Predictive and prognostic value of preoperative serum tumor markers is EGFR mutation-specific in resectable non-small-cell lung cancer

Background The predictive and prognostic value of carcinoembryonic antigen (CEA), cytokeratin-19 fragments (Cyfra21-1), squamous cell carcinoma antigen (SCCA) and neuron-specific enolase (NSE) has been investigated in non-small-cell lung cancer (NSCLC) patients. However, few studies have directly focused on the association between these markers and epidermal growth factor receptor (EGFR) mutation status or mutation subtypes. Patients and methods We retrospectively analyzed 1016 patients with stage I-IIIA NSCLC who underwent complete resection between 2008 and 2012. Correlations between serum tumor marker levels and EGFR mutations and survival parameters were analyzed and prognostic factors were identified. Results Cyfra21-1 levels (P = 0.032 for disease-free survival [DFS]; P < 0.001 for overall survival [OS]) and clinical stage were identified as independent predictive and prognostic factors in EGFR-mutated adenocarcinoma patients. CEA levels (P < 0.001 for DFS; P = 0.002 for OS) and clinical stage were independently predictive and prognostic in EGFR wild-type adenocarcinoma patients. Further stratification analysis revealed that in EGFR exon 19 deletion adenocarcinomas, elevated Cyfra21-1 was an independent prognostic factor (P = 0.002). Within the Leu858Arg substitution subgroup, increased CEA (P = 0.005) and clinical stage were predictive factors of DFS, while elevated CEA (P = 0.005) and Cyfra21-1 (P = 0.027) were independent prognostic factors. Conclusion Cyfra21-1 and CEA exhibit different predictive and prognostic values between EGFR-mutated and wild-type adenocarcinomas, as well as between EGFR mutation subtypes. The prognostic impact of preoperative serum tumor markers should be evaluated together with EGFR mutation status.


INTRODUCTION
Despite significant progress in the last decade, lung cancer remains the leading cause of cancer-related death worldwide. Non-small cell lung cancer (NSCLC) is predominantly comprised of adenocarcinomas, squamous cell carcinomas and large cell lung carcinomas, and accounts for 75-85% of all lung cancers. Recent advancements in molecular targeted therapy have exploited the discovery of distinct cancer subsets with epidermal growth factor receptor (EGFR) mutations and led to a major paradigm shift in the treatment of NSCLC [1,2]. EGFR exon 19 deletion (del19) and exon 21 Leu858Arg substitution (L858R) make up around 90% of all EGFR mutation-positive lung adenocarcinomas, and are strongly associated with robust responses and improved progression-free survival (PFS) to EGFR tyrosine kinase inhibitors (EGFR-TKIs) [3]. Moreover, several studies demonstrated that EGFR del19 mutation consistently associated with better EGFR-TKI therapeutic outcomes compared with L858R substitution [4][5][6]. Preclinical studies have shown that EGFR del19 and L858R mutants have distinct biochemical properties that affect response to EGFR-TKIs [7], thus likely belong to different molecular subsets and should be studied independently [4].
EGFR mutations are most common in Asian populations, nonsmokers, females and those with adenocarcinoma histology [1]. The EGFR mutation rate in squamous cell carcinoma is reported to be approximately 5%, therefore, most squamous cell carcinoma patients do not benefit from EGFR targeted therapy [8]. Moreover, even in squamous cell carcinoma patients who achieved a partial response by EGFR-TKI, the median PFS was shorter than patients with adenocarcinoma harboring EGFR mutations [9].
Serum tumor markers including carcinoembryonic antigen (CEA) [10,11], neuron-specific enolase (NSE) [12], cytokeratin-19 fragments (Cyfra21-1) [13,14] and squamous cell carcinoma antigen (SCCA) [15], have been considered to be predictive or prognostic in NSCLC, although no consensus has been reached. Several studies have shown that serum tumor markers were associated with EGFR mutation status and capable of predicting the efficacy of EGFR-TKI therapy in advanced NSCLC. Higher serum CEA level has been associated with higher EGFR mutation rate [16][17][18], higher disease control rate (DCR) and longer survival time in advanced adenocarcinoma patients treated with EGFR-TKI [19]. Furthermore, in advanced NSCLC patients treated with EGFR-TKIs, high CEA and/or low Cyfra21-1 levels significantly correlated with higher responses and longer survival, especially in patients with unknown EGFR mutation status or with squamous cell carcinoma diagnosis [20]. Similarly, CEA reduction after 1 month of EGFR-TKI therapy was significantly correlated with DCR and PFS in EGFR wild-type/ unknown but not in mutated NSCLC cases [21]. However, Fiala et al. reported that high pre-treatment levels of CEA and/or Cyfra21-1 were associated with poor outcome for advanced NSCLC patients treated with erlotinib [22]. Moreover, Tanaka et al. demonstrated that high pretreatment Cyfra21-1, but not CEA levels, closely associated with shorter PFS in EGFR-TKI treated EGFRmutated NSCLC patients [8]. Therefore, prognostic significance of CEA and Cyfra21-1 in EGFR mutated NSCLC remains largely unknown. Furthermore, the relationship between NSE, SCCA and EGFR mutation has not been well investigated.
We investigated indicative prognostic factors among serum tumor markers (CEA, NSE, SCCA and Cyfra21-1) and the clinical characteristics for NSCLC patients who underwent complete surgical resection of stage I, II and IIIA tumors. Particularly we addressed the impact of respective tumor markers in patients with EGFR mutations as potential prognostic factors.

