|Home | About | Journals | Submit | Contact Us | Français|
DNA methylation is an important epigenetic mechanism in prostate cancer (PCa) progression. Given the role of even-skipped homeobox 1 (EVX1) in the regulation of multiple genes during embryogenesis, we postulated that EVX1 methylation is altered in PCa progression.
Bisulphite sequencing and quantitative MethyLight were used to assess methylation in human prostate epithelial cells, four PCa cell lines, liver, lung, spleen, kidney, 35 paired tumour and tumour-associated benign tissues, and 11 normal prostate tissues. Prostate cancer cell lines were treated with 5-azacytidine (AzaC) or trichostatin A (TSA), and expression of EVX1 transcript and variants was assessed by qPCR. Hypermethylation was compared with clinicopathological features in a validation set of 58 patients using microarray.
Even-skipped homeobox 1 hypermethylation was observed in all four PCa cell lines and 57% of tumours. High-grade tumours exhibited increased methylation compared with intermediate-grade tumours. Even-skipped homeobox 1 expression was induced in PCa cell lines after treatment with AzaC or TSA. In the validation set, 83% of tumours were hypermethylated and hypermethylation was associated with worse recurrence-free survival.
In this first evaluation of EVX1 methylation in human cancer, EVX1 is one of the most commonly hypermethylated genes observed in PCa and predicted treatment failure in moderate risk patients.
The predominant tools for predicting prostate cancer (PCa) outcomes consist of clinical and pathologic nomograms incorporating PSA, clinical stage, Gleason grade (GG) and biopsy results (Kattan et al, 1999). The biological behaviour of PCa, especially intermediate-grade cancer (GG 5-7), exhibits a wide range of aggressive behaviour (Stark et al, 2009). Clearly, identifying new biological markers that allow improved detection and prognosis of PCa is critical. Prostate cancer, like many other malignancies, arises through an interplay of genetic and epigenetic alterations (Schulz and Hoffmann, 2009). DNA alterations at CpGs occur early in PCa development and represent a promising area for screening and prognostic marker development (Bastian et al, 2004).
There are increasing number of reports detailing the epigenetic gene modifications in PCa pathogenesis (Yegnasubramanian et al, 2004; Hanson et al, 2006; Henrique et al, 2006; Aitchison et al, 2007; Kawamoto et al, 2007). The most well-known target, GSTP1, is subject to epigenetic silencing in PCa, but its clinical applications are currently limited by its low sensitivity (Ellinger et al, 2008). In addition to cancer detection, epigenetic markers have the potential to predict cancer progression for selected diseases (Hartmann et al, 2009; Tanaka et al, 2011; Sun et al, 2012). In PCa, altered methylation at specific loci has been reported to correlate with clinicopathological features, but its role in predicting outcomes, especially for intermediate-grade, remains unknown (Sathyanarayana et al, 2003; Enokida et al, 2004). In addition to improving PCa detection, diagnosis, staging and risk stratification, epigenetics provides an opportunity to expand our knowledge of cancer biology (Nelson et al, 2009).
Recent studies have found aberrant methylation of embryonic developmental genes in cancer, suggesting a link between the cancer epigenome and developmental programs (Baylin and Jones, 2011). The Wnt signalling pathway, which is critical for normal cell growth and differentiation, was found to have a role in the development of urological cancers (Costa et al, 2010). Even-skipped homeobox 1 (EVX1) is a developmental homeobox domain protein located at 7p15 that is closely related to the Drosophila gene even-skipped (eve). Even-skipped homeobox 1 protein is expressed in the visceral endoderm and is required for normal blastocyst development and differentiation (Spyropoulos and Capecchi, 1994). Its transcripts are detected in undifferentiated F9 stem cells and disappear upon differentiation into parietal endoderm cells after the administration of retinoic acid (Faiella et al, 1991). Most notably, EVX1 functions as a potent repressor of gene transcription at numerous promoters in mammalian cells, suggesting a central role in the maintenance of normal gene expression (Briata et al, 1995). Currently, the role of EVX1 in cancer development is completely unknown. Given its important role in embryogenesis as a widespread gene repressor, we hypothesise that aberrant EVX1 methylation may have a role in prostate tumourigenesis.
