MZF1 mediates oncogene-induced senescence by promoting the transcription of p16INK4A.

Oncogene induced senescence is a tumor suppressing defense mechanism, in which the cell cycle-dependent protein kinase (CDK) inhibitor p16INK4A (encoded by the CDKN2A gene) plays a key role. We previously reported that a transcriptional co-activator chromodomain helicase DNA binding protein 7 (CHD7) mediates oncogenic ras-induced senescence by inducing transcription of the p16INK4A gene. In the current study, we identified myeloid zinc finger 1 (MZF1) as the transcriptional factor that recruits CHD7 to the p16INK4A promoter, where it mediates oncogenic ras-induced p16INK4A transcription and senescence through CHD7, in primary human cells from multiple origins. Moreover, the expression of MZF1 is induced by oncogenic ras in senescent cells through the c-Jun and Ets1 transcriptional factors upon their activation by the Ras-Raf-1-MEK-ERK signaling pathway. In non-small cell lung cancer (NSCLC) and pancreatic adenocarcinoma (PAAD) where activating ras mutations occur frequently, reduced MZF1 expression is observed in tumors, as compared to corresponding normal tissues, and correlates with poor patient survival. Analysis of single cell RNA-sequencing data from PAAD patients revealed that among the tumor cells with normal RB expression levels, those with reduced levels of MZF1 are more likely to express lower p16INK4A levels. These findings have identified novel signaling components in the pathway that mediates induction of the p16INK4A tumor suppressor and the senescence response, and suggested that MZF1 is a potential tumor suppressor in at least some cancer types, the loss of which contributes to the inactivation of the p16INK4A/RB pathway and disruption of senescence in tumor cells with intact RB.

Introduction

Replicative senescence (RS) is a stable proliferative arrest resulted from exhaustion of replicative potential of normal cells [1]. Senescence can also be induced prematurely by cellular stresses, including oncogenes such as activated ras (termed oncogene induced senescence, or OIS) [1]. Senescence is accompanied by biomarkers, including senescence-associated-β-galactosidase (SA-β-Gal) activity, senescence-associated secretory phenotypes (SASP), senescence-associated heterochromatin foci (SAHF), telomere shortening, DNA damage responses, and altered expression of cell cycle inhibitors, such as p16INK4A, p53 and p21WAF1 [1-5]. p16INK4A/retinoblastoma protein (RB) and p53/p21WAF1 are two important effector pathways that cooperatively and independently regulate senescence [6, 7]. Senescence is also associated with global 3D genome reorganizations, such as impaired heterochromatic topologically associating domains (TADs) representing self-interacting genomic regions, altered genomic contacts and increased global chromosome density [8, 9].

Senescence is a double-edged sword in cancer development. OIS is a defense mechanism against malignant transformation, which suppresses proliferation of cells with activated oncogenes. In both human and mouse cancer models, senescence occurs in premalignant lesions associated with oncogenic mutations in K-ras, BRAF, PTEN and NF1, but is absent in malignant tumors [10, 11] [11, 12]. OIS thus represents a barrier to tumorigenesis, the disruption of which accelerates cancer development. On the other hand, senescent cells can enhance cancer development via SASP, which promotes formation of tumor amiable microenvironment [13-16]. Despite advancements in recent years, the signaling pathways that mediate senescence induction have not been fully delineated.

Myeloid zinc finger 1 (MZF1) is a member of the SCAN domain containing zinc finger transcriptional factor family. It functions as either a transcription activator or transcription repressor. There are two alternatively-spliced MZF1 isoforms in human [17-19]. The long isoform of MZF1 has 734 amino acids containing an N-terminal SCAN domain mediating dimerization or oligomerization, a linker regulatory region containing phosphorylation sites, and 13 C-terminal zinc finer (ZF) motifs that bind DNA [19, 20]. The 290-amino-acid short isoform consists of the SCAN domain and a regulatory domain that differs from that of the long isoform by an alternatively spliced exon, but without DNA binding domain. MZF1 was initially identified as a tumor suppressor in hematopoietic malignancy, as MZF1 knockout in mice promoted autonomous proliferation and long-term hemopoiesis in hemopoietic progenitor cells, and led to development of lethal neoplasias characterized by infiltration and disruption of liver architecture by cells of myeloid origin reminiscent of human chloromas [21]. In cultured cancer cells, MZF1 can either suppress [22-25] or promote [19–21, 26–31] tumorigenic phenotypes such as cell proliferation, migration, invasion and xenograft tumor formation in a cancer type- and context-depend manner.

We reported previously that a transcriptional co-activator chromodomain helicase DNA binding protein 7 (CHD7) mediates oncogenic ras-induced senescence by inducing transcription of the p16INK4A gene upon binding to the p16INK4A promoter [32]. CHD7 is a chromatin remodeling factor with an ATP-dependent helicase activity, but without DNA binding domain[33]. It can be recruited to enhancers by methylated histone 3-Lys4 (H3K4me) or to promoters by transcription factors such as Oct4, Sox2, Sox9, Sox4, Sox11, Twist, p300 and PBAF [34-37]. In this study, we identified the long isoform of MZF1 as the transcriptional factor that recruits CHD7 to the p16INK4A promoter and as an essential mediator of oncogenic ras-induced p16INK4A transcription and senescence. Moreover, the expression of MZF1 is induced by oncogenic ras in senescent cells through c-Jun and Ets1 transcriptional factors upon their activation by the Ras-Raf-1-MEK-ERK pathway.

Activating mutations of K-ras are the most frequent mutations and initiating genetic event in pancreatic cancer, and also occur at high frequency (~30%) in non-small cell lung cancer (NSCLC) [11, 38, 39]. We found that MZF1 expression was reduced in tumors from patients with pancreatic adenocarcinoma (PAAD) or NSCLC as compared to normal tissues, and that reduced MZF1 expression correlated with poor survival in these patients. Thus, MZF1 is a potential tumor suppressor in PAAD and NSCLC. Analysis of single-cell RNA-sequencing data from PAAD patients revealed that MZF1 expression was reduced in tumor cells as compared to normal ductal cells, and that in tumor cells with intact RB1 expression, those with reduced MZF1 are more likely to express lower p16INK4A, suggesting that loss of MZF1 expression is at least one mechanism responsible for inactivation of the p16INK4A-RB pathway in PAAD.

Our findings have thus identified MZF1 as a novel regulator of p16INK4A and OIS, which may act as a tumor suppressor in certain cancers.

