Objective Understand the role of ZEB1 in the tumour initiation and progression beyond inducing an epithelial-to-mesenchymal transition.
Design Expression of the transcription factor ZEB1 associates with a worse prognosis in most cancers, including colorectal carcinomas (CRCs). The study uses survival analysis, in vivo mouse transgenic and xenograft models, gene expression arrays, immunostaining and gene and protein regulation assays.
Results The poorer survival determined by ZEB1 in CRCs depended on simultaneous high levels of the Wnt antagonist DKK1, whose expression was transcriptionally activated by ZEB1. In cancer cells with mutant TP53, ZEB1 blocked the formation of senescence-associated heterochromatin foci at the onset of senescence by triggering a new regulatory cascade that involves the subsequent activation of DKK1, mutant p53, Mdm2 and CtBP to ultimately repress macroH2A1 (H2AFY). In a transgenic mouse model of colon cancer, partial downregulation of Zeb1 was sufficient to induce H2afy and to trigger in vivo tumour senescence, thus resulting in reduced tumour load and improved survival. The capacity of ZEB1 to induce tumourigenesis in a xenograft mouse model requires the repression of H2AFY by ZEB1. Lastly, the worst survival effect of ZEB1 in patients with CRC ultimately depends on low expression of H2AFY and of senescence-associated genes.
Conclusions The tumourigenic capacity of ZEB1 depends on its inhibition of cancer cell senescence through the activation of a herein identified new molecular pathway. These results set ZEB1 as a potential target in therapeutic strategies aimed at inducing senescence.
Statistics from Altmetric.com
Significance of this study
What is already known on this subject?
Aberrant activation of Wnt is involved in the pathogenesis and progression of several human cancers, most notably colorectal carcinomas (CRCs).
A key target of Wnt is the transcription factor ZEB1, whose expression promotes tumour initiation and progression by triggering an epithelial-to-mesenchymal transition (EMT) that induces a more motile and stem-like phenotype in cancer cells at the tumour invasive front.
However, the role of ZEB1 in tumour initiation and progression beyond the induction of an EMT remains poorly understood.
What are the new findings?
The worst survival effect of ZEB1 in patients with CRC depends on high levels of the Wnt antagonist DKK1. ZEB1 transcriptionally activates DKK1 and their expression correlates in human CRCs and ZEB1-deficient mice.
ZEB1 inhibits senescence in cancer cells through activation of a newly identified ZEB1-DKK1 mutant p53-Mdm2-CtBP pathway to repress macroH2A1.
ZEB1 and DKK1 associate in Wnt-active cells at the invasive front of CRCs in an inverse pattern with macroH2A1. In a mouse model of CRC, partial downregulation of ZEB1 is sufficient to induce macroH2A1 and trigger in vivo senescence, thus resulting in fewer tumours and better survival. In a xenograft mouse model, the tumourigenic capacity of ZEB1 requires its repression of macroH2A1.
The worst survival effect of ZEB1 in human patients with CRC ultimately depends on low expression of senescence-inducing genes and macroH2A1.
How might it impact on clinical practice in the foreseeable future?
The study establishes a new mechanism through which ZEB1 promotes tumour progression and sets ZEB1 as a target in therapeutic strategies aimed at inducing senescence.
Aberrant activation of canonical Wnt signalling, most often by mutations in components of the pathway, is involved in the pathogenesis of numerous cancers.1 Wnt signalling is also triggered by the engagement of cell surface receptors by Wnt ligands, which prompts the nuclear translocation of β-catenin that cooperates with TCF/LEF factors in the transcriptional activation of Wnt target genes. In turn, Wnt antagonists of the sFRP (sFRP1-sFRP5) and Dickkopf (DKK1-DKK4) families inactivate Wnt signalling by sequestering Wnt ligands or binding to Wnt receptors, respectively.1 Wnt antagonists most commonly function as tumour suppressors and their loss of expression/function promotes tumour growth and invasiveness. However, in certain cancers and stages of tumour progression, some Wnt antagonists can promote oncogenesis.1 ,2 Notably, the Wnt antagonist DKK1 is induced by TCF4/β-catenin and is overexpressed in some carcinomas.3 ,4
Senescence involves an irreversible cell cycle arrest and constitutes an important tumour suppressor mechanism.5 Repression of proliferative genes during senescence is in part achieved through chromosome condensation into senescence-associated heterochromatin foci (SAHF).6–8 Wnt signalling delays both replicative and oncogene-induced senescence and downregulation of the Wnt3a ligand is both necessary and sufficient to trigger SAHF assembly and to drive several cell types into senescence.9 Nevertheless, the role of senescence in Wnt-induced cancer initiation and progression is still not fully understood.
A key target of Wnt in tumour progression is the transcription factor ZEB1 (also known as δEF1), whose expression in malignant cells at the invasive front of carcinomas triggers an epithelial-to-mesenchymal transition (EMT), endows cancer cells with a proinvasive and stem-like phenotype, and determines a worse clinical prognosis in most human cancers.10 ZEB1 can either directly activate or repress gene expression by recruitment of coactivators (eg, p300) or corepressors (eg, CtBP).10–12 Interestingly, ZEB1 is downstream of several classical pathways involved in oncogenesis (eg, transforming growth factor (TGF)β, Wnt, Ras), whose activity can be in turn regulated by ZEB1.12–17 Thus, ZEB1 is induced by Wnt signalling and also activates several downstream Wnt target genes.14 ,15 The binding of ZEB1 to DNA-binding transcription factors that act as effectors of some of these pathways (eg, R-Smads, TCF4) can turn ZEB1 from a transcriptional repressor into an activator.12 ,13 ,15
Beyond the induction of an EMT, the mechanisms by which ZEB1 promotes tumour initiation and progression remain to be elucidated. We found here that the maximum effect of ZEB1 as a determinant of decreased survival in patients with colorectal cancer (CRC) unexpectedly depends on its coexpression with high levels of the Wnt antagonist DKK1. We also found that ZEB1 upregulates DKK1 expression by direct activation of its promoter and that both genes are inversely correlated with senescence-associated genes in expression arrays of human CRCs. ZEB1, through its induction of DKK1, inhibited SAHF formation and expression of senescence-associated genes in cancer cells. In cancer cells with mutant TP53, we found that DKK1 upregulates the expression of TP53, MDM2 and CTBP and that inhibition of senescence by ZEB1 ultimately depends on the DKK1/mutant p53/Mdm2/CtBP-mediated repression of histone variant macroH2A1 (encoded by H2AFY). Of note, activation of this pathway by DKK1 is independent of its role as a Wnt antagonist. ZEB1 and DKK1 display an inverse correlation pattern with macroH2A1 in malignant cells at the tumour front of CRCs. Using a transgenic mouse model of CRC, we found that deletion of a single Zeb1 allele was sufficient to trigger both macroH2A1 expression and cancer cell senescence that resulted in reduced tumour load and higher survival in Zeb1 (+/−) mice vis-à-vis Zeb1 (+/+) littermates. Importantly, in a xenograft mouse model, we found that the capacity of ZEB1 to promote tumourigenesis requires its inhibition of macroH2A1. In parallel, the role of ZEB1 as a determinant of poorer survival in patients with CRC depends on the simultaneous low expression of H2AFY and senescence-associated genes.
