Impact of Serum Hepatocyte Growth Factor on Treatment Response to Epidermal Growth Factor Receptor Tyrosine Kinase Inhibitors in Patients with Non–Small Cell Lung Adenocarcinoma

Selective epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKI) block signal transduction pathways implicated in the proliferation and survival of cancer cells (1–3), and show clinical activity against non–small cell lung carcinoma (NSCLC, refs. 4–6). To date, four populations are known to have a better response to EGFR-TKIs: females, never-smokers, adenocarcinoma histology, and patients of East Asian ethnicity (4–6). Active EGFR mutation apparently confers a hyperresponsiveness to gefitinib among NSCLC patients (7, 8). Such mutations affect the ATP-binding cleft of EGFR, and EGFR mutants exhibit constitutive tyrosine kinase activity. Most reported EGFR mutations are either point mutations in exons 18 (G719A/C) and 21 (L858R and L861Q) or in-frame deletions in exon 19 that eliminate five amino acids (ELREA) located at position 745 (9).

A large-scale randomized study, the IRESSA Pan-Asia study (IPASS), comparing gefitinib monotherapy with carboplatin/paclitaxel for 1,217 previously untreated patients with lung adenocarcinoma, was recently completed, and the results showed that the progression-free survival (PFS) curves of the two arms crossed each other. A subset analysis of the 437 patients whose EGFR mutation status was known revealed that the presence of an EGFR mutation in the tumor was a strong predictor of a better outcome after gefitinib treatment (10). Two recent phase III trials targeting adenocarcinoma patients with EGFR mutations showed that the gefitinib group had a significantly longer PFS than the platinum-doublet therapy group (11, 12). These data indicated EGFR mutation status as a powerful predictor of the tumor response to EGFR-TKIs.

Even among patients with EGFR mutations, however, not all patients respond to EGFR-TKI treatment in like manner; the objective response to EGFR-TKI treatment has remained at 62% to 75% (11–13). In addition, no effective biomarker is currently available for patients with wild-type EGFR tumors (14). These facts motivated us to investigate molecular biomarkers that can be utilized independently of EGFR status to predict the efficacy of EGFR-TKIs. Identifying such a marker would contribute to the further individualization of treatment for NSCLC. In this report, we retrospectively studied the serum concentrations of several molecules in patients with non–small cell lung adenocarcinoma who underwent treatment with EGFR-TKIs.

A total of 104 patients with histologically confirmed non–small cell lung adenocarcinoma who had been treated with EGFR-TKIs at three centers (Kanazawa University, Japan; Cancer Institute Hospital, Japan; and Tokyo Medical University, Japan) between 2002 and 2009 were included in this study. Six patients were excluded because their tumor response was not evaluated. A complete clinical data set was not available for three additional patients. Thus, 95 patients were included in the final analysis (Fig. 1A). The tumor response was evaluated every 2 to 3 months using computerized tomography according to the Response Evaluation Criteria in Solid Tumors; the response was then classified as a complete response (CR), a partial response (PR), stable disease (SD), or progressive disease (PD). Clinicopathologic features including age, gender, Eastern Cooperative Oncology Group (ECOG) performance status (PS), tumor-node-metastasis stage, smoking status, and EGFR mutation status were recorded. Direct sequencing of a tumor sample was done to detect active EGFR mutations in 43 patients; 20 of these samples were found to harbor an EGFR mutation (exon 19, n = 14; exon 21, n = 6), whereas the remaining 23 samples exhibited wild-type EGFR. The mutation status of the other 52 patients was not evaluated. The present study was approved by the institutional review boards of all the centers.

Blood samples were collected before the initiation of EGFR-TKI treatment. Separated serum was stocked at −80°C until use.

Serum HGF concentrations were determined using a Human HGF Quantikine ELISA Kit (R&D Systems) according to the manufacturer's instructions. A 50-μL aliquot of serum per well was examined in duplicate, and the average was used for subsequent analysis. The absorbance of the samples at 450 nm was measured using VERSAmax (Japan Molecular Devices).

