INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Non-small-cell lung cancer (NSCLC) exhibits a variable clinical phenotype, even in those patients with apparently limited stage disease who undergo resective surgery. Numerous reports suggest that a variety of tumor cell markers predict survival in patients who have undergone surgery (1). However, some reports conflict, and, in general, the associations do not appear sufficiently strong to be of value in formulating clinical treatment plans. Assuming that the putative biological markers are independent variables, multiple markers might be more informative than any single marker. A few published reports support such a possibility (2, 3). To further test this hypothesis we evaluated seven biological markers in resected NSCLC. One or more previous reports suggests that each of these markers might be useful in prognosis. These include the presence of K-ras (4, 5) and p53 gene mutations (6, 7), as well as expressions of the p53 protein (8, 9), the bcl-2 protein (10, 11), the c-erbB-2 protein (12, 13), and the epidermal growth factor receptor or c-erbB-1 protein (14). We also assessed an antigen in the H/Le^y/Le^b complex by immunostaining with the MIA-15-5 antibody. Antigen overexpression in NSCLC strongly predicted poor prognosis in a single prior study (15).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Patients and Tissue Samples

The study involved 101 consecutive patients with NSCLC who underwent exploratory thoracotomy with curative intent at the Minneapolis VA Medical Center between July, 1990, and April, 1995. Sections of the primary tumor from each case were snap-frozen in liquid isopentane and stored at 60° C, and other samples were preserved as formalin-fixed, paraffin-embedded sections. None of the patients received radiation or chemotherapy prior to surgery. The postsurgical pathological stage of each tumor was classified according to the tumor-node-metastasis (TNM) classification system (16). Survival was assessed from review of medical charts and computerized databases.

Histopathological and Immunohistological Examination

Diagnosis of NSCLC and categorization as to cell type were based on conventional morphological criteria. All immunostains were performed on paraffin-embedded sections according to the following protocols. Appropriate positive and negative controls were included with each set of stains.

Formalin-fixed, paraffin-embedded sections were cut at 5 microns, mounted on silane-coated slides, and baked overnight at 57° C. Sections were then deparaffinized in Americlear (Baxter, McGaw, IL) and rehydrated through graded alcohols into distilled water. Endogenous peroxidase was quenched by immersion for 10 min in phosphate buffered saline (PBS), pH 7.4, containing 0.3% hydrogen peroxide. Slides to be stained for c-erbB-1 protein were pretreated with pronase (Sigma Chemical Co., St. Louis, MO), 1 mg/ml in Tris-HCl buffer, pH 7.4, for 5 min at 37° C. Slides to be stained for p53 or bcl-2 protein were immersed in Coplin jars filled with 10 mM citrate buffer, pH 6.0, heated to boiling in a 6 L pressure cooker (Decor USA, Palatine, IL) at 12 psi, allowed to boil for 10 min, and then left to cool in the citrate buffer for 20 min at room temperature. No pretreatment was applied to slides to be stained for c-erbB-2 protein or M15-5, or to negative controls. After a rinse in PBS, slides were incubated overnight with one of the primary antibodies described below or with non-immune mouse ascites (Sigma), diluted 1:100, as a negative control. Subsequent steps utilized the Vectastain mouse Elite ABC immunoperoxidase kit (Vector Laboratories, Burlingame, CA), according to manufacturer's instructions, except that for MIA-15-5 antigen, biotinylated goat anti-mouse IgM (diluted 5 µl/ml PBS; Vector Laboratories), was substituted for the regular secondary reagent (biotinylated horse antimouse IgG, 5 µl/ml PBS). Color development was accomplished by a 5-min incubation with a solution of 0.22 mg/ml diaminobenzidine tetrahydrochloride (PolySciences, Warrington, PA) containing 0.003% hydrogen peroxide. Development was terminated by immersion in tap water, followed by color enhancement with osmium tetrachloride. Slides were counterstained with hematoxylin, dehydrated through graded alcohols into Americlear, and coverslipped with Permount (Fisher, Pittsburgh, PA).

