Transcriptome analysis of human cumulus cells reveals hypoxia as the main determinant of follicular senescence

Abstract

Study question

Can RNA sequencing of human cumulus cells (CC) reveal molecular pathways involved in the physiology of reproductive aging?

Study finding

Senescent but not young CC activate gene pathways associated with hypoxia and oxidative stress.

What is known already

Shifts in socioeconomic norms are resulting in larger numbers of women postponing childbearing. The reproductive potential is sharply decreased with aging, and the reasons are poorly understood. Since CCs play an integral role in oocyte maturation and direct access to human oocytes is limited, we used whole transcriptome analysis of these somatic cells to gain insights into the molecular mechanisms playing a role in follicular senescence.

Study design, samples/materials, methods

Twenty CC samples (from a total of 15 patients) were obtained from oocytes of either male factor or egg donor patients. RNA sequencing and bioinformatic tools were used to identify differentially expressed genes between CCs from seven aged and eight young patients (<35 (years old) y.o. vs >40 y.o.). Quantitative-PCR and immunoflourescent staining were used for validation.

Main results and the role of chance

RNA sequencing identified 11 572 genes expressed in CC of both age cohorts, 45 of which were differentially expressed. In CC collected from patients >40 y.o., genes involved in the hypoxia stress response (NOS2, RORA and NR4A3), vasculature development (NR2F2, PTHLH), glycolysis (RALGAPA2 and TBC1D4) and cAMP turnover (PDE4D) were significantly overexpressed when compared with CC of patients younger than 35 y.o.

Limitations, reasons for caution

This study focused almost exclusively on assessing the genetic differences in CC transcriptome between young and older women. These genetic findings were not fully correlated with embryonic development and clinical outcome.

Wider implications of the findings

Our data provide a new hypothesis—follicular hypoxia—as the main mechanism leading to ovarian follicular senescence and suggest a link between cumulus cell aging and oocyte quality decay. If specific molecular findings of hypoxia would be confirmed also in oocytes, genetic platforms could screen CC for hypoxic damage and identify healthier oocytes. Protocols of ovarian stimulation in older patients could also be adjusted to diminish oocyte exposure time to hypoxic follicles.

Large scale data

GEO accession number: GSE81579

Study funding and competing interest(s)

Funded in part by EMD Serono Grant for Fertility Innovation (GFI).

Introduction

In the last two decades, shifts in socioeconomic norms have resulted in increasing numbers of women postponing childbearing while pursuing educational goals and securing financial stability. As a consequence of this trend, women 40 years or older are requesting assistance with in vitro fertilization (IVF) to reproduce (Wyndham et al., 2012). In the context of IVF treatments, oocyte quality is the most crucial determinant for the capacity to achieve a pregnancy, (Patrizio and Sakkas, 2009; Li and Albertini, 2013), and it is well established that the quality of the oocyte deteriorates with aging. In fact, live birth rates after assisted reproductive techniques in women 40 years or older are significantly lower than those of women younger than 35 years (Bromer et al., 2009; Cetinkaya et al., 2013) (www.SART.org accessed March 2015). In addition to increased aneuploidy rates, oocytes from older women also display abnormal levels of mitochondrial DNA and reduced oxidative phosphorylation capability (Murakoshi et al., 2013; Fragouli et al., 2015). What remains elusive, however, is definitive evidence about the mechanisms responsible for oocyte senescence. Some evidence suggests that the breakdown of molecular cooperation between the cumulus cells (CC) and oocyte is a signal for the loss of oocyte quality (Lamar et al., 2013; Fragouli et al., 2012). CC are highly specialized somatic cells that are in close contact with the oocytes via transzonal projections and gap junctions, and this communication is critical for oocyte nuclear and cytoplasmic maturation and to acquire competency for fertilization and early stages of embryo development (Park et al., 2004; Wells and Patrizio, 2008; Robinson et al., 2012; Fragouli et al., 2014).

A number of investigators have successfully studied the CC transcriptome to gain insight into the health of the oocyte (Adriaenssens et al., 2011; Wathlet et al., 2011; McReynolds et al., 2012; Ola and Sun, 2012; Iager et al., 2013; Pacella-Ince et al., 2014). However, what has not yet been investigated is the transcriptome of CC in relationship to aging. Several studies have reported that the oocyte meiotic spindle and mitochondria appear progressively damaged with advancing maternal age (Battaglia et al., 1996; Duran et al., 2011; Tilly and Sinclair, 2013) or that there is a decrease in the cohesion of sister chromatids (Chiang et al., 2010; Duncan et al., 2012). However, these injuries, which account for an increased rate of oocyte aneuploidy and impaired metabolic activity, and affect the ability of older patient's oocytes to either fertilize or develop into a competent embryo, could be triggered by other additional events.