These results establish the mechanism by which ZEB1, through the subsequent induction of DKK1, mutant p53, Mdm2 and CtBP and the downregulation of macroH2A1, inhibits cancer cell senescence and promotes tumour progression, thus offering a new entry point to interfere with ZEB1 expression and function.
Cell lines and cell culture
The origin and culture of cell lines in the study is described in online supplementary information.
Antibodies, plasmids, oligonucleotides and short hairpin RNAs
Description and source of antibodies, plasmids, DNA and RNA oligonucleotides and short hairpin RNA (shRNA) lentivirus used in this study are detailed in online supplementary information. Transient and stable interference of gene expression by small interfering RNA (siRNA) and shRNA are also described in online supplementary information.
Gene expression array data and survival plots
Analysis of the association between ZEB1 and DKK1 expression in CRCs and gene signatures associated with different cohorts of patients with CRC is described in online supplementary information. Correlation between the expression of selected genes and relapse-free survival was assessed as detailed in online supplementary information.
Human primary CRC tissue
Paraffin-embedded sections of human primary CRCs were obtained as described in online supplementary information.
Transgenic mouse model of CRC and mouse tumour xenograft model
Description of the transgenic mouse model of CRC is detailed in online supplementary information. Likewise, the set up and analysis of the mouse tumour xenograft model were performed as described in online supplementary information.
Cell viability and senescence assays
Overall cell viability and senescence were assessed by MTT assays and staining for senescent-associated β-galactosidase (SA β-gal), respectively, as detailed in online supplementary information.
Determination of protein and RNA expression and transcriptional assays
Analysis of protein expression by immunostaining or Western blot is described in online supplementary information. Quantitative real-time PCR, mutagenesis of promoters as well as chromatin immunoprecipitation (ChIP) and transcriptional assays were performed as described in online supplementary information.
Statistical analysis of the data shown in this study was performed as described in online supplementary information.
ZEB1's role as a determinant of worse cancer survival depends on the joint coexpression of the Wnt antagonist DKK1
ZEB1 has an independent prognostic value for decreased survival in a number of cancers.14 Since ZEB1 directly activates several Wnt target genes, we hypothesised that ZEB1 would inhibit the expression of Wnt antagonists and decided to explore first whether ZEB1 and Wnt antagonists have opposing effects on survival of patients with CRC.
Examination of the relapse-free survival associated with ZEB1 and Wnt antagonists in gene expression arrays of CRCs18 ,19 revealed that, as expected, high expression of ZEB1 correlated with poorer survival. In turn, expression of Wnt antagonists was either associated with better prognosis (sFRP3) or had no significant association with survival (see online supplementary table S1). Unexpectedly, we found that DKK1 and sFRP4 were associated with poorer prognosis, indicating a potential tumour-promoting role of these Wnt antagonists in CRCs. Given that DKK1, like ZEB1, is induced by canonical Wnt signalling,3 we investigated whether the joint expression of DKK1 and ZEB1 affects overall survival in CRCs. Interestingly, using survival data for 928 patients from published databases (see Methods), we found that patients displaying high expression of both DKK1 and ZEB1 had a lower survival rate than those with high levels of ZEB1 but low DKK1 (figure 1A). In other words, the maximum effect of ZEB1 as a predictor of reduced survival requires high levels of DKK1. This suggests that, at least for some of its tumour-promoting functions, ZEB1 depends on DKK1 expression.
In light of these results, we next investigated whether downregulation of ZEB1 and/or DKK1 affects overall cell viability of SW480 cells, a commonly used Wnt-active CRC cell line, which harbour a mutation in APC at 1338 bp that results in a stop codon and a truncated protein, methylated p16INK4A (CDKN2A), phosphorylated Rb protein and mutant TP53 (R273H and P309S). Compared with an siRNA control (siCtl), knockdown of either ZEB1 or DKK1 with specific siRNAs (siZEB1 and siDKK1, respectively) reduced overall cell viability in SW480 CRC cells (figure 1B and online supplementary figure S1A). Simultaneous interference of ZEB1 and DKK1 did not have an additional effect over ZEB1 knockdown alone suggesting that in cell viability—although not in relapse-free survival—DKK1 may act in the same functional pathway as ZEB1 (figure 1B). We next tested whether recombinant human DKK1 (rhDKK1) was able to revert the effect of siDKK1 on cell viability. Despite the fact that SW480 cells have a truncated APC, they are still responsive to Wnt ligands and antagonists.20 As control we used recombinant human DKK3 (rhDKK3), a protein homologous to DKK1 that through a different receptor and mechanism also reduces nuclear β-catenin in SW480 cells and, as other Wnt antagonists, can either activate or repress Wnt-mediated signalling depending on the cell type.21 ,22 We found that although both rhDKK1 and rhDKK3 displayed similar effectiveness blocking Wnt signalling in SW480 cells, only rhDKK1 reverted the effect of siDKK1 on cell viability (see online supplementary figures S1B and S1C).
We then examined whether ZEB1 and DKK1 were correlated in primary CRCs. Examination of 1557 cases from expression arrays of CRCs (see supplementary figure S1D). This positive correlation between ZEB1 and DKK1 was also observed in a panel of CRC cell lines (figure 1D). SW620 cells originated from CRC metastasis in the same patient than SW480 cells and although they carry identical mutations they have epigenetic differences. Expression of DKK1 did not associate with ZEB2, a second member of the ZEB family and highly homologous to ZEB123 (see online supplementary figure S1E). Altogether, these results indicate that ZEB1 and DKK1 are coexpressed in CRCs, where they jointly determine a worse prognosis.
