The Stem Cell Factor HMGA2 Is Expressed in Non-HPV–Associated Head and Neck Squamous Cell Carcinoma and Predicts Patient Survival of Distinct Subsites Free

Head and neck squamous cell carcinomas (HNSCC) are epithelial tumors representing a heterogeneous disease entity, which encompasses a variety of tumors originating in the oral cavity, larynx, and pharynx with differences in epidemiology, etiology, and therapeutic approach. Over the past decades, HNSCC incidence rates increased in several countries in Eastern and Northern Europe and among females in Southern and Western Europe (1–3). Smoking, alcohol consumption, and poor oral hygiene, the most easily preventable cancer causes, are associated with an elevated HNSCC risk (4–6). HNSCC induced by human papilloma virus (HPV) via the oncoproteins E6/7 and HNSCC caused by other factors (such as smoking and alcohol consumption) are two separate entities, with distinct etiologies, clinical characteristics, prognoses, and a different epidemiology and molecular basis.

The molecular pathogenesis of HNSCC is not yet completely understood, a fact that impairs the improvement therapeutic approaches. Different genetic biomarkers have been proposed to identify patients who are at risk for recurrent tumors or aggressive disease progress. Most prominently, the cell-cycle proteins cyclin D1, p21, p53, and MDM2 have been described (7, 8). Other targets comprise growth factor receptors such as FGFR, VEGF, and EGFR (9–11).

The architectonic transcription factor high-mobility AT-hook 2 (HMGA2) is expressed during early embryogenesis and in cell differentiation (12–15). It has been attributed to the p53 cell-cycle pathway (16, 17). Targeted inhibition of HMGA2 induces apoptosis and chemosensitization in p53 mutant tumorspheres (18). In neuronal cells, HMGA2 was reported to regulate stemness (19). HMGA2 expression in tumor tissues has been detected in many cancer sites, that is, the digestive tract, lung, thyroid gland, urinary bladder, liver, testis, and pancreas (20–26). In HNSCC of the oral cavity, HMGA2 expression was associated with poor survival in a small cohort of 42 patients of Asian origin (27).

No systematic analysis of HMGA2 expression in a larger cohort stratified by HNSCC subsites and exposure to risk factors has been done so far. Therefore, the aim of this study was to investigate whether HMGA2 expression in tumor tissue predicts recurrence-free, overall, and tumor-specific survival in a larger group of German HNSCC patients. An understanding of these mechanisms may offer new strategies for cancer treatment.

As part of the European multicenter hospital-based case–control study ARCAGE (28, 29), a total of 287 cases were recruited in Germany from 2002 to 2005. The cancer sites comprised (ICD-10): oral cavity (C01-C06), tonsils (C09), pharynx (C10-C13), and larynx (C32). We had no patients with nasopharyngeal carcinomas (C11) in our study population. The patients were interviewed face-to-face about their sociodemographic characteristics, medical history, and life-style factors, including smoking and drinking history. Between 2011 and 2012, a mortality and morbidity follow-up was conducted. In the follow-up period (range: 6–10 years), all hospitalizations/new diagnoses (recurrence, secondary tumor) and ambulant examinations were recorded (pathologic/histologic records). The primary endpoint was time from enrollment in the ARCAGE study to all-cause death, occurrence of metastasis or relapse (tumor progression), or last follow-up. Events were determined by local health departments and medical practitioners, death certificates were confirmed through the respective public health department or the Bremen mortality index (BreMI). The BreMI is an electronic database providing all information recorded on death certificates of Bremen citizens who have died since January 1, 1998.

BreMI is based on the law of the Bremen Cancer Registry and follows the example of the National Death Index (NDI) in the United States (30). One participant emigrated and was censored at date of emigration. All other information concerning the survival or causes of death are complete.

Only patients for whom formalin-fixed, paraffin-embedded (FFPE) tumor tissue was available were included in the molecular biological analysis for the genetic biomarker HMGA2 (n = 202). All these samples were obtained for diagnostic purposes during the ARCAGE study (baseline). Our sample size allows the detection of a HR of 1.79 for patients with a high expression of HMGA2 in the tumor compared to those with low expression of HMGA2 at a statistical significance level of 5% with a power of 80%. We performed a sensitivity analysis that only included HPV negative tumors (n = 159) to check whether the observed associations were influenced by HPV status. Only one HPV+ case with a high HMGA2 expression level was detected. That is the reason we could not adjust for HPV status.