In this study, we sought to identify these events by assessing the CC transcriptome in aging women (40 or older). We hypothesized that age-related molecular changes in the CC could be the initial signaling event for the oocyte to set in motion molecular pathways responsible for oocyte aging. In this study, we collected CCs derived from patients of two different age cohorts (<35 y.o. and >40 y.o.) and analyzed their transcriptomes using RNA sequencing (RNAseq). This high-throughput approach allowed us to determine the differential gene expression (DGE) at a genome-wide scale, providing an unprecedented depth of investigation. Moreover, we applied a comprehensive bioinformatics analysis to define the molecular pathways and networks of genes implicated in cellular senescence. This study, however, was not intended to fully evaluate the clinical outcomes.

Materials and methods

Ethical approval

The study was approved by the Institutional Review Board of Yale University (protocol number HIC# 0601000994), and written informed consent was obtained from all participants.

Patients and stimulation protocols

Fifteen patients undergoing assisted reproduction treatments were enrolled in the study. They were allocated to one of two cohorts based on maternal age: (i) ‘Younger’ group, age < 35 y.o. [n = 8] and (ii) ‘Older’ group, age > 40 y.o. [n = 7]. One patient in the older group repeated the treatment 6 months after the first retrieval, and in both cases provided one cumulus oocyte complexes (COC). Inclusion criteria were male factor infertility (as defined by World Health Organization (Cao et al., 2011)), oocyte donation program (exclusively for patients belonging to the younger group) and elective oocyte freezing for postponing childbearing. Exclusion criteria were patients with polycystic ovarian syndrome, unexplained infertility, endometriosis, decreased ovarian reserve. All patients had a body mass index value within normal/overweight ranges (18.5–29.9 kg/m²).

Controlled ovarian hyperstimulation was carried out with a luteal phase agonist (n = 14) or with an antagonist protocol (n = 2) as described elsewhere (Cetinkaya et al., 2013; Doherty et al., 2014). When at least three follicles reached 18 mm in diameter, the patients received human chorionic gonadotropin (Ovidrel, EMD Serono, USA), and oocyte retrieval was performed 36 hours later by transvaginal ultrasound-guided needle aspiration.

CCs collection and RNA extraction

COC were identified under a stereomicroscope, washed in Hepes buffered IVF medium (global with Hepes, Life Global, USA) to remove blood contaminants and incubated (37°C, 7.5%CO₂) in regular IVF media (global, Life Global, USA). Two hours after oocyte retrieval, COCs were placed under a stereomicroscope and a portion corresponding to roughly one-third of the total expanded cumulus oophorus mass was biopsied with a 21G needle. CC biopsies were then individually collected into cryogenic tubes, washed in Dulbecco's phosphate buffered saline (PBS; Gibco, Thermo Scientific, Waltham, MA) and centrifuged at 200g for 5 minutes at room temperature. Supernatant was discarded and remaining pellet snap frozen in liquid nitrogen. Labeled samples were then stored in −80°C until RNA extraction.

The corresponding oocyte was then enzymatically and mechanically denuded from the remaining CC/corona radiata as described previously (Wathlet et al., 2011), and maturation stages were recorded. Only COCs corresponding to mature metaphase II oocytes were used for the analysis. Concerning embryological development, after ICSI the oocytes included in the study (except those electively cryopreserved) were individually cultured and embryo developmental stages recorded. Blastocysts were graded according to Gardner scoring system (Gardner and Schoolcraft, 1999). However, detailed information about pregnancy outcome was not available either because the resulting embryo was transferred together with others (resulting from non-study oocytes) or because they were not transferred and cryopreserved.

Total RNA was extracted from CC with the Pico Pure kit (Arcturus; Life Technologies, Carlsbad, CA) according to the manufacturer's instructions, and genomic DNA was removed by DNAse I treatment (Qiagen, Valencia, CA, USA) directly on the Pico Pure column. Eluted total RNA concentration was verified using a Nanodrop 1000 Spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA), whereas integrity was assessed using an Agilent 2000 Bioanalyzer and RNA 6000 Pico LabChip according to the manufacturer's instructions (RNA 6000 Pico assay kit; Agilent Technologies, Santa Clara, CA, USA). The generated RNA integrity number above eight was considered as an acceptable threshold for RNA sequencing submission. A total of 20 CC samples from the 15 patients were then used for cDNA preparation and sequencing (10 CC samples from ‘Younger’ and 10 CC samples from ‘Older’, see Table 1 for details).

Table I.

Patient's demographics and baseline characteristics.

Variable	Unit	Younger (n = 8)	Older (n = 7*)	P-valuea
Age	Year	29 (1.0)	43 (0.5)	<0.0001
AMH	ng/ml	4.67 (1.56)	1.07 (0.44)	ns
Basal FSH	mIU/ml	7.4 (0.5)	8.0 (0.8)	ns
Basal E2	ng/l	33 (4)	41 (10)	ns
E2 peak	ng/l	3923 (413)	1888 (396)	<0.05
Gonadotropins administered	IU tot	2050 (207)	4505 (565)	<0.01
Oocytes collected	Number	22 (2)	9 (2)	<0.01
Etiology Infertility		Egg donors (n = 4)	Male factor (n = 7)
		Male factor (n = 4)
Stimulation protocol		Agonist (n = 8)	Agonist (n = 6)
			Antagonist (n = 2)
qPCR validation (Batch 1)	CC samples	9	6	ns
qPCR validation (Batch 2)	CC samples	4 (3 patients)	4 (3 patients)	ns

Standard error of the mean are given in parentheses.

