Expression levels of ROS1/ALK/c-MET and therapeutic efficacy of cetuximab plus chemotherapy in advanced biliary tract cancer

Aberrant expression of ROS1, ALK or c-MET (RAM) is implicated in carcinogenesis and cancer drug resistance. We retrospectively evaluated the effect of RAM expression on outcomes for advanced biliary tract cancer patients, who were treated with gemcitabine plus oxaliplatin (GEMOX), with or without cetuximab, in a randomized phase II trial. RAM expression levels on archived tissue sections were scored using immunohistochemistry (IHC). Of 110 tumors with IHC staining for all three markers, 18 were RAMhigh (IHC intensity 3+ for any markers). Ninety-two tumors were RAMlow (IHC intensity <3+ for all markers). All RAMhigh tumors were intra-hepatic cholangiocarcinomas (IHCC). Of the patients with IHCC (n = 80), median overall survival (OS) of RAMhigh group was inferior to that of the RAMlow group (5.7 vs. 11.7 months, p = 0.021). In multivariate analysis RAMhigh remained an independently adverse prognostic factor, with a hazard ratio of 2.01 (p = 0.039). In the RAMlow group, GEMOX treatment with cetuximab significantly improved the disease control rate (68% vs. 41%, p = 0.044), median progression-free survival (7.3 vs. 4.9 months, p = 0.026), and marginally prolonged median OS (14.1 vs 9.6 months, p = 0.056), compared to GEMOX treatment alone. Future trials of anti-EGFR inhibitors for IHCC may consider RAM expression as a patient stratification factor.

Scientific RepoRts | 6:25369 | DOI: 10.1038/srep25369 chemotherapy 4,5,7 . Identification of predictive and prognostic biomarkers is crucial to better define patients who may benefit from targeted molecular therapies such as anti-EGFR inhibitors.
Various biomarkers predictive for anti-EGFR inhibitor efficacy have been explored for different cancer types. For example, RAS/BRAF mutations predicted poor response to cetuximab in patients with metastatic colorectal cancers 8 . In another study, high EGFR expression predicted incremental benefit from cetuximab combine with chemotherapy in patients with advanced non-small cell lung cancer (NSCLC) 9 . However, two previous randomized studies with either pre-planned or post-hoc analyses failed to demonstrate an association between the status of RAS/BRAF mutations and the therapeutic efficacy of cetuximab in ABTC, or between the latter and EGFR expression 4,7 . Additional molecular alternations have been identified to account for the resistance to EGFR inhibition in NSCLC and provide opportunities for new drug development. Among them, over-expression of c-MET, which could result in activation of hepatocyte growth factor (HGF)-c-MET signaling correlates with resistance to anti-EGFR therapy in patients with NSCLC 10 . c-MET overexpression has been detected in 11.7% and 16.2% of intra-and extra-hepatic cholangiocarcinoma (IHCC and EHCC) cases, respectively, and associated with inferior post-resection recurrence-free survival than those without 11 . Chromosome rearrangements involving either anaplastic lymphoma kinase (ALK) or ROS1 genes have been identified in wild-type EGFR NSCLCs that were refractory to anti-EGFR therapy 12,13 . ROS1 rearrangement has been detected in 8.7% of IHCC cases and shown to be oncogenic in the BTC mouse model, whereas ALK rearrangement or aberrant protein expression has not been described in ABTC 14,15 . In NSCLC, ALK or ROS1 rearrangement-harboring tumors exhibited compatible response to ALK inhibitors because of the high concordance of kinase domain sequences 16,17 . Whether alternations in c-MET, ROS1, and ALK signaling pathways may impact the clinical outcomes of patients undergoing anti-EGFR therapy and provide opportunities for new ABTC drug development, remains unexplored.
The present post-hoc study evaluates predictive and prognostic significances of aberrant ROS1, ALK or c-MET (RAM) expression in a prospective cohort of ABTC patients who received gemcitabine plus oxaliplatin (GEMOX) chemotherapy with or without cetuximab in a randomized phase II study.

