RANTES Expression Is a Predictor of Survival in Stage I Lung Adenocarcinoma¹

Lung cancer is the leading cause of cancer-related death and is the second most common malignancy in both men and women. Surgery, the mainstay of therapy for NSCLC,³ accounts for only a 14% 5-year survival in all patients (1). In patients who undergo resection, no single variable has been shown to more strongly predict survival than the tumor-node-metastasis staging system (2). Given the variability of survival within a given stage, many have searched for alternative or adjunctive variables to more accurately predict patient prognosis. These variables have been both of a clinical and a molecular nature (3, 4, 5, 6).

Patients who have an ALR+ within their NSCLC tumor demonstrate an improved prognosis compared with those patients without ALR (ALR−). This relationship is especially true for tumors with high levels of intratumoral CD3⁺ and S100⁺ cells (7). This relationship has also been shown for stage I NSCLC tumors with peritumoral B-lymphocytes (8) and for advanced lung cancer patients (9).

The mechanisms underlying the presence or absence of lymphocytic infiltration within a lung tumor mass are unknown. The immune system may tolerate a tumor because of the tumor’s failure to produce the needed “danger” signals required for recognition and inflammation (10). On the other hand, tumors may also secrete signals that actively down-regulate the immune response and allow the tumor to escape immune surveillance (11). Regardless of how these cells arrive in the tumor, stimulation of peritumoral lymphocytes results in the release of tumoricidal cytokines (12). Some therapies have tried to harness the antitumor activity of the immune system by providing stimulation to these peritumoral lymphocytes (13). Other experiments have used fusion antibodies that target tumor antigens and contain a superantigen to activate T cells by linking the T-cell receptor to the MHC class II receptor on an antigen-presenting cell (14). To date, immune therapies for lung cancer are experimental and have achieved limited success.

Chemokines, a structurally similar but functionally diverse family of molecules, are principally produced by, and target cells of, the immune system (15). Chemokines are also produced by a variety of normal tissues including lung tissue. These lung-derived chemokines are produced by pulmonary epithelium, especially in the setting of acute injury from infectious or chemical insult (16, 17). The effect of chemokines on other cells is usually dependent on the presence of chemokine receptors on those target cells. Such cell surface receptors have been found on a variety of cells including endothelial, neuronal, smooth muscle, and epithelial cells (18).

In addition to normal tissues, a variety of tumors have been shown to secrete chemokines (19, 20). These molecules are now known to affect the behavior of cells in tumors. Chemokine receptors are expressed by breast cancer cells and are thought to be responsible for the distinct pattern of metastasis in breast cancer (21). Chemokines and their receptors have also been implicated in the migration of ovarian cancer cells (22).

Some chemokines with ascribed “antitumor” activity, particularly RANTES, IP-10, MCP-1, and MIP-3α, have been used in animal trials in an attempt to induce tumor-specific immune responses (23, 24). In NSCLC, chemokines secreted by the tumor cells are responsible for the infiltration of macrophages (25). Sharma et al., (26) has shown that secondary lymphoid organ (SLO) chemokine reduced tumor burden and increased CD4, CD8, and dendritic cell infiltrates in the tumors when injected into mice with metastatic pulmonary adenocarcinoma.

The aim of this study was to identify differences in gene expression between lung adenocarcinomas that correlate with ALR. Global gene expression in stage I tumors was examined using oligonucleotide microarrays. Differences in tumor cell expression of genes with known chemotactic, or immune-activating potential, may provide additional avenues for immune therapy in the future, or markers for prognosis that may direct therapy and more accurately stratify patients based on risk profile.

