Ethical approval and informed consent

This study was approved by the Medical Review Ethics Committee of the Keio University Hospital (#20150473) and written informed consent was obtained from all subjects who agreed to use used embryo culture medium for research purposes.

Patients

Informed consent was obtained from each participant prior to enrollment. We included 60 patients who underwent thawed embryo transfer of blastocysts after intracytoplasmic sperm injection (ICSI) between May 2016 and April 2019. Patients having undergone conventional in vitro fertilization (IVF) were excluded to avoid contamination of granulosa cells, as were patients with structural and tissue abnormalities or infectious etiologies in the uterus and adnexa. Used embryo culture medium (SAGE 1-Step, Origio) was collected on the fifth day of culture during the IVF cycle. Samples were classified into four groups according to the transferred embryo grade and implantation outcome: I) good morphology blastocyst (*) and positive pregnancy result; II) good morphology blastocyst (*) and negative pregnancy result; III) poor morphology blastocyst (**); and IV) control samples of culture medium only (negative control) (*: Gardner grade ≥ 3BB, **: embryo transfer was not performed for Group III). All patients whose blastocysts had been classified into group II became pregnant within the next three ET cycles, which suggests infertility was not due to uterine conditions.

The mean age of patients whose blastocysts were classified into group I and group II for the RNA sequencing analysis was 36.1 ± 3.6 years (n = 12) and 36.7 ± 4.2 years (n = 18), respectively. The mean age of patients whose blastocysts were classified into group I, group II, and group III for the RT-qPCR analysis was 34.1 ± 3.2 years (n = 17), 36.3 ± 2.8 years (n = 17) and 36.2 ± 3.6 years (n = 17), respectively. All patients were below 41 years of age.

The stimulation methods were antagonist, short, long, or mild. Ovarian stimulation was initiated on the third day of the menstrual cycle, with 2.5 mg letrozole for 5 days, together with 225 IU of highly purified hMG (Ferring) daily in most cases. A daily dose of 0.25 mg Ganirelix Acetate (Organon) was given when follicle size reached an average diameter of 13–14 mm. Administration of gonadotropin-releasing hormone (GnRH) antagonists was performed when the follicle size was 14 mm or larger. All patients underwent transvaginal ultrasonography and regular monitoring of follicle-stimulating hormone, luteinizing hormone, estrogen, and progesterone. Mature oocytes were collected 34–36 h after human chorionic gonadotropin injection or intranasal GnRH agonist (600 mcg of buserelin, Sanofi) or dual trigger (GnRH agonist plus 250 mcg of choriogonadotropin alfa, Merck Biofarma). Mature oocytes were fertilized in vitro using ICSI methods. All good morphology blastocysts were frozen using the vitrification method. Single embryo transfer (SET) was performed by hormone replacement therapy cycles. If the pregnancy test was positive, follicular hormone and progesterone supplementation were continued until 8 weeks of gestation. If the result was negative, medication was discontinued.

IVF protocol and sample collection

After ICSI, embryos with two pronuclei were selected by fertilization check. One ICSI embryo each was placed in a microdrop of 30 μL of culture medium (SAGE 1-Step, Origio) sealed with mineral oil and cultured for 5 days in a timelapse incubator (Embryo Scope, Vitrolife) without medium refreshment. When a good blastocyst was obtained, 25 μL of spent medium was collected and cryopreserved. After subsequent thawed embryo transfer, if the outcome was a clinical pregnancy with a fetal heartbeat at ≥ 8 weeks, the spent medium was classified into the pregnancy group (I). In the absence of an increase in hCG blood levels, the medium was classified as belonging to the non-pregnancy group (II). Microdrops of culture medium without embryos were cultured in the same way for 5 days to serve as the control group (IV). Blastocyst grade was evaluated by the Gardner classification just before cryopreservation. The same manufacturer lot of culture medium was used for all groups.

Library preparation and next-generation sequencing

To investigate the biological relationship between miRNA expression profiles and successful implantation, RNA sequencing was performed on miRNAs in the pregnancy and non-pregnancy groups (I and II), as well as in the control group (IV) to correct for the presence of non-specific background miRNAs, including those derived from serum albumin in the culture medium. Among the non-coding RNAs obtained, the percentage of miRNAs was very small, about 0.04–0.2%. To ensure sufficient miRNA abundance, three pooled samples each were prepared by combining four and six culture media in group I and II, respectively.

