Identification of a Five-Gene Signature for Predicting Survival in Malignant Pleural Mesothelioma Patients

Our plain-English summary of Identification of a Five-Gene Signature for Predicting Survival in Malignant Pleural Mesothelioma Patients, by Y. Bai, X. Wang, J. Hou, et al (Frontiers in Genetics, 2020).

The study

• Authors: Y. Bai, X. Wang, J. Hou, L. Geng, X. Liang, Z. Ruan, H. Guo, K. Nan, L. Jiang
• Journal: Frontiers in Genetics
• Published: 2020;11:899
• Cited by: 11 articles, as counted by NIH iCite
• Read it: PubMed · Publisher · Free full text

Abstract

Malignant pleural mesothelioma (MPM), predominantly caused by asbestos exposure, is a highly aggressive cancer with poor prognosis. The staging systems currently used in clinics is inadequate in evaluating the prognosis of MPM. In this study, a five-gene signature was developed and enrolled into a prognostic risk score model by LASSO Cox regression analysis based on two expression profiling datasets (GSE2549 and GSE51024) from Gene Expression Omnibus (GEO). The five-gene signature was further validated using the Cancer Genome Atlas (TCGA) MPM dataset. Univariate and multivariate Cox analyses proved that the five-gene signature was an independent prognostic factor for MPM. The signature remained statistically significant upon stratification by Brigham stage, AJCC stage, gender, tumor size, and lymph node status. Time-dependent receiver operating characteristic (ROC) curve indicated good performance of our model in predicting 1- and 2-years overall survival in MPM patients. The C-index was 0.784 for GSE2549 and 0.753 for the TCGA dataset showing moderate predictive accuracy of our model. Furthermore, Gene Set Enrichment Analysis suggested that the five-gene signature was related to pathways resulting in MPM tumor progression. Together, we have established a five-gene signature significantly associated with prognosis in MPM patients. Hence, the five-genes signature may serve as a potentially useful prognostic tool for MPM patients.

Copyright © 2020 Bai, Wang, Hou, Geng, Liang, Ruan, Guo, Nan and Jiang.

