Nuclear Factor-κB–Related Serum Factors as Longitudinal Biomarkers of Response and Survival in Advanced Oropharyngeal Carcinoma

Purpose: Cytokines and growth factors modulated by transcription factor nuclear factor-κB and secreted by tumor and stromal cells are detectable in serum of patients with advanced cancers, including head and neck squamous cell carcinomas (SCC). Longitudinal changes in these serum factors could be early biomarkers of treatment response and survival.

Experimental Design: Interleukin (IL)-6, IL-8, growth-related oncogene-1 (GRO-1), vascular endothelial growth factor (VEGF), and hepatocyte growth factor (HGF) concentrations were determined by Luminex multiplex assay using serum obtained at baseline and every 3 months in a prospective study of 30 patients with locally advanced (stage III/IV) oropharyngeal SCC receiving chemoradiation therapy. The relationship between baseline and direction of change in individual and multiple cytokines with cause-specific and disease-free survival was determined by Cox proportional hazards models and Kaplan-Meier survival analysis. Statistical analyses included adjustment for smoking status and response to chemoradiation.

Results: Three-year cause-specific and disease-free survival was 74.4% and 68.9%. Nonsmoking history (P = 0.05) and higher baseline VEGF (P = 0.003) correlated with increased survival. Longitudinal increases in levels of individual factors predicted decreased cause-specific survival when adjusted for smoking history [IL-6: relative risk (RR), 3.8; 95% confidence interval (95% CI), 2.0-7.4; P = 0.004; IL-8: RR, 1.6; 95% CI, 1.2-2.2; P = 0.05; VEGF: RR, 3.0; 95% CI, 1.6-5.6; P = 0.01; HGF: RR, 2.9; 95% CI, 1.9-4.4; P = 0.02; and GRO-1: RR, 1.2; 95% CI, 1.1-1.3; P = 0.02]. For a given individual, large increases in the upper quartile for any three or more factors predicted poorer cause-specific survival compared with patients with two or fewer large increases in factor levels (P = 0.004).

Conclusions: Pretreatment VEGF levels and longitudinal change in IL-6, IL-8, VEGF, HGF, and GRO-1 may be useful as biomarkers for response and survival in patients with locally advanced oropharyngeal and head and neck SCC treated with chemoradiation.

Malignant and stromal cells have been shown to secrete multiple factors and cytokines that play a critical role in promotion of cell growth, inflammation, angiogenesis, and other complex interactions within the tumor microenvironment (1–3). Cytokine “footprints” or “signatures” may be characteristic of different histologic types and subsets of cancer and transcriptional regulatory mechanisms (3, 4). The presence of increased levels of one or more of these factors in serum of patients with advanced cancer makes them potential candidates as serum biomarkers for response, relapse, and survival.

In head and neck squamous cell carcinoma (HNSCC) cell lines and/or tumor specimens, we previously detected increased expression of a repertoire of cytokines and growth factors, including interleukin (IL)-1, IL-6, IL8, growth-related oncogene-1 (GRO-1), granulocyte macrophage colony-stimulating factor, vascular endothelial growth factor (VEGF), and hepatocyte growth factor (HGF; refs. 5–7). Among these, increased concentrations of IL-6, IL-8, VEGF, HGF, and GRO-1 are detected in sera of patients with SCC when compared with control subjects (5, 7–12). Evidence indicates that these factors functionally contribute to the phenotype and pathogenicity of SCC and other cancers (2, 3, 7, 10, 11, 13–17). Thus, the elevated concentration and functional importance makes them important candidates for evaluation as individual serum markers for HNSCC.

The frequent coexpression and elevation of several of these factors in serum suggested they could be coregulated by common underlying mechanism(s) and together serve as a biomarker panel. We showed that the coexpression of several of these cytokines and factors can be modulated, directly or indirectly, by aberrant activation of transcription factor nuclear factor-κB (NF-κB), in cooperation with other transcription factors in HNSCC (18–20). Genetic or pharmacologic inhibition of NF-κB inhibited expression of IL-1α, IL-6, IL-8, GRO-1, and/or VEGF (19–21) by HNSCC cells and tumorigenesis of xenografts in immunodeficient mice (19, 21). HGF was shown to be secreted by tumor-associated stromal fibroblasts in response to NF-κB–regulated cytokines IL-1α and IL-6 produced by HNSCC cells in vitro.⁴

NF-κB and proinflammatory cytokines may also be inducibly activated in epithelial cells by exposure to carcinogens in tobacco products (22, 23), indicating that smoking may contribute to coexpression of these factors during and after development of cancer.