The vital status and the cause of death were determined for each participant. The vital status and current place of residence were determined by contacting the respective registration office. Death certificates were collected from the public health department or the local mortality index (30). Data on the clinical course of the tumor disease were collected in cooperation with the chief physicians of ear, nose, and throat (ENT) hospital departments or the patients' general practitioners. Data from paper files, histologic and pathologic findings and discharge letters, electronic hospital files, mortality index records, general practitioners' and public health department questionnaires were transferred to a computer-based standardized form by qualified study nurses (28).

The study endpoints were overall survival, progression-free survival, and tumor-specific survival. Overall survival time was defined as time the patient stayed alive from the date of HNSCC diagnosis to the end of the study or to death. Progression-free survival time was defined as time alive from HNSCC diagnosis without any local recurrence or newly diagnosed metastasis. Tumor-specific survival was defined as time alive from HNSCC diagnosis to the end of the study or to death which was not related to the primary tumor.

If no further visits to health departments or medical practitioners were documented (loss to follow-up), the last day of a documented visit to a health department/medical practitioner was the time point of censoring. TNM classification was performed according to the 2010 International Union against Cancer (UICC) guidelines (31).

Total RNA was isolated from the tissue using an RNeasy FFPE Kit (Qiagen) in a QIACube (Qiagen) and measured in triplicates, as previously described (16, 32). For calibration and HMGA2 quantification, HPRT was used as control gene, as has been recommended for head and neck tumors (33). The short length of both the amplicon of interest (HMGA2 61 Bp) and the control (HPRT1 81 Bp) are advantageous when FFPE tissue is examined.

HPV-DNA was detected using the primer system GP5+/6+ developed by de Roda Husman and colleagues (34). HPV-type specific PCR and primers were used as previously described [ref. 32; HPV16 (ATATAAGGGGTCGGTGGACCG, GCAATGTAGGTGTATCTCCATGC) and HPV18 (AAGGATGCTGCACCGGCTGAA, CACGCACACGCTTGGCAGGTTT)]. The PCR was performed with a PCR Core Kit^PLUS (Roche) according to the manufacturer's instructions. Prior to the PCR, the reaction mixture was incubated with 0.5 U uracil-DNA-glycosylase (UNG) for 5 minutes at 20°C followed by thermal inactivation of UNG for 2 minutes at 95°C. The PCR was performed with an initial denaturation of 30 s at 98°C, 35 repeats of 20 s denaturation at 98°C, 15 s annealing at 55°C, and 20 s elongation at 72°C, followed by a final elongation step for 3 minutes at 72°C. Unfortunately, p16 expression was not assessed.

Smoking behavior was quantified as pack-years [number of packs consumed (20 cigarettes) per day multiplied by the number of years the person had smoked]. Alcohol drinking behavior was quantified as drink-years (drinks per day multiplied by the number of years of alcohol consumption). The definition of one alcoholic drink equivalent was 18 mL of pure alcohol, corresponding to 330 mL of beer, 150 mL of wine, or 36 mL of hard liquor (35).

We used the 75th quantile of each tumor site to dichotomize the logarithmized and normalized expression values of HMGA2 into a low and high expression group, because on the one hand, we wanted to identify high-risk patients and thus choose a high cutoff, and on the other hand we wanted to obtain stable estimates. Because of the low number of cases, we could not use the 90th percentile. That is why we decided to use the upper quartile as a conventional cutoff for HMGA2 expression a priori, as has been done in previous biomarker studies (36, 37). A sensitivity analysis with other cutoffs (upper tertile, upper quintile) is now presented in Supplementary Table S1. Two-sided χ²- and t-tests were performed to compare the frequency of clinicopathologic parameters between patients with low and high expression levels of HMGA2. Differences in survival between groups were assessed by the Kaplan–Meier method and the log-rank test. Cox proportional hazards (PH) regression models were applied to investigate the effects of low versus high HMGA expression levels for the defined endpoints.

