The transcription factor FOXM1 regulates the balance between proliferation and aberrant differentiation in head and neck squamous cell carcinoma

Sustained expression of FOXM1 is a hallmark of nearly all human cancers including squamous cell carcinomas of the head and neck (HNSCC). HNSCCs partially preserve the epithelial differentiation program, which recapitulates fetal and adult traits of the tissue of tumor origin but is deregulated by genetic alterations and tumor‐supporting pathways. Using shRNA‐mediated knockdown, we demonstrate a minimal impact of FOXM1 on proliferation and migration of HNSCC cell lines under standard cell culture conditions. However, FOXM1 knockdown in three‐dimensional (3D) culture and xenograft tumor models resulted in reduced proliferation, decreased invasion, and a more differentiated‐like phenotype, indicating a context‐dependent modulation of FOXM1 activity in HNSCC cells. By ectopic overexpression of FOXM1 in HNSCC cell lines, we demonstrate a reduced expression of cutaneous‐type keratin K1 and involucrin as a marker of squamous differentiation, supporting the role of FOXM1 in modulation of aberrant differentiation in HNSCC. Thus, our data provide a strong rationale for targeting FOXM1 in HNSCC. © 2019 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd.


Introduction
Transcription factors of the Forkhead family are crucial regulators of embryonic development and many vital functions in adult tissues [1]. These proteins also play a pivotal role in the progression of cancers. The most important Forkhead Box (FOX) transcription factor involved in a multitude of pro-tumorigenic processes is FOXM1 [2,3]. The main gene regulatory activity of FOXM1 is attributed to cell cycle and mitotic control [4,5]. However, recent studies revealed additional activities of FOXM1 as a co-regulator of oncogenic pathways [3,6]. In addition, FOXM1 expression is a major predictor of poor outcome across a large set of cancer types [7]. This strongly supports the concept of FOXM1 as an actionable target for therapy [3,6] and encourages further exploration of FOXM1 function.
Squamous cell carcinomas of the head and neck (HNSCC) are a heterogeneous group of deadly cancers associated with loco-regional recurrences and distant metastases caused by the invasive and metastatic behavior of cancer cells [8]. The most striking feature of all SCCs is their ability to preserve an epithelial differentiation program [9], which recapitulates many specific traits of the tissue of origin of the tumor. However, in contrast to normal tissues, tight control over the fate and size of the proliferating and differentiating cell populations is lost in SCCs due to genetic and epigenetic alterations in critical regulatory mechanisms [9], thus causing, in HNSCC, aberrant epithelial differentiation and stratification featured in particular by altered and heterogeneous expression of keratins [10][11][12].
Despite a large body of evidence that FOXM1 has a role in HNSCC [13][14][15][16], its oncogenic functions in this type of cancer are not yet sufficiently understood. In 108 V Roh et al that regard, it is noteworthy that FOXM1 is upregulated in premalignant oral lesions and persisting in primary HNSCCs and lymph node metastases [17]. Ectopic FOXM1 overexpression in oral keratinocytes induces a hyperproliferative phenotype and suppresses the expression of epithelial differentiation-specific genes [18] via a mechanism that includes differential regulation of DNA methylation [15]. Also, upon loss of p53 or overexpression of c-Myc, ectopic FOXM1 partially rescued primary human keratinocytes from terminal squamous differentiation, while promoting proliferation and genome instability [19]. In turn, differentiating normal human keratinocytes lose FOXM1 expression [20], thus suggesting that downregulation of FOXM1 is part of the keratinocyte differentiation program. However, the effect of FOXM1 knockdown or overexpression on the differentiation state of HNSCC cells remains elusive so far. Given that FOXM1 is expressed in HNSCCs also early on during tumorigenesis [17,18], the demonstration of an effect on tumor growth and phenotype would support FOXM1 as a novel actionable target for adjuvant treatment of these cancers. In the present work, we thus sought to clarify the role of FOXM1 in HNSCC cell lines using cell culture and tumor xenograft models.