Results

MZF1 mediates oncogenic ras-induced recruitment of CHD7 to the p16INK4A promoter and CHD7-stimulated p16INK4A transcription

We previously reported that an ATP-dependent helicase CHD7 mediates oncogenic ras-induced p16INK4A transcription [32]. As CHD7 is a transcriptional coactivator without a DNA-binding domain [40], we hypothesize that CHD7 is recruited to the p16INK4A promoter by a transcription factor. We thus searched for the consensus binding sites for transcription factors in the 119 ~ +20 region containing the CHD7 binding peak [32] on the p16INK4A promoter, using the JASPAR database. We found that this region contained binding sites for multiple transcription factors (Table S1). In order to narrow down the candidates for the transcription factors that regulate p16INK4A transcription, we examined expression of these transcription factors and p16INK4A (CDKN2A) in senescent cells in datasets from the GEO database, including one data profile comparing replicative senescent cells and control growing cells and 8 profiles comparing cells undergoing oncogene-induced senescence and control growing cells. MZF1 was the only transcriptional factor consistently upregulated together with CDKN2A/p16INK4A in all these 9 GEO profiles (Fig. S1). We thus focused on MZF1 and investigated its role in oncogenic ras-induced p16INK4A transcription and senescence.

We introduced either deletion or point mutations into the 3 MZF1 binding sites in a luciferase reporter driven by a 1kb p16INK4A promoter (Fig. 1A), which was induced by CHD7 and oncogenic ras [32]. In 293T cells, deletions or point mutations of the MZF1 binding sites (Fig. 1B) and knockdown of MZF1 (Fig. 1C) reduced stimulation of transcription from the p16INK4A-luciferase reporter by cotransfected CHD7, as compared to the reporter containing wild type MZF1 binding sites and the shRNA control (SC), respectively. Similarly, in BJ primary fibroblast cells stably transduced with the p16INK4A luciferase reporters, oncogenic ras stimulated transcription for the wild type p16INK4A promoter, which was abrogated by deletions or point mutations in the MZF1 binding sites (Fig. 1D). The MZF1 binding sites are thus required for ras- and CHD7-induced p16INK4A transcription.

Fig. 1.

MZF1 mediates oncogenic ras-induced recruitment of CHD7 to the p16INK4A promoter and CHD7-stimulated p16INK4A transcription.

(A) A diagram of the p16INK4A promoter luciferase reporter used in the study, showing the CHD7 binding peak (−119 ~ −12), MZF1 binding sites (site 1–3, #) and transcription start site (TSS).

(B) The p16INK4A promoter luciferase reporter with wild type MZF1 binding sites (WT) or with deletion mutations (Del) or point mutations (Mut) in the MZF1 binding sites were transfected to 293T cells with pcCDNA3-CHD7 or vector.

(C) The wild type p16INK4A promoter luciferase reporter was transfected with pCDNA3-CHD7 or vector into 293T cells transduced with shRNA control (SC) or MZF1 shRNAs targeting both long and short isoforms (sh-T1 and sh-T2). Insert shows knockdown efficiency of the shRNAs, measured by qRT-PCR.

(D) BJ cells stably transduced with the wild type p16INK4A promoter reporter (WT) or that containing deletion mutations (Del) or point mutations (Mut) in the MZF1 binding sites were transduced with HaRasV12 (Ras) or vector (WH).

(B-D) Luciferase activity was measured 48h after transfection (B-C) or 7 days after ras transduction (D) and normalized to protein concentrations determined by BCA assays. Relative p16-Luc activity was calculated by dividing values of CHD7 or Ras cells by that of vector control cells. Values are means±SDs for triplicates. ** p < 0.01 vs WT (B, D) or SC (C) in Student’s t test.

(E) Chromatin was immuneprecipitated from BJ cells transduced with HaRasV12 (Ras) or vector (WH) using an anti-CHD7 or anti-MZF1 antibody or control IgG (rabbit IgG).

(F) Chromatin was immuneprecipitated from BJ cells transduced with the long MZF1 isoform (MZF1-L) or vector (pLv) using an anti-CHD7 antibody or control IgG (rabbit IgG). Insert shows overexpression of the long MZF1 isoform as determined by Western blotting.

(E-F) DNA present in immunoprecipitated complexes was quantified by real-time PCR using primers that amplify the −119 ~ +50 region of p16INK4A promoter and normalized to input. Values are means±SDs for triplicates. ** p < 0.01 vs vector (WH or pLv) by Student’s t test.

(G-H) Expression levels of the short (G) and long (H) MZF1 isoforms in BJ cells transduced with shRNA control (SC) or shRNAs specific for the long (sh-L), short (sh-S) or both (sh-T1 and sh-T2) isoforms, as determined by qRT-PCR. Values are means±SDs for triplicates. ns, not significant, p > 0.05; ** p < 0.01 vs SC by Student’s t test.

(I) Expression levels of the MZF1 isoforms in BJ cells transduced with shRNA control (SC) or shRNAs specific for the long (sh-L), short (sh-S) or both (sh-T1 and sh-T2) isoforms, as determined by Western blot analysis.

(J) Chromatin was immuneprecipitated from BJ cells transduced with shRNA control (SC) or MZF1 shRNAs targeting both isoforms (sh-T1) or the long isoform (sh-L) and with HaRasV12 (Ras) or vector (WH), using an anti-CHD7 antibody or control rabbit IgG (IgG). DNA present in immunoprecipitated complexes was quantified by real-time PCR using primers that amplify the −119 ~ +50 region of the p16INK4A promoter and normalized to input. Values are means±SDs for triplicates. ** p < 0.01 vs SC by Student’s t test.

(K) Western blot analysis of CHD7 and MZF1 co-immunoprecipitated by an anti-MZF1 antibody or control IgG (top) or an anti-CHD7 antibody or control IgG (bottom) from BJ cells transduced with HaRasV12 (Ras) or vector (WH).

Using chromatin immunoprecipitation coupled with quantitative real time PCR (ChIP-qPCR), we showed that in BJ cells, both MZF1 and CHD7 were enriched in the −119 ~ +50 region of the p16INK4A promoter, a region containing the previously identified CHD7 binding peak [32], and that binding of both proteins to this region was increased by oncogenic ras (Fig. 1E). Moreover, in BJ cells, CHD7 enrichment in the −119~+50 region of the p16INK4A promoter was increased by the ectopically expressed long isoform of MZF1 (Fig. 1F). We constructed shRNAs silencing both long and short isoforms (sh-T1 and sh-T2), only the long isoform (sh-L), or only the short isoform (sh-S) of MZF1 (Fig. 1G–I). CHD7 enrichment in the −119~+50 region of the p16INK4A promoter was decreased by shRNAs that silenced the expression of both isoforms (shT-1) or only the long isoform of MZF1 (shL) (Fig. 1J). CHD7 co-immunoprecipitated with MZF1 by an anti-MZF1 antibody, and MZF1 co-immunoprecipitated with CHD7 by an anti-CHD7 antibody (Fig. 1K), indicating that endogenous MZF1 and CHD7 interact with each other. The interaction between MZF1 and CHD7 was enhanced by oncogenic ras, likely because ras increased the expression of these proteins (Fig. 1K). Therefore, CHD7 is recruited to the p16INK4A promoter by MZF1 upon ras activation.