ZEB1 transcriptionally activates DKK1 expression
The above data prompted us to investigate whether ZEB1 may activate DKK1 expression. We first examined whether deletion of the Zeb1 gene alters Dkk1 expression in the intestinal tract of Zeb1-deficient mice. Compared with wild-type littermates, Dkk1 expression was reduced in the intestinal tract of Zeb1 (+/−) mice—Zeb1 (−/−) mice die perinatally24—corroborating the in vivo association between both genes (figure 2A).
We then examined the effect on endogenous DKK1 expression of knocking down endogenous ZEB1 in three CRC cell lines expressing high levels of ZEB1. Compared with siCtl, transient knockdown of ZEB1 with two specific siRNAs (siZEB1-A and siZEB1-B) resulted in the downregulation of endogenous DKK1 mRNA levels (figure 2B, left panel, and online supplementary figures S2A and S2B). Similar downregulation of DKK1 was observed when ZEB1 was stably knocked down in SW480 cells using two different shRNAs (to generate SW480-shZEB1-A and SW480-shZEB1-B cells) and compared with cells transfected with an shRNA control (SW480-shCtl cells) (figure 2B, right panel). In contrast, knockdown of ZEB2 in ZEB2-positive COLO320 CRC cells did not affect DKK1 mRNA levels (see online supplementary figure S2C). Positive regulation of DKK1 by ZEB1 was also examined at the protein level. Compared with control siRNA or shRNA, expression of DKK1 protein was downregulated upon transient and stable knockdown of ZEB1 in SW480 CRC cells (figure 2C, left and right panels, respectively). Altogether, these results indicate that ZEB1 induces DKK1 expression in CRC cell lines.
ZEB1 binds to a subset of E-box and E-box-like sequences in the regulatory regions of its target genes.10 We therefore investigated whether ZEB1-mediated induction of DKK1 occurs at the transcriptional level. Examination of the first 1 kb of the human DKK1 promoter revealed the existence of several ZEB1 consensus binding sites including two high-affinity sequences at −490 and −880 bp. Accordingly, the transcriptional activity of serial deletions of the DKK1 promoter increased when the site at −490 bp was included (see online supplementary figure S2D). The ability of endogenous ZEB1 to directly bind to the DKK1 promoter was tested for the site at −490 bp by ChIP assay. An antibody against ZEB1, but not its respective matched specie IgG control, immunoprecipitated a region of the human DKK1 promoter containing the site at −490 bp, but not a region lacking consensus binding sites for ZEB1 (figure 2D). ZEB1 also failed to bind to the GAPDH promoter, whose expression is not regulated by ZEB1 (figure 2D).
To test the functionality of the ZEB1 binding site at −490 bp, this sequence was mutated within the −535 bp DKK1 promoter reporter to a sequence known not to bind ZEB1 and the resulting mutant DKK1 promoter (DKK1ZEB1mut) was compared with the wild-type DKK1 promoter (DKK1wt) for its regulation by ZEB1. Mutation of the ZEB1 site in DKK1wt promoter reduced its basal activity in SW480 and SW620 CRC cells (figure 2E and online supplementary figure S2E). Transient and stable knockdown of ZEB1 in both CRC cell lines resulted in a downregulation of the activity of DKK1wt but not of DKK1ZEB1mut promoter (figure 2E, left and right panels, respectively, and online supplementary figure S2E). We also examined the response of the wild-type and mutant versions of the DKK1 promoter to ZEB1 overexpression in SW480 and SW620 cells. While exogenous ZEB1 further activated transcription of the DKK1wt promoter, it had no effect on DKK1ZEB1mut promoter (figure 2F and online supplementary figure S2F).
ZEB1 activates transcription through recruitment of the p300 histone acetyltransferase to its N-terminal region.12 ,13 Overexpression of ZEB1 upregulates Wnt target genes in a p300-dependent manner in cells with active Wnt signalling, but not in cells where this pathway is inactive.15 Therefore, we investigated the involvement of p300 in ZEB1-mediated activation of DKK1—that as noted earlier is a direct Wnt target—in SW480 and SW620 cells, two Wnt-active CRC cell lines that therefore express high levels of ZEB1 (figure 1D). A version of p300 where its cDNA (EP300) is fused to the activation domain of the herpes simplex virus VP16 protein (VP16-p300) enhanced ZEB1-induced activation of DKK1wt promoter (figure 2G). In contrast, deletion of the N-terminal region of ZEB1 (ZEB1ΔNterm) eliminated most of the transcriptional activation effect of ZEB1 on the DKK1wt promoter (figure 2G). The recruitment of p300 by full-length ZEB1 was confirmed by a ChIP assay where a p300 antibody, but not a matched specie IgG control, precipitated the fragment of the DKK1 promoter harbouring the ZEB1 binding site at −490 bp, but not a fragment of this promoter lacking ZEB1 binding sites (figure 2H).
ZEB1 and TCF4 mutually modulate their transcriptional activities in the regulation of selected Wnt target genes,15 and we found here that both transcription factors cooperate in the regulation of the DKK1 promoter in SW480 and SW620 CRC cells (figure 2I and online supplementary figure S2G). Altogether, the above results demonstrate that ZEB1 binds to the DKK1 promoter to directly drive its transcription through a mechanism involving p300 and in cooperation with TCF4.
ZEB1, through its induction of DKK1, inhibits senescence-associated genes
The 1557 cases of CRC in figures 1C and online supplementary figure S1D were classified in four cohorts based on their expression of ZEB1 and DKK1 above or below the upper quartile. Analysis of the gene signature of these cohorts revealed that, compared with CRCs where both ZEB1 and DKK1 were above the upper quartile (cohort 1), 429 genes were upregulated when the expression of either ZEB1 or DKK1 was below that quartile (cohorts 2 and 3, respectively) (see online supplementary table S2). The expression of these genes was further increased when both ZEB1 and DKK1 levels were in the three lower quartiles (cohort 4). Interestingly, within this subset of genes upregulated when ZEB1 and DKK1 expression was low, around 10% of them were associated with senescence, suggesting that ZEB1 and DKK1 may cooperate in the repression of cellular senescence (figure 3A).