Stratification by tumor site (“oral cavity,” “tonsils,” “pharynx,” “larynx”) was used to detect possible differences between HNSCC of distinct subsites, because tumor sites are known to differ regarding biological behavior and HPV status. HPV infection status of the tumor and tumor stage are established predictors for HNSCC patient survival. Therefore, we excluded HPV+ tumors (HPV 16 and 18) and stratified by tumor stage (“1/2”, “3/4”) in a sensitivity analysis. To determine whether the prognostic levels of HMGA2 expression are independent of clinicopathologic parameters, multivariate Cox regression models were adjusted for sex, age (continuous), treatment (surgery, radiotheraphy, and chemotherapy), tobacco (pack-years; continuous), and alcohol consumption (drink-years; continuous).

We tested the PH assumptions by including time-by-covariate interactions in the multivariate Cox PH models by assuming a significance value of α = 0.05. The results indicated that tumor stage violated the PH assumption in all models. We also detected time-varying effects for the variable surgery in the subgroup analysis on tumor-specific survival. We therefore stratified all Cox models by tumor stage and where needed also by treatment (surgery). Stratification, however, led to a loss of power. This resulted in unstable effect estimates regarding HMGA2 and overall survival in the subgroup “oral cavity.” In general, P-values <0.05 were considered statistically significant and two-sided 95% confidence intervals (CI) were calculated. All statistical analyses were performed with SAS software (v9.3; SAS Institute Inc.).

Table 1 shows the distribution of patient characteristics by tumor site. Of a total of 202 HNSCC patients, 62 had tumors of the oral cavity (30.7%), tonsils (12.9%), pharynx (27.7%), or larynx (28.7%).

Fifty-five percent of HNSCC patients for whom complete information was available for TNM staging according to UICC (n = 153) were classified as TNM stage 3 and 4. This does not reflect the high number of locally limited tumors, because many cases lacked values referring to distant metastasis (M-stage). A total of 24 HPV-associated HNSCC (11.9%) were detected. The prevalence of HPV infection differed substantially with a maximum in tonsillar tumors (42.3%) and the minimum in laryngeal HNSCC (3.5%). Tobacco and alcohol consumption were highest among pharyngeal cancer patients (45.96 pack-years/98.96 drink-years).

HMGA2 expression was dichotomized into low and high, based on the 75th quantile of logarithmized expression values. Table 2 shows the relationship to expression status, clinicopathologic features, and patient characteristics. HMGA2 expression levels did not differ with regard to age, sex, tumor site, tumor stage, and smoking habits (Table 2). A statistically significant association between HMGA2 expression and negative HPV status was observed (P < 0.01). HMGA2 expression levels also differed by cumulated amount of alcohol drinking and N status, but these differences were not statistically significant.

Shorter overall survival was observed in patients with high HMGA2 expressing HNSCC, regardless of the subsite (P = 0.02; Fig. 1). The median survival times were 63.5 months in the low expression group and 21.4 months in the high expression group.

The stratification by T stages (Fig. 2) showed that patients with low HMGA2 expression and locally limited tumors had the best prognosis (P < 0.001). In the latter, the survival time was 82.8 months compared to 55.6 months in the group with locally limited tumors and high HMGA2 expression. The shortest overall survival was seen in patients with advanced tumors and high HMGA2 expression (13 months).

Subsite-specific Kaplan–Meier curves revealed a significant survival benefit for patients with HNSCC of the oral cavity and the larynx compared to patients with tumors in other subsites (Fig. 3).

Table 3 summarizes the results of the Cox PH regression analysis. The 1.5-fold risk increase in the total of 202 patients with HMGA2+ HNSCC was observed without adjustment for confounders. The analysis stratified by tumor site showed that the associations varied between localizations, corroborating the Kaplan–Meier results (Fig. 3). In the Cox regression model stratified by tumor stage and fully adjusted for age, sex, tumor site, treatment, tobacco, and alcohol consumption (Table 3), an up to four-fold increased risk for the adverse outcome (endpoint overall survival) was seen in patients with high HMGA2 expressing HNSCC of the larynx (HR = 4.00; 95% CI, 1.18–13.62). The risk for reduced overall survival was also increased for tumors of the oral cavity (HR = 1.99; 95% CI, 0.92–4.29), but not statistically significant. The sensitivity analysis with other cutoffs (upper tertile, upper quintile) for high and low HMGA2 expression levels is presented in Supplementary Table S1. The resulting effect estimates varied but did not change their direction.