Taken together, our results indicate that the ability of CHD7 to mediate ras-induced p16INK4A transcription relies on its recruitment to the p16INK4A promoter by MZF1.

MZF1 mediates oncogenic ras-induced p16INK4A transcription.

We investigated the role of MZF1 in p16INK4A transcription during OIS. Like oncogenic ras, ectopic expression of the long, but not the short, MZF1 (Fig. 2A–C), induced the expression of p16INK4A mRNA and protein in BJ cells (Fig. 2D–E). While oncogenic ras induced p16INK4A expression at both mRNA and protein levels, MZF1 shRNAs (sh-T1 and sh-T2) that knocked down both MZF1 isoforms (Fig. 1G–I, 2F) abrogated ras-induced p16INK4A expression (Fig. 2G–H). Neither ectopic expression nor knockdown of MZF1 altered levels of total p53 or phosphorylation of p53 at Ser15, one of the key activation sites, in BJ cells with or without ras (Fig. 2E, H), suggesting that MZF1 specifically mediates induction of p16INK4A, but not p53, by oncogenic ras.

Fig. 2.

The long MZF1 isoform mediates oncogenic ras-induced p16INK4A transcription.

(A-C) BJ cells transduced with the long (MZF1-L) or short (MZF1-S) MZF1 isoform or vector (pLv) were analyzed for MZF1 expression by qRT-PCR (A and B, for the long and short isoforms, respectively) and Western blotting (C). Values are means±SDs for triplicates. ** p<0.01 vs pLv by Student’s t test.

(D-E) BJ cells transduced with the long (MZF1-L) or short (MZF1-S) MZF1 isoform or vector (pLv) were analyzed for p16INK4A mRNA levels by qRT-PCR (D) and for p16INK4A and phospho- and total p53 proteins by Western blotting (E). Values are means±SDs for triplicates. ** p < 0.01 vs pLv by Student’s t test.

(F) BJ cells transduced with shRNA control (SC) or shRNAs targeting both MZF1 isoforms (sh-T1 and sh-T2) were analyzed for knockdown efficiency by qRT-PCR. Values are means±SDs for triplicates. ** p < 0.01 vs SC by Student’s t test.

(G-H) BJ cells transduced with shRNA control (SC) or shRNAs targeting both MZF1 isoforms (sh-T1 and sh-T2) and HaRasV12 (Ras) or vector (WH) were analyzed p16INK4A mRNA level by qRT-PCR (G) and protein levels by Western blotting (E). Values are means±SDs for triplicates. ** p < 0.01 vs SC by Student’s t test.

(I) The wild type p16INK4A promoter luciferase reporter was co-transfected with the long (MZF1-L) or short (MZF1-S) MZF1 isoform, CHD7 or vector (pLv or V) into 293T cells.

(J-K) The wild type (WT) p16INK4A promoter luciferase reporter or that with deletion mutations (Del) or point mutations (Mut) in the MZF1 binding was transfected into 293T cells with the long (MZF1-L) (J) or short (MZF1-S) (K) MZF1 isoform.

(L) BJ cells stably transduced with the wild type p16INK4A promoter luciferase reporter was transduced with shRNA control (SC) or MZF1 shRNAs (sh-T1 and sh-L).

(I-L) Luciferase activity was measured 48h after transfection (I-K) or 7 days after ras transduction (L) and normalized to protein concentrations determined by BCA assays. Relative p16-luc activity was calculated by dividing values of MZF1-L, MZF1-S, CHD7 or Ras cells by that of vector control cells. Values are means±SDs for triplicates. ns, not significant, p > 0.05; * p < 0.05; ** p < 0.01 vs pLv (I), WT (J-K) or SC (L) by Student’s t test.

Using p16INK4A promoter-driven luciferase reporters, we tested whether MZF1 mediated p16INK4A expression at the transcription level. In 293T cells, cotransfection of CHD7 or the long but not the short MZF1 stimulated transcription from the p16INK4A reporter (Fig. 2I). Induction of p16INK4A transcription by the long MZF1 was abolished by deletions or point mutations in the MZF1-binding sites on the p16INK4A promoter, while p16INK4A transcription in the presence of the short MZF1 was unaltered by the deletions or point mutations in the p16INK4A promoter (Fig. 2J–K). Moreover, in BJ cells, ras-induced transcription from the p16INK4A reporter was abrogated by deletions or point mutations of the MZF1-binding sites in the p16INK4A promoter (Fig. 1D) and by MZF1 shRNAs that silenced both isoforms (sh-T1) or the long isoform (sh-L) of MZF1 (Fig. 2L).

Thus, the long MZF1 isoform mediates ras-induced transcription of p16INK4A.

MZF1 mediates oncogenic ras-induced senescence through CHD7.

Consistent with the importance of p16INK4A and p53 in senescence, ras-induced senescence was disrupted by the knockdown of p16INK4A or p53 (Fig. S2A–D). We thus investigated whether MZF1 was also important for ras-induced senescence. In control BJ cells (SC), Ha-rasV12 induced proliferative arrest and accumulation of SA-β-gal, indicative of senescence induction; shRNAs silencing both long and short isoforms (sh-T1 and sh-T2) or only the long isoform (sh-L) of MZF1, but not that targeting the short isoform (sh-S) (Fig. 1G–I), abrogated Ha-rasV12-induced senescence (Fig. 3A–D), demonstrating that the long MZF1 is essential for OIS. Eectopic expression of the long (MZF1-L) but not the short (MZF1-S) MZF1 isoform also induced senescence, similar to Ha-rasV12 (Fig. 3E–F).

Fig. 3.

MZF1 mediates oncogenic ras-induced senescence through CHD7.

(A) Growth curves of BJ cells transduced with shRNA control (SC) or shRNAs targeting both MZF1 isoforms (sh-T1 and sh-T2) and HaRasV12 (Ras) or vector (WH).

(B) Percentage of SA-β-Gal positive cells in populations in (A) on day-12 of the growth curves.

(C) Growth curves of BJ cells transduced with shRNA control (SC) or MZF1 shRNAs targeting both isoforms (sh-T1), or the long (sh-L) or short (sh-S) isoform, and HaRasV12 (Ras) or vector (WH).

(D) Percentage of SA-β-Gal positive cells in populations in (C) on day-12 of the growth curves.

(E) Growth curves of BJ cells transduced with the long (MZF1-L) or short (MZF1-S) MZF1 isoforms or vector (pLv).

(F) Percentage of SA-β-Gal positive cells in populations in (E) on day-12 of the growth curves.

(G) BJ cells transduced with shRNAs for CHD7 or shRNA control (SC) were analyzed for knockdown efficiency by qRT-PCR.

(H) Growth curves of BJ cells transduced with control shRNA (SC) or CHD7 shRNAs (sh2 and sh3) and the long MZF1 isoform (MZF1-L) or vector (pLv).