In a recent study, we identified aberrant downstream hypermethylation of the developmental gene EVX1 on genome-wide methylation arrays to be associated with an epigenetic field defect in PCa (Yang et al, submitted for publication). In the present study, we extensively examine EVX1 methylation profiles in PCa within the promoter, exon 1 and intron 1 using bisulphite sequencing. Here, we provide evidence that EVX1 is frequently methylated in PCa and that its expression is regulated by DNA methylation and histone modification. Furthermore, EVX1 hypermethylation is an independent predictor of PSA recurrence following prostatectomy in intermediate-grade cancer.
LNCaP, PPC-1, DU145 and PC3 cells were purchased from the American Type Culture Collection and cultured in DMEM (Life Technologies, Inc., Grand Island, NY, USA) with 10% fetal bovine serum and 1% penicillin-streptomycin. Human prostate epithelial cells (HPEC) were established on collagen-coated dishes in Ham's F-12 supplemented medium containing 1% fetal bovine serum as previously described (Schwarze et al, 2001).
A total of 11 normal prostate specimens, termed ‘non-tumour associated' (NTA) tissue, were obtained from either organ donation or cystoprostatectomy and OCT-embedded. Histological evaluation ruled out the presence of cancer cells. We also obtained 35 prostate specimens from patients with PCa who underwent radical prostatectomy. Microdissection was performed under the direction of a genitourinary pathologist to obtain tumour and benign ‘tumour-associated' (TA) tissue apart from the tumour foci as previously described (Bhusari et al, 2011). In total, 19 intermediate-grade and 16 high-grade paired specimens along with 11 controls were obtained for DNA methylation analysis. Tissues were obtained in accordance with institutional review board policies.
DNA was isolated from prostate tissue samples and PCa cell lines using DNeasy Blood and Tissue Kit (QIAGEN, Valencia, CA, USA). Sodium bisulphite conversion was performed using Epitect Bisulfite Kit (QIAGEN).
Three sets of primers were designed to amplify bisulphite-converted DNA spanning 25–450, 532–1018 and 1098–1484bp downstream of the EVX1 transcriptional start site (TSS), using Methyl Primer Express Software (Applied Biosystems, Foster City, CA, USA). The following primer sets were used to amplify the three regions specified above: 5′-AGTGTGTGTGTTTTTTTGTTTG-3′ and 5′-AAAACCCCAAATAAATTCCTAA-3′ 5′-TTGGTATTTTGGTTGGTAAGAG-3′ and 5′-CAACCATCCTACCCTAAAAAA-3′ and 5′-GAGGGTGATAGTAGTGTTTTGA-3′ and 5′-AACAAAAACAATTAAACCATT-3′, respectively. Two separate HPEC DNA samples from individual patients (patients 1 and 2) and pooled HPEC DNA samples from multiple patients (patients 3–5) were bisulphite converted. PCR reactions were performed to amplify bisulphite-converted DNA using HotStarTaq (QIAGEN). The following cycling conditions were used: 95°C for 15min; 50 cycles of 95°C for 15s, 55°C for 30s, 72° for 90s. PCR products were cloned into pCR2.1 using TOPO TA Cloning Kit (Invitrogen, Carlsbad, CA, USA). We collected 10–20 colonies for each of the three regions for each cell type being assessed. Plasmid isolation was performed using PureYield Plasmid Miniprep System (PROMEGA, Fitchburg, WI, USA) kit and sequenced. A total of 4–7 sequencing results per region per cell type were obtained and aligned using BiQ Analyzer (Bock et al, 2005).
In total, 96 CpGs within 2 CpG islands were analysed spanning a total of 1300bp. The entire set of CpGs was compared using two-way analysis of variance (ANOVA) with Bonferroni post hoc testing for multiple comparisons. Fisher's statistical analyses were performed using GraphPad Prism Version 5.03 (La Jolla, CA, USA).
As we demonstrated with bisulphite sequencing that PCa cell lines and HPEC were differentially methylated, we designed MethyLight probes 460bp downstream of the TSS in a region where PCa cell lines were highly methylated and HPEC were highly unmethylated. Probes sequences were designed using Beacon Designer (Premier Biosoft, Palo Alto, CA, USA).
