and young ones. We harvested RNAs from NHEK cultures at exponential growth phase (Y), senescence plateau (S) or post-senescence emergence (E) (Figure 1A), reverse transcribed the RNAs and then hybridized them to RNG-MRC human set 25K microarrays as described in the methods section. Comparison between E vs Y and E vs S were performed with three biological replicates (from the same donor) and using a dye-swap strategy to limit color bias. The R package limma was used to test for differential expression. Variances were modelled using an empirical Bayesian approach in order to increase the sensitivity of the experiment and thus mitigate the low number of replicates. The visual observation of the MAPlots and the shape of p-values histograms (Additional file 1: Figure S1A and B) validate the statistical modelling, warranting further analyses and interpretation of differentially expressed genes. For delineating the differentially expressed genes, we applied as a first filter a range of False Discovery Rate (FDR) from 0.1 to 1% (Figure 1B). We retained the most stringent FDR value of 0.1%. As a second filter, we applied log (2) fold change (E/Y or E/S) ≥ 1.1. Hence, 417 genes differentially expressed between E vs Y and 375 genes between E vs S were selected. A Venn diagram analysis indicated that 286 genes were common to the two conditions. These 286 genes were all either up-regulated both in E vs Y and E vs S or down-regulated both in E vs Y and E vs S, suggesting that they represent the PSNE driver genes (Figure 1C). This gene set, hereafter called the PSNE signature, is presented in Additional file 2: Table S1. Validation of the array data was carried out on a subset of up- and down-regulated genes by quantitative real-time PCR (Additional file 1: Figure S2), hence validating the overall signature. The PSNE signature was then the basis of bioinformatics analyses. The flow chart of these analyses is depicted in Figure 2.

In a first step (Figure 2), we wanted to evaluate whether the PSNE signature was relevant to cancerogenesis. To this end, we performed an analysis of the over-represented genes associated to diseases and disorders using Ingenuity software (Ingenuity Pathway Analysis (IPA), http://​www.​ingenuity.​com). This analysis revealed that 36% of the PSNE genes are annotated in at least one cancer type (Figure 3A) and this was expanded to 50.7% (145/286 genes) by performing extensive literature mining. These 145 genes are highlighted in grey in Additional file 2: Table S1. It is worth to note that 18% of the PSNE genes are involved in psoriasis, a benign skin disease characterized by hyperproliferation of keratinocytes (Figure 3A), directly showing that PSNE can be used as a model for studying the initial steps of tumorigenesis. The comparison of the PSNE signature with those of common cancer types showed that the genes of the PSNE signature annotated in cancer (50.7%) are preferentially associated with the most frequent carcinomas, including colorectal, breast, skin and female genital organs carcinomas (Figure 3B), with a range of 4–54 genes (1.4-18.9%) within each cancer type category (Figure 3B). Finally, to determine whether the PSNE signature is relevant of early steps of carcinogenesis, and especially of early steps of non-melanoma skin carcinogenesis (NMSC), we compared by literature mining the PSNE signature with array-based transcriptional profiles of actinic keratosis (AK) and arsenic-induced cutaneous hyperplastic lesions, two premalignant state related models of NMSCs [14‐19]. The PSNE gene set overlaps with a total of 28 genes belonging to one or more profiles of the aforesaid premalignant models (Figure 3C).

In a second step, we searched for new pathways which could be involved in the pre-malignant steps. To this goal, the PSNE signature was analyzed according to over-represented Gene Ontology (GO) terms within the “molecular function” and “biological process” groups. The top five of this ontology analysis is listed in Table 1. The “molecular function” GO analysis revealed that the most over-represented GO category is the oxidoreductase activity category (GO: 0016491) in which almost all genes are up-regulated. The four other best statistically over-represented categories comprise much less genes than in the oxidoreductase activity category. These are the phosphoric diester hydrolase activity category (GO: 0008081), the transaminase activity category (GO: 0008483), the neutral amino-acid transporter activity category (GO: 0015175) and the unfolded protein binding category (GO: 0051082) (Table 1 and Additional file 2: Table S1). The “biological process” GO analysis emphasized numerous genes involved in response to stress, and to a less extent response to chemical stimulus, external stimulus, cellular ketone metabolism and epidermis development (Table 1). Next, the PSNE signature was analyzed by functional annotation using the Kyoto Encyclopedia of Genes and Genome (KEGG), a compendium of genes annotated and organized by signalling pathway [20]. The most significant functional identified pathway concerns the metabolism of xenobiotics by cytochrome P450 (Table 2). It includes AKR1C3, AKR1C2, GSTA4, ADH5, GSTP1, MGST2, ALDH3A1 that are all up-regulated in E vs Y and E vs S. Interestingly, all these genes belong to the GO oxidoreductase activity category (Table 1 and group 3 in Additional file 2: Table S1) and AKR1C2, AKR1C3 in particular, intersected with NMSC pre-malignant gene sets (Figure 3C). The second most significant identified pathway is that of aminoacyl-tRNA biosynthesis including the IARS, WARS, AARS, GARS, MARS genes that are all down-regulated in E vs Y and E vs S (Table 2). The other pathways identified within the top five KEGG pathways are Glycine, serine and threonine metabolism, Glycolysis/Gluconeogenesis and Prion diseases. The same kind of analysis was performed using IPA Canonical pathways. Three amongst the top five pathways that were highlighted by the analysis were the same as those highlighted by the KEGG analysis, namely with the best p value the Metabolism of xenobiotics by cytochrome p450 including ADH5, AKR1C2, AKR1C3, MGST2, GSTA4, CYP4B1, ALDH3A1 and GSTP1, in second Glycine, serine and threonine metabolism and in the forth position Aminoacyl-tRNA biosynthesis. The other two are Interferon signaling and Aldosterone signaling in epithelial cells (Table 3).

