Personalized analysis of minimal residual cancer cells in peritoneal lavage fluid predicts peritoneal dissemination of gastric cancer

Abstract

Peritoneal dissemination (PD) is a major type of gastric cancer (GC) recurrence and leads to rapid death. Current approaches cannot precisely determine which patients are at high risk of PD to provide early intervention. In this study, we developed a technology to detect minimal residual cancer cells in peritoneal lavage fluid (PLF) samples with a personalized assay profiling tumor-specific mutations. In a prospective cohort of 104 GC patients, the technology detected all the cases that developed PD with 100% sensitivity and 85% specificity. The minimal residual cancer cells in PLF were associated with a significantly increased risk of PD (HR = 145.13; 95% CI 20.20-18,435.79; p < 0.001), which was the strongest independent predictor over pathologic diagnosis and cytological diagnosis. In pathologically high-risk (pT4) patients, the PLF mutation profiling model exhibited a greater specificity of 91% and a positive predictive value of 88% while retaining a sensitivity of 100%. This approach may help in the postsurgical management of GC patients by detecting PD far before metastatic lesions grow to a significant size detectable by conventional methods such as MRI and CT scanning.

Keywords: Gastric cancer; Minimal residual disease; Peritoneal dissemination; Peritoneal lavage fluid; Personalized mutation assay.

Conflict of interest statement

Yuchen Jiao, Dongbing Zhao, Pei Wang, Qianqian Song and Pinli Yue have filed patents/patent applications based on the technology and data generated from this work. Yuchen Jiao is one of the cofounders, has owner interest in Genetron Holdings, and receives royalties from Genetron. The remaining authors disclose no conflicts.

© 2021. The Author(s).

Figures

Fig. 1

Cancer cell fraction model and background noise. a Cancer cell fraction model. A model for estimating the cancer cell fraction based on allele frequency and sequencing depth of somatic mutations in tumor tissue and paired PLF samples. MAF: mutant allele frequency; Pti : MAF in solid tumor tissue; Pci : MAF in corresponding peritoneal lavage fluid (PLF); Di : sequencing depth in PLF; Ai : mutation read number in PLF; Xi : observed reads with a mutation in PLF; and R: overall cancer cell fraction. b The linear correlation between the theoretical and estimated cancer cell fractions. Each dilution was repeated three times. The blue dots highlight the fractions above the limit of detection (PLC/PRF/5 cell fraction = 0.001%, 0.005%, 0.05%, 0.5%, 5% and 33%). The red dots highlight the fractions under the limit of detection (PLC/PRF/5 cell fraction = 0.0001%, 0.0003%). c Background noise observed in the cancer cell fraction model at 0% PLC/PRF/5 cell input among the 20 independent replicates. d The distribution of mutations detected in the cancer cell fraction model. We profiled 20 SNPs to calculate the estimated dilution ratio with the model. Left panel: Heatmap illustrating detected (yellow) and undetected mutations (green) for each dilution. Right panel: the number of detected mutations. e Biological noise of the 104 PLF samples from patients. The cancer cell fraction for each sample was calculated based on nontumor-specific mutations

Fig. 2

The performance of the cancer cell fraction model in gastric cancer patients. a Clinical and histopathologic parameters, somatic mutations and cancer cell fraction for all patients. Top panel, summary of the frequencies of the tracked mutations in tumor and matched PLF samples from 104 patients. Blue bar, the tumor frequency of each tracked mutation. Frequency values are shown on the left vertical axis. Red bar, the detected peritoneal lavage fluid frequency of each tracked mutation. Frequency values are shown on the right vertical axis. The clinical outcome of patients is indicated under the bar. Middle panel, the cancer cell fraction of each patient. Bottom panel, clinical and histopathological characteristics. b The distribution of the cancer cell fraction in patients with peritoneal dissemination (n = 27), lymphatic metastasis (n = 6) or no recurrence (n = 71). Reported p values were computed using a 2-tailed Wilcoxon Mann–Whitney U test. ***p < 0.001; *p < 0.05. c Binary results of the PLF mutation profiling model and clinical risk factors. Pathologic diagnosis was defined as high (pT4) or low (pT0–3) according to the standard criteria. PD, peritoneal dissemination; PPV, positive predictive value; NPV, negative predictive value. d The Kaplan–Meier survival analysis shows the probability of recurrence-free survival as determined by MRD analysis of PLF (n = 98). e Kaplan–Meier estimates of overall survival for 104 gastric cancer patients based on MRD analysis of PLF. f The Kaplan–Meier survival analysis shows the probability of recurrence-free survival (RFS) as determined by MRD analysis of PLF in stage pT4 patients (n = 56) (for peritoneal dissemination). Shaded areas in the Kaplan–Meier plots indicate 95% CIs. HR: hazard ratio; CCF: cancer cell fraction