The methylation-specific probe was tagged with 6-FAM and the unmethylated-specific probe was tagged with HEX. The following methylated-specific and unmethylated-specific probe sequences were synthesised: 5′-6-FAM-CCCTAAATCTCCAACGACCGAACGAATCT-3′ and 5′-HEX-CCCTAAATCTCCAACAACCAAACAAATCTT-3′, respectively (Integrated DNA Technologies, Coralville, IA, USA). Forward and reverse primers that do not contain any CpGs were designed: 5′-GTTTTATAAGGTGATTTTAGTTT-3′ and 5′-ACCCCAAATAAATTCCTAAC-3′, respectively. The primer annealing temperatures were optimised and 56°C was ideal. With the exception of our optimised annealing temperature, the PCR reaction was carried out in duplicate using the Epitect MethyLight PCR Kit (QIAGEN) according to the manufacturer's instructions. Real-time amplification and measurements of threshold cycles (Ct) for 6-FAM and HEX was performed using the CFX96 Real-Time PCR Detection System (BioRad, Hercules, CA, USA). The % methylation was calculated using the formula: , where CtCG and CtTG are the measured Ct value from the real-time detection of 6-FAM and HEX, respectively (Eads et al, 2000). We used a t-test to compare groups. We generated a receiver operating characteristic (ROC) curve to determine the optimal cutoff value to define hypermethylation, which maximised the sum of sensitivity and specificity in predicting cancer in this cohort. This cutoff was then used to determine what percentage of tumour specimens were hypermethylated compared with the normal specimens. Receiver operating characteristic analysis was carried out using Graphpad Prism Version 5.03.
Experiments were performed for assay validation. DNA mixing of methylated LNCaP DNA and unmethylated kidney DNA was performed and was highly linear across a wide range of methylation (Supplemental Figure 1). Detection of methylation was highly accurate with as low as 40pg of DNA template. Furthermore, the % methylation did not change with differing amounts of template DNA or between multiple experiments, demonstrating high efficiency and reproducibility of the MethyLight assay.
After performing a dose-ranging study, PCa cell lines were treated with 2.5μ of the demethylating agent AzaC (Sigma-Aldrich, St Louis, MO, USA) for 48h or 25n of the histone deacetylase TSA (Sigma-Aldrich) for 16h. Cells were redosed with AzaC after the first 24h of treatment. 5-Azacytidine stock solution was prepared in sterile water and TSA stock solution was prepared in DMSO, stored in aliquots at −20°C.
Total RNA was isolated from drug-treated cells using the RNeasy RNA isolation kit (QIAGEN) and treated with DNase I (QIAGEN). The Omnicript Reverse Transcript Kit (QIAGEN) was used to synthesise cDNA using 200ng of RNA per sample. Even-skipped homeobox 1 mRNA transcripts and variants were identified from expressed sequence tags (EST) found in the UCSC genome database (Dreszer et al, 2011). Primers were designed that spanned exons 1 and 2 (primer set 1) and exons 2 and 3 (primer set 2) of the forward expressing mRNA transcripts RefSeq NM_001989, EST AK314155 and EST-DC36605. We also designed a set of primers (primer set 3) that spanned exons 2 and 3 of EST-DB260732 and exons 1 and 2 of EST-DB455833, both of which are reverse-transcribing mRNA variants (Figure 4A). The following conditions were used for PCR amplification: 95°C for 3 min; 40 cycles of 95°C for 10s, 55°C for 30s as described (Bhusari et al, 2011). All reactions were performed in duplicate. Quantification was performed by monitoring real-time fluorescence of Syto 9. Amplification and detection of PCR products was performed and threshold cycles (Ct) were measured using CFX96 Real-Time PCR Detection System (BioRad). Target gene expression was normalised to 18S expression and calculated to generate fold change. A t-test was used to compare the mean expression in treatment vs control groups.
Kobayashi et al (2011) used the HumanMethylation27 (Illumina, San Diego, CA, USA) methylation microarray to identify hypermethylated genes in PCa. Using this patient population as a validation cohort, we assessed the association of clinicopathological features and the presence of EVX1 methylation. Using two probes, two CpGs were assayed, one located in the promoter region 70bp upstream of the TSS (Microarray Probe no. 1) and one CpG in exon 1, 10bp downstream of the TSS (Microarray Probe no. 2).
