Abstract

   Decidualization is a crucial process for successful embryo implantation
   and pregnancy in humans. Defects in decidualization during early
   pregnancy are associated with several pregnancy complications, such as
   pre-eclampsia, intrauterine growth restriction and recurrent pregnancy
   loss. However, the mechanism underlying decidualization remains poorly
   understood. In the present study, we performed a systematic analysis of
   decidualization-related genes using text mining. We identified 286
   genes for humans and 287 genes for mice respectively, with an overlap
   of 111 genes shared by both species. Through enrichment test, we
   demonstrated that although divergence was observed, the majority of
   enriched gene ontology terms and pathways were shared by both species,
   suggesting that functional categories were more conserved than
   individual genes. We further constructed a decidualization-related
   protein-protein interaction network consisted of 344 nodes connected
   via 1,541 edges. We prioritized genes in this network and identified 12
   genes that may be key regulators of decidualization. These findings
   would provide some clues for further research on the mechanism
   underlying decidualization.

Introduction

   In mammals, pregnancy begins with embryo implantation into uterus
   [[25]1]. In some species such as humans and mice, invasive embryo
   implantation is accompanied by a rapid remodeling process in the
   stromal compartment of uterus known as decidualization. Upon
   decidualization, stromal cells undergo proliferation and subsequent
   differentiation into large epithelioid cells characterized by
   cytoplasmic accumulation of glycogen and lipid droplets, as well as an
   expansion of Golgi complex and rough endoplasmic reticulum [[26]2,
   [27]3]. This process is marked by the secretion of decidual prolactin
   (PRL) and insulin-like growth factor binding protein 1 (IGFBP1)
   [[28]4]. From a functional perspective, decidualization contributes to
   uterine angiogenesis and hemostasis during trophoblast invasion and
   placenta formation [[29]5]. It also enables establishing maternal
   immunological tolerance to embryonic antigens [[30]6]. Defects in
   decidualization during early pregnancy are associated with several
   pregnancy complications, such as pre-eclampsia, intrauterine growth
   restriction and recurrent pregnancy loss [[31]7]. Therefore, it is
   imperative to gain a clear understanding of the molecular mechanism
   underlying decidualization in order to improve reproductive health.

   In humans, decidualization is initiated spontaneously in the secretory
   phase of menstrual cycle [[32]8]. If pregnancy is obtained,
   decidualization continues as the embryo undergoes implantation;
   otherwise, menstruation occurs. Most knowledge about human
   decidualization has come from studies using in vitro model systems. It
   is well established that decidualization can be induced in cultured
   endometrial stromal cells by incubation with progesterone after proper
   estrogen priming [[33]9]. Decidualization is mediated by a gradual
   increase in intracellular cAMP level and addition of cAMP analogues
   leads to a boost of this process [[34]10, [35]11]. The main advantage
   of the in vitro model systems is the ability to provide key information
   on a single cell type reaction. However, a cell growing as a layer in a
   dish does not have the complexity that a cell growing in vivo has. Most
   importantly, the uterus is a complex organ comprised of many cell
   types. Cultured stromal cells lack whole organ physiology and
   interacting microenvironment.

   Because of ethical restrictions and experimental difficulties, it is
   not practical for in vivo study of decidualization in humans. Direct
   analysis of decidualization heavily relies on mice. Unlike humans, the
   decidual reaction in mice is an embryo-dependent process [[36]8].
   Decidualization begins shortly after the blastocyst attaches to the
   uterine luminal epithelium. Interestingly, hormonally primed uterus can
   be stimulated by mechanical means (e.g. sesame oil) to trigger
   decidualization in the absence of an embryo [[37]12]. The mechanically
   decidualized endometrium, known as the deciduoma, is morphologically
   similar to the embryo-induced decidua, making it a good model of in
   vivo decidualization free of embryo contamination [[38]13, [39]14]. A
   previous study has compared the global gene expression profiles between
   deciduoma and decidua [[40]15]. Approximately 1,500 genes were
   differentially expressed by at least 1.2 folds. However, only 53 genes
   exhibited 2.5 folds or more, indicating that deciduoma is also similar
   to decidua at the transcriptome level.

   Nevertheless, a comprehensive analysis of the molecular mechanism
   underlying decidualization is lacking. A wealth of information remains
   hidden within published research articles, the number of which is
   growing fast. Recently, the text mining methodology has been
   implemented, providing a necessary means to retrieve these data in an
   automated way [[41]16]. Here we reported a systematic analysis of
   decidualization-related genes in humans and mice using text mining. Our
   study provides in-depth insights into the molecular mechanism
   underlying decidualization from a comparative aspect.

Methods

Text mining

   The PubMed database was used. We conducted a search with the following
   combinations of query key words: “decidualization OR decidual OR
   decidua OR deciduas OR deciduoma OR decidualized OR decidualizing”. The
   search tag “[Title/Abstract]” was added after each key word. The
   relevant articles were retrieved in XML format. This format makes
   information extraction more precise owning to the use of enclosed
   contents within tag pairs. For each article, titles and abstract texts
   were fetched using the dom4j XML parser class in JAVA. Abstract texts
   were further divided into sentences through a sentence tokenizer
   implemented in LingPipe (Alias-I, Inc). Text mining was performed at
   the sentence level. Species names were parsed based on a lexicon
   [[42]17]. All articles were classified into two categories according to
   species names mentioned in the texts: those studying human
   decidualization (including the monkey) and those studying mouse
   decidualization (including the rat). When no species name or multiple
   species names were detected, articles were classified manually.

   Gene mention recognition was performed using two different gene mention
   taggers, the hidden Markov model (HMM) tagger implemented in LingPipe
   and the ABNER tagger [[43]18] based on a machine learning system of
   conditional random fields (CRF). Gene mentions detected by both taggers
   were merged. Because researchers name the genes in a highly variable
   manner, we built a gene synonym dictionary from Entrez gene database
   [[44]19]. This dictionary was used for the gene name normalization
   process during which gene mentions were mapped to unique Entrez genes
   using exact string match. If multiple Entrez genes were linked to the
   same gene mention, the ambiguity was resolved manually. In order to
   reduce false positives, we required co-occurrence of decidualization
   mention and gene mention within a single sentence. In general, the
   abstract is sufficient for our text mining task, as it contains the
   most important findings of an article. However, articles on high
   throughput experiments often reveal a large number of genes which
   cannot be fully listed in the abstracts. For these articles, we
   downloaded full texts (as well as supplementary files if needed) and
   extracted gene mentions by hands. Finally, we compiled two gene sets:
   one is associated with human decidualization and the other one is
   associated with mouse decidualization. To ensure accurate and complete
   recording, each gene was checked manually and additional references