A 25-gene classifier predicts overall survival in resectable pancreatic cancer

BACKGROUND

Pancreatic carcinoma is one of the most lethal human cancers. In patients with resectable tumors, surgery followed by adjuvant chemotherapy is the only curative treatment. However, the 5-year survival is 20%. Because of a strong metastatic propensity, neoadjuvant chemotherapy is being tested in randomized clinical trials. In this context, improving the selection of patients for immediate surgery or neoadjuvant chemotherapy is crucial, and high-throughput molecular analyses may help; the present study aims to address this.

METHODS

Clinicopathological and gene expression data of 695 pancreatic carcinoma samples were collected from nine datasets and supervised analysis was applied to search for a gene expression signature predictive for overall survival (OS) in the 601 informative operated patients. The signature was identified in a learning set of patients and tested for its robustness in a large independent validation set.

RESULTS

Supervised analysis identified 1400 genes differentially expressed between two selected patient groups in the learning set, namely 17 long-term survivors (LTS; ≥ 36 months after surgery) and 22 short-term survivors (STS; dead of disease between 2 and 6 months after surgery). From these, a 25-gene prognostic classifier was developed, which identified two classes ("STS-like" and "LTS-like") in the independent validation set (n = 562), with a 25% (95% CI 18-33) and 48% (95% CI 42-54) 2-year OS (P = 4.33 × 10-9), respectively. Importantly, the prognostic value of this classifier was independent from both clinicopathological prognostic features and molecular subtypes in multivariate analysis, and existed in each of the nine datasets separately. The generation of 100,000 random gene signatures by a resampling scheme showed the non-random nature of our prognostic classifier.

CONCLUSION

This study, the largest prognostic study of gene expression profiles in pancreatic carcinoma, reports a 25-gene signature associated with post-operative OS independently of classical factors and molecular subtypes. This classifier may help select patients with resectable disease for either immediate surgery (the LTS-like class) or neoadjuvant chemotherapy (the STS-like class). Its assessment in the current prospective trials of adjuvant and neoadjuvant chemotherapy trials is warranted, as well as the functional analysis of the classifier genes, which may provide new therapeutic targets.

Identification of a naturally processed HLA-A*02:01-restricted CTL epitope from the human tumor-associated antigen Nectin-4

Nectin-4 is a tumor antigen present on the surface of breast, ovarian and lung carcinoma cells. It is rarely present in normal adult tissues and is therefore a candidate target for cancer immunotherapy. Here, we identified a Nectin-4 antigenic peptide that is naturally presented to T cells by HLA-A2 molecules. We first screened the 502 nonamer peptides of Nectin-4 (510 amino acids) for binding to and off-rate from eight different HLA class I molecules. We then combined biochemical, cellular and algorithmic assays to select 5 Nectin-4 peptides that bound to HLA-A*02:01 molecules. Cytolytic T lymphocytes were obtained from healthy donors, that specifically lyzed HLA-A2(+) cells pulsed with 2 out of the 5 peptides, indicating the presence of anti-Nectin-4 CD8(+) T lymphocytes in the human T cell repertoire. Finally, an HLA-A2-restricted cytolytic T cell clone derived from a breast cancer patient recognized peptide Nectin-4145-153 (VLVPPLPSL) and lyzed HLA-A2(+) Nectin-4(+) breast carcinoma cells. These results indicate that peptide Nectin-4145-153 is naturally processed for recognition by T cells on HLA-A2 molecules. It could be used to monitor antitumor T cell responses or to immunize breast cancer patients.