Association of clinical features before prostatectomy between the two groups of patients with or without hypermethylated tumours were assessed using Student's t-test for normally distributed variables, whereas Wilcoxon rank sum test was used for non-normal distributed variables. For nominal variables, Fisher exact test was used. Kaplan–Meier survival analysis was performed to test for differences in time to biochemical recurrence between these two patient groups, using the Log-rank test. A multivariate Cox proportional hazards model was used to assess whether individual variables were independent predictors of biochemical recurrence using a forward selection method with a model entry criterion of P<0.05. Analyses were performed using SAS 9.3 for Windows (SAS Institute, Cary, NC, USA). All tests of significance were at the P<0.05 level, and two-tailed.
We identified four CpG islands within the EVX1 gene using MethPrimer software (Li and Dahiya, 2002). Previously described CpG island criteria include a region greater than 200bp with a observed/expected CpG ratio of >0.6 and CG percentage of >50% (Gardiner-Garden and Frommer, 1987). Three of the four CpG islands overlap exonic regions (Figure 1). We surveyed promoter methylation encompassing 96 CpGs found in CpG islands no. 1 and 2. In a validation set, microarray probes surrounding the TSS were used to compare methylation in tumour vs TA tissue. Even-skipped homeobox 1 CpG islands and the gene regions assayed are detailed in Figure 1.
Bisulphite conversion and cloning of DNA from PPC-1 and PC3 PCa cell lines was compared to HPECs cultured from non-tumour prostate tissue across 96 CpGs within CpG islands no 1 and 2, as described previously (Schwarze et al, 2001). Comparing the entire set of CpGs, PCa cell lines were significantly hypermethylated compared with HPEC lines (P<0.0001, two-way ANOVA with Bonferroni post-test). From CpGs no. 59–96, there was no difference between HPEC and PCa lines in 45/48 (94%) of the CpGs (P>0.05). However, in the region adjacent to the transcription start site, HPEC lines were significantly unmethylated. From CpGs no. 1–58, HPEC lines were hypomethylated in 54/58 (93%) of CpGs compared with PCa cell lines (P<0.05). There was no statistical difference between PPC-1 and PC3 (P=0.95). Thus, HPEC lines exhibited increasing methylation approaching intron 1, whereas PPC-1 and PC3 demonstrated a high density of methylation throughout the sequences analysed (Figure 2). This differential methylation between cancer lines and normal epithelial cells led to further analysis of the 5′ promoter region of CpG island no. 1.
MethyLight is a quantitative assay utilising PCR amplification of bisulphite-converted DNA and binding of fluorescent labelled unmethylated and methylated probes (Eads et al, 2000). Prostate cancer cell lines PPC-1, DU145, PC3 and LNCaP demonstrated high levels of methylation (100.0%, 72.1%, 100.0% and 96%, respectively), whereas normal HPEC, liver, lung, spleen and kidney demonstrated low levels of methylation (26.3%, 22.8%, 22.8% and 24.9%, respectively) at the three CpGs assayed by the MethyLight probe (Figure 3A). A region in CpG island no. 1 was assayed by MethyLight (Figure 1). The HPEC DNA treated with SSSI methyltransferase, a cytosine-methylating enzyme, was used as a positive control.
Prostate cancer specimens were then analysed using MethyLight. Methylation was increased in tumour specimens compared to matched TA tissue, as well as normal peripheral prostate tissue from prostates not containing cancer (NTA; P=0.02 and P=0.001, respectively). We then stratified paired specimens according to grade and found that high-grade tumours (Gleason8) demonstrated significantly higher methylation compared with intermediate-grade tumours (Gleason7, P=0.04; Figure 3B). Patient characteristics for these prostate specimens are provided in Supplemental Table 1.
To determine the proportion of tumours specimens that were hypermethylated, a ROC curve was constructed yielding an area under the curve of 0.73 (P=0.023). The optimal cutoff of 18.2% methylation yielded a sensitivity of 55.9% and specificity of 90.9% for differentiating tumour vs NTA specimens, and was used to define hypermethylation (Figure 3C). Using 18.2% as a cutoff, 20/35 (57%) of tumours were hypermethylated, while only 1/11 (9%) NTA tissues were hypermethylated (P=0.006, two-tailed Fisher's exact test).