Biomarkers that reflect functionally important molecular alterations are needed for patients with HNSCC and other cancers. Most patients with HNSCC present at advanced stage and many suffer significant morbidity and poor survival rates despite advances in multimodality and organ preservation therapies (24–26). Tumor, nodal staging, and clinical response in patients with such advanced disease do not adequately predict survival, particularly in individual patients. We previously observed an increase or decrease in concentration of IL-6, IL-8, VEGF, HGF, and GRO-1 in posttreatment serum corresponding to tumor progression or reduction in patients treated with chemotherapy and radiation (8). These observations suggested that longitudinal monitoring of one or a panel of these NF-κB–related factors could potentially serve as serum markers of response, relapse, and survival.

In the present prospective study, we determined concentrations of NF-κB–modulated factors IL-6, IL-8, GRO-1, VEGF, and HGF in baseline and posttreatment serum samples collected at 3-month intervals during the first year from a group of 30 patients with local-regionally advanced oropharyngeal SCC, who were treated with chemotherapy and radiation. The potential relationship of tumor stage, smoking status, and pretreatment and posttreatment longitudinal changes in cytokines with cause-specific and disease-free survival was evaluated.

Study subjects. Patients were participants in the University of Michigan Head and Neck Cancer Specialized Program of Research Excellence trial. Patients were recruited at the University of Michigan Medical Center, Veteran's Administration Hospital (Ann Arbor, MI) or Henry Ford Medical Center (Detroit, MI) from 2003 to 2005. Human subject approval was obtained from the Institutional Review Boards at all three study sites, and all subjects gave written informed consent. Patients were enrolled at diagnosis and followed every 3 months. Epidemiologic survey data, history, physical examination, and serum collection were done at each time point. From 561 patients, 30 patients were selected based on the following criteria: a pathologic diagnosis of SCC arising from the oropharynx, American Joint Committee on Cancer stage III or IV disease, and treatment with chemotherapy and radiation but not surgery. Patients were excluded if they were missing the pretreatment or more than one follow-up serum sample during the first year, unless this was due to death from early progressive or recurrent disease. Smoking status, tumor, nodal, metastatic stage, and treatment at entry, and disease status and recurrence dates were abstracted from study data forms and medical records by the study medical assistants, and notations were made about local, regional, or distant recurrences/metastatic disease. Patients for whom the treatment resulted in incomplete (partial) response were assigned a time to progression/recurrence of 1 day. By contacting patients every 3 months, study personnel were able to keep track of patient status (dead or alive).

Serum cytokine analysis. Peripheral blood samples (30 mL) were collected using routine venipuncture technique before treatment and at 3, 6, 9, and 12 months after treatment. Sera were collected after centrifugation of blood and stored in 0.5 mL aliquots at −80°C until testing. To ensure patient confidentiality and to blind laboratory personnel, all serum samples were immediately barcoded and assigned a numerical identifier before being stored in the University of Michigan Specialized Program of Research Excellence Tissue Core.

Serum samples were analyzed using the Luminex 100 instrument. Individual bead kits for IL-6, IL-8, VEGF, HGF, and GRO-1 were purchased from Biosource (Invitrogen). Lyophilized standards were chosen from individual bead kits to include but not replicate all five cytokines and were reconstituted with assay diluent using serial 1:3 dilution. Incubations were done on a microtiter plate shaker (Fisher) at 550 rpm. Filtration after incubations was done with a vacuum filtration manifold (Millipore).