In plain English

This was a computer analysis of public gene-activity data from malignant pleural mesothelioma (MPM) tumors. It did not study asbestos exposure, and the authors note the datasets held no asbestos exposure history. Using two datasets (GSE2549: 40 tumor and 9 normal specimens; GSE51024: 41 tumors with 41 paired normal tissues), the authors found 225 genes that differed between tumor and normal tissue. They built a risk score from five of them (CDH2, CKS2, KIF11, KIF18B and LOX) and tested it in 85 MPM patients from the Cancer Genome Atlas (TCGA). High-score patients had shorter survival than low-score patients in GSE2549 (hazard ratio 3.909, 95% CI 1.797 to 8.503) and in TCGA (hazard ratio 3.929, 95% CI 2.295 to 6.726). The score stayed statistically significant after accounting for other clinical factors. The authors described its predictive accuracy (C-index) as moderate: 0.784 (95% CI 0.674 to 0.894) in GSE2549 and 0.753 (95% CI 0.698 to 0.808) in TCGA. The authors concluded the score could serve as a prognostic tool complementing current staging. They said clinical information was incomplete and that large prospective studies are needed.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: Bioinformatics analysis of publicly available gene expression datasets: a prognostic gene-signature model was built in two discovery datasets and checked in a separate validation dataset
• Who was studied: Malignant pleural mesothelioma tissue samples from public databases. GSE2549 had 40 MPM tumor specimens and 9 normal specimens (4 normal lung, 5 normal pleura). GSE51024 had 41 MPM tumor tissues and 41 paired normal tissues. The TCGA validation set had 85 MPM patients. Locations and dates of sample collection are not given.
• Exposure: Asbestos exposure was not studied. The study analysed tumor gene expression. The authors state that no asbestos exposure history was available for the patients in GSE2549 or TCGA.
• Main results:
  • 225 candidate genes differed between MPM tumor and normal tissue in both discovery datasets: 92 upregulated and 133 downregulated.
  • A five-gene risk score was built from CDH2, CKS2, KIF11, KIF18B and LOX.
  • In the GSE2549 discovery set, high-risk patients had shorter survival than low-risk patients (hazard ratio 3.909, 95% CI 1.797 to 8.503, log-rank P less than 0.0001).
  • In the TCGA validation set of 85 cases, high-risk patients had significantly shorter overall survival than low-risk patients (hazard ratio 3.929, 95% CI 2.295 to 6.726, log-rank P less than 0.0001).
  • After controlling for other clinical factors, the five-gene score remained an independent prognostic factor in both datasets: P less than 0.001 in GSE2549 and P less than 0.0001 in TCGA. Clinical data were missing for 1 GSE2549 patient and 19 TCGA patients, who were left out of these analyses.
  • Within both early-stage and advanced-stage Brigham subgroups, the score separated patients into better- and poorer-prognosis groups.
  • In GSE2549, the area under the curve for predicting survival was 0.858 (95% CI 0.727 to 0.990) at 1 year and 0.959 (95% CI 0.918 to 1) at 2 years.
  • In the TCGA set, the area under the curve was 0.821 (95% CI 0.726 to 0.916) at 1 year and 0.852 (95% CI 0.766 to 0.938) at 2 years.
  • The authors described the C-index of 0.784 in GSE2549 and 0.753 in TCGA as showing moderate predictive accuracy.
  • In GSE2549, the five-gene model's C-index was 0.784 (95% CI 0.674 to 0.894), against 0.770 for an earlier three-gene model by Zhou et al. (2019). No statistical difference was found.
  • In TCGA, the five-gene model's C-index was 0.753 (95% CI 0.698 to 0.808), higher than the 0.608 of Zhou et al.'s three-gene model (P 0.000649).
  • Compared with Zhou et al.'s model in TCGA, the areas under the curve were higher for the five-gene model, but the confidence intervals overlapped, indicating no statistical difference.
  • Several cell-division-related pathways were significantly enriched in high-risk patients in both the discovery and validation sets: RHO GTPases activate formins, mitotic spindle checkpoint, PLK1 pathway, and resolution of sister chromatid cohesion.
• The authors' conclusion: The authors concluded that they had identified a five-gene risk score model that is an objective and practical prognostic tool for MPM patients, independent of clinicopathological factors, which complements the current MPM staging system.
• Limitations the authors state:
  • Clinical information was incomplete; the GSE51024 discovery set had no accessible clinical information.
  • No age, asbestos exposure history or treatment information was available in GSE2549 or TCGA, so other possible prognostic factors could not be evaluated.
  • Only two discovery datasets were available, and the uneven tumor-to-normal split in GSE2549 may have biased results.
  • Only one validation dataset (TCGA) was used.
  • The model was built on microarray data but validated on RNA-seq data, which might affect performance; microarrays may miss some genes.
  • Rigorous validation in large prospective studies is needed.
  • The biological mechanisms of the five genes in MPM need further research.

Topics

Medical terms in this study, explained in our Lexicon: Pleural Mesothelioma, Asbestos Exposure, Prognosis, Staging, AJCC, Overall Survival.

Related studies

• Mesothelioma survival prediction based on a six-gene transcriptomic signature — K. Behrouzfar, S.E. Mutsaers, et al. (2024)
• Identification of prognostic and bone metastasis​-related alternative splicing signatures in mesothelioma — R. Huang, Z. Zheng, et al. (2021)
• Osteopontin modulates inflammation, mucin production, and gene expression signatures after inhalation of asbestos in a murine model of fibrosis — T. Sabo-Attwood, M.E. Ramos-Nino, et al. (2011)
• MicroRNA gene expression signatures in long-surviving malignant pleural mesothelioma patients — R.C. Lin, M.B. Kirschner, et al. (2016)
• Towards a potential pan-cancer prognostic signature for gene expression based on probesets and ensemble machine learning — D. Chicco, A. Alameer, et al. (2022)
• A gene signature linked to fibroblast differentiation for prognostic prediction of mesothelioma — J. Liu, Y. Lu, et al. (2024)

Browse more research in Frontiers in Genetics, or return to medical research.