Introduction

In contrast to other gastrointestinal malignancies such as colorectal or gastric cancer, no predictive biomarkers have been defined for patients with advanced pancreatic cancer (PC) [1]. Several, mainly small and retrospective, studies have shown that the KRAS exon 2 mutation status may have a prognostic role in PC, with KRAS wildtype (WT) patients having an improved overall survival (OS) [2, 3]. Recently, KRAS data from a large, well conducted Japanese study were reported (n = 242). Within a single-center analysis, it was shown that KRAS WT patients with unresectable PC have a survival benefit (HR 1.78, p = 0.03) compared to patients with a KRAS mutation [4]. Of note, none of the patients included in this study received treatment with the novel anti-epidermal growth factor receptor (EGFR) tyrosine kinase inhibitor erlotinib [1]. This is important, since a Korean study published in 2011 found that KRAS may rather be a predictive biomarker for the efficacy of erlotinib than a prognostic factor in patients with advanced PC [5]. In this report of Kim and co-workers (n = 136), PC patients with a KRAS mutation showed a worse response compared to WT patients (11 vs. 26 %) and also a reduced OS (5.8 vs. 8.0 months, p = 0.001). However, the survival advantage was observed only in the subgroup of patients treated with gemcitabine/erlotinib (5.2 vs. 9.7 months; p = 0.002), whereas no OS difference based on KRAS mutation status was obvious in patients treated without erlotinib (7.0 vs. 7.0 months; p = 0.121) [5]. Thus, it currently remains unclear if KRAS is a prognostic biomarker (as suggested by Ogura and colleagues [4]) or if KRAS is rather a predictive marker for erlotinib efficacy (as hypothesized by Kim and co-workers [5]). Furthermore, within the Ogura study, early evidence for a difference in OS outcome based on the type of KRAS mutation was also described, as PC patients with tumors carrying the G12V mutation had a significantly longer OS compared to patients with a G12D or a G12R mutation [4].

Between May 2006 and December 2008, the AIO study group conducted a large, multicenter, cross-over phase III trial in advanced PC (AIO-PK0104). The clinical study results were reported recently [6]. This trial was accompanied by a large translational research project that investigated the KRAS exon 2 mutation status, EGFR expression by immunohistochemistry (IHC), EGFR amplification by fluorescent in situ hybridization (FISH), PTEN expression by IHC, as well as the polymorphisms in the EGFR intron 1 and exon 13 (R497K). These translational data were correlated with the clinical endpoints time-to-treatment failure, OS and skin rash and were published recently [7]. The main conclusion derived from this large translational project was that KRAS WT status was correlated with an improved OS (HR 1.68, p = 0.005). Of note, within AIO-PK0104, all patients received a 1st-line erlotinib containing treatment, making it impossible to definitely assess if the observed survival benefit for KRAS WT patients in our study population was a prognostic or a predictive phenomenon.

However, as one might expect that a predictive biomarker for the efficacy of a specific drug also shows a significant correlation with objective treatment response (e.g., KRAS and cetuximab in colorectal cancer, EGFR mutation status and erlotinib in non-small cell lung cancer), we re-analysed our data within a post hoc explorative biomarker analysis for this endpoint specifically in front of the recent intriguing results of the Ogura group [4, 8, 9]. Second, the correlation of KRAS status with OSc (the OS calculated from the start of the pre-defined 2nd-line chemotherapy) was also assessed. Of note, all patients receiving 2nd-line treatment with either capecitabine or gemcitabine within this cross-over study had a previous exposure to erlotinib, but the 2nd-line regimens were free of erlotinib. Third, a subgroup analysis on the role of the type of KRAS mutation on OS was performed also in our patient population.

Methods

The clinical trial results as well as the translational data from AIO-PK0104 on KRAS, EGFR and PTEN have been reported previously [6, 7]. In brief, AIO-PK0104 was a prospective, randomized, cross-over phase III trial including 281 patients with a histologically or cytologically confirmed diagnosis of treatment-naïve, advanced PC. Patients were treated with either 1st-line gemcitabine/erlotinib followed by 2nd-line single-agent capecitabine (‘reference arm’) or with 1st-line capecitabine/erlotinib followed by 2nd-line single-agent gemcitabine (‘experimental arm’). The study had ethics committee approval and was conducted according to GCP/ICH guidelines and to the Declaration of Helsinki. AIO-PK0104 was registered at ClinicalTrials.gov, number NCT00440167.

