NOVA1 induction by inflammation and NOVA1 suppression by epigenetic regulation in head and neck squamous cell carcinoma

Neuro-oncological ventral antigen 1 (NOVA1) is known as a neuron-specific pre-mRNA binding splicing factor. Previously, it was shown to be highly upregulated in T lymphocytes, as well as fibroblasts/stromal spindle cells, in tertiary lymphoid tissues formed by the benign immune-inflammatory process, while it was frequently downregulated in tumor cells and other cells within the tumor microenvironment. Here, we sought to identify the mechanisms of NOVA1 modulation in head and neck squamous cell carcinoma (HNSCC). NOVA1 was induced by inflammatory-immune signals within the tumor microenvironment and was suppressed by epigenetic dysregulation, such as that with miR-146. We found attenuated expression of NOVA1 to be associated with non-oropharynx sites such as oral cavity, hypopharynx, and larynx, human papilloma virus (HPV)-negative SCC defined by immunohistochemistry for p16INK4a expression, fewer tumor infiltrating lymphocytes, and poor patient outcomes. Moreover, changes were discovered in epithelial mesenchymal transition-associated markers according to NOVA1 status. This study provides some insights to the underlying mechanism of NOVA1 regulation and suggests that NOVA1 may serve as a prognostic biomarker and potential therapeutic target for HNSCC in the future.

According to comprehensive genetic analyses using The Cancer Genome Atlas Network (TCGA, https://cancergenome.nih.gov) and the Catalogue Of Somatic Mutations in Cancer (COSMIC, http://cancer.sanger.ac.uk/ cosmic), genetic mutations and epigenetic hypermethylation eliciting NOVA1 dysregulation appear to be rare in most cancers. Despite marked variations in expression levels of NOVA1 in HNSCC, recent comprehensive studies have revealed that genetic alteration of NOVA1 is very rare, at a frequency of about 2% ( Supplementary  Fig. S2). From this and our previous study 11 , we conjectured that epigenetic regulation, specifically with microR-NAs (miRNA), may be involved in the dysregulation of NOVA1 in HNSCC.
In the present study, we sought to determine whether NOVA1 is induced by inflammatory signals and epigenetically suppressed within the tumor microenvironment in HNSCC.
NOVA1 expression varied in tumor cells and lymphocytes and stromal spindle cells from the tissue microenvironment of 396 HNSCC specimens. Attenuated NOVA1 expression, which showed an H-score < 200 and indicated a loss of NOVA1 in more than one-third of the cell population, was more frequent in tumor cells than in lymphocytes and stromal spindle cells ( Fig. 3C and Supplementary Fig. S6). Generally, NOVA1 expression status in tumor cells was correlated with that of T lymphocytes and stromal spindle cells (tumor cells and T lymphocytes, r = 0.769, p < 0.001; tumor cells and stromal spindle cells, r = 0.753, p < 0.001; Fig. 3D). NOVA1 expression in association with anatomical sites, HPV infection status, T cell infiltration, epithelial mesenchymal transition markers, and patient outcomes. Attenuated expression (H-score < 200) of NOVA1 in tumor cells was frequent in non-oropharynx SCC such as SCC arising in oral cavity, hypopharynx, larynx, and p16 (HPV)-negative SCC (p < 0.001, respectively; Fig. 4A,B), and it was most frequently observed in p16-negative non-oropharynx SCC, when comparing oropharynx SCC and/or HPV-positive cases (p < 0.001; Fig. 4C). Regarding TILs, which are observed within tumor cell nests, specimens showing attenuated NOVA1 expression in tumor cells showed lower cell densities of CD3+ or CD8+ TILs than those showing strong NOVA1 expression (Fig. 4D). A marker of epithelial mesenchymal transition (EMT), Twist expression in tumor cells was related to NOVA1 expression status in tumor cells, but not to NOVA1 status in stromal spindle cells/fibroblasts or T lymphocytes ( Fig. 4E; Supplementary Fig. S7). SNAI1/SLUG expression (another marker of EMT) in tumor cells was related to NOVA1 expression status in T lymphocytes and stromal spindle cells/fibroblasts, but not to NOVA1 status in tumor cells ( Fig. 4F; Supplementary Fig. S7).
