Body mass index in relation to extracellular vesicle-linked microRNAs in human follicular fluid.

This study consisted of 133 women, mainly Caucasians (92%), while the other 8% consisted of women from other ethnicities (Asian, African, and Hispanic). On average, the women in our study were 30.9 ± 3.7 years of age (mean ± standard deviation), had a BMI of 23.6 ± 4.7 kg/m 2 , and 9.4 ± 5.1 oocytes retrieved during the IVF cycle ( Table 1 ). The majority of women were never smokers (54%) and 26% were current smokers. Most of the women were attempting IVF for the first time (65%). We screened for EV-linked miRNAs from a panel of 754 microRNAs in follicular fluid samples collected from follicles that contained mature oocytes. We detected a signal for 320 EV-linked miRNAs in at least one of the 133 follicular fluid samples analyzed. Of these, 192 EV-linked miRNAs met our pre-determined threshold for inclusion and were therefore considered as expressed. A list of the selected EV-linked miRNAs with their percent expression levels are provided in Supplemental Table S1 . Possible associations between follicular fluid EV-linked miRNAs and BMI were investigated using regression for each EV-linked miRNA. Models adjusted for age, smoking status, race, batch, and SVA surrogate variables and modeled BMI continuously. Eighteen EV-linked miRNAs were marginally significantly associated with BMI (p < 0.05) ( Table 2 and Supplemental Table S2 ) and after adjusting for multiple testing, hsa-miR-328 remained significant. A one unit increase in BMI was associated with an increase of hsa-miR-328 expression value (ΔCt) in 0.03 standard deviation [95% CI: 0.02, 0.05; q-value: 0.03]. Global variability of EV-linked miRNAs was explored using PCA, which revealed that the first three PCs explained over 50% of the variability in the outcome data set (PC1: 40%, PC2: 6%, PC3: 5%) ( Supplemental Figure S2 ). No associations were identified between smoking status, BMI, and age with these PCs in unadjusted univariate analyses; however, mutually adjusted models identified PC1 to be significantly associated with BMI (p-value: 0.01) ( Supplemental Table S3 ). This association remained significant after FDR (q-value: 0.04). To further examine the makeup of PC1, the absolute values of the PC loadings were examined ( Supplemental Figure S3 and Supplemental Table S4 ). Twenty-five EV-linked miRNAs in PC1 with a PC loading in the 90 th percentile or higher (90 th percentile threshold > 0.10) were identified. Six EV-linked miRNAs were identified in both the top 25 EV-linked miRNAs from PC1 and the 18 top EV-linked miRNAs from the BMI analyses ( Figure 2 ). We performed two KEGG analysis with these EV-linked miRNAs associated with BMI. The first examined the 18 EV-linked miRNAs that were associated with BMI in the EV-linked miRNA-by-EV-linked miRNA analysis. For this analysis, the miRWalk tool recognized 1872 unique genes that were predicted from at least four prediction programs and the TargetScan software. The DAVID software identified 18 FDR significant KEGG pathways that were associated with unique genes related to these EV-linked miRNAs. Twelve of these identified KEGG pathways are related to oocyte and follicle development and maturation ( Figure 1 ). The second BMI KEGG analysis examined the 25 EV-linked miRNAs having a PC loading in the 90 th percentile or above in PC1. The miRWalk software recognized 4092 unique genes and the DAVID software found 43 KEGG pathways significantly associated with these 4092 genes (q-value < 0.05). Among these 43 KEGG pathways, 24 were involved with biological processes of interest to our study ( Figure 1 ).