Taken together, these results suggest that genes encoding proteins carrying oxidoreductase activity and involved in the metabolism of xenobiotics may be crucial for the early steps of transformation in the PSNE cellular model.

We then investigated by genetics and pharmacological approaches whether the AKR1C genes activated at senescence in primary keratinocytes may function as oncogenes, i.e. promote senescence evasion in the form of pre-transformed cells. We first verified by RT-qPCR the levels of AKR1C2, −C3 mRNAs in senescent NHEKs and emerging cells compared with young NHEKs. From RT-qPCR analyses performed with extracts from three different donors, we observed an increase in AKR1C2, −C3 levels at senescence followed by a much stronger increase at emergence (Figure 4A), hence confirming the microarray results. We next assessed whether the up-regulated transcription of AKR1C2, −C3 genes resulted in increased AKR1C2, −C3 protein levels. Western Blot analysis revealed strongly increased protein levels at senescence, further increased at emergence (Figure 4B).

To establish whether AKR1C2 and –C3 play a role in PSNE, we examined whether decreasing their expression at senescence could affect the emergence frequency. To this goal, individual siRNAs directed against AKR1C2 or -C3 were transfected in senescent cells. As shown in Figure 5A and B, AKR1Cs siRNA transfection of senescent NHEKs resulted in strong decreased of AKR1C2, −C3 at both mRNA and protein levels. We first examined whether AKR1Cs siRNAs affected the senescent state itself. The results show that four days after transfection neither the population doubling numbers nor the percentage of SA-β-Gal positive cells were affected (Additional file 1: Figures S3A and S3B). We then monitored these siRNA-transfected senescent NHEKs for the emergence frequency. For that, siRNA-transfected senescent NHEKs were plated at low density, and, once the emergence has occurred about 7 days after transfection, the clones of emergent cells were manually counted and the emergence frequency calculated as the number of emergent clones on the number of initially plated senescent cells. The results show that knocking-down AKR1Cs diminished by a twofold factor the emergence frequency (Figure 5C and 5D). We also examined whether AKR1Cs could be involved in the proliferation of the emergent NHEKs. For that, emergent NHEKs were transfected with the siRNAs against AKR1C2 or –C3 or the control siRNAs and their proliferation was followed during 7 days. The results show that AKR1C2, −C3 knock-down inhibited proliferation of emerging cells from four days after transfection (Figure 5E). In contrast AKR1C2, −C3 silencing did not impact on cell proliferation in young NHEKs (Figure 5E). Taken together, these results indicate that AKR1Cs are involved both in the process of emergence from senescence and in the proliferation of the neo-formed transformed cells.

We next wanted to determine whether AKR1C2, −C3 activity is required for post-senescence emergence. To this goal, we treated senescent NHEKs with the AKR1C2 inhibitor ursodeoxycholic acid [21] and AKR1C3 inhibitor 3-(4-(Trifluoromethyl) phenylamino) benzoic acid [22]. We first checked the efficacy of the inhibitors using coumberone, a non-fluorescent probe reporter, which is specifically metabolized by the AKR1C enzymes into fluorescent coumberol (see the methods section). We quantitatively measured coumberone metabolism in presence or absence of ursodeoxycholic acid or 3-(4-(Trifluoromethyl) phenylamino) benzoic acid in senescent NHEKs. We show that addition of both AKR1C inhibitors decreased coumberone metabolism, validating the inhibitory effects of the compounds (Figure 6A and B). We then determined the effects of AKR1C2, −C3 activity inhibition on senescence and emergence as above. Population doublings and emergence frequency were then determined as above. Both inhibitors did not affect senescence growth arrest, whatever the concentration used (Additional file 1: Figures S4A and B). Ursodeoxycholic acid significantly decreased the emergence frequency at the same rate whatever the dose (Figure 6C), while the AKR1C3 inhibitor was almost inefficient in impacting emergence (Figure 6D).

To further support a role of AKR1C2, −C3 in pre-transformation, we investigated whether overexpressing AKR1C2, C3 in senescent NHEKs could also increase the emergence frequency. We used recombinant qdenovirus particles encoding V5-tagged AKR1C2, V5-tagged AKR1C3, or, as control, GFP. The overexpression of AKR1C2, −C3 were checked by western Blot using an anti-V5, 48h after infection (Figure 7A). We next followed the emergence frequency in AKR1C2, −C3 expressing cells. We showed that AKR1C2 but not AKR1C3 overexpression increased significantly the emergence frequency (Figure 7B).