Sequential patients seen by the General Thoracic Surgery Section at the University of Michigan Hospital between May 1994 and July 2000 for resection of stage I or stage III lung adenocarcinoma were evaluated for inclusion in this study. Consent was received from all of the patients and the project approved by the Institutional Review Board. Patients’ medical records were reviewed and patient identifiers removed. Tumor and uninvolved lung tissue were obtained immediately at the time of surgery and transported to the laboratory in DMEM (Life Technologies, Inc., Gaithersburg, MD) on ice. Tumor samples were obtained from the advancing edge of resected lung carcinomas. A portion of each tumor tissue was embedded in OCT (Miles Scientific, Naperville, IL) for cryostat sectioning, frozen in isopentane cooled to the temperature of liquid nitrogen, and stored at −80°C. Hematoxylin-stained cryostat sections (5 μm), prepared from tumor pieces to be used for mRNA isolation, were evaluated by a study pathologist (T. J. G.) for comparison with H&E sections made from paraffin blocks of the same tumors. All of the stage I specimens included in the study had a minimum 70% tumor cellularity. The study pathologist diagnosed the tumors that had ALR from those that did not. The pathological diagnosis of ALR was made without knowledge of the gene expression data. Demographic and follow-up data were collected on all of the patients. Survival, as measured in months, was calculated from the time of resection to the time of latest follow-up or death.

At least two of the original H&E-stained slides of each tumor were evaluated for the presence of lymphocytic response. Tumors were divided by the study surgical pathologist (T. J. G.) into two broad categories based on the absence or presence of a significant peritumoral and intratumoral lymphocytic infiltrate that was readily apparent at low magnification (ALR− versus ALR+).

RNA isolation and oligonucleotide microarray analysis was carried out according to the methods of Giordano et al., (27).

Trimming of the dataset was performed by excluding genes from analysis if the Affymetrix measure of their 75th percentile value was less than 100. Restriction of the dataset in this manner reduced the number of probe sets from 7129 to 4578. Tumors were divided into two groups: those with ALR+ and those without (ALR−). Next, using Microsoft Excel, we performed a Student’s t test to compare the ALR+ and ALR− groups. To restrict the dataset further, a P of <0.02 was considered statistically significant. An average-linkage hierarchical clustering of an uncentered Pearson correlation similarity matrix was applied to the list of significant genes with the program Cluster, and figures were generated with the program TreeView (Eisen Lab Software; Stanford, Palo Alto, CA). From the clusters generated, genes that were overexpressed by the ALR+ group were further analyzed. With the use of the Locus Link website,⁴ genes in this group were subcategorized into functional classes. Genes with functional potential to incite ALR were chosen for validation. A correlation analysis between mRNA expression and protein expression was performed using the Spearman method. Survival analysis was performed using the Cox univariate proportional hazard regression model. Median gene expression for all tumors for a given gene was used as a cutoff value to designate “high expression” or “low expression” groups.

IHC was performed on tissue microarrays containing 90 lung adenocarcinomas to simultaneously examine the majority of cases in the study. From the diagnosed adenocarcinomas, the areas of the tumor that best represented the overall morphology were selected and a tissue microarray (TMA) block was constructed according to the method of Kononen et al., (28). Deparaffinized sections, along with sections from a commercially available normal-tissue tissue microarray block (NO50; Clinomics Laboratories, Pittsfield, MA) were microwave pretreated in citric acid buffer for antigen retrieval. The sections were incubated with blocking solution for 60 min at room temperature before being exposed to the primary antibodies. The antibodies used included: anti-MIP-1β, anti-RANTES (R&D Systems, Minneapolis, MN), and anti-IP-10 (Santa Cruz Biotechnologies, Santa Cruz, CA). The antibodies were incubated with the tissue microarrays overnight at 4°C. The immunocomplex was visualized by the immunoglobulin enzyme bridge technique using Vector ABC-kit (Vector Laboratories, Burlingame, CA), with 3,3-diaminobenzidine tetrachloride as the enzyme substrate. Sections were lightly stained with hematoxylin and permanently mounted.

