Poor nutrition is a frequent finding in areas with high risk of cancer, and poor nutrition can be associated with impaired immunologic function. Poor nutrition may result in an inadequate intake of immunologically essential nutrients, such as selenium, zinc, vitamin C and vitamin E (8), that influence Th1 and Th2 responses (9) and that are particularly important during critical periods of gestation and neonatal maturation (10).

Our group has conducted multiple studies of the association of nutrition and esophageal and gastric cancers in Linxian, China, a region with some of the highest rates of esophageal squamous cell carcinoma (ESCC) and gastric cardia adenocarcinoma in the world (11). These include the Linxian General Population Nutrition Intervention Trial (12;13). This study supplemented 29,584 residents for 5.25 years with different combinations of nutrient supplements or placebo. It found that subjects supplemented with selenium, beta-carotene, and alpha-tocopherol had a statistically significant reduction in all-cause mortality (9%) and in total cancer mortality (13%) (13). A prospective study of this cohort during the 5.25 yr trial period found an inverse relation between baseline serum selenium concentrations and risk of ESCC; the relative risk (95% CI) of the highest to lowest quartiles was 0.56 (0.44-0.71) (14). Continued follow-up of this cohort for 10 years after the trial period confirmed the persistence of this inverse association (15). We subsequently conducted a randomized, controlled 10 month intervention of selenomethionine 200ug QD and/or celecoxib 200mg BID (2 × 2 factorial design) that included 238 residents of Linxian with biopsy-proven mild or moderate squamous dysplasia (16). This Linxian Chemoregression Study (CRS) was designed to determine whether supplementation with the antioxidant micronutrient selenium and/or the COX-2 inhibitor celecoxib would affect the progression of esophageal squamous dysplasia, the precursor lesion of invasive ESCC. Among the 115 subjects who had mild dysplasia at baseline, selenomethionine treatment favorably altered dysplasia grade (p=0.02). No effect on dysplasia was seen with celecoxib treatment (16). Further molecular evaluation of the CRS suggested that the beneficial treatment effect may have been related to modulation of immune function (17;18). Individuals whose lesions regressed had higher RNA expression in genes associated with immune stimulation, including antigen presentation, T-cell activation and T-cell survival (17). However, selenomethionine supplementation did not have a measurable effect on the gene expression profiles of normal squamous esophageal mucosa (18).

Both the seleniomethionine and celecoxib supplements in the CRS have possible associations with immune function. Selenium is incorporated as selenocysteine into a variety of selenoproteins and has been shown to enhance cell-mediated immune responsiveness and to inhibit NF-kappa B activation, which leads to reduced expression of NF-kappa B target genes, including inflammatory cytokines (19;20). Celecoxib's immunologic role relates to its effect on COX-2 and is well illustrated by its role in colon carcinogenesis and treatment (21). This includes the association between NSAID use and regression of colorectal polyps in patients with familial adenomatous polyposis and, of particular relevance to the current study, COX-2's role in prostaglandin synthesis and related inflammatory mediators, downstream effectors on gene expression and apoptosis.

Subjects completed a baseline questionnaire and underwent venipuncture at t0 and t10. The t0 blood samples were all collected within 2 weeks of each other, and the t10 blood samples were all collected during a similar, narrow time frame and immediately placed on ice prior to centrifugation. Seven cc of whole blood was collected in EDTA for hemoglobin, WBC and platelet counts; and an additional 7 cc was collected in a separate tube without anticoagulant which was left to clot and then centrifuged for 10 minutes. An aliquot of the separated serum was sent for chemistry analysis, and the remainder of the serum was stored at −80C until cytokine analysis.

