GRAINYHEAD-LIKE-2, AN EPITHELIAL MASTER PROGRAMMER, PROMOTES INTERFERON INDUCTION AND SUPPRESSES BREAST CANCER RECURRENCE

Abstract

Type-I and -III interferons play a central role in immune rejection of pathogens and tumors, thus promoting immunogenicity and suppressing tumor recurrence. Double strand RNA is an important ligand that stimulates tumor immunity via interferon responses. Differentiation of embryonic stem cells to pluripotent epithelial cells activates the interferon response during development, raising the question of whether epithelial vs. mesenchymal gene signatures in cancer potentially regulate the interferon pathway as well. Here, using genomics and signaling approaches, we show that Grainyhead-like-2 (GRHL2), a master programmer of epithelial cell identity, promotes type-I and -III interferon responses to double-strand RNA. GRHL2 enhanced the activation of IRF3 and relA/NF-KB and the expression of IRF1; a functional GRHL2 binding site in the IFNL1 promoter was also identified. Moreover, time to recurrence in breast cancer correlated positively with GRHL2 protein expression, indicating that GRHL2 is a tumor recurrence suppressor, consistent with its enhancement of interferon responses. These observations demonstrate that epithelial cell identity supports interferon responses in the context of cancer.

INTRODUCTION

Type-I/III interferons form a central hub of immune signaling, with crucial functions in innate and adaptive immunity. ⁵⁷ They also support endogenous and immunotherapy-induced tumor rejection, and interferon signaling defects occur frequently in tumor cells. ^12,18,63 In general, the expression of interferon-I (IFN-I) and the functionally similar IFN-III genes – the latter expressed specifically in epithelial cells – is activated by signaling in response to the ligation of Pattern Recognition Receptors (PRRs) by ligands of two classes, Pathogen Associated Molecular Patterns (PAMPs) and Damage-Associated Molecular Patterns (DAMPs). ¹ DAMPs and PAMPs also can induce interferon-mediated chronic inflammation and cell senescence. ³⁰ Because interferons interfere with normal cellular homeostasis, ⁵⁷ the differentiation state of a cell intuitively might be expected to affect interferon responses, but these effects are not well characterized.

Differentiation of embryonic stem cells to epiblast-like epithelial cells and then to mesenchymal cells is one of the most fundamental and early cell lineage commitments during development. This first commitment is programmed substantially by an epithelial “master programmer” transcription factor, Grainyhead-like-2 (GRHL2). GRHL2 acts as a pioneer transcription factor – i.e., a factor that binds DNA in even closed chromatin regions and opens them -- landmarking over 1,000 epithelial gene enhancers to permit more specialized transcription factors such as nuclear receptors to activate gene transcription subsequently in a tissue-specific manner. ^7,44 Interestingly, the embryonic stem cell to epiblast transition is accompanied by the acquisition of type-I interferon expression, suggesting that the epithelial phenotype might prove important for interferon pathways in other contexts such as cancer. ⁹³

The oncogenic Epithelial-Mesenchymal Transition (EMT) is a global, albeit transient, transcriptional reprogramming of a subpopulation of carcinoma cells that confers phenotypic plasticity, promoting tumor metastasis and propensity to recurrence, as well as resistance to chemotherapeutic drugs, anoikis, and NK cytotoxicity. ⁴⁹ The oncogenic EMT endows immune resistance to tumors via diverse mechanisms. ^80,94 The GRHL2 gene is widely down-regulated during EMT of tumors and tumor cell lines. ^16,17 Conversely, experimental restoration of GRHL2 expression both suppresses and reverses EMT, consistent with its function as an epithelial identity programmer in development. ^15-17 This is achieved through the protein’s functional interactions with the epigenetic remodeling factors p300 and KMT2C/D, a histone acetyltransferase adaptor and histone methyltransferases, respectively. These interactions enforce an epithelial transcriptional phenotype, thus reversing some cancer-promoting effects in experimental models. ^58,70 In this connection, clinically relevant mutations in the KMT2D gene were recently shown to promote phenotypic plasticity, favoring EMT and metastasis. ⁹⁷ One of the key GRHL2-repressed genes is ZEB1, a potent transcriptional driver of EMT. ^15,16 Interestingly, GRHL2 and ZEB1 repress each other in a negative feedback loop, which is affected by TGF-β or Wnt antagonists, favoring ZEB1 or GRHL2, respectively. ^16,65 Additional targets mediating the EMT-suppressive and tumor suppressive effects of GRHL2 have been identified. ³⁶

The tumor-suppressive effects of type I interferons in cancer, including breast cancer, are documented extensively. ^4,5,22,50 Double-strand RNA derived from retrotransposons is a major driver of interferon responses that promotes tumor rejection in immunotherapy. ^9,13,42,96 Tumor cell interferon responses are important for maintaining tumor dormancy thus suppressing re-activation and recurrence. ^40,67,75

Here, we report that GRHL2 promoted the induction of IFN-β and IFN-λ by double-strand RNA. GRHL2 enhanced the phosphorylation of IRF3 and relA/NF-kB, and the induction of IRF1 expression, transcription factors critical for IFN-I/III expression. Ectopic expression of IRF1 or a constitutively active mutant of IRF3 rescued interferon induction in GRHL2-knockout cells. Furthermore, GRHL2 expression correlated with increased time to recurrence in patient samples. These observations connect GRHL2 – an epithelial master programmer – with interferon-mediated innate immunity driven by double strand RNA.