Antibody suspension bead arrays for determining the serum concentrations of 12 molecules were obtained commercially (WideScreen Human Cancer Panel 2, Merck). The markers used in this panel were as follows: amphiregulin, β-cellulin, epidermal growth factor (EGF), EGFR, epiregulin, fibroblast growth factor-basic (FGF-basic), heparin-binding EGF-like growth factor (HB-EGF), platelet-derived growth factor β polypeptide (PDGF-BB), placental growth factor (PIGF), tenascin C (TnC), transforming growth factor α (TGF-α), and vascular endothelial growth factor (VEGF). Data were obtained using a Bio-Plex suspension array system (Bio-Rad Laboratories). The assay was done according to the manufacturer's instructions and a previously described method (15). Serum samples were diluted 1:4 with the appropriate diluents prior to assay. The samples were tested in duplicate, and the averages were used for analysis.

The primary objective was to investigate novel markers correlated with efficacy independently of EGFR-status. If a molecule was very strongly associated with survival after adjustments for the EGFR status and important prognostic factors, then that molecule was deemed as warranting further prospective study to determine whether it is a predictive factor, a prognostic factor, or both.

The distributions of the clinical factors and molecules were compared between patients with PD and those without PD using the Wilcoxon test. In terms of the analysis for survival time [PFS and overall survival (OS)], clinical factors including age, gender, ECOG PS, clinical stage, and smoking status were examined using the Cox proportional hazards model. After selecting the important clinical variables, we considered these variables fixedly in a Cox proportional hazards model and then determined which, if any, of the molecules was associated with survival using the backward selection method; thus, any molecules remaining in the final model were significant in a manner that was independent of the important clinical variables at a two-sided level of 0.05 using the Wald test. Log-transformed values were used for the molecules in the Cox models. The proportional hazards assumption was assessed graphically, and an individual time-dependent component was included for each covariate. In the multivariate Cox models, EGFR status (wild-type/mutant/unknown) was treated as a stratified variable. We applied the above analyses to all the cases, to the cases in which the EGFR status was evaluated, and to the cases with wild-type EGFR to check the robustness of the conclusions. The survival curves for PFS and OS were estimated using the Kaplan-Meier method. The Kaplan-Meier curves were shown just to visualize the trends of the association between the molecules and PFS/OS, as any determination of the optimal cutoff point for the molecules relative to the PFS/OS was beyond the scope of the present study. All the statistical analyses were done using SAS for Windows (ver. 9.1.3) and Medcalc for Windows (ver. 11.1.1).

Of the 95 patients evaluated in this study, all the patients were of Asian ethnicity and had been treated with an EGFR-TKIs (gefitinib, n = 91; erlotinib, n = 4). Seventy-five (79%) and eight (8%) patients had received prior chemotherapy and radiotherapy, respectively. After the start of EGFR-TKI treatment, a PR was observed in 37 (39%) patients, SD in 25 (26%) patients, and PD in 33 (35%) patients; none of the patients exhibited a CR. The median follow-up period was 29 months. The patient characteristics are shown in Table 1. Regarding tumor response, male gender (P = 0.0002), a PS of 2 to 4 (P = 0.0498), a positive smoking history (P = 0.0010), and a wild-type EGFR status (P < 0.0001) were significantly less sensitive to EGFR-TKI treatment (Table 1). These results are consistent with those of a previous report (16).

We measured the serum concentrations of 13 molecules: amphiregulin, β-cellulin, EGF, EGFR, epiregulin, FGF-basic, HB-EGF, HGF, PDGF-BB, PIGF, TnC, TGF-α, and VEGF. Among them, many samples (>80%) showed a value that could not be measured or that was below the standard range for amphiregulin, β-cellulin, epiregulin, FGF-basic, and TGF-α; these molecules were omitted from subsequent analyses. Therefore, only the correlations between clinical outcome and the serum concentrations of EGF, EGFR, HB-EGF, HGF, PDGF-BB, PIGF, TnC, and VEGF were analyzed.