Regarding associations with patient outcomes, attenuated NOVA1 expression in tumor cells was associated with inferior OS (Fig. 4G, p = 0.018) and PFS in Kaplan-Meier analysis (Fig. 4H, p = 0.030). In multivariate Cox regression analysis, attenuated NOVA1 expression in tumor cells was identified as an independent poor prognostic factor for OS (Hazard ratio = 2.104, p = 0.005) and PFS (Hazard ratio = 1.599, p = 0.03), as were other clinicopathologic factors, such as older age, non-oropharynx site, advanced pathologic T stage (pT stage 3-4), advanced pathologic N stage (pN stage 2-3), and lower cell density of tumor infiltrating CD8 + cytotoxic T cells (Supplementary Tables S5 and S6).

Microenvironment Cell Populations-counter analysis. Microenvironment Cell Populations-counter
(MCP-counter) analysis using a gene set for HNSCC generated from TCGA (https://cancergenome.nih.gov, n = 348) indicated that high NOVA1 expression is significantly correlated with greater abundance of immune cells, including T cells, CD8+ T cells, cytotoxic lymphocytes, NK cells, B cells, monocytes, and myeloid dendritic cells, as well as stromal cells of fibroblasts and endothelial cells. Upregulation of the CD8+ T cell-related genes CD8A and CD8B was significantly related to high NOVA1 expression (  Table S8). In summary, lower abundances of immune and stromal cells, downregulation of CD8+ T cell-related genes, downregulation of TWIST and SNAI1, and upregulation of SNAI2 and TGFB1 were all found to be related to low NOVA1 expression (   www.nature.com/scientificreports www.nature.com/scientificreports/ microenvironments (palatine tonsils and base of the tongue) and is primarily associated with HPV infection. Non-oropharynx SCC such as SCC of oral cavity, hypopharynx, and larynx, however, arises from an immune cell-poor tissue microenvironment and is generally unrelated to HPV infection. Although the lymphoid and immune cell structures of the oropharynx are physiologically formed as secondary lymphoid structures, inflammatory stimuli in response to HPV infection are thought to induce NOVA1 expression in tumor cells and the surrounding microenvironment. Nonetheless, in non-oropharynx SCC, NOVA1 could still be induced in tumor cells and the surrounding microenvironment upon formation of tertiary lymphoid structures within the tumor microenvironment as an immune response to tumor growth [12][13][14][15][16] . As expected, in the present study, most oropharynx SCCs showed strong NOVA1 expression in tumor cells, while non-oropharynx SCC frequently showed attenuated expression. Likewise, HPV (p16 immunohistochemistry)-positive SCC mostly exhibited strong NOVA1 expression, while HPV-negative SCC frequently showed attenuated expression in tumor cells. Such high NOVA1 expression in oropharynx SCC may be related to an abundance of lymphoid structures within the tissue microenvironment.
As stated above, most oropharynx SCCs show HPV positivity. Thus, immune reactions within the lymphoid structures, which are activated by HPV infection, might stimulate NOVA1 induction. HPV type 16 is the most prevalently detected virus type in HPV-mediated oropharynx SCC, accounting for about 80% of all HPV-positive SCC cases, and HPV E6 and E7 are the most potent oncogenes affecting the tumorigenesis of HPV-mediated tumors 17 . In the present study, transfection of HPV 16 E6/E7 genes to originally HPV-negative SCC cells did not significantly induce NOVA1 in tumor cells. Accordingly, we deemed that HPV infection, triggering immune reactions within the surrounding microenvironment, is only indirectly involved in NOVA1 induction.
As a limitation of our study, a single tumor cell line could not fully represent actual tissue conditions, comprising activated inflammatory and lymphoid structures formed within the tissue microenvironment. To address this limitation, we sought to simulate immune signals by treating cells with poly(dA: dT). Transfected poly(dA: dT) is recognized by several cytosolic DNA sensors, activates immune responses, and triggers the formation of the inflammasome [18][19][20][21] . Indeed, by mimicking inflammatory signals, poly(dA:dT) upregulated NOVA expression in the present study. In line with this result, we noted higher infiltrating cell densities of CD3+ T lymphocytes and CD8+ cytotoxic T lymphocytes in HNSCC specimens showing strong NOVA1 expression among tumor cells, compared to specimens showing attenuated NOVA1 expression. MCP-counter analysis using gene sets for HNSCC from TCGA also indicated that high NOVA1 expression may be related to increases in immune cells, www.nature.com/scientificreports www.nature.com/scientificreports/ such as CD8+ T cells, and stromal cells, such as fibroblasts. Overall, these findings support our hypothesis that inflammatory and immune signals within the microenvironment may induce NOVA1 upregulation.