We therefore validated the potential regulation by ZEB1 and/or DKK1 of some of these senescence-related genes. To that effect, SW480-shCtl, SW480-shZEB1-A and SW480-shZEB1-B cells used in previous figures were stably transfected with an expression vector for DKK1 (to generate SW480-shCtl+DKK1, SW480-shZEB1-A+DKK1 and SW480-shZEB1-B+DKK1 cell lines, respectively) or their corresponding empty vector equivalents (to generate SW480-shCtl+vector, SW480-shZEB1-A+vector and SW480-shZEB1-B+vector cell lines, respectively) (see online supplementary figure S3A for ZEB1 and DKK1 levels in these stable cell lines).
The upregulation of senescence-associated genes in CRCs with low expression of ZEB1 and/or DKK1 (figure 3A) suggested that ZEB1 and DKK1 are repressing their expression. Using these stable cell lines, we validated the regulation by ZEB1 and/or DKK1 of several of these senescence-associated genes. Namely, histone variant macroH2A1 (encoded by H2AFY)—an integral component of SAHF and whose knockdown blocks SAHF formation and cellular senescence,6 GLB1—that encodes lysosomal β-galactosidase and whose mRNA levels as well as enzymatic activity at pH 6.0 increase in senescent cells, commonly referred as SA β-gal activity25 ,26—RPL5, EIF3K, EHMT2 and AATF.27–30 Indeed, we found that mRNA levels for all these genes increased in SW480-shZEB1-A and SW480-shZEB1-B cells with respect to SW480-shCtl cells (figure 3B). In turn, overexpression of DKK1 reverted the effect of ZEB1 knockdown (figure 3B).
Cell cycle inhibitors p16INK4A (CDKN2A) and p15INK4b (CDKN2B) are upstream of Rb1 and accumulate in senescent cells.5 Although CDKN2A is methylated in SW480 cells, expression of CDKN2A and CDKN2B was upregulated in SW480 cells knocked down for ZEB1 (SW480-shZEB1-A+vector and SW480-shZEB1-B+vector) compared with SW480-shCtl+vector cells (figure 3C). In that line, CDKN2A and CDKN2B are also repressed by ZEB1 in lung epithelial cells and mouse embryo fibroblasts (MEFs).13 ,31 In contrast, overexpression of DKK1 in SW480-shZEB1-A+DKK1 and SW480-shZEB1-B+DKK1 cells failed to revert this increase, suggesting that CDKN2A and CDKN2B are regulated by ZEB1 but not by DKK1.
ZEB1, through its induction of DKK1, inhibits SAHF formation and cell senescence
Downregulation of Wnt3a is both necessary and sufficient for the translocation of the chaperone HIRA to nuclear bodies and accelerates the assembly of SAHF, which are required to trigger senescence.9 Since ZEB1 is both a target and an effector of Wnt signalling, we first tested whether ZEB1 could mediate Wnt3a-induced inhibition of senescence in SW480 CRC cells. As noted earlier, SW480 cells harbour a stop codon mutation in APC. Nevertheless, truncated forms of the APC protein can still bind β-catenin and cells carrying them are responsive to Wnt ligands.20 SW480-shCtl and SW480-shZEB1-A cells were cultured in the presence or absence of Wnt3a and their senescence was assessed by staining for SA β-gal activity.25 We found that ZEB1 knockdown increased the number of SA β-gal-positive cells and that Wnt3a inhibited basal SA β-gal activity in SW480-shCtl cells but not in SW480-shZEB1-A cells (figure 4A). ZEB1 interference yielded similar results in SW620 CRC cells treated with Wnt3a (see online supplementary figure S4A). Activation of Wnt signalling by Wnt3a was confirmed by the upregulation of Wnt targets ZEB1, DKK1 and AXIN2 (see online supplementary figure S4B). These results indicate that ZEB1 mediates the Wnt-induced inhibition of senescence and that ZEB1 knockdown triggers senescence in CRC cells.
We next investigated whether DKK1 also inhibits senescence in cancer cells. SW480 cells were transiently interfered with siCtl, specific siRNAs for either ZEB1 (siZEB1-A) or DKK1 (siDKK1) or a combination of siZEB1-A and siDKK1 and then examined for SA β-gal staining. Compared to siCtl, single interference of either ZEB1 or DKK1 resulted in an increase in the number of SA β-gal-positive cells (figure 4B). Induction of senescence by siDKK1 was reverted by rhDKK1, but not by rhDKK3 (see online supplementary figure S4C). These results indicate that inhibition of senescence by DKK1 occurs independently of its role as a Wnt antagonist since both rhDKK1 and rhDKK3 were equally efficient blocking Wnt signalling (ref 22 and online supplementary figure S1B). The number of senescent cells following joint knockdown of both ZEB1 and DKK1 was similar to that observed with just single ZEB1 knockdown (figure 4B), thus suggesting that DKK1 inhibits senescence through the same functional pathway as that by ZEB1. Lastly, the induction of senescence by siDKK1 was corroborated with an anti-DKK1 blocking antibody (figures 4C and online supplementary figure S4D).
We then examined whether exogenous overexpression of DKK1 could revert the senescence induced by ZEB1 knockdown. To that effect, the stable SW480 cells used in figures 3B,C were assessed for senescence by SA β-gal staining. Like in transient interference, stable knockdown of ZEB1 increased the number of SA β-gal-positive cells that, interestingly, was reverted by overexpression of DKK1 (figure 4D). The increase in SA β-gal-positive cells induced by shZEB1 was reverted by rhDKK1 but not by rhDKK3 (figure 4E and online supplementary figure S4E). Once again, the failure of DKK3 to inhibit senescence indicates that DKK1 mediates ZEB1-induced inhibition of senescence through a mechanism that is independent of its role as Wnt antagonist. It is worth noting that since SW480 cells are Wnt-active and express high levels of ZEB1, exogenous DKK1 also inhibited basal levels of senescence in shCtl cells. The above data allow us to conclude that inhibition of senescence by ZEB1 is mediated, at least in part, by activating DKK1 expression.
Lastly, we examined whether manipulation of ZEB1 and/or DKK1 expression alters SAHF formation triggered by macroH2A1. We found that knockdown of ZEB1 or DKK1 increased the assembly of SAHF (figures 4F,G and online supplementary figure S4F). In parallel with figure 4B, simultaneous knockdown of ZEB1 and DKK1 did not increase SAHF formation vis-à-vis single ZEB1 knockdown suggesting, once again, that DKK1 may exert its effects on senescence through the same functional pathway than ZEB1.
