Paired Box-1 (PAX1) Activates Multiple Phosphatases and Inhibits Kinase Cascades in Cervical Cancer

DNA methylation alteration, such as global hypomethylation and localized hypermethylation, within the promoters of tumor suppressor genes, is an important risk factor in cervical cancer. The potential use of DNA methylation detection, in cervical cancer screening or triage of mildly abnormal cytology, has recently been demonstrated. In particular, PAX1 DNA methylation testing was approved as an adjunct to cytology, in Taiwan, and is now undergoing registration trials in China. However, the function of PAX1 in cancer biology remains largely unknown. Here, we show that PAX1 inhibits malignant phenotypes upon oncogenic stress. Specifically, PAX1 expression inhibited the phosphorylation of multiple kinases, after challenges with oncogenic growth factors such as EGF and IL-6. Analogously, PAX1 activated a panel of phosphatases, including DUSP1, 5, and 6, and inhibited EGF/MAPK signaling. PAX1 also interacted with SET1B, increasing histone H3K4 methylation and DNA demethylation of numerous phosphatase-encoding genes. Furthermore, hypermethylated PAX1 associated with poor prognosis in cervical cancer. Taken together, this study reveals, for the first time, the functional relevance of PAX1 in cancer biology, and further supports the prospect of targeting multifold oncogenic kinase cascades, which jointly contribute to multiresistance, via epigenetic reactivation of PAX1.

www.nature.com/scientificreports www.nature.com/scientificreports/ biomarker for cancer detection, including cervical and colon cancer 13,14 , with a methylation test for the advanced disease being approved in 2016 by the U.S. Food and Drug Administration (FDA) 15 .
Similar to the above, DNA methylation as a biomarker for cervical cancer early detection is also promising [16][17][18] . Cytology-based cervical cancer screening has been used for nearly a century, albeit with unsatisfactory sensitivity and a 20% false positive rate 19 . And, while a quadravalent vaccine against HPVs 5,11,16, and 18 has reduced infection by those strains, it is minimally effective against other strains, and is little available in developing nations 3 . To more accurately detect cervical cancer, DNA methylation, as an adjunct for cytology or human papilloma virus (HPV) testing, is becoming the standard for molecular cervical cancer screening 13,20 . The methylation of the paired box-1 (PAX1) gene in cervical cancer was firstly reported in 2008 16 . Serial validation studies performed worldwide have demonstrated its potential in cervical cancer molecular screening [21][22][23][24] . In 2016, the Taiwanese FDA approved the application of PAX1 methylation as an adjunct of conventional cytology in cervical cancer screening 25 . Despite this diagnostic association with cervical cancer, however, there remains a gap in the mechanistic understanding of the precise role of PAX1 in disease etiology.
The PAX gene family is named for the paired box DNA-binding domain that is critical for tissue development and cellular differentiation in embryos 26 . In cancer, however, the function of PAX genes is controversial and has not been well characterized 27 . Four PAX family subgroups exist, based on their structural domains: paired domain, octapeptide, and homeodomain 26 . PAX1 and PAX9 belong to subgroup I, which has a DNA-binding paired domain without a homeodomain, and these show functional redundancy during embryogenesis 28 . PAX9 is also amplified and highly expressed in lung cancer, and its knockdown reduced lung cancer formation in a xenograft study, supporting a role in oncogenesis 29 . On the contrary, PAX1 was aberrantly methylated, and downregulated, in cervical, oral, and ovarian cancer, suggesting that PAX1 is a tumor suppressor gene 16,30,31 . Nevertheless, the mechanism underlying the role of the PAX gene family in cancer biology has not been elucidated. Here, we present the first comprehensive, functional investigation of the mechanistic role of PAX1 in cancer.

PAX1 Inhibits malignant phenotypes of cervical cancer cell lines upon EGF stimulation.
To explore the role of PAX1 in cervical carcinogenesis, we used immunohistochemical staining to examine its expression over the full spectrum of cervical lesions, including normal epithelium, precancerous lesions, and invasive cancer. The results of this experiment revealed that PAX1 nuclear expression was weak in normal cervix, moderate in low-grade squamous intraepithelial lesion (LSIL), strongest in high-grade squamous intraepithelial lesion (HSIL), but weak in invasive cancer cells (Fig. 1A,B). These data suggest that PAX1 functions as a tumor suppressor, especially at the transition from in situ to invasive cancer. Consequently, we overexpressed PAX1 in two cervical cancer cell lines, HeLa and SiHa, and found that it did not significantly inhibit malignant phenotypes, including proliferation, migration, and invasion, in vitro, and tumor growth, in vivo ( Fig. 1C-F). However, the function of a tumor suppressor gene may not become obvious until the existence of oncogenic stressors, such as the expression of certain cytokines or growth factors. Thus, we hypothesized that PAX1 might inhibit malignant phenotypes in the presence of growth factors. In cervical cancer, the epidermal growth factor (EGF) was reported to induce the epithelial-to-mesenchymal transition (EMT), a metastasis-related phenotype that includes cancer cell invasion 32 . Hence, we examined the effects of PAX1 in the presence of EGF. As expected, PAX1 expression inhibited EGF-induced EMT phenotypes, including mesenchymal spindle shape, migration, and cell invasion ( Fig. 2A-C). Analogously, we observed that the EMT marker Snail was downregulated by PAX1, while the epithelial marker CDH1 was upregulated ( Supplementary Fig. S1). Moreover, PAX1 significantly inhibited tumor growth in cells pretreated with EGF ( Fig. 2D). Together, these results support our hypothesis that PAX1 is a tumor suppressor that is responsive to environmental factors, such as EGF, and inhibits EGF-induced EMT and malignant phenotypes.