(I) Percentage of SA-β-Gal positive cells in populations in (H) on day-12 of the growth curves.

(A-I) Values are means±SDs for triplicates. ns, not significant, p > 0.05; ** p < 0.01 vs SC-Ras (A-D), WH/pLv (E-F), SC (G) or SC-MZF1-L (H-I) by Student’s t test.

(J) BJ cells transduced with control shRNA (SC) or CHD7 shRNAs (sh2 and sh3) and the long MZF1 isoform (MZF1-L) or vector (pLv) were analyzed for p16INK4A expression by Western blotting.

Similar observations were made in HSAEC and primary human pancreatic cells (Fig. S3, S4), in that the long but not the short of MZF1 induced p16INK4A protein and mRNA levels and senescence (Fig. S3A–D, S4A–D), while the MZF1 shRNAs silencing both isoforms or only the long isoform, but not that silencing only the short isoform, abrogated Ha-RasV12-induced p16INK4A expression and senescence (Fig. S3E–H, S4E–H)

Thus, the long MZF1 isoform is required for induction of p16INK4A expression and senescence by ras, and is also sufficient to induce p16INK4A expression and senescence, though its effect was not as robust as that of oncogenic ras (Fig. 3E–F), likely because posttranslational modifications, beside increased expression, induced by ras may also contribute to MZF1 function in senescence.

Furthermore, induction of senescence and p16INK4A expression by the long MZF1 was abrogated by knockdown of CHD7 in BJ cells (Fig. 3G–J), which indicates that MZF1 acts upstream of CHD7 to mediate senescence induction.

Oncogenic ras induces MZF1 expression through Raf1-MEK1-ERK.

Besides CHD7 [32], Ha-rasV12 also increased expression of MZF1 isoforms at mRNA and protein levels in BJ cells (Fig. 4A–B), consistent with the upregulation of CDKN2A/p16INK4A and MZF1 in cells undergoing RS or OIS in GEO datasets (Fig. S1). Thus, ras may induce MZF1 expression, which in turn mediates p16INK4A transcription and senescence.

Fig. 4.

Oncogenic ras induces MZF1 expression through the Raf1-MEK1-ERK pathway.

(A-B) BJ cells transduced with HaRasV12 (Ras) or vector (WH) were measured for MZF1 mRNA and protein levels by qRT-PCR (A) and Western blotting (B), respectively. Positions of the long (MZF1-L) and short (MZF1-S) MZF1 isoforms are marked by arrows.

(C) Western blot analysis of MZF1 levels in BJ cells on day-12 after transduction of HaRasV12 (Ras), constitutively active forms of the indicated Ras downstream effectors, or vector controls (BP or WN). Positions of the long (MZF1-L) and short (MZF1-S) MZF1 isoforms are marked by arrows.

(D-E) Cells in (C) were measured for the mRNA levels of the long (MZF1-L) (D) and short (MZF1-S) (E) MZF1 isoforms by qRT-PCR.

(F) Western blot analysis of BJ cells transduced with HaRasV12 (Ras) or vector (WH) and treated with 10 μM of U0126 or vehicle for 24h starting on day-10 after ras transduction.

(A, D-E) Values are means±SDs for triplicates. ns, not significant, p > 0.05; **p<0.01 vs WH (A) or BP/WN (D-E) by Student’s t test.

Global genome organization is altered in senescent cells and is associated with altered expression of senescence-regulating genes [8, 41, 42]. We analyzed the GSE118494 dataset containing in situ Hi-C and ChIP-seq data of growing IMR90 cells and those undergoing OIS, to profile DNA-DNA and DNA-protein interactions in chromosomal regions near the MZF1 and CDKN2A/p16INK4A genes [8]. Analysis of the Hi-C data revealed that contact between the CDKN2A gene and its neighboring regions was significantly decreased in senescent cells compared to growing cells (Fig. S5A–B). The same trend was observed with the MZF1 gene (Fig. S5C–D). Thus, the chromatin structures surrounding MZF1 and CDKN2A/p16INK4A genes are reorganized to a state consistent with active transcription in OIS cells [43].

Condensin II and cohesin are protein complexes consisting of structure maintenance of chromosomes (SMC) proteins, which regulate gene expression via high-order genome organization [41, 44]. Condesin II reinforces euchromatic A compartments where highly expressed genes are enriched, while the condensing II subunit CAP-H2 and cohesion (SMC1) are preferentially localized to active promoters and enhancers [8, 45]. In senescent cells, CAP-H2 and SMC1 are enriched at upregulated, senescence-associated genes, including those encoding SASP factors and p21WAF1, as compared to proliferating cells, and are required for their upregulation and for senescence [8, 44]. Analysis of ChIP-seq data in GSE 118494 indicated that endogenous CAP-H2, ectopically expressed CAP-H2-FLAG and SMC1 were highly enriched near the CDKN2A and MZF1 genes in OIS cells, as compared to proliferating cells (Fig. S5E–F), consistent with the reported predominant localization of condensing II and cohesin at active promoters and enhancers [8, 46] and the higher CDKN2A and MZF1 expression in senescence cells than in proliferating cells (Fig. 4A–B, S1). Collectively, these results suggest that both MZF1 and CDKN2A/p16INK4A genes are localized to chromosomal compartments undergoing active expression.

Oncogenic ras activates multiple downstream effector pathways, including Raf1-MEK-ERK, MKK3/6-p38, MKK4/7-JNK, and PI3K-AKT [40]. We analyzed the effects of constitutively active forms of these effectors [40] on MZF1 expression in BJ cells. Active Raf-1 and MEK1 induced expression of the long and short MZF1 isoforms at protein (Fig. 4C) and mRNA levels (Fig. 4D–E). In contrast, active MKK3, MKK4, MKK7, PI3K and AKT failed to increase the long MZF1 expression at protein or mRNA levels (Fig. 4C–D). Active MKK4, MKK7, PI3K and AKT increased expression of the short MZF1 (Fig. 4C, E), which does not contribute to ras-induced p16INK4A expression or senescence (Fig. 2D–E, 2I, 3C–F). Moreover, a MEK1 inhibitor U0126 abrogated ras-induced upregulation of MZF1 and p16INK4A (Fig. 4F). Thus, oncogenic ras induces MZF1 expression through the Raf-1-MEK-ERK pathway.

C-Jun and ETS1 are essential for ras-induced MZF1 transcription.

