Performance and prognostic utility of the 92-gene assay in the molecular subclassification of ampullary adenocarcinoma

Adenocarcinomas of the ampulla of Vater are uncommon gastrointestinal cancers with heterogeneous outcomes. Histological classification of ampullary adenocarcinomas, and subsequent clinical management, is challenging due to the anatomical complexity and convergence of three distinct epithelial types in this region [1]. While previous studies suggest a role for histological classification of ampullary carcinoma into intestinal and pancreaticobiliary subtypes through examination of morphological features, the variability in subtype frequencies observed across studies has contributed to the unclear prognostic utility of histological examination alone [2–4]. Moreover, inter-pathologist concordance for subtyping ampullary adenocarcinomas is reportedly poor, with k value of 0.57 for intestinal and pancreaticobiliary subtypes, and a k value of 0.09 for the mixed subtype [5].

In addition to histological examination of morphological features, immunohistochemical (IHC) stains have been utilized to substantiate the intestinal versus pancreaticobiliary morphology of these adenocarcinomas including CK7, CK17, CK20 and CDX-2, and MUC1/2 [6]. Recently, Chang et al. described clinically relevant histomolecular phenotypes using both histomorphology and differential protein expression of CDX2 and MUC1 in patients with resected ampullary cancers [3]. While the histomolecular phenotype represents progress in classifying ampullary adenocarcinomas, IHC has challenges including tissue heterogeneity and antigenicity, interpretation of staining patterns, and inter/intra- observer variability [5, 7, 8]. Given the inherent subjective nature of both morphological evaluation and IHC, there exists a need for standardized approaches for tumor classification.

Genomic classifiers based on gene expression profiling of known reference tumor types have reported performance accuracies in the range of 80–90 % [9–13]. The 92-gene assay (CancerTYPE ID® Biotheranostics, Inc.) is a clinically validated cancer classifier that measures and interrogates the collective expression of 92 genes to determine tumor type and subtype utilizing a computational algorithm trained on a reference database of more than 2000 tumors [8, 13]. The objectives of the current study were to 1) evaluate the performance of the 92-gene classifier to subtype ampullary adenocarcinomas into intestine and pancreaticobiliary types, and 2) compare the prognostic utility of histological examination (morphology alone), histomolecular phenotype (morphology plus IHC) and the 92-gene classifier (genomic classification) to stratify ampullary adenocarcinomas into intestinal and pancreaticobiliary subtypes.

Identification of patient subgroups based on clinicopathological and molecular characteristics is a validated approach for both tumor subclassification [17] and for the optimal use of targeted therapies [18–23]. The ability to stratify patients into different prognostic and/or treatment groups is particularly important in cancers associated with a diverse range of outcomes such as ampullary adenocarcinomas, which have classically been separated into intestinal or pancreaticobiliary subtypes based on histomorphological assessment [3, 4, 24, 25]. In this blinded study, we demonstrate that a 92-gene cancer classifier showed favorable performance in the classification of ampullary adenocarcinomas into intestinal and pancreaticobiliary histological subtypes, and that these subtypes are prognostically distinct with discrete survival outcomes. The 92-gene cancer classifier represents a non-subjective classification method that demonstrated comparative prognostic performance to histological subtyping (morphology alone) and histomolecular phenotyping (morphology plus IHC).

Demonstration that the 92-gene assay classified ampullary tumors into two statistically significant prognostic groups is a novel finding of this study. Although this work did not demonstrate a single method to have significantly improved ability to prognostically stratify ampullary adenocarcinoma, the comparable prognostic utility of the 92-gene classier with histological subtyping and histomolecular phenotyping suggests that the prognostic capability is partly consistent with histological subtyping. Intriguingly the 92-gene classifier demonstrated an improved ability to identify prognostic subgroups within poorly differentiated ampullary adenocarcinomas, though the sample size for this analysis was small.

Currently, the classification of ampullary adenocarcinomas is performed by non-standardized examination of tumor histomorphology. However, given the subjective nature of histological classification it remains uncertain which methodological approach to ampullary stratification best reflects biology. This variation is highlighted by the substantial differences in histological subtype designations that were observed amongst the three classification methods. Whereas histology alone designated 48 % of the cohort as pancreaticobiliary, histomolecular phenotyping designated 26 % and the 92-gene classifier designated 36 % of the cohort as pancreaticobiliary. Though histomolecular phenotyping also considers differential protein marker expression, this method fundamentally relies upon histomorphological examination. In addition, the challenges of immunohistochemical staining interpretation and the establishment of optimal cut-points for both CDX2 and MUC1 immune reactivity are limitations for the application of histomolecular phenotyping [3, 26]. The 92-gene classifier exhibited sensitivities of 64 % (0.46–0.79) and 85 % (0.66–0.94), and specificities of 88 % (0.70–0.98) and 68 % (0.48–0.84) for the intestinal and pancreaticobiliary histological subtypes, respectively. The differential sensitivities and specificities for the 92-gene assay reflect the increased number of false positive pancreaticobiliary predictions (N = 9) within the ampullary-intestine histological subtype in contrast to the false positive intestine predictions (N = 3) within the ampullary-pancreaticobiliary histological subtype. Unlike the setting in which the primary cancer site is used as the clinical truth, the use of morphological subtype as the gold-standard reference may be limited by subjective determination.

