Genomic profiling of the transcription factor Zfp148 and its impact on the p53 pathway

Recent data suggest that the transcription factor Zfp148 represses activation of the tumor suppressor p53 in mice and that therapeutic targeting of the human orthologue ZNF148 could activate the p53 pathway without causing detrimental side effects. We have previously shown that Zfp148 deficiency promotes p53-dependent proliferation arrest of mouse embryonic fibroblasts (MEFs), but the underlying mechanism is not clear. Here, we showed that Zfp148 deficiency downregulated cell cycle genes in MEFs in a p53-dependent manner. Proliferation arrest of Zfp148-deficient cells required increased expression of ARF, a potent activator of the p53 pathway. Chromatin immunoprecipitation showed that Zfp148 bound to the ARF promoter, suggesting that Zfp148 represses ARF transcription. However, Zfp148 preferentially bound to promoters of other transcription factors, indicating that deletion of Zfp148 may have pleiotropic effects that activate ARF and p53 indirectly. In line with this, we found no evidence of genetic interaction between TP53 and ZNF148 in CRISPR and siRNA screen data from hundreds of human cancer cell lines. We conclude that Zfp148 deficiency, by increasing ARF transcription, downregulates cell cycle genes and cell proliferation in a p53-dependent manner. However, the lack of genetic interaction between ZNF148 and TP53 in human cancer cells suggests that therapeutic targeting of ZNF148 may not increase p53 activity in humans.

Activation of the tumor suppressor p53 may have beneficial effects on tumors with wild-type p53 1 . The first drugs targeting the p53 repressor murine double minute 2 (MDM2) have entered clinical trials, but their clinical benefit is still under evaluation 2 . Because knockout of Mdm2 in mice causes widespread and lethal activation of p53 in normal tissues 3 , efficient pharmacological inhibition of MDM2 may have adverse effects. Thus, identifying alternative ways to activate the p53 pathway with less adverse effects is warranted.
Zinc finger protein 148 (Zfp148, Zbp-89, BFCOL, BERF1, htβ) is a transcription factor that binds to GC-rich DNA sequences, thus activating or repressing transcription of target genes [4][5][6][7][8][9][10][11] . Earlier studies have shown that Zfp148 is a potent repressor of p53 activity in mice. Firstly, deletion of Zfp148 caused p53-dependent proliferation arrest of cultured mouse embryonic fibroblasts (MEFs) and prenatal lung tissue that was rescued by reducing oxidative stress 12 . Moreover, deletion of one copy of Zfp148 in the Apc Min/+ model of intestinal adenomas reduced tumor numbers and increased survival by increasing p53 activity 13 . In line with this, conditional deletion of one or both alleles of Zfp148 in the gut epithelium of Apc FL/+ mice reduced tumor formation through a mechanism

Results
Zfp148 deficiency downregulates expression of E2F-responsive cell cycle genes. To define mechanisms that cause proliferation arrest of Zfp148 gt/gt MEFs, we used transcript profiling to identify genes that were differentially expressed between Zfp148 gt/gt MEFs and wild-type controls. We also compared Zfp148 gt/gt MEFs infected with adenovirus carrying Zfp148 cDNA or empty vector to verify that differentially expressed genes depended on Zfp148. The analysis revealed a set of more than 300 genes that were downregulated in Zfp148 gt/gt MEFs compared to wild-type controls, and rescued in Zfp148 gt/gt MEFs after adenoviral expression of Zfp148 cDNA compared to empty vector ( Fig. 1A and Supplementary Data S1). Overrepresentation analysis showed strong enrichment of cell cycle-related genes (Fig. 1B). Since this category of genes is regulated by E2F transcription factors, we identified known E2F targets from earlier publications 17,19 and investigated their rank in the gene list. The majority of known E2F targets clustered at the top of the list confirming that Zfp148 deficiency downregulated E2F-dependent cell cycle genes (Fig. 1A).
To investigate the impact of p53 on gene expression, we repeated the analysis with RNA extracted from MEFs generated on a Trp53-null genetic background. Downregulation of the cell cycle-enriched gene set was markedly attenuated on this background (Fig. 1A), confirming that p53 plays a dominant role. However, the rank of differentially expressed genes was similar to the rank obtained on Trp53 wild-type background (Spearman correlation, r = 0.56, P < 0.0001) with strong enrichment of cell cycle-related gene ontology (GO) terms among the downregulated genes (Fig. 1B). We therefore conclude that Zfp148 deficiency downregulates cell cycle genes by both p53-dependent and p53-independent mechanisms.
