Comprehensive Evaluation of Nuclear Factor-κΒ Expression Patterns in Non-Small Cell Lung Cancer

Nuclear factor (NF)-κB signalling is required for lung adenocarcinoma development in mice, and both of its subunits RelA and RelB were independently reported to be highly expressed in human non-small cell lung cancer (NSCLC). To comprehensively examine NF-κB expression in NSCLC, we analyzed serial sections of primary tumor samples from 77 well-documented patients (36 adenocarcinomas, 40 squamous cell carcinomas and 3 large cell carcinomas) for immunoreactivity of RelA, RelB, P50, and P52/P100. Tumor and intratumoral stroma areas were discriminated based on proliferating cell nuclear antigen immunoreactivity and inflammatory infiltration was assessed in intratumoral stroma areas. NF-κB immunoreactivity was quantified by intensity, extent, and nuclear localization and was cross-examined with tumor cell proliferation, inflammatory infiltration, and clinical-pathologic data. We found that the expression of the different NF-κB subunits was not concordant, warranting our integral approach. Overall, RelA, RelB, and P50 were expressed at higher levels compared with P52/P100. However, RelA and P50 were predominantly expressed in intratumoral stroma, but RelB in tumor cells. Importantly, tumor area RelA expression was correlated with the intensity of inflammatory infiltration, whereas RelB expression was identified in proliferating tumor cells. Using multiple logistic regression, we identified that tumor RelB expression was an independent predictor of lymph node metastasis, and tumor P50 was an independent predictor of TNM6 stage IIB or higher, whereas tumor RelA was an independent predictor of inflammatory infiltration. We conclude that pathologic studies of NF-κB expression in cancer should include multiple pathway components. Utilizing such an approach, we identified intriguing associations between distinct NF-κB subunits and clinical and pathologic features of NSCLC.

Introduction the University Hospital of Patras were retrospectively enrolled in the study. Some of these patients had also been studied in a previous report from our Institution [28]. Patients, all Caucasian, 69 men and 8 women aged from 46 to 84 years, were carefully selected based on the availability of full clinical and pathologic data extracted from the primary pathology reports and patients' files, including age, sex, histologic type, grade, and detailed pathologic stage. Patients selected included 34 patients with lung adenocarcinomas, 40 with squamous cell carcinomas, and three with large cell carcinomas. All patients were staged according to the sixth edition of the TNM classification for NSCLC, current at the time of their diagnosis [29]. The study's observational protocol was conducted according to the Declaration of Helsinki, was approved by the Ethics Committee of the University Hospital of Patras, Greece, and all patients gave written informed consent.

Statistics
Probability values (P) less than 0.05 were considered significant. Human NF-κΒ subunit scores were not normally distributed, as tested by the Kolmogorov-Smirnov test (P < 0.05), and are shown in bar graphs as median with boxes indicating interquartile range and whiskers indicating 95% percentiles. Matched numerical NF-κΒ subunit scores of the same tumors were compared using Friedman's tests followed by Dunn's post-tests. Categorical NF-κΒ subunit scores were compared using χ 2 tests followed by Fisher's exact tests. Correlations between numerical NF-κΒ subunit scores and clinical-pathologic data were done using Spearman's tests, whereby the overall threshold of statistical significance (P < 0.05) was adjusted to the total number of correlations examined (n = 153) by the Bonferroni method to P < 0.05/153 = 0.000327. Comparisons of numerical NF-κΒ subunit scores by inflammatory infiltration degree and clinicalpathologic categories were done using Wilcoxon signed rank tests or Kruskal-Wallis tests followed by Dunn's post-tests, for two or multiple comparison groups, respectively. To determine whether numerical NF-κΒ subunit scores independently predict clinical-pathologic variables, the latter were dichotomized by their median and binary logistic regression analyses were performed using the backward Waldman method. For this, NF-κΒ subunit scores were used as the input (independent variables) and clinical-pathologic variables and inflammatory infiltration scores as the output (dependent variables). Mouse NF-κΒ subunit scores at progressive stages of carcinogenesis are presented as mean ± SD and were compared using two-way ANOVA followed by Bonferroni post-tests. Statistical analyses were performed using Prism v5.0.0 (GraphPad, San Diego, CA) and the Statistical Package for the Social Sciences v20 (IBM SPSS Statistics, Chicago, IL, USA).