RESULTS

GRHL2 promotes the induction of IFNB1 and IFNL1 in response to double strand RNA

MCF10aneoT is a KrasV12-transformed cell line derived from basal-like, non-tumorigenic MCF10a cells. ⁵⁹ Previously, we generated cell lines derived from MCF10aneoT cells with knockouts of both GRHL2 alleles via CRISPR/cas9, which induced Epithelial-Mesenchymal Transition (EMT). ⁵⁸; validation of the knockouts using two independent guide RNAs is shown in figure S1. PolyIC is a double-strand (ds) RNA, mimicking dsRNA viruses, that induces interferon pathway genes via the Toll-like Receptor TLR3, after internalization by endocytosis from growth medium. ⁶ PolyIC rapidly induced the transcription of the (type-I) IFNB1 and (type-III) IFNL1 genes in MCF10aneoT cells (figure 1A). This induction was blunted by GRHL2 knockout, using two independent sgRNAs sequences that were characterized previously. ⁵⁸ Quantitation of IFNB1 and IFNL1 proteins using ELISA assays confirmed that GRHL2 promoted their expression (figure 1B). The interferon protein concentrations obtained here by ELISA were within the same log-expression range as those reported previously in MCF10a and other cell lines. ^8,60,74 PolyIC also stimulates a set of Interferon Stimulated Genes (ISGs) either indirectly, through interferons, or directly, through transcription factors shared with interferon genes. ³ A selected set of ISG mRNAs was also found to be GRHL2/polyIC-dependent (figure 1C). Among these ISGs were two potential dsRNA receptors, MDA5/IFIH1 and RIG1. Neither individual nor combined siRNA-mediated knockdown of these receptors hindered IFN-λ1 mRNA induction by PolyIC in our cell system, however, indicating no role for MDA5 or RIG1 in the response to PolyIC administered in growth medium (figure S2), consistent with the unique role of TLR3 documented previously. ⁶ Moreover, GRHL2 knockout, surprisingly, increased the expression of TLR3, while not affecting the adaptor proteins MAVS and TRIF (figure S2).

Figure 1. GRHL2 promotes the induction of IFNB1, IFNL1 and ISGs by double strand RNA.

A). qRT-PCR results for IFN-β and IFN-λ mRNAs following induction with soluble PolyIC in wild-type control vs. GRHL2-KO cells, for the indicated times.

B). ELISA assays on cell lines stimulated with PolyIC as in (A).

C). qRT-PCR and western blot data on the expression of the indicated Interferon Stimulated Genes following stimulation with PolyIC (50 μg/ml) for the times indicated; western blot data represented the SG1 cell line.

D). (left panel): HT1080 cells or BT474 cells with or without over-expressed GRHL2 were compared by qRT-PCR for IFN-β1 and IFN-λ1 mRNA expression after PolyIC stimulation. (right panel): Induction of stable EMT with transient TGF-β treatment down-regulated IFN-β1 and IFN-λ1 expression.

To determine whether the effect of GRHL2 on interferon genes was limited to MCF10aneoT cells, we expressed GRHL2 in the (GRHL2-low) fibrosarcoma cell line HT1080 and the HER2-positive breast cancer cell line BT474; over-expression was validated by western blotting (figure S1). In both cell lines, GRHL2 over-expression enhanced IFNB1 and IFNL1 gene expression following PolyIC stimulation (figure 1D). The results demonstrated that GRHL2 promoted interferon expression in representative cell lines of both epithelial and mesenchymal origins. Conversely, the down-regulation of GRHL2 by transient TGF-β treatment of MCF10aneoT cells (reported previously) blunted the induction of these interferons (figure 1D). ¹⁶ These results indicated that GRHL2 promoted the expression response of interferon genes to polyIC; this promotion was neither mediated by an up-regulation of three double strand RNA receptors, nor their adaptor proteins, in response to GRHL2. The generality of this conclusion across cell lines was supported further by our previous report showing that GRHL2 expression correlated positively with interferon signaling pathway enrichment in tumors. ⁵⁸

We then utilized genomic approaches to identify pathways that were affected by GRHL2 in the presence or absence of PolyIC stimulation. To investigate the regulatory role of GRHL2 in gene expression changes in response to PolyIC treatment, we generated RNAseq libraries for control and GRHL2-knockout MCF10aneoT cells, with and without PolyIC treatment. The loss of GRHL2 down-regulated 650 genes and up-regulated 791 genes in MCF10aneoT cells without PolyIC stimulation (Figure 2A). The number of differentially expressed genes after PolyIC stimulation increased by 2-to-3 fold due to the loss of GRHL2 (Figure 2B). GSEA identified an overall down-regulation in Reactome Interferon Signaling in the GRHL2-KO cells, especially after PolyIC stimulation (NES=−2.08 and FDR=0.0; Figure 2C and 2F). A visual inspection of the expression of leading genes down-regulated by GRHL2 after PolyIC stimulation identified a substantial portion being up-regulated by the double strand RNA in the control cells (Figure 2D). This observation led us to the identification of 187 polyIC-induced and GRHL2-dependent genes, namely genes that are doubly dependent upon PolyIC and GRHL2 for expression (Figure 2E). This subset of genes was strongly enriched in pathways related to cytokine/interferon/antiviral response/toll-like receptor signaling. This inference was supported by TFEA.ChIP data analysis identified Interferon Response Factors (IRFs), STATs, and NF-κB/relA as potential transcription regulators for polyIC-induced and GRHL2-dependent genes (figure 2G).

Figure 2: polyIC-induced and GRHL2-dependent genes are enriched in interferon signaling response.

A) Scatter plots for averaged expression level and fold change of gene expression comparing GRHL2 KO to control cells, all before PolyIC stimulation. Black: upregulated in KO; Red: downregulated in KO; blue: all expressed genes.

B) Scatter plots for averaged expression level and fold change of gene expression comparing GRHL2 KO to control cells, all after PolyIC stimulation.

C) Gene set enrichment analysis by sorting genes by fold change of expression (GRHL2 KO/Ctrl) from high (left side) to low (right side) for cells before (green line) or after (blue line) PolyIC stimulation and then aligning against genes from the MisgDB gene set of Reactome interferon signaling. A valley on the right indicated an expression downregulation of the interferon signaling pathway. NES = normalized enrichment score.

D) Heat map visualization of gene expression values across GRHL2 KO and control cells, with or without PolyIC, for leading genes that explained the most downregulation of the Reactome interferon signaling pathway by GRHL2 deletion after the polyIC stimulation (panel C).

E) Venn diagram of PolyIC-induced genes, defined as upregulated by PolyIC stimulation in the control cells, and GRHL2-dependent genes, defined as downregulated in GRHL2-depleted cells as compared to control cells after polyIC stimulation. Indicated were genes overlapping with the Reactome interferon signaling pathway.

F) Reactome pathway enrichment analysis for polyIC-induced and GRHL2-dependent genes. Shown are top 10 enriched pathways.

G) TFEA.ChIP inference for transcriptional regulators of PolyIC-induced and GRHL2-dependent genes. Highlighted are IRF TFs and STAT TFs.