Serum cytokine analysis utilized the Human Cytokine-22-Plex Antibody Bead Kit (‘Beadlyte’, Upstate, Charlottesville, VA). This approach combines the principle of a sandwich immunoassay with Luminex's fluorescent-bead technology. More specifically, beads with defined spectral properties were conjugated to analyte-specific capture antibodies. Samples were pipetted into a 96-well filter-bottom microplate and incubated with the capture antibodies for 2 hours. During a second incubation, the analyte-specific biotinylated detector antibodies were added and incubated with the beads for 1 hour. The analyte-specific biotinylated detector antibodies recognize their epitopes and bind to the appropriate immobilized analytes. Then streptavidin conjugated to the fluorescent protein, R-Phycoerythrin (Streptavidin-RPE), was added and incubated for 30 minutes. The Streptavidin-RPE binds to the biotinylated detector antibodies associated with the immune complexes on the beads, forming a four-member solid phase sandwich. A Luminex 100 instrument (Luminex Corp, Austin, TX) evaluated the spectral properties of the beads and the amount of associated RPE fluorescence. Sixteen control wells per plate were used for a standard serial dilution curve (8 point curve in duplicate).

We utilized a quartet-based design for the sample wells and included four samples from a QC pool on each plate. Each quartet contained a participant from each of the 4 treatment groups. The subjects from each treatment group were randomly permuted and the first person from each of the 4 permuted lists was used to complete the first quartet, then the second person from each list was used to complete the second quartet, and so forth. The t0 and t10 samples of each quartet were included together on the same plate.

The statistical analysis explored 3 objectives: (1) whether selenium or celecoxib treatments altered cytokine levels; (2) whether baseline cytokine levels were associated with baseline dysplasia, dysplasia at t10, or change in dysplasia, after controlling for treatment effects; and (3) whether the effect of treatment on dysplasia (regression, stable disease, progression) may have been mediated by cytokine levels.

For (1), we compared the t10-t0 differences in median MFI values between treated and untreated subjects for each intervention agent. We used Wilcoxon rank tests with plate-specific ranks of cytokine levels (PROC TWOSAMPL, StatXact 7, Cytel Inc.) because the cytokine distributions had many outliers and were strongly non-normally distributed even under log or square root transformations. We repeated all Wilcoxon tests without plate-specific ranks, but the results were always similar to the results of the Wilcoxon tests with plate-specific ranks (data not shown). We did not formally adjust for multiple hypothesis testing because this study was hypothesis-generating in nature.

Treatment effects were quantified as the difference of the median t10 cytokine level minus the median t0 cytokine level, between participants on versus not on the individual treatments. For example, the selenium treatment effect on t10-t0 cytokine changes was calculated as (median(c_t10,sel) − median(c_t0,sel)) − (median(c_{t10,no sel}) − median(c_{t0,no sel})) where c denotes a cytokine level, and the subscript “sel” denotes selenium treatment and “no sel” denotes no selenium treatment. These treatment effects are expressed both as actual change in median MFI values and as percent change in median MFI values.

For (2), we conducted trend tests for each cytokine, using linear models with dysplasia outcomes coded as consecutive integers reflecting ordinality (for dysplasia at one time point: 0=no dysplasia, 1=mild dysplasia, 2=moderate dysplasia, 3=severe dysplasia; for change in dysplasia: −1=regression (mild to no dysplasia, or moderate to mild or no dysplasia), 0=no change, 1=progression (mild to moderate or severe dysplasia, or moderate to severe dysplasia)). Separate trend test models were run for each cytokine at t0, with the cytokine levels categorized into tertiles. The MFI tertiles were coded 1, 2, and 3 for the first, second, and third tertiles, respectively. For the t10-t0 cytokine differences, we subtracted each cytokine's tertile coding at t0 from its tertile coding at t10. Adjusting covariates included plate number and treatment. We fit these models including everyone, and also including only those with mild dysplasia at t0 or those with moderate dysplasia at t0, to look for different effects by initial dysplasia.

For (3), we conducted trend tests using linear models (as in objective (2)) to see if changes in cytokine levels (t10-t0) were associated with changes in dysplasia. If an individual cytokine's t10-t0 change was associated with a change in dysplasia and that cytokine's level was also associated with a treatment that changed dysplasia in the same direction (in objective 1), then we considered this to be evidence that this cytokine may have mediated the treatment effect. For cytokines with significant trend tests, we fit proportional odds logistic regression models (22) to estimate odds ratios for a worse change in dysplasia as the cytokine increased by one tertile from t0 to t10.