Inhibition of senescence by ZEB1 requires repression of H2AFY
Contrary to ZEB1, macroH2A1 suppresses tumour initiation and progression and associates with better survival in multiple types of cancer.32 Therefore, out of all the genes validated in figure 3B, we focused on H2AFY. First, we confirmed the inhibition of macroH2A1 by ZEB1 and DKK1 at the protein level. Knockdown of ZEB1 and DKK1—with siZEB1-A and siDKK1—upregulated macroH2A1 protein expression (figure 5A). Repression of macroH2A1 by ZEB1 was also tested in vivo. Compared with their wild-type littermates, expression of H2afy was upregulated in the intestine of Zeb1 (+/−) mice (figure 5B).
Next, we investigated whether inhibition of H2AFY by ZEB1 involves direct binding of ZEB1 to the H2AFY promoter. We identified two high-affinity binding sites for ZEB1 at −1059 and −1687 bp and tested the ability of the former to recruit ZEB1 in ChIP assays (figure 5C and online supplementary figure S5A). An antibody against ZEB1, but not its corresponding matched specie IgG control, immunoprecipitated a region of the H2AFY promoter containing this binding site but not a region lacking consensus binding sites for ZEB1 (figure 5C). The GAPDH promoter, which is not regulated by ZEB1, did not recruit ZEB1.
Repression of H2AFY by ZEB1 and DKK1 prompted us to investigate whether inhibition of senescence by ZEB1 and DKK1 depends on macroH2A1. The stable SW480 cell lines used in figures 3B,C—with ZEB1 and DKK1 expression knocked down or overexpressed, respectively—were now transiently transfected with siCtl or a siRNA against H2AFY (siH2AFY) and tested for SA β-gal staining (see online supplementary figure S5B for H2AFY knockdown efficiency). As shown in figure 5D and online supplementary figure S5C, siH2AFY inhibited both basal SA β-gal staining and the senescence induced by ZEB1 knockdown. Once again, DKK1 overexpression reverted the senescence induced by shZEB1, which was further reduced by siH2AFY. These results indicate that the ability of ZEB1 and DKK1 to repress senescence depends on their repression effect on H2AFY.
ZEB1 represses H2AFY expression and senescence through the subsequent induction of DKK1, mutant p53, Mdm2 and CtBP
ZEB1 represses gene expression through the recruitment of non-DNA binding corepressors, chiefly CtBP.11 ,13 ,33 Therefore, we first checked whether CtBP represses H2AFY expression. Indeed, as shown in figure 5E, transient knockdown of CtBP with a specific siRNA (siCtBP) upregulated H2AFY mRNA levels. Furthermore, mutation of the CtBP binding sites in ZEB1 (ZEB1-CIDmut) hampered ZEB1's repression of H2AFY (figure 5F).
The ability of DKK1 to mediate ZEB1 inhibition of H2AFY prompted us to investigate whether DKK1 could induce CtBP expression. Indeed, overexpression of DKK1 upregulated CtBP expression at the mRNA and protein levels (figure 5G and online supplementary figure S5D). Likewise, addition of rhDKK1, but not rhDKK3, upregulated CTBP expression (figure 5H). We then examined whether the increased CtBP expression induced by DKK1 enhances ZEB1 repressor activity. The region of ZEB1 containing the CtBP interacting domain (CID), either wild-type or mutated to a sequence unable to bind ZEB1, was fused to the DNA binding domain of the yeast protein Gal4 to create Gal4-ZEB1-CID and Gal4-ZEB1-CIDmut. Both constructs were then cotransfected with a luciferase reporter containing UAS DNA binding sites and driven by the SV40 promoter into SW480 cells stably expressing DKK1 (SW480+DKK1 cells) or its corresponding empty vector (SW480+vector cells). DKK1 overexpression greatly enhanced the ability of ZEB1 to repress transcription (figure 5I).
CtBP binding to 2-keto-4-methylthiobutyrate (MTOB), an intermediate in the methionine salvage pathway, relieves CtBP-mediated transcriptional repression in breast cancer cells.34 We found that blocking of CtBP activity in SW480 cells by MTOB induced senescence (figure 5J and online supplementary figure S5E) and upregulated H2AFY (see online supplementary figure S5F). Altogether, these results indicate that DKK1 cooperates with ZEB1 in the repression of H2AFY and senescence by upregulating CtBP.
Mdm2 is an E3 ubiquitin-ligase best known for targeting the p53 tumour suppressor for proteasomal degradation, although Mdm2 also promotes tumour progression through transcriptional regulation and histone modification in a p53-independent manner.35–37 Induction of senescence is in fact one of the key mechanisms through which p53 suppresses tumourigenesis and, consequently, Mdm2 antagonists like nutlin-3a trigger senescence.38 ,39 Of note, Mdm2 binds to and cooperates with CtBP to repress transcription.40 This evidence prompted us to investigate whether mutant p53 and/or Mdm2 are involved in the CtBP-mediated repression of H2AFY and senescence by ZEB1 and DKK1.
It is now well established that mutations in TP53 not simply eliminate wild-type TP53 tumour suppressor functions but also confer cancer cells with oncogenic hallmarks to promote tumour progression such as overcoming of senescence.41 ,42 Wild-type and mutant p53 proteins differ in their target genes and also in the direction (eg, activation or repression) and degree (eg, deficient or enhanced transcription) of their regulation.43 Interestingly, mutant p53—including some variants with mutations in the DNA binding domain—retain the ability to bind and transcriptionally activate the MDM2 promoter.44 ,45 Thus, some mutant TP53 cells, like SW480 CRC cells, still express MDM2.46
We found that overexpression of DKK1 or addition of rhDKK1, but not rhDKK3, in mutant SW480 cells upregulated mutant TP53 and MDM2 expression in these cells (figure 6A, B). In turn, DKK1 knockdown downregulated both mutant TP53 and MDM2 (figure 6C) and this downregulation was reverted by addition of rhDKK1, but not of rhDKK3 (see online supplementary figure S6A). The downstream effect of DKK1 via mutant p53 was corroborated by the knockdown of the latter with a specific siRNA (siTP53). siTP53, but not siCtl, downregulated mRNA expression levels of MDM2 and CTBP while increasing those of H2AFY (figure 6D).
Interestingly, wild-type and mutant p53 have been described to have opposite functions with regard to EMT.47 Wild-type p53 inhibits the EMT by direct transactivation of miR-200c, which targets ZEB1 mRNA for decay. Meanwhile, mutant p53 proteins induce an EMT.47 Accordingly, we found that siTP53 downregulated ZEB1 in SW480 cells (figure 6D).