GRHL2 is a pioneer transcription factor that interacts with numerous regions of the genome to generate open chromatin conformations at epithelial genes specifically. ^7,44. To investigate the potential role of GRHL2 in regulating gene expression changes through modulating chromatin accessibility, we generated ATAC-seq libraries to characterize the chromatin accessibility landscape for GRHL2 knockout MCF10aneoT cells and control cells, with and without PolyIC treatment (paired with the RNAseq data). Initial analyses focused on ATAC-seq data generated without PolyIC treatment, aiming to define GRHL2’s role in regulating chromatin accessibility and identify the affected molecular pathways. Further analyses combined ATAC-seq data generated with PolyIC treatment, aiming to define genomic regions where polyIC-induced increase in chromatin accessibility occurs in the control cells but not in the GRHL2 KO cells and identify the associated genes. This analysis when combined with the RNAseq data help to define a subset of genes that are polyIC-induced and GRHL2 dependent for both gene expression and the underlying regulatory chromatin accessibility.

We investigated the potential of GRHL2 to affect epithelial vs. mesenchymal gene accessibility prior to PolyIC exposure (figure 3) and, subsequently, to affect interferon pathway-related genes after PolyIC exposure (figure 4). GRHL2 depletion induced a substantial re-organization of accessible chromatin with 17,655 regions increased and 20,000 regions decreased in accessibility (Figure 3A). ChIP-Atlas TF analysis with public ChIP-seq data identified a positive enrichment of GRHL2 binding for regions deceased in accessibility but a depletion for regions increased in accessibility (Figure 3B). Genomic region enclosing CDH1, TP63, and IL1A were shown as examples of decreases in chromatin accessibility, while those for TWIST2 and PPARG were included as examples of increases (Figure 3C). As expected, a decrease or increase in accessibility at regulatory regions predicted transcription down-regulation or up-regulation of their potential target genes (Figure 3D). Gene ontology analysis of open chromatin regions enriched in GRHL2-knockout cells, using Metascape against MSigDB hallmark gene sets demonstrated an enrichment in EMT pathway-related genes and TGF-β signaling, validating our approach (figure 3E). Conversely, the GRHL2-knockout cells showed decreased accessibility at NF-κB pathway genes, with potential detriment for interferon responses following stimulation (see data below). ²¹ Prediction of transcription factors enriched in open chromatin regions that decreased accessibility in the absence of GRHL2 using ChIP-Atlas indicated that GRHL2 promoted open chromatin associated with the transcription factors TP53, TP63, GRHL2, KMT2C and TFAP2c (Figure 3B) – factors associated with an epithelial phenotype and/or related to GRHL2 itself, ^58,62,90,97 further validating the approach.

Figure 3. GRHL2 depletion reduces chromatin accessibility and induces EMT.

A) Scatter plots for averaged chromatin accessibility and fold change of chromatin accessibility comparing GRHL2 KO to control cells at open chromatin regions, all before Poly IC stimulation. Yellow: genomic regions showing an increase in chromatin accessibility; Blue: genomic regions showing a decrease in chromatin accessibility; gray: no change in accessibility.

B) ChIP-atlas transcription factor enrichment analysis for genomic regions showing an increase in accessibility (yellow) and for genomic regions showing a decrease in accessibility (blue).

C) UCSC genome browser images for representative genes (CHD1, TP63, IL1A) that are associated with multiple genomic regions (at least seven) showing a decrease in accessibility (shaded in blue) and representative genes (TWIST2 and PPARG) that are associated with multiple genomic regions showing an increase in accessibility (shaded in yellow).

D) Gene set enrichment analysis by sorting genes by fold change of expression (GRHL2 KO/Ctrl) from high (left side) to low (right side) and then aligning against genes associated with multiple genomic regions that exhibited an increase in chromatin accessibility (left panel) or against genes associated with multiple genomic regions that exhibited a decrease in chromatin accessibility (right panel). NES = normalized enrichment score.

E) Functional enrichment analysis against MSigDB hallmark gene sets for genes associated with multiple genomic regions that exhibited an increase in chromatin accessibility (left panel) or for genes associated with multiple genomic regions that exhibited a decrease in chromatin accessibility (right panel).

Figure 4. polyIC-induced and GRHL2-dependent changes in accessibility are linked to interferon signaling response.

A) Venn diagram of PolyIC-induced genomic regions, defined as increase in chromatin accessibility by PolyIC stimulation in the control cells, and GRHL2-dependent genomic regions, defined as decrease in accessibility in GRHL2-depleted cells as compared to the control cells after polyIC stimulation. The overlapping defined PolyIC-induced and GRHL2-depedent genomic regions.

B) Heat map visualization of chromatin accessibility across samples for PolyIC-induced and GRHL2-depedent genomic regions. The level of chromatin accessibility measured by ATAC-seq read density for each open chromatin region (column) was z-score transformed across samples.

C) ChIP-atlas transcription factor enrichment analysis for PolyIC-induced and GRHL2-depedent genomic regions.

D) Top 10 hits on biological processes from gene ontology enrichment analysis for targets genes predicted by GREAT for PolyIC-induced and GRHL2-depedent genomic regions.

E) Venn diagram of polyIC-induced and GRHL2-induced genes defined from RNA-Seq analysis and from ATAC-seq analysis as in panel D.

F) UCSC genome browser images for examples of polyIC-induced and GRHL2-induced genes shared between RNA-Seq analysis and ATAC-seq analysis (IFNL1, IKBKE, and CXCL10). Shaded in yellow are genomic regions showing an polyIC-induced increase in chromatin accessibility in the control cells but no such increase when GRHL2 is depleted.

G) A GRHL2 binding site in the IFNL1 promoter was identified by CHIP-seq (see Materials and Methods); it aligned with a peak of accessible chromatin identified by our ATAC-seq data. Deletion of this binding site compromised polyIC/IL1 induced IFNL1 promoter activity in transient reporter assays, graphed here as ratio of IFN-λ-LUC to thymidine kinase-Renilla activity.

