Cell Adhesion Molecules, Vascular Endothelial Growth Factor, and Basic Fibroblast Growth Factor in Patients with Non–Small Cell Lung Cancer Treated with Chemotherapy with or without Bevacizumab—an Eastern Cooperative Oncology Group Study

Materials and Methods

In this randomized phase II/phase III trial, treatment assignments were designed to achieve balance between the two treatment groups using the following stratification factors: measurable versus nonmeasurable disease, prior radiation versus no prior radiation, prior weight loss of <5% versus ≥5%, and stage IIIB with pleural effusion versus stage IV or recurrent disease. The primary study end point was overall survival and has been reported (6). The protocol was approved by the institutional review boards of all participating institutions and was carried out in accordance with the Declaration of Helsinki, current Food and Drug Administration Good Clinical Practices, and local ethical and legal requirements.

Patients were randomized to receive paclitaxel (200 mg/m²) and carboplatin (AUC, 6 each) given i.v. on day 1 without (carboplatin + paclitaxel, PC) or with bevacizumab (15 mg/kg, i.v.) on day 1 (PC + bevacizumab, BPC). Chemotherapy was repeated every 21 days up to a total of six cycles, unless there was evidence of disease progression or patient intolerance. Patients randomized to BPC were continued on bevacizumab monotherapy every 3 weeks until evidence of disease progression or unacceptable toxicity.

The first objective of this correlative study was to determine if differences in the pretreatment levels of plasma VEGF predict response to chemotherapy ± bevacizumab. For the correlative study, 80 plasma samples would be available from each arm (first 160 consecutive patients enrolled on study). Subjects were divided into two groups based on their baseline VEGF levels using the median value as the cutoff. For this analysis, response was dichotomized into responders (best response of partial or complete response per RECIST criteria) and nonresponders. Within each of the VEGF groupings, a Fisher exact test with a two-sided 5% type I error rate was used to detect an association between response and treatment. With the planned sample size, this test would have 80% power to detect a difference in response rates of 15% on PC versus 46% on BPC. The secondary objective of this correlative study was to perform exploratory analysis of baseline levels and serial changes in ICAM, E-selectin, and bFGF. These factors are also categorized as high/low using median cutoffs for all analyses. Overall survival is defined to be time from randomization to death from any cause, censored at the date of last contact. PFS is time from randomization to documented progression per RECIST or death, censored at the date of the last documented disease evaluation for patients without a PFS event reported. The Kaplan-Meier method is used to estimate survival and PFS distributions. Log-rank tests are used for univariate comparisons of survival and PFS end points. Cox proportional hazards models are used to estimate hazard ratios in regression models. Tests for differences in treatment effects between marker groups use interaction Wald tests from logistic regression models (for response) and Cox models (for PFS and survival). All P values are two-sided. Analyses of individual factors used all cases with values for that factor, and models containing more than one factor included only cases with data on all factors in the model. Because of the exploratory nature of the analyses (other than the primary objective), no adjustment for multiple testing was made.

Plasma samples were collected from patients before cycle 1 and after completion of cycle 2 (precycle 3; termed hereon as “week 7” samples). Pretreatment samples were analyzed for levels of ICAM, E-selectin, bFGF, and VEGF using commercially available ELISA kits (R&D Systems). Same factors were measured at week 7 except for VEGF (repeated measures of VEGF in the setting of treatment with bevacizumab show consistently high levels due to binding of antibody to ligand and increase in half life.) Three milliliters of peripheral blood were drawn into citrated Vacutainer tube and mixed immediately by inverting the tube 10 to 15 times. Samples were then centrifuged within 30 min of collection for 10 min at 4°C at 3,000 × g. Plasma was removed and transferred to cryogenic storage tubes. Samples were stored immediately stored at −70°C. They were then shipped on dry ice to University Hospitals Case Medical Center (A. Dowlati) for analysis. Samples were run after the first thaw only, in duplicate, and the average was recorded. The lower limits of detection for ICAM, E-selectin, VEGF, and bFGF were 0.35 ng/mL, 0.1 ng/mL, 12.5 pg/mL, and 1.0 pg/mL, respectively. The intraassay coefficient of variation was 4.6%, 5%, 4%, and 4.3%, respectively.

Results

Measurement of factor levels at baseline and week 7.Table 1 summarizes the distribution of the measurements on these factors. The baseline ICAM levels are significantly higher on PC than BPC arm (P = 0.02, Wilcoxon rank sum test). The distributions of the baseline E-selectin, bFGF, and VEGF levels are not significantly different between the treatment arms. E-selectin shows a significant decrease from baseline at week 7 (P < 0.0001, Wilcoxon signed-rank test), and bFGF shows a significant increase from baseline (P = 0.004) in the combined sample (Supplementary Fig. S1). In both cases, the magnitude of the change is not significantly different between the treatment arms (Wilcoxon rank sum test). The change from baseline in ICAM is not significant.