PAX1 activates multiple phosphatases and inhibits EGF-mediated signaling.
To examine the extent to which PAX1 inhibits the EGF-associated kinome, we tested for kinome changes using an array-based assay. The results of this experiment demonstrated that PAX1 inhibited the activity of multiple oncogenic kinases, including the MAPK pathway (ERK1/2 and P38), and the SRC pathway (HCK, FYN, and SRC) (Fig. 3A,C, and Supplementary Fig. S2A). In addition to EGF, IL-6 is another oncogenic cytokine, and its activation can induce STAT3 and AKT signaling, resulting in tumor growth and invasion in numerous cancers, including cervical 33,34 . Furthermore, IL-6 expression has been correlated with poor prognosis in cervical cancer 35 . In this context, we observed PAX1 to inhibit IL6-mediated activation of several oncogenic signaling kinases, such as AMPK, P70 S6 kinase, AKT, β-catenin, and PRAS40 (Fig. 3B,C and Supplementary Fig. S2B). ERK1/2 is most sensitive to PAX1 inhibiting its phosphorylation by about 80%. Western blotting thus confirmed that PAX1 delayed the timing and amplitude of ERK1/2 phosphorylation in SiHa cells (Fig. 3D). Since inhibition of EGF signaling may be achieved via activation of protein phosphatases, and a previous study revealed the protein tyrosine phosphatase, receptor type R (PTPRR), to inhibit ERK phosphorylation in cervical cancer 17 , we tested whether PTPRR is a transcriptional target of PAX1. Those results demonstrated that PTPRR was indeed upregulated upon PAX1 expression, especially in the presence of EGF ( Supplementary Fig. S4A). Further, ChIP-PCR confirmed PAX1 binding to the PTPRR promoter ( Supplementary Fig. S4B). These results reveal that PAX1-activated PTPRR inhibits the EGF signaling pathway, via inhibition of ERK1/2 phosphorylation.
Two major enzyme families are involved in H3K4 trimethylation in humans: the SET and MLL families. Both share common components, such as WDR5, but target diverse genes 38 . To understand how PAX1 might induce H3K4 trimethylation, we tested possible protein-to-protein interactions between PAX1 and components of the SET and MLL complexes, including SET1A, SET1B, MLL1, MLL2, MLL4, and WDR5, in addition to the DNA demethylases, TET1 and TET2. Coimmunoprecipitation assays indicated direct binding of PAX1 to SET1B and WDR5, but not to other H3K4 HMTs and DNA demethylases (Fig. 5E). Analogously, the reverse experiment also demonstrated WDR5 binding to PAX1 (Fig. 5F), which also bound the PTPRR, DUSP1, DUSP5, and DUSP6 promoters ( Fig. 5G and Supplementary Fig. S6). Taken together, these data suggest that PAX1 activates phosphatase genes, via interacting with SET1B and WDR5, to change histone modifications to a transcriptionally permissive state.  www.nature.com/scientificreports www.nature.com/scientificreports/ PAX1 methylation associates with poor prognosis in cervical cancer. To understand further the status of PAX1 promoter methylation in cervical cancers, we extracted and analyzed DNA methylation profiles from a TCGA dataset. The median methylation levels (beta-values) of the non-tumor specimens and tumors were 0.05 and 0.58, respectively (Fig. 6A), while PAX1 promoter methylation levels in non-tumor tissues were significantly lower than those in tumors (Mann-Whitney U test, p < 0.0001). In addition, hypermethylated PAX1 associated with poor progression-free survival (PFS), and poor overall survival (OS), in patients with cervical squamous cell carcinoma (Fig. 6B).