This study was approved by Sheba Medical Center IRB in accordance with the Declaration of Helsinki ( 35 ). All participants provided written informed consent upon enrollment. Between January 2014 and August 2016, women undergoing IVF were recruited in a tertiary care university-affiliated hospital. Eligibility criteria were women aged 19 to 38 years old, undergoing their 1st to 6th IVF attempt with a basal FSH less than 10 IU. To avoid potential confounding by the stimulation regimen or fertility diagnosis, we enrolled only women that were treated using GnRH antagonist protocol as well as women undergoing IVF due to unexplained infertility, mild male factors or women undergoing IVF for pre-gestational diagnosis of autosomal recessive disorders. We excluded those with a diagnosis of polycystic ovarian syndrome (PCOS), endometriosis or poor responders according to Bologna criteria ( 36 ) which might impair oocyte quality. All women contributed a single IVF cycle. Participant’s height and weight were measured at the beginning of the IVF cycle and were used to calculate body mass index (BMI, kg/m 2 ). Body mass index was calculated by dividing the weight (kg) of a person with the square of their height (m 2 ). Follicular fluid (otherwise discarded material) was collected during oocyte retrieval from a single follicle ≥18 mm, per patient and centrifuged at 1500×g for 15 min. FF was aliquoted to 0.5 ml tubes So that the volumes of FF that were analyzed were equivalent among patients. Samples were pre-cleaned using a 0.80 um pore-size polyethersulfone filter (StericupRVP, Merck Millipore) to remove larger proteins and debris and aliquoted into 500 uL for immediate storage at −80°C ( 37 ). Only mature (MII) oocytes were examined for RNA analysis. Methods for RNA extraction from biological fluids have been previously described ( 38 ). In short, samples were thawed, centrifuged for 15 min at 1200 × g at room temperature and then centrifuged three times at 1000, 2000, and 3000 × g, respectively, for 15 min at 4°C Following this steps, samples were ultracentrifuged (Beckman Coulter Optima-MAX-XP) at 110,000 × g for 75 min at 4°C for the extraction of EV, as ultracentrifugation is considered the standard according to International Society for Extracellular Vesicle recommendations ( 28 , 39 ). The pellets obtained were kept at −80°C until use. EV-linked miRNAs were extracted from the ultracentrifuged pellets using the miRNAeasy Kit and RNeasy CleanUp Kit per the manufacturer (Qiagen, Valencia, CA, USA). The final purified EV-linked miRNA-enriched RNA was eluted into 20 uL of RNAse-free water and stored at −80°C until further use. This protocol has been tested and samples have been examined by both Nanosight and Flow Cytometry for CD63 (an EV-specific marker) (results unpublished). We screened for 754 microRNAs using the TaqMan Open Array® system. We obtained 758 Crt values for each follicular fluid sample, which included 754 unique miRNAs and four internal controls (ath-miR159a, RNU48, RNU44 and U6). Methods of Real-Time Quantitative Polymerase Chain Reaction (RT-qPCR) for screening EV-linked miRNAs using the microRNA array in biological fluids are published elsewhere ( 38 ). QuantStudio™ 12K Flex is a fixed-content panel containing validated human TaqMan® MicroRNA Assays derived from Sanger miRBase release v.14. All 754 assays have been functionally validated with miRNA artificial templates. The panel is specifically designed to provide specificity for only the mature miRNA targets. TaqMan MicroRNA Assays (spotted in the panel) incorporate a target-specific stem–loop reverse transcription primer allowing to work despite the short length of mature miRNAs (~22 nucleotides) which prohibits conventional design of primers. Briefly, we prepared 3.3 uL of each RNA sample and then reverse-transcribed to cDNA (complementary DNA) and pre-amplified. Pre-amplified samples were mixed with the TaqMan Open Array® Real Time PCR Master Mix (Life Technologies, Foster City, CA) and loaded onto a TaqMan™ OpenArray® Human miRNA panel with the QuantStudio™ AccuFill System Robot (Life Technologies, Foster City, CA). RT-qPCR was performed on the QuantStudio™ 12K Flex Real-Time PCR System with the OpenArray® Platform [QS12KFLEX] (Life Technologies, Carlsbad, CA) according to the manufacturer’s instructions. Expression levels were calculated in relative cycle threshold values (Crt), estimating the amplification cycle at which the fluorescence levels for each of the analyzed EV-linked miRNAs exceed the background fluorescence threshold ( 40 ). Participant’s age, number of previous IVF attempts, and number of oocytes retrieved at the start of IVF cycle were extracted from participant’s medical charts. Cigarette smoking status was determined by a questionnaire in which participants were asked about cigarette smoking history and intensity. Smoking status was categorized as never smokers, former smokers, and current smokers. Those who self-identified as former smokers stopped smoking prior to the beginning of their IVF cycle. Follicular fluid samples were analyzed for EV-linked miRNAs in two batches and in our subsequent statistical analyses we controlled for batch (batch 1 vs batch 2). Thermo Fisher Cloud Relative Quantification software was used to extract the EV-linked