Aging endometrium in young women: molecular classification of endometrial aging-based markers in women younger than 35 years with recurrent implantation failure

Abstract

Background

To explore the differences between a population with premature endometrial aging and a population with normal endometrial status in young women with recurrent implantation failure (< 35 years).

Methods

Systematic analysis of the endometrium transcriptome of 274 RIF women. The NMF algorithm was used for classification based on endometrial-specific aging markers in CellAge, and the endometrial receptivity, gene expression patterns, and clinical data were compared between the classifications.

Results

Two hundred forty-five young RIF women could be divided into two clusters, in which the aging gene expression pattern of cluster 2 was closer to the reference cluster. Cluster 1 was characterized by high immune activity, while cluster 2 was characterized by high metabolic activity. Combined with clinical data, cluster 2 was worse than cluster 1 in window of implantation deviation rate and endometrial receptivity.

Conclusion

Premature aging of the endometrium exists in young women with RIF, and premature aging of the endometrium was associated with poor reproductive outcomes.

Supplementary Information

The online version contains supplementary material available at 10.1007/s10815-022-02578-x.

Introduction

Despite the counting refinements in embryo culture, implantation of embryos into the endometrium remains a huge obstacle for the 15% of couples currently undergoing assisted reproductive technology (ART) [1, 2]. It is estimated that the role of the endometrium is responsible for two-thirds of recurrent implantation failure (RIF) [3]. Thus, improving the endometrial environment is key to improving pregnancy rates per transplant and reducing the cost of fertility treatment. Among others, the importance of endometrial aging in RIF has been underestimated. However, age on the endometrium and the impact of endometrial aging on implantation failure have not been established [4–6]. In addition, it has been reported that older women can become pregnant from donor oocytes. Therefore, the effect of age change on the endometrium has not been considered a key factor in reproductive aging. However, a recent study confirmed age development on the endometrium by using a genome-wide functional approach and clarified that endometrial gene expression changes in women starting at age 35 years [7].

However, the inconsistency between the biological age of the endometrium and the individual’s age is also an area that needs to be looked at [8–11]. Mia S Olesen et al. used Horvath’s epigenetic clock to calculate the biological age of the endometrium. They found that the biological age of the endometrium was separated from the individual’s age. One of the most significant female ages differed from the biological age of the endometrium by 13 years [8]. This suggests that the biological age of the endometrium contributes more to reproductive aging relative to individual age.

CellAge is a database that summarizes information on genes associated with cellular senescence [12]. Researchers have validated aging-related genes in each major human organ with age, including the endometrium [12]. This study, based on endometrial aging-related genes combined with artificial intelligence algorithms, proposes a new molecular classification for women younger than 35 years of age with RIF.

Method

Patient collection and ethics statement

Two hundred forty-five women younger than 35 years old with recurrent implantation failure (RIF) were recruited at the Reproductive Medicine Center, The Sixth Hospital of Sun Yat-sen University from 2019 to 2021. We also recruited 29 women older than 40 years with RIF as a reference group (cluster 3). The Ethics Committee approved the study of Sixth Affiliated Hospital of Sun Yat-Sen University. All patients were fully informed and voluntarily participated in the study and signed an informed consent form. At the same time, this study did not cover any research products.

Sample collection, RNA extraction, and sequencing

Endometrial specimens were collected on LH + 7 with a sterile single-use endometrial sampler (Type I, Run Ting). The sample was carefully packed in Corning® 2 ml External Threaded Polypropylene Cryogenic Vial (430,659, CORNING Inc. USA) with RNA later (R0901, Sigma-Aldrich). The process of total RNA extraction and library construction was as described previously [13]. Total RNA was briefly extracted using the RNeasy Micro Kit (74,004, Qiagen). Subsequently, we used MALBAC® Platinum Single Cell RNA Amplification Kit (KT110700796; Xukang Co., Ltd.) for RNA amplification and reversed transcription. Libraries were constructed using the gene sequencing and library preparation kit (XY045, Xukang Co., Ltd.). Paired-end sequencing was performed on the HiSeq 2500 platform (Illumina).

Pre-processing of data

The processing flow of the downstream raw data is as described in our previous study [13]. Briefly, FASTQ files are de-conjugated and filtered for low-quality sequences by Trim Galore (bioinformatics.babraham.ac.uk, parameter: -q 25 –phred33 –length 25 -e 0.1 –stringency 3) [14]. Next, HISAT2 software was used to map the paired-end clean reads to the human genome GRCh38 from Gencode v26 [15]. Finally, the read count table was obtained by the feature counts function of the subread software [16]. The expression of genes was normalized to TPM (Transcripts Per Kilobase of exon model per Million mapped reads) for comparison and classification. Genes were annotated by Annoprobe R package (https://github.com/jmzeng1314/AnnoProbe/).

