SEPROGADIC - serum protein-based gastric cancer prediction model for prognosis and selection of proper adjuvant therapy

Abstract

Gastric cancer (GC) patients usually receive surgical treatment. Postoperative therapeutic options such as anticancer adjuvant therapies (AT) based on prognostic prediction models would provide patient-specific treatment to decrease postsurgical morbidity and mortality rates. Relevant prognostic factors in resected GC patient's serum may improve therapeutic measures in a non-invasive manner. In order to develop a GC prognostic model, we designed a retrospective study. In this study, serum samples were collected from 227 patients at a 4-week recovery period after D2 lymph node dissection, and 103 cancer-related serum proteins were analyzed by multiple reaction monitoring mass spectrometry. Using the quantitative values of the serum proteins, we developed SEPROGADIC (SErum PROtein-based GAstric cancer preDICtor) prognostic model consisting of 6 to 14 serum proteins depending on detailed purposes of the model, prognosis prediction and proper AT selection. SEPROGADIC could clearly classify patients with good or bad prognosis at each TNM stage (1b, 2, 3 and 4) and identify a patient subgroup who would benefit from CCRT (combined chemoradiation therapy) rather than CTX (chemotherapy), or vice versa. Our study demonstrated that serum proteins could serve as prognostic factors along with clinical stage information in patients with resected gastric cancer, thus allowing patient-tailored postsurgical treatment.

Figure 1

Study workflow and design. (a) Standard care for gastric cancer patients. MRM assay is performed on serum collected after surgery and used to build SEPROGADIC prognosis models. (b) Multistage MRM-MS workflow for discovery, triage, assays, and prognosis module-building. The 284 marker candidates were obtained by combining data from literature search and MS/MS profiling of GC plasma. MC: Marker Candidate; NP: Normalization Protein; ND: Non-Depletion; MD: MARS14 affinity column Depletion of high abundant proteins.

Figure 2

Serum protein abundances in Plasma Proteome Database (top) and normalized protein amount measured by MRM-MS (bottom). The first 20 proteins are ND panel proteins and the rest 73 proteins are MD panel proteins. Data are presented with box plots.

Figure 3

Kaplan-Meier plots of MD panel in SEPROGADIC prognostic module. (a) Classification of all the patients into two risk groups by the median of prognostic module (PM) scores. (b) All patients: low risk group (n = 114; observed: 7), high risk group (n = 113; observed: 37), P = 1e-07. (c) Patients at stage 1b: low risk group (n = 47; observed: 2), high risk group (n = 11; observed: 2), P = 0.063. (d) Patients at stage 2: low risk group (n = 44; observed: 2), high risk group (n = 44; observed: 9, P = 0.021. (e) Patients at stage 3: low risk group (n = 22; observed: 3), high risk group (n = 39; observed: 14), P = 0.069. (f) Patients at stage 4: low risk group (n = 1; observed: 0), high risk group (n = 19; observed: 12), P = 0.308.

Figure 4

Kaplan-Meier plots of MD panel in SEPROGADIC-AT selection module. (a) AT module scores at the three endpoints in the CCRT group (n = 120) and CTX group (n = 107). Patients are divided into four subgroups (true CCRT, false CCRT, true CTX and false CTX) based on the median of AT scores. NR: no recurrence, C: censored, R: recurrence. (b) K-M plots of four subtypes. True CCRT (n = 63; observed: 3), false CCRT (n = 57; observed: 14), true CTX (n = 56; observed: 7), false CTX (n = 51; observed: 20).