Serum cytokine concentrations for all samples from individual patients were determined in the same assay to minimize interassay variability. Samples were thawed on ice, homogenized, and centrifuged at 14,000 rpm (10 min) to clear residual cellular debris. After prewetting a 96-well filter bottom plate, 25 μL of bead mixture and 50 μL of incubation buffer were added to each well. Each of the seven standards (100 μL each) and a blank were run in duplicate. Each patient serum sample was run in triplicate (1:2 dilution in assay diluent) and incubated for 2 h. After washing, biotinylated detector antibody was added (100 μL/well) and incubated for 1 h. Following additional washes, 100 μL of a streptavidin-R-phycoerythrin solution were added and incubated for 30 min. Final washes were done before suspending bead mixtures in 100 μL/well wash solution. Plates were read by the Luminex cytometer after cleansing and calibration of the instrument.

The instrument was programmed to not extrapolate cytokine values below the standard curve; therefore, the most dilute standard represents the lower limit of detection for each cytokine. Lower limits of detection for each cytokine are as follows: IL-6, 6 pg/mL; IL-8, 5.9 pg/mL; VEGF, 14.3 pg/mL; HGF, 5.3 pg/mL; and GRO-1, 13.7 pg/mL. Internal controls were done using known concentrations of recombinant protein for each cytokine.

Statistical analyses. The mean cytokine concentrations determined from triplicate assay values were used. All cytokine concentration measurements were log transformed (log10), resulting in symmetrical distributions that were nearly normally distributed. Consistent with common practice, we imputed values of one half the lower limit when values were below the level of detection for a given cytokine. The relationship between dichotomized tumor and nodal characteristics (T stage, N stage), smoking status, and survival was assessed using log-rank tests. We examined the relationship between baseline cytokine measurements and survival using Cox proportional hazards models, treating the cytokine measurements as continuous variables. To examine the relationship between longitudinal changes in cytokines and survival, we estimated a slope for each cytokine on every patient by fitting a least-squares regression line to longitudinal log-transformed cytokine measurements and examined the relationship between changes in these slopes and survival using a Cox proportional hazards model. To assess differences in survival between patients with three or more large longitudinal cytokine increases and those with two or fewer large increases, we used log-rank tests. These analyses were done without and with adjustment for smoking history. Kaplan-Meier curves were constructed to assess cause-specific overall and disease-free survival and relationship between cytokines and survival. All statistical tests are two sided. All analyses were done using SPLUS v7.0 software (Insightful Corp.).

Patient and tumor characteristics and survival. The characteristics of the patients, tumors, treatment, and smoking status are shown in Table 1. The median age was 56 years (range, 39-84), and the majority were men (72%). No statistically significant association was observed between age and cause-specific survival (P = 0.93, Cox model). The majority had a smoking history (67%). A significant correlation was found between a positive smoking history before diagnosis and decreased cause-specific survival (P = 0.05, log-rank test). Resumption of smoking was documented in only 11% during the first year, and no statistically significant correlation was observed between smoking during treatment, or smoking after treatment at any follow-up period, and cause-specific survival (P = 0.43 and 0.49, log-rank test). Similarly, there were no associations between these patient characteristics and disease-free survival (data not shown).

All patients had advanced oropharyngeal SCC with a majority having stage IV disease (87%) by tumor-node-metastasis (TNM) staging. Using Cox proportional hazards models, we found no association between T classification (P = 0.46), N classification (P = 0.81), or stage (P = 0.99) and cause-specific survival. All patients received radiation therapy and either concurrent, adjuvant, or neoadjuvant chemotherapy.

Patient clinical outcomes. Based on clinical evaluation during the first year of follow-up, 25 of 30 patients (83%) had a complete response to therapy. Five (17%) patients progressed and died of their malignancy; one died before the 3-month follow-up, one died before 6 months, one died before 12 months, and two died 16 months after initial presentation. Of the 25 patients that had a complete response to therapy, 4 (16%) developed a recurrence. One patient with a complete response died within 6 months of an unrelated cause. Based on this, we evaluated cause-specific survival rather than overall survival. Figure 1 displays Kaplan-Meier survival curves illustrating a 3-year cause-specific overall survival rate of 74.4% (Fig. 1A) and a 3-year disease-free survival rate of 68.9% (Fig. 1B).