Previous studies showed that ras activates c-Jun and ETS1 transcription factors through phosphorylation mediated by MEK-ERK [47]. Indeed, in BJ cells, Ha-rasV12 induced phosphorylation of c-Jun-S73 and ETS1-T38, the ERK substrate sites[47]; U0126 abrogated ras-induced phosphorylation of these sites (Fig. 4E), confirming that oncogenic ras induces c-Jun, and ETS1 phosphorylation in a MEK1/ERK-dependent fashion.

shRNA-mediated silencing of ETS1 (Fig. 5A–C) or c-Jun (Fig. 5D–F) disrupted induction of MZF1 expression by ras at both mRNA (Fig. 5B, E) and protein (Fig. 5C, F) levels in BJ cells. Ectopic expression of TAM67, a dominant negative c-Jun mutant lacking the transactivation domain but retaining the DNA binding domain [47], also reduced induction of MZF1 mRNA and protein by Ha-rasV12 (Fig. 5G–H). Thus, c-Jun and ETS1 are required for ras-induced MZF1 expression.

Fig. 5.

Oncogenic ras-induced MZF1 transcription requires c-Jun and Ets1.

(A) Relative Ets1 mRNA levels in BJ cells transduced with shRNA control (SC) or ETS1 shRNAs (sh247 and sh257) as measured by qRT-PCR.

(B-C) BJ cells transduced with a shRNA control (SC) or Ets1 shRNAs (sh247 and sh257) and HaRasV12 (Ras) or vector (WH) were measured for MZF1 mRNA and protein levels by qRT-PCR (B) and Western blotting (C), respectively, on day-10 after ras transduction.

(D) Relative c-Jun mRNA levels in BJ cells transduced with shRNA control (SC) or a c-Jun shRNA (shJun1555) as measured by qRT-PCR.

(E-F) BJ cells transduced with a shRNA control (SC) or a c-Jun shRNA (shJun1555) and HaRasV12 (Ras) or vector (WH) were measured for MZF1 (MZF1-L) mRNA and protein levels by qRT-PCR (E) and Western blotting (F), respectively, on day-10 after ras transduction.

(G-H) BJ cells transduced with a dominant negative mutant of c-Jun (TAM67) or vector (BP) and HaRasV12 (Ras) or vector (WH) were measured for MZF1 (MZF1-L) mRNA and protein levels by qRT-PCR (G) and Western blotting (H), respectively, on day-10 after ras transduction.

(A-B, D-E, G) Values are means±SDs for triplicates. ** p < 0.01 vs SC (A, D), SC-Ras (B, E) or BP-Ras (G) by Student’s t test.

To determine the roles of c-Jun and ETS1 on ras-induced MZF1 transcription, we constructed luciferase reporters driven by a 2.5 kb (−2512 ~ +130 relative to the TSS) or 1.5 kb (−1500 ~ +130) region of the MZF1 promoter (Fig. 6A). When co-transfected with Ha-rasV12 into 293T cells, the 1.5 kb promoter was sufficient to mediate the ras responsiveness (Fig. 6B). We generated a retroviral luciferase reporter for the 1.5kb MZF1 promoter and transduced it into BJ cells to generate a stable MZF1 promoter reporter cell line. Knockdown of ETS1 or c-Jun (Fig. 6C) or ectopic expression of TAM67 (Fig. 6D) disrupted ras-induced transcription from this 1.5kb promoter, indicating requirement of c-Jun and ETS1 for ras-induced MZF1 transcription.

Fig. 6.

The c-Jun and Ets1 binding sites on the MZF1 promoter are required for oncogenic ras-induced MZF1 transcription.

(A) Diagrams of the MZF1 promoter luciferase reporters used in the study, showing positions of the predicted ETS1 (ETS1-A and ETS1-B) and c-Jun binding sites (c-Jun-A and c-Jun-B), the transcriptional start site (TSS), and the amplicons used in ChIP-qPCR in (E-F). The actual ETS1 and c-Jun binding sites (ETS1-A, c-Jun-A and c-Jun-B) confirmed by ChIP are shown in bold.

(B) 293T cells were co-transfected with the 2.5kb or 1.5kb MZF1 promoter reporter and pCDNA3-HaRasV12 (Ras) or vector (pCDNA3).

(C-D) BJ cells stably transduced with the 1.5kb MZF1 promoter reporter were transduced with a shRNA control (SC) or shRNAs for ETS1 (sh247 and 257) or Jun (shJun1555), or TAM67 or vector (BP), and HaRasV12 (Ras) or vector (WH).

(B-D) Luciferase activity was measured 48h (B) or 10 days (C-D) after ras transfection (B) or transduction (C-D) and normalized to protein concentrations determined by BCA assays. Relative MZF1 promoter activity was calculated by dividing values of Ras cells by that of vector (pCDNA3 or WH) cells

(E-F) ChIP analysis of Ets1 and c-Jun binding to the MZF1 promoter. Chromatin was immunoprecipitated from BJ cells transduced with HaRasV12 (Ras) or vector (WH) using an anti-ETS1 (E) or anti-c-Jun (F) antibody or control rabbit IgG (IgG). DNA present in immunoprecipitated complexes was quantified by qPCR using primers amplifying the −900~−729 and −422~−242 regions (A) of the MZF1 promoter and normalized to inputs.

(G) BJ cells stably transduced with the wild type 1.5kb MZF1 promoter reporter (WT) or that containing point mutations in the c-Jun binding sites (MUT-c-Jun-A, MUT-c-Jun-B and MUT-c-Jun-A-B) or the ETS1 binding site (MUT-ETS1-A) were transduced HaRasV12 (Ras) or vector (WH). Luciferase activity was measured 10 days after ras transduction and normalized to protein concentrations determined by BCA assays. Relative MZF1 promoter activity was calculated by dividing values of Ras cells by that of WH cells.

(B-G) Values are means±SDs for triplicates. ns, not significant, p > 0.05; *p < 0.05; ** p < 0.01 vs 1.5k-Ras (B), SC-Ras (C), BP-Ras (D), WH-Ets1 (E), WH-c-Jun (F) or WT-Ras (G) by Student’s t test.

A search in JASPAR identified 2 c-Jun binding sites (c-Jun-A, −744 ~ −730 and c-Jun-B, −333 ~ −323) and 2 ETS1 binding sites (ETS1-A, −886 ~ −877 and ETS1-B, −256 ~ −247) in this 1.5 kb promoter region (Table S2, Fig. 6A). We analyzed c-Jun and ETS1 binding to these sites by ChIP-qPCR using primers amplifying the −900~−729 region containing c-Jun-A and ETS1-A and primers amplifying the −422~−242 region containing c-Jun-B and ETS1-B. ETS1 bound to ETS1-A, and c-Jun bound to both c-Jun-A and c-Jun-B in a ras-dependent manner, while ETS1 did not bind to ETS1-B either in the presence or absence of oncogenic ras, on the MZF1 promoter (Fig. 6E–F). Point mutations that disrupted the consensus binding sites in c-Jun-A, c-Jun-B, c-Jun-A-B or ETS1-A (Fig. 6A, Table S2) abrogated Ha-rasV12-inducted transcription from the stable 1.5kb-MZF1 promoter reporter in BJ cells (Fig. 6G).