Discussion
The present study demonstrated that PAX1 plays a tumor suppressor role in response to the onset of oncogenic stress, including growth factor stimulation. We further demonstrated that PAX1 inhibited EGF and IL-6 signaling, through epigenetic activation of multiple phosphatases (Fig. 6C). Kinase-phosphatase network is rigidly regulated in normal physiology [39][40][41] . The kinase cascade may reciprocally activate the phosphatase to keep growth signaling in check. In cancer, however, the regulation loop between kinases and phosphatases is little known. Our data show that PAX1 inhibits EGF-ERK signaling, via epigenetic activation of multiple phosphatases. Furthermore, it was reported the ERK1 mediates activation of the PAX1-like protein in Giardia lamblia 42 , suggesting that PAX1 mediates a physiologic balance between kinases and phosphatases. Taken together, we speculated that PAX1 might activate multiple phosphatases that keep oncogenic kinases in check.
Our previous study demonstrated that the methylation of PAX1 is associated with HPV L1 region methylation in the ≥CIN2 lesions 25 . Under nitroxidative stress, full-length HPV transformed precancerous cell lines can undergo epigenomic modifications including DNA methylation of an EMT inhibition phosphatase, PTPRR 43 . Indeed, there were reports demonstrating HPV infection is associated with higher nitric oxide synthase expression 44 and nitric oxide (NO) concentration in cervix 45 , suggesting the indirect effects of HPV infection on epigenomics. Furthermore, the methylation of PAX1 was increased in the keratinocyte after immortalized by the full-length HPV16 46 . Together with the present study, the infection of HPV may cause the promoter methylation of PAX1 and result in the reduction of phosphatases' expression. Therefore, hypermethylated PAX1 may derepress the oncogenic kinase cascade, thus promoting cancer progression.
Previously, the PAX family was reported to participate in development, but its role in cancer has been little studied. It was suggested that PAX2 and PAX5 play tumor suppressor roles in ovarian cancer and leukemia 47,48 , while conversely, PAX3, PAX7, and PAX6 were reported to be oncogenic in embryonal rhabdomyosarcomas and retinoblastoma 49,50 . However, these reports were all based only on correlations between expression and malignant www.nature.com/scientificreports www.nature.com/scientificreports/ phenotypes, and the underlying mechanisms were not investigated. Although PAX1 is known to participate in neuronal development 51 , there have been no reports of a possible role in cancer biology. Previous studies regarding the PAX family and chromatin modifiers (histone deacetylases (HDACs) and histone lysine methyltransferases (HMTs)), in developmental biology, revealed interactions between PAX2 and MLL2/3 (KMT2C/D, an HMT), in embryonic kidney cell lines, PAX5 and HMTs/HDACs, in mouse B cell development, and PAX7 and HMTs, in mouse myogenesis [52][53][54][55] . In cancer, PAX-interacting proteins remain largely unknown. Here, we demonstrate that PAX1 interacts with SET1B, but neither MLL1-MLL4 nor SET1A. Further, the cooperation between PAX1 and the SET1B histone H3K4 methyltransferase complex c activated the phosphatome and suppressed www.nature.com/scientificreports www.nature.com/scientificreports/ kinase signaling. These findings represent the first indication of PAX1-modulated epigenetic regulation of the kinase-phosphatase loop in cancer biology.
In addition, resistance to targeted therapy in cancer is a notorious problem. For example, tyrosine kinase inhibitors (TKIs), designed to target mutant EGFR, have shown dramatic therapeutic efficacy in lung cancer; www.nature.com/scientificreports www.nature.com/scientificreports/ however, patients often develop resistance to therapy 56 . Additionally, compensatory oncogenic kinase cascades, such as the MAPK 57 , PI3K/AKT 58 , and SRC/AKT 59 pathways, may render TKIs ineffective. In that regard, our study revealed that PAX1-activated phosphatases can inhibit multiple MAPK pathways and the SRC kinase. Consequently, reactivation of DNA methylation-silenced PAX1 may be a strategy to inhibit multiple oncogenic kinase pathways, which could reverse multi-pathway resistance to antineoplastic TKI drugs.
In conclusion, PAX1 is responsible for maintaining an on-and-off homeostasis between kinases and phosphatases, within the cervical epithelium. Loss of PAX1 expression, however, through DNA methylation, may disrupt this balance, leading to cancer development, supporting the further study of PAX1 DNA methylation as a biomarker for cervical cancer detection. Moreover, reactivation of PAX1 may overcome the failure of current target therapies targeting single kinases in cervical cancer treatment. We strongly contend that such epigenetic Cell migration assay and invasion assays. Cell migration assay and invasion assays were described in detail in our earlier study 43 . Cell migration was assessed using the scratch wound-healing assay. Cells were seeded into 6-well dishes until confluent and then wounded using a pipette tip. The migration area was photographed at 0 and 24 h. Cell migration was quantified by measuring the migrated area using ImageJ. Cell invasion was measured in the Transwell system with Matrigel (BD Bioscience). 2000 cells were seeded in the upper chamber in culture medium without serum. The same medium with serum was added to the lower chamber. Cells that had permeated the Matrigel and migrated to the bottom of the insert were fixed with methanol and stained with Giemsa's azur eosin methylene blue solution (Merck) after 24 h. The cells in each chamber were photographed and counted.
RNA extraction, cDNA synthesis, and quantitative real-time PCR (qPCR). RNA extraction, cDNA synthesis, and quantitative real-time PCR were performed as described in our earlier studies 17,43 . Total RNA was isolated by Qiagen RNeasy kit with DNase I digestion (Qiagen) and reverse-transcribed to cDNA using the Super Script III first-strand synthesis system for RT-PCR with Oligo(dT) as the primer (Thermo Fisher Scientific). qPCR was performed by Roche SYBR Green Real-Time PCR System (Roche). The glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used as the internal reference gene. The comparative threshold cycles (Ct) method was used for relative quantification. All values are expressed as mean ± SD. The primers used in this study are shown in Supplementary Table S3.