Murine monoclonal IgG antibodies utilized in this study were those to c-erbB-1 protein (clone 31G7, used at 0.9 µg/ml; Zymed, South San Francisco, CA); c-erbB-2 protein (clone TAB-250, used at 0.5 µg/ml; Zymed); bcl-2 protein (clone 124, used at 1.4 µg/ml; Dako, Carpinteria, CA); and p53 (clone p1801, used at 0.1 µg/ml; Oncogene Research Products, Cambridge, MA). MIA-15-5 (gift from Dr. Sen-itiroh Hakomori) is a mouse IgM monoclonal, supernatant diluted 1:10 in PBS.

The same blinded observer (G.A.N.) assessed all immunostained sections according to these criteria:

p53 protein: 0 = no nuclear staining; 1+ = < 50% nuclei stained; 2+ = > 50% nuclei stained.

All others: 0 = < 5% cells stained; 1+ = weak staining of > 5% cells; 2+ = intermediate-to-strong staining of 5-20% cells; 3+ = strong staining of > 20% cells.

Genetic Analyses

We utilized single-strand conformational polymorphism analysis of polymerase chain reaction products (PCR-SSCP) to detect mutations in exons 5 through 8 of the p53 gene and in codons 12 and 13 of the K-ras gene (17). Genomic DNA was extracted from several 10 µm sections of frozen tumor using Puregene kits (Gentra Systems, Minneapolis, MN). Genomic DNAs (0.2 µg) were amplified by PCR with primer sets purchased from Clontech, Palo Alto, CA, closely following their suggested protocol. Products ranged in size from 139 to 211 base pairs.

p53

exon 5 sense: 5'-CTC TTC CTG CAG TAC TCC CCT GC-3'

anti: 5'-GCC CCA GCT GCT CAC CAT CGC TA-3'

exon 6 sense: 5'-GAT TGC TCT TAG GTC TGG CCC CTC-3'

anti: 5'-GGC CAC TGA CAA CCA CCC TTA ACC-3'

exon 7 sense: 5'-GTG TTG TCT CCT AGG TTG GCT CTG-3'

anti: 5'-CAA GTG GCT CCT GAC CTG GAG TC-3'

exon 8 sense: 5'-GGA CAG GTA GGA CCT GAT TTC CTT AC-3'

anti: 5'-TGC ACC CTT GGT CTC CTC CAC-3'

K-ras

codons 12-13 sense: 5'-GGC CTG CTG AAA ATG ACT GA-3'

anti: 5'-TTC CTA CAG GAA GCA AGT AC-3'

The amplified DNA products thus obtained were heated to 95° C for 5 min followed by rapid cooling, and they were then subjected to electrophoresis on nondenaturing polyacrylamide gels to detect mutant single stands (18). Products were visualized with silver stains (19). Optimal electrophoretic conditions were established for exon 5 (12% polyacrylamide, 275 V, 3.5 h, room temperature), exon 6 (15% polyacrylamide, 150 V, 18 h, 4° C), exon 7 (15% polyacrylamide, 275 V, 3.5 h, 4° C), exon 8 (15% polyacrylamide, 275 V, 18 h, 4° C) of the p53 gene and for the K-ras gene (12% polyacrylamide, 275 V, 3.5 h, room temperature) by analysis of cell lines with defined mutations.

Statistical Analysis

Associations between categorical variables were analyzed by ² analysis. Confidence intervals were calculated for the differences in sample proportions. Survival curves were calculated by the Kaplan-Meier method and survival differences were compared with the log rank test. A Cox proportional hazards model, incorporating pathological stage and markers as covariates, was used to determine the independent association between these variables and survival. Calculations were performed using SPSS^® for Windows (release 6.1; Chicago, IL). All p values are two-tailed.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

All 101 subjects included in the study were men and their average age at the time of operation was 67 yr (Table 1). Tumor cell types included 51 adenocarcinomas, 44 squamous cell carcinomas, and six large cell, undifferentiated carcinomas. The number of cases by pathological stage were: stage I = 61, stage II = 21, stage IIIA = 15, stage IIIB = 3, and stage IV = 1. As expected, survival was strongly related to pathological stage with p < 0.001 (Figure 1).