Thus, oncogenic ras induces MZF1 transcription by activating the ETS1 and c-Jun transcriptional factors through the Raf-1-MEK1-ERK pathway.

MZF1 is a potential tumor suppressor in non-small lung cancer (NSCLC) and pancreatic adenocarcinoma (PAAD)

Analysis of data from the Cancer Genome Atlas (TCGA) database indicated downregulation of MZF1 expression in NSCLC, including lung adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC) (Fig. 7A, S6), and other cancer types such as breast cancer (BRCA), head and neck cancer (HNSC), kidney chromophobe renal cell carcinoma (KICH), kidney papillary renal cell carcinoma (KIRP), and thyroid carcinoma (THCA) (Fig. S6), as compared to paired normal tissues.

Fig. 7.

Reduced MZF1 expression in tumors correlates with poor patient survival in NSCLC and PAAD.

(A) MZF1 expression is decreased in tumor tissues of LUAD and LUSC, the 2 major subtypes of NSCLC, as compared with paired non-tumor tissues.

(B) Reduced MZF1 expression correlates with poor overall survival in LUAD (left) and LUSC (right) patients based on Kaplan-Meier analysis of GEO data.

(C) MZF1 expression is reduced in tumors from stage II and II/IV PAAD patients as compared to those from stage I patients.

(A, C) Logarithmically (log2) transformed MZF1 expression levels based on RNA sequencing data from PAAD patients in TCGA were compared by paired t-tests between tumor and normal tissues (A) or among tumors of different stages determined at diagnosis (C).

(D) MZF1 expression is decreased in tumors as compared to adjacent non-tumor tissues from PAAD patients, based on analysis of GEO datasets GSE101448 (left), GSE62452 (middle) and GSE71729 (right). Box plots represent quartile of normalized gene expression levels with relative intensity (log2) of microarray data from PAAD patients in these GEO datasets that compared between tumor and normal tissues.

(E) Reduced MZF1 expression correlates with poor overall survival (left) and disease-free survival (right) in PAAD patients based on Kaplan-Meier analysis of TCGA data.

(A, C-D) Horizontal lines represent median values. The bottom and top of the box are the first and third quartiles, with whiskers extending to 1.5 Interquartile range of the lower and upper quartile respectively. The blue hollow circles represent outliers. ns, not significant, p > 0.05; *p < 0.05; **p < 0.01, ***p < 0.001 vs normal tissues (A, D) or for indicated comparisons (C) by Student’s t test.

(B, E) p values from log-rank tests and Hazard ratio (HR) are indicated in the graphs

Activating ras mutations occur in 30% of NSCLC patients and over 90% of PAAD patients, and are the drivers of these cancers [11, 38, 39]. We thus investigated the role of MZF1 in these 2 cancer types. Analysis of GEO data using Kaplan-Meier Plotter [48, 49] showed that low MZF1 expression correlated with poor survival in LUAD and LUSC patients (Fig. 7B). Since the RNA-seq data for PAAD in TCGA only included 4 normal tissues, we were unable to determine confidently whether MZF1 expression was downregulated in PAAD. We thus compared MZF1 expression in patients at different disease stages based on initial diagnosis before treatments began. MZF1 expression was remarkable lower in patients with higher-stage (stages II and III/IV) diseases than in those with stage I disease (Fig. 7C). Moreover, analysis of 3 PAAD datasets (GSE101448, GSE62452 and GSE71729) from the GEO database revealed significant downregulation of MZF1 in PADD tissues as compared to paired normal tissues (Fig. 7D). Kaplan-Meier analysis of clinical data in TCGA showed correlation between low MZF1 expression and poor overall and disease free survival in PAAD patients (Fig. 7E). Results from these analyses of human patient data thus suggest that MZF1 may be a tumor suppresser in NSCLC and PAAD.

MZF1 and p16INK4A expression are positively correlated in PAAD

We investigated whether reduced MZF1 expression contributes to p16INK4A downregulation in human cancer. We failed to detect significant correlations between MZF1 and p16INK4A expression in any cancer types in TCGA (Fig. S6B and data not shown). Surprisingly, p16INK4A expression was upregulated or unchanged in tumors as compared to normal tissues in multiple cancers including LUAD, LUSC and PAAD in TCGA (Fig. S6C–D). Increased p16INK4A expression has indeed been reported in malignant tumors of multiple cancers, likely due to loss of a negative feedback resulted from deregulation of RB1, the p16INK4A downstream effector [50, 51]. RB1 loss correlated with high p16INK4A expression in cancers such as breast cancer and lung cancer [50, 52–55].

To dissect the relationship between MZF1 and p16INK4A expression among heterogeneous tumor cells from each patient and among different tumors with heterogeneous genetic backgrounds and treatment histories, we analyzed recently published single-cell RNA-sequencing (scRNA-seq) data from 24 PAAD samples from untreated patients and 11 healthy pancreases [56]. The original 57530 single cells were annotated as 10 different cell types based on known markers. We focused on the type 2 ductal cells that are the malignant ductal cells and the type 1 ductal cells that are the major normal origin of pancreatic tumor cells [56] (Fig. 8A).

Fig. 8.

Analysis of single cell RNA-seq data reveals a correlation between low MZF1 expression and reduced CDKN2A/p16INK4A expression in PAAD.

(A) The t-distributed stochastic neighbor embedding (t-SNE) map of 16084 ductal cells grouped into ductal cell type 1 (normal cells) (orange) and ductal cell type 2 (tumor cells) (aqua blue).

(B-D) Expression levels of MZF1 (B), CDKN2A/p16INK4A (C) and RB1 (D) are plotted onto the t-SNE map, showing decreased expression of MZF1 and RB1 and increased expression of p16INK4A in the type 2 as compared to the type 1 ductal cells (left). MZF1 (B), CDKN2A/p16INK4A (C) and RB1 (D) expression levels are directly compared between the type 1 and type 2 ductal cells in the box plots (right), with each dot representing expression level in each cell. The horizontal lines in the box are the median values, and the bottom and top of the box are the first and third quartiles. *** p < 0.001 vs type 2 by Student’s t test.

(E) Percentage of type 2 ductal cells with high or low levels of CDKN2A/p16INK4A that express high (orange) or low (aqua blue) levels of RB1. Type 2 ductal cells were divided into CDKN2A/p16INK4A-high and -low groups using the median CDKN2A/p16INK4A expression level as threshold; the number of cells with higher than/equal to or lower than median RB1 expression levels were counted in each group to determine the percentage of cells with high- or low-RB1 expression, respectively. The indicated p value was from Fisher’s exact tests comparing the difference between the CDKN2A/p16INK4A-high and -low groups. Number of single cells in each group is indicated in the columns.

(F) The t-SNE plot showing type 2 ductal (tumor) cells with normal RB1 expression levels, defined as those in which RB1 expression levels were higher than or equal to the median RB1 level in the type 1 normal ductal cells.