*One patient from the older group repeated the treatment over the time of the study.

^aP-values were calculated using the Student's t-test.

Fixation and fluorescent staining

A total of six CC samples from three additional patients (n = 2 older, n = 1 younger, three CC collected per patient) from the same age cohorts/inclusion criteria as described above were mechanically biopsied, washed in PBS and immediately spread on a microscope slide. After air drying at room temperature for 5 minutes, the slides were fixed in 4% paraformaldehyde (Sigma-Aldrich, USA) in PBS for 20 minutes at room temperature. After blocking unspecific binding sites with 10% donkey serum in PBS with 0.1% Triton X-100 for 45 minutes at room temperature, cells were incubated in the primary antibody (mouse anti-human nitric oxide synthase 2 (NOS2); Santa Cruz Inc, CA, USA) made up in phostaphate buffered saline with Tween 20 (PBST) with 5% serum at 4°C overnight. Primary antibody was detected using Alexa-Fluor-conjugated secondary antibody from Invitrogen (Thermo Fisher, Carlsbad, CA, USA). All the samples were collected, fixed and stained the same day. DAPI was used for nuclear staining (VectaShield Mounting Media; Vector labs Inc, USA). Fluorescence 3D optical sectioning and bright-field imaging used a Carl Zeiss Apotome microscope and Axiovision 4.8 software (Zeiss Inc, Thornwood, NY, USA). To avoid batch effects, all samples were processed and imaged at the same time.

Image acquisition

Images were taken under an epifluorescence microscope at x20 magnification using identical camera exposure time for all samples (densitometry analysis was performed by an observer blind to the experimental conditions). For each sample, image stacks were acquired from at least six randomly selected sites and their intensity values averaged. For both DAPI (blue) and NOS2 (red) channels, image stacks consisted of 10, 1-μm-thick optical grayscale slices, which were then collapsed using maximum intensity projection filter and used as input for densitometry in NIH ImageJ (Schneider et al., 2012). Signal intensity was estimated by manually applying the same threshold filter to all images, then measuring the area covered by each thresholded signal. For each image, NOS2 signal was normalized over DAPI and presented as a percentage to correct for differences in cell profiles number in each microscopic field.

Generation of cDNA library and sequencing

RNA samples were converted to sequencing libraries using standard Illumina RNA-Seq preparation protocols. Briefly, mRNA was purified from 20 ng total RNA with oligo-dT coated magnetic beads (Thermo Fisher Scientific) and sheared by incubation at 94°C in the presence of fragmentation buffer. Following first-strand synthesis with random primers, second-strand synthesis was performed with deoxy-UTPs (dUTPs) for generating strand-specific sequencing libraries. The cDNA library was end-repaired and A-tailed, adapters were ligated and second-strand digestion was performed by Uracil-DNA-Glycosylase. Indexed libraries were quantified on an Agilent 2000 Bioanalyzer (High Sensitivity DNA Chip; Agilent Technologies). Samples were then multiplexed and sequenced on Hiseq2000 platform at a read length of 75 bp using paired-end (PE) chemistry. Sample multiplexing was designed to achieve a sequencing depth of 200 million PE reads on each flowcell lane.

Bioinformatic analysis

Sequenced reads were checked for quality control using the FastQC tool (Babraham Bioinformatics, Babraham Institute, Cambridge, UK): reads with a Phred (Richterich, 1998) quality score below a cutoff of 30 were removed using the program Trimmomatic (Bolger et al., 2014). Good quality reads were mapped to the reference genome (hGRC37/hg19) using the software package Tophat v 2.0.6 (Trapnell et al., 2009). Default settings were used, except for the mean inner mate distance between the pairs that was set to 150 bp. A maximum of two mismatches and a minimum length of 36 bp per segment were allowed. The BAM files from Tophat were then converted to SAM format by Samtools (v 1.4) (Li et al., 2009), and raw counts estimated by the Python script HTSeq-count (http://www.-huber.embl.de/users/anders/HTSeq/) (Anders, 2014).

Reference GTF file for gene annotation was downloaded from iGenomes website (http://cufflinks.cbcb.umd.edu/igenomes.html).

We generated approximately 1500 M PE sequencing reads by multiplexing 21 samples on six lanes of Illumina HiSeq2000 platform. The RNA-sequencing results were output as FASTQ format and mapped to the Ensembl GRCh37 release using Tophat. On average, each sample achieved a sequencing depth of 70 million reads (Table S1). Trimming low-quality reads improved the overall mapping rate against the reference transcriptome up to 97%. After genomic coordinates were obtained for each mapping read pair, we used HTSeq-count script program to obtain gene count information. Normalized gene expression levels, represented as counts per million (CPM), were obtained for 62 069 transcripts from Ensembl database and were processed for statistical analysis with EdgeR (Robinson et al., 2010).