In PCR-based whole amplification, 30 cycles of PCR amplification of the DNA library with adapters allowed sequencing. The miRNA sequencing library was constructed using the TruSeq® Small RNA Sample Prep Kit according to the manufacturer’s protocol (Illumina, San Diego, CA, USA). To the total RNA of each sample we added miRNAs (dme-miR-308-3p and dme-miR-316-5p) that were spiked as carrier RNA, to prepare the miRNA sequencing library in the following steps: 3′-adaptor ligation, 5′-adaptor ligation, cDNA synthesis, PCR amplification and AMPure beads (Beckman Coulter, Brea, CA, USA) purification for PCR-amplified fragments. After quantification with RT-PCR, the libraries were captured on cBOT (Illumina, San Diego, CA, USA) to be amplified in situ as clusters. Finally, they were sequenced on the HiSeq™ 1500 System (Illumina, San Diego, CA, USA) as per the manufacturer’s instructions.

After sequencing, the adaptor sequences were trimmed and removal of non-small RNA-related reads such as null-insert (insert size < 10 nt), 5′ adapter contamination and poly-A containing, were performed by custom perl scripts. The trimmed reads were aligned to the human reference genome (hg19) using COBWeB aligner implemented in StrandNGS ver. 2.6 (Agilent Technologies, Santa Clara, CA). The UCSC transcripts ver.2013.12.3 dataset was used for gene annotation. Up to one mismatch per each sequencing read was allowed in the alignment. The raw read count values estimated for each small RNA were normalized using the Trimmed Mean of M-value (TMM) method [28].

RT-qPCR

The spent medium obtained from each embryo was analyzed separately without pooling and miRNAs secreted or absorbed from each embryo were quantitatively analyzed by quantitative RT-PCR. We measured Ct values in 17 samples in the good morphology/pregnancy group (I), 17 samples in the good morphology/non-pregnancy group (II), 17 samples in the poor morphology (III), and 11 samples in the control group (IV) using only culture medium. The TaqMan Advanced miRNA cDNA Synthesis Kit (ThermoFisher) was used to synthesize cDNA according to the manufacturer’s protocol, followed by pre-amplification. To 5 uL of pre-amplified cDNA we added 10 uL of TaqMan Fast Advanced Master Mix (2X), 1 uL of TaqMan Advanced miRNA Assay (20X), and RNase free water to make 20 uL. Then, the PCR was run with 40 cycles of enzyme activation at 95 °C for 20 s, one denaturation cycle of 95 °C for 3 s, and an anneal/extend stage at 60 °C for 30 s, and the fluorescence intensity of FAM was detected with a 7500 real-time PCR system (ThermoFisher). Normalization was performed using pre-spiked C. elegans-derived miRNA as a control gene.

Data analysis

Obtained sequence data was trimmed and aligned to the human genome sequence using the COBWeb aligner tool (Agilent Technologies, Santa Clara, CA, USA). miRNA annotation information was obtained from UCSC ver2013.12.3 database. Alignments were allowed under single nucleotide mismatches, and normalization was performed for raw read counts using the Trimmed Mean of M-value (TMM) method [27].

Logistic regression analysis

A prediction model was generated based on the results from expression levels of indicated miRNAs by logistic regression analysis to predict implantation outcome. A ROC curve was generated by plotting the true positive rate against the false positive rate, which was calculated by comparing the predicted value using the prediction model with the true result. Five-fold cross validation was used to validate the generalization performance of the model. Each dataset was randomly divided into five subsets of equal size, and the model was trained using four of the five subsets and validated on the remaining holdout subset. All combinations were performed, and the AUC was calculated for each validation and the final AUC was reported as the average AUC of five individual training trials.

Target prediction and functional analysis

We performed a web-based database search for miRNAs indicated to be involved in the implantation process. The databases used were DIANA miRPath-v3.0 (http://www.microrna.gr/miRPathv3) and miRTargetLink Human (https://ccb-web.cs.uni-saarland.de/mirtargetlink). These databases give an overview of miRNAs and provide information on the predicted target genes and their targets.