Macrophage Migration Inhibitory Factor and CXC Chemokine Expression in Non-Small Cell Lung Cancer

Normal Human Lung and NSCLC Tumor Tissue.

Tissue specimens were obtained from consecutive individuals undergoing thoracotomy (n = 87) for suspected NSCLC. All of the patients provided informed consent, and tissue was obtained with approval of both the University of Michigan Tissue Procurement Core and Institutional Review Board. Standardized punch biopsy samples of tumor and normal lung distal to the tumor were homogenized and sonicated in Complete antiprotease buffer (Boehringer) on recovery from the operating room. Specimens were centrifuged at 900 × g for 15 min, filtered through 0.45-μm Sterile Acrodiscs (Gelman Sciences, Inc., Ann Arbor, MI), and frozen at °70°C until thawed for assay by a specific ELISA. Additionally, an adjacent portion of tumor was fixed in 10% zinc formalin and embedded in paraffin for histological analysis. University of Michigan Hospital pathologists determined the final histological diagnosis. For each patient, smoking status was determined as current (smoking within the last 6 months), former (quit for >6 months), or never. Only cases of NSCLC were included in our analysis (squamous cell carcinoma, n = 29; adenocarcinoma, n = 50; other non-small cell or undifferentiated, n = 8). An H&E-stained section of each specimen of normal lung tissue was examined and used as a control only if the tissue was devoid of significant inflammation or malignant cell invasion (n = 60).

ELISA.

Purified and biotinylated antibody for MIF, VEGF, ENA-78/CXCL5, and GROα/CXCL1 ELISAs were obtained from R&D Systems, Inc. (Minneapolis, MN). Antibodies for GRO-γ (CXCL3) were obtained from PeproTech Inc. (Rocky Hill, NJ). Purified and biotinylated antibodies for IL-8 (CXCL8) were produced as described previously (19) . Aqueous tumor and normal lung extracts were subjected to specific ELISA according to our protocol published previously (17) . All of the tissue samples were run in parallel for total protein content (Pierce, Rockford, IL), and results were standardized by expressing values as ng of cytokine per mg total protein.

Because the angiogenic CXC chemokines are all of equivalent potency in terms of angiogenic activity (20) and mediate their angiogenic effects via the same receptor (CXC receptor 2; Ref. 21 ), we generated a combined “ELR score” for each of the tumors, representing the sum of the levels of all of the measured CXC chemokines bearing the ELR motif (glu-leu-arg, shared by the angiogenic CXC chemokines). That is total levels of IL-8/CXCL8, ENA-78/CXCL5, GROα/CXCL1, and GROγ/CXCL3.

Angiogenesis Bioassays.

Endothelial cell chemotaxis was performed as described previously (17 , 22) , using various concentrations of MIF from 0.1 to 100 ng/ml as the angiogenic stimulus (120 μl). Each experiment was accompanied by a positive control (VEGF 100 ng/ml) and a negative control (nonconditioned medium) to account for variable chemotaxis from one assay to the next. Results were expressed as the mean number of endothelial cells per high-powered field ± SE. Each sample was assessed in triplicate wells. Each experiment was performed three times, and the results combined as the chemotactic index (ratio of cell migration in the experimental condition to negative control).

Microcarrier bead assays were performed using the method of Fife et al. (23) . Briefly, 100 mg of Cytodex-3 microcarrier beads (Sigma) were washed three times in 20 ml PBS and resuspended in 10 ml of serum-free endothelial cell medium. Eight ml of a cell suspension containing HMEC-1 cells at 10⁵ cells/ml were suspended with 440 μl of the bead slurry. This suspension was placed in a siliconized (Sigmakote; Sigma) glass Petri dish, and the cells were allowed to adhere to beads overnight. Fresh medium was added, and the beads were allowed to incubate until the beads were coated with cells. Forty μl of beads were then added to 0.5 ml of a fibrinogen suspension (Sigma; 2.5 mg/ml in serum-free HMEC-1 medium), and placed into a well of a 24-well plate. The fibrinogen was then polymerized with thrombin (Sigma; 0.625 u/ml) and aprotinin (Sigma; 0.15 IU/ml). The bead suspensions were allowed to clot for 1 h, and test medium was added (VEGF 10 ng/ml, MIF 5 or 50 ng/ml, or negative control medium, 0.5 ml/well for a total volume of 1 ml/well). Bead formation was assessed at 4 days by counting the number of tube projections per bead. Data represents the average of 5 beads/well, and 3 wells/condition. The experiment was performed twice.