miRNA qPCR data. To ensure accuracy of in our normalization methods of the EV-linked miRNAs, we ran algorithms to identify the best normalization strategy. We first applied the NormFinder and geNorm algorithms to select the best normalization strategy among global mean (arithmetic and geometric), RNU48, RNU6, or the average of the four miRNAs with the lowest standard deviation (SD) among subjects. Based on these algorithms, we found that global mean was the best method to normalize the data. EV-linked miRNAs data was normalized using the global mean (GM) method (ΔC rt_EV-linked miRNAi = (C rt_EV-linked miRNAi – C rt_EV-linked miRNAi_global_mean )) ( 38 ). All the EV-linked miRNA with a Crt value > 28 and/or an amplification score ≤ 1.24 were identified as unexpressed. For the global mean, we coded all those EV-linked miRNAs that were unexpressed as 28 as suggested by Pergoli et al. ( 41 ). We calculated the delta Crt based on the global mean across all the miRNAs within that subject and dividing it by the total miRNAs (N=754). All subsequent analyses were performed exclusively on those EV-linked miRNAs that had expressed values, which were detected in at least 15% of our samples, based on the histogram distribution of the percent detected in our samples ( Supplemental Figure S1 ). This cut-off was chosen to maximize the number of EV-linked miRNAs examined while retaining a large enough sample to assess the associations (at least 20 individuals), while removing those EV-linked miRNAs that had a very large missing rate. Standard descriptive statistics were used to explore the characteristics of the study participants and exposure data. Spearman’s correlation coefficients were used to examine correlations between covariates and exposure variables. Linear regression models were performed to elucidate top hit EV-linked miRNAs. Models were adjusted for continuous age (years), continuous BMI, categorical smoking status, binary race/nationality, binary batch, and continuous SVA surrogate variables, identified by the SVA package in R. The SVA package can help identify and remove any batch effects or unwanted sources of variation, seasonal, meteorological, exposure, or technical variables, which are unknown but might be differently distributed in the two batches of samples. It creates surrogate variables, accounting for the unmeasured variation, that act as covariates in our models that would account for any unknown, un-modeled, or other sources of ( 42 , 43 ). EV-linked miRNA outcomes were inverse normally transformed to ensure normality with a standard deviation of 1. To account for multiple-testing, we applied the Benjamini-Hochberg FDR “p.adjust” function in R ( 44 ). All statistical analyses were performed in R-version 3.4.0 ( 45 ). Statistical significance was set at a p-value < 0.05. We used the Benjamini-Hochberg False-Discovery Rate (FDR) ( 44 ) to account for multiple testing and chose a threshold, FDR q-value < 0.05. Principal component analyses (PCA) were performed to examine the overall variation of EV-linked miRNA profiles. All of the selected 192 EV-linked miRNAs were included in the analysis. The function “prcomp” in R was used to calculate the principal components (PCs) and the PC loadings for each component. All variables were scaled and centered with a mean of 0 and a standard deviation of 1 prior to the PCA. PCs were inverse normally transformed for normality. Separate unadjusted univariate analyses were performed on age, smoking status, and BMI using linear regression with each principal component as the outcome. Mutually adjusted linear regressions were performed with the principal components as the outcomes and adjusted for age, BMI, smoking status, race, batch, and SVA surrogate variables. Absolute values of the PC loadings were taken and the EV-linked miRNAs that contributed a loading in the 90 th percentile or higher were examined further. This threshold was chosen based on both the histogram distribution of the absolute values of the PC loadings ( Supplemental Figure S3 ) and the percentile cutoffs ( Supplemental Table S5 ). This would allow us to examine those EV-linked miRNAs that are the most important to the principal component while keeping the number of EV-linked miRNAs manageable. An in-silico analysis using a web-based tool, miRWalk2.0 ( http://mirwalk.umm.uni-heidelberg.de/ ) was performed to investigate gene targets of EV-linked miRNAs ( 46 ). Target predictions included a comparative analysis of seven prediction programs: DIANAmT, miRanda, miRDB, miRWalk, PICTAR5, RNA22, and Targetscan. To decrease the number of false positive pathways, targeted genes were identified that were validated in a total of five prediction programs including Targetscan and any combination of at least four of the other prediction programs. The predicted target search examined promoter, 5’-untranslated regions, 3’-untranslated regions, and coding sequences and included a minimum seed length of 7 nucleotides ( 46 ). Using the genes selected by miRWalk, we ran a Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis using the web-based tool DAVID ( https://david.ncifcrf.gov/summary.jsp ) ( 47 , 48 ). We restricted our results to only those with 10 or more genes listed in the in silico predicted targets and an FDR q-value < 0.05.