Translational biomarker study

All biomarker analyses from AIO-PK0104 were performed centrally at the University of Munich, Department of Pathology (Max-Borst Laboratory for Cancer Research) by AJ, TK and JN employing archival formalin fixed paraffin embedded (FFPE) tumor tissue from 206 (72 %) of the randomized patients. The detailed molecular methods for the analyses of KRAS exon 2 (codon 12 and 13) mutation status, EGFR expression by IHC, EGFR gene amplification by FISH, PTEN expression by IHC, as well as the analyses on the EGFR intron 1 and the exon 13 R497K polymorphism have been published in detail elsewhere [7].

Statistical analyses

Statistical analyses were performed at the University of Munich, Institute of Medical Informatics, Biometry and Epidemiology (by RPL). For the current post hoc retrospective subgroup analysis, the already available biomarker data (as dichotomous variables) were correlated with objective response to the erlotinib containing 1st-line treatment (which was determined by RECIST, version 1.0, in this study). These analyses were performed for the whole translational study population and also separated for each of the two treatment arms. Furthermore, a time-to-event Kaplan–Meier analysis for the exploratory endpoint OSc was conducted. OSc thereby was defined as the time frame form start of the pre-defined 2nd-line chemotherapy (without erlotinib) until death from any cause. Finally, an exploratory subgroup analysis on the impact of different types of the KRAS codon 12 mutation on OS was also carried out. Differences between groups were compared applying Fisher’s exact test and the log-rank test with a two-sided p value of less than 0.05 being regarded as statistically significant.

Results

Correlation of KRAS status with objective response

Objective response (to 1st-line treatment) was correlated with molecular marker results (dichotomous variables) in the translational study population (Table 1). For all patients (irrespective of the study arm) no significant association of the KRAS mutation status with treatment response (CR + PR for WT vs. mutation: 16 vs. 9 %, p = 0.40) was found. In the subgroup analyses with regard to the two different treatment arms, also no correlation of KRAS with response was detectable. KRAS WT patients treated within the gemcitabine/erlotinib arm (n = 30) had an objective response rate (CR + PR) of 20 % compared to a response rate of 15 % seen in KRAS mutant patients (n = 62; p = 0.77). A similar correlation (although with lower response rates) was observed in patients receiving 1st-line capecitabine/erlotinib: in this subgroup, KRAS WT patients (n = 22) had a response rate of 9 % compared to 3 % for patients with KRAS mutated tumors (n = 58; p = 0.30).

Table 1 Correlation of objective response (to 1st-line treatment) with molecular marker results (dichotomous variables) in the AIO-PK0104 translational study population
Full size table

For EGFR protein expression determined by IHC, EGFR gene amplification determined by FISH, and for the PTEN protein expression level (IHC), no significant response correlations were found. This also holds true for the two investigated EGFR gene polymorphisms in intron 1 and exon 13 of the EGFR gene (see detailed results in Table 1).

Correlation of KRAS status with OSc

In our previous investigation, KRAS was the only biomarker that had a statistically significant impact on OS in univariate analysis (KRAS mutation vs. KRAS WT: 5.7 vs. 7.9 months; HR 1.68, 95 % CI 1.17–2.41, p = 0.005) [7]. Overall, 140 out of the 281 (50 %) randomized study patients started the pre-defined 2nd-line chemotherapy within the AIO-PK0104 trial [6]; the KRAS mutational status was available from 91 of the 140 patients. These 91 patients were finally included in the analysis of OSc: the median survival time estimated form the start of the 2nd-line cross-over—non-erlotinib containing—chemotherapy in patients with a KRAS mutation (n = 62) was estimated with 3.3 months, whereas in patients with KRAS WT (n = 29) median OSc was 5.4 months (HR 1.47, 95 % CI 0.93–2.33, p = 0.10; Fig. 1).

Fig. 1
figure 1

Overall survival calculated form the start of 2nd-line cross-over (non-erlotinib containing) chemotherapy (OSc) by KRAS mutation status

Full size image

Impact of the type of KRAS mutation on survival outcome

Of the 173 patients analyzed for KRAS mutation status, 52 (30 %) had a KRAS WT and 121 (70 %) a KRAS mutation [6]. Within the KRAS mutation subgroup 81 (67 %) patients had the G12D mutation, 26 (21 %) the G12V mutation, 12 (10 %) the G12R mutation, and one (1 %) patient each the G12A or G12C mutation, respectively. Median OS was estimated with 5.3 months in the G12D subgroup, with 6.6 months for patients with the G12V mutation and 7.7 months in patients with tumors carrying the G12R KRAS mutation (log rank tests: global p value = 0.18; G12D vs. G12V: p = 0.07; G12D vs. G12R: p = 0.50; G12V vs. G12R: p = 0.52).