Cell culture
The human CAL27 cell line was purchased from ATCC (Manassas, VA, USA; CRL-2095) and its authenticity was confirmed by STR profiling (Microsynth Gmbh, Balgach, Switzerland). Other cancer cell lines (LAU2083, LAU2088, and LOD148) and cancerassociated fibroblasts (CAF142) were established from tumor biopsies from patients undergoing surgery at the CHUV hospital. The procedures were all approved by the Cantonal Department of the Swiss Ethics Committees and the relevant informed consent forms were signed by the patients. All cells were cultured in a humidified atmosphere of 5% CO 2 at 37 ∘ C. A full list of the culture media used is provided in supplementary material, Supplementary materials and methods.

Inducible shRNA knockdown
Inducible FOXM1 knockdown in CAL27, LAU2083 or LOD148 cells was achieved using the pTRIPZ system (Dharmacon, Lafayette, CO, USA). Cells were transduced with a lentivirus carrying an shRNA sequence targeting FOXM1 (shFOXM1, CloneId V3THS_396939; Dharmacon) or a non-silencing shRNA control (shCtr, RHS4743; Dharmacon) and further selected by 1.5 μg/ml puromycin (540222-25; VWR International GmbH, Dietikon, Switzerland). After transient activation of the tetracycline response element by its transactivator (rtTA3) with 2 μg/ml doxycycline (Dox, D9891; Sigma-Aldrich Chemie GmbH, Buchs SG, Switzerland) for 2 days, cells were sorted by FACS to enrich for those expressing a high level of the pTRIPZ reporter (TurboRFP high ) and were further maintained without Dox. Prior to any in vitro experiment, cells were treated again for 5 days with 2 μg/ml Dox to induce FOXM1 downregulation and were further maintained in Dox-containing medium throughout the course of the experiment.
3D cell culture (organotypic culture, spheroids, and tumorspheres) Organotypic cell culture was performed as previously described [21] and is briefly outlined in supplementary material, Supplementary materials and methods. After 8-16 days of culture, the organotypes were collected either in 4% PFA/PBS for histological staining or in TRI reagent (T9424; Sigma-Aldrich Chemie GmbH) for RT-qPCR analysis. As a simplified version of the organotypic culture, cells can be embedded inside the collagen instead of on top of it, in which case cells will grow as spheroids. In brief, 200 000 tumor cells were admixed into a regular collagen I solution at a final concentration of 2.8 mg/ml before seeding them in one well of a poly-HEMA-coated 24-well plate. After 30 min at 37 ∘ C, regular culture medium was added on top of the collagen and cells were kept in tissue culture plates for 9 days to allow the development of spheroids. Spheroid-containing collagen gels were collected either in TRIzol for RNA analysis or with a PFA/Sucrose/OCT protocol for immunofluorescence staining. As a third alternative, cells were seeded in StemXVivo semi-solid medium (CCM012; R&D Systems Europe Ltd, Abingdon, UK) and cultured in poly-HEMA-coated 24-well plates in order to grow as tumorspheres. guidelines. HNSCC cells were implanted in the neck of NMRI-nu immunodeficient mice (Janvier Labs, Le Genest-Saint-Isle, France) as previously described [21]. Eight hundred thousand cells were resuspended in 30 μl of HBSS [14175053; Gibco (Life Technologies Europe BV, Zug, Switzerland)], mixed with 20 μl of BD Matrigel (356234; Becton Dickinson, Reinach, Switzerland), and subcutaneously injected along the submental midline of 8-week-old female NMRI-nu mice. Tumor volume was monitored using a Vernier caliper and calculated using the modified ellipsoid formula volume = 1/2 × [length × (width) 2 ]. Five days after implantation, FOXM1 downregulation was initiated by providing 1 mg/ml Dox in drinking water containing 2% sucrose [35580.02, Serva (Witec AG, Sursee, Switzerland)]. When tumors reached 150 mm 3 , they were dissected in pieces that were either collected by snap freezing in liquid nitrogen for downstream RNA extraction or fixed in 10% formalin (HT501128; Sigma-Aldrich Chemie GmbH) for histological evaluation.

RNA-seq and gene set enrichment analysis (GSEA)
TruSeq stranded RNA library preparation and paired-end 100 bp read sequencing on an Illumina HiSeq 2500 machine were performed at the genomics facility of the University of Lausanne. The ENA study accession number is PRJEB29013. Transcripts were quantified using Salmon (v.0.8.2) [22]. Given the xenogeneic nature of the samples analyzed, the quantification was performed using an artificial hybrid transcriptome generated by merging Homo sapiens GRCh38 and Mus musculus GRCm38 cDNA sequences to efficiently distinguish reads from human or mouse origin. Downstream analysis was performed in R (https://www.Rproject.org) and is described in supplementary material, Supplementary materials and methods.