To examine the effects of methylation on the expression of EVX1 mRNA and its variants, we treated PCa cell lines with the demethylating agent AzaC and the HDAC inhibitor TSA. A number of sense and antisense transcripts are expressed from the EVX1 gene (Figure 4A). 5-Azacytidine and TSA treatments were optimised for maximal gene expression and cell viability (data not shown). Using primer set 1, we found that AzaC treatment for 48h resulted in a 2.6- and 3.9-fold induction of EVX1 in PPC-1 and PC3 cell lines (P=0.047 and P=0.014, respectively). 5-Azacytidine did not induce expression of downstream or reverse transcripts (Figure 4B). The HDAC inhibitor TSA induced the EVX1 expression with the use of primer set 1 in PPC-1 and PC3 (3.2- and 8.7-fold induction; P=0.048 and P=0.0095, respectively). Trichostatin A treatment also resulted in a 1.9-fold induction of the downstream transcript using primer set 2 (P=0.039) and 2.4-fold induction of the reverse transcripts using primer set 3 in PC3 cells (P=0.0023), an effect that was not observed with AzaC treatment (Figure 4C). Induction of transcripts was not significantly altered in LNCap and DU145 cells. 5-Azacytidine and TSA can often act synergistically to re-express certain genes (Cameron et al, 1999); however, this was not observed in any cell lines for EVX1 (data not shown).
Previously, genome-wide methylation profiling was conducted in 70 matched tumours and TA prostate tissues from patients that underwent radical prostatectomy for clinically localised PCa between 1998 and 2007 (Kobayashi et al, 2011). We analysed EVX1 methylation data at two different CpGs in 56 pairs where clinical data was available. These samples were primarily intermediate-grade (93%), comprising of GG six (19%) and GG seven (74%). Of paired samples assessed for EVX1 methylation differences, 48 out of 58 (83%) exhibited hypermethylation at one or both sites in tumour compared to adjacent tissue. Serum PSA was measured in these patients on follow-up visits after surgery (mean 56 months, median 49 months) and recurrence was defined as serum PSA>0.07ngml−1. Using the Kaplan–Meier analysis and the Log-rank test, we found that patients with hypermethylated tumours using the upstream Probe no. 1 had a significantly worse recurrence-free survival compared with patients without hypermethylation (51 vs 91% P=0.04; Figure 5). Furthermore, we analysed age, PSA, Gleason sum, SV involvement, extracapsular extension, peripheral node involvement, margins, pT category and EVX1 methylation at Probe no. 1 and 2 using multivariate Cox regression survival analysis and found that only pT category and EVX1 methylation (Probe no. 1) were independent predictors of PSA recurrence (Table 1). Probe no. 2 was not predictive of biochemical recurrence (P=0.6).
Epigenetic alterations, such as gene promoter hypermethylation, show promise as potential tumour biomarkers. Even-skipped homeobox 1 (EVX1) is a gene important in blastocyst development and differentiation (Spyropoulos and Capecchi, 1994) that functions as a potent repressor of gene transcription in mammalian cells (Briata et al, 1995). Previously, a role for EVX1 in cancer had not been examined. In this study, EVX1 methylation was extensively surveyed across the promoter and 5′ regions in cell lines and tumour samples. We demonstrated that hypermethylation occurs in the majority of PCa samples, is more extensive in high-grade samples, and correlates with the risk of PSA failure in a validation cohort. This suggests a previously unrecognised biological role for this gene in PCa and as a marker for cancer and of risk of progression.
Mapping of methylation in cell lines demonstrated dense methylation of intron 1 and the 3′ region of exon 1 in HPECs, but absent methylation in upstream exonic regions including those surrounding the TSS. Prostate cancer cell lines were found to be uniformly densely methylated throughout both of these regions. These findings are consistent with the phenomenon of epigenetic spreading during aging and cancer in which CpG islands have the propensity to become methylated de novo by extension from surrounding methylated regions (Jones and Takai, 2001). Our data suggest this may occur early in the disease, as differential methylation at a downstream site was previously recognised in TA histologically normal tissues (i.e., field defect) when compared with prostate tissue from men without cancer (Yang et al, submitted for publication).