In conclusion, these results indicate that the early steps of transformation in the PSNE model, including emergence from senescence and proliferation of the just emerged neoplastic cells are dependent, at least in part, on the expression and activity of AKR1C2 and to a less extent AKR1C3.

Normal human epidermal keratinocytes (NHEKs) were purchased from Clonetics (CC-2501) and Promocell. We used cells from 3 different donors of different race and age (referred as 4F0315, 2F1958, and K1MC). Cells were obtained anonymously and informed consent of each skin donor was obtained by the supplier. Cells were grown at 37°C in an atmosphere of 5% CO2 in a KGM-2 BulletKit medium consisting of modified MCBD153 with 0,15 mmol/L calcium, supplemented with bovine pituitary extract, epidermal growth factor, insulin, hydrocortisone, transferrin, and epinephrin (Clonetics). Such a serum-free low-calcium medium was shown to minimize keratinocyte terminal differentiation [29]. In all experiments, cells were seeded at 3500 cells/cm2 and always splitted at 70% confluence. The number of population doublings (PD) was calculated at each passage by means of the following equation: PD = log (number of collected cells/number of plated cells)/log2. Ursodeoxycholic acid and 3-(4-(Trifluoromethyl) phenylamino) benzoic acid were obtained from Sigma and Calbiochem, and diluted in ethanol and DMSO respectively.

After siRNA transfection, adenoviral infection or chemical treatment, senescent NHEKs were seeded into 10 cm dishes at the limit density for emergence (10,000 cells per dish). A few days later, the dishes were fixed and stained by crystal violet, the emergent clones were counted and the emergence frequency was calculated as the ratio of the number of clones on the number of seeded senescent cells [12].

Total RNAs were extracted using RNeasy mini-columns (QIAGEN). One μg of RNA was reverse-transcribed using random hexamers, Superscript III and dNTPs (Invitrogen) in a final volume of 20 μl according to manufacturer’s instructions. Quantitative real-time PCR reactions were performed using the Mx3005P Real-time PCR system (Stratagen). Primers used were designed with qPrimerDepot (http://​primerdepot.​nci.​nih.​gov/​) and are listed in Additional file 3: Table S2. PCR products were measured by SYBR Green fluorescence (SYBR Green Master Mix, Applied Biosystems). Experiments were performed in triplicates for each data point. Results were analyzed with the MxPro software (Agilent). The expression of each gene was normalized to 18S gene, and fold expression relative to the control is shown.

Invalidation experiments for AKR1C2 and AKR1C3 were performed using pools of siRNAs from Dharmacon (ON TARGET Plus Smart pool, Dharmacon, USA). A non-targeting siRNA pool (Dharmacon) was used as control. Young, Senescent or Emergent NHEKs were plated at 70,000 cells per well in six-well plates and were transfected by using the RNAiMAX Lipofectamine reagent (Invitrogen Corp., Carlsbad, CA, USA) or by using PrimeFect siRNA Transfection Reagent (Lonza) according to manufacturer’s instructions.

Equal numbers of young, senescent or emergent NHEKs were lysed in Laemmli buffer 2× containing SDS 4%, Glycerol 20%, Tris–HCl 50 mM pH 6.8, bromophenol blue 0.02% and 2-Mercaptoethanol 5%. Samples were denatured by heating at 95°C for 5 min. Proteins were resolved by SDS-PAGE and transferred to nitrocellulose membranes (Hybond-C extra, Amersham). Primary antibodies used against AKR1C2 and AKR1C3 were from Abcam and anti-human GAPDH mouse monoclonal antibody was from Chemicon International, anti-V5 was from Life Technologies. Secondary antibodies used were peroxidase-conjugated (Jackson ImmunoResearch Laboratories). Peroxidase activity was revealed using ECL (enhanced chemiluminescence) or ECL advanced kit (Amersham Biosciences).

Adenoviral particles expressing AKR1C2-V5 (#SL170932) or AKR1C3-V5 (#SL177696) were purchased from SignaGen (Gaithersburg, MD, USA). AdGFP construction was previously described [35]. In all experiments, cells were infected at an input of 50 viral infectious particles/cells by adding virus stocks directly in the culture medium. Four hours later, the medium was replaced by fresh medium.

The enzyme activity of AKR1C was assayed according to the procedure described by Halim et al. [36], with slight modifications. Briefly, cells were added to 96-well plates at a density of 6000 cells/well in phenol red-free media. Concentrations of inhibitors (as indicated) were added to the plate and incubated for 1 h, prior to addition of coumberone at concentration of 10 μM. Final volumes were 200 μl/well. The plates were returned to the incubator and fluorescence intensity was read 7 h after coumberone addition with excitation at 385 nm and emission at 510 nm on a FluoStar Optima platereader (BMG Labtech, Cary, NC). Results were expressed as a percentage +/− s.d. and normalized to control.

Human cells used in this study provide from people whose informed consent was obtained by the cell suppliers (Clonetics, Promocell). Cells were obtained anonymously. No ethics approval was necessary for the experiments performed therein.