The previous data panels show qualitative changes. To obtain quantitative data, we determined mRNA levels with real-time RT PCR for 17 genes arbitrarily selected from the downregulated gene set. Expression of genes related to the GO term cell cycle was reduced by half in Zfp148 gt/gt MEFs compared to controls (Fig. 1C). Deletion of one copy of Trp53 is sufficient to rescue cell proliferation of Zfp148-deficient MEFs 12 . In line with this, expression of 11 genes was restored to near normal levels in Zfp148 gt/gt Trp53 +/− compared to Zfp148 gt/gt MEFs, and 9 of those genes were linked to the GO-term cell cycle (Fig. 1C). Only 1 of the 6 remaining genes that were not restored on Trp53 +/− background was linked to the cell cycle. These results suggest that p53-mediated repression of cell cycle genes causes the proliferative arrest of Zfp148 gt/gt MEFs.
To assess whether downregulation of the gene set observed in response to Zfp148 deficiency in MEFs also occurred in vivo, we isolated RNA from neonatal lungs and brains and from bone marrow-derived macrophages of Zfp148 gt/gt and wild-type mice. We determined gene expression by microarrays and showed that the gene set was markedly downregulated compared to wild-type controls in Zfp148 gt/gt lungs but not in brains or macrophages (Fig. 1D). Previous studies showed that Zfp148 deficiency induces cell cycle arrest in prenatal lungs in a p53-dependent manner 12 . In contrast, brains and bone marrow derived macrophages from Zfp148 deficient mice displayed no overt phenotypes 12,20,21 . Thus, the observed downregulation of cell cycle genes in lungs but not brain or bone marrow derived macrophages from Zfp148 deficient mice compared to controls makes sense (Fig. 1D).
the Cdkn2a transcript ARF is required for cell proliferation arrest of Zfp148 gt/gt MEFs. The downregulation of cell cycle genes in Zfp148 gt/gt MEFs by p53-dependent and -independent mechanisms prompted us to investigate mRNA levels of the cyclin-dependent kinase inhibitor Cdkn2a. Cdkn2a produces two gene products, ARF and p16, which downregulate cell cycle genes by activating the p53 pathway and inactivating the CDK4/6 complex, respectively ( Fig. 2A). ARF and p16 mRNA were markedly increased in Zfp148 gt/gt MEFs compared to controls (Fig. 2B), supporting a role for Cdkn2a in the downregulation of cell cycle genes.
We have previously demonstrated that Zfp148 deficiency induces cell proliferation arrest in MEFs after a few passages, and that p53 is required for the arrest 12 . To further investigate the role of Cdkn2a in the proliferation arrest, Zfp148 gt/+ mice were bred on a Cdkn2a +/− background to generate Zfp148 gt/gt Cdkn2a +/− MEFs. Deletion of one copy of Cdkn2a restored normal proliferation of Zfp148 gt/gt MEFs (Fig. 2C), demonstrating that Cdkn2a is required for proliferation arrest of these cells. To determine the relative importance of ARF and p16 for the proliferation arrest, we used lentiviral guide-RNAs and CRISPR to generate Zfp148 gt/gt MEFs lacking either ARF or p16 (Fig. 2D, E and Supplementary Fig. S1). Selective ARF knockout rescued growth of Zfp148 gt/gt cells (Fig. 2F) despite increased expression of p16, which is an expected consequence of ARF inactivation 22 . Knockout of p16 had no effect on cell growth (Fig. 2F). Collectively, these data show that ARF but not p16 is required for p53 activation and proliferation arrest of Zfp148 gt/gt MEFs.
Zfp148 binds to clustered cytosine-rich DNA elements in basal promoters of other transcription factors. To understand the primary basis of ARF and p53 activation in Zfp148 gt/gt MEFs, we used chromatin immunoprecipitation-on-chip (ChIP-chip) assays to identify target sites of FLAG-tagged Zfp148 (AdZfp148FLAG) in Zfp148 gt/gt Trp53 −/− MEFs. The experiment was performed on a Trp53-null background to   www.nature.com/scientificreports/ overcome senescence and obtain sufficient amounts of starting material. The CMV-driven adenovirus vector used produces Zfp148 protein amounts that are similar to those found in wild-type MEFs 12 .
