Spontaneous preterm birth (sPTB) mainly consists of spontaneous preterm labor (sPTL) and preterm premature rupture of the membranes (PPROM). It refers to delivery that occurs before 37 gestational weeks (GWs) and is the leading cause of perinatal morbidity and mortality worldwide (Lawn et al., 2005). Etiologically, sPTB has many causes, including intra-amniotic infection, decidua senescence, and breakdown of maternal-fetal tolerance. The recognized risk factors underlying PPROM include physiologic weakening of the fetal membranes associated with apoptosis near term; dissolution of the amniochorionic matrix exacerbated by contraction-induced shearing forces; infection and inflammation resulting from ascending genital tract colonization initiating a cytokine cascade that triggers membrane degradation; protease production and dissolution of the extracellular matrix (ECM); placental abruption with decidua thrombin expression triggering thrombin-thrombin receptor interactions and increasing choriodecidual protease production; and membrane stretching that may increase amniochorionic cytokine and protease release (Charles and Edwards, 1981; Skinner et al., 1981; Lavery et al., 1982; Taylor and Garite, 1984). The degradation of fetal membranes involved in sPTB is mediated through the activation of Toll-like receptors (TLRs) and causes an increase of matrix metalloproteinases (MMPs; Geraghty et al., 2011; Sandig and Bulfone-Paus, 2012). MMP1 and MMP8 are collagenases that have been found to degrade collagen types I–III and are upregulated in the amnion and chorion (Menon and Fortunato, 2004), which leads to collagenolysis and a decrease in the collagen content of fetal membranes (Draper et al., 1995). The increase in collagen solubility contributes to the remodeling of the ECM and further results in cervical softening and fetal membrane activation (Pollock et al., 1991). Collagen provides the major structural support for the fetal membranes, which is formed by the amnion and chorion. In addition, preterm contractions can accelerate the separation of the amnion and chorion, and then reduce membrane tensile strength, whereas cervical dilation can cause exposure of the membranes to vaginal microorganisms and reduce underlying tissue support (Strohl et al., 2010).

Genetic factors associated with PPROM have been reported. A significant association of a single nucleotide polymorphism (SNP) was found at the genes MMP1 and MMP8, CARD15, TLR4, and SERPINH1 among PPROM cases (Fujimoto et al., 2002; Wang et al., 2004, 2006). More studies have been carried out on sPTL. The gene loci of ABCB11, BBS5, FSTL5, CSMD3, NTS, KLHL1, and NCAM2, in addition to duplications at the loci of OR4P4, OR4S2, OR4C6, and RASSF7, have been shown to be associated with sPTL (Biggio et al., 2015).

Although non-coding RNAs (ncRNAs) are defined by the lack of a protein-coding potential, they have been found to play important roles in many biological processes (Mattick, 2009; Lipovich et al., 2010). The long non-coding RNAs (lncRNAs) are a subtype of ncRNAs with transcripts that are more than 200 nucleotides long without obvious protein-coding potential. Occasionally, lncRNAs may be translated to produce short peptides of unknown function (Fatica and Bozzoni, 2014; Ingolia et al., 2014). LncRNAs predominantly localize to the nucleus and have a lower level of expression than protein-coding regions of genes (Djebali et al., 2012). Based on the biological characteristics of transcription loci and their relationship with the associated genes, lncRNAs can be classified as exonic, intronic, or intergenic overlapping transcripts, in either sense or antisense orientation. LncRNAs may modify the expression of genes and be involved in diverse cellular processes including cell differentiation, imprinting control, and immune responses (Wilusz et al., 2009; Archer et al., 2015). The regulatory function of lncRNAs lies in their ability to alter the expression of DNA in a site-specific manner and, at the same time, bind to different proteins, bridging chromosomes, and protein complexes (Rinn and Chang, 2012; Geisler and Coller, 2013). Evidence increasingly supports the linkage of dysfunctions of lncRNAs to many human diseases, including neurodegenerative, psychiatric diseases (Faghihi and Wahlestedt, 2009), cardiovascular disease (Annilo et al., 2009), and immune dysfunction and auto-immunity (Kino et al., 2010). In our previous study, lncRNAs that are differentially expressed in human placentas delivered from PPROM and sPTL were found to be involved in more than 20 functional pathways (Luo et al., 2013). The patterns of differentially expressed lncRNAs and pathways identified from placentas of PPROM and sPTL were similar to those we observed in our study of human miscarriages (Wang et al., 2014) and of a viral-infected mouse model (Pan et al., 2015), suggesting that deregulation and dysfunction of the ubiquitin-proteasome-collagen (CUP) pathway may be one of the pathogenic mechanisms underlying the adverse outcomes of pregnancies, including PPROM.