Despite our meaningful results, the biological significance of NOVA1 upregulation in cancer remains uncertain. To identify possible roles for NOVA beyond its known role as a neuron-specific pre-mRNA binding splicing factor, we focused on EMT, since NOVA1 upregulation seems to occur by interacting with microenvironment inflammatory signals and since NOVA1 is known to be enriched in normal fibroblasts 5 . In HNSCC tissue samples, we found Twist and SNAI1 expression (EMT markers) to be related to NOVA1 status in tumor cells, as well as stromal spindle cells and T lymphocytes. Moreover, MCP-counter analysis using the gene set of HNSCC from TCGA also indicated that NOVA1 expression is related to upregulation or downregulation of EMT signature genes, including SNAI1, TWIST, SNAI2, and TGFB1. Although further studies of direct signal pathways are needed, these findings suggest that changes in NOVA1 expression in tumor cells or the tissue microenvironment may affect EMT and that NOVA1 may act as a signaling factor between tumor cells and interacting stromal cells.
Regarding the clinical implications of NOVA1 in HNSCC, we discovered that attenuated NOVA1 expression in tumor cells is related to poor patient prognosis via univariate and multivariate survival analysis. This suggests that NOVA1 suppression might be related to an aggressive cancer cell phenotype. As a mechanism of NOVA1 suppression, we suspected that epigenetic regulations could affect genetically normal lymphocytes and stromal spindle cells. Several studies have found that altered miRNA expression likely occurs in the tumor microenvironment as the result of crosstalk among stromal spindle cells and immune cells through various signaling autocrine and/or paracrine methods 22 . In HNSCC tissue samples, we noted that variable expression of hsa-miR-146a-5p and miR-146b-5p, which are predicted to target NOVA1, was positively correlated between tumor and stromal areas: the expression levels of individual miRs were relatively higher in tumor areas than in stromal areas, suggesting that a major source of upregulated miRs may be tumor cells. To confirm NOVA1 inhibition by miRs in vitro, we experimented on FaDu cells, because this cell line shows higher expression of miR-146 and NOVA1. Because the sequences of miR-146a-5p and miR-146b-5p were very similar and their predicted target sites of NOVA1 were identical (-AGUUCUC-: Supplementary Fig. S9), transfection experiments were performed with miR-146b mimic and anti-miR-146b inhibitors. In doing so, we found that miR-146 (miR-146a and miR-146b) could suppress NOVA1 in tumor cells, even in the presence of inflammatory signals simulated by poly(dA:dT) transfection. When simultaneously treating cells with poly(dA:dT) and blocking miR-146 with anti-miR-146b-5p and anti-miR-146b-3p, NOVA1 expression was increased. From these findings, we discerned that miR-146 may be potent suppressors of NOVA1, even in the presence of inflammatory signals. These results could reflect a mechanism by which NOVA1 is epigenetically attenuated in tumor cells, even when immune-inflammatory lymphoid structures are present, within the tumor microenvironment.
In the process of cancer development and progression, HNSCCs adapt several mechanisms of immune escape, such as T cell tolerance, inhibition of inflammatory cytokines, and exhaustion of effective cytotoxic T cell function within the tumor microenvironment [23][24][25] . In this study, we found that NOVA1 is induced by inflammatory-immune signals within the tissue microenvironment and is suppressed by epigenetic dysregulation, potentially by miR-146. Although epigenetic suppression of NOVA1 and impaired immune surveillance within the tissue microenvironment according to tumor progression should be investigated further, the present study provides some insights into the underlying mechanism of NOVA1 regulation and raises the translational potential of NOVA1 as a prognostic biomarker and therapeutic target for HNSCC in the future.