Histology, immunohistochemistry, and immunofluorescence
Hematoxylin and eosin staining was performed on formalin-fixed, paraffin-embedded (FFPE) tumors or organotypes using the standard procedure. Immunohistochemistry staining was performed on FFPE samples; details of the antibodies are listed in supplementary material, Supplementary materials and methods. Involucrin immunofluorescence was performed on PFA-fixed spheroids. Sections were permeabilized for 10 min at room temperature with 0.1% Triton X-100 and then blocked for 30 min in PBS 0.5% BSA containing 1% donkey serum before incubation for 2 h at room temperature with a 1/200 antibody dilution in 0.5% BSA/PBS. After incubating for 1 h at room temperature with donkey anti-mouse-A488 (A-21202; Life Technologies Europe BV) antibody diluted 1/500 in 0.5% BSA/PBS, sections were mounted in ProLong Gold antifade reagent with DAPI (P-36931; Life Technologies Europe BV) and evaluated by fluorescence microscopy.

Statistics
The experimental results presented in the figures are generally representatives of at least three observations. Unless stated otherwise, the experimental data points in the figures represent the mean from triplicate experiments and the statistical significance of the data was determined using a two-tailed Student's t-test.
Details of the cell culture, 3D cell culture, RNA-seq and gene set enrichment analysis, chromatin immunoprecipitation, Alamar Blue assays, time-lapse cell tracking, double thymidine blocking, RNA isolation, RT-qPCR, image data analysis, immunohistochemistry, and western blotting may be found in supplementary material, Supplementary materials and methods.

Impact of FOXM1 knockdown on HNSCC cells in 2D and 3D cultures
To validate well-known in vitro activities of FOXM1 such as regulation of cell proliferation [23], migration, and cellular invasion [24], we generated stable doxycycline (Dox)-inducible shFOXM1-and shCtr-expressing CAL27, LAU2083, and LOD148 cell lines ( Figure 1A). FOXM1 knockdown using a validated shRNA sequence [14] induced only slight (CAL27 and LOD148) or no difference (LAU2083) in cell proliferation ( Figure 1B). Correspondingly, there were fewer cells in the S phase and slightly more cells in the G2 phase upon shRNA-mediated FOXM1 knockdown ( Figure 1C). Next, we monitored the expression of the cell cycle-related genes GTSE1 and CCNB2 (a known FOXM1 target) as read-out for the transcriptional activity of FOXM1. GTSE1 was identified amongst the top genes that correlated with FOXM1 expression in the TCGA HNSCC dataset (Spearman's rho 0.784), and we hypothesized that its promoter is a potential direct target of FOXM1. Using qChIP, we confirmed FOXM1 binding to the GTSE1 promoter region in CAL27 cells as well as to the CCNB2 promoter ( Figure 1D). However, GTSE1 expression in FOXM1-depleted cells was only slightly lower than in control cells ( Figure 1E), while the effect of FOXM1 downregulation on CCNB2 expression was more pronounced. Thus, we were unable to clearly confirm that FOXM1 regulates proliferation of HNSCC cells and expression of cell cycle-related genes in vitro.
There was also no major difference with respect to HGF-stimulated migration of control and FOXM1-depleted cells in a scratch wound healing assay (Figure 2A). Since HGF triggers transcription of the FOS gene [25], encoding an early-response transcription factor that stimulates expression of migrationand invasion-related genes such as PLAU [26] and PLAUR [27], we looked at the expression of FOS and its target genes by RT-qPCR in HGF-stimulated cells. The transcript level of FOS was strongly induced 1 h after HGF stimulation, although it did not show a significant difference between control and FOXM1-depleted cells ( Figure 2B). Three hours after HGF stimulation, the expression of PLAUR in shCtr was slightly upregulated when compared with shFOXM1 cells, demonstrating that HGF-induced FOS activity seems to decline upon FOXM1 knockdown to some extent. Surprisingly, neither PLAU transcript levels nor the abundance of its secreted product (uPA) changed upon FOXM1 knockdown in CAL27 cells (not shown).
To further evaluate the effect of FOXM1 knockdown on invasion, we used 3D organotypic culture (OT culture) and assessed whether cancer-associated fibroblasts (CAFs), known to secrete HGF [28], would impact the invasive phenotype of the organotypes' epithelial layer in a FOXM1-dependent manner. Indeed, mixing control cells with CAFs and seeding them on top of collagen I gel resulted in a more invasive phenotype when compared with OT cultures developed from shFOXM1 cells, as demonstrated by the quantification of budding or detaching cell groups ( Figure 2C,D).
Thus, our data demonstrate that shRNA-mediated FOXM1 knockdown in HNSCC cells did not significantly impact on their capacity to proliferate and migrate under conditions of 2D culture. However, FOXM1 depletion resulted in a significant inhibition of cellular invasion in organotypic co-cultures with fibroblasts.