MDM2 expression was also knocked down with a specific siRNA (siMDM2). siMDM2 downregulated CtBP mRNA and protein levels (figure 6E and online supplementary figure S6B) and increased H2AFY mRNA expression (figure 6E). MDM2 knockdown was also used to confirm the sequencing of the regulatory cascade from DKK1 to CtBP. SW480 cells interfered with siCtl or siMDM2 were transfected with an expression plasmid for DKK1 or its corresponding empty vector. As shown in figure 6F, MDM2 knockdown blocked the upregulation of CTBP by DKK1 as shown in figure 5G. Altogether, the above results indicate that in mutant TP53 cells DKK1 cooperates with ZEB1 in the repression of H2AFY through the subsequent induction of mutants TP53, MDM2 and CTBP.
Downregulation of Zeb1 is sufficient to trigger in vivo senescence, reduce tumour load and improve survival in a mouse model of colon cancer
ZEB1 is not expressed in the normal colonic mucosa or by well-differentiated cancer cells in the tumour centre of CRCs. Instead, ZEB1 is found in dedifferentiated malignant cells at the CRC invasive front.14 ,48 ZEB1 is also expressed by stromal cells in both normal colorectal mucosa and CRCs.10 ,14 ,48 In contrast, macroH2A1 is strongly expressed in normal colonic mucosa, but not in invasive CRCs (ref. 49, and online supplementary figure S7A). In light of this reverse pattern of expression, we tested the correlation of ZEB1, DKK1 and macroH2A1 at the invasive front of sporadic CRCs. Evaluation of ZEB1 and DKK1 staining in a tissue microarray of 53 primary human CRCs stages I–IV revealed a positive correlation between both proteins (Spearman's ρ=0.46). Conversely, macroH2A1 expression displayed an inverse correlation with ZEB1 and DKK1 (figures 7A, B).
We then examined whether expression of ZEB1 in CRCs represses senescence in vivo. To that effect, Zeb1 (+/+) and Zeb1 (+/−) mice were used in a well-established mouse model of CRC induced by administration of azoxymethane and cycles of dextran sodium sulfate (reviewed in ref. 50) (figure 7C). Colorectal tumours collected at the end of the protocol (day 70) were examined for cellular senescence. Interestingly, we found that tumours from wild-type mice barely had senescent cells, whereas those from Zeb1-deficient ones displayed larger number of SA β-gal-positive cells (figure 7D). Likewise, SAHF were only observed in tumours from Zeb1 (+/−) mice but not in those from Zeb1 (+/+) mice (figure 7E and online supplementary figure S7B). Given the role of senescence as a tumour suppressor programme, we examined mice from both genotypes for the total number of colonic tumours. In line with the above results, Zeb1 (+/−) mice developed a lower number of tumours than Zeb1 (+/+) mice and the percentage of Zeb1 (+/−) mice that survived until the end of the protocol was also higher than among Zeb1 (+/+) mice (figure 7F).
Colorectal tumours from Zeb1 (+/+) and (+/−) mice were also stained for macroH2A1. ZEB1 was strongly expressed by cancer epithelial cells in colorectal tumours from Zeb1 (+/+) mice but only a few epithelial cells were stained for macroH2A1 (figure 7G). In contrast, in the tumours of Zeb1 (+/−) mice, expression of ZEB1 was greatly reduced in the malignant epithelial cells—ZEB1 was mostly confined to a few stromal cells—while they displayed strong expression of macroH2A1. In addition, tumours in Zeb1 (+/−) mice displayed neoplastic glands with partially preserved goblet cells and polarisation and scarce signs of mitotic activity (figure 7G). In contrast, the tumours arising in Zeb1 (+/+) mice showed altered tubular structures with almost complete loss of polarisation and atypical nuclei and mitosis homogeneously distributed throughout the gland, even reaching the lumen (figure 7G). Altogether, these data indicate that expression of ZEB1 in CRCs represses senescence. Importantly, these results also show that partial downregulation of Zeb1, even to only heterozygous levels, is sufficient to induce macroH2A1 expression and SAHF formation, to trigger in vivo cancer cell senescence, to reduce tumour load and to confer a more differentiated histological pattern.
The tumourigenic capacity of ZEB1 critically depends on its repression of H2AFY
Next, we wondered whether ZEB1's role promoting tumourigenesis requires in vivo repression of macroH2A1. The SW480-shCtl and SW480-shZEB1-A cells used above were now stably cotransfected with retroviral shRNA vector against H2AFY (pSUPER-shH2AFY) or its corresponding shRNA control (pSUPER-shCtl) to generate the following cell lines: SW480-shCtl/shCtl, SW480-shZEB1-A/shCtl, SW480-shCtl/shH2AFY and SW480-shZEB1-A/shH2AFY (see protein levels of macroH2A1 and ZEB1 in these cell lines in online supplementary figure S8A). The four cell lines were then used in a xenograft mouse model by inoculation in 10-week-old SCID mice and tumour growth was monitored for up to 23 days. As expected, single knockdown of ZEB1 (SW480-shZEB1-A/shCtl cells, blue line in figure 8A, B and online supplementary figure S8B) drastically reduced tumour growth. Interestingly, when cells were simultaneously knocked down for both ZEB1 and H2AFY (SW480-shZEB1-A/shH2AFY cells, green line in figures 8A, B and online supplementary figure S8B), the ability of shZEB1 to reduce tumourigenesis was reversed. Lastly, single knockdown of H2AFY (SW480-shCtl/shH2AFY cells, red line in figure 8A, B and online supplementary figure S8B) accelerated tumour growth compared with control cells (SW480-shCtl/shCtl cells, yellow line in figure 8A, B and online supplementary figure S8B). Overall, these results indicate that in vivo expression of macroH2A1 inversely correlates with ZEB1 (figure 7A, G) and ZEB1 induces tumourigenesis through repression of macroH2A1 (figure 8A).
It is worth remarking here that there was no statistical difference between the tumourigenic capacity of SW480-shZEB1-A/shH2AFY and SW480-shCtl/shH2AFY cells (green line vs red line in figure 8A) or between SW480-shCtl/shCtl and SW480-shZEB1-A/shH2AFY cells (yellow line vs green line in figure 8A). Importantly, the lack of significant difference between these experimental condition pairs suggests that macroH2A1 is a key ZEB1 target in tumourigenesis and also that most of ZEB1's role in tumourigenesis and tumour progression is contingent on its repressor effect on macroH2A1.
