Abstract

   Clinicopathological characteristics alone are not enough to predict the
   survival of patients with cervical squamous cell carcinoma (CESC) due
   to clinical heterogeneity. In recent years, many genes and non-coding
   RNAs have been shown to be oncogenes or tumor-suppressors in CESC
   cells. This study aimed to develop a comprehensive transcriptomic
   signature for CESC patient prognosis. Univariate, multivariate, and
   Least Absolute Shrinkage and Selection Operator penalized Cox
   regression were used to identify prognostic signatures for CESC
   patients from transcriptomic data of The Cancer Genome Atlas. A
   normalized prognostic index (NPI) was formulated as a synthetical index
   for CESC prognosis. Time-dependent receiver operating characteristic
   curve analysis was used to compare prognostic signatures. A prognostic
   transcriptomic signature was identified, including 1 microRNA, 1 long
   non-coding RNA, and 6 messenger RNAs. Decreased survival was associated
   with CESC patients being in the high-risk group stratified by NPI. The
   NPI was an independent predictor for CESC patient prognosis and it
   outperformed the known clinicopathological characteristics,
   microRNA-only signature, gene-only signature, and previously identified
   microRNA and gene signatures. Function and pathway enrichment analysis
   revealed that the identified prognostic RNAs were mainly involved in
   angiogenesis. In conclusion, we proposed a transcriptomic signature for
   CESC prognosis and it may be useful for effective clinical risk
   management of CESC patients. Moreover, RNAs in the transcriptomic
   signature provided clues for downstream experimental validation and
   mechanism exploration.

   Keywords: transcriptome, signature, prognosis, CESC, angiogenesis

Introduction

   Cervical cancer (CC) is still the fourth most common cancer in women
   ([31]Ferlay et al., 2015). Despite developed countries are low epidemic
   areas of CC by virtue of easier accessibility of routine screening test
   and human papillomavirus (HPV) vaccination, CC is still the second
   leading cause of cancer death among women aged 20–39 years in the
   United States in 2015 ([32]Siegel et al., 2018). At present, clinical
   stage is the leading predictive characteristic for CC prognosis,
   although useful, significant variability is observed and the 5 years
   survival rate is still poor for women with advanced CC (30–40% for
   stage III and 15% for stage IV). Theoretically, clinicopathological
   characteristics are macroscopic emergence of molecules (e.g., genes,
   proteins) and CC patients with homogeneous clinical status may have
   completely diverse molecular patterns. Therefore, identification of
   robust and accurate molecular biomarkers for CC patient prognosis is
   valuable and in urgent need.

   By comprehensively characterizing various molecules (DNA-level,
   RNA-level, protein-level) in 100s of CC samples, The Cancer Genome
   Atlas (TCGA) has provided a comprehensive way to understand CC (The
   [33]Cancer Genome Atlas Research Network et al., 2017). Enormous
   multiple omics data make the discovery of potential biomarkers for CC
   diagnosis, treatment and prognosis possible. Several studies have
   investigated the molecular signatures for CC prognosis based on the
   expression of CC genome. [34]Hu et al. (2010) profiled 96
   cancer-related microRNAs (miRNAs) in 102 CC samples and firstly
   proposed a two-miRNA expression signature for predicting the overall
   survival (OS) of CC patients. [35]How et al. (2015) measured the miRNA
   omics of CC samples by miRNA arrays and proposed a prognostic
   nine-miRNA expression signature in their training set. However, the
   prognostic value of the nine-miRNA expression signature could not be
   validated in an independent cohort ([36]How et al., 2015). [37]Liu et
   al. (2016), [38]Liang et al. (2017), [39]Ma et al. (2018), and [40]Ying
   et al. (2018) proposed a seven-miRNA expression signature, a
   three-miRNA expression signature, a three-miRNA expression signature,
   and a 2 two-miRNA expression signature for CC prognosis based on TCGA
   miRNA sequencing data, respectively. [41]Huang et al. (2012) profiled
   1440 human tumor related gene transcripts using custom oligonucleotide
   microarrays in 100 CC samples and identified a prognostic seven-gene
   expression signature. Based on TCGA gene sequencing data, [42]Li et al.
   (2017b, [43]2018) proposed a two-histone family gene signature and
   further proposed another independent gene signature to predict the OS
   of CC patients.

   However, some limitations should be noticed: (1) Previous studies
   focused on single omics independently, and there lacks a whole
   transcriptomic analysis which may provide more comprehensive and robust
   discovery ([44]Hu et al., 2010; [45]Huang et al., 2012; [46]How et al.,
   2015; [47]Liu et al., 2016; [48]Li et al., 2017b, [49]2018; [50]Liang
   et al., 2017; [51]Ma et al., 2018; [52]Ying et al., 2018). (2)
   Prognostic miRNA signatures identified based on the same data source
   without cross-references are very different ([53]Liu et al., 2016;