FOXM1 knockdown reduces the number of proliferating cells in 3D OT cultures
Next, we reasoned that the conditions of OT culture are more physiological and hence more adequate to address the effect of FOXM1 knockdown. In OT cultures,

V Roh et al
both shCtr and shFOXM1 CAL27 cells plated on top of collagen I (this time without CAFs to allow RT-qPCR analysis of cancer cells only) formed multilayered structures that in part resembled squamous epithelium ( Figure 3A). Both control and shFOXM1 organotypes demonstrated widespread expression of keratins K14 (a marker of basal keratinocytes) and K8 (a marker of fetal epithelial differentiation in oral mucosa [29] that is aberrantly expressed in HNSCC [10]). In contrast, slightly stronger expression of involucrin (a marker of squamous differentiation) and cells positive for keratin K4 (a marker of mucosal differentiation) were observed in shFOXM1 organotypes, while K18 (another marker of fetal epithelial differentiation) was substantially downregulated in FOXM1-depleted organotypes ( Figure 3B). This phenotype was associated with upregulation of IVL transcripts and its upstream regulator JUND [30], as well as with a strong decrease in expression of the FOXM1 target genes CCNF [31] and GTSE1 in shFOXM1 cells ( Figure 3C). Although below the standard statistical significance threshold, a trend for upregulation of IVL upon FOXM1 knockdown could be confirmed in a large series of OT cultures ( Figure 3D). Importantly, FOXM1 expression already decreased in shCtr cells upon transfer from 2D to OT culture ( Figure 3E), indicating that FOXM1 downregulation is intrinsically associated with the activation of a differentiation program triggered upon 3D conditions of OT culture. In the 3D context and in contrast to the minimal effect of FOXM1 knockdown on proliferation observed in 2D culture, the association of FOXM1 with a proliferative state was confirmed by Ki-67 IHC staining of shCtr and shFOXM1 OT cultures. Although at day 8 we observed a comparable number of Ki-67-positive cells mostly enriched at the basal layer ( Figure 3F,G), at day 16, Ki-67-positive cells expanded throughout the control OT culture but not upon FOXM1 knockdown ( Figure 3F,G). Thus, in OT culture, the level of FOXM1 is decreased upon formation of a multilayered, epithelial-like tissue. Nonetheless, further reduction of FOXM1 expression by shRNA-mediated knockdown results in a decreased amount of proliferating cells mostly trapped in the basal layer and facilitates progression towards a more differentiated-like phenotype.