On the basis of these findings, we hypothesized that the epigenetic regulatory role of lncRNAs in the ubiquitin proteasome system (UPS) and collagen remodeling is that they are involved in the CUP pathway in sPTB, including PPROM (Zhong et al., 2015). To test our hypothesis, we studied the lncRNAs and lncRNA-associated messenger RNAs (mRNAs) and identified gene mutations/variations associated with the CUP pathway.

The current study is a continuing investigation of the regulatory role of lncRNAs in the molecular pathogenesis of sPTB, based on our previous studies, in which we identified differential expression of lncRNAs in sPTL and PPROM (Luo et al., 2013; Pan et al., 2015). In the current study, we analyzed the DEPs of the lncRNAs and the mRNAs that are associated with the CUP pathway that may also associate with PPROM (Zhong et al., 2015). We also identified gene mutations/variations among sPTBs, although a larger sample size will be needed to make a definitive conclusion. These findings provided evidence of the involvement of the CUP pathway in the pathogenesis of PPROM, which could be further supported by the protein interaction network, as shown in Figure 2. Although the algorithms of GeneMANIA (Warde-Farley et al., 2010) and STRING (Szklarczyk et al., 2015) yielded results with slight differences, a clear connection between the collagen and the UPS by an E3 ligase-like protein, PIAS3, was observed (Figure 2B).

Currently, the cause of preterm weakening of fetal membranes leading to PPROM remains unclear. Several studies have concluded the collagen degradation to be the major factor in remodeling of fetal membrane, as the collagen content was lowered in the ruptured membranes (Kanayama et al., 1985; Hampson et al., 1997). The strength of amnion and chorion is basically due to collagen fiber, and the process above reduces the physical strength of the fetal membranes. The major strength in the amnion was shown to be derived from collagen I (extensively in the compact layer and adjacent mesoderm) and collagen IV (a major component of the basement membrane and of the bundles connecting the mesenchymal layer and the epithelium; Bachmaier and Graf, 1999). COL18A1, COL1A1, and EMID2 were all within the collagen family, among which COL1A1 was involved in most human connective tissues, and COL18A1 and EMID2 were shown to be directly associated with the formation and remodeling of the extracellular matrix (Rebhan et al., 1993; Hoffmann et al., 2015). In this study, we have identified both gene mutations and abnormal gene expression. A missense mutation (Table 5) was found within the coding sequence of gene COL23A1 from sPTL blood samples. The mutation causes the residue of the 102nd amino acid to be changed from glycine to arginine (Wang et al.), which could cause an interruption in the formation of the normal structure of collagen (Lee et al., 1997). It would be interesting to introduce this mutation into the mouse model to investigate whether the missense mutation may generate the sPTL phenotype. Several InDels were identified in the intronic region of collagens from the sPTL cases, indicating that these SNVs may have a genetic predisposition that might function in gene-environmental interactions, in which the environmental factor(s) may induce the epigenetic regulation that consequently may trigger the DEP and influence transcription. In fact, our data showed that mRNAs of COL1A1, COL18A1, and EMID2 (COL26A1) were all downregulated in ruptured membranes in the PPROM subgroup. Interestingly, the expression of lncRNAs 504601 and CR602937, which overlap with COL1A1 and COL18A1, respectively, has been shown to be upregulated in PPROM (Figure 1). Both lncRNAs 504601 and CR602937 are located at the lagging strand as the antisense, opposite from the leading strand of coding genes COL1A1 and COL18A1. Like microRNAs (miRNAs; Jalali et al., 2013), these lncRNAs may also function as a suppressor to down-regulate their complementary mRNAs. Should this hypothesis be confirmed, a novel therapeutic strategy with small interfering RNA could be designed for prevention of sPTB. It would be worth to expand the sPTB cases to replicate our findings in a larger sample size among different ethnic populations globally.

Differential expression of the lncRNAs and mRNAs of ubiquitin-conjugating protein identified from expression array and qPCR suggested the involvement of the UPS in sPTL and PROM. The UPS is an ATP-dependent, non–lysosomal-proteolytic system. The whole process is shown in Figure 3. The protein product of three genes screened by qPCR belongs to ubiquitin ligase, functioning at different stages of ubiquitination. UBE2B, a member of the ubiquitin-conjugating enzyme family, works as an E2. RNF14 contains a RING zinc finger and can interact with E2s, acting as a ubiquitin-ligase (E3). These genes were found to be overexpressed in PPROM, suggesting that the ubiquitination process was boosted with rupture of the membrane. In the paired lncRNA-mRNA of uc173-UBE2B and G4299-RNF14, there is a clear correlation, in terms of the DEP, between lncRNAs and mRNAs. Apparently, lncRNAs were present as an activator, whereby when the lncRNA is up-regulated in PPROM, the mRNA is up-regulated accordingly. Both lncRNAs uc173 and G4299 are natural antisense. However, they may have a different epigenetic regulatory mechanism, compared to lncRNAs 504601 (COL1A1) and CR602937 (COL18A1). It is likely that the lncRNAs uc173 and G4299 might function as a scaffold to bind to transcriptional factors and facilitate the gene transcription (Engreitz et al., 2016). Previously, Faghihi et al. studied a similar phenomenon in Alzheimer's disease, describing antisense transcripts that can increase mRNA stability by making the binding sites (Faghihi et al., 2008; Faghihi and Wahlestedt, 2009).