(G-H) The t-SNE plots of type 2 ductal cells with normal RB1 expression levels, showing cells with higher than/equal to the median (Hi, orange) or lower than the median (Lo, aqua blue) expression levels of MZF1 (G) or CDKN2A/p16INK4A (H).

(I) Percentage of type 2 ductal cells with normal RB1 expression levels and high or low levels of MZF1 that express high (orange) or low (aqua blue) levels of CDKN2A/p16INK4A. Type 2 ductal cells with normal RB1 expression levels (as shown in F) were divided into MZF1-high (Hi) and -low (Lo) groups using the median MZF1 expression level as threshold; the number of cells with higher than/equal to or lower than the median CDKN2A/p16INK4A expression levels were counted in each group to determine the percentage of cells with high (Hi) - or low (Lo)-CDKN2A/p16INK4A expression, respectively. The indicated p value was from Fisher’s exact tests comparing the difference between the MZF1-high and -low groups. Number of single cells in each group is indicated in the columns.

(J) A proposed signaling pathway underlying the role of MZF1 in oncogenic ras-induced senescence. Upon activation of the Raf-MEK1-ERK cascade by oncogenic ras, ERK activates ETS1 and c-Jun through phosphorylation, which bind to the MZF1 promoter and induce MZF1 transcription. MZF1 in turn binds to the CDKN2A/p16INK4A promoter and recruits CHD7, which stimulates CDKN2A/p16INK4A transcription, leading to induction of senescence. It is likely that in NSCLC and PAAD cells with normal RB1 expression, decreased MZF1 expression contributes to reduction in p16INK4A expression, disruption of senescence and enhanced tumorigenesis. Events that occur in cancer cells are indicated in red.

Consistent with results from analyses of the TCGA and GEO datasets (Fig. 7D and S6C–D), MZF1 expression was lower, while p16INK4A expression was higher, in type 2 (tumor) than in type 1 (normal) ductal cells (Fig. 8B–C). Moreover, RB1 expression was reduced in type 2 cells as compared to type 1 cells (Fig. 8D), and among the type 2 cells, high CDKN2A/p16INK4A expression correlated with low RB1 expression (Fig. 8E). These findings were consistent with the previously reported inverse correlation between CDKN2A/p16INK4A and RB1 expression, and suggested that in tumor cells with low RB1 expression, the CDKN2A/p16INK4A level is increased due to loss of negative feedback and may no longer be relevant to cell proliferation since its downstream effector RB1 is already inactivated. After excluding the type 2 cells with lower-than-normal RB1 expression (Fig. 8F) from analyses, we found that low MZF1 expression indeed correlated with low CDKN2A/p16INK4A expression in tumor cells with normal RB1 levels (Fig. 8G–I). Therefore, collectively, our data indicate that at least in some PAAD cells, such as those with normal RB1 levels, decreased MZF1 expression may contribute to the reduction in CDKN2A/p16INK4A and inactivation of the RB1 pathway, leading to disruption of senescence induction.

Discussion

In this study, we demonstrate that CHD7, a transcriptional co-activator without DNA binding domains, is recruited to the p16INK4A promoter by a transcriptional factor MZF1, which mediates oncogenic ras-induced p16INK4A transcription and senescence in a CHD7-dependent fashion. Further investigations revealed a novel signaling pathway mediating OIS and p16INK4A expression, in which oncogenic ras activates the c-Jun and ETS1 transcription factors through phosphorylation mediated by the Raf-MEK-ERK cascade, which in turn stimulate transcription of the MZF1 gene, leading to increased MZF1 expression, enhanced CHD7 recruitment to the p16INK4A promoter, elevated p16INK4A expression, and senescence induction (Fig. 8J).

In NSCLC and PAAD, where activating ras mutations occur frequently [11, 38, 39], MZF1 expression is decreased in tumors as compared to normal tissues, and reduced MZF1 expression correlates with advanced disease stages and poor patient survival (Fig. 7A–E, S1), suggesting that MZF1 is a potential tumor suppressor in these cancers (Fig. 8J). Consistent with this notion, MZF1 mediates ras-induced p16INK4A expression and senescence in primary human small airway lung epithelial cells and pancreatic epithelial cells, the origins of NSCLC and PAAD, respectively (Fig. S3A–H, S4A–H). Our initial attempt to determine the effect of MZF1 on p16INK4A expression in human cancer was hindered by the lack of reduction (or even increase) in p16INK4A expression in tumors with defective RB as reported previously, likely due to the loss of a negative feedback loop [50-55]. By focusing on single-cell RNA-seq data from untreated PAAD patients, we were able to demonstrate a correlation between reduced MZF1 levels and low p16INK4A expression in tumor cells with normal RB levels, suggesting that decreased MZF1 expression contribute to cancer development through reduction in CDKN2A/p16INK4A and inactivation of RB, at least in some PAAD tumor cells with functional RB.

Although oncogenic ras induced expression of both long and shout MZF1 isoforms, likely because they are derived from the same transcript, only the long isoform with DNA binding capacity [17-19] mediates ras-induced p16INK4A transcription and senescence. Therefore, the DNA-binding activity is essential for the function of MZF1 in p16INK4A transcription and senescence. Interestingly, while the long MZF1 isoform was essential for ras-induced p16INK4A expression and senescence, its overexpression only partially mimicked the ras effects and induced moderate levels of senescence as compared to ras itself (Fig. 3E–F). It is likely that ras-induced posttranslational modifications, besides increased expression, is important for the role of MZF1 in senescence.

MZF1 expression can be regulated by transcription factors AP4 and c-Myb and by microRNAs including Let-7, miR-492, miR-337–3p in various biological settings in cancer cells [31, 57, 58]. Here, we identified a new regulatory pathway for MZF1 expression, in which oncogenic ras induced MZF1 transcription through the c-Jun and ETS1 upon their activation by the Raf1-MEK-ERK pathway. We previously reported that the same Ras-Raf1-MEK-ERK-c-Jun/ETS1 axis stimulated transcription of p38δ that mediates OIS via a p16INK4A/p53-independent mechanism [34]. Thus, in response to oncogenic ras, c-Jun and ETS1 are activated by Raf1-MEK-ERK and stimulate transcription of multiple targets that coordinate to mediate senescence induction.

Elevated p16INK4A level isa crucial for cellular senescence and tumor suppression [59, 60], and is a key effector in OIS (Fig. S2) [61]. It has been reported that transcriptional regulators such as Id1, YY1, PcG and YB1 suppress senescence by repressing p16INK4A transcription, while others including FOXA1, Sall2, ETS1/2 and histone H3K4 methyltransferase MLL1 bind to the p16 INK4A locus and induce p16INK4A during senescence [62]. Our study has identified MZF1 as a novel transcriptional factor essential for p16INK4A transcription during OIS. Since senescence is a barrier to oncogenic transformation and is mediated at least partly by p16INK4A [63, 64], regulators of p16INK4A are likely to control tumorigenesis. Indeed, loss of FOXA1 that mediates p16INK4A transcription leads to silencing of p16INK4A, bypass of senescence, and ras-mediated oncogenic transformation in prostate and breast cancer cells [65]; up-regulated Id1 expression decreased p16INK4A expression, leading to increased proliferation in prostate cancer cells [66]. Investigations on the role of MZF1 in tumorigenesis is underway.