Discussion

Based on the novel data from Ogura and colleagues, the AIO investigators decided to re-analyse their translational KRAS data from AIO-PK0104 for the endpoints objective response (to 1st-line therapy), overall survival calculated from the start of the pre-defined 2nd-line (“cross-over”) chemotherapy (OSc) and for the correlation of KRAS codon 12 mutation subtypes with OS. The previous translational investigations from this prospective multicenter phase III study were analyzed for the endpoints time-to-treatment failure, OS and skin rash [7]. Within the current exploratory analysis we found no significant correlation of KRAS status with objective response (determined by RECIST) to erlotinib-containing 1st-line therapy, supporting the assumption of Ogura et al. that KRAS may represent a prognostic biomarker in advanced PC. In this context one should keep in mind that in PC (in contrast to other diseases like colorectal and non-small cell lung cancer) no data are available that show an increase of the objective response rate with the use of an anti-EGFR treatment like erlotinib in an unselected patient population [8, 9]. Within the pivotal trial PA.3 however, a trend for an increase of the disease control rate with the use of erlotinib compared to placebo (58 vs. 49 %, p = 0.07) was reported [1].

Within the AIO-PK0104 translational study population a KRAS WT status was associated with an improved OS in univariate analysis, resulting in a nearly identical HR as the one reported by Ogura and co-workers (1.68 and 1.78, respectively) [4, 6]. All patients treated with 1st-line therapy in our study received the novel anti-EGFR tyrosine kinase inhibitor erlotinib, making it impossible to assess if KRAS had a value as a prognostic or predictive biomarker in the AIO-PK0104 population. The latter assumption was supported by data from a Korean study, showing that a survival advantage for KRAS WT was obtained only in the subgroup of patients treated with gemcitabine/erlotinib, whereas no OS difference obvious in patients treated without erlotinib [5]. When we assessed the OSc in our patient population a similar trend for an OS advantage of KRAS WT patients is apparent as in the analysis for OS in AIO-PK0104 as well as in the Ogura study (Fig. 1). This difference however was statistically not significant (HR 1.47, p = 0.10), possible related to the smaller number of patients analyzed. The clearly separating Kaplan–Meier curves however indicate that there may be a biological KRAS effect also in this patient population that received a non-erlotinib containing 2nd-line chemotherapy with either gemcitabine or capecitabine. Of note, this observation should be regarded carefully since all patients included in this analysis received previous erlotinib-based 1st-line therapy. Nevertheless, the lack of correlation with response for KRAS and the trend for a survival benefit in non-erlotinib treated patients both support the assumption that KRAS may rather be a prognostic than a predictive biomarker in advanced PC. Of note, data on the role of KRAS in PC still remain inconclusive, since there also are studies in the literature that showed no impact of KRAS on OS: in a subgroup analysis of the erlotinib pivotal trial PA.3 KRAS was neither a prognostic nor a predictive biomarker as reported by Da Cunha Santosa and colleagues [10].

Based on preliminary data from colorectal cancer there is evidence that different KRAS codon 12 mutation subtypes may be related to differences in outcome. However, in colorectal cancer the G12D mutation is possibly related to a more favorable survival outcome, whereas in PC this mutation was associated with an inferior OS compared, e.g., to the G12V mutation [4, 11]. Our exploratory data on this topic showed no statistically significant differences and the interpretation of these subgroup data should—at least in our opinion—be performed critically (also with regard to the low numbers analyzed). At least a trend for a difference in OS between G12D and G12V was observed (p = 0.07), similar to the observation of Ogura and co-workers [4].

Pre-clinical evidence suggested that EGFR gene polymorphisms may have an impact on the response to EGFR inhibitors like erlotinib [12, 13]. Specifically, the EGFR intron 1 polymorphism was suspected to mediate—at least in vitro—the erlotinib response in PC [13]. These data derived from cell lines and mouse models were not confirmed by our translational analysis from AIO-PK0104. None of the five makers assessed besides KRAS had a significant impact on response to 1st-line treatment (as summarized within Table 1). The reasons for these conflicting data remain unclear as the hypothesis-generating pre-clinical data could not be transferred into the clinical setting [13].

In conclusion, data obtained from the current post hoc, exploratory translational subgroup analysis from AIO-PK0104 support the hypothesis that in PC KRAS should rather be regarded as a prognostic than a predictive biomarker for the efficacy of erlotinib. An external prospective validation (ideally conducted within a multicenter biomarker trial) of these findings remains pivotal.