Following stimulation with polyIC, interferon-related effects of GRHL2 upon chromatin accessibility were observed (figure 4). Of the 4,872 genomic regions with chromatin accessibility increased by polyIC, 45% were found to be GRHL2-dependent (Figure 4A,B). High-confidence transcription factors identified in the doubly-dependent genomic regions using CHIP-ATLAS included NF-kappaB/relA, with high significance for interferon pathways (Figure 4C). Pathway analysis of genes associated with doubly-dependent genomic regions, utilizing GREAT, ⁶¹ identified apoptosis signaling (death receptors), leukocyte immunity, and type I interferons as high confidence pathways (figure 4D). About 36% (67/187) of the doubly dependent genes identified by RNAseq were present in our list of doubly dependent genes identified by ATAC-seq, demonstrating that the transcription of a significant number of these genes was regulated by chromatin accessibility; many of these genes were interferon pathway-related (figure 4E). GRHL2/polyIC dependent regions of increased chromatin accessibility were identified in the regulatory regions of several interferon pathway-related genes, including IFNL1, IKBKE, and CXCL10 (Figure 4F).

Through combining our ATAC-seq data with public databases reporting GRHL2 binding sites using CHIP-seq, we also identified a region close to the IFN-λ promoter (but not the IFNB1 promoter) transcription start site (−487 to −475) containing a consensus GRHL2 binding site, GAAACAGGATCT.. Deletion of this binding site attenuated the induction of the IFNL1 gene in reporter assays, suggesting that GRHL2 may also activate IFNL1 transcription by direct interaction of the protein with its cognate binding site (figure 4G).

GRHL2 enhances IRF3 and relA activation and IRF1 expression, promoting interferon responses.

Interferon Response Factor-3 (IRF3) is ubiquitously critical for activating the transcription of the IFNB1 and IFNL1 genes, while more variable roles have been reported for NF-κB/relA and IRF-1. ^38,68,94 Using siRNA knockdowns, we confirmed that IRF3, IRF1 and relA were important for interferon induction in MCF10aneoT cells (Figure S3). We focused on IFNL1 for mechanistic studies in light its greater responsiveness to GRHL2. Dimerization and nuclear translocation of activated IRF3 requires serine phosphorylation on several sites, including serine-386, for which the kinase TBK1 is primarily responsible. ⁵⁶ Using western blotting with IRF3 phosphoserine-386-specific antibody, we observed that IRF3 phosphorylation was defective in GRHL2-knockout cells (figure 5A). The upstream phosphorylation of TBK1 itself on serine-172 indicates activation of TBK1, via the MAVS or TRIF adaptor proteins. TBK1 phosphorylation/activation was defective in GRHL2-knockout cells as well (figure 5A). The siRNA-mediated knockdown of IRF3 blunted interferon gene responses in GRHL2-WT cells, consistent with its established function (figure S3). Conversely, a previously characterized constitutively active mutant of IRF3 --in which five serine phosphoacceptor sites were replaced by aspartates (IRF3-5D) ⁵⁶-- was ectopically expressed using a doxycycline-inducible lentiviral vector in GRHL2-KO cells. In this cell line, doxycycline induction reconstituted IFN-λ induction to a level comparable to GRHL2-WT cells (figure 5B). These results confirmed the gene ontology pathway analysis, indicating that GRHL2 promoted the TBK1/IRF3 pathway. Additionally, IRF1, which was up-regulated by PolyIC selectively in GRHL2-expressing cells, also rescued IFN-λ induction in GRHL2-knockout cells when over-expressed in a doxycycline-inducible vector (figure 5B). When combined with the observation that IRF1 was crucial for IFN-λ induction (figure S3), this result implicated IRF1 in the GRHL2-promoted induction of IFN-λ.

Figure 5. GRHL2 promotes the activation of IRF-3 and relA by polyIC.

A) Phosphorylation of IRF3 (serine386) and TBK1 (serine-172) are promoted by GRHL2. Lysates from GRHL-WT or GRHL-KO MCF10aneoT cells induced with PolyIC for the indicated times were analyzed on Western blots using the indicated antibodies. The graphs represent quantitation of ratios of phosphorylated to total proteins, averaged over replicates.

B) Transient over-expression of IRF3-5D (left) or IRF1 (right) rescues IFNL1 expression in GRHL2-knockout cells. Cell lines with inducible expression (Materials and Methods) were treated for 12 hours with doxycycline, followed by treatment with polylC; qRT-PCR results are shown. (boxed graph). Doxycycline treatment failed to induce significant IFNL1 levels after PolyIC treatment, in control GRHL2-KO cells without the IRF1 or IRF3 transgene.

C). RelA phosphorylation (S536) is promoted by GRHL2.

In light of its important role in interferon induction by double strand RNA and the interaction of NF-kB with IRF3 at the IFN-β promoter, ^19,68 and guided by the gene ontology pathway results above, we then investigated the activation of NF-κB. We chose to use IKK complex-mediated relA phosphorylation at serine-536 – which activates the intranuclear association of relA with critical co-factors for transactivation -- to report NF-κB activation; this crucial phosphorylation is considered a more reliable reporter for NF-κB activation than the phosphorylation of IkB, which is confounded by rapid IkB degradation. ¹⁴ Western blotting for relA phosphoserine-536 revealed a significant positive effect of GRHL2 after PolyIC stimulation (figure 5C). In this connection, we functionally tested the effect of stimulating NF-κB activity with IL-1α protein. Interestingly, exogenous IL-1α reconstituted the ability of GRHL2-knockout cells to induce interferon expression in response to polyIC, whereas IL-1α alone had only minimal effect (figure S4). These results suggested that defective NF-κB signaling following PolyIC stimulation was significant for the attenuated interferon induction observed in GRHL2-knockout cells, although how IL1 receptor activation overcomes the defect in IRF3 activation noted above remains to be addressed.

GRHL2 expression level correlates with delayed recurrence in breast cancer

Type-I interferon signaling confers dormancy, suppressing or delaying tumor recurrence. ^40,67,75 We therefore investigated the correlation of GRHL2 expression in a 554-sample human breast tumor tissue microarray (TMA) representing all classes of breast cancer and containing comprehensive clinical data that was characterized previously. ⁶⁹ Representative IHC images are shown in figure S5A and a link to clinical data is in figure S5B. Interestingly, the time to recurrence correlated positively with GRHL2 expression in the range of 0 to +2/+3, with no significant difference between +2 vs. +3 expression levels (figure 6A). This analysis was not amenable to stratification by subtype or disease stage, as sample sizes became insufficient for statistical significance. These data implicate GRHL2 as potential recurrence suppressor in breast cancer, consistent with its positive effects on interferon induction.