Pretreatment factor levels and survival. Serum concentrations of IL-6, IL-8, VEGF, HGF, and GRO-1 determined from 30 pretreatment samples are shown in Table 2. We detected no significant relationships between differences in baseline IL-6, IL-8, HGF, or GRO-1 serum levels and cause-specific survival. However, higher baseline VEGF levels (geometric mean, 24.3; 95% confidence interval, 13.2-44.7 pg/mL) were strongly correlated with prolonged survival (P = 0.003). Based on this correlation, and the observation that the median baseline VEGF concentration approximated the lower limit of the VEGF assay (7.1 pg/mL), we evaluated survival differences between those patients who had VEGF concentrations above or below the median and limit of the assay. Figure 1C displays a Kaplan-Meier survival plot illustrating significantly prolonged cause-specific survival (P = 0.01, log-rank test) in patients with serum VEGF levels >7.1 pg/mL compared with patients with undetectable VEGF levels. Of note, 11 of 12 of the patients with baseline elevation of VEGF had a complete response and showed a decrease in VEGF after chemoradiation as indicated below.

Longitudinal serum factor patterns corresponding to response, progression, recurrence, and nonspecific events in individual patients. In a previous retrospective study, increase or decrease in concentration between pretreatment and posttreatment measurements of one or more serum growth or cytokine factors was observed in patients undergoing tumor progression, recurrence, or other pathologic events (8). This suggested the hypothesis that individually, or as a panel, these factors could potentially serve as biomarkers for response and survival. In the present study, 29 of 30 patients had at least one evaluable serum sample beyond the pretreatment sample because 1 patient died of progressive disease before the 3-month follow-up. Concentrations of all five cytokines were plotted against time with the vertical axis in log10 scale to enable visual comparison of cytokine changes in individual patients. The three patients in Fig. 2A show patterns representative of decrease in cytokine levels observed in the majority of the patients who achieved a complete response and long-term survival. Measurable levels of several factors at baseline, and decreases in three or more of the cytokines, including HGF, VEGF, IL-6, and/or IL-8, are observed after treatment, although some variability and delay in timing of the decline in cytokine responses was observed after chemoradiation in different patients. Figure 2B shows patterns representative of increase in cytokine levels that preceded or occurred concurrently in patients with early tumor progression or recurrence. Increases from pretreatment levels in three or more of the factors HGF, IL-6, IL-8, GRO-1, and/or VEGF are seen. In some instances, changes in levels of individual cytokines were not specific for progression or recurrence of SCC but rather indicated some other pathologic process occurring in the patient. Figure 2C shows one patient with a recurrence of a childhood pleomorphic sarcoma diagnosed at 12 months and another that had developed an aspiration pneumonitis at 6 months with corresponding alterations in cytokines. These observations are consistent with the hypothesis that longitudinal changes in individual or three or more of these factors could serve as early biomarkers for tumor response and survival, another malignancy, or of other inflammatory complications.

Relationship between longitudinal changes in serum factors, response, and survival. These observations indicated that overall directional change in serum factor levels could reflect early tumor response and potentially predict survival. After fitting a least-squares regression line to longitudinal log-transformed cytokine concentrations for each individual, we estimated quartiles of slopes for each cytokine (summarized in Table 3). Further, we estimated the relative risk (RR or hazard ratio) of cause-specific death of going from the first to the third quartile of slope using a Cox model, with estimated slope being treated as a continuous covariate.

We observed that, for each of the five factors, patients with large slopes (i.e., highest increases in cytokines) had poorer cause-specific survival. Secondary analysis revealed that patients with a smoking history tended to have larger slopes than those patients who did not (data not shown), so RRs of death were reestimated after controlling for a positive smoking history. Although RRs of cause-specific death decreased slightly when adjusted for smoking, the association of increasing cytokine slopes with decreased cause-specific survival remained statistically significant for each cytokine, consistent with expression of these cytokines by tumor in the absence of smoking.