FOXM1 knockdown limits tumor growth and affects tumor phenotype in a xenograft model
To find out whether the effects of inducible FOXM1 knockdown observed in OT cultures could be reproduced in vivo, control and shFOXM1-expressing cells were tested in the submental mouse model of HNSCC. In this model, human HNSCC cells injected into the submental region of immunodeficient mice initiate tumors featured by squamous differentiation and heterogeneous expression of epithelial differentiation markers such as involucrin and keratins 8, 18, 14, and 4 (supplementary material, Figure S1). In line with the reduced number of proliferating cells seen in OT cultures, FOXM1 knockdown (induced from day 7 after implantation) significantly slowed down the growth kinetics of xenograft tumors ( Figure 4A). In addition, FOXM1 knockdown also affected tumor invasion ( Figure 4B). For each sagittal section through the tumor attached to the floor of the mouth (FOM), we selected a series of regions of interest (ROI) and examined the orientation of the tumor border relative to the horizontal axis in these regions ( Figure 4B, left panel). Upon FOXM1 knockdown, the tumor borders were significantly less invasive, as demonstrated by the quantified reduction in their dispersion factor (a measure of the curvature of the border) ( Figure 4B, right panel). In addition, the ability of FOXM1-depleted cells to form large cohesive cords was significantly reduced.
Histological analysis by H&E staining revealed accelerated squamous differentiation by FOXM1 knockdown, as evidenced by the formation of keratin pearls involving large areas ( Figure 4C,D). However, immunostaining-based quantification of slightly upregulated keratin 1 expression (a marker of terminal squamous differentiation) in shFOXM1 tumors did not reach statistical significance due to widespread expression of this marker in well-differentiated shCtr tumors ( Figure 4E,F). In shCtr tumors, multiple cells positive for FOXM1 staining were observed in the less differentiated areas ( Figure 4G,1) and sometimes located in a cell layer close to tumor stroma ( Figure 4G,2). Surprisingly, rare single cells ( Figure 4G,3) or patches of cells ( Figure 4G,4) showing nuclear expression of FOXM1 were detected in shFOXM1 tumors, indicating incomplete knockdown in a fraction of cells.
To analyze how FOXM1 knockdown globally affects the cancer cell transcriptome, 11 tumor samples were subjected to transcriptome sequencing (RNA-seq). As xenograft tumors include both human and mouse cells, reads were aligned using an artificially created human-mouse hybrid reference genome to ensure efficient and separate mapping of both human and mouse RNA sequencing reads. Owing to the different degrees of histologically evident squamous differentiation in shCtr samples and, respectively, to the varying level of FOXM1 expression and significant variations in FOXM1 knockdown efficiency in shFOXM1 samples, paired group analysis was not possible. Therefore, we used DEseq2 [32] to perform a correlative analysis of genes that were positively or negatively associated with FOXM1 expression and used the resulting Wald statistics value (supplementary material, Table S1) as a metric to define a ranked list of genes for downstream gene set enrichment analysis (GSEA). Figure 5A   was enriched in proliferation gene sets (e.g. G2-M cell cycle and EGFR signaling), while genes at the bottom of the ranked list were enriched in differentiation-related gene sets ( Figure 5C and supplementary material, Table S2). Furthermore, clustering analysis using a differentiation-related subset of genes including keratin-coding genes revealed differential expression in our RNA-seq samples (supplementary material, Figure S2) and in the TCGA HNSCC dataset (supplementary material, Figure S3), which can  likely be attributed to deregulated mechanisms of squamous-like differentiation. Interestingly, in the TCGA dataset, FOXM1 expression tends to be associated with cancers originating at the base of the tongue and tonsils including HPV-related cases. In addition, FOXM1 clustered with simple epithelial-type keratins (K7, K8, K18, and K19) and basal keratinocyte keratin K15, suggesting a possible association of FOXM1 with traits of the fetal differentiation program.
Altogether, by analyzing the effects of FOXM1 knockdown in the HNSCC xenograft model, we concluded that FOXM1 plays a key role in the tightly regulated balance of cancer cell proliferation versus differentiation.

Ectopic FOXM1 reduces expression of differentiation markers in HNSCC cell lines
Finally, we provide complementary evidence that ectopic FOXM1 overexpression is accompanied by a less differentiated phenotype in a series of four head and neck cancer cell lines ( Figure 6). Following immunoblotting in 2D cultures, we confirmed the downregulation of involucrin upon FOXM1 overexpression in LOD148, LAU2083, and LAU2088 cells, whereas keratin 1 was downregulated in LAU2083 and LAU2088 cells ( Figure 6A). To further validate the effect of ectopic FOXM1 on squamous-like differentiation, cells were cultured in 3D, either using collagen embedding or in semi-solid StemXVivo medium. LAU2083 and LAU2088 cells did not grow as spheroids; however, CAL27 and LOD148 grew as spheroids (in collagen I gel) or as aggregates of cells (in StemX-Vivo medium) and confirmed the downregulation of IVL expression following FOXM1 overexpression ( Figure 6B). Finally, immunostaining confirmed involucrin reduction upon FOXM1 overexpression in CAL27 collagen-embedded spheroids, as demonstrated by the reduced area of involucrin-positive cells within the spheroids ( Figure 6C).