Patient characteristics
The initial study cohort included 1016 consecutive patients with completely resected NSCLC, and the followup period ranged from 2 to 72 months. Clinicopathological characteristics of the patients are shown in Table 1

Clinicopathological characteristics and serum tumor marker levels
Among all patients, preoperative CEA, Cyfra21-1, SCCA, and NSE levels were elevated in 33.5%, 43.7%, 17.9% and 34.5% of patients, respectively. These rates differed between adenocarcinoma and squamous cell carcinoma histological types. The positive rate of CEA was significantly lower in squamous cell carcinomas than in adenocarcinomas (27.0% versus 58.5%, χ 2 = 3.981, P < 0.001), while the positive rates of increased Cyfra21-1, SCCA as well as NSE were significantly higher in squamous cell carcinomas compared to adenocarcinomas (P < 0.001 for all comparisons). According to clinical stage, patients with stages II and IIIA tended to have higher tumor marker values than those with stage I. The distributions of all serum markers tested according to tumor stage and histology are shown in Table 2. www.impactjournals.com/oncotarget Median levels and positive rates for CEA, Cyfra21-1 or SCCA were similar regardless of EGFR mutation status in adenocarcinoma patients (Table 2). Similarly, no differences were found in those marker levels between del19 and L858R adenocarcinoma patient subgroups. However, median NSE levels in patients with EGFR mutations were higher than those with wild-type EGFR (median: 14.27 versus 13.25 ng/ml, P = 0.007). In addition, elevated NSE was observed in 32.1% EGFR mutated patients compared to 24.4% in EGFR wild-type patients (χ 2 = 3.981, P = 0.046). No difference was found in median levels and positive rates of NSE between del19 and L858R adenocarcinoma subgroups.

Association of increased CEA/Cyfra21-1/NSE with DFS and OS based on EGFR mutation status
EGFR was mutated in 218 adenocarcinoma patients, and among them CEA, Cyfra21-1, SCCA, and NSE increased in 86, 68, 12, and 70 patients, respectively. Due to the low frequency of elevated SCCA in EGFR-mutated NSCLC, we thereafter only investigated whether other markers could predict clinical outcome.

Univariate and multivariate analysis of prognostic factors in adenocarcinoma patients
By univariate analysis, adenocarcinoma patients harboring EGFR mutations had DFS and OS significantly associated with clinical stage, tumor size, regional lymph node (LN) metastasis, adjuvant treatment, CEA, and Cyfra21-1 (Table 3). When stratifying EGFR-mutated cases into del19 (105 cases) and L858R (113 cases) subgroups, we found that DFS and OS were longer in the Cyfra21-1 normal group than in the elevated Cyfra21-1 group for both subgroups. However, elevated CEA level was associated with shorter DFS and OS only in L858R patients. In addition, we found that DFS and OS significantly associated with clinical stage, tumor size, regional LN metastasis, adjuvant treatment, CEA, and Cyfra21-1 in the EGFR wild-type adenocarcinoma patients (Table 3).