Of these same tumor specimens, 25 had sufficient tumor material to perform protein-based assays. This included 12 in the ALR+ group and 13 in the ALR− group. Samples of tumor and normal lung distal to tumor were homogenized and sonicated in anti-protease buffer on recovery from the operating room. Specimens were centrifuged at 900 × g for 15 min, filtered through 0.45-μm Sterile Acrodiscs (Gelman Sciences, Ann Arbor, MI), and frozen at −70°C until thawed for assay by a specific antibody (MIP-1β, RANTES, IP-10). Proteins were quantitated using a modification of a double ligand method as described previously (29). All of the antibodies were polyclonal goat antihuman as listed above. The IP-10 antibody used for the ELISA was purchased from R&D Systems. Briefly, flat-bottomed 96-well microtiter plates (Nunc Immuno-Plate I 96-F) were coated with 50 μl/well containing the primary antibody [1 ng/μl in 0.6 m NaCl, 0.26 m H₃B0₄, and 0.08 n NaOH (pH 9.6)] for 24 h at 4°C and then washed with PBS (pH 7.5), 0.05% Tween 20 (wash buffer). Microtiter plate nonspecific binding sites were blocked with 2% BSA in PBS and incubated for 60 min at 37°C. Plates were rinsed three times with wash buffer. Fifty μl of tumor homogenate were added (neat and 1:10 dilution), followed by incubation for 1 h at 37°C. Plates were washed three times, 50 μl/well biotinylated polyclonal rabbit anti-IP-10 [3.5 ng/μl in PBS (pH 7.5), 0.05% Tween 20, and 2% FCS] were added, and plates were incubated for 45 min at 37°C. Plates were washed three times, streptavidin-peroxidase conjugate (Bio-Rad Laboratories, Richmond, CA) added, and the plates incubated for 30 min at 37°C. Plates were washed three times and chromogen substrate (Bio-Rad Laboratories, Richmond, CA) added. The plates were incubated at room temperature to the desired extinction, and the reaction terminated with 50 μl/well H₂SO₄ (3 m) solution. Plates were read at 490 nm in an automated microplate reader (Bio-Tek Instruments, Inc., Winooski, VT). Standards were dilutions of IP-10 from 100 ng/ml to 1 pg/ml (50 μl/well). This protocol consistently detected protein (MIP-1β, RANTES, IP-10) concentrations greater than 50 pg/ml in a linear fashion. Tissue samples were run in parallel for total protein content (Pierce, Rockford, IL), and results were expressed as nanograms of protein per milligram.

Sixty-three stage I adenocarcinomas were analyzed using the Affymetrix oligonucleotide arrays that interrogates ∼6800 genes and included tumors of various cell differentiations and derivations. Tumors were stratified into 27 with ALR+ and 36 ALR−. Pathological and demographic data for the two groups are represented in Table 1. The presence of ALR was the only other clinical or pathological variable in this patient cohort that was even marginally associated with survival (P = 0.06). Trimming of the data was performed to remove genes that showed low or no expression in the lung tumor samples and resulted in 4578 genes for further analysis. The Student’s t test was then used to compare the average expression of these genes between tumors that were ALR+ and ALR− tumors; 303 were found to significantly differ between ALR+ and ALR− tumors, even when a fairly stringent P of ≤0.02 was used as a cutoff.

Hierarchical clustering analysis of the 303 genes that were found to be significantly different between ALR+ and ALR− revealed two main clusters of genes, with Cluster 1 representing the 146 genes overexpressed in the ALR+ tumors (Table 2) and Cluster 2 representing the 157 genes overexpressed in the ALR− tumors. The genes in both clusters were classified according to their function with the use of the Locus Link website.⁴ Among the 146 genes in Cluster 1, there were at least 40 genes that were directly related to lymphocytes (Fig. 1). Other genes in Cluster 1 included cell adhesion molecules, signal transduction proteins, metabolic proteins, cytostructural proteins, and oncogenes. Any of these classes of genes may have proved interesting for further study, however, in an attempt to identify genes that may represent chemotactic signals secreted by tumor cells to help incite an ALR, we focused on the three chemokines overexpressed in the ALR+ group in Cluster 1: SCYA4 (MIP-1β), SCYA5 (RANTES), and SCYA10 (IP-10). The expression of the corresponding receptors for these three chemokines was not significantly different between ALR+ and ALR− groups (CXCR3, CCR1, CCR3, CCR5). MIP-1β showed an average level of expression of 786.3 in the ALR+ group but a level of only 476.0 in the ALR− group, which was significantly different (P = 0.001). Similarly, RANTES was more highly expressed in the ALR+ tumors with an average expression level of 816.4 versus 494.1 in ALR− tumors (P = 0.01). IP-10 also showed a higher level of average expression in those tumors that had ALR: 713.8 versus 322.7 in the ALR− group (P = 0.013; Table 3).