Clinical samples. From the database of Severance Hospital Cancer Registry Data, Seoul, South Korea, cases of HNSCC were retrieved. Affected anatomical sites included the oropharynx (tonsils, base of tongue, soft palate, and oropharynx) and non-oropharynx. Non-oropharyngeal sites included oral cavity (anterior two-thirds of the tongue, mouth floor, hard palate, buccal mucosa, and unspecified oral cavity), hypopharynx, larynx (lingual surface of the epiglottis, glottis, supraglottis, subglottis, and larynx), and nasal cavity/paranasal sinuses. Carcinomas of nasopharynx, ear cavity, salivary gland, or other anatomical sites were excluded from the present study. Formalin-fixed paraffin-embedded tissue (FFPE) specimens from 116 consecutive oropharyngeal SCC and 280 consecutive non-oropharyngeal SCC patients were included: these patients underwent surgical resection with a curative aim from 2005 to 2012 at Severance Hospital Seoul, South Korea. Specimens that underwent decalcification were excluded for accurate immunohistochemical analysis. Part of the present cohort had been included in our previous studies [26][27][28][29][30] . Information on various clinical factors, including age at operation, sex, smoking, and alcohol consumption, was obtained by reviewing medical records. Pathologic factors, including lymphovascular invasion, perineural invasion, pathologic TNM staging according to the 7 th American Joint Committee on Cancer, and tumor classification by the World Health Organization system 31,32 , were obtained from slide review by three individual pathologists (E.K. Kim, Y.A. Cho, and S.O. Yoon). The median follow-up period was 50 months (range from 1 to 110 months). Clinicopathologic characteristics are described in Supplementary Table S1. Also, using the TransIT-X2 ® Dynamic Delivery System (MIR 6000; Cambridge Bioscience, UK), we transfected repetitive synthetic double-stranded DNA sequences of poly(dA-dT)•poly(dT-dA) (Sigma-Aldrich, MO, USA) into each cell line at a dose of 0, 10, 100, 1000, or 10000 ng. RNA was isolated 24 and 48 hours thereafter. All experiments were performed independently three or more times.

Quantification of miRNA in clinical samples of HNSCC. The expression patterns of candidate miR-
NAs in the microenvironment were analyzed in 50 FFPE specimens of HNSCC that were randomly selected and showed variable NOVA1 protein expression in tumor cells and stromal cells. Information on NOVA1 protein expression status in these 50 specimens is summarized in Supplementary Table S3. Tumor areas and surrounding stromal areas were separately dissected under a microscope. RNA, including microRNA, was isolated using miRNeasy FFPE Kits (Qiagen, Hilden, Germany). cDNA was synthesized using looped reverse transcription primers specific to individual miRNA species. Looped reverse transcription and forward and reverse primers for each microRNA species (hsa-miR-146a-5p, -146b-5p, -181a, -181b, -181c, -181d, -27a, and -27b, as well as a housekeeping gene, U6sn) are described in Supplementary Table S4 Immunohistochemistry and interpretation. Tissue microarray analysis was conducted by selecting two or three different tumor areas from 396 HNSCC samples. Core specimens 3 mm in diameter were obtained from donor tissue blocks and arranged in recipient TMA blocks using a trephine apparatus. Immunohistochemistry Expression of NOVA1 was scored respectively in tumor cells, stromal spindle cells (fibroblasts, support cells, and endothelial cells), and immune cells (T lymphocytes) (Fig. 6). Nuclear expression of NOVA1, Twist, and SNAI1/SLUG were analyzed according to the semi-quantitative H-score method: this method yields a total score range of 0 to 300 33 , which is obtained by multiplying the dominant nuclear staining intensity score (0, no staining; 1, weak or barely detectable nuclear staining; 2, distinct brown nuclear staining; 3, strong dark brown nuclear staining) by the percentage (0-100%) of positive cells. Densities of tumor infiltrating T lymphocytes (TILs) were semiquantitatively scored as follows 28,34 : the five most representative sections at high-power magnification (x400) were selected. Intact lymphocytes expressing CD3 and CD8 were counted manually, and the numbers of counted cells were averaged ( Supplementary Fig. S10A,B). Conventionally accepted criteria for HPV-positivity were used for p16 IHC, and positivity was defined as the presence of strong and diffuse nuclear and cytoplasmic staining in >70% of the HNSCC cells. All other staining patterns were scored as negative ( Supplementary Fig. S10C) 35 .
Gene expression profile analysis. Gene expression profiles with normalized RSEM values from the TCGA-HNSC cohort (https://cancergenome.nih.gov, n = 348) were downloaded using the TCGAbiolinks R package and were analyzed. Samples were classified into upper one-third and lower one-third sample groups according to expression levels of NOVA1. Immune and stromal cell abundance in individual specimens was calculated by the Microenvironment Cell Populations-counter (MCP-counter) R package 36 , and the Wilcoxon