Classification based on nonnegative matrix factorization (NMF)

CellAge (http://genomics.senescence.info/cells/) is a manually curated database of 279 human genes that drive cellular senescence. Sixteen of these markers were shown to change with age in the endometrium, including CPEB1, DDB2, PTTG1, DHCR24, AR, CCND1, E2F1, RPS6KA6, ZMAT3, EZH2, KCNJ12, VEGFA, ASPH, BMI1, BRCA1, and TBX2 [12]. Based on the expression of these markers, we classified the samples using a nonnegative matrix factorization (NMF) algorithm by using NMF R package [17]. We use the lee algorithm for matrix decomposition and choose the rank (k) based on the coeval correlation [18]. To ensure the stability of the results, we set the number of repetitions as 200.

Principal component analysis (PCA) was used to observe the distance of each classification from the advanced age reference group (> 40 years group).

Differential expression analysis and functional enrichment analysis

DEseq2 R package v.1.34.0 [19] was used to perform differential expression analysis between classifications with |fold change|> 2 and p < 0.05. Functional enrichment analysis was implemented using OmicShare tools (www.omicshare.com) with two databases: Kyoto Encyclopedia of Genes and Genomes (KEGG) [20] and Gene Ontology (GO) [21]. p-values for each gene and function were corrected for false discovery rate (FDR) [22]. When FDR < 0.05, the gene or function was considered significantly altered.

Statistical analysis

All analyses were implemented using R v.4.0.5 [23]. Statistics of clinical data implemented by GraphPad Prism 8.

Results

Participant selection and NMF unsupervised clustering

After excluding samples that failed the experiment due to the low total number of samples, 214 samples out of 245 RIF participants were included in this study. Also, this study recruited 29 women older than 40 years with RIF as a reference group. After follow-up, 171 of these women had complete clinical data.

The TPM expression of 16 endometrial senescence-related markers was subsequently extracted for NMF (nonnegative matrix factorization) clustering. We applied the nonnegative matrix decomposition (NMF) to these 16 markers and determined the optimal rank based on the coeval coefficients and profile values as 2 (Fig. 1A-C). Therefore, NMF analysis identified two patient subtypes (cluster 1 and cluster 2). Eighty-two of the 214 (38.3%) RIF women were clustered in cluster 1 and 132 (61.7%) in cluster 2. We defined the reference group containing 29 women > 40 years of age with RIF as cluster 3.

Fig. 1.

NMF unsupervised clustering. Cophenetic coefficient value (A) and silhouette value (B) were determined by scale values from 2 to 10, with consensus over 200 runs per scale. The highest co-occurrence and contour values both occur at k = 2 (C). Consensus clustering matrix at k = 2, dark red indicates samples are always in a cluster, dark blue indicates samples are never in a cluster (D). The 2D PCA plots of the centroids of each group describe the distances between the three clusters. Cluster 2 was closer to cluster 3 (the reference cluster consisting of women > 40 years old) (E). The 3D PCA map adds a dimension to describe the distance between the three cluster samples (F). The number of differentially expressed genes between each cluster, orange represents the number of upregulated genes, and blue represents the number of downregulated genes

Characteristics and differences of each classification

Based on 16 endometrial aging-related markers, we used principal component analysis (PCA) to analyze the distance between cluster 1, cluster 2, and cluster 3. The 2-D PCA plot suggested that cluster 2 is closer to cluster 3 (Fig. 1D). We could observe this feature more clearly in the 3-D PCA plot (Fig. 1E). This result indicated that the endometrial expression pattern in the cluster 2 group was more similar to that of women older than 40 years with RIF. The results of differential expression analysis also support this feature (Fig. 1F).

The results of differential expression analysis showed 184 differential expressed genes (DEGs) between cluster 1 and cluster 3 (176 upregulation genes and 8 downregulation genes). In contrast, there were only 83 DEGs between cluster 2 and cluster 3 (28 upregulation genes and 63 downregulation genes). The expression trend of DEGs was also very interesting, with 95.7% of DEGs between cluster 1 and cluster 3 being significantly upregulated (Fig. 2A). In comparing cluster 2 and cluster 3, this percentage decreased to 33.7%, while the rate of downregulated genes increased to 75.9% (Fig. 2B).

Fig. 2.

Functional enrichment analysis of differential genes between cluster 1/2 and cluster 3 (A). Ranking of differentially expressed genes between cluster 1 and cluster 3. GO enrichment analysis described the differentially expressed genes among them mainly focused on the relevant biological processes of immune activity (C, B). Ranking of differentially expressed genes between cluster 2 and cluster 3 (D). The biological process of cluster 2 showed an opposite trend to that of cluster 1, which was downregulated during immune activity and active during metabolic activity

The enrichment analysis results demonstrate the characteristics of biological processes in both groups (Fig. 2C ). Cluster 1 was highly active in regulating the immune system, including the activation and differentiation of immune cells. On the contrary, cluster 2 was mainly activated in metabolic processes while downregulated genes were concentrated in immune system regulatory processes. We, therefore, defined cluster 1 as the immunologically active classification and cluster 2 as the metabolically active classification.