Exploratory data analysis showed that the cytokine distributions were greatly skewed, with many large outliers. Consequently, we present the summary data using medians, rather than means, with interquartile range (IQR) as a measure of spread, rather than variance.

The within-plate CVs for MFI readings for each cytokine averaged 23% (range: 7%-33%) and the overall CVs averaged 32% (range: 7%-58%). Our analyses reduced the effect of between-plate variation by either stratifying on plate or including plate as a covariate for regression adjustment. Thus, the within-plate CVs are more relevant to our analyses.

First, we tested if selenomethionine or celecoxib treatments were associated with changes in cytokine levels from t0 to t10. Selenomethionine treatment was associated with a lower t10-t0 difference in IL-2 (a median effect (IQR) of −1.0 (7.4) MFI, a −9.0% change, p=0.04) (Table 3). Celecoxib treatment was associated with decreased IL-7 (a median effect (IQR) of −2.5 (18) MFI, a −11% change, p=0.006) and increased IL-13 (1.5 (4.0) MFI, a 17% change, p=0.008) (Table 3).

Second, we tested for an association between a trend in baseline cytokine tertiles with baseline dysplasia, t10 dysplasia, and change in dysplasia between t0 and t10. None of these trend tests was significant (Table 4), and there were no significant Treatment by Cytokine interactions for any trend-test model for any of the outcomes on either scale (MFI or concentration), including the most complex model of t0 or t10 cytokine level, and dysplasia, stratified by plate, treatment intervention, and baseline dysplasia level (data not shown).

Third, we tested whether an increase in cytokine MFI tertile from t0 to t10 was associated with a change in dysplasia during that time. All 22 cytokines were evaluated (data not shown); the results for IL-2, IL-7 and IL-13, the three cytokines that were significantly associated with intervention treatments (Table 3), are shown in Table 5. An increase in cytokine IL-2 tertile was non-significantly associated with an increase in dysplasia grade for all participants (OR=1.32, p=0.098), and was significantly associated with worsening dysplasia among those with mild (OR=2.0, p=0.01), but not moderate (OR=1.04, p=0.8), dysplasia at baseline. An increase in IL-7 tertile was also significantly associated with an increase in dysplasia grade both overall and among those with mild dysplasia at t0 (all participants OR=1.47 (p=0.03); mild dysplasia at t0 OR=2.53 (p=0.001), moderate dysplasia at t0 OR=0.84 (p=0.5)) (Table 5).

None of the baseline cytokine levels were associated with dysplasia status at baseline or after 10 months of intervention, or with the change in dysplasia from baseline to 10 months after intervention. However, an increase in IL-2 or IL-7 between baseline and the end of intervention was associated with an increase in dysplasia grade. These associations were most evident in participants who had mild dysplasia at baseline and were given selenomethionine treatment. Since selenomethionine treatment was associated with a reduction in dysplasia grade in the parent trial and a reduction in IL-2 levels in this study, and changes in IL-2 level paralleled the changes in dysplasia grade, it is plausible that the beneficial effect of selenomethionine treatment on dysplasia observed in the original trial may have been partially mediated by its effects on lowering IL-2. This hypothesis would be consistent with reports of a direct relationship between IL-2 serum levels and disease progression in other tumors (23). The same hypothesis cannot be made for IL-7, because selenomethionine treatment did not affect its levels in this trial.

There are several possible mechanisms for anticarcinogenic effects of selenium compounds, including immunologic modulation (24). The potential therapeutic importance of enhancing immunologic function, as may have occurred with the selenomethionine treatment in this trial, stems from the fact that host-derived cytokines can modulate an anti-tumor response and, consequently, can play a beneficial role in suppressing tumor formation. This response has been associated with a reduced frequency of metastasis and improved patient survival in early-stage cancers of several types (25).