Zfp148 target sites were recorded in four independent experiments with a strong preference for binding close to transcription starts ( Fig. 3A and Supplementary Data S2). To determine the DNA-binding preferences of Zfp148, we constructed a position weight matrix (PMW) model from 16 known Zfp148 binding sites ( Table 1). As expected, the model predicted a cytosine-rich binding motif (Fig. 3B). We next applied the model to DNA regions surrounding known transcription starts. While there was no correlation between peak positions of ChIP-chip data and PWM-predicted single motifs, Zfp148 showed a strong preference for binding to regions with clustered PWM-predicted motifs (Fig. 3C,D). These data indicate that Zfp148 binds to clustered motifs in cytosine-rich basal promoters.
Overrepresentation analysis of 3,488 genes that bound to Zfp148 at 1% false discovery rate in all experiments showed strong enrichment of functional terms. Binning genes according to their rank in the ChIP-chip experiment revealed a striking correlation between binding scores and the enrichment of functional terms; the enrichment declined abruptly after 600 genes (Spearman correlation r = 0.74, P < 0.001, Fig. 3E  www.nature.com/scientificreports/ connectivity is an expected property of co-regulated genes, the number of Zfp148 target genes is likely much lower than the number of binding genes. The top-ranked genes (top 600) were enriched for functional terms related to transcriptional regulation, suggesting that Zfp148 mainly operates upstream of other transcription factors (Fig. 3F). There was no overrepresentation of cell cycle-related terms. By integrating ChIP-chip and gene expression data, we confirmed a small but significant upregulation of Zfp148 target genes in Zfp148 gt/gt MEFs after infection with AdZfp148 (Fig. 3G). However, there was no significant overlap between Zfp148-binding genes and the set of cell cycle-enriched genes that was downregulated in Zfp148 gt/gt MEFs (Fig. 3H). Notably, Zfp148 bound to the ARF promoter in all experiments with binding regions centered at 800 base pairs upstream of the transcription start (Fig. 3I) suggesting that Zfp148 may repress ARF transcription. However, the preferential binding of Zfp148 to promoters of other transcription factors indicates that deletion of Zfp148 could have pleiotropic effects that activate ARF and p53 indirectly.
No genetic interaction between ZNF148 and TP53 in human cancer cell lines. Since the integrated transcriptomic and promoter occupancy analysis suggested pleiotropic and indirect mechanisms of p53 and ARF activation in Zfp148 gt/gt MEFs, we decided to explore the genetic interaction between ZNF148 (the human orthologue of Zfp148) and TP53 in CRISPR and siRNA data from human cancer cell lines.
We downloaded CRISPR and siRNA scores for all genes in 487 human cancer cell lines from the Cancer Dependency Map database (https ://depma p.org). A negative score indicates that knockout (CRISPR) or knockdown (siRNA) of that gene has a negative impact on cell growth or survival. Genes with CRISPR scores below − 0.6 are considered to be essential 23 . The median CRISPR and siRNA scores for ZNF148 were 0.05 (± 0.01 95% CI) and − 0.12 (± 0.02 95% CI), suggesting that cancer cells are marginally affected by ZNF148 inactivation (Fig. 4A). ZNF148 was not essential for the growth or survival of any of the tested cell lines (Fig. 4B).
We next divided the cell lines into two groups depending on their TP53 genotype. As expected, inactivation of TP53 increased the growth of cells with wild-type TP53 but had no impact on cells with TP53 mutations (Fig. 4C). The p53 repressor MDM2 showed a reverse pattern with low scores in TP53 wild-type cells and scores near zero in cells with TP53 mutations (Fig. 4D). In contrast, there was no marked difference in ZNF148 scores between TP53 wild-type and mutant cell lines (Fig. 4E).
We finally ranked the genes according to their similarity with TP53 as judged by the Pearson correlation of CRISPR or siRNA scores across the set of human cancer cell lines. Canonical activators of the p53 pathway such as USP28, CDKN1A, and TP53BP1 showed strong correlation with TP53 and were positioned at the top of the list, while repressors including MDM2, PPM1D, and MDM4 were positioned at the end (Fig. 4F). In contrast, ZNF148 did not show strong correlation or anti-correlation with TP53 (Fig. 4F), suggesting that there is no genetic interaction between ZNF148 and TP53 in human cancer cell lines under standard culture conditions.
To exclude the possibility that ZNF148 exerts mild regulation of p53 that is compensated under standard conditions, we investigated the role of ZNF148 under DNA damaging conditions. We used CRISPR-Cas9 to delete ZNF148 in H838, A549, and H1299 (TP53 null) human lung cancer cells ( Fig. 5A and Supplementary Fig. S3), and confirmed that p53-induction by etoposide was intact ( Fig. 5B and Supplementary Fig. S3). To uncover quantitative differences in the DNA damage response, we established dose-response curves for cisplatin and etoposide that activates p53 through ATM/ATR, and for TH588 that activates p53 though USP28 24 . There was no difference in IC50 values between ZNF148 knockout cells and controls in any of the tested cell lines (Fig. 5C).