Because we observed that changes in three or more of these factors are often observed together with response or progression (Fig. 3A and B), and have previously shown that expression of these cytokines in serum may be linked together by activation of transcription factors NF-κB in HNSCC (18–21), such cytokine signatures could have combined pathologic effects that contribute to poor patient outcomes. We thus hypothesized that an individual patient with longitudinal changes in three or more cytokines could predict survival. To examine this possibility, we defined any slope above the upper quartile (across patients) as a large change for each cytokine and determined differences in cause-specific survival between patients with three or more large cytokine slopes and patients with two or fewer large slopes. Figure 3A shows a Kaplan-Meier plot illustrating poorer survival in patients who showed large slopes for three or more cytokines (P = 0.004, log-rank test), with a calculated RR of death of 2.2 (P = 0.01) when adjusted for positive smoking history.

To evaluate whether our findings are related to tumor response, we examined differences in cause-specific survival between patients with three or more large cytokine slopes and patients with two or fewer large slopes among those patients that had a complete response only. Figure 3B shows a Kaplan-Meier survival plot indicating that, although patients with two or fewer large longitudinal increases in serum factors tended to survive longer, the difference was not statistically significant when response was taken into account (P = 0.19, log-rank test). Repeating this analysis for disease-free survival in patients who had a complete response (Fig. 3C) revealed that patients with three or more large increasing cytokine slopes had significantly worse disease-free survival (P = 0.04, log-rank test), with a RR of death of 1.9. However, when also controlled for positive smoking history, the difference in disease-free survival between the two groups did not achieve statistical significance (P = 0.13, log-rank test; RR of death, 1.6). These findings suggest that response and smoking history may be contributing determinants of cause- and disease-specific survival.

We show that longitudinal increase in serum levels of individual factors IL-6, IL8, VEGF, HGF, and GRO-1 is significantly associated with decreased cause-specific survival in patients with local-regionally advanced oropharyngeal SCC undergoing primary treatment with combined modality chemotherapy and radiation. We found that patients who had a history of smoking before the start of treatment tended to have greater slopes and decreased survival, a finding potentially consistent with inducible activation of the transcription factor NF-κB and proinflammatory cytokines by tobacco products (22, 23). After adjustment for smoking history, large increases in serum IL-6, VEGF, and HGF concentrations (upper quartile compared with the lower quartile) over time still conferred a 3.8-, 3.0-, and 2.9-fold RR of death, respectively. Furthermore, we observed that patients who had large increases in three or more in any of their longitudinal factor levels (defined as above the upper quartile across the population) had worse clinical outcomes than patients with large increase in two or fewer large factors, imparting a 2.2-fold RR of death after adjustment for smoking history. The finding that increasing levels in serum of three or more NF-κB–modulated factors are associated with progression and decreased survival is consistent with previous studies by us and others, implicating constitutive NF-κB activation and cytokine factors in pathogenicity and radiation resistance of HNSCC and other cancers (27–29). The finding that increase in individual or multiple serum factors is associated with decreased response, recurrence, and death during the first year after treatment suggests that longitudinal monitoring of these cytokines could be a useful adjunct to clinical monitoring and imaging during a time when surgical salvage and adjuvant therapy is still possible.

This is the largest and most definitive study in patients with HNSCC or any subsite that shows a statistical relationship between longitudinal changes in serum cytokines individually, or as a NF-κB–related panel in any cancer type, with cause-specific and disease-free survival. Uchida et al. (30) observed decreased longitudinal HGF levels in six patients with oral SCC who had a complete response. In our previous pilot study of 21 patients with HNSCC from different subsites (8), we found evidence for a statistically significant relationship between longitudinal decreases in IL-6 and HGF levels and overall survival, whereas decreases in IL-8 and VEGF only approached significance. In the current study, we found that changes in all four of these cytokines were significantly related to cause-specific survival, after addressing many of the limitations of the previous study, which included retrospective analysis, differences in site of tumor origin, treatment modality, limited sample of patients with posttreatment serum, variability in the intervals of sampling, and an insufficient numbers of smokers relative to nonsmokers to analyze the potential effects of smoking on outcome.