Of particular relevance to the current findings are several possible mechanisms by which a selenium-induced reduction in IL-2 might protect against tumor development. Selenium inhibits NF-kappa B expression, which can result in reduced expression of NF-kappa B's target genes, including IL-2 (19;20). A reduction in serum IL-2 might be especially protective against IL-2 receptor alpha (IL-2Ra)-expressing tumors (26;27). These IL-2 promoted tumors include squamous cell carcinoma of the lung and cervical intraepithelial neoplasias, entities with histologic and etiologic similarities to esophageal squamous cell carcinoma (23;27). And in the evaluation of cervical intraepithelial neoplasia, IL-2Ra expression was also associated with the grade of squamous dysplasia (27), suggesting that serum IL-2 may specifically help regulate pre-invasive squamous disease, which was the focus of the Linxian chemoprevention trial.

Selenium-induced reduction in IL-2 might also protect against tumor development by reducing the activity of regulator T-cells (Tregs), which are T-lymphocytes that can impair cancer immunosurveillance. IL-2 activated Tregs protect against autoimmune reactivity but suppress the activation, proliferation, and effector functions of tumor infiltrating lymphocytes (TILs) (28-31), which may be particularly relevant in modulating immune responses during the earliest stages of neoplastic disease. Thus, the reduction in IL-2 seen with selenomethionine treatment could lead to reduced Treg activity and enhanced tumor immune surveillance.

In a separate set of studies, we have previously shown that the people of Linxian are highly exposed to polycyclic aromatic hydrocarbons (PAHs), carcinogenic products of incomplete combustion of organic matter (32-34). It is interesting to note that PAHs can also have harmful effects on immunologic function. The intracellular receptor for PAHs is the aryl hydrocarbon receptor (AhR), and increased exposure to PAHs stimulates the production of AhR target genes (35). AhR in turn can be immunotoxic, because it stimulates Tregs, which impair cancer immunosurveillance. Thus PAH exposure and selenium exposure are probably both impacting cancer immunosurveillance in Linxian (in opposite directions) via their effects on Tregs.

IL-7 decreased with celecoxib supplementation and was positively associated with a worsening of preneoplastic disease. IL-7 is a member of the common gamma-chain family of cytokines which are involved in homeostatic proliferation and survival (36;37). It has a potential regulatory role affecting the interaction between mucosal lymphocytes and intestinal epithelium (38-40), is constitutively expressed by gastrointestinal epithelial cells (39), and is involved in the development and persistence of chronic inflammatory bowel disease (40). Thus, the decrease in IL-7 seen with celecoxib supplementation could serve to lessen a potentially adverse chronic inflammatory response that might potentiate cancer formation (41).

IL-13, which was increased with ten months of celecoxib supplementation, also plays an important role in various inflammatory diseases, including cancer, asthma, and allergy (42;43). In this study, IL-13 was not associated with a change in preneoplastic disease and, thus, the potential clinical significance of its association with celecoxib could not be further assessed.

A significant strength of the current study was the original intervention trial, with its accurate, paired, endoscopic categorization of existing esophageal disease (44) before and after an intervention that showed a significant beneficial effect. Thus, this study provided an opportunity to explore potential mechanisms to explain the observed changes in preneoplastic disease in the parent intervention trial.

A limitation of the study was the relatively high within-plate and plate-to-plate variability, the reasons for which remain unclear. This variation was similar to that recently reported by others but still deemed successful at yielding high intraclass correlation coefficients that were useful for epidemiologic studies (45). Another limitation of all cytokine studies is the fact that serum cytokine levels may change rapidly over time in an individual in response to changes in diet and other factors, so that a single serum measurement may not represent long-term cytokine levels. This issue may be less important in a population with chronic nutrient deficiencies such as our study population. Finally, it is also possible that some of the associations that we observed may have occurred by chance, since we examined 22 cytokines and did not adjust for multiple comparisons. Thus our observations must be regarded as exploratory and need independent confirmation.