The lncRNAs for PSMA3 and PSMD14 were both intronic antisense, but were demonstrated to have a distinct pattern. LncRNA ENST00000482477 presented with the most varied expression pattern, and its level was shown to be greatly increased in PPROM, whereas levels of the lncRNAs AK125314 and NR_029434 remained steady among groups. Like PIAS3, both the PSMG4 mRNA and the lncRNA AK125314 that overlapped with PSMG4 were located at the same (sense) strand, and both were upregulated. Because both lncRNAs and mRNAs are transcribed as the sense strand, the possible mechanism of lncRNAs regulating mRNAs is that lncRNAs may bind to miRNAs, functioning as a sponge, which protects mRNAs from miRNA targeting and repressing. As a result, the transcription of mRNA is upregulated, and the level of mRNA is increased. In this case, the lncRNAs may act as a sink of miRNAs (Poliseno et al., 2010). The paired lncRNA of PSMA3 was intronic antisense. Hawkins and Morris (2010) reported that antisense lncRNAs can bind chromatin and chromatin-modifying proteins, facilitating epigenetic regulation. The expression pattern of PSMD14 and its paired lncRNA ENST00000482477 was similar to that of UBE2B and RNF14; the lncRNAs might regulate their mRNAs by interacting with the miRNA-binding sites.

Whether the differentially expressed lncRNAs that were identified from the human placentas derived from the sPTB are the etiological cause for, or the result of, PPROM is unknown. To better understand this, further investigation of CUP-associated lncRNA expression at the early stage of pregnancy and of the dynamic expression profile of lncRNA longitudinally during the entire pregnancy will be necessary. The lncRNAs in the maternal circulation could be assessed through quantitation of placenta-originated exosomes isolated from maternal blood. Comparing the lncRNA expression profile at early pregnancy and at term-labor may shed light on this important question.

Both pathway analyses indicated the importance of PSMA4 and PSMD14, as they appear to be the central bridge linking the collagens and ubiquitin enzymes to the proteasome proteins. PSMA4 harbors more connections with ubiquitin enzymes (Figure 2). Together with the expression level, we hypothesize that the suppression of collagen and the upregulation of the UPS were functionally connected and then were associated with rupture of the membranes, as a cause or a result, and within the UPS, the proteolysis process was mediated through PSMA4, rather than via PSMD14.

CNVs have been studied extensively for years, and their associations with various diseases have been proved, but the possible association between CNVs and sPTL/PPROM was studied very little in comparison. We applied maternal sera samples in the screening and identified four CNVs affecting seven genes. Results of the study by Biggio et al. of CNVs in more than 1,000 American maternal and neonatal preterm birth samples and term controls showed that only neonatal, not maternal, CNVs were associated with early sPTL (Biggio et al., 2015). However, our previous work on the Chinese population indicated the contribution of maternal CNVs at the RYR1 locus to sPTL (Liu et al., 2013). Research on the Danish population reported the association of maternal CNVs in GSTT1/GSTT2 with smoking, preterm delivery, and low birth weight (Zheng et al., 2013). In the current study, the sample size may not be large enough for statistical evaluations, but the findings nevertheless suggest both race specificity for sPTL and areas for further study on the effects of these involved genes.

In conclusion, differentially expressed lncRNAs and mRNAs, polymorphisms, mutations, and CNVs identified from human placenta and fetal membrane of PPROM supported the involvement of the UPS in the development of PPROM and the possible regulatory pattern of the lncRNAs to their associated mRNAs. The results may also indicate that the loss of collagen content in PPROM was the result of not only degradation, but also of the suppressed expression, of collagen mRNAs. Furthermore, we studied the functional links of collagen to the UPS in PPROM and identified the central connector in proteasome proteins. However, the detailed mechanisms through which lncRNAs regulated their associated mRNAs in sPTL and PPROM must be further studied. On the basis of our data presented here, we propose a “two-hit” hypothesis in which genetic variations/mutations including SNVs and CNVs present as the first hit, which is genetic predisposition. Epigenetic regulation, such as lncRNAs, present as the second hit to modulate the outcome of pregnancy through the lncRNAs' epigenetic regulatory function. Our findings provide a new path for investigating the pathogenesis of sPTL and PPROM.

The work presented here was carried out in collaboration among all authors. NZ defined the research theme and designed the experiments. XZ, XL, JP, XD, WJ, MZha, and YY were involved in sample collection and carried out the experiments. XZ, XD, PW, WB, and NZ analyzed the data, interpreted the results, and drafted the manuscript. NZ finalized the manuscript. All authors have approved the manuscript.