NF-κΒ subunit expression patterns in NSCLC
The clinical and pathologic features of the study patients are summarized in Table 1. The raw study results are appended to this article as S1 Table. We first examined NF-κΒ subunit expression in normal lung areas adjacent to our surgically resected tumor samples. In bronchial and alveolar epithelium, RelA, P50 and P100/P52 exhibited low or moderate cytoplasmic immunoreactivity, while RelB displayed higher expression levels. In juxta-tumoral bronchial and alveolar hyperplasias, immunoreactivity for all NF-κB subunits showed a stronger expression pattern, with RelB showing both cytoplasmic and nuclear localization. In tumors, all NF-κB subunits were highly expressed relative to normal and hyperplastic areas, with RelB scoring highest and showing the highest degree of nuclear localization. However, RelA, RelB, and P50 immunoreactivity was stronger compared with P100/P52 (Fig 1A and 1B). When NF-κB subunit scores were subdivided into low (0-4), intermediate (5)(6), or high (7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18), no relationship was evident between the different subunits ( Fig 1C). No statistically significant differences were noted between the different NSCLC histologic subtypes. However, statistically significant differences were observed in NF-κB subunit expression levels within each tumor: tumor areas displayed different expression patterns compared with intratumoral stroma areas (Fig 1A; representative panels in red frame). These results suggested that multiple NF-κΒ subunits are progressively overexpressed in NSCLC, showed that not all NF-κΒ subunits are expressed equally,  and identified a discordance in the expression of multiple NF-κΒ subunits between tumor and intratumoral stromal areas, warranting further investigation.

Differential NF-κΒ expression in tumor and stromal compartments of NSCLC
To study this in more detail, all samples were stained for PCNA, which clearly identified tumor areas (high abundance of PCNA+ cells) from the intratumoral stroma compartment (very low abundance of PCNA+ cells), and NF-κB subunit scoring was repeated separately for these two areas in serial sections of all patients (Fig 2A). These analyses revealed that RelB was the subunit expressed the strongest in areas of tumor cells displaying a both nuclear and cytoplasmic expression pattern. On the other hand, the remaining subunits studied showed only moderate cytoplasmic expression in tumor areas (Fig 2A and 2B). When tumor compartment NF-κB subunit scores were subdivided into low (0-4), intermediate (5)(6), or high (7-18) and compared within each tumor, RelB and P50 showed significant discordance, with tumors with strong RelB immunoreactivity exhibiting low P50 scores ( Fig 2C). Surprisingly, evaluation exclusively of the intratumoral stroma yielded different results: here, RelA and P50 subunits were most predominantly expressed in the nuclei of stromal cells. In stark contrast, RelB and P100/P52 showed only weak cytoplasmic immunoreactivity (Fig 2A and 2D). When stroma NF-κB subunit scores were subdivided into low (0-4), intermediate (5)(6), or high (7-18) and compared within each tumor, RelB and P100/P52 showed significant concordance, with 49/77 tumors displaying simultaneously low scores for both subunits, likely reflecting the low expression levels of both proteins (Fig 2E). We finally examined the relationships between numerical and categorical tumor and stroma scores for each NF-κB subunit, finding that only P100/P52 expression in tumor and the related stroma were concordant and correlated (Fig 2F and 2G).

Correlation of Rel subunit expression with cellular proliferation and inflammatory infiltration
Since NF-κB is mechanistically implicated in the proliferation and inflammatory signalling of lung tumor cells in mouse models [21][22][23][24], we sought to determine the degree of cellular proliferation and inflammatory infiltration in our 77 patients with NSCLC. For this, serial sections were subjected to PCNA immunostaining as above, with simultaneous hematoxylin staining of isotype control sections, and the percentage of PCNA-positive cells in tumor areas was assessed as a measure of cellular proliferation rate. In addition, the percentage of tumor-infiltrating PCNA-negative inflammatory cells in stroma areas was determined for each patient (Fig 3A). These analyses identified that, although expression levels of RelA in tumor areas were quite low, they were increased in tumors with high levels of inflammatory infiltration (Fig 3B). Moreover, although the percentage of PCNA-positive cells was not different between the different histologic subtypes (Fig 3C), and was not associated with the expression of any NF-κB subunit (data not shown), dual immunofluorescence for RelA or RelB combined with PCNA revealed that PCNA+ proliferating cells predominantly expressed RelB in their nucleus ( Fig  3D). These data indicated that tumor RelA expression was associated with inflammatory infiltration and that tumor RelB expression was linked with cellular proliferation within each tumor. Since RelB expression has been found to be regulated by both the canonical as well as Overall scoring of NF-κB subunit expression levels. Data presented as median with boxes indicating interquartile range and whiskers indicating 95% percentiles. ns, * and ***: P > 0.05, P < 0.05, and P < 0.001 for indicated comparisons by Friedman's test followed by Dunn's post-tests. (C) Co-expression matrixes of categorical NF-κB subunit expression levels. For this, NF-κB scores from (B) were categorized into low (0-4), intermediate (5)(6), and high (7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18). ns: P > 0.05 by χ 2 tests followed by Fisher's exact tests. the alternative NF-κB pathways, we next sought to examine a possible correlation between RelA expression in tumor or stroma compartments, and RelB expression in tumor cells.  However, no statistically significant correlation was found when the data were assessed either in a parametric or a non-parametric fashion, suggesting that tumor RelA and RelB expression occur independently in NSCLC.