In the present study, we identified 18 EV-linked miRNAs associated with BMI. Of those, only hsa-miR-328 remained statistically significant after FDR-adjustment. Additionally, principal component analysis was applied to examine the global variability in the profile of 192 EV-linked miRNAs. The first principal component explained 40% of the variability in the EV-linked miRNA dataset and was significantly associated with an increased BMI in a mutually adjusted model. It has been well established that an increased BMI is associated with decreased fertility outcomes, such as reduced pregnancy and fertilization rates ( 10 - 12 , 49 ). We found hsa-miR-328 was significantly associated with BMI after adjusting for covariates and taking into account multiple testing. Further examination of hsa-miR-328 identified associations with cancer ( 50 , 51 ). MiR-328 has been known to play a role in myocardial infarction and cancers in adults, where obesity is an established risk factor ( 52 ). A study examining childhood obesity and circulating miRNAs found circulating concentrations of hsa-miR-328 were higher in obese compared to lean children ( 53 ). This suggests that hsa-miR-328 might play a role in adipose regulation, obesity, and obesity related diseases. A rodent study found that reducing expression of miR-328 blocked pre-adipocyte commitment and inducing miR-328 instigated differentiation of brown adipose tissue ( 54 ). Few studies have examined the underlying molecular mechanisms linking BMI with fertility. One previous study assessed the associations between circulating plasma microRNAs and BMI and found 19 significant microRNAs, including hsa-miR-193b-3p ( 55 ). This was the only overlapping EV-linked miRNA with our analysis, but the directionality of the association with hsa-miR-193b-3p was the opposite of that previously reported ( 55 ). Another study examined pre-pregnancy BMI and circulating miRNAs during pregnancy and identified 27 significant miRNAs in two separate cohorts. Specifically, hsa-miR-28-5p, hsa-miR-376a, has-miR-139-5p, and hsa-miR-423-5p were found to overlap with those EV-linked miRNAs which were significantly associated with BMI in our analysis ( 56 ). Even more EV-linked miRNAs overlapped with Enquobahrie et al, however, the direction of association was not consistent for hsa-miR-376a and hsa-miR-139-5p ( 56 ). We also examined whether the global variability of EV-linked miRNAs were associated with BMI. EV-linked miRNAs can come from different non-coding sequences but can target the same messenger RNA transcripts ( 57 ). Thus, one EV-linked miRNA can interfere with the expression of many genes and that one gene can be negatively regulated or silenced by many EV-linked miRNAs. Our PCA identified principal components that would explain a certain percentage of the variability within our EV-linked miRNA dataset and found PC1 was significantly associated with BMI. This suggests that the biological variability captured by PC1 was significantly associated with BMI. From these two analyses, six EV-linked miRNAs overlapped as significant, hsa-miR-99b, hsa-miR-29a, hsa-miR-28, hsa-miR-328, hsa-miR-152, and hsa-miR-331. The EV-linked miRNA that was FDR significant in the EV-linked miRNA by EV-linked miRNA analysis, hsa-miR-328, was identified as significant in both analyses. This suggests that this hsa-miR-328 might be an important EV-linked miRNA that could be impacted by an increase in BMI. Our KEGG pathway analysis of the top EV-linked miRNAs contributing to PC1 revealed several pathways associated with pathways of cellular signaling, follicular growth, oocyte maturation ( 58 - 62 ). We also performed KEGG pathway analyses on the 18 EV-linked miRNAs associated with BMI in the EV-linked miRNA-by-EV-linked miRNA regression analyses. We identified 16 significant pathways that overlapped between the three KEGG analyses. Eleven of these were associated with oocyte and follicle maturation and development: ECM-receptor interaction, focal adhesion, FoxO signaling, oocyte