The resulting raw counts per gene were processed by the EdgeR program to perform normalization, clustering and estimate differential expression (DE). EdgeR (Bioconductor release 3.4.2) performs differential abundance analysis using a pairwise design based on negative binomial model, as an alternative to the Poisson estimates (Oshlack et al., 2010). This allows accounting for biological variability as well as correcting for batch effect. Normalization of the sequenced libraries was performed to remove effects due to differences in library size; additionally, genes with <1 CPM in all samples were excluded from the analysis. DE analysis was performed after fitting the generalized linear model (GLM) approach to the normalized counts. The resulting P-values were corrected for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) approach, and genes having corrected P-values < 0.05 were called as being DE.

Weighted gene co-expression network analysis

We conducted signed co-expression network analysis using the R Weighted Gene Co-expression Network Analysis (WGCNA) package as previously described (Langfelder and Horvath, 2008). WGCNA is based on topological overlap measurements derived from pairwise correlation-based adjacency values to estimate the neighborhood similarity among genes, followed by hierarchical clustering to identify co-expression modules. To detect modules of co-expressed transcripts, the adjacency matrix (derived from the correlation analysis) was clustered using the ‘dynamic tree cut’ algorithm. This tool allows the operator to manipulate the internal structure of the dendrogram by cutting its branches, which ultimately define the modules. We then validated module membership by a permutation procedure, checking whether average module adjacency is higher than average adjacency of random groups of transcripts (Iancu et al., 2013). Among the advantages of using this approach, WGCNA allows a more comprehensive analysis of the transcriptome: instead of focusing on individual genes, WGCNA is highly effective for characterizing the features of co-expressed gene modules each of which is represented by a color classifier.

A soft threshold value, power of 16, was used to transform the adjacency matrix to meet the scale-free topology criteria for optimal clustering. The minimum module size was set to 30 genes, and the height for merging modules was set to 0.75, which required at least 25% dissimilarity among modules in expression.

Gene ontology analysis

Gene ontology (GO) enrichment was performed using DAVID (Huang da et al., 2009) (https://david.ncifcrf.gov) and ConsensusPathDb (Kamburov et al., 2009) (http://consensuspathdb.org). A hypergeometric test with the Benjamini and Hochberg FDR was performed using the default parameters to adjust the P-value.

Transcription factor analysis

Transcription factor (TF) enrichment analysis was performed by submitting the list of differentially expressed genes (DEGs) to Expression2kinases (X2K) software (Chen et al., 2012). X2K calculates P-values (applying Fisher's exact test) and the Z-score of the TF compared to the background expected rank if the enrichment was applied to a random set of genes. It then combines the two enrichment scores and ranks the list of TFs (obtained from the annotations in ChEA database), their scores and the genes that they putatively bind from the input list (Lachmann et al., 2010).

Validation of RNAseq results by Reverse transcription polymerase chain reaction

A selection of genes displaying a statistically significant difference of expression (displaying a fold-change >±1.5) between older and younger cohort was validated with real-time PCR. Since our goal was to confirm the results with the same input RNA used to generate the RNAseq data, we had restricted the number of tested genes due to low input. To further validate the first data set, a second batch of CC from ‘sibling’ oocytes (obtained in the same IVF cycles) was also analyzed. A 15 ng sample of extracted RNA was obtained from 16 samples out of the 21 submitted for sequencing (Batch 1) and 8 sibling samples (Batch 2), and reverse-transcribed using SuperScript III first-strand synthesis system for reverse transcription polymerase chain reaction (RT-PCR) (Invitrogen, Life Technologies) with oligo-dT(20) primers following manufacturer's instructions. The primers used for RT-PCR were designed using Primer Blast (Ye et al., 2012). Primer sequences are provided in Table S2. Real-time PCR was performed using LightCycler 480 Sybr Green I Master and the LightCycler 480 (Roche Diagnostics, Indianapolis, IN, USA). The PCR conditions used for all genes were as follows: denaturing cycle for 10 s at 95°C; 50 PCR cycles (denaturing, 95°C for 10 s, annealing for 10 s, extension, 72°C for 20 s), a melting curve (95°C for 1 s, 65°C for 1 s, and a step cycle starting at 72°C up to 97°C at 0.11°C/s), and a final cooling step at 40°C. Each sample was analyzed in triplicates. Glyceraldehyde 3-phosphate dehydrogenase was chosen as a reference gene. Gene expression quantification was performed according to the 2^−ΔΔCt method.

Results

Patients’ demographics and baseline characteristics and embryonic development of the oocyte-derived CCs are outlined in Table 1 and Table S3, respectively.