Association of NF-κΒ subunit expression with clinical and pathologic features of NSCLC
We next examined whether NF-κB subunit expression in tumor and stroma compartments is correlated with clinical and pathologic features of our patients with NSCLC. For this, correlation analyses were done, which revealed no significant associations (data not shown). NF-κB scores were further subdivided according to clinical and pathologic features and subgroups were compared (Fig 4A). The stroma score of P100/P52 was significantly lower in well-differentiated tumors, as compared with low or moderately differentiated ones. In addition, stromal P100/P52 expression was decreased in patients with N2 disease, as compared with patients without lymphatic involvement. No other significant differences were identified regarding the intratumoral stromal expression of any other subunit. When tumor area NF-κB scores were examined this way, RelA was found to be expressed at higher levels in men compared with women, in patients aged 65 and over compared with younger ones, and in patients with squamous cell carcinoma compared with adenocarcinoma. Importantly, tumor P50 scores were elevated in patients with N2 and pTNM6 stage III disease compared with patients without lymphatic involvement and stage II disease, respectively. Finally, tumor P100/P52 expression was statistically significantly increased in patients with pTNM6 stage II disease compared with patients with stage I disease. In a third level of investigations, clinical and pathologic NSCLC features were dichotomized by their median and were entered as dependents into binary logistic regression analyses, using tumor and stromal NF-κB scores as the input (independent) variables ( Fig 4B). These analyses revealed that tumor RelA score was an independent predictor of age 65 or older, of squamous histology, and of inflammatory infiltration. Interestingly, tumor and stroma RelB expression were identified as positive and negative predictors of nodal involvement, respectively. Finally, tumor P50 score was found to be a independent predictor of poor differentiation and pTNM6 stage IIB or higher. Hence distinct NF-κB subunit (RelB and P50, but not RelA) expression was associated with advanced disease, and distinct NF-κB subunits were linked with defined features of NSCLC, validating our integrated approach.

NF-κΒ subunit expression in mouse models of NSCLC
We finally sought to verify whether our findings could also be recapitulated in two different mouse models of human NSCLC, namely the model of urethane-induced pulmonary adenoma and of mutant KRAS-induced pulmonary adenocarcinoma. Analyses, done as for human samples, identified that RelB was overexpressed in urethane-induced hyperplastic lesions, but not in adenomas. Moreover, RelB was overexpressed in both mutant KRAS-induced hyperplastic lesions and adenocarcinomas (Fig 5). These findings were consistent with the human data showing focal overexpression of RelB in tumor areas of human NSCLC (Fig 2).