View larger version (20K): [in this window] [in a new window]   Figure 1.   Kaplan-Meier estimates of survival stratified by postsurgical pathological stage. The three patients with Stage IIIB disease and the one patient with Stage IV disease are excluded. The difference among stages is significant (p < 0.001).

Of the 101 cases, we identified 40 as having a p53 mutation by PCR-SSCP. These included 23 of 51 adenocarcinomas, 16 of 44 squamous cell carcinomas, and one of six large cell carcinomas. Twenty-one of the p53 mutations were on exon 5, nine on exon 6, five on exon 7, and five on exon 8. K-ras mutations were closely associated with cell type with positive findings in 17 of 51 adenocarcinomas, zero of 44 squamous cell carcinomas, and two of six large cell carcinomas.

As shown in Table 2, expression was variable for each of the antigen markers that was evaluated. No significant relationship existed between antigen expression and histological cell type for any of the markers. Overexpression of p53 protein was significantly related to the presence of a p53 gene mutation. Five of 28 tumors with a p53 immunostain grade of 0 were positive for a p53 mutation, seven of 25 with an immunostain grade of 1+ were mutation positive, and 28 of 48 with an immunostain grade of 2+ were mutation positive (p < 0.001 by ² analysis). For purposes of the subsequent analyses we reclassified each immunostain as being lightly stained or heavily stained, choosing cutoff values to provide roughly equal numbers in each category for all markers. After reclassification the lightly stained to heavily stained ratio for p53 protein is 53/48, for MIA-15-5 is 31/70, for bcl-2 is 59/42, for c-erbB-1 is 35/66, and for c-erbB-2 is 43/58.

We analyzed the relationships between survival and molecular markers by confidence intervals and by the more conventional Kaplan-Meier estimates. Figure 2 shows the means and 95% confidence intervals of the survival difference at two years for each marker. Follow-up data were available for all subjects at that time and 55 of the 101 subjects remained alive. By those analyses, p53 mutation predicted a significantly shorter survival, whereas overexpression of both MIA-15-5 antigen and c-erbB-2 protein predicted longer survivals. Given the multiple comparisons, a more stringent criterion to judge statistical significance might be appropriate. When calculated at the 99% level, all confidence intervals intersect 0% survival difference, meaning that none of the differences would be judged significant.

View larger version (17K): [in this window] [in a new window]   Figure 2.   Means and 95% confidence intervals for the differences in survival at two years for each of the molecular markers. A positive value indicates longer survival in the presence of a gene mutation or with antigen overexpression.

Comparison of survivals from Kaplan-Meier estimates showed that only MIA-15-5 antigen expression (p < 0.001) and p53 gene mutation (p < 0.05) significantly correlated with postsurgical survival. The presence of a p53 mutation predicted worse survival, while overexpression of MIA-15-5 antigen predicted a better survival (Figure 3). If we again make allowance for multiple comparisons, only the difference in MIA-15-5 antigen expression would be judged significant. Neither a K-ras mutation nor c-erbB-2 protein overexpression was predictive of survival, even when analyses were restricted to adenocarcinomas. In a multivariate analysis, postsurgical pathological stage entered the model as the best single predictor of survival (p < 0.001). Of the molecular markers evaluated, only MIA-15-5 antigen remained in this model at a significant level (p < 0.01).

View larger version (19K): [in this window] [in a new window]   Figure 3.   Kaplan-Meier estimates of survival statified by expression of MIA-15-5 antigen. Class 0 represents least expression and class 4 represents most expression. The difference among classes is significant (p < 0.001).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

We found that multiple NSCLC tumor markers provided little prognostic information beyond that from surgical stage alone in a consecutive series of 101 patients from this institution. We were unable to confirm claims previously made for several markers, and, in the case of MIA-15-5 antigen, our findings were directly contradictory. Whereas Miyake and associates reported that MIA-15-5 antigen overexpression was closely associated with poor survival (p < 0.001), we found a similarly strong statistical association but in the opposite direction (15). While population composition might account for some of the observed differences, particularly in the case of the MIA-15-5 antigen, we suggest that other factors are probably more important.