MZF1 is a multifaceted transcription factor that either promotes or suppresses tumorigenesis in a cancer type- and context-dependent fashion, likely by regulating different transcriptional targets under different conditions. MZF1 decreased viability, proliferation, migration, and anchorage-independence by inhibiting IGF-IR transcription in lymphoma cells [67], reduced transcription of the Mtor oncogene in both fibroblasts and mouse plasmacytoma cells [68], and suppressed migration and invasion of cervical cancer cells by inhibiting transcription of matrix metalloproteinase-2 [28]. These studies in cell culture are consistent with the cancer prone phenotype of the MZF1 knockout mice [21], suggesting that MZF1 is a tumor suppressor. On the other hand, silencing MZF1 suppressed migration, invasion and tumorigenicity by reducing PKCα expression in human hepatocellular carcinoma cells [22, 26], and overexpression of MZF1 in glioma cells induced cell proliferation and promoted growth of xenograft tumors implanted into mouse brains by inducing c-Myc transcription [69], suggesting a tumor promoting role of MZF1. The function MZF1 is controversial in some cancer types, likely due to differential experimental conditions. In breast cancer, MZF1 mediates invasion of Her2-positive cells by inducing expression of cysteine cathepsins B and L (CTSB and CTSL1) and PKCα [27, 70], and also inhibits brain metastasis of Her2 positive cancer cells by repressing βIII-tubulin expression [48]. In prostate cancer cells, MZF1 promotes tumorigenesis by up-regulating transcription of CDC37 [25], but also inhibits prostate cancer growth by inducing ferroportin transcription and ferroportin-driven iron egress [57]. The differential roles of MZF1 in cancer is consistent with our finding that MZF1 expression is upregulated in some, but downregulated in other, cancer types (Fig. S6A).

We found that MZF1 mediates OIS in primary HSAECs (Fig. S3), and that reduced MZF1 expression occurs in tumors from LUAD and LUSC patients and correlates with poor survival (Fig. 7A–B), suggesting that MZF1 suppresses NSCLC development. These results appears to contradict a previous report that MZF1 expression was increased in LUAD cells as a result of loss of liver kinase B1 (LKB1), which in turn induced c-Myc expression, leading to enhanced anchorage-independent growth, migration and invasion of LUAD cells [24]. It is possible that this previously reported tumor-promoting activity of MZF1 operates mainly in LUAD cells with loss of LKB1. Similarly, reduction in MZF1 expression and correlation between reduced MZF1 levels and poor patient survival were observed in PAAD, another cancer type with high-frequency oncogenic ras mutations, suggesting that loss of MZF1 is crucial for abrogating oncogenic ras-induced p16INK4A expression and senescence, and that MZF1 is a tumor suppressor in cancers driven by activated ras mutations. However, we failed to observe correlations between ras mutations and reduced MZF1 expression and between reduced MZF1 expression and reduced p16INK4A expression in NSCLC and PAAD using TCGA data, likely due to heterogeneity among tumor cells from each patient and among different tumors with heterogeneous genetic backgrounds and treatment histories. We thus analyzed scRNA-seq data from untreated PAAD patients and normal individual [56], and found that upon exclusion of tumor cells with reduced RB1 expression, in which regulation of p16INK4A expression may no longer be relevant to cell proliferation and tumorigenesis due to inactivation of its downstream effector RB, reduced MZF1 expression indeed correlated with reduced p16INK4A expression, suggesting that at least in some PAAD tumor cells where p16INK4A is crucial, decreased MZF1 expression contributes to reduction in p16INK4A and thus inactivation of the RB pathway.

Of caution, most of our studies were performed using Ha-RasV12 and primary BJ fibroblast cells as models for senescence induction, which may differ from the oncogenic K-ras genes present in NSCLC and PAAD and primary lung and pancreatic epithelial cells, respectively. However, oncogenic K-ras also induces senescence[71, 72]. In addition, while most detailed mechanisms were studied in BJ cells, we showed that MZF1 mediates oncogenic ras-induced p16INK4A expression and senescence in primary human small airway epithelial cells (HSAECs) and primary human pancreatic epithelial cells (Fig. S3, S4). Thus, it is highly likely that the mechanistic insights obtained from the senescence models are relevant to specific cancers such as NSCLC and PAAD, although additional studies in more cancer specific systems are needed. Moreover, although analysis of the GEO data showed significant correlation between reduced MZF1 expression and poor survival in both LUAD and LUSC, analysis of the TCGA data showed a strong trend of opposite correlation in LUSC (p=0.064). The difference may lie in the different cohorts or different experimental approaches (gene chip vs RNA-seq) for generating data in the 2 databases.

In summary, our findings have identified a novel regulator of p16INK4A transcription, a new signaling component in the pathway that mediates OIS, and a mechanism underlying the tumor suppressing function of MZF1. Since MZF1 is a potential tumor suppressor in some cancers, the signaling pathway in which MZF1 exerts its function may serve as target for new anti-cancer therapies.

Materials and Methods

Cell lines and cell culture

BJ, HSAEC and primary human pancreatic epithelial cells were purchased from ATCC, Life Cell Technology and Cell Biologics, respectively. They were authenticated twice a year by expression profiling and by mycoplasma tests. Senescence was analyzed as reported [32, 47]. Details are described in Supplemental Methods.

Plasmids.

Plasmids are described in Supplemental Methods.

Retrovirus- and lentivirus-mediated gene transduction.

Cells were transduced with recombinant retroviruses and lentiviruses as previously described [32] and selected as described in Supplemental Methods.

Western blot analysis.

Western blot was performed as described [32].

RNA isolation and quantitative Real-time PCR (qRT-PCR).

RNA was extracted and analyzed by qRT-PCR as described in Supplemental Methods using gene-specific primers (Table S3).

Luciferase reporter assay.

Luciferase reporters were reported previously [32] or generated as described in Supplemental Methods using primers in Table S3. Luciferase assays are described in Supplemental Methods.

ChIP-qPCR

Details of ChIP-qPCR are described in Supplemental Methods.

Genomic data analyses

Analyses of data from TCGA and GEO databases and single cell RNA-seq data are described in Supplemental Methods.

Statistical Analyses

Each experiments were performed in triplicates and repeated at least 3 times. Exact sample size, and methods and results from statistical analyses for each experiment are indicated in figure legends.