Figure 6. GRHL2 expression affects time to recurrence in breast cancer.

A. GRHL2 expression level correlates with time to recurrence in breast cancer. The Indiana University breast cancer tumor tissue microarray (N=554, all subtypes) characterized previously 69 was subjected to immunohistochemistry using a GRHL2 monoclonal antibody, scored and quantitated as described in Materials and Methods. B. Expression of a GRHL2 transgene suppresses tumor recurrence in a mouse model of breast cancer. In MTB/TAN mice, mammary tumors are generated by doxycycline induction of a tet-inducible HER2/neuNT transgene. The removal of doxycycline results in rapid tumor regression and the eventual appearance of recurrent tumors at or near the original injection site. These recurrent tumors are HER2/neuNT-null, driven by epithelial-mesenchymal transition, and GRHL2-low. 17,64 To determine the effect of GRHL2 upon tumor recurrence, a stable MMTV-GRHL2 transgenic mouse line was generated and validated by genotyping and western blotting (Materials and Methods, figure S6). This transgenic line, or a control line, was crossed with MTB/TAN mice and genotyped as described in Materials and Methods. Primary tumors were induced in both groups by doxycycline; growth and frequency of these tumors was unaffected by the GRHL2 transgene. After removal of doxycycline (when tumor reached about 200mm2 cross-sectional area), tumors regressed rapidly to an undetectable state, as reported previously. {Moody, 2005 #1168) At variable time points, recurrent tumors appeared, which were scored in the GRHL2-transgenic vs. GRHL2-WT (MTB/TAN) mice. The experimental scheme and recurrence frequencies are shown. See text for explanation of significance values.

To functionally test the effect of GRHL2 on breast cancer recurrence, we utilized a mouse model (MTB/TAN) in which the expression of the neuNT oncogene is doxycycline inducible. ⁶⁴ In this model, mice are administered doxycycline, inducing neuNT expression and large, although localized, mammary tumors. Removal of the doxycycline causes rapid tumor regression, but recurrent tumors, which demonstrate an EMT-like phenotype, appear with 20-40% frequency in a Snail-dependent fashion, demonstrating that EMT drives tumor recurrence in this model. ⁶⁴ Previously, we reported that GRHL2 mRNA was significantly down-regulated in the recurrent tumors compared to the initial tumors. ¹⁷ To test the potential of GRHL2 to suppress recurrence, we generated a stable MMTV-GRHL2 transgenic mouse line (Materials and Methods, figure S6 and figure 6). This transgenic line was crossed with MTB/TAN mice and the recurrences of GRHL2-transgenic vs. GRHL2-WT mice were scored. Primary tumor growth occurred in the presence or absence of the GRHL2 transgene, and tumors regressed upon the removal of doxycycline in both groups, as reported previously; ⁶⁴ the experimental scheme and recurrence frequencies are depicted in figure 6. A single mouse in the GRHL2-transgenic group – in which GRHL2 expression could not be confirmed – had a recurrence with unique tumor characteristics (widespread subcutaneous metastases); all twelve other GRHL2-transgenic mice were recurrence-negative. Including this unique mouse, the recurrence frequency was 7.7% in the GRHL2-transgenic group (1/13). In the control group, 27.8% of the mice had recurrences, a trend indicating that GRHL2 suppressed tumor recurrence that did not, however, reach statistical significance (p<0.16 by Chi-squared test, p<0.36 by Fisher’s exact test). Exclusion of this single mouse with unique tumor characteristics yielded statistical significance (p=0.05 by Fisher’s exact test). When combined with the patient data, this trend indicated that GRHL2 suppresses the recurrence of breast cancer, via interferon induction and/or other mechanisms.

DISCUSSION

This study connects two concepts. First, the epithelial-mesenchymal transition (EMT) protects tumor cells against immune surveillance by a variety of mechanisms. ^{2,24,41,46,48,58,78,80,86,87} Secondly, interferon response gene signatures correlate with improved tumor surveillance and longer time to recurrence. ^12,18,63 Prior to this study, possible connections between oncogenic EMT -- or conversely, the epithelial phenotype enforced by GRHL2 -- and type-I/III interferon gene regulation in cancer were unclear. Nevertheless, previous reports suggested a potential link. First, interferon-III induction by Respiratory Syncytial Virus (RSV) is attenuated by TGF-β induced EMT in lung alveolar epithelial cells – modeling asthma -- through repression of IRF1, contributing to RSV infectivity. ⁹⁴ Also, an EMT-like mammary epithelial cell line suppressed IFN-β gene transcription via the secretion of the IL-6 family member Oncostatin-M (OSM), through unknown mechanisms; ²³ OSM was not expressed in our cell culture models, however.

Herein, we have shown that GRHL2 -- a potent EMT suppressor -- enhances interferon responses to double-strand RNA. These observations may be understood in the context of the developmental biology of GRHL2 and interferon gene regulation. Developmentally, in its capacity as a pioneer transcription factor (i.e. a factor capable of specific, functional binding to DNA sequences in closed chromatin), GRHL2 is critical for driving the first commitment to a generic epithelial lineage, the embryonic stem cell-to-epiblast transition. ⁷ Importantly, embryonic stem cells fail to mount interferon responses to viral infection or Pathogen-Associated Molecular Patterns, relying instead on the constitutive expression of a small subset of antiviral proteins. ⁹³ The interferon response pathway becomes functional upon implantation, occurring shortly after the onset of GRHL2 expression. ^7,93 These data, combined with our current study, suggest that the developmental acquisition of the interferon response pathway requires GRHL2 expression, which will be interesting to test in subsequent mouse developmental studies.

