Extracellular vesicle RNAs reflect placenta dysfunction and are a biomarker source for preterm labour

Abstract Preterm birth (PTB) can lead to lifelong complications and challenges. Identifying and monitoring molecular signals in easily accessible biological samples that can diagnose or predict the risk of preterm labour (PTL) in pregnant women will reduce or prevent PTBs. A number of studies identified putative biomarkers for PTL including protein, miRNA and hormones from various body fluids. However, biomarkers identified from these studies usually lack consistency and reproducibility. Extracellular vesicles (EVs) in circulation have gained significant interest in recent years as these vesicles may be involved in cell‐cell communication. We have used an improved small RNA library construction protocol and a newly developed size exclusion chromatography (SEC)‐based EV purification method to gain a comprehensive view of circulating RNA in plasma and its distribution by analysing RNAs in whole plasma and EV‐associated and EV‐depleted plasma. We identified a number of miRNAs in EVs that can be used as biomarkers for PTL, and these miRNAs may reflect the pathological changes of the placenta during the development of PTL. To our knowledge, this is the first study to report a comprehensive picture of circulating RNA, including RNA in whole plasma, EV and EV‐depleted plasma, in PTL and reveal the usefulness of EV‐associated RNAs in disease diagnosis.

related conditions. 4 In recent years, cell-free circulating nucleic acids, especially circulating microRNA (miRNA), have garnered much attention for their potential as a disease biomarker. MiRNAs are evolutionary conserved, small non-coding RNAs ranging in size from 19 to 24 nucleotides. They regulate various biological activities by modulating the cellular transcriptome and proteome. 5,6 Besides their regulatory roles in the cell, miRNAs can also be detected in various body fluids. These cell-free circulating miRNAs are either bound to RNAbinding proteins, such as NPM1 or Ago2, 7 or lipoproteins, such as HDL or LDL, 8 or encapsulated into extracellular vesicles (EVs) to escape RNase degradation. 6,9 Some circulating miRNAs are already used as markers for disease diagnosis or prognosis. For example, the circulating miR-122 level is closely associated with different liver diseases 10 and miR-208 level is associated with heart conditions. 11,12 It has been suggested that at least some of the EVs, including exosomes, are involved in cell-cell communication. 13 Therefore, characterizing the molecular content and studying the function of EVs have been of great interest in recent years.
A number of studies report putative biomarkers for PTL including protein, miRNA and hormones from various body fluids such as serum and plasma, 14,15 cervical vaginal fluid, 15 saliva and amniotic fluids. 15 They can largely be grouped into three main categories: inflammatory-related molecules, placenta or foetal-derived molecules and stress-related molecules. For example, several inflammationrelated proteins, including C-reactive protein and cytokines, including IL6, IL8 and IL10, show PTL-associated concentration changes. [16][17][18][19] However, biomarkers identified from these studies lack consistency, especially for cell-free miRNA-based biomarkers. For example, multiple studies report changes of specific miRNA concentrations in serum or plasma of PTL patients, but the results are inconsistent or even contradictory among studies. [20][21][22][23][24] Elovitz et al, using microarray, concluded that PTL has very little effect on serum-derived miRNA; however, Gray et al, using the nanostring platform, identified several miRNAs that can be used to predict the development of PTL. 23,24 The major causes of inconsistency are different types of sample used (serum vs. plasma), low RNA concentration in samples and lack of robust measurement technology. In the past few years, next-generation sequencing (NGS) has become the major platform for miRNA analysis, especially for body fluid samples. Yet studies have shown significant sequence bias among different small RNA library preparation protocols. 25, 26 We adapted a small RNA library construction protocol using adapters with 4 degenerated nucleotides at miRNA-adapter ligation ends to reduce ligation-associated sequence bias. In addition, ultracentrifugation has been the method of choice for EV purification but the high centrifugation force may alter the natural state and content of EVs. Furthermore, the method is low throughput and requires large sample volume. We tested a newly developed size exclusion chromatography (SEC)-based EV purification protocol that provides higher throughput and cleaner EVs compared to other methods. 27,28 We are using these improved approaches to gain more reliable profiles of circulating RNA in body fluid as well as its associated EVs and EV-depleted plasma to explore the possibility of using circulating miRNAs, specifically those encapsulated in EVs, as a non-invasive biomarker for PTL.