RNAseq analysis of CC reveals differences between patients of different age cohorts

RNA was extracted from CC biopsies and the transcriptomes of the two groups of patients (older and younger) were analyzed using RNAseq. Prior to performing the differential gene expression analysis, and in order to increase the detection power of our study, we applied an independent filtering method to remove low-expression transcripts (CPM < 1) from the complete set of samples studied. The filtered data set showed 11 572 expressed genes across all samples. Data were then normalized and controlled for both groups and batch effects using EdgeR (Robinson et al., 2010) (see Methods section) to fit a GLM. Each experiment was conducted in three batches, and each sample was normalized with respect to the mean in the corresponding batch.

DGE analysis reveals up-regulation of hypoxia-induced genes in CCs from women > 40 y.o.

We analyzed the CC RNAseq data set using a rigorous statistical approach to define differences in gene expression at the transcriptomic level. We found 45 DEGs using the edgeR software package (Fig. 1). Of these, 37 genes were up-regulated and 8 were down-regulated in the older group. The modest number of DE genes and the overall similarity between transcriptomes support the robustness of patient inclusion selection and provide confidence in the interpretation of our results (Fig. S1).

Figure 1.

Expression levels of the 45 differentially expressed genes. (A) A heatmap comparing counts per million (CPM) count of each gene for each sample. The intensity of the pseudocolors reflects gene expression levels (red indicates high CPM count, navy indicates low CPM count). Hierarchical clustering of the DE genes shows separation between older and younger groups (B) Plot of differentially expressed genes (left side) comparing samples from younger and older individuals. Differential expression is represented as a fold-change, calculated as the ratio between the CPM of older vs. younger: a positive value indicates a higher expression in the older and consequent lower expression in younger, while a negative value indicates a higher expression in younger as opposed to older. The asterisks indicate genes validated with qPCR (Fig. S2).

Most of the genes identified in the DGE analysis have protein-coding potential and associated biological functions that are informative for considering possible mechanisms of oocyte aging. One notable exception is CTA-217C2.1, which is an uncharacterized, putative long-intergenic non-coding RNA (lincRNA) that is down-regulated in the older group. Additionally, some of the protein-coding genes have not yet been functionally characterized in the literature and were therefore excluded from further analysis. To gain insights into the biological implications of the DGE analysis, we analyzed our curated list using the gene annotation enrichment programs DAVID and ConsensusPathDb (Huang da et al., 2009; Kamburov et al., 2009). GO enrichment by DAVID highlighted genes involved in cyclic nucleotide metabolic processes, vasculature development, respiratory tube development and the extracellular matrix (adjusted P-value < 0.05) (Table 2 and S4). A similar functional enrichment was obtained by ConsensusPathDb, which found significant GO annotations for regulation of blood pressure, cell proliferation, respiratory tube development and extracellular matrix organization. Pathway enrichment analysis revealed over-representation of genes involved in hypoxia inducible factor 1 alpha (HIF-1A), epidermal growth factor receptor-1 (EGFR1) and translocation of glucose transporter type 4 (GLUT4) to the plasma membrane in CCs from women > 40 y.o.

Table II.

David and ConsensusPathDb Gene Ontology (GO) and pathway enrichment analysis of differentially expressed genes.

Genes	Category	GO Term	P-value	q-value
A: DAVID GO enrichment analysis
PTHLH; PDE4D; RORA	GO:0009187	Cyclic nucleotide metabolic process	3.18E-03	4.19E-01
CTGF; EFNB2; NOS2; NR2F2; GJA5	GO:0001944	Vasculature development	2.17E-03	4.26E-01
PTHLH; ADAMTS14; CTGF; FBLN2; CCDC80; PCSK5	GO:0005576	Extracellular region	2.94E-02	9.49E-01
PTHLH; CTGF; NR4A3; GJA5; PCSK5	GO:0035295	Tube development	1.34E-03	4.96E-01
PTHLH; CTGF; PCSK5	GO:0030323	Respiratory tube development	2.13E-02	8.41E-01
PTPRK; CTGF; UNC5C; NR2F2	GO:0006928	Cell motion	8.68E-02	9.64E-01
PTHLH; NR4A3; RORA; NR2F2	GO:0045935	Positive regulation of nucleobase, nucleoside, nucleotide and nucleic acid metabolic process	1.59E-01	9.85E-01
B: ConsensusPathDb GO enrichment analysis
NOS2; PDE4D; GJA5; PCSK5	GO:0008217	Regulation of blood pressure	3.32E-04	2.48E-02
MFI2; FBLN2; CTGF; CCDC80; PTPRK	GO:0031589	Cell-substrate adhesion	3.49E-04	2.48E-02
EFNB2; NOS2; MFI2; PTPRK; PTHLH; CLEC11A; CTGF; TFRC; NR2F2; PDE4D	GO:0042127	Regulation of cell proliferation	7.19E-04	3.94E-02
PDE4D; GJA5; PCSK5	GO:0003073	Regulation of systemic arterial blood pressure	7.37E-04	3.94E-02
CTGF; PDE4D; PTHLH; PCSK5	GO:0030323	Respiratory tube development	7.48E-04	3.43E-02
MFI2; CTGF; TFRC; CCDC85B; CRIM1	GO:0001558	Regulation of cell growth	7.83E-04	3.94E-02
EFNB2; NR2F2	GO:0001945	Lymph vessel development	9.80E-04	3.43E-02
MFI2; FBLN2; CTGF; CCDC80; ADAMTS14	GO:0030198	Extracellular matrix organization	1.27E-03	3.94E-02
CTGF; CRIM1	GO:0005520	Insulin-like growth factor binding	1.28E-03	2.18E-02
MFI2; FBLN2; CTGF; CCDC80; ADAMTS14	GO:0043062	Extracellular structure organization	1.29E-03	3.43E-02
PDE4D; GJA5; NR2F2	GO:0003158	Endothelium development	1.29E-03	3.94E-02
C: ConsensusPathDb pathway enrichment analysis
	(pathway)
NOS2 ;TFRC; RORA		HIF-1-alpha transcription factor network	2.96E-04	6.81E-03
NR2F2; RORA		Nuclear receptors	2.08E-03	1.35E-02
NEDD9 ;PTPRK; EFNB2; CRIM1; TFRC		EGFR1	2.18E-03	1.35E-02
TBC1D4; RALGAPA2		Translocation of GLUT4 to the plasma membrane	2.35E-03	1.35E-02
RORA; NR2F2		NHR	3.40E-03	1.55E-02
TFRC; TBC1D4; RALGAPA2	Membrane trafficking	4.04E-03	1.55E-02