The serum concentrations of EGF, HB-EGF, HGF, TnC, and VEGF were significantly higher among patients with PD than among those without PD (Table 2). Among these molecules, HGF and VEGF exhibited marked differences in tumor response. These results were unchanged when subsets of patients with a known EGFR status (n = 43) as well as patients with wild-type EGFR (n = 23) were analyzed (Fig. 1B-D). These observations suggested that HGF and VEGF were significantly associated with tumor response, independent of the EGFR mutation status. The sensitivity and specificity of HGF and VEGF for discriminating PD from PR plus SD were determined using the cutoff values according to a previously reported methodology (17). The cutoff values for HGF and VEGF for discriminating PD from PR plus SD were 1,228 pg/mL and 1,187 pg/mL, respectively. The sensitivity and specificity of HGF for discriminating PD from PR plus SD were 0.848 and 0.677, and those for VEGF were 0.788 and 0.710, respectively.

The numbers of observed events were 85 cases for PFS and 70 cases for OS. The median PFS and OS were 4.0 and 8.9 months, respectively. Among the clinical factors that were examined, a male gender, a poor PS, and a positive smoking history were significantly related with a poor PFS and OS (Table 3). Regarding the serum concentrations of the molecules, higher concentrations of HGF, HB-EGF, TnC, and VEGF were significantly associated with a shorter PFS and OS (Table 3). Similar to the results for tumor response, HGF and VEGF were associated with both PFS and OS when all the cases were analyzed, with all tests exhibiting a hazard ratio (HR) with a P value <0.0001. Similar results were obtained when subsets of patients with a known EGFR status and patients with wild-type EGFR were analyzed (data not shown). Figure 2 shows the Kaplan-Meier estimates for PFS and OS with regard to the concentrations of serum HGF and VEGF. Although a determination of the optimal cutoff values was beyond the scope of this study, all the patients were divided into one of four groups according to the quartile values for each molecule. The curves provided a clear trend indicating that the outcomes became poorer as the concentrations of these molecules increased. To compare these results with those for patients with a known EGFR status in Table 3, we calculated the univariate HRs for the EGFR status (wild-type versus mutant) in the subset of patients with a known EGFR status. These HRs were 3.49 [95% confidence interval (95% CI), 1.76-6.92; P = 0.0004] and 3.32 (95% CI, 1.52-7.24; P = 0.0026) for PFS and OS, respectively.

The purpose of this study was to investigate novel markers correlated with the response to EGFR-TKI treatment independently of the EGFR status. However, the EGFR status (wild-type/mutant/unknown) did not satisfy the assumption of proportional hazards in our data. Hence, we adopted a stratified model, and the multivariate Cox analyses always included the EGFR status as a stratification factor when estimating the adjusted HRs (ref. 18; Table 4). Gender and PS remained statistically significant at a level of 0.05 in a multivariate model that included the three clinical variables with a small P value (<0.20) in a univariate analysis. Smoking status was no longer significant (P = 0.46 and P = 0.40 for PFS and OS, respectively) because it was highly correlated with gender (P < 0.0001, Fisher's exact test). Thus, we fixed gender and PS in the Cox model and identified significant molecules by backward selection with P ≥ 0.05 as a removal criterion. Table 4 shows the final model using this procedure. HGF (HR, 6.31; 95% CI, 2.60-15.3; P < 0.0001) and VEGF (HR, 2.01; 95% CI, 1.14-3.57; P = 0.0165) were significant for PFS, whereas only HGF (HR, 4.01; 95% CI, 2.20-7.32; P < 0.0001) was significant for OS; a high concentration of HGF showed a shorter PFS and OS independently of the clinical variables of EGFR status, gender, and PS (Table 4). In the final model, no interaction was shown between the EGFR status and HGF (P = 0.84 and P = 0.20 for PFS and OS, respectively). The results shown in Table 4 were also stable in analyses of subsets of patients with a known EGFR status as well as patients with wild-type EGFR (data not shown).

HGF was identified as a natural ligand of the MET receptor and belongs to the plasminogen family (19). It contains a hairpin loop followed by four kringle domains flanked by an activation portion and a serine protease domain devoid of proteolytic activity (20). In cancer cells, activation of the Met receptor increases invasion and metastasis, and allows the survival of cancer cells in the bloodstream in the absence of anchorage (20). In addition, HGF is well known as a potent angiogenic cytokine, and Met activation can modify the microenvironment to facilitate cancer progression (21). Therefore, HGF-MET signaling is regarded as an oncogenic signaling pathway, and intensive therapeutic approaches focusing on this signaling pathway are ongoing in the field of cancer treatment (22).