We focused on double-strand RNA as an interferon inducer. Retrotransposon-derived double-strand RNA intermediates contribute substantially as a driver of interferon responses that promote endogenous and immunotherapy-enhanced tumor rejection. ^9,13,42,96 Mechanistically, the rapid transcriptional induction of IFN-β and IFN-λ genes following exposure to double strand RNA occurs downstream of the three major receptors, RIG-1, MDA-5 and TLR3, via multiple pathways. Adaptor proteins (MAVS or TRIF) activate the TBK1-mediated phosphorylation of IRF-3, permitting its phosphorylation, homodimerization and nuclear translocation. The activation of NF-κB pathway downstream of these receptors plays a critical supporting role, and relA functionally interacts with IRF3 at the IFN-β promoter for optimal transcription. ⁶⁸ Remarkably, GRHL2 promoted the activation of both pathways to enhance IFN-β and IFN-λ induction. TBK1, a member of the IKK family of kinases, can directly or indirectly activate both the IRF3 and NF-κB pathways, which then synergize to induce interferon gene expression. ^43,81,95 TBK1 activation was attenuated in the GRHL2-knockout cells; while the underlying mechanism is unclear, GRHL2 could potentially control both signaling pathways through TBK1.

In addition, IRF1 induction by PolyIC was enhanced by GRHL2. Moreover, IRF1 knockdown attenuated interferon induction and ectopic IRF1 expression rescued interferon expression in GRHL2-knockout cells, formally establishing the critical role of IRF1 in this effect of GRHL2. IRF1 as well as IRF3 and p65/relA) was previously reported to be critical for IFN-λ expression, consistent with these observations. ^28,94 Interestingly, IRF1 contributes to IRF3 phosphorylation, by blocking IRF3 interactions with a phosphatase, PP2A ⁸⁹. GRHL2 could thus promote IRF-3 activation both through stimulating TBK1 activation as well as through induction of IRF1 expression. Superimposed upon this regulation of cell signaling, the IFN-λ promoter was found to contain a CHIP-validated consensus GRHL2 binding site that enhanced the transcriptional activity of the promoter.

The previously reported effects of GRHL2 on tumor growth, metastasis and drug resistance are highly contextor even cell line-dependent. ^{29,33,36,47,71,84,91,92} The basis for this high degree of context dependence is multifactorial. First, GRHL2 inhibits TGF-β signaling, a notorious “double edged sword” in cancer. ¹¹ Secondly, GRHL2 activates cell proliferation through cell cycle genes in certain tumor types (e.g., luminal breast cancer). ²⁷ Perhaps relatedly, GRHL2 activates estrogen receptor target genes, substituting for estrogen receptor itself and promoting resistance to anti-estrogen therapy in breast cancer; in this connection, functional relationships between GRHL2 and steroid receptors are well characterized. ^71,72 GRHL2 also induces hTERT expression, which has been proposed to promote head and neck cancer progression. ¹⁰ On the other hand, by reversing EMT, GRHL2 suppresses tumor progression in EMT-dependent mouse tumor models of breast and ovarian cancer. ^15,16 A recent report revealed a negative correlation between GRHL2 expression and overall survival in breast cancer. ⁹¹ This latter discrepancy of our results vs. this latter report may perhaps be attributed to differences between protein expression data vs. mRNA data. Additionally, GRHL2 mRNA expression was previously reported ⁹¹ to affect the survival effect of basal – not luminal -- breast cancer patients, whereas our recurrence data included all subtypes of breast cancer. We were unable to stratify our tissue microarray data based on breast cancer subtype, due to incomplete data and lack of statistical significance in individual subtype comparisons. Finally, the previous study⁹¹ reported on (HER2-negative) 4T1-derived tumors, whereas we used HER2-positive tumors, perhaps explaining this discrepancy.

EMT is one of several common mechanisms of post-treatment recurrence in cancer, including the HER2⁺ breast cancer represented in our mouse model. ^{20,34,35,49,53,66} The effect of GRHL2 as a recurrence suppressor that we observed is consistent with the tumor tissue microarray data indicating that GRHL2 protein levels correlate positively with time to recurrence. Although interferon signaling capacity portends better outcome across cancer types, it most likely is only one of a large set of factors affected by GRHL2 with downstream consequences for tumor progression.

Recently, intratumoral microbiome compositions have been shown to exert important tumor-regulatory effects. ^26,45,85 In light of the potent cell cycle arrest and pro-apoptotic effects of interferons downstream of certain microbial infections, it is not inconceivable that the tumor microbiome could select for EMT-like/GRHL2-negative cells, because these cells fail to express interferons. In this connection, GRHL2 critically regulates epithelial cell polarity, perhaps affecting the localization and function of pattern recognition receptors, as reported recently. ^79,82

Immune cell cytokines, particularly arising from tumor macrophages, are thought to drive EMT and metastasis of tumor cells, which can be reversed clinically by CSF-1R inhibition, restoring tumor interferon responses in vivo. ⁷⁷. Additionally, IFN-λ has recently been shown to play an important role in immunotherapy-mediated tumor rejection. ⁵² The GRHL2 gene is shown here to enforce interferon responses. Subsequent studies addressing the regulation of the GRHL2 gene so as to maintain its expression, or development of molecular mimics of its function, may provide tools to enhance immunotherapy.

MATERIALS AND METHODS

Cell culture, siRNA, generation of cell lines and Western blotting

Culture conditions and characterization of GRHL2 knockdown cells derived from MCF10aneoT and GRHL2-overexpressing cells derived from HT1080 and BT474 were as described previously. ^58,70. Briefly, we utilized lentiviral constructs in the pLenti-Crisprv2 vector that were validated and provided by the laboratory of Brigid Hogan ³². Lentiviruses were packaged by transfection of 293T cells (with co-transfection of psPAX2 and pCMV-VSV-G), collected and filtered 48 hours post-transfection. Lentiviruses were infected as we described previously, ^31,58,70 followed by western blotting of total puromycin-resistant cells (figure S1). For induction of interferons, PolyIC (Sigma P9582, 50 μg/ml) or IL1-α (R and D/Biotechne; 10ng/ml) were added to culture media. To generate cell lines with doxycycline-inducible IRF1 or IRF3-5DA, MCF10aneoT cells were first infected with pLenti-CMV-rTTA3-hygro lentivirus (kindly provided by John Karijolich, Vanderbilt University) ⁹⁹ and selected with 300 ug/ml hygromycin. IRF1 and IRF3-5D sequences were PCR amplified from clones kindly provided by Alan Brasier (University of Wisconsin )⁹⁴ or John Hiscott (Pasteur Institute, Rome, via Matthew Frieman, University of Maryland). ⁷⁶ PCR fragments were subcloned into the BstBI-XbaI site of pLenti CMVtight eGFP Blast (Addgene), thus removing the EGFP insert, and sequence-verified. Following lentivirus packaging and infection, cells were selected in Blasticidin (25 ug/ml) and doxycycline-induced (1 μg/ml) expression was confirmed by Western blotting.