WGCNA analysis confirmed the up-regulation of hypoxia-related genes in the CC of older women

WGCNA was used to identify clusters of co-expressed genes whose expression was different between CC from younger and older patients. The WGCNA analysis relies on the assumption that strongly correlated expression levels for a group of genes indicates that these genes work cooperatively in related pathways, and that this contributes to a resulting phenotype or condition. In addition, genes may also cluster together if they share a common set of TFs. We identified 17 modules in total, with 3 modules showing significant overlap with the list of DE genes (Fig. 2). Among these, two modules displayed significant overlap with up-regulated genes in CCs from older women, whereas one module overlapped exclusively with genes up-regulated in the younger group.

Figure 2.

Weighted gene co-expression network analysis (WGCNA) and immune fluorescent staining of NOS2 (A) WGCNA analysis revealing co-expressed genes (or modules), which are compared to differentially expressed (DE) genes using a Venn diagram. Modules overlapping with DE genes (modules 0, 1 and 2) are shown. Module 0 displays 28 genes, Module 1 displays 9 genes and Module 2 displays 5 genes overlapping with the list of differentially expressed transcripts. The squares list the overlapping genes. (B) Fluorescent immunostaining of CCs from patients of different age. Nuclei are labeled in blue (DAPI staining). The red signal represents NOS2 protein. (C) Relative NOS2 protein expression in older and younger groups (see Methods). The asterisk indicates P-value significance of <0.05; bars represent standard deviation of the mean.

To better elucidate the nature of these three modules, we analyzed their functional annotations with DAVID and ConsensusPathDb. The first module, Module 0, was enriched for vasculature development, HIF-1a, EGFR1 and oxidative stress, all pathways directly related to hypoxia (Table S5). Interestingly, several pathways previously linked to oocyte aging, RAGE (advanced glycation end-products receptor) and ceramide signaling, as well as the Hedgehog pathway, were also significantly enriched (Eichenlaub-Ritter et al., 2011; Tatone et al., 2011). Biological processes such as vasculature development, apoptosis and respiratory tube development were enriched by GO annotation (Table S6). The second module, Module 1, displayed enrichment in pathways involved in cell cycle control, protein synthesis, RNA transcription, GLUT4 translocation and glycolysis (Tables S7 and S8). Network analysis of Module 2 revealed over-representation of genes involved in mRNA editing, as well as extracellular matrix remodeling and mitochondrial translation pathways (Table S9). GO performed with DAVID showed enrichment in biological processes such as transcription, translation and chromatin modification (Table S10). Thus, the WGCNA analysis revealed two distinct signatures of CCs from women of different age: a significant up-regulation of genes involved in stress-induced response in the older group as opposed to cell proliferation and protein biosynthesis in the younger. Consistently, additional gene networks already associated with oocyte aging by other groups, such as the ceramide and RAGE pathways, were also observed in this study and found to be over-represented in the older group.