meiosis, PI3K-Akt signaling, adipocytokine signaling, AMPK signaling, cGMP-PKG signaling, ErbB signaling, gap junction, and GnRH signaling pathways. These 11 aforementioned pathways contribute to oocyte development, maturation, and signaling. More notably the Pl3K-Akt pathway has been associated with recruitment of primordial follicles, granulosa proliferation, and ovarian function ( 63 ). It also activates the mTOR and FoxO pathways ( 64 ). The FoxO pathway plays an important role in intra-oocyte signaling and can negatively regulate oocyte growth and follicular development by acting as a regulatory switch to initiate primordial follicle activation ( 65 - 68 ). The pathway of oocyte maturation is induced to transform an immature oocyte into a mature or fertilizable egg ( 69 ). ECM-receptor interaction (extracellular matrix) is believed to play an essential role in regulating follicle development ( 70 ) and significantly contributes to follicle microenvironments that allow intracellular communication to occur between cells and the oocyte ( 70 ). Focal adhesion is necessary to establish or maintain the oocyte-granulosa cell contact in the follicle ( 71 ). Pathways identified by the significant EV-linked miRNAs play major roles in follicle and oocyte development and maturation, suggesting that BMI could disrupt these processes. This could provide a mechanistic explanation for the reduced fertility rates and increased spontaneous abortion rates that are observed with an increased BMI. This study has limitations. First, we quantified EV-linked miRNA profiles from a single follicle, it might not accurately represent the EV-linked miRNA profile of the whole cohort of follicles. In a previous analysis (unpublished data) we compared the profile of EV-linked miRNA from two follicles per patients that contained mature oocytes, and their profiles were highly correlated. As the single follicle did not always contain the oocyte that yielded the embryo that was transferred and in many of the cases more than one embryo was transferred, we were limited in testing associations of BMI, EV-linked BMI and pregnancy outcomes. Second, we ran our KEGG pathway analyses on a relatively small number of EV-linked miRNAs. Thus, our results may be subject to overfitting of the selected pathways by the software. To minimize this, we set our inclusion of genes to those found in at least four separate prediction software packages in addition to the Target scan database. We also included those EV-linked miRNAs that were marginally significant but did not meet the FDR threshold for multiple comparisons. EV-linked miRNAs are highly correlated and have high biological variability between samples, so to prioritize biological relevance in our pathway analyses we chose to include all those EV-linked miRNAs that met the marginally significant threshold (p < 0.05). In addition, while we can assure that we have a pellet enriched in EVs, it is possible that the pellet included not only miRNAs that are encapsulated within the EV, but also miRNAs that were extracellularly attached to EVs ( 28 ). Additionally, there is a possibility that some of the miRNAs in the pellet derived either from necrotic cells ( 28 , 72 ). To ensure we extract EV-linked miRNAs, we have optimized our protocol by running thousands of samples in previous large studies investigating EV-linked miRNAs as well as assessed random follicular fluid samples by Nanosight and flow cytometry. Our study has significant strengths as well. First, to our knowledge, this is the first study to examine BMI with respect to EV-linked miRNAs in follicular fluid. Second, we used PCA to examine how EV-linked miRNAs contribute to variability in our dataset, allowing us not only to expand upon work done by others, but also to examine the global variability of the outcomes. In conclusion, these identified associations between an increased BMI and follicular fluid EV-linked miRNAs may provide an insight as to how higher BMI may disrupt fertility on a mechanistic level.