Discussion
In the present study, we characterized the expression and the subcellular localization of several NF-κB subunits in human NSCLC and experimental models of the disease. We found different patterns of immunoreactivity for each of the subunits studied, justifying our integrated approach. In specific, RelB was highly expressed in tumor cells, while RelA and P50 in stromal cells. Interestingly, tumor RelA expression was correlated with inflammatory infiltration and tumor RelB expression was collocalized with proliferating tumor cell nuclei. These results were consistent with binary logistic regression analyses that revealed that tumor RelA expression was an independent predictor of inflammatory infiltrates, while tumor RelB immunoreactivity predicted more advanced disease with nodal involvement. Finally, this pattern of predominant RelB expression in human NSCLC was recapitulated in mouse models of the disease. Collectively, our results indicate a multi-modal activation of NF-κB in NSCLC and possibly suggest divergent functions for RelA and RelB in driving paracrine induction of inflammation and cellautonomous or autocrine promotion of cellular proliferation, respectively (Fig 6). Although previous reports from various human cancers support a predominantly canonical NF-κB activation pathway mostly mediated by RelA/P50 [32][33][34][35][36][37], recent data along with ours highlight the possibility for a non canonical NF-κB activation pattern in different cancers [38,39]. In 2005, Lessard and colleagues reported investigations on the expression of multiple NF-κB subunits in prostate cancer tissue arrays; they found nuclear subunit combinations such as RelB-P100/P52 and RelA-RelB, introducing for the first time a different NF-κB pattern associated with the progression of the disease [39]. In addition, a more recent study of head and neck squamous cell carcinoma proposed a combined effect of both IKKα and IKKβ on the Overall scoring of NF-κB subunit expression levels from four mice per group. Data presented as mean ± SD. ** and ***: P < 0.01, and P < 0.001 for the indicated color-coded subunit compared with normal bronchial and alveolar epithelium by two-way ANOVA followed by Bonferroni post-tests. Non-significant comparisons are not indicated. nuclear localization of canonical RelA and alternative RelB and P100/P52 subunits [40]. These studies, collectively with our results, suggest a possible alternative NF-κB activation pattern in malignant versus benign cells. This could imply altered intracellular and paracrine signalling from tumor cells in response to alternative NF-κB activation, since RelA-P50 and RelB-P100/ P52 complexes bind to NF-κB binding sites of different promoters [41]. In addition, several studies correlate the nuclear membrane transporter chromosomal region maintenance/expor-tin1 protein (CRM1) with tumor progression in several types of cancers and CRM1 is known to export RelA from the nucleus into the cytoplasm in ovarian cancer, a phenomenon which could explain the observed cytoplasmic localization of RelA in the present and other studies [42][43][44][45].
Both RelA and RelB nuclear immunoreactivity have been independently identified in human NSCLC by two different groups [20,28]. We report here a possible explanation for the different results: we identified for the first time that tumor and intratumoral stroma compartments display distinct NF-κB subunit expression patterns, predominantly signified by noncanonical and canonical subunits, respectively. This contrasting expression pattern could account for the divergent results, if intratumoral stromal areas were to be interpreted as reflecting tumor areas in terms of NF-κB subunit expression, and warrant careful discrimination of these distinct areas within each tumor when looking for NF-κB immunoreactivity. Moreover, we show how this can be readily done using PCNA immunostaining.
A large body of evidence supports that inflammation fuels the progression of bodily tumors, including lung cancer [46,47]. In addition to promoting tumor cell proliferation, NF-κB is a cardinal mediator of innate immune responses in the lungs and other organs, by driving the expression of several genes involved in cell cycle progression and non-cell-autonomous inflammatory signalling [48]. Our findings suggest dual functions for canonical and alternative NF-κB pathway components in these two processes: RelA expression in tumor areas was elevated in tumors with higher degrees of inflammation and independently predicted severe tumorassociated inflammation, while nuclear RelB coincided with proliferating cells within each tumor. Hence, in NSCLC, RelA may promote the expression of secretory genes encoding proinflammatory cytokines and chemokines responsible for attracting inflammatory cells to the tumor site, with possible pro-tumor or anti-tumor functions [48]. In turn, RelB may facilitate the expression of cell cycle genes that promote cellular proliferation in a cell-autonomous fashion.
Studies in animal models are in accord with our findings and consistent with the hypothesis that non-canonical NF-κB activation pathways are at play in NSCLC. In particular, a previous study from our group identified that the proteasome (and hence canonical NF-κB pathway) inhibitor bortezomib dramatically enhanced instead of halting chemical-induced lung carcinogenesis in mice [49]. Bortezomib has also shown limited efficacy against human NSCLC [50]. Collectively, our data presented here and other reports from the literature indicate that noncanonical NF-κB transcriptional activity may exist in lung and other tumors.
In addition to triggering tumor-associated inflammation, NF-κΒ is positioned to critically impact tumor cell proliferation. We found that RelB was expressed in dividing cells of NSCLC, suggesting that this NF-κΒ subunit may be linked with NSCLC cellular proliferation in an either positive or negative fashion. Although our study's design does not allow for the deduction of functional conclusions on RelB effects on NSCLC cell proliferation, a recent study of head and neck squamous carcinomas revealed that IKKα, cooperatively with IKKβ, promotes cellular growth as well as nuclear RelB translocation in this tumor type, suggesting possible pro-tumor effects for RelB [40]. In addition, Jacque and co-workers showed that RelB exerts inhibitory effects on tumor cell proliferation by up-regulating tumor protein 53 (TP53) expression, which are lost upon the occurrence of TP53 mutations or loss [51]. We speculate that RelB may exert pivotal effects on tumor cell proliferation in our patients with NSCLC, since many patients with this tumor type have been consistently and repetitively shown to feature all kinds of TP53 perturbations, including copy number alterations and point mutations, while a significant proportion retain the wild-type tumor suppressor. In any case, our results collectively with the literature warrant further investigation of RelB functions in cancer.
The limitations of our study are not to be overlooked. First, the descriptive nature of our findings do not allow for functional implications prior to experimental validation. Second, since the majority of our study's participants were male, our conclusions should be regarded as male-specific. In addition, the limited number of patients with large cell carcinoma does not allow for conclusions to be drawn for this tumor type.
In summary and despite their inherent limitations, our findings support that alternative modes of NF-κB activity are functional in human and murine NSCLC, in addition to canonical NF-κB activity. This non-canonical NF-κB activity may be responsible for the lack of efficacy of canonical NF-κB inhibitors against NSCLC and warrants further investigation.
Supporting Information S1