ZEB1's role as determinant of worse cancer survival depends on its inhibition of H2AFY and senescence
In vivo evidence in human and mouse tumours indicates that senescence is associated with lower malignancy, slower tumour growth and, consequently, better clinical prognosis.5 Therefore, we tested whether the worse survival effect of ZEB1 in CRCs depends on its inhibition of senescence.
Altogether, the results shown above indicate that ZEB1 inhibits senescence in CRCs by induction of DKK1 and repression of macroH2A1. We therefore first examined whether expression of H2AFY modulates the effect of ZEB1 on the prognosis of patients with CRC. Analysis of the 928 human CRCs used in figure 1A revealed that those cases displaying high expression of ZEB1 and low expression of H2AFY have lower survival probability than those with high levels of ZEB1 but high levels of H2AFY (yellow and red lines in figure 8C). This implies that the maximum effect of ZEB1 as a predictor of poorer survival in CRC requires of low levels of H2AFY or that, in other words, high levels of ZEB1 are not sufficient to determine worse survival if not accompanied by low levels of H2AFY. Similar results were found for the effect of H2AFY expression on the survival probability determined by DKK1: high levels of DKK1 only determined maximum levels of reduced survival when expression of H2AFY was low (see online supplementary figure S8C). Of note, overall survival depends on multiple variables beyond tumour growth thus contributing to explain that H2AFY knockdown promoted tumour growth in the xenograft mouse model independently of ZEB1 levels (red and green lines in figure 8A), while lower overall survival in patients with CRC associated with low levels of H2AFY is worsened by high levels of ZEB1 (red and green lines in figure 8C). From all these data it could be therefore concluded that, at least for some of its tumour-promoting and worse-survival functions, ZEB1 depends on its activation effect on DKK1 (figure 1A) and also on its repression of H2AFY (figure 8C).
Likewise, we also tested whether expression of GLB1, encoding for SA β-gal and whose mRNA levels increase in senescent cells,26 alters the predictor value of ZEB1 in CRC survival. Indeed, among patients with CRC with high levels of ZEB1, low levels of GLB1 determined poorer survival (figure 8D). Altogether, these results indicate that ZEB1 promotes tumour progression and determines worse prognosis in patients with CRC, at least in part, through its effect as an inhibitor of senescence.
ZEB1 triggers an EMT in invading cancer cells at the tumour invasive front conferring malignant cells a more motile and stem-like phenotype. However, beyond the induction of an EMT, the mechanisms through which ZEB1 promotes tumour initiation and progression and determines a worse survival in most cancers remain to be elucidated. Here, we found that the maximum effect of ZEB1 as a predictor of reduced survival in CRCs requires the joint expression of high levels of the Wnt antagonist DKK1 and low levels of H2AFY and GLB1. We showed that ZEB1 activates DKK1 by direct transcriptional activation of its promoter and that both genes inversely correlate with a number of senescence-inducing genes, whose expression they repress. Inhibition of senescence by ZEB1 and DKK1 depended on their repression of histone macroH2A1 and of SAHF formation through the subsequent induction of mutant p53, Mdm2 and CtBP (see model in figure 8E). Expression of ZEB1 and DKK1 at the invasive front of CRCs displays an inverse correlation with that of macroH2A1. In a mouse model of CRC, partial deletion of Zeb1 is sufficient to trigger senescence, SAHF formation and expression of macroH2A1. Importantly, using a tumour xenograft mouse model, we found that most of the tumourigenic effect of ZEB1 depends on its repression of macroH2A1.
Although Wnt antagonists are frequently silenced in human cancers, resulting in activation of the Wnt signalling and carcinogenesis, they can also have oncogenic roles.2 ,4 In this line, DKK1 is overexpressed in several carcinomas where it promotes tumour growth and metastasis4 and we found that high DKK1 expression is required for ZEB1 to impose its maximum impact on poorer survival in patients with CRC. In fact, DKK1 expression and function can be modulated during tumour progression. For instance, compared with normal prostate tissue, DKK1 expression increases in early malignant lesions and primary prostate carcinomas but declines in metastatic prostate carcinomas.51 In colorectal tumours, DKK1 is expressed at higher levels in Wnt-active CRCs than in adenomas and it is silenced by methylation in CRCs with low Wnt activity.52 As referred earlier, DKK1 is induced by Wnt signalling3 and our results here indicate that DKK1 is induced, along with ZEB1, in invasive malignant cells at the tumour front of CRCs. These invasive cells are characterised by the expression of nuclear β-catenin and by the display of active Wnt signalling and a proinvasive and stem-like phenotype.10
Our results also showed that DKK1 mediates the repression by ZEB1 of senescence-associated genes and of senescence. In SW480 cells, DKK1 increased endogenous levels of mutant TP53 and MDM2 and of the corepressor CTBP, thus enhancing ZEB1 transcriptional repression of H2AFY. Notably, wild-type and mutant p53 compete in the regulation of EMT: while wild-type p53 inhibits EMT through the direct transactivation of miR-200c, which in turn targets ZEB1 mRNA for decay, mutant p53 triggers an EMT.47 Our results in SW480 cells indicate that ZEB1, through DKK1, activates mutant TP53 and that ZEB1 is also under positive regulation by TP53, indicating the existence of a positive feedback loop between both transcription factors. Of note, EMT-activating and ZEB1-activating mutations of TP53 (at least, R175H and R273H) are common in primary CRCs and CRC cell lines, including those used in this study.42 ,46 Wild-type and mutant p53 vary in their target genes and also in their transcriptional activities for a given gene with some genes being regulated in different directions (eg, activation vs repression) and others to different degree (eg, deficient or enhanced activation).43
We showed here that DKK1, through induction of mutant p53 and Mdm2, activates CtBP. CtBP is overexpressed in human cancers and represses a number of tumour suppressors.53 As it does not bind to DNA, CtBP regulates gene expression through its recruitment by transcription factors, particularly ZEB1.11 ,53 To the extent that many functions of ZEB1—in cancer and beyond (eg,12 ,33)—depend on its repressor activity through CtBP, it could be expected that induction of CtBP by DKK1-mutant p53-Mdm2 may participate in other ZEB1-mediated processes beyond senescence such as regulation of cell differentiation, proliferation, apoptosis and/or cell migration.