Human infertility is influenced by a broad range of physical, hormonal, genetic and environmental stressors ( 1 ). Despite enormous advances in the technical aspects of IVF, the success rates of the procedure remain relatively low. Whereas much has been published about non-modifiable factors associated with outcomes of IVF such as female age, fertility diagnosis, and ovarian reserve, less attention has been devoted to modifiable behavioral risk factors, such as body mass index (BMI) that may influence IVF outcomes ( 2 - 4 ). Studies have shown that an increased BMI and obesity alter follicular development, oocyte maturation, and reduced numbers of oocytes retrieved ( 5 - 9 ). Women with a BMI ≥ 25 kg/m 2 undergoing IVF have a significantly reduced chance of a clinical pregnancy compared to women with BMI < 25 kg/m 2 ( 10 - 12 ). Despite the ample epidemiologic evidence, the molecular mechanisms underlying how increased BMI contributes to infertility are still unknown. The ovarian follicle houses the oocyte itself and as it matures, cellular differentiation occurs, creating cellular layers of thecal, mural granulosa, and cumulus cells. Theca and granulosa cells create the membrane of the follicle itself, while cumulus cells encapsulate the oocyte ( 13 , 14 ). Follicular fluid is a crucial microenvironment for the developing oocytes, containing hormones, metabolites, ions, proteins, and various biological structures including non-coding RNAs and extracellular vesicles (EVs) ( 14 - 21 ). EVs, membrane enclosed structures ( 22 ), have been identified as a means for intercellular communication between the granulosa cells and oocyte in the follicular fluid ( 15 ). They are nanosized (30-700 nm in diameter) and can contain numerous molecules involved in cell signaling and cell to cell communication, including microRNAs, which are short non-coding RNA molecules (approximately 22 nucleotides long) ( 23 ). MicroRNAs are short, non-coding RNA molecules that can post-transcriptionally regulate gene expression ( 24 - 26 ) and can be free-floating or packaged in extracellular vesicles (EV-linked miRNAs). Extracellular vesicles (exosomes, micro-vesicles, and other membrane-bound vesicles) have been detected in almost every biofluid, including follicular fluid ( 22 ), and can act as a vehicle carrying proteins, messenger RNAs (mRNAs), and microRNAs (miRNAs) ( 27 ). These extracellular vesicle linked microRNAs (EV-linked miRNAs) can be encapsulated within the EV itself, but also associated with the EV extracellularly ( 28 ). These EV-linked miRNAs can influence gene expression and may be important for follicular signaling ( 15 , 29 , 30 ). Several studies to date have reported possible associations between BMI and EV-linked miRNA profiles either in adipocytes or serum linking some of these EV-linked miRNAs with pathways of inflammation and insulin resistance ( 31 - 34 ). To the best of our knowledge, no study has examined associations between BMI and EV-linked miRNAs found in follicular fluid. Investigating EV-linked miRNAs in follicular fluid might help elucidate the mechanisms and pathways by which these modified risk factors that can impact fertility. Therefore, this study was aimed to investigate the effects of BMI with EV-linked miRNA expression isolated from follicular fluid in women undergoing IVF treatment.

Supplemental Figure S1: Histogram of percent detected levels of expression of each EV-linked miRNA. Red line denotes the 15% threshold which separates the majority of expressed EV-linked miRNAs (to the right of the line) from those that did not express (to the left of the line). Supplemental Figure S2: Principal components from PCA analysis. On left, all 133 principal components calculated with the y-axis showing the percent variability each principal component explains in the EV-linked miRNA dataset. On right, a close up of the top three principal components with their percent variability described. Supplemental Figure S3. Histogram of the absolute values of the PC loadings from principal component 1. Gray shaded bars indicate the 90 th percentile or higher of PC loadings. EV-linked miRNAs examined in this percentile are listed.