Western blots were performed on total cell lysates dissolved directly in 1X Lithium Dodecyl Sulfate Sample Buffer (Life Technologies) + 100 mM dithiothreitol as described previously using a BioRad Transblot apparatus and GE/Amersham 680 Imager for image acquisition and quantitation. ^31,58,70 Antibodies were as follows:

Antigen	Application	Source	Catalog number
MDA-5	WB	Cell Signaling Technologies	5321
RIG-1	WB	Cell Signaling Technologies	3743
IRF3	WB	Cell Signaling Technologies	11904
P-IRF3 (386)	WB	Cell Signaling Technologies	37829
IRF-1	WB	Santa Cruz Biotechnology	Sc-74530
relA/p65	WB	Cell Signaling Technologies	8242
P-relA/p65	WB	Cell Signaling Technologies	3033
TBK1	WB	Cell Signaling Technologies	3504
P-TBK1	WB	Cell Signaling Technologies	5483
MAVS	WB	Cell Signaling Technologies	3993
GRHL2	IHC	Life Technologies	MA5-31388
TRIF	WB	Cell Signaling Technologies	4596
IFN-λ	ELISA	Biotechne	DY1598B
IFN-β	ELISA	Biotechne	DY815-05

SiRNA transfections were performed in 12-well confluent dishes of confluent cells, using Lipofectamine RNAimax according to the manufacturer’s protocol (Invitrogen/Life Technologies) and pre-designed Silencer Select siRNAs, with the following sequences:

Target	siRNA ID (Life Technologies Cat# 4427037 or 4427038)
IL1A-1	S7268
IL1A-2	S7267
relA-1	s11916
relA-2	s11915
IRF1-1	S7503
IRF1-2	S7501
IFIH1 (MDA-5)-1	S34498
IFIH1 (MDA-5)-2	S34499
DDX58 (RIG1)-1	S223614
DDX58 (RIG1)-2	S223615
IRF3-1	S7507
IRF3-2	S7508

After overnight incubation, cells were passaged into 6-well dishes for PolyIC treatment, protein lysate preparation and RNA preparation, 36 hours post-transfection.

qRT-PCR

Quantitative RT-PCR was performed as described previously ⁵⁸, using the following primer sets:

Gene	Forward Primer	Reverse Primer
IFNL1	CTAGACCAGCCCCTTCACAC	AAGGTGACAGATGCCTCCAG
IFNB1	CATTACCTGAAGGCCAAGGA	CAGCATCTGCTGGTTGAAGA
CXCL10	TTCCTGCAAGCCAATTTTGT	TTCTTGATGGCCTTCGATTC
IRF1	CGATACAAAGCAGGGGAAAA	GTGGAAGCATCCGGTACACT
IFIH1	AGGAAATCGCAAAGAACGTG	CATCATCACCACCCTCATCA
IL1A	TGCCTGAGATACCCAAAACC	AACAAGTTTGGATGGGCAAC
GAPDH	TGCACCACCAACTGCTTAGC	GGCATGGACTGTGGTCATGAG
mGeno	GTGGTTTCCTGACTTGGTTTGG	TTGGCTGTCACTTGCTTTGC

Data were represented in expression units = 2^−ΔCt, using GAPDH as internal control.

Transgenic mice and tumor recurrence

Mice bearing the doxycycline-inducible (rat) neuNT oncogene (tetO-neuNT,or “TAN”) and the recombinant tetactivator under the control of an MMTV promoter (MMTV-rTTA, or “MTB”) were kindly provided by Lewis Chodosh (University of Pennsylvania); these were maintained and genotyped as described previously ⁶⁴. MMTV-GRHL2-FLAG transgenic mice were generated by subcloning mouse GRHL2-FLAG coding sequence into the EcoRI site of MMTV-SV40-BSSK (Addgene). Linearized MMTV promoter-GRHL2 cDNA fragment (SaII-SpeI) was microinjected into one pronucleus of one-cell FVB embryos. After culturing to the two-cell stage, embryos were implanted into the oviduct of pseudo-pregnant CD-1 females and allowed to develop to term. Founder pups were genotyped using primers mGeno-F2 and mGeno-R2, yielding an 850bp PCR product. Upon weaning, pups were genotyped for transgene integration using tail biopsies and PCR amplification. Once founders were identified, they were backcrossed to wild-type FVB and maintained as hemizygotes.

MTB/TAN mice, either with wild-type GRHL2 (“control”) or with MMTV-GRHL2 were induced to form mammary tumors with 1mg/ml doxycycline (drinking water) until tumors reached an area of 200 mm²; removal of the doxycycline result in rapid regression of tumors as reported previously. ⁶⁴ Mice were monitored for recurrences for one year following regression. Palpable recurrent tumors localized to the site of injection were scored. All procedures were approved under IACUC protocol number 11-0706.

Tumor microarray and Immunohistochemistry

IHC staining was performed using a Dako Plus autostainer (Dako, Inc, Carpinteria, CA, USA), on the breast tumor tissue microarray described previously.⁶⁹ The clinical data with corresponding individual GRHL2 IHC scores are in Supplementary figure S5.

In brief, following antigen retrieval in citrate buffer of pH 6.0 in 97-98°C water bath for 37 min with 12 min cooling at RT and blocking steps: Peroxidazed 1 - for 5 min, Background Sniper for 10 min, sections were incubated in a mouse monoclonal GRHL2 (CL3760) Ab for 60 min (1:500), followed by incubation in a MACH 4 mouse probe and HRP-Polymer for 20 min each. Immunostaining was detected using 3,3′-diaminobenzidine (DAB) peroxidase chromogen substrates for 10 min. Sections were counterstained with hematoxylin, dehydrate in 95% Ethanol, 3x100% Ethanol, xylene and cover-slipped. Reagents: Peroxidazed 1 - Biocare Medical, PX968M; Background Sniper - Biocare Medical, BS966MM; Mouse monoclonal GRHL2 (CL3760) Ab primary – Life Technologies, MA5-31388; MACH 4 Universal HRP-Polymer kit - Biocare Medical, Concord, CA, USA, M4U534; DAB Quanto, 3,3′-diaminobenzidine -Thermo Scientific Lab Vision, TA-125-QHDX. TMAs were de-identified at Indiana University prior to transfer to WVU, so no additional approvals for Human Subjects were required.