Validation of the results obtained by an analysis of mRNA was attempted using quantitative ELISA; and the cellular localization of these proteins was determined using IHC. ELISA was used to validate 25 samples from the tumor set. This included 12 ALR+ tumors and 13 ALR− tumors. Using a standardized ELISA for quantitation of the protein, RANTES showed the highest overall protein levels and also showed higher protein levels in the ALR+ group of tumors (Table 3). MIP-1β also showed a higher level of protein expression in the ALR+ tumors, yet showed much lower levels than the level of RANTES expression. IP-10 protein expression was higher in the ALR− group although the IP-10 mRNA levels were higher in the ALR+ group. A Spearman correlation was used to compare the mRNA expression and the protein expression of MIP-1β, RANTES, and IP-10. RANTES expression in the ALR+ tumors was the most strongly correlated among the three genes and had the highest correlation between mRNA and protein (r = 0.38). IP-10 had a negative correlation between mRNA and protein for the ALR+ tumors (r = −0.19). Because of the sample size, none of the correlation coefficients were statistically significant.

MIP-1β, RANTES, and IP-10 protein expression was then examined using an immunohistochemical analysis of tissue microarrays containing the majority of the stage I tumors in this study. These arrays also contained samples of normal lung, squamous cell carcinoma, advanced-stage lung adenocarcinoma, and adenocarcinoma cell line samples. For all three of the proteins tested, there was little to no staining in the normal sample, with the exception of positive-staining alveolar macrophages (Fig. 2,A). The A549 lung adenocarcinoma cell line showed positive cytoplasmic staining for all three proteins. Similarly MIP-1β, RANTES, and IP-10 antibodies stained the cytoplasm of tumor cells, and these IHC results demonstrated that the expression of these gene products are localized to the adenocarcinoma cells. There was no appreciable staining for IP-10, RANTES, or MIP-1β within lymphocytes (Fig. 2, B–D).

Using univariate Cox modeling, MIP-1β, RANTES, and IP-10 mRNA expression values were tested as predictors of survival. Neither MIP-1β nor IP-10 was a predictor of survival; however, elevated RANTES expression correlated with an improved survival. The median value of RANTES expression for all of the stage I tumors that were tested was 515. This median value was used to stratify patients into “high” or “low” expression groups. Patients in the low-expression group, whose tumors had a RANTES mRNA expression level below the median value, had a higher mortality (P = 0.002; Fig. 3). One patient, whose tumor had a RANTES expression level of 811, died before 2 months, but his death was unrelated to his cancer. When this patient was omitted from the survival analysis, the analysis showed a more dramatic difference in survival based on the RANTES level of mRNA expression (P = 0.0001).