Immunohistochemistry.

A portion of each tumor specimen was fixed in 10% zinc formalin. Paraffin-embedded tissue sections were dewaxed in xylene and rehydrated through graded concentrations of ethanol. Antigen retrieval was performed with 0.1% trypsin/CaCl₂ for 30 min at 37°. Samples were then stained for MIF using a modification of our technique described previously (24) . In brief, using peroxidase staining kits (VectaStain Elite, ABC Peroxidase kit for goat primary antibody; Vector Laboratories, Burlingame, CA), nonspecific binding sites were blocked. Sections were then overlaid with a either antihuman MIF or control goat IgG (0.2 μg/ml in PBS; R&D Systems) for 1 h. Slides were then rinsed and overlaid with secondary biotinylated antibody (Vectastain ABC kit) and incubated for 30 min. After washing twice with PBS, slides were overlaid with ABC reagent, and incubated for 30 min. 3,3′-Diaminobenzidine substrate (Vector) was used to localize antigenic MIF, yielding a red-brown product. After optimal color development, sections were immersed in sterile water, counterstained with Mayer’s hematoxylin for 5 min, and coverslipped with aqueous mounting solution. Sections were photographed with a Spot RT color camera coupled to a Nikon microscope, and captured on a Dell Optiplex computer using Adobe Photoshop software.

Quantitation of Vessel Density.

Quantitation of vessel density was performed using the method described previously (24) . In brief, tissue sections were dewaxed and rehydrated. Slides were stained for endothelial cells using antibody to Factor VIII-related antigen conjugated to horseradish peroxidase (EPOS staining reagent; DAKO Corporation, Carpinteria, CA) for 1 h at room temperature. 3,3′-Diaminobenzidine, (Vector) was used for localization of Factor VIII-related antigen. After optimal color development, sections were immersed in water and coverslipped. Tumor specimens were scanned at low magnification (×40) to identify vascular hot spots. Areas of greatest vessel density were then examined under higher magnification (×400) and counted. Each distinct area of positive staining for Factor VIII-related antigen was counted as a single vessel. Vessel density for each patient tumor was expressed as the average number of vessels per high power (×400) field from three nonoverlapping microscopic fields.

Statistical Analysis.

All of the normal lung tissue samples were examined under the microscope after paraffin embedding without knowledge of the levels of cytokines in the corresponding tissue homogenate. Any normal tissue containing significant inflammation (n = 32) or unsuspected tumor invasion (n = 1) on histological examination was excluded from analysis of cytokine levels. Results for MIF were compared between tumor and paired normal tissue using a nonparametric analysis, the Mann-Whitney t test, because the values of MIF in tumors did not follow a normal distribution. Wilcoxon signed rank test was used to compare all of the tumors to all of the normal tissues in a nonpaired fashion (given that not all of the tumors had a paired normal lung specimen from the same patient). To determine the correlation between cytokine levels with one another and with vessel density, we performed a correlation analysis and a Fisher’s r to z translation to determine the degree of statistical significance of the correlation using the StatView statistical software package. Patients were stratified into groups with high or low levels of MIF or VEGF based on tumor-associated levels above (high) or below (low) the median for that particular cytokine. Univariate Cox proportional hazards modeling was used to determine the relation of the following variables to disease-free survival, based on their association with angiogenesis: MIF, individual CXC chemokines, total CXC chemokine levels (the ELR score), and VEGF. Kaplan-Meier survival plots were generated to describe the disease-free survival of the cohorts of patients defined by the presence of high MIF or high VEGF, both cytokines being high, or neither being high.

MIF Levels Are Increased in Tumor Tissue Relative to Normal Lung.