RNAseq, ATAC-seq and bioinformatics

To investigate the regulatory role of GRHL2 in gene expression changes in response to PolyIC treatment, we generated RNAseq libraries for GRHL2 knockout cells (MCF10aneoT) and control cells, with and without PolyIC treatment (50 μg/ml, 3 hours). Total cellular RNA from each condition (in triplicate) was extracted from ~2x10⁶ cells using Qiagen RNeasy Plus kit (Qiagen). Library preparations, sequencing reactions and bioinformatic analysis were conducted at GENEWIZ, LLC. (South Plainfield, NJ, USA). RNA was quantified using Qubit 2.0 Fluorometer (Life Technologies, Carlsbad, CA) and integrity was assessed using Agilent TapeStation 4200 (Agilent Technologies, Palo Alto, CA). Libraries were prepared using the NEBNext Ultra RNA Library Prep Kit for Illumina (E7760, NEB, Ipswich, MA, USA), with initial oligo(dT) enrichment. Enriched mRNAs were fragmented by incubation in first-strand synthesis buffer for 15 minutes at 94 °C. First strand and second strand cDNA were subsequently synthesized. cDNA fragments were end repaired and adenylated at 3’ends, and universal adapters were ligated to cDNA fragments, followed by index addition and library enrichment by PCR with limited cycles. The sequencing library was validated on the Agilent TapeStation (Agilent Technologies, Palo Alto, CA, USA), and quantified by using Qubit 2.0 Fluorometer (Invitrogen, Carlsbad, CA) as well as by quantitative PCR (KAPA Biosystems, Wilmington, MA, USA). The sequencing libraries were clustered on a single lane of a flowcell. After clustering, the flowcell was loaded on the Illumina HiSeq 4000 and sequenced using a 2x150bp Paired End (PE) configuration.

RNA-Seq data analysis followed established pipeline as previously. ²⁵ RNA-Seq read alignment to the human genome (hg38) was performed by subread-aligner, ⁵⁵ read count summarization at gene level based on RefSeq annotation by featureCount, ⁵⁴ RPKM quantification of gene expression by in-house script, prediction of differentially expressed (DE) genes by EdgeR (FC > 1.5 and FDR < 0.05) ⁷³, functional enrichment analysis against Reactome pathways (by GSEA or Metascape) ^83,100, visualization of gene expression (by MeV) ³⁷ and transcriptional regulator prediction by TFEA.ChIP. ³⁹ For DE gene analysis, we included genes that were expressed at least in three samples (RPKM > 3). Functional enrichment analysis by Metascape targeted differentially expressed genes, which should be expressed at least in one condition from a comparison. Therefore, we used genes expressed at least in one condition as the background, instead of using all genes, (that would include some not expressed in either condition).

ATAC-seq data analysis used pipelines established previously. ²⁵ Briefly, ATAC-seq libraries were prepared in-house following the Omni-ATAC protocol using a Tagment DNA Enzyme and Buffer Large Kit (Illumina, San Diego, CA, USA, 20034198) with ~50K cells per sample. The libraries were sequenced at the Genomics Core Facility at Marshall University with Illumina Nextseq2000. ATAC-seq data analysis used procedures established previously ²⁵. ATAC-seq reads were mapped to the human reference genome using bowtie 2. ⁵¹ Reads mapped to multiple genomic positions were removed. For reads mapped to the same genomic coordinates, only one copy was kept. ATAC-seq read enriched genomic regions (or accessible chromatin regions) were predicted by MACS3. ⁹⁸ Differential accessible regions (DARs) were predicted by EdgeR (FC > 2 and FDR < 0.01). TF enrichment analysis using public ChIP-Seq data for DARs were conducted with the online ChiP-atlas Webserver. ¹⁰¹ Potential target genes associated with a DAR were annotated by GREAT. ⁶¹ Visualization of ATAC-seq reads across genomic locus was done with IGV. ⁸⁸ Sequencing data were deposited to GEO with the accession # GSE241852 for RNAseq and GSE241853 for ATAC-Seq. (Available at present using a secure token for review purposes for both data sets: cbergqamjlgltir.)

Reporter assays

The wild-type IFNL1 promoter region was amplified from human genomic DNA using IFNL-F:TAAACCAATGGCAGAAGCTCCTTC and IFNL-R: GCCATGGCTAAATCGCAACT, yielding a 2,228 bp fragment which was gel purified and re-amplified with similar primers containing NheI (for F) and XhoI (for R) sites. The PCR fragment was subcloned into pGL4.14 and sequenced. For deletion of the GRHL2 consensus binding site, the forward primer TTTAAAGGATCCCATCACCCAGGCTGGAGTGC and reverse primer TATATAGGATCCAGTTTTGCTCTTGTTGCCCA were used in conjunction with the reverse and forward primers above, generating two sub-fragments that were ligated together via the BamHI site and subcloned into pGL4.14; this insert was sequence-verified. Transient transfections were performed using Lipofectamine 2000 as per the manufacturer’s protocol (Life Technologies) including a 1:1 ratio of reporter plasmid to internal control plasmid (pRL-TK-Ren). Cells were re-plated 16 hours after the transfection, grown for an additional 24 hours, lysed in Glo-lysis buffer and assayed in luciferase and Renilla assay reagents (Life Technologies).

Statistics

Comparisons between quantitative values were evaluated using two-sample unpaired t-tests with Welch correction in Microsoft Excel. Frequency of recurrent tumor in mice were compared using a Fisher Exact Test using R version 4.2.1. P-values less than 0.05 were considered statistically significant in all tests.

highlights.

This manuscript demonstrates that Grainyhead-like-2, a core master programmer of epithelial cell identity, promotes type-I/III interferon responses to double strand RNA and suppresses tumor recurrence. The work is significant because it connects the epithelial gene expression program with interferon response capability, with implications for cancer and development.

Data availability

Sequencing data were deposited to GEO with the accession # GSE241852 for RNAseq and GSE241853 for ATAC-Seq. (Available at present using a secure token for review purposes for both data sets: cbergqamjlgltir.)