DNA extraction, bisulfite modification, quantitative methylation-specific PCR (qMS-PCR).
qMS-PCR was performed as described by in our former study 43  www.nature.com/scientificreports www.nature.com/scientificreports/ Immunoblot and kinase array analysis. Cells used for immunoblot were serum-starvated for 16 h and treated with medium with EGF for 0, 5, 15 and 30 min. The process of immunoblot were described in detail in our earlier study 43 . The antibodies used in the immunoblot analysis were: anti-phospho-ERK 1/2 (9106), anti-ERK 1/2 (9102) from Cell Signaling; anti-Set1A (sc-515590), anti-Set1B (sc-248564), anti-MLL1 (sc-377274), anti-MLL2 (sc-292359), anti-MLL4 (sc-517017), anti-TET2 (sc-136926), anti-TET3 (sc-139186) from Santa Cruz; anti-TET1 (GT1462) from GeneTex. Kinome was analyzed by Human Phospho-Kinase Array Kit (ARY003B, R & D systems) according to the manufacturer's instructions. Cells used for kinome analysis were serum starvated for 16 h and treated with medium with EGF or IL-6 for 15 min. In vivo tumorigenicity model. Six-week-old nude mice were used in the tumorigenicity analysis. 10 6 cells from each stable line were resuspended in 0.1 ml PBS and injected subcutaneously into both flanks of each mouse. The mice were sacrificed at day 30. Tumors were removed from the mice and weighed. The protocol for this animal experiment was approved by the Institutional Animal Care and Use Committee (IACUC) of the National Defense Medical Centre, Taipei, Taiwan. All animal procedures and animal care were performed according to institutional animal research guidelines.
Global gene expression analysis. Total RNA samples were isolated and the absorbance ratio of A260/ A280 and A260/A230 analyzed by Nanodrop 2000 (Thermo Scientific) should bigger than 1.8 and 1.0, respectively. The further qualification, the ratio of 28S/18S and RNA quality indicator (RQI) was analyzed by Agilent 2100 Bioanalyzer (Agilent Technologies). Samples with ratio of 28S/18S is bigger than 1 and RQI is bigger than 7 were send to the core service unit (Health GeneTech Corp.) for whole-genome expression analysis. We measured gene expression profiles using Illumina HumanHT-12 v4 Expression Beadchip (Illumina, Inc.). The data had been deposited in the Gene Expression Omnibus (GEO) database (GSE102986). After quantile normalization, we removed the probes with detecting p-values > 0.05 in all of the samples. The average of signal (AVG_signal) of each probe was calculated from duplication of expression data. Differentially expressed genes were identified as having a ≥1.25-fold change in the average expression in the PAX1-1 set and PAX1-2 set compare to EGF set ( Supplementary Fig. S5A,B). The phosphatase gene list was obtained from the human DEPhOsphorylation database (DEPOD, http://depod.bioss.uni-freiburg.de/). Statistical analysis. The Mann-Whitney U test, two-tailed, was used to compare data groups for cell proliferation, migration, invasion, and tumor formation. Standard deviations were used for error bars and various comparisons. Kaplan-Meier analysis and log-rank tests were used to calculate survival, and to compare differences between curves for progression-free survival (PFS) and overall survival (OS). p-values < 0.05 were considered to be statistically significant. Methylomics profiles of cervical cancer cases, obtained from The Cancer Genome Atlas (TCGA), were based on data generated using Human Methylation 450 BeadChips (Illumina). The results were downloaded from the Broad Institute GDAC Firehose (http://gdac.broadinstitute.org/), and used in compliance with TCGA's data usage policy.

Data Availability
All data generated or analysed during this study are included in this published article and its supplementary information files. The whole-genome expression data had been deposited in the Gene Expression Omnibus (GEO) database (GSE102986).