Meanwhile, a high serum HGF concentration has been associated with a poor prognosis in many malignant neoplastic diseases, including colorectal, esophageal, gastric, and prostate cancer, as well as malignant myeloma (23–25). Where cancer treatment is concerned, high peripheral and portal HGF serum levels are related to a poor prognosis after hepatic resection in hepatocellular carcinoma (HCC) patients (26). The HGF level is also elevated in patients with myeloma, and patients with higher HGF levels tend to have a poorer prognosis; furthermore, treatment with high-dose chemotherapy reduced the serum HGF level in the majority of these patients (27). The above-mentioned evidence suggests that a high serum HGF level is a relatively well-established predictor of poor prognosis. Ishikawa et al. have shown that, besides HGF, amphiregulin and TGF-α are predictors of a poor response to gefitinib in EGFR status-unknown advanced non–small cell lung cancer (17). Although we did not obtain similar results, these molecules are expected to be noninvasive predictive biomarkers.

On the other hand, few reports have examined the predictive value of HGF with regard to the clinical outcomes of chemotherapy or molecular-targeting drugs. A recent study showed that the pretreatment HGF concentrations in the peripheral blood plasma as well as in the bone marrow plasma of patients who achieved a complete or very good partial response were significantly lower than those in patients who had a partial or worse response (28). In this study, we showed, for the first time, that a high pretreatment serum HGF level was associated with poor clinical outcomes, including tumor response and survival time, in patients treated with EGFR-TKIs. The robust association of this parameter with outcome regardless of the EGFR mutation status clearly suggests that the serum HGF level could be useful for predicting the response to EGFR-TKI treatment on an individual basis. The direct sequencing for EGFR mutations could have produced false-negative results, presenting a bias for the relevance of HGF in patients with supposed wild-type EGFR. We plan to evaluate EGFR mutations using a highly sensitive detection method in a future study.

Although we do not have any definitive biological data to address the question of why a high pretreatment serum HGF level is associated with a poor clinical outcome after EGFR-TKI treatment, several studies are worth noting. Engelman et al. showed that the amplification of MET causes gefitinib resistance by driving the ERBB3 (HER3)-dependent activation of phosphoinositide 3-kinase, and they proposed that MET amplification may promote drug resistance in other ERBB-driven cancers (29). Yano et al. clearly showed that HGF-mediated MET activation is involved in gefitinib resistance in lung adenocarcinoma with EGFR-activating mutations (30). In addition, a recent study has clearly shown that HGF accelerates the development of MET amplification both in vitro and in vivo, mediating the EGFR kinase inhibitor resistance caused by either MET amplification or autocrine HGF production (31). These studies indicate that the activation of HGF-MET signaling confers resistance to EGFR-TKIs. Therefore, lung cancer cells producing high amounts of HGF might mediate drug resistance to EGFR-TKIs, leading to poor treatment outcomes.

In this study, a multivariate analysis revealed that the serum HGF level was independently associated with a poor treatment outcome. This finding indicates the possibility of using the serum HGF level to refine the indications for patients who are likely to respond to EGFR-TKI treatment. In particular, patients with wild-type EGFR are considered to be less sensitive to EGFR-TKIs, but our results might allow the identification of a selective subgroup of these patients who might actually benefit from this treatment. When used in combination with EGFR mutation status, the serum HGF level might increase the clinical benefit of EGFR-TKIs by allowing the further individualization of this treatment. We plan to conduct a prospective study to validate the ability of the serum HGF level to predict the response to EGFR-TKI treatment.

No potential conflicts of interest were disclosed.

We thank Ms. Tomoko Kitayama for technical assistance.

Grant Support: Third-Term Comprehensive 10-Year Strategy for Cancer Control, a Grant-in-Aid for Scientific Research, the Program for the Promotion of Fundamental Studies in Health Sciences of the National Institute of Biomedical Innovation (NiBio), a Grant-in-Aid for Cancer Research (H20-20-9) from the Ministry of Health, Labour and Welfare and a Grant-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science and Technology of Japan (19209018).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.