We have shown recently that NSCLC cell lines express MIF constitutively and that MIF is responsible for inducing CXC chemokine-dependent angiogenic activity from monocytes in coculture. The resulting coculture-conditioned medium, when tested in angiogenesis bioassays, was markedly more angiogenic than medium generated by either monocytes or NSCLC cells cultured independently (17) . Many studies have shown that CXC chemokines are major determinants of the angiogenic bioactivity in human NSCLC (24, 25, 26, 27, 28, 29, 30, 31) . Given these findings, we hypothesized that human lung tumor tissue would be associated with expression of MIF in situ. To address this question, we performed immunohistochemical staining for MIF in surgical specimens of human NSCLC. We found significant, specific staining of tumors for MIF in specimens of NSCLC (Fig. 1)⇓ . To investigate this additionally, we measured levels of MIF in aqueous tumor homogenates taken from 87 patients with lung cancer. Tumor-associated levels of MIF did not follow a normal distribution. Rather, it was apparent that there was a bimodal distribution, with approximately half the tumors having levels of MIF that were comparable with that found in normal lung tissue (n = 42), and half having levels of MIF markedly elevated above “normal” (n = 45; Fig. 2⇓ ). We found no significant difference among histological types (adenocarcinoma versus squamous versus poorly differentiated), age, sex, or smoking status (current, former, or nonsmokers; data not shown). We also measured levels of angiogenic factors (the CXC chemokines IL-8/CXCL8, ENA-78/CXCL5, GROα/CXCL1, and GROγ/CXCL3) and VEGF. Unlike MIF, levels of these cytokines in the tumors followed a normal distribution. Confirming our observations published previously, we found elevated levels of CXC chemokines in tumor tissue relative to normal lung. Levels of VEGF were also elevated in tumor tissue relative to normal lung tissue (Table 1)⇓ .

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Representative immunohistochemistry of MIF in human NSCLC tumors. Staining with goat antihuman MIF (left column) or negative control goat IgG (right column). A, adenocarcinoma (×40 original magnification). B, squamous cell carcinoma (×200 original magnification), and C, squamous cell carcinoma (×400 original magnification).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

MIF expression is increased in human lung cancer tumors relative to normal lung tissue. A, MIF levels expressed in ng MIF per mg total protein (P = 0.01 by nonparametric paired comparison, P = 0.03 by nonparametric unpaired comparison). B, frequency distribution of MIF expression in tumors (solid line) and normal lung tissue (dashed line) demonstrating the bimodal distribution of MIF levels among tumor tissue.

Previous publications have suggested that MIF can promote endothelial cell migration or proliferation (14 , 16) . We tested recombinant human MIF for in vitro angiogenic activity in two different assays, endothelial cell migration, and tube formation in fibrin gels. MIF did not demonstrate any angiogenic bioactivity in vitro either in endothelial cell chemotaxis assays (Fig. 3A)⇓ or in tube formation assays using endothelial cell-coated microcarrier beads embedded in fibrin gel (Fig. 3B)⇓ .

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

MIF is not angiogenic in vitro. Over a range of concentrations, MIF did not induce endothelial cell migration in Boyden chambers (A, composite data from three experiments) or tube formation in fibrin gels by endothelial cell-coated microcarriers (B, representative data from two experiments, and an average of 5 beads/well, and 3 wells/condition). In each experiment, VEGF served as a positive control; bars, ±SD.

MIF Levels Correlate with ELR CXC Chemokine Levels and Vessel Density.

The above findings suggest that MIF does not directly promote angiogenesis in vitro. However, given that MIF induces CXC chemokine-dependent angiogenic activity by monocytes in vitro (17) , we hypothesized that MIF expression by NSCLC tumors in vivo would be associated with higher levels of tumor-associated CXC chemokines. The ELR score, reflecting total levels of tumor-associated angiogenic CXC chemokines, correlated highly with the levels of MIF in the corresponding tumor (r = 0.76; P < 0.0001; Fig. 4⇓ ).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

MIF correlates with the ELR score. In each tumor, MIF levels are plotted against the ELR score. With all values taken together the correlation is strong (r = 0.72; P < 0.0001). However, the correlation is strengthened when only high MIF tumors are considered (in which MIF levels are higher than the median, 138 ng/mg; closed symbols; r = 0.79; P < 0.0001). Note, because of the scale, 1 tumor was omitted from the figure where the ELR score was 251 and the level of MIF was 3331 ng/mg.