Patients with HNSCC of the oral cavity and the larynx showed a risk of earlier death due to the tumor (tumor-specific survival), in accordance with the overall survival results (Table 3). Recurrence-free survival was also negatively associated with HMGA2 expression status (progression-free survival), although the association did not reach statistical significance. On the contrary, in patients with pharyngeal HNSCC, an inverse association between HMGA2 expression and the risk of earlier death was detected (HR = 0.39; CI 95%, 0.17–0.89). A sensitivity analysis excluding HPV+ tumors confirmed the increased risk of earlier death in patients with high HMGA2 expression. Effect estimates show only minor changes after removal of HPV+ participants (Supplementary Table S2).

In this follow-up study of 202 incident HNSCC cases, the stem cell factor HMGA2, a p53 modulating candidate molecular biomarker of tumor aggressiveness, was negatively associated with survival of patients with HNSCC of the larynx.

For many different cancer types, previous studies demonstrated that higher HMGA2 expression, possibly through p53 mutagenesis effects, is associated with a more aggressive tumor type, for example, gastric cancer (38, 39), colon cancer (40), pancreatic cancer (25), and carcinomas of the oral cavity (27, 41). A study by Miyazawa and colleagues (27) showed that HMGA2 expression was associated with poor survival in 42 patients with HNSCC of the oral cavity. In that study, the 5-year survival rate was 100% among patients who did not show high expression of HMGA2.

Chang and colleagues (42) suggested that HMGA2 could be a potential prognostic tissue biomarker for HNSCC of the oral cavity. Their analysis indicated that HMGA2 expression was an independent predictor of overall survival, disease-specific survival, and disease-free survival (42). The multivariate analysis was adjusted for age, sex, overall stage, and perineural invasion.

In our study of 202 HNSCC patients involving four different subsites, we confirmed that HMGA2 predicts overall survival in HNSCC of the larynx independent of covariates like age, sex, treatment, alcohol, and tobacco consumption. Interestingly, HMGA2 was predominantly expressed in non-HPV–associated HNSCC, which supports the notion that HPV- and non-HPV–associated tumorigenesis differ dramatically from each other. HNSCC is a heterogeneous group of tumors with different clinical characteristics, which may therefore have distinct pathways of tumorigenesis. Biological heterogeneity in HNSCC has been suggested by molecular analyses that have uncovered distinct classes of HNSCC with unique mRNA expression patterns or differences in DNA copy number alterations. It may be speculated that topical noxae like alcohol and tobacco smoke have a more direct effect on the mucosal tissue of the oral cavity. The elevated cell turnover caused by topical noxae might explain why a stem cell gene like HMGA2 is upregulated. In fact, HMGA2 expression could identify a subgroup of HNSCC, characterized by a stem cell phenotype of tumor cells with higher cell turnover.