Non-adenocarcinoma patients harboring EGFR mutations
Among non-adenocarcinoma patients, EGFR mutations were detected in 15 squamous cell carcinomas, 8 adenosquamous carcinomas 8 and 4 large cell lung carcinomas (

DISCUSSION
NSCLC is a heterogeneous and complex disease with high genetic, epigenetic, and phenotypic diversity. Preoperative serum tumor markers might reflect substantial intratumor subclonal variability, and have been evaluated as predictive or prognostic factors, either alone or in combination with clinicopathological parameters. Moreover, several serum tumor markers have been shown to be associated with EGFR mutation status and efficacy of EGFR-TKI treatment. However, there is not a consensus regarding the predictive and prognostic significance of these markers. In addition, their association with EGFR mutation subtypes (del19 and L858R) remains largely unknown.
In this study, we demonstrated that elevated Cyfra21-1 and advanced clinical stage were independently associated with shorter DFS and OS in EGFR-mutated adenocarcinoma patients, while increased CEA and advanced clinical stage were independently associated with worse DFS and OS in wild-type adenocarcinoma patients. Within the EGFR del19 subgroup, elevated Cyfra21-1 was correlated with shorter OS. However in L858R group, CEA and clinical stage were significantly associated with DFS, while elevated CEA and Cyfra21-1 were significantly unfavorable prognostic factors.
The correlation between baseline tumor marker values and NSCLC survival is controversial [10-13, 15, 21, 23-26]. We observed a direct relationship between high CEA and unfavorable prognosis, but only for adenocarcinoma patients with wild-type EGFR. Preoperative CEA may not be predictive or prognostic for patients with EGFR mutations due to the high sensitivity to EGFR-TKI and platinum-based doublet chemotherapy [27,28].
NSCLC diagnoses are based on biopsies from small regions of the tumor, thus may underrepresent rare subclonal squamous cell carcinoma or neuroendocrine cell populations. Simultaneous measurement of several serum tumor markers might provide a more comprehensive measure of a tumor's biological potential. Cyfra21-1 was previously shown to be useful for predicting clinical outcome of NSCLC patients [25], especially those with EGFR mutations [8]. Consistently, we observed that only in EGFR-mutated adenocarcinoma patients, elevated Cyfra21-1 was associated with worse DFS and OS. It is tempting to speculate that histologic heterogeneity plays a critical role in the elevated treatment efficacy for these patients compared to wild-type EGFR patients where Cyfra21-1 is not prognostic.
We observed squamous cell carcinoma patients had significantly shorter DFS and OS than those with adenocarcinomas, especially in the context of EGFR mutations, consistent with previous reports [8,25]. Elevated Cyfra21-1 reliably predicted worse DFS and OS in EGFR-mutated squamous cell carcinoma patients.
Previous studies have linked EGFR mutation status to elevated CEA levels [29,30]. However, we were unable to found a difference in the CEA or Cyfra21-1 levels between EGFR-mutated or wild-type adenocarcinoma patients [20]. This discrepancy could be attributed to our large sample size that partly avoided possible selection bias.
To our knowledge, the association between serum tumor markers and EGFR mutation subtypes (del19 and L858R) has not been reported. Previous studies have demonstrated that unlike the L858R substitution, EGFR del19 mutation is consistently associated with improved EGFR-TKI therapeutic outcomes [4][5][6]. This suggests that EGFR del19 and L858R NSCLC possess different biochemical properties and belong to distinct molecular subsets. Our findings are consistent with this notion as only Cyfra21-1 in correlated with shorter OS in the EGFR del19 group while both CEA and Cyfra21-1 were prognostic in L858R group. Even when using a more stringent cut-off (20.0 ng/ml), elevated CEA serum levels were still not significantly associated with either DFS (P = 0.149) nor OS (P = 0.211) for del19 patients.
In our study, patients with squamous cell carcinoma had significantly higher rates of elevated SCCA than those with adenocarcinoma. Only in the squamous cell carcinoma group did serum SCCA level serve as an independent predictive and prognostic marker, consistent with a previous report [15]. Within these patients, SCCA levels did not vary depending on EGFR mutation status.
Due to the low frequency of elevated SCCA in EGFRmutant adenocarcinoma patients, we could not analyze its prognostic value.
A previous study found that the prognostic value of NSE for stage I NSCLC is limited [31], while another reported that NSE was correlated with lactate dehydrogenase, tumor diameter, and disease extent as well as a predictor of survival [12]. Similarly, Yu et al. demonstrated that high levels of preoperative serum NSE correlated with worse survival in NSCLC patients [32]. We found that NSE was significantly higher in squamous cell carcinoma than in adenocarcinoma cases, although was unable to predict recurrence or OS. NSE levels were higher in adenocarcinoma cases with EGFR-mutations and only within L858R subgroup did NSE level have marginal predictive and prognostic significance. The discrepancy between our study and others could be attributed to different sample size of patients, or the inconstancy of covariables introduced in the proportional hazards model in previous studies. Our retrospective study here has some limitations, as confounding factors cannot be reduced as much as in prospective, randomized studies. It should be noted that treatment after recurrence clearly had an impact on overall survival, especially the use of EGFR-TKI for patients harboring EGFR mutations. In our cohort most recurrent patients received cisplatin-based doublet chemotherapy, while only 54 received EGFR-TKI therapy. However, recurrent patients treated with EGFR-TKI were evenly distributed into EGFR del19 and L858R subgroups. This may minimize the impact of the therapies administered after recurrence. Secondly, only EGFR del19 and L858R were examined in this study; therefore, uncommon EGFR mutations might have been miscategorized during analysis. Third, it is unclear whether alterations in serum tumor marker levels actually represent the intratumor heterogeneous components in each patient, although Tanaka et al. found that EGFR-TKI efficacy in EGFRmutated patients depended on the initial Cyfra21-1 level and concluded this represented a squamous-rich component in NSCLC [8]. Despite of these limitations, the cases analyzed in our study were in accordance with the uniform inclusion and exclusion criteria, which strengthens our confidence in the results and provides potentially useful information for clinical practice.
In conclusion, Cyfra21-1 is a predictive and prognostic marker in resectable adenocarcinoma patients harboring EGFR mutations, and a prognostic factor in EGFR del19 or L858R group. However, CEA was an independent predictive and prognostic factor only for EGFR wild-type adenocarcinoma patients and EGFR L858R adenocarcinoma patients. A prospective clinical trial is necessary to validate our present findings. (2) preoperative chemotherapy or radiotherapy; (3) history of second primary cancer diagnosed within 5 years; and (4) patients who died within 30 days after resection. Preoperative evaluation included physical examination, blood chemistry analysis, measurement of serum tumor markers, bronchofiberscopy, chest radiograph, computed tomography (CT), brain magnetic resonance imaging (MRI) or CT and bone scintigraphy. All patients underwent a wedge resection, lobectomy or pneumonectomy and systematic dissection of hilar and mediastinal lymph nodes for resection of the primary lesion. Complete resection was defined as complete removal of all tumors with negative margin proven by histopathological examination. Diagnosis of malignant disease was confirmed pathologically and classified according to World Health Organization histological classification (3rd edition) and staged according to the TNM classification of the Union for International Cancer Control (7th edition). Primary adjuvant treatment after surgery was chemotherapy or radiotherapy, either alone or in combination. Platinumbased adjuvant chemotherapy was routine for stage IB-IIIA NSCLC. Patients with multistation mediastinal N2 disease received sequential radiotherapy following completion of adjuvant chemotherapy. Follow-up information was collected directly from the outpatient clinic records or from family contact. Patients were evaluated every 3 months by chest CT scans for the first 2 years after surgery and annually thereafter. The date of recurrence, treatment for recurrence, date of death or last visit, and cause of death were recorded. Patient characteristics are shown in Table 1.