Using a quantitative MethyLight probe, we extended these findings and found that promoter hypermethylation occurred in over 57% of tumours compared to histologically normal tissues, and that hypermethylation at this locus is associated with a 6.2-fold increase in the presence of cancer, using a ROC curve. These results were confirmed by an independent validation set demonstrating that 83% of tumours exhibited hypermethylation. We observed that a higher frequency of methylation (83%) in this validation cohort, which may, in part, be explained by the close proximity of these probes to the TSS compared to our MethyLight probe (Figure 1). Alternatively, it may be due to the use of different assays. Gleason grade is a strong prognostic indicator and is linked to a number of important clinical end points, including pathological stage, progression to metastatic disease and survival (Egevad et al, 2002; Humphrey, 2004). Previously, Maruyama et al (2002) reported genes that had a qualitatively higher methylation in tumours with Gleason 7 or greater compared to Gleason 5–6. Liu et al (2011) found higher methylation of AKAP12 in Gleason 8–10 tumours compared with Gleason 4–7. Here, we demonstrate that EVX1 can quantitatively differentiate high-grade Gleason 8–10 vs intermediate-grade Gleason 6–7 tumours, thereby separating indolent from more aggressive tumours.
We extended our analysis to evaluate whether EVX1 has prognostic significance using an independent validation cohort. Notably, this cohort consisted primarily of intermediate-grade cancers (Table 2), a group that demonstrates biological variability (Stark et al, 2009), and in which it is more difficult to predict behaviour. We found that 83% of tumours were hypermethylated and that patients with EVX1 hypermethylation had only a 51% PSA recurrence-free survival compared with 91% in patients without hypermethylation. The high proportion of patients without recurrence (91%) when hypermethylation is absent indicates that EVX1 is a highly specific methylation marker for detecting recurrence, compared to other methylation markers such as AIM1 (Rosenbaum et al, 2011). Even in a cohort consisting of 93% intermediate-grade tumours (Gleason 7), pT category and EVX1 methylation were independent predictors of biochemical recurrence with odds ratios of 7.742 and 3.967, respectively, using a multivariate Cox proportional hazards model. In another study, pT category and PGP9.5 methylation were found to be independent predictors of biochemical recurrence with odds ratios of 5.352 and 3.526, respectively (Mitsui et al, 2012). In this cohort, 29% of the patients were high-grade (Gleason >7). Previously, HOXD3 was reported to interact with stage (pT) in for predicting recurrence, but was not an independent predictor on multivariate analysis (Kron et al, 2010). Incorporation of EVX1 as a predictive marker into clinical practice may help to differentiate aggressive from indolent tumours in biopsy specimens, as well as provide a more accurate prediction of PSA recurrence following prostatectomy in intermediate-grade patients.
Finally, treatment of PCa cell lines with AzaC or TSA can induce EVX1 expression. However, despite dose-ranging studies, generating re-expression in Du145 and LNCaP was difficult. These genes may contain mutations or other alterations that lead to gene inactivation. Unlike genetic mutations, the reversibility of these epigenetic changes makes them attractive targets for cancer therapy. Previous groups have shown that reversal of aberrant methylation with restoration of silenced gene expression using demethylating agents and HDAC inhibitors can decrease tumour growth in vivo and in vitro (Butler et al, 2001; Kuefer et al, 2004; Li et al, 2005; Rokhlin et al, 2006). 5-Azacytidine is currently in phase II clinical trials for the treatment of PCa (Jerónimo et al, 2011). The drawback of these agents is that they modulate the epigenome through widespread, non-specific changes in gene expression. Improving our understanding of the biology of these aberrantly methylated genes would allow for the therapeutic targeting of specific genes and signalling pathways.
In summary, our data indicate that EVX1 is epigenetically altered in PCa and may serve as a biological marker of the disease that can differentiate high-grade and intermediate-grade tumours and independently predict biochemical recurrence. Furthermore, these epigenetic alterations may influence the expression of EVX1, a gene important in embryological development. Future studies will focus on elucidating the biological role of this aberrantly-methylated gene in PCa.
We would like to thank Dr Wei Huang for pathological review of specimens and Pushpa Weeratunga for technical assistance. MT was funded by the Institute for Clinical and Translational Research Training Grant (NIH KL2 RR02012) and the University of Wisconsin School of Medicine Shapiro Grant. BY and DJ were funded by NIH 5RO1CA 097131.
Supplementary Information accompanies the paper on British Journal of Cancer website (http://www.nature.com/bjc)
This work is published under the standard license to publish agreement. After 12 months the work will become freely available and the license terms will switch to a Creative Commons Attribution-NonCommercial-Share Alike 3.0 Unported License.