Methodology is probably not a major factor in these apparent discrepancies. We prepared all immunostains with appropriate positive and negative controls by standard methods. Dr. Hakomori kindly provided us with the same MIA-15-5 antibody that had been used in the study by Miyake and associates (15), and he confirmed that the antibody retained its biological activity. We procured other antibodies from reputable commercial sources. A single blinded observer evaluated all slides by pre-established criteria. Consistent with previous reports, each of the antigens assessed by immunostain was variably expressed in NSCLC.

The prevalences of mutations detected in our study cohort, 40% of the p53 genes in all cell types and 33% of K-ras genes in adenocarcinomas, were consistent with previously reported values (4). Under optimal conditions PCR-SSCP may only be about 90% sensitive (20). We may also have missed a few cancer-related mutations occurring in exons 4 and 9 of the p53 gene, in codon 61 of the K-ras gene, and in the H-ras and N-ras genes (21, 22). However, it is unlikely that small numbers of false negatives would have a major impact on our conclusions.

We suggest that chance might play a very important role in explaining the variability of published results. The confidence intervals shown in Figure 3 indicate that the precision of our sample estimates are relatively poor, and, at 2 yr, we might not have detected survival differences as large as 40%. Given this level of imprecision, it is not surprising that a series of studies of similar size might yield variable but not necessarily incompatible results.

The relationship of p53 protein expression to clinical features of NSCLC has been the subject of many published reports and it serves as example. We identified more than 20 such studies from a MEDLINE search, each of which contained between 19 and 271 cases. The authors of the majority of these articles, including the single largest study (2), concluded that overexpression of p53 protein adversely affects the clinical course. However, some of these studies used endpoints other than survival, several other studies reported no significant survival difference, and one relatively large study (n = 156) found that p53 protein overexpression predicted a substantially longer survival (23). Similarly variable results characterize other putative NSCLC markers, and we suggest that much of this variability can be attributed simply to chance.

Nonetheless, a preponderance of the articles showed an adverse effect of p53 protein expression on survival, and it can be argued that a formal meta-analysis would most likely show a statistically significant difference. This conclusion would presume an absence of publication bias, a well recognized phenomenon in which studies with statistically significant results were more likely to be published than those showing null effects (24, 25). We have no evidence for or against this factor, but the wide availability of tumor specimen banks and a relatively simple methodology suggest that all completed studies might not have been published. Failure to report all "negative" outcomes would bias results in favor of published "positive" outcomes, even though the latter might be only chance occurrences.

In two studies the authors concluded that p53 expression adversely affected adenocarcinoma but not other NSCLC cell types (26, 27). We found many other examples where conclusions about molecular markers and survival are based on subset analyses. To cite only a few, expression of blood group antigen A, but not other blood group antigens, favorably affected prognosis in NSCLC (28), the adverse impact of K-ras mutations was restricted to certain specific amino acid substitutions (29), c-erbB-2 protein expression adversely effected adenocarcinomas but not squamous cell carcinomas (12), and bcl-2 protein expression favorably affected survival in squamous cell carcinomas but not adenocarcinomas (10). In many of these reports the main effect of tumor marker on survival was either not statistically significant or was not reported. Indiscriminate subgroup analyses subverts conventional statistical rigor and it may seriously distort scientific objectivity (30).

In summary, we were unable to show that a battery of seven putative molecular markers in NSCLC had much prognostic value beyond that provided by pathological stage alone. The only marker with apparent value predicted survival in a direction opposite to that shown in the one prior report. Our results do not preclude the possibility that the particular markers chosen for this study contain some prognostic information. More definitive evidence requires a far larger study in which specific hypotheses and statistical methods have been clearly described before the study was undertaken. However, the magnitudes of any such effects are likely to be small. If many hundreds of cases are required to discern an effect, one might reasonably conclude that the effect is unlikely to be of major clinical significance.