Results
RAM protein expression and rearrangement of ROS1 and ALK genes. Of the tumor tissue sections from 122 patients who participated in the prospective randomized phase II trial, sections from 110 patients had sufficient tissue for IHC of all three markers. The staining of the RAM markers was mainly localized in the cytoplasm of tumor cells (Fig. 1A). Among them, 18 of cases (16%) were categorized as RAM high , with 3+ IHC staining of ROS1 in 9 cases, ALK in 5, and c-MET in 6 (Supplementary Table 1). One tumor overexpressed all three markers. ALK rearrangement was detected in 1 of the 5 tumors presenting ALK overexpression (Fig. 1B), whereas none of the 9 tumors with ROS1 overexpression had detectable ROS1 rearrangement.

Correlations between RAM expression and clinicopathological factors.
As of December 31, 2014, the median follow-up was 10.3 months (range 0.9-45.4 months). The demography of the 110 ABTC patients with RAM high or RAM low tumors is listed in Table 1. Compared with RAM low tumors, RAM high tumors were more likely to be IHCC (100% vs. 67.4%, p = 0.015), without prior surgery (0% vs. 48.9%, p < 0.0001), and with concurrent mutant KRAS (55.6% vs. 31.5%, p = 0.063). The trend of differences in the baseline characteristics between RAM high and RAM low groups was similar in the 80 patients with IHCC (Supplementary Table 2).
Comparison of treatment efficacies. Of the 110 patients included in this study, those who received C-GEMOX had a tended to have better ORR, better long-term disease control rates (DCR ≥16wk , objective response according to RECIST plus stable disease [SD] ≥ 16 weeks), and longer survival compared to patients who received GEMOX (Supplementary Table 3) 2 . Because all RAM high tumors were IHCCs, further analysis was performed on the IHCC population. Despite similar DCR ≥16wk and PFS, IHCC patients with RAM high tumors had a significantly inferior median OS than patients with RAM low tumors (5.7 vs. 11.7 months, p = 0.021, Table 2). In the RAM low IHCC subgroup, C-GEMOX treatment achieved significantly better DCR ≥16wk (68% vs. 41%, p = 0.044) and median PFS (7.3 vs. 4.9 months, p = 0.026) than GEMOX alone did. A trend toward improved OS was also observed (14.1 vs. 9.6 months, p = 0.056, Table 3 and Fig. 2). Both KRAS wt and KRAS mut patients in the RAM low IHCC subgroup showed similar trends of improvement in DCR ≥16wk and median PFS after C-GEMOX treatment (Supplementary Table 4). This KRAS-independent improvement suggests that KRAS mutation did not interfere with the benefits derived from add-on cetuximab to GEMOX in RAM low cases. The findings were consistent with our previous report on the entire ABTC intention-to treat population 4 . However, in the RAM high subgroup, the DCR ≥16wk , PFS and OS were similar between C-GEMOX and GEMOX-treated patients (Table 3).
Multivariate analysis showed that male gender and RAM high subgroup were independent adverse prognostic factors for OS in 80 IHCC patients (Table 4). Male gender and GEMOX alone had only a statistically non-significant tendency to prognosticate worse PFS (Supplementary Table 5).
Recommendations for phase III trial design based on phase II trial with biomarkers. In the RAM low IHCC subgroup, C-GEMOX achieved better median PFS than GEMOX did, with a hazard ratio (HR) of 1.78 at one-sided α = 0.1, and the null hypothesis was rejected, according to Freidlin et al. 17 . The 2-sided 80% confidence interval (CI) for the HR in the RAM high subpopulation was 0.39 to 1.50 (Supplementary Table 6). For RAM high cases, this range is consistent with the null hypothesis that the efficacy of C-GEMOX is no different from that of GEMOX. Therefore, future large-scale clinical trials should consider using RAM expression as a patient stratification factor.