Association between factor levels, clinical characteristics and response rate among all patients.Table 2 examines the association among factor levels at baseline with selected baseline clinical characteristics and with objective tumor response rate. Throughout, low and high levels are defined as less than or equal to the median value and more than the median. There are no correlations between any of the baseline factor levels, gender, age of ≥65 or <65, or performance status. Only ICAM shows a significant association with response. In patients with a low ICAM level, the response rate was 32% versus 14% in the high ICAM group (P = 0.02).

Association between factor levels and response according to treatment assignment.Table 3 examines response by treatment assignment (BPC or PC) separately for groups defined by low and high baseline levels. The primary end point of the correlative study which was to determine if baseline VEGF levels is predictive of response to chemotherapy ± bevacizumab was met. The logistic regression interaction test, which tests whether the effect of treatment is different within the low and high levels, has P = 0.04 for VEGF measurements, suggesting that patients with high baseline VEGF levels are more likely to have an increased probability of response with the addition of bevacizumab (33% response on BPC arm versus 7.7% on PC arm, P = 0.01) than those with low VEGF levels (28.6% response on BPC arm versus 29% on PC arm, P = not significant). The interaction tests are not significant for any of the other factors.

Association between baseline factor level and progression-free or overall survival among all patients.Figures 1 and 2 give survival and PFS by baseline factor levels. Baseline ICAM seems to be a significant prognostic factor for survival (P = 0.00005). This difference seems to be consistent for the two treatment arms (data not shown). Patients with a low baseline ICAM have a 1-year survival of 65% versus 25% for patients with high ICAM levels. Overall survival was not associated with any other factor. Baseline VEGF was, however, a significant predictor of PFS. Patients with a low VEGF level had a better PFS (median, 6.0 months) compared with patients with a high level (median, 4.5 months; P = 0.04). Other comparisons were not statistically significant.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Survival based on baseline factor levels.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

PFS based on baseline levels.

For survival, the variables baseline ICAM (estimated hazard ratio for high/low, 1.83; P = 0.001), performance status (1/0 estimated hazard ratio, 1.73; P = 0.005), sex (male/female estimated hazard ratio, 1.60; P = 0.01), bone involvement (estimated hazard ratio, 2.15; P = 0.00007), liver involvement (estimated hazard ratio, 1.80; P = 0.006), nonrecurrent (versus recurrent, estimated hazard ratio, 1.91; P = 0.01), and adrenal involvement (estimated hazard ratio, 1.75; P = 0.02) were all significant in a joint proportional hazards model. None of the variables baseline E-selectin, baseline bFGF, or baseline VEGF added significantly to this model.

For PFS, the variables VEGF (estimated hazard ratio, 1.53; P = 0.009), bone involvement (estimated hazard ratio, 1.48; P = 0.03), recurrent disease (versus nonrecurrent stage IV, estimated hazard ratio, 0.62; P = 0.03), and nonrecurrent stage IIIB (versus nonrecurrent stage IV, estimated hazard ratio, 0.46; P = 0.04) were all significant in a joint model. None of the variables baseline E-selectin, baseline bFGF, or baseline ICAM added significantly to this model.

Predictive value of baseline factor levels or changes in levels on the benefits of progression-free or overall survival with bevacizumab.Table 4 examines overall survival and PFS by treatment separately for groups defined by low and high baseline levels and for groups defined by low and high levels of changes from baseline (week 7 minus baseline levels). As can be seen in this table, the hazard ratio (PC/BPC) of death or disease progression is >1 in almost all subcategories, indicating the superiority of BPC over PC. There are two exceptions: patients with high baseline ICAM levels have a hazard ratio of 1.0 for PFS (95% confidence interval, 0.62-1.6), whereas those with low baseline ICAM levels have a hazard ratio of 2.14 for PFS (95% confidence interval, 1.31-3.48; P < 0.001 from Cox partial likelihood Wald tests for a treatment difference in each subgroup). The second exception is that patients with a drop in E-selectin of >5.35 ng/mL have a hazard ratio of 0.94 for survival (95% confidence interval, 0.54-1.63), whereas patients with a drop of E-selectin of ≤5.35 have a hazard ratio of 1.98 for overall survival (95% confidence interval, 1.1-3.57; P = 0.02). Here (and elsewhere), the interpretation of differences in change from baseline is limited to patients continuing on study and progression-free for at least two cycles. Tests for treatment by factor interactions, which test whether the effect of treatment is different within the low and high levels, are significant for the difference in E-selectin levels for survival (P = 0.05) and for baseline ICAM for PFS (P = 0.04), although these interaction tests do not have good power for the small sample sizes here. The results suggest a substantial benefit from bevacizumab for patients with decreases in E-selectin levels of <5.35 (49% reduction in the mortality hazard rate) or with low baseline ICAM levels (53% reduction in the PFS hazard rate).