Measurement of serum CEA, NSE, SCCA and Cyfra21-1 levels and EGFR mutations
Serum concentrations of CEA, NSE, SCCA and Cyfra21-1 were measured within 2 weeks before surgery by electrochemiluminescence immunoassay on Roche Analytics E170 Immunology Analyzer (Roche Diagnostics, China). Based on manufacturer recommendation, the following cut-offs for serum marker levels were used: CEA 5.0 ng/ml, NSE 15.2 ng/ml, SCCA 1.5 ng/ml, and Cyfra21-1 3.3 ng/ml. EGFR mutations (del19 and L858R) were identified by real-time PCR or DNA sequencing as previously described [33]. Other EGFR mutations were not tested.

Statistical analysis
Continuous variables were described using mean ± SD or median and range. Comparison of average value between groups was detected with rank sum test. Categorical variables were compared by chi-square test or Fisher's exact test. DFS was defined as the interval between lung resection and local recurrence and/or occurrence of distant metastases. Overall survival time was calculated as the interval between surgery and death or last clinical evaluation. Survival was estimated by the Kaplan-Meier method and compared using the Log-rank test for univariate analysis. The hazard ratio (HR) and the 95% confidence intervals (CI) were estimated using the www.impactjournals.com/oncotarget Cox proportional hazard model, and the multivariate Cox model was developed using stepwise regression (backward selection) to adjust for potential confounding factors. P values of less than 0.05 were considered statistically significant. SPSS (version 18.0; SPSS, Inc., Chicago, IL) was used for all analyses.