ALR is known to be a favorable prognostic factor in patients with NSCLC. In this study, we examined 63 stage I patients and demonstrated that patients with ALR do indeed have improved survival. Oligonucleotide microarrays revealed 303 genes that significantly separated ALR+ and ALR− tumors. These differences are not attributable to tumoral differentiation or derivation, and the two study groups have similar demographics. In fact, there were more patients with poorly differentiated tumors in the ALR+ group, but this group still had a trend toward improved survival (P = 0.06). There were 146 genes that were overexpressed in the tumors that had ALR. From these 146 genes, the three chemokines were chosen for further analysis because this family of molecules is a known group of chemotactic cytokines and offered a logical signal for ALR. Immunohistochemical analysis showed that these gene protein products were expressed by the tumor cells themselves and not by infiltrating WBCs. RANTES consistently showed more intense staining in the ALR+ group. Protein quantification via ELISA showed a trend toward significance that MIP-1β and RANTES were overexpressed in ALR+ tumors. IP-10 was overexpressed at the mRNA level in the tumors but did not show a corresponding increase in protein expression by ELISA.

IP-10, MIP-1β, and RANTES, are known chemotactic cytokines. IP-10 is a chemotactic cytokine that can activate the immune system but that also inhibits angiogenesis. Monocytes/macrophages, T-lymphocytes, endothelial cells, fibroblasts, and keratinocytes produce and secrete IP-10 (30). As an α-chemokine, IP-10 is primarily chemotactic for Th-1 lymphocytes and macrophages but has little effect on basophils or eosinophils (31). From our data, the mRNA expression of IP-10 was higher in the ALR+ group; however, this pattern was not supported by the protein expression as evidenced by the ELISA data. This result may have been influenced by the relatively small sample size tested by ELISA. IHC staining of all stage I tumors did show a higher number of strongly staining tumors in the ALR+ group; however, there was no correlation between the strength of staining and the mRNA level. One explanation for this discrepancy may be that, if only a small number of cells account for the high mRNA expression in a given tumor, this difference in expression may not be reflected in the protein staining. Another possibility is that it is difficult to determine quantitative information regarding protein expression levels using IHC alone. IP-10 may also be induced because of some interaction between lymphocytes and tumor cells. IFNγ up-regulates IP-10 expression. As evidenced by the array data, IFNγ levels in both ALR+ and ALR− tumors are very low. IP-10 is not a product of the lymphocytes themselves, as indicated by the IHC results. The presence of a significant amount of lymphocytes in a tumor in the ALR+ tumor group, as well as the tumor/immune system interplay, may account for the higher level of IP-10 message and may potentially account for the difference in expression between the ALR+ and ALR− groups. In a murine model, IP-10 has been implicated as the key player in the tumor-protective immunity conferred by IL-12 by affecting Th-1 type CD8⁺ cells (32).

MIP-1β is produced by mononuclear cells, endothelial cells, and mast cells. It functions as a chemotactic and activating factor of the immune system and serves to inhibit the CD4+/CCR5-mediated HIV cell entry (30). MIP-1β is a β chemokine and affects all leukocytes except neutrophils (31). In one study, MIP-1β was implicated as the major chemotactic cytokine responsible for macrophage chemotaxis and activation during acute lung injury (15). MIP-1β was also shown to be produced by NSCLC and was highly correlated with macrophage chemotaxis (25).

RANTES is another known chemotactic cytokine that is produced by many cell types including T-lymphocytes, monocytes, platelets, eosinophils, epithelial cells, dendritic cells, and mast cells (30). In cell cultures of bronchial epithelial cells infected with influenza A, the transcription and translation of RANTES is increased within 24 h after infection (33). In vivo studies have shown the release of RANTES in lung to be a response to a variety of noxious stimuli including bacterial and viral infection, and chemotherapy (34, 35, 36). Although it typically works through the CCR5 receptor, multiple RANTES molecules can form a complex that can directly stimulate T cells, outside of the usual receptor (37). This direct stimulation is a reminder that RANTES, as do other chemokines, exerts its effects in a dose-dependent fashion. Arenberg et al. (25) showed that NSCLC cells produced RANTES but that RANTES levels did not correlate significantly with macrophage chemotaxis. RANTES, like MIP-1β, is a β chemokine and affects all leukocytes except neutrophils (31). RANTES has been used as a prognostic indicator in both breast and cervical cancers (38). However, in contrast to the findings of our study, higher levels of RANTES in those malignancies are correlated with a poor outcome. The reason for this difference in tumor behavior is unknown and may be worthy of future study.