Discussion

A plethora of tumor markers have been reported in recent years. Significant variability in methodologic assessment of markers exists, and therefore, recent reporting recommendations for tumor marker prognostic studies have been published (21, 22). Our study follows those recommendations. In this prospective study, ICAM was a strong independent prognostic factor for overall survival. This is in concordance with initial body of evidence relating to ICAM in other malignancies (10–12). Whereas the source of soluble ICAM has not been totally elucidated, it is felt that endothelial cells are an important source. In vitro studies using cultured endothelial cells established that soluble ICAM may simply reflect ICAM expression on these cells (23). Therefore the better survival of patients with lower ICAM levels may reflect the tumor angiogenic load, with highly angiogenic tumors having higher levels and worse prognosis. The magnitude of difference in survival in patients categorized as having low ICAM levels (≤260.5 ng/mL) as opposed to high levels is quite significant with a 1-year survival of 65% versus 25%, respectively. In addition, ICAM was predictive for response to chemotherapy with or without bevacizumab. Patients with low ICAM levels had an increased probability of response to chemotherapy. This is the first reported study showing such a finding. Although we also had hypothesized that an increase in E-selectin and ICAM would be seen with treatment, we saw no significant change in ICAM at week 7 compared with baseline and actually also saw a statistically significant decrease in E-selectin that was similar on both arms.

VEGF levels on the other hand were not prognostic for overall survival. This is in concordance with data from two smaller studies, where serum VEGF was not prognostic for survival in early stage non–small cell lung cancer (24) nor did it correlate with the presence of distant metastases (25). VEGF levels were, however, predictive of response to bevacizumab in our study, and therefore, the primary end point of this correlative study was reached. Patients with high levels of baseline plasma VEGF (>35.7 pg/mL) had an increased probability of response to the BPC arm. The response rate in patients with low VEGF levels (≤35.7 pg/mL) was similar in PC and BPC arms. It is important to note, however, that VEGF levels were not predictive of the survival benefit afforded by the addition of bevacizumab to chemotherapy, and therefore, VEGF measurements have no clinical utility in this setting. This discrepancy between the ability of plasma VEGF to predict response, but not survival, may simply indicate that the improvement in survival seen with the addition of bevacizumab may not be related to the ability of this agent to increase response rate when added to chemotherapy.

In almost all subgroups, the hazard of death and progression were higher on the PC arm compared with BPC arm indicating the superiority of BPC over PC. For two groups, this hazard ratio was near or above 2: patients with low baseline ICAM and patients with a drop in E-selectin of ≤5.35 ng/mL at week 7. Tests for treatment by factor interactions, which test whether the effect of treatment is different within the low and high levels, were significant for the difference in E-selectin levels (week 7 minus baseline) for survival (P = 0.05) and for baseline ICAM for PFS (P = 0.04). It is important to note that these interaction tests do not have good power given the small sample sizes. Nevertheless, the results suggest a substantial benefit from bevacizumab for patients with decreases in E-selectin levels of <5.35 ng/mL (49% reduction in the mortality hazard rate) or with low baseline ICAM levels (53% reduction in the PFS hazard rate). We had previously observed an increase in E-selectin in patients (of whom all progressed on therapy) receiving the VEGF receptor inhibitor SU5416. Our findings in E4599 related to E-selectin may simply suggest the patients with significant progression of their disease (i.e., more aggressive tumor biology) are less likely to benefit from bevacizumab-based treatments. The finding that low ICAM level patients had a 53% reduction in PFS hazard rate when treated with bevacizumab may also indicate again that patients with lower tumor burden are those most likely to benefit from bevacizumab-based therapy. Other studies have looked at predictive factors on the benefit of bevacizumab and failed to show any predictive value. Jubb et al. looked at tissue expression of VEGF-A, thrombospondin-2, and microvessel density and found no predictive benefit in patients with metastatic colon cancer treated with 5-fluorouracil–based therapy ± bevacizumab (8). Another group failed to show a statistically significant relationship between mutations of k-ras, b-raf, or p53 and the increase in median survival associated with the addition of bevacizumab to IFL in metastatic colorectal cancer (26). Of importance, unlike our trial, which was prospectively planned, all other studies were retrospective.

The search for predictive markers for response and benefit from antiangiogenic agents continues. One promising marker is circulating endothelial progenitor cells. Levels of these progenitor cells decrease in response to antiangiogenic agents. Another marker under active investigation is the soluble VEGF receptor-2. The utility of these markers remain under active investigation (27, 28).