Discussion
In this study, we demonstrated that the cell cyclepromoting function of FOXM1 is contextdependent in HNSCC and only evident under conditions of 3D culture and in tumor xenografts. Although mechanisms of FOXM1 action in the regulation of cell cycle in cancer cells have been well explored [4], its role in controlling their cell fate decisions is largely unknown, in particular in SCCs which are characterized by complex patterns of deregulated differentiation states. Our work provides novel insight into the function of FOXM1 in such tumors, demonstrating its supporting role in counterbalancing an aberrant squamous differentiation of HNSCC.
Mechanisms that govern normal differentiation towards stratified mucosal epithelia are impaired in FOXM1 limits aberrant differentiation in HNSCC 117 HNSCC due to genetic and epigenetic alterations. An aberrant squamous differentiation program is evident in HNSCC by intra-and inter-tumor heterogeneity in morphological appearance paralleled by variable expression of differentiation-specific proteins such as keratins [10,11]. In this respect, acquisition of the traits of fetal tissues (such as expression of simple epithelial-type keratins 8, 18, and 19) and developmental plasticity could be beneficial for growth and dissemination of cancer cells, as it may confer a survival advantage in a distinct microenvironment and upon anti-cancer treatment. Therefore, unraveling most critical pathways that trigger and control differentiation programs in HNSCC is urgently needed to discover actionable targets and to redirect the cancer cell phenotype towards a less aggressive state. Accordingly, targeted induction of terminal differentiation may reduce tumor invasion, as has been demonstrated by ectopic overexpression of S100A16, an EF-hand calcium binding protein of the S100 protein family, in HNSCC cell lines [33].
The loss of a malignant phenotype upon the induction of differentiation has been exploited in the past, in particular in hematological malignancies and coined differentiation therapy [34]. The most pertinent example is the treatment of acute promyelocytic leukemia (APL) with all-trans retinoic acid (ATRA), which results in a high rate of complete remissions [35]. Differentiation strategies to treat solid cancers are more difficult to realize because of their genetic heterogeneity and dependence on a complex combination of oncogenic pathways [34]. In HNSCC cells, retinoids are effective as anti-proliferative agents [36] that inhibit the growth of tumor xenografts [37]. Not surprisingly, the proliferation-inhibitory effect of ATRA and bexarotene (a pan-RXR agonist) is associated with FOXM1 downregulation [38]. How FOXM1 regulates the tightly coordinated balance between proliferation and differentiation of HNSCC cells and whether this mode of action depends on direct or indirect interactions with other transcriptional regulators of squamous differentiation such as p63 [39] or the ETS2/ELK3-AP1-SOX2-KLF5 network [40] remains to be elucidated.
It is intriguing to speculate that the retained ability to trigger a fetal differentiation program of oral mucosa and eventually acquire a reversible differentiation state may be advantageous to SCCs serving as an alternative to epithelial-to-mesenchymal transition (EMT) for survival. While proliferating cells are more sensitive to genotoxic drugs, differentiation may rescue a fraction of non-dividing, committed cancer stem cells (CSCs) from the treatment, especially if the process can be halted before the cells reach a state of permanent differentiation and reverse back to a proliferative state. In support of this idea, it has been reported that tumor cells regrowing after chemotherapy seem to originate from layers of cells directly surrounding hypoxic well-differentiated areas, which due to scarce vascularization are insufficiently accessible for cytotoxic drugs [41]. Also, we hypothesize that the survival of residual cancer cells, which remain after surgery in the tumor surrounding area [21], may depend on their aptitude to switch reversibly into a dormant state via incomplete squamous differentiation and resume proliferation upon favorable conditions. In this respect, the targeting of FOXM1 using specific inhibitors, such as RCM-1 [42], combined with a standard-of-care concurrent chemo-radiotherapy may potentially give an advantageous therapeutic effect.
In summary, FOXM1, being widely expressed in human cancers and in particular HNSCCs, emerges as an interesting actionable target. Before that, preclinical investigations will have to address the potential and mechanisms of differentiation-promoting and invasion-inhibiting therapy using FOXM1 inhibitors in HNSCCs.  Table S1. List of genes and their associated correlation factor with FOXM1 as calculated by DEseq2 Table S2. GSEA report for gene sets correlating with FOXM1 expression