Importantly, this study found that deletion of a single Zeb1 allele is sufficient to trigger senescence in cancer cells and therefore reduce tumour growth and improve survival in a mouse model of CRC. Our results here in CRC cells support earlier evidence that MEFs with heterozygous and homozygous deletion of ZEB1 display replicative senescence at earlier passages than wild-type ones.31 Senescence in Zeb1-deficient MEFs occurs in parallel with an upregulation of Cdkn2b and Cdkn1a (p21WAF1/CIP1), direct targets of ZEB1.13 ,31 We found that while DKK1 cooperates with ZEB1 in the repression of H2AFY, repression of CDKN2A and CDKN2B by ZEB1 is independent of DKK1 levels. Evidence in the extant literature for a role of DKK1 in senescence is inconclusive as DKK1 expression decreases in senescent endothelial cells but DKK1 mediates senescence in oesophagous epithelial cells,54 ,55 indicating once again that DKK1, as other Wnt antagonists, plays opposing functions under different physiological and pathological contexts. In contrast to DKK1, DKK3 failed to induce the mutant p53-Mdm2-CtBP pathway or to inhibit senescence in CRC cells. Since both DKK factors displayed similar inhibitory effect on Wnt signalling (ref 22 and online supplementary figure S1B), these results indicate that activation of the mutant p53-Mdm2-CtBP pathway and inhibition of macroH2A1 and senescence by DKK1 occurs independently of DKK1's role as a Wnt antagonist.
In contrast to ZEB1, we found that ZEB2 does not regulate DKK1. ZEB2 is structurally and functionally highly homologous to ZEB1, induces an EMT and has been associated with increased invasiveness in several cancers.10 ,23 However, ZEB2 represses hTERT in hepatocarcinoma and induces replicative senescence.56 ZEB1 and ZEB2 also play opposing roles in the regulation of TGFβ signalling during differentiation and development.12 ,13 In some cancers (eg, melanomas), ZEB2 functions as tumour suppressor displaying an inverse expression pattern with ZEB1.57 Therefore, it is possible that the divergent roles of ZEB factors in tumour progression are related to their opposing effects on senescence.
In a wide range of cancers, expression of macroH2A1 is reduced in tumours compared with their respective normal tissues and in invasive cancers with respect to cancer precursor lesions.32 ,49 Contrary to ZEB1, macroH2A1 suppresses tumour initiation and metastasis and associates with better survival. We found here that ZEB1 binds to the H2AFY promoter whose expression represses in a CtBP-dependent manner. In addition to the binding site we validated here, we identified other ZEB1 binding sites in the H2AFY gene in the ENCODE Project database (http://www.genome.gov/encode). We showed that ZEB1 and H2AFY display an inverse pattern of expression in CRCs and that downregulation of Zeb1 in a mouse model of CRC results in upregulation of macroH2A1. This correlation was validated at the mechanistic level in a xenograft model, where most of the tumourigenic activity of ZEB1 depended on its repression of macroH2A1. Of relevance, the worst survival effect of ZEB1 in patients with CRC depends on low levels of H2AFY.
Senescence is a key tumour suppressor mechanism that malignant cells must bypass to sustain tumour growth. Thus, while senescent cells are abundant in preneoplastic lesions, they disappear as tumours progress into malignant stages.5 Nevertheless, malignant cells can still undergo senescence if the expression and/or function of tumour suppressors are restored or that of oncogenes is inactivated. In that regard, senescence plays an important role in the response of cancer cells to many conventional chemotherapy agents and also a number of therapeutic strategies are currently being designed to specifically induce senescence.58 We show here that partial downregulation of Zeb1 is sufficient to trigger in vivo senescence in tumours and to reduce their growth and that the role of ZEB1 as determinant of worse prognosis in CRCs depends on low levels of senescence markers GLB1 and H2AFY. Therefore, ZEB1 could represent a new and important target in cancer therapy approaches aimed at inducing senescence.
This work has established a new role for ZEB1 as an inhibitor of senescence in cancer cells, thus expanding the set of cancer cell hallmarks regulated by ZEB1. By dissecting the molecular mechanisms by which ZEB1 contributes to overcome this cancer safeguard programme, the results presented here offer new opportunities to target the tumour-promoting functions of ZEB1.
We are grateful to all researchers who provided us with reagents (see Methods and online supplementary information) and regret that some articles were only cited indirectly through reviews due to space limitations. The bulk of the study was conducted in the Centre de Recerca Biomèdica Cellex at IDIBAPS.
Contributors OdB performed most of the experimental work in the study, designed and interpreted experiments and wrote the manuscript. BG carried out the bioinformatics analysis of arrays for survival and gene expression. MJFA facilitated the collection of human samples of colorectal carcinomas and conducted the pathological analysis of their immunostaining. EST, LSM and LS performed some of the experimental work. AEA and GR assisted in the setting of the xenograft mouse model. JIC, DSD and AC supplied critical materials to the study. AP conceived and supervised the study, designed and interpreted experiments, obtained funding and wrote the manuscript. All authors critically reviewed the manuscript.
Funding The different parts of this study were independently funded by grants to AP from Fundació La Marató de TV3 (201330.10), Ministry of Economy and Competitiveness (BFU2010-15163 and SAF2014-52874-R, the latter part of the 2013–2016 National Scientific and Technical Research and Innovation Plan, that is co-financed by the European Regional Development Fund of the European Union Commission), Avon Foundation S.A.U. (ACSAU), Catalan Agency for Management of University and Research Grants (AGAUR, 2014-SGR-475), Leukemia Research Foundation (Hollis Brownstein Grants 2014), Academy of Medical and Health Sciences of Catalonia and the Balearic Islands (Recerca Bàsica 2013), European Commission of the European Union, Olga Torres Foundation, La Caixa Foundation (LCF) and Spanish Association Against Cancer. OdB's stipend was subsequently funded by a PhD scholarship from the Spanish Ministry of Education, Culture and Sports (MECS) (FPU programme, AP2010-5489), AGAUR and a scholarship from University of Barcelona. EST's salary was subsequently funded by the grants from LCF and ACSAU, a CIBERehd postdoctoral contract and a Miguel Servet contract (MS13/00200) from Instituto de Salud Carlos III. LSM, LS and AEA are also recipients of PhD scholarships from the MECS (FPU14/06217, AP2010-4495 and FPU12/05690, respectively).
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.