To assess the role of Zfp148 under DNA damaging conditions in vivo, we exposed Zfp148 heterozygous mice and controls to ionizing radiation and quantified the depletion of immature thymocytes, a process that is entirely p53-dependent 25 . As expected, ionizing radiation reduced the number of thymocytes in a dose dependent manner (Fig. 5D). However, there was no difference in the total number of thymocytes, or the distribution of different thymocyte subsets, between Zfp148 gt/+ and control mice 72 h after exposure to ionizing radiation (Fig. 5D). We conclude that Zfp148 does not regulate the p53-pathway under DNA damaging conditions.

Discussion
We have previously shown that Zfp148 deficiency arrests proliferation in MEFs by activating p53 12 , and here we investigated the underlying mechanism by analyzing the transcriptional output and promoter occupancy of Zfp148 in MEFs. Our results suggest that Zfp148 deficiency, by derepressing ARF transcription, downregulates cell cycle genes and cell proliferation in a p53-dependent manner. The model is supported by three lines of evidence. Firstly, Zfp148 deficiency downregulated cell cycle genes in a p53-dependent and -independent manner, with p53 playing a dominant role. Secondly, proliferation arrest of these cells required increased expression of the Cdkn2a transcript ARF, which is a major regulator of the p53 pathway. Finally, Zfp148 interacted physically with the ARF promoter, indicating that Zfp148 may regulate ARF transcription.
Zfp148 did not bind to promoters of cell cycle genes more often than predicted by chance, suggesting that regulation of cell cycle genes is indirect. p53 can downregulate cell cycle genes by inducing the cyclin-dependent kinase inhibitor p21, which is also regulated by ZNF148 8,9 . Although regulation of the p21 promoter by ZNF148 could theoretically explain the p53-independent repression of cell cycle genes, the promoter was negatively regulated by ZNF148 (in response to butyrate) in most published experiments 8,9 , which would not explain our current finding. Given that the second transcript of the Cdkn2a locus, p16, was markedly increased in Zfp148 gt/gt MEFs compared to controls, a more likely explanation is that p53-independent downregulation of cell cycle genes is mediated by p16.
The cell cycle arrest in Zfp148 gt/gt MEFs was rescued by heterozygous deletion of Cdkn2a or selective knockout of the ARF transcript, indicating a plausible mechanism for p53 activation by ARF. In contrast to homozygous deletion of Cdkn2a, which causes hyper-proliferation of MEFs and spontaneous escape from senescence, heterozygous deletion does not increase cell proliferation on its own 26  www.nature.com/scientificreports/ likely part of the mechanism underlying proliferation arrest in Zfp148 gt/gt MEFs. We have previously shown that Zfp148 gt/gt MEFs are rescued by exogenous antioxidants or culture at reduced oxygen concentration showing that oxidative stress contributes to the proliferation arrest in these cells 12 . Because ARF suppresses the transcriptional activity of NRF2 (NFE2L2) 27 -the master regulator of endogenous antioxidants-increased ARF transcription may explain the oxidative stress weakness of Zfp148 gt/gt MEFs. Zfp148 interacted physically with the ARF promoter, raising the possibility that ARF is a direct target of Zfp148. Thus, deletion of Zfp148 may activate p53 by derepressing ARF transcription, similar to knockouts of polycomb members Bmi1, M33, Mel18, and Phc2 28-31 . In line with this, inhibition of ZNF148 potentiates butyrate-induced senescence of human HT116 colon cancer cells by derepressing the p16 promoter 32 . However, increased expression of ARF could have many underpinnings. The strong enrichment of transcription factors among Zfp148-binding genes suggests that Zfp148 operates at a high level of the transcription factor hierarchy 33 . Small orchestrated alterations of many transcripts may therefore contribute to the activation of ARF and p53 in Zfp148 gt/gt MEFs, possibly by affecting cellular homeostasis. Collectively, our data suggest that Zfp148 deficiency activates p53 indirectly, by derepressing ARF and by altering the activity of other transcription factors.