The relationship between elevated baseline serum VEGF and improved survival was unexpected, given the role of this cytokine in tumor angiogenesis (2). Shang et al. (31) observed a positive correlation between increased baseline serum VEGF and the rate of cervical metastasis and clinical stage in patients with oral SCC. Further, VEGF itself has been reported to promote radioresistance rather than radiosensitization through protection of tumor neovasculature from cytotoxic effects of radiation (32). However, we observed a similar disparity between baseline VEGF and outcome in a previous study between patients with laryngeal cancer randomized to surgery and chemoradiation therapy (11). Although higher baseline serum VEGF correlated with poorer overall survival in patients with advanced laryngeal SCC, when the two arms of this study were analyzed individually, higher VEGF levels significantly correlated with poor prognosis in the surgery arm but not in the chemoradiotherapy arm. A possible explanation for a relationship between high baseline VEGF and increased survival could be greater dependence in these tumors of expression of VEGF by a NF-κB–independent mechanism(s) that does not confer the same radioprotective effects as NF-κB. Consistent with this possibility, we have observed differences in the relative activation and contribution of transcription factors NF-κB and activator protein-1 to IL-8 and VEGF expression in HNSCC lines that differ in radiosensitivity (20, 28). Another transcription factor, Sp1, and translational factor, hypoxia-inducible factor-1α, have also been implicated in regulation of VEGF expression and radiosensitivity (33). Hypoxia-induced hypoxia-inducible factor-1α may be more likely to contribute to expression of VEGF in advanced stage III/IV tumors in our series (Table 1), as Winter et al. (34, 35) have shown that increased hypoxia-inducible factor-1α expression is associated with stage III/IV disease in specimens of patients with HNSCC undergoing surgery. It is important to note that, of patients with higher baseline VEGF levels, 11 of 12 (92%) showed decreased longitudinal VEGF levels with treatment, consistent with our other finding that longitudinal decrease in VEGF also represents a predictor of response of oropharyngeal SCC to chemoradiation therapy. Examination of the potential relationship of serum VEGF levels to response and survival and to NF-κB–independent mechanism(s) in tumor in a larger series of patients could validate if high baseline VEGF and other mechanisms, such as hypoxia-inducible factor-1α, represent independent biomarkers for selection for chemoradiation therapy.

Although the changes in serum cytokines overall and in most of our patients individually corresponded with tumor response to therapy, there were a few patients with increases in serum cytokines that were nonspecific for HNSCC. One patient showed increased cytokines at the time of diagnosis of recurrence of a salivary neoplasm. One patient developed an aspiration pneumonitis accompanied by sharp increases in several cytokines; however, all cytokine levels returned to preinfection levels by the next follow-up visit. One patient showed increases in four of five cytokines at the 12-month follow-up but has had no identifiable cause and remains clinically disease-free. Two additional patients showed transient spikes at the 6-month collection point and, on further chart reviews, were found to have had these blood samples obtained at the time of a biopsy procedure, consistent with a possible stress response. Consequently, sampling should be done before procedures that could induce increased serum factor levels, and longitudinal increases in individual or multiple serum cytokine in individual patients merit investigation for neoplastic, other inflammatory conditions, or complications.

The finding that longitudinal changes in the panel of cytokines and factors investigated in this study may have broader application for cancer detection, as elevated serum levels of one or more of the factors IL-6, IL-8, VEGF, and HGF have been detected in serum of patients with other types type of cancer, including non–small cell lung, breast, cervical, ovarian, gastric, and bladder carcinomas and melanoma (36–45). Delayed increase in levels of one or more of these factors after complete clinical response of SCC was observed in one patient in our previous study who developed a second primary lung cancer (8) and in the patient in this study who developed a recurrent pleomorphic sarcoma.

Our study and that of Hathaway et al. (46) have shown that multiplex technologies, such as Luminex, can provide profiles and quantitation of specific soluble protein factors, which are implicated in pathogenesis of HNSCC from patient sera. When combined with clinical evaluation, measuring NF-κB–related serum cytokines may be useful to monitor tumor response to therapy, influence threshold of suspicion for treatment failure, and ultimately lead to earlier intervention and improved patient outcomes.

We thank John Morris, Martin Gutierrez, and Shivaani Kummar for their review and comments on the manuscript.