The performance of the 92-gene classifier is promising for this off-panel tumor, particularly given that the reference training database did not include ampullary adenocarcinoma cases. The observed performance of the 92-gene classifier to subtype these ampullary tumors can be attributed to the discovery methodology of the classifier, which included a data-dependent search for gene combinations from whole genome expression profiling and a genetic algorithm that allowed the gene panel to evolve as a combination [27]. While reduced performance in off-panel tumor types such as ampullary adenocarcinoma is a potential limitation of gene expression based classifiers, these findings support the inherent scalability of the 92-gene assay to recognize a diverse range of tumor types. These data are generally consistent with the utilization of a dichotomous subclassification for ampullary adenocarcinomas. This finding is in agreement with a previous microarray analysis [25] and provides information on cellular context, which points to distinct epithelial origins of ampullary tumors with different outcomes.

Limitations of this study are that, in contrast to the recent histomolecular phenotyping by Chang et al. [3], this current study involves a smaller cohort of patients, likely contributing to the lack of significance for the 3 classifiers in multivariate analysis (Additional file 1: Table S1). Additional limitations of this study were its retrospective nature and use of both untreated and neoadjuvantly treated primary tumors. Further validation of these findings in a larger dataset is needed.

Based on the ability of the 92-gene classifier to measure gene expression in biologically-relevant pathways such as lineage commitment and signal transduction, an exploratory analysis was performed using the cases from this study as a training set to identify a 3-gene expression model that stratified this ampullary cohort into two statistically significant survival groups (HR = 3.42, P = 0.0012). Although a majority of cases in the poor survival group had a pancreaticobiliary histological subtype, a significant portion (33 %) of cases in this group had an intestinal histological subtype. Similarly, 30 % of the cases in the improved survival group had a pancreaticobiliary histological subtype, whereas the majority was intestinal (Additional file 2: Table S2). This 3-gene expression model suggests that the collective expression of IRX3, PYCR1, and TMPRSS3 reflects the additional biological importance of tumor invasion, intestinal differentiation, and cellular metabolism [28–34], in the prognosis of ampullary cancers. The increased expression of IRX3 and TMPRSS3 in the poor prognosis ampullary subset is consistent with the biological functions of these genes in promoting tumorigenesis and tumor invasion, respectively [29, 32]. The exploratory 3-gene signature represents a hypothesis-generating findings and further validation is needed.

At the present time the main clinical utility from stratifying ampullary adenocarcinomas results from improved prognostication. The clinical impact of classifying ampullary tumors to inform therapeutic decisions in the adjuvant or metastatic settings has not been proven. Histological subtyping of some ampullary cases in the European Study Group for Pancreatic Cancer (ESPAC)-3 prospective trial demonstrated improved DFS (P = 0.01), but not OS (P = 0.28). However, due to small number of cases with known histological subtype the outcomes stratified by chemotherapy type and histological subtype were not analyzed [35, 36]. Based upon the reproducible identification of not only distinct prognostic but also biologically unique ampullary subgroups, a clinical trial utilizing subtype guided therapy [e.g. fluoropyrimidine/oxalipliatin for the intestinal ampullary subtype and gemcitabine/cisplatin for pancreaticobiliary ampullary subtype has merit. Additionally, the role of novel approved agents and biomarkers with therapeutic intent needs to be explored in ampullary cancers (e.g. her-2 and RAS for intestinal subtype and IDH1, ARID-1, FGFR for biliary subtypes). One can envision that progress made in pancreas, intestinal and cholangiocarcinomas may directly impact subtypes of ampullary cancers.

In conclusion, the results of this study demonstrate that the 92-gene assay offers a standardized and reproducible approach beyond the review of morphology and immunohistochemistry to classify ampullary cancers [5, 8]. The 92-gene assay may potentially provide an improved method to classify the molecular subtype of unresectable periampullary cancers in which primary resection tissue is not present and thus the anatomic source of the cancer (pancreatic vs. biliary vs. duodenal vs. ampullary) cannot be determined. A potential next step in the clinical utility of the 92-gene assay is to integrate its use into the broader diagnostic armamentarium of periampullary cancers and allow the molecular subtype designation to guide tailored therapy.