As demonstrated by T-stage-specific Kaplan–Meier curves with and without HMGA2 expression, the HMGA2 phenotype can already be diagnosed in early tumor stages, indicating a strong impact on postoperative patient management. HMGA2 is independently associated with tumor-specific survival in HNSCC of the oral cavity and the larynx. The 10-fold increased risk of tumor-specific death in patients with laryngeal HNSCC has to be interpreted cautiously considering the wide CI and due to the small number of cases in this group. As has been reported, patients with HPV+ HNSCC (which are 42% of the tonsillar HNSCC) are known to have a longer survival per se (43), which could explain why HMGA2 is of limited prognostic value in patients with tonsillar HNSCC. Furthermore, the small sample size of tonsillar HNSCC resulted in unstable effect estimates in the multivariate analysis. The reason for the inverse, i.e., protective, effect of HMGA2 expression resulting in longer recurrence-free survival in patients with pharyngeal HNSCC remains unclear, although we cannot rule out a random finding. Otherwise, the result implies that HNSCC of the pharynx differ in terms of tumorigenesis pathways from other HNSCC subsites, regardless of their epithelial origin. Finally, it is worthy of note that the classical clinicopathologic factors (TNM stage, local extent of the tumor disease, lymph node involvement, and surgical resection of the tumor) in patients with HNSCC are still the most useful independent prognostic factors. A limitation of this study is that the stratified analysis yielded small sample sizes and data on TNM stage were missing in 25% of all cases. Only FFPE tissues were available. RNA isolated from archived material can be degraded and formalin fixation can cause chemical modification in the RNA, like cross-linking to proteins and addition of monomethylol groups. Therefore, qRT-RCR performed on FFPE tissue consistently obtains lower CT values than for fresh frozen tissues with the same input RNA (44). Nevertheless, experiments comparing qRT-PCR results with fresh frozen tissue and FFPE tissue suggest that mRNA expression levels derived from FFPE tissue reflect the actual expression levels in the original tissue samples, regardless of the variable effects of fixation (45, 46). In addition, we could not investigate the effect of tumor stage on survival in multivariate models because the PH assumptions were not fulfilled for tumor stage.

In contrast to previous studies, we did not use immunohistochemistry to detect HMGA2 (over-)expression, because sensitivity and specificity of HMGA2 immunohistochemistry have been an issue, for example, in mesenchymal tumors (47). Because our study employed quantifications of HMGA2 expression relative to the nonregulated house-keeping gene HPRT, cutoff points were defined by data-driven quartiles to discriminate between low and high HMGA2 expression levels. The major limitation of our study is the small sample size for some variables leading to insufficient power to detect smaller associations. Despite small the sample size, the findings are relatively consistent across clinicopathologic parameters. Furthermore, the importance of our findings in providing new insight into potential preventive and treatment strategies for HPV− HNSCCs outweighs the limitations.

Our findings indicate that HMGA2 plays a major role in non-HPV-associated HNSCC, especially of the larynx. The extensive adjustments for confounders in the Cox regression model we used support the validity of our findings. The role of HMGA2 in stem cell renewal facilitates sustained proliferative signaling in cancer cells and may identify a stem cell phenotype within the heterogeneous group of HNSCC. The previously reported antitumorigenic effects achieved through targeted HMGA2 inhibition (18), could offer new strategies for cancer treatment in HMGA2 expressing early-stage HNSCC.

No potential conflicts of interest were disclosed.

We obtained written informed consent from all patients. The study was approved by the ethics committee of the medical association of Bremen.

The funders played no role in the design or conduction of the study, collection, management, analysis or interpretation of the data, preparation, review, or approval of the manuscript, or the decision to submit the manuscript for publication.

Conception and design: K. Günther, J. Bullerdiek, R. Nimzyk, T. Behrens, W. Ahrens

Development of methodology: K. Günther, J. Bullerdiek, R. Nimzyk, T. Behrens

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K. Günther, R. Nimzyk, D.N. Markowski, T. Behrens

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K. Günther, R. Foraita, J. Friemel, F. Günther, J. Bullerdiek, D.N. Markowski

Writing, review, and/or revision of the manuscript: K. Günther, R. Foraita, J. Friemel, F. Günther, R. Nimzyk, T. Behrens, W. Ahrens

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K. Günther, J. Bullerdiek

Study supervision: K. Günther, J. Bullerdiek, T. Behrens, W. Ahrens

The authors would like to thank the patients and their families for their participation and gratefully acknowledge the excellent technical assistance of Beate Schütte, Anja Bergmann and Marina Resnikov (Leibniz Institute for Prevention Research and Epidemiology, BIPS), and Sabrina Dorschner (Center for Human Genetics). We are also grateful to our clinical colleagues in hospitals and primary care who supported this study. We also thank the team of the BreMI and Carola Lehmann for her technical support.

This project was funded by the German Research Foundation [DFG research grant AH 82/6-1 (to W. Ahrens) and BU 592/11-1 (to J. Bullerdiek)] and by the German Cancer Aid [No. 110276 (to W. Ahrens, J. Bullerdiek)].

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.