Discussion
The present post-hoc analysis is the first report to describe the prognostic significance of aberrant ROS1, ALK, or c-MET expression in ABTC patients undergoing front-line therapy. We found that patients with tumors Scientific RepoRts | 6:25369 | DOI: 10.1038/srep25369 over-expressing any of the ROS1, ALK, or c-MET proteins (RAM high ) had inferior OS, regardless of the assigned treatment. Patients with RAM low tumors, notably of IHCC, were likely to benefit from the addition of cetuximab to GEMOX chemotherapy.
The findings of 18 RAM high tumors that were all IHCC cases and without prior surgery at study entry imply that aberrant RAM expression can be site-specific and associated with aggressive tumor biology. Recent comprehensive genomic profiling studies of BTCs also suggested distinct etiologies and molecular characteristics among IHCC, EHCC and gallbladder carcinoma (GBC) [18][19][20][21] . IDH mutation and FGFR rearrangement are usually specific to IHCC, whereas BAP1 and IDH mutations are more frequently detected in non-parasite related cholangiocarcinoma than in those related to parasite infection. Compared with IHCC, EHCC has a significantly higher frequency of TP53 and KRAS mutations. These BTC subtype-specific genetic alternations indicate that studies of molecular pathogenesis and therapeutic development may be most fruitful on a subtype-specific basis. The observation of higher KRAS mutation rate in RAM high tumors indicates that the RAM high phenotype might be related to the 'proliferation' class of IHCC, which is associated with worse clinical outcomes than other IHCC classes 22,23 .
Because of the lack of association between the status of KRAS mutation or EGFR positivity and therapeutic benefit of cetuximab as shown in our primary report, we strongly encourage a search for other predictive biomarkers for anti-EGFR therapy in ABTC. ROS1, ALK and c-MET, appear to be reasonable choices because the known crosstalk between EGFR signaling and all three signaling pathways, and their genetic alterations or aberrant expression may confer resistance to EGFR inhibition in NSCLC or other cancer types [24][25][26][27][28][29] . For example, c-MET activation involves the acquisition of resistance to anti-EGFR therapy via bypass of the EGFR pathway 26 . EGFR activation can mediate the resistance to ROS1 inhibition in NSCLC 25 . In addition, ROS1 activation mediates resistance to EGFR inhibition in glioblastoma cells 24 . ALK kinase is another druggable target in NSCLC patients who do not respond to EGFR inhibitors or have EGFR wild-type tumors; ALK activation may be associated with resistance to EGFR tyrosine kinase inhibitor 27,28 . Although aberrant ALK signaling has not been reported in BTC, ROS1 rearrangement and c-MET over-expression have been detected in cholangiocarcinoma; the latter observations are associated with aggressive biological behavior 11 Table 3. Therapeutic efficacies stratified by treatment arms in IHCC with RAM low and RAM high expression. RAM, ROS1/ALK/c-MET; IHCC, intrahepatic cholangiocarcinoma; C-GEMOX, cetuximab plus gemcitabine and oxaplatin; PFS, progression-free survival; OS, overall survival. # Disease control rate = complete response+ partial response+ stable disease >16 weeks. ## Disease control rates were compared using Fisher's exact test, 2-tailed; PFS and OS were compared using log-rank test. in our analysis. Finally, due to the low respective incidence of aberrant ROS1, ALK, or c-MET expression, a pooled analysis of all 3 molecules was performed to increase the statistical power of our analysis. Although our previous studies demonstrated a high concordance between ALK rearrangement and protein over-expression in surgically resected NSCLC samples 32 , FISH detected chromosomal translocation for only 1 of 5 ALK over-expression tumors in the present study (ALK rearrangement, Fig. 1B). However, our present finding of discord between the frequency of ROS1 over-expression and that of ROS1 rearrangement is consistent with results of previous IHCC and NSCLC studies. Seventy-two of 194 surgically resected IHCC tumors (37%) exhibited high expression level of ROS1; none of the 102 samples successfully analyzed using FISH had detectable ROS1 rearrangement 33 . Analyses of NSCLC tumors found that the frequency of ROS1 over-expression and ROS1 rearrangement was 22% (9 of 399) and 1.7% (18 of 1037), respectively 12,34 . Lee et al. further showed that a small fraction of NSCLC tumors over-expressing ROS1 exhibited hypo-methylation in the ROS1 promoter region; however, the mechanisms of this dysregulation remain largely un-elucidated 34 . Our discordant IHC and FISH results could be caused by the technical difficulty posed by FISH analyses of small IHCC biopsy tissue samples. Another possible explanation of the discordance could be that translational or post-translational regulation enhances the expression of ROS1/ALK proteins.
Our data indicate that RAM low expression predicts therapeutic efficacy of add-on cetuximab in IHCC with GEMOX-based therapy. These results are applicable to this statistical model proposed by Freidlin et al., which aims to clarify the values off biomarkers in designing phase III randomized trials of targeted therapies 35 . RAM expression status should be considered as a stratification factor (RAM low vs. RAM high subgroups) for future trials of anti-EGFR inhibitors in IHCC. However, the wide 80% CI of the HR estimate (Supplementary Table 6), resulting from the small sample size of patients in the RAM high subgroup, possibly misled the conclusion according to Freidlin's algorithm. Therefore, we strongly recommend adoption of this algorithm at the interim analysis of phase III studies, through monitoring the HR of biomarker-defined subgroups. Such monitoring should be used to determine whether RAM should be used continuously as a stratification factor or shifted to the role of an enrichment factor. Based on the dismal survival outcome of RAM high patients and the well-known clinical activity of ALK inhibitors in NSCLC with ALK and/or ROS1 rearrangement, an investigator-initiated phase II trial is ongoing. This trial evaluates the efficacy and safety of ceritinib in patients with IHCC over-expressing ROS1 or ALK (clinicaltrials.gov NCT2374489).
For IHCC patients, ROS1, ALK and c-MET expression levels have prognostic significance on clinical outcomes. Although this finding may require further validation, it has led to proposal of a new stratification or enriched biomarker for future phase III trial of anti-EGFR therapy in IHCC. This example demonstrates the importance of prospective tissue sample collection from well-designed clinical trials for companion exploratory biomarker studies. Such exploratory studies can facilitate further drug development.