DGE and WGCNA analyses provide valuable information about the biological processes altered in our two cohorts, but these analyses lack a mechanistic explanation for the altered mRNA levels. Therefore, we searched for possible upstream regulators of DEGs using the ExpressiontoKinases (X2K) algorithm (Chen et al., 2012) and derived a list of TFs that are the likely upstream regulators of these genes. Of note, 3 of the top 10 TFs, EP300, NRF2, and PPARG, have previously been linked to hypoxia (Table S11). EP300 has been identified as co-activator of HIF-1A thus promoting vasculature development through VEGF signaling. NRF2, on the other hand, is a nuclear factor known to be expressed under oxidative stress and to promote the transcription of antioxidative genes. PPARG is a TF known for its role in adipocyte differentiation and mitochondrial biogenesis. It can also regulate the expression of vascular endothelial growth factor (VEGF) and thereby affect angiogenesis (Yamakawa et al., 2000). It has been demonstrated that the hypoxic stimulus has a direct influence over the mitochondrial biogenesis either via AMPK or NO pathways (Mason et al., 2007; Gutsaeva et al., 2008). HIF-1A is known to inhibit the process of mitochondrial biogenesis by promoting c-Myc degradation (Zhang et al., 2007). Although no other genes involved in the mitochondrial biogenesis pathways (either up or downstream of PPARG) were found to be differentially expressed between the two groups of patients, we examined the levels of expression of some key regulators of this biological process, such as nuclear respiratory factor 2 (NRF2), sirtuin 1 (SIRT1) and mitochondrial transcription factor A (data not shown).

A careful review of the literature of several of the individual DEGs exemplifies the connection with hypoxia-related pathways. In particular, CRIM1, CTGF and NR2F2 have been described previously as mediators of vasculature development, which is a process associated with hypoxia (Pereira et al., 1999; Higgins et al., 2004; Fan et al., 2014). Additionally, several genes directly involved in the hypoxia–response pathway were enriched in the older group, such as NR4A3, PTHLH, NOS2, TFRC, PDE4D, CRIM1, RORA and NEDD9 (Tacchini et al., 2002; Chauvet et al., 2004; Bruder et al., 2008; Martorell et al., 2009; Kim et al., 2010). Among these, NOS2, TFRC and RORA participate in the canonical HIF-1A pathway. Other transcripts are shown to be involved in the translocation of GLUT4 to the plasma membrane, such as TBC1D4 and RALGAPA2 (Table S4). The glycolytic process is known to be associated with hypoxia as an alternative source of ATP production (Robin and Theodore, 1984).

Among the few genes down-regulated in the older group is ADAMTS14, which is a metallopeptidase with aminoprocollagen activity involved in extracellular matrix (ECM) turnover.

Validation of individual gene expression

We used RT-qPCR to validate the results obtained by RNAseq and selected transcripts involved in hypoxia-induced stress response (NOS2, RORA), tissue development (CRIM1, PTHLH) and extracellular matrix remodeling (ADAMTS14) (Fig. S2).

In order to assess whether the observed increase in NOS mRNA levels in older patients was paralleled by a concomitant increase in NOS protein expression, we performed a qualitative assessment on fresh, formalin-fixed CC samples by immunofluorescence, followed by densitometric estimate of NOS signal intensity in the two experimental cohorts.

As expected, CC obtained from older women displayed a significant increase in NOS2 protein expression, with percentages of positive staining ranging between 12–50% as compared to the younger cohort (4–20% NOS2 staining). A representative image of the differential NOS2 activation is shown in Fig. 2.

Discussion

To the best of our knowledge, this is the first study to employ a comprehensive analysis of whole-genome mRNA expression using high-throughput sequencing to compare CC of oocytes from older versus younger individuals. Furthermore, the study is uniquely complemented and extended by the addition of WGCNA, qPCR validation and whole cell imaging of target gene expression. Taken together, this multi-pronged approach has revealed an enrichment of transcripts involved in hypoxia-related stress as well as elements of the HIF-1A canonical pathways (NR4A3, NOS2, PDE4D) in CCs of older (>40 y.o.) as opposed to younger (<35 y.o.) patients. Our findings strongly suggest that hypoxia in the follicular environment is one of the main contributors to the senescence of CC.

Cellular processes such as loss of DNA repairing functionality, impaired ability to mitigate cellular stress ad altered mitochondria biogenesis are generally described as a hallmark of tissue aging. However, some investigators suggest that different molecular pathways are associated with organ-specific aging (Ori et al., 2015). Tissue hypoxia is known to occur as a natural feature of the aging process and has been described in senescent tissues such as brain, carotid body, prostate, kidney, liver and endothelial cells (Frenkel-Denkberg et al., 1999; Di Giulio et al., 2005; Tanaka et al., 2006; Kang et al., 2010). Reproductive senescence may recognize the same tissue hypoxic events due to mechanisms such as poor vascular follicular flow. Under hypoxic conditions, both intracellular pH and oocyte metabolism are reduced, influencing meiotic spindle stability and ultimately determining chromosomal scattering or abnormal cohesion between sister chromatids as observed in some solid tissue cancers (Yamamoto et al., 2006).

Prior hypotheses regarding the adverse influence of hypoxia on the formation of the meiotic spindle have been postulated by Gaulden (1992). In addition, an elegant study by Van Blerkom et al. demonstrated that poorly vascularized human ovarian follicles (with a dissolved oxygen content ≤3%) were associated with chromosomally abnormal oocytes and developmentally impaired embryos (Van Blerkom et al., 1997).