A previous publication showed that ZNF148 interacts physically with p53 16 , thus identifying another possible mechanism behind the activation of p53 in Zfp148 gt/gt MEFs. However, our current data on ZNF148 and TP53 dependencies in human cancer cells strongly argue against a physical interaction between ZNF148 and p53 playing a significant role. There was no evidence of genetic interaction between ZNF148 and TP53 in CRISPR and siRNA screens of hundreds of human cancer cell lines, performed under conditions that revealed strong dependencies between TP53 and known repressors of p53 activity including MDM2, MDM4, and PPM1D. Moreover, the was no evidence of genetic interaction between ZNF148 and TP53 under DNA damaging conditions in human lung cancer cells or mice. However, it remains possible that regulation of ARF is important. For reasons not well understood, ARF appears to play a more prominent role as a tumor suppressor in mice than in humans 34 . Moreover, accumulating evidence indicate that ARF has important roles that are independent of p53 34 . In line with this, we observed that the correlation between CDKN2A and TP53 was weak in the CRISPR screen and absent in the siRNA screen (data not shown).
In the light of these results, we need to reevaluate the role of Zfp148 and its impact on the p53 pathway. Without evidence of genetic interaction between ZNF148 and TP53 in human cancer cells, the incentive for therapeutic targeting of ZNF148 becomes weak. The recent finding that inhibition of ZNF148 potentiates butyrate-induced senescence of human HT116 colon cancer cells 32 suggests that core functions of Zfp148 are evolutionarily conserved. Further investigation of the interaction between Zfp148 and Cdkn2a promoters may provide valuable mechanistic insights. However, transcriptomes are shaped by combinatorial interactions of transcription factors forming complex network motifs, including feedforward and feedbackward loops, that cooperatively govern gene expression 35 . Our current data suggest that Zfp148 operates at a high level of the transcription factor hierarchy. Thus, future studies of Zfp148 warrant a more integrated and comprehensive approach.

Material and methods
Mice. Cdkn2a tm1rdb (Cdkn2a −/ ) and Trp53 tm1Tyj (Trp53 −/ ) mice were obtained from The Jackson Laboratory and Zfp148 gt/+ mice were produced by us 12 . The mice were kept on a 129/Bl6 mixed genetic background and all experiments were performed with littermate controls. Genotyping was performed by PCR amplification of genomic DNA from mouse biopsies obtained during earmarking. PCR primers used for genotyping are listed in Supplementary Table S1. Mice were fed on a regular diet and had unlimited supply of food and water. All animal procedures used in this study were approved by The Animal Research Ethics Committee in Gothenburg and performed in accordance with relevant guidelines and regulations.
Cell culture. Primary MEFs were isolated from E13.5-15.5 embryos as described 12 . Cells were cultured in DMEM low glucose medium with 10% fetal bovine serum, 100 µg/ml penicillin and streptomycin, 1% non-  Gene expression analysis. Total RNA from MEFs, lungs, brains and macrophages was isolated using the Gene Elute kit (Sigma) and hybridized to Affymetrix Mouse Gene 1.0 ST chips. Data were normalized with the Robust Multi-Array Analysis (RMA) method 36 . The heat-map was constructed using the Hierarchical Clustering Explorer 3.0 software. Gene ontology statistics was calculated using the DAVID software 37,38 .
Real-time quantitative PCR. TaqMan and SYBR Green assays were performed as described 39 using TaqMan/SYBR Green universal PCR mastermix (A25741, Applied-biosystem) and the pre-designed TaqMan assays (Applied Biosystems) or RT-PCR primers listed in Supplementary Table S1. CRISPR-Cas9 knockouts. Pre-designed guide-RNA (gRNA) sequences targeting p16 or ARF or ZNF148 or control (does not target a sequence in the genome) were cloned into the Lenticrisprv2 vector, and co-transfected with pCMV-dR8.2 (Addgene # 8455) and pCMV-VSV-G (Addgene # 8454) vectors into HEK293T cells to produce complete Cas9 and gRNA coding lentivirus. Virus infected Zfp148 gt/gt MEFs or human lung cancer cells (H838, A549, or H1299) were selected with puromycin for 3 days to obtain batch clones that were used for experiments. The gRNA sequences used are listed in Supplementary Table S1. Lenticrisprv2 was a kind gift from Feng Zhang (Addgene # 52961) 40 .

Statistical analyses.
Values are mean ± SEM. Statistics were performed with Student's t-test for comparisons between two groups; one-way ANOVA for multiple groups; Fisher exact test for contingency tables; twosample Kolgomorov-Smirnov test for cumulative distribution function plots; Spearman correlation coefficient for ranked values; and Pearson correlation coefficient for continuous variables. Differences between groups were considered significant when P < 0.05.