Methods
Patients. The current study analyzed archival tumor tissues from patients with ABTC who participated in a Taiwan Cooperative Oncology Group (TCOG)-sponsored randomized phase II trial comparing first-line GEMOX alone versus GEMOX plus cetuximab (C-GEMOX) 4 . In brief, 122 patients were 1:1 randomized, stratified by tumorous KRAS status, ECOG performance status, and location of primary tumor, and then 1:1 randomized. The patients received GEMOX or C-GEMOX until disease progression or the occurrence of unacceptable toxicity. The primary end point of the study was objective response rate (ORR) by RECIST. Trends toward improvement in ORR (27% vs. 15%, p = 0.12) and median PFS (6.7 vs. 4.1 months, p = 0.05) favoring C-GEMOX were observed. The trends of improvement did not correlate with tumorous KRAS mutation status. However, C-GEMOX did not improve OS (10.6 vs. 9.8 months, p = 0.91) at primary analysis with a cut-off date on April 30, 2013 4 . In the present study, survival follow-up was updated to December 31, 2014. Tumor procurement was approved by the institutional review board of National Health Research Institutes (EC1040427-E). Written informed consent was obtained from all subjects and the study was conducted in accordance with International Conference Statements for Good Clinical Laboratory Practice and the declaration of Helsinki of the World Health Medical Association. Samples subjected to FISH in which more than 15% of tumor nuclei had wide apart of red and green signals for more than 2 nuclei, or an absent green signal, were considered break-apart positive. Positive break-apart samples were considered to have ALK or ROS1 rearrangements.

Decision algorithm for recommendations of phase III trial design.
To determine whether the RAM expression status may serve as a patient enrichment or patient stratification factor in future randomized trials of EGFR inhibitors for ABTC, the median PFS of patients who received GEMOX or C-GEMOX was further analyzed using an algorithm proposed by Freidlin B et al. (see supplementary information for detailed methods) 35 . Briefly, the median PFS was analyzed first to test whether the GEMOX and C-GEMOX treatment groups had the same PFS in the RAM low subpopulation. If the null hypothesis was rejected, then the 2-sided 80% CI for the HR in the RAM high subpopulation was analyzed to determine whether RAM expression was suitable to be used as a patient-enrichment or patient-stratification biomarker in future clinical trials. If the null hypothesis was not rejected, then the 2-sided 80% CI for the HR for all patients was analyzed to determine the value of RAM expression in future clinical trials.
Statistical Analysis. Correlations between the results of IHC and clinicopathological factors were assessed using Fisher's exact probability test or t-test. The associations between objective response rate stratified by treatment (GEMOX or C-GEMOX) and RAM expression status (RAM high vs RAM low ) were evaluated using the Chi-square test or Fisher's exact test. PFS and OS were calculated using the Kaplan-Meier method and compared using the log-rank test. Univariate and multivariate Cox regression was used to estimate prognostic parameters for PFS and OS. For multivariate analysis of PFS and OS, all predictors with a p value < 0.2 in univariate analysis were incorporated. Two-sided p-values less than 0.05 were considered statistically significant. All analysis was performed using the SAS 9.1 software package (SAS Institute, Inc., Cary, NC, USA).