In viewing this study in the context of previous studies, it is very possible that MIP-1β and RANTES could be signals to incite an ALR. In the tumors that were ALR+, both RANTES and MIP-1β mRNA were significantly increased. Similarly, the protein levels of these gene products, as confirmed by ELISA, were increased in the tumors with ALR. Immunohistochemical analysis shows that RANTES and MIP-1β are produced by NSCLC tumor cells and are not associated with lymphocytes.

When the types of inflammatory cells that infiltrate lung tumors are considered, RANTES is seen as an even more plausible signal for ALR. Controversy regarding the type of cells that infiltrate lung tumors exists in the literature. There are reports that both Th-1 and Th-2 profiles exist (39, 40). One study that looked at the lymphocytes themselves concluded that the predominant peritumoral lymphocyte in NSCLC is an activated memory T cell that is capable of synthesizing Th-1 cytokines (4). RANTES was first described as a chemoattractant for memory T cells in vitro (41). Other more recent data suggest that, in certain lung tumors, the CD4+, CD25+ T cells predominate; these may function to depress the host immune system and allow for tumor progression (42). One possible future direction for investigation would be to try to further characterize the T-cell population in each tumor and correlate the various T-cell profiles to the gene expression profile of each tumor to determine a more causal relationship among chemokine expression, T-cell population, and tumor behavior. Our data suggest that RANTES is a major chemotactic signal for infiltrating lymphocytes and is produced by NSCLC cells.

The proper antitumor cytokine milieu is surely the result of lymphocytes that recognize the danger of tumor tissue and, in turn, start an inflammatory cascade to destroy the tumor. RANTES expression by the tumor itself may be one possible event to aid in the initiation of this inflammatory cascade in lung adenocarcinoma. Although the Th-1 (IL-2 and IFNγ) and Th-2 (IL-4, IL-5, IL-9, and IL-13) representative cytokines are on the oligonucleotide arrays, their expression levels were found to be extremely low (data not shown). In addition, there was no discernable difference in the expression of any of these cytokines between the ALR+ and ALR− groups. Given these limitations, the prevalence of a Th-1 or Th-2 profile in the peritumoral lymphocytes of NSCLC tumors could not be determined.

If RANTES and MIP-1β represent signals needed to incite ALR, it is unclear as to why some tumors produce these cytokines and others do not. MIP-1β is located at 17q21 and RANTES is located at 17q11.2-q12. Chromosome 17p is a known region of loss in NSCLC that also includes the region in which the p53 tumor suppressor gene resides. It is possible that a portion of chromosome 17 could be lost in those tumors with lower levels of RANTES and MIP-1β. Alternatively, these differences may reflect regulation of gene transcription.

Regardless of the mechanism for the differential expression of these genes, it is clear that, in the case of RANTES, overexpression of this gene is associated with improved prognosis in NSCLC. Patients with a lower expression of this gene had an increased mortality using Cox modeling. If RANTES is a key signal to help promote chemotaxis of lymphocytes to a tumor, it may have potential therapeutic benefit. Irrespective of its therapeutic potential, RANTES could be considered a prognostic factor and may help to further stratify patients with stage I NSCLC. In recent studies, the use of genetic profiling and gene microarray technology has yielded families of genes that have delineated the various phenotypes of lung cancer as well as stratifying high- and low-risk expression profiles (43, 44, 45). Like those studies, this study also has been able to show a global difference in gene expression profiles between stage I tumors that are ALR+ and those that are ALR−. In concentrating on and further investigating genes that may play a role in recruiting lymphocytes, this study has shown RANTES to predict survival in stage I patients. Recognition of high-risk stage I patients may afford the use of adjuvant therapies in an attempt to positively impact survival in NSCLC patients.

We extend special thanks to Drs. S. L. Kunkel, J. F. Moran, and S. A. Moran for their review of the manuscript and their suggestions.