Clinical and gene expression characteristics of cluster 1 and cluster 2

The clinical data of the two groups suggested that there were no significant differences in age, BMI, endometrial thickness before transplantation and basal hormones (included E2, LH, AMH) (Table 1). There were no significant differences in age, BMI, and basal hormones between the two clusters. Endometrial thickness before transplantation in cluster 1 was greater than in cluster 2, although the difference was not statistically significant. Based on the rsERT tool developed by Aihua He et al. to detect window of implantation (WOI) [24], we found that the WOI excursion rate of the cluster 1 group was significantly lower than that of cluster 2 (22.2% vs 31.3%, Table 1). All the included subjects met the diagnostic criteria for RIF (pregnancy failure with 3 or more transfer or transfer of 4 to 6 high-scoring blastomere stage embryos or 3 or more high-scoring blastocysts).

Table 1.

Clinical feature of cluster 1 and cluster 2

N	Cluster1	Cluster2	p	SMD
	63	106
Age (mean (SD))	30.49 (2.55)	31.15 (2.49)	0.101	0.262
BMI (mean (SD))	21.73 (2.68)	21.52 (2.38)	0.592	0.084
AMH (ng/ml) (mean (SD))	4.50 (3.35)	4.08 (2.86)	0.392	0.134
E2 (pg/ml) (mean (SD))	40.14 (15.98)	67.13 (25.11)	0.4	0.151
LH (IU/l) (mean (SD))	5.22 (3.12)	6.05 (2.91)	0.084	0.274
Endometrial thickness (mm) (mean (SD))	1.83 (1.24)	1.66 (1.24)	0.404	0.133
WOI (rsERT) = offset (%)	14 (22.2)	33 ( 31.1)	0.284	0.202

The results of DEseq2 differential expression analysis showed that there were significant differences in gene expression patterns between cluster 1 and cluster 2. A total of 481 differentially upregulated genes and 12 differentially downregulated genes were found (Fig. 1F). GO enrichment analysis of upregulated genes shows that biological processes were mainly concentrated in the entries related to the immune response process (Fig. 3A). The downregulated genes mainly focus on the biological process of material metabolism (Fig. 3B).

Fig. 3.

Biological process characteristics of cluster 1 and cluster 2 (A). Cluster 1 was significantly active in the process of immune activity (B). Cluster 2 was significantly active in metabolically active processes (C). The abundance of M1 macrophages, Treg cells, and Mast cells in cluster 1 was significantly higher than that in cluster 2

CIBERSORT tool (https://cibersort.stanford.edu/) [25] was then used to calculate immune cell abundance information for both groups with perm = 1000. STAMP (Statistical Analysis of Metagenomic Profiles) [26] was used to calculate group differences in immune cells. The results showed that the abundance of macrophages M1, T regulatory cells (Treg), and activated Mast cells in cluster 1 was significantly higher than that in cluster 2 (Fig. 3C). This result was consistent with the results of the GO functional enrichment analysis.

Another interesting point in the results, the DEGs contain 13 genes in the endometrial receptivity array [27] (Fig. 4A), and these 13 genes are all significantly highly expressed in cluster 1 (Fig. 4B). And, these 13 genes are mainly enriched in the biological process of cell cycle (Fig. 4C, D ).

Fig. 4.

Comparison of endometrial receptivity between cluster 1 and cluster 2 (A). Venn diagram between endometrial receptivity array (ERA) and DEGs between cluster 1/2 (B). Comparison of expression levels of 13 endometrial receptivity-related DEGs between cluster 1 and cluster 2. (*p < 0.05; **p < 0.01; ***: p < 0.001) (C, D). Thirteen receptivity-related DEGs were mainly active in the cell cycle

Discussion

With the intensification of population aging, reproductive aging has gradually become the focus of attention. For a long time, the endometrium was not thought to age, and embryo implantation failure was more attributed to egg senescence [28, 29].

Recently, A Devesa-Peiro et al. confirmed that the endometrium is gradually aging from the age of 35 in a normal state [30], which means that the endometrium is aging, even if there is a previous pregnancy in older women after receiving oocytes from age-old donors [28]. A previous study showed that one of the important reasons for the significant decline in reproductive potential with age is that the endometrium undergoes marked age-related physiological and morphological changes [10]. Another study based on endometrial DNA methylation data found that the biological age of the endometrium of some women is not consistent with the actual age of the individual [8]. Therefore, endometrial aging and separation from the individual’s chronological age appear to be a concern in young women (< 35 years) with RIF as well.

In this study, we classified 241 women younger than 35 years based on 16 proven markers of endometrial aging and proposed two new Molecular classifications (cluster 1: immune active classification and cluster 2: metabolic active classification). Interestingly, we observed that cluster 2 closely matched aging gene expression patterns in our reference group (consisting of 29 RIF women over 40 years old). Therefore, although there was no significant difference in individual age between cluster 1 and cluster 2, cluster 1 seems to be “younger” than cluster 2 in terms of the biological age of