To determine whether MIF levels correlated with vascular density, each tumor was subjected to staining with antibody to Factor-VIII-related antigen to delineate microvessels. We found a statistically significant positive, but weak, correlation between vessel density and levels of tumor-associated MIF (r = 0.23;P = 0.05; Table 2⇓ ; Fig. 5A⇓ ). Correlation analysis to determine the degree to which variation in cytokine levels between tumors accounted for the variability in vessel density was performed. Surprisingly, tumor-associated levels of VEGF had the weakest correlation with tumor vessel density (r = 0.07; P = 0.58; Table 2⇓ ). With the exception of GROγ/CXCL3, levels of individual CXC chemokines correlated positively and significantly with tumor vessel density, which confirms our findings published previously (Table 2⇓ ; Ref. 24 ). However, the strongest predictor of tumor vessel density was the ELR score (r = 0.47; P < 0.0001; Table 2⇓ ; Fig. 5B⇓ ).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

MIF and CXC chemokine levels correlate with tumor vessel density. Vessel density is plotted on the vertical axis as determined by staining for Factor VIII-related antigen (see “Materials and Methods”). A, MIF levels in the corresponding tumor homogenate. MIF correlates weakly with vessel density (r = 0.23; P = 0.05; open and closed symbols). When only tumors with MIF levels above those found in normal lung are considered, the correlation between MIF and vessel density is strengthened significantly (closed symbols only; r = 0.52; P = 0.005). B, the strongest correlate with vessel density is the total levels of angiogenic CXC chemokines (ELR score). Levels of VEGF correlated poorly with tumor vessel density (r = 0.07; P = 0.58; data not shown).

One possible interpretation of our data is that tumors may achieve their angiogenic requirement through a variety of pathways. For example, one tumor may predominantly secrete VEGF, whereas another may secrete one or more of the CXC chemokines or one of the many angiogenic factors not measured in our study (e.g., fibroblast growth factor or platelet derived endothelial cell growth factor). Yet another type of tumor may induce the expression of angiogenic factors by infiltrating stromal cells. An example of this would be expression of MIF by tumor cells, which leads to expression of angiogenic CXC chemokines from tumor-associated monocytes. If this were true, one would expect a tumor that exploits one angiogenic mechanism might not require the presence of another angiogenic mechanism. In this scenario, vascularity in tumors in which levels of MIF are low (similar to normal tissue for example) would depend on an alternative angiogenic strategy. In these low MIF tumors, one would not expect vessel density to correlate with levels of MIF. On the other hand, in tumors with high MIF levels, one might expect a stronger correlation between MIF levels and vessel density. Consistent with this, in tumors in which MIF levels are above the median (138 ng/mg, or higher than normal lung tissue), the correlation of MIF with vessel density is stronger (r = 0.42) and statistically significant (P = 0.02; Table 2⇓ ; Fig. 5A⇓ ). In this group, the correlation of MIF with the ELR score is also strengthened (r = 0.79; P < 0.0001; Fig. 4⇓ , closed symbols). In the low MIF cohort, VEGF still correlated poorly with vessel density (r = 0.12; P = 0.45).

MIF, ELR Score, and VEGF Levels Are Associated with Worse Prognosis.

In the course of follow-up of the cohort in this study (median, 16 months; range, 2–32), 27 of the 87 patients have had recurrence of their lung cancer. In a univariate Cox proportional hazards model, factors associated with recurrence of lung cancer after resection included MIF, the ELR score, and VEGF levels above than the median (Table 3)⇓ . When both VEGF and MIF were elevated above the median, disease-free survival was significantly worse (Fig. 6)⇓ .

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6.

Kaplan-Meier survival curves describing disease-free survival experience of patients with both VEGF and MIF levels below the median (n = 22; solid line), either VEGF or MIF levels above the median (n = 42; dashed line), or both VEGF and MIF levels above the median (n = 23; dotted line); last follow up visit = +, death = ○; P = 0.005. The relative risk of recurrence of patients whose tumors had both high MIF and high VEGF compared with both low was 2.54 (95% CI, 1.21–5.31; P = 0.01). For patients with tumors with both low MIF and low VEGF, the hazard relative to those with either high MIF or high VEGF was 0.26 (95% CI, 0.06–1.1; P = 0.07). The risk for recurrence with both high MIF and high VEGF compared with patients with only one variable high was 3.54 (95% CI, 1.51–6.94; P = 0.003).