Cellular processes that are typically associated with hypoxia include the following: (i) the activation of oxygen sensor HIF (through heterodimerization of its two subunits α and β), (ii) induction of neovascularization by VEGF and (iii) promotion of ATP synthesis through glycolysis. Increased levels of cAMP are also reported as a consequence of hypoxic environment (for review (Lendahl et al., 2009)). Previous studies showed that HIF-1A is typically turned off in cumulus–granulosa cells before ovulation and activated as a consequence of luteinizing hormone (LH) signaling to the follicle, where it promotes vascularization, leukocyte migration, inflammation and follicle wall rupture (Neeman et al., 1997; Kim et al., 2009). Our data show that HIF-1A itself is expressed at similar levels in both age groups, consistent with the fact that all samples were collected at the same point in the ovulation cycle. However, genes downstream of HIF-1A, such as neuron-derived orphan receptor-1 (NOR-1 or NR4A3), nuclear receptor subfamily 2 (NR2F2) and cAMP-specific 3’,5’-cyclic phosphodiesterase 4D (PDE4D), were found to be up-regulated in older women. These findings suggest that follicles of older patients have a suboptimal concentration of oxygen perhaps due to chronic hypoxia. Since vascular development is one of the major responses to hypoxia, the up-regulation of genes involved in neovascularization is another inference supporting the follicular environment of older patients as deprived of oxygen. NR2F2 (also known as TCOUP-F2) is a nuclear receptor, which plays a critical role in the development of a number of tissues and organs, including limbs, heart and blood vessels (Pereira et al., 1999). More recently, Al-Edani and collaborators observed an increased NR2F2 expression in CCs from women of advanced maternal age compared to younger patients through microarray study (Al-Edani et al., 2014).

Another indirect piece of evidence supporting a hypoxic ovarian environment in older women is the CC up-regulation of TBC1D4 and RALGAPA2 genes involved in the translocation of GLUT4 to the plasma membrane. To the best of our knowledge, no previous reports of their expression in ovarian follicle cells have been described.

Interestingly, CC from older women display an increased level of NOS2, whose altered expression has also been confirmed at the protein level. This enzyme is known to produce NO as a consequence of hypoxic or inflammatory stimuli (Wild et al., 1986; Nussler et al., 1992). A recent study described an increased NOS2 expression in CC of oocytes that failed to fertilize, suggesting that suboptimal oocytes are exposed to oxidative stresses within the follicles (Bergandi et al., 2014).

In addition to the molecular signatures of hypoxia in ovaries of older women, previous studies of oocyte aging revealed that a compromised perifollicular vascularization could determine a lack of oxygen supply to the growing follicles (Gaulden, 1992; Van Blerkom, 1996) and cause both spindle and chromosome abnormalities as well as catastrophic genetic mosaicism to the developing embryos (Van Blerkom et al., 1997). Moreover, studies of perifollicular vascularity before oocyte aspiration by transvaginal power Doppler ultrasonography reported a positive correlation between high-grade vascularity and improved IVF outcome (Bhal et al., 1999; Bhal et al., 2001) and a negative correlation between age and ovarian perifollicular blood flow (Jarvela et al., 2003; Costello et al., 2006). Also, increased VEGF expression in follicular fluid from older women has been reported in a number of studies (Friedman et al., 1997; Klein et al., 2000; Artini et al., 2003; Ng et al., 2004). However, in our study, VEGF expression was not up-regulated in CC of older patients; therefore an increased follicular VEGF must reflect a mural and/or thecal cells origin. To reconcile the hypoxic insult with the increased follicular VEGF (Friedman et al., 1997; Klein et al., 2000; Artini et al., 2003; Ng et al., 2004) and the lack of appropriate neoangiogenesis, it is plausible to infer that the aged follicle fails to respond to non-CC-derived growth factors, perhaps because of a maternal age receptor defect of scarcity. Moreover, some other investigators ascribe low responsiveness of endothelial cells as a result of defective signaling pathways or increased distance between the perifollicular bed and the wall of the growing follicle as a function of maternal age (Tatone et al., 2008).

In summary, the findings of our study inferred by detailed CC transcriptome provide a new hypothesis—follicular hypoxia—as the main mechanism leading to ovarian senescence and suggest a link between CC aging and oocyte quality decay. If these molecular findings would be confirmed also in the aged oocytes, it could be useful to adjust protocols of ovarian stimulation by anticipating egg retrieval so as to limit the exposure of oocytes to the hypoxic environment (Wu et al., 2015).

Supplementary data

Supplementary data are available at http://molehr.oxfordjournals.org/.

Authors’ roles

E.M. collected the samples, performed the experiments, analyzed the data and wrote the manuscript, H.B. performed the statistical analysis, A.M.P. analyzed the data and helped writing the manuscript, P.P. designed the study, analyzed the data and helped writing the manuscript.

Funding

The study was partially funded by the EMD Serono Grant for Fertility Innovation (GFI).

Conflict of interest

None declared.