| Sample collection, extracellular vesicle isolation and electron microscopy
Blood samples were collected from women who participated in the GAPPS study into EDTA-treated blood collection tubes, and the plasma was prepared according to standard protocol. Briefly, wholeblood samples were centrifuged for 15 minutes at 20009 g at 4°C.
The resulting supernatant is designated plasma. The plasma was transferred to clean polypropylene tubes in 100 lL aliquots and frozen at À80°C. Prior to exosome isolation or RNA extraction, plasma was spun at 10 0009 g for 15 minutes at 4°C to remove platelets and large particles. EVs were isolated from 100 lL of plasma from a total of 22 selected samples, 11 from each group (Table S1)

| RNASeq data analysis
The raw sequence data have been deposited in NCBI's Gene Expression Omnibus 30 and are accessible through GEO Series accession number GSE106224 (https://www.ncbi.nlm.nih.gov/geo/query/acc.c gi?acc=GSE106224). The results were analysed using an in-house small RNA analysis pipeline, sRNAnalyser (available at http://srnan alyzer.systemsbiology.net/), which contains three major components: data preprocessing, sequence mapping and results summarization. In the data preprocessing step, the adaptor sequences were trimmed and the low-quality sequences such as low nucleotide complexity reads-homopolymer sequences or di-and tri-nucleotide repeat sequences were removed. The processed sequences were then mapped against various databases including known human miRNA, human transcripts, followed by human genomic sequence. We also applied three different levels of error tolerance: 0 mismatch, 1 mismatch and 2 mismatches when aligning read sequence to databases.
To be considered as detectable RNA species, the RNA has to have more than 10 mapped reads in at least 70% of the samples.

| Validation of small RNASeq Results
The changes of miRNA concentrations determined by NGS were validated using the Taqman miRNA Assay kit. In brief, 2 ll of isolated RNA from individual samples was reverse transcribed using the Taq-Man microRNA RT kit (Thermo Fisher, Waltham, MA). Real-time qPCR amplification was performed on the BioRad C1000 Touch thermocycler. Enzyme was activated at 95°C for 10 minutes followed by 40 2-step cycles of amplification at 95°C for 15 seconds and 60°C for 60 seconds. MiR-16 (hsa-miR-16-5p) was used as a normalization control for each assay, as miR-16-5p did not show significant concentration changes across samples.

| Functional enrichment and network analyses
The biological impacts of PTL-associated circulating miRNAs were assessed using predicted and validated miRNA-mRNA interactions from miRTar database. 31,32 To focus on the possible functional changes in placenta, we filtered the miRNA target gene list with protein/transcript enriched in placenta based on human protein atlas. 33 To gain a more complete view of perturbed network in PTL, we expanded the initial miRNA-mRNA interaction with information from protein-protein interaction databases. [34][35][36] Functional enrichment analyses were conducted with DAVID (Database for Annotation, Visualization and Integrated Discovery) webtool. 37 Cytoscape and KEGG pathway information were used to generate the network. 38,39 The process is illustrated in Figure S1.

| Characteristics of study participants
Demographics of the 67 pregnant women (47 controls and 20 PTLs) who participated in the study are presented in Table S1. Samples from 11 preterm pregnancies and 11 matched normal pregnancies were selected for further processing of EVs and EV-depleted plasma.
Among the selected samples for EV analysis, all women reported previous pregnancies while half of the women had previously delivered a child and none of them had a previous preterm birth. The mean gestational age at delivery among the 11 PTL cases was 27.9 weeks and ranged between 23.3 and 33 weeks. The mean gestational age at delivery among the 11 controls was 39.9 weeks and ranged between 38.1 and 41.7 weeks. Characterization of SEC performance and EVs purified with the column is described in Supporting information and Figure S2.

| General statistics for small RNA sequencing results
For sequencing results, we obtained on average about 12 million raw reads in whole plasma and 8.5 million in EV-depleted plasma samples (Table 1). The average read count in the EV fractions is much higher because one control sample has more than 150 million reads. We did not observe any significant difference in raw read counts between controls and PTL samples. The number of observed and detectable miRNAs is highest for plasma and lowest for EV in both the control and PTL sets ( Table 1) Table S2). Among the detectable miRNAs (at least 10 mapped reads in 70% of samples), 481 of them are present in all three sample types ( Figure S3). Interestingly, 15 unique miRNAs were detected in the EV-depleted plasma and 9 unique miRNAs were identified in EV; these miRNAs may have been too dilute in the whole plasma and therefore below our detectable limit. The miRNAs in these clusters have been implicated in placenta development. 20 The chromosome 14 microRNA cluster (C14MC) is located at the imprinted, maternally expressed DLK-DIO3 region on the human chromosome 14q32 and is the largest known miRNA cluster with 52 miRNA precursors (miRBase, www.mirbase.org) having the potential to be processed into 104 mature miRNAs. 40 Whole-plasma control (47) Whole-plasma PTL (20) EV control (11) EV PTL (11) EV-dep plasma control (11) EV-dep plasma PTL(11)  , there are some miRNAs showing significant concentration differences associated with PTL. These concentration differences may be significant in one sample type (whole plasma) yet become insignificant in another (EV) demonstrating that the mechanisms by which the placenta releases them into circulation (either protein bound or packaged into exosomes) are differently affected by PTL

| DISCUSSION
For this study, we overcame several challenges associated with the profiling and analysis of extracellular miRNA and identified several circulating miRNAs (164 in total) whose concentrations were changed in PTL. The results feature many of the same PTL-affected miR-NAs in placenta or maternal plasma reported in prior studies; however, the direction of concentration change is not always in agreement. 45,46 Nevertheless, this is the first comprehensive characterization of RNA in whole plasma, EV and EV-depleted plasma fractions in PTL and provides insight into the effect of PTL on the spectrum of circulating RNA and how they may be used as biomarkers for PTB. As we do not have earlier samples from the individuals we analysed, the study was limited to the characterization of circulating RNA profiles from women already symptomatic for PTL. It is possible that the labour itself affects the spectrum of circulating RNA; therefore, we need to be careful when interpreting the results.

| Some miRNAs are enriched in EVs
Profiling both EV and EV-depleted plasma allowed us to unequivocally determine the partition of specific miRNA between in and outside of EV in body fluid samples. Based on our sequencing results, the miRNA distribution between in and outside of EVs is different and more miRNAs are located outside of EVs (in EVdepleted plasma; Table S3). There are at least four different reported methods for cells to sort miRNAs into EVs, especially for exosomes. These include protein-mediated processes-one by neural sphingomyelinase 2 (nSMase2) 49 and the other by AGO2 50 and sequence-dependent processes-uridine at 3 0 end 51 or a sequence motif (GGAG) recognized by sumoylated heterogeneous nuclear ribonucleoproteins (hnRNPs). 52 It is unclear whether the two protein-mediated processes, nSMase 2 and AGO2 mediated miRNA sorting, are based on specific sequence motif(s). Examining the miRNA sequences between EV and EV-depleted fractions, we could not find any common sequence motif(s) including the known exosomal miRNA-associated motif GGAG. 52 It is difficult for us to examine the non-template addition of nucleotide(s) at the 3 0 end due to the 4N adapter used in library preparation; however, the miRNAs showing higher concentrations in EVs have a higher per cent of U at the 3 0 end compared to the EV-depleted fraction (52% vs. 37%) ( Figure S4A). In addition, the overall nucleotide composition of miRNAs preferentially packaged in EVs has a higher percentage of U while sequences outside of EVs have a higher percentage of G ( Figure S4B). These findings suggest that some miR-NAs showed higher concentrations in EVs, and these miRNAs are preferentially packaged into EVs by processes yet to be fully determined.

| Circulating RNA may reflect the condition of placenta
Placenta-associated mRNAs such as CSH1 (placental lactogen), CGB1 (chorionic gonadotropin beta 1) and PLAC1 (placenta-specific protein 1), have been detected in maternal plasma, and their concentration changes in plasma can be used to reflect the health of the placenta. [53][54][55] In this study, we did not detect sufficient reads mapped to various placenta-enriched transcripts in either whole plasma, EV or EV-depleted plasma. This is probably due to our library size selection step which focuses on RNA molecules that are around 20 nucleotides in length. A library with a larger insert size may reveal more reads that map to protein-coding transcripts.
Nearly 50% of all PTL-affected miRNAs, especially in the EV fraction, identified in the study belong to two large imprinted miRNA clusters, C19MC and C14MC that are known to be expressed by the placenta suggesting that circulating miRNA profile is a relevant resource to reflect the condition of placenta. The miRNAs from the   33 We further expanded the 354 placenta-enriched proteins with its first neighbour using protein-protein interaction information, [34][35][36] (Table 3). To illustrate the complexity of miRNA-mRNA interactions, Figure 5 shows a detailed miRNA-mRNA interaction network based on 4 cell proliferation- hsa-miR-369-3p and hsa-miR-410-3p (from C14MC), and PTEN by hsa-miR-519d-3p, hsa-miR-520d-3p and hsa-miR-524-5p (from C19MC; Figure 5).

RNAs in circulation
Besides miRNA, our pipeline also reports other types of RNA in circulation. Like miRNA, we observed that the concentration of other RNAs including small nucleolar RNA (SnoRNA), piwi-interacting RNA (piRNA) and long non-coding RNA (lncRNA) was affected in patients with PTL (Table S4). The two affected SnoRNAs: SNORD22 and SNORD26 are encoded by small nucleolar RNA host gene 1 (SNHG1) on chromosome 11, and both of their concentrations increased in PTL plasma samples. Unlike miRNA, we know very little about the function of these RNAs; however, they probably also participate in foetal development. For example, the C14MC cluster is colocalized with a large cluster of SnoRNAs. One of them, the SNORD114 promotes cell cycle progression and overexpressing SNORD114 induces K562 and HCT116 cell proliferation. Even though we do not know their function, these PTL-affected non-coding RNAs identified in this study may lead to future functional studies on their involvement in normal foetal development.
In conclusion, we provide evidence for an altered profile of circulating RNA including miRNA and other small RNAs in the plasma from women with PTLs as compared to normal pregnancies and confirm the levels of some differentially expressed miRNAs (DEmiRNAs) in the whole plasma, EVs and EV-depleted plasma by real-time qPCR. We show that some of the EVs in plasma from pregnant women most likely originate from the placenta and make EV-associated molecules a useful and relatively non-invasive source of biomarkers for PTL. Further investigation with longitudinal and larger number of samples is required to validate a specific EV-associated miRNA panel that can be used towards this goal.
F I G U R E 5 Schematic diagram of perturbed gene network in placenta that may be reflected by the changes of circulating miRNAs in PTL. The network is built based on the KEGG pathway maps: focal adhesion (hsa04510), PI3K-Akt signalling pathway (hsa04151), gap junction (hsa04540) and VEGF signalling pathway (hsa04370) which were from results of the enrichment analysis of miRNA targets. The genes are indicated by circles and miRNAs by diamonds. The predicted miRNA-mRNA interactions are indicated by light grey lines, and the blue lines are validated miRNA-mRNA interactions. The identity of genes and miRNAs involved in the process are indicated, and the red colour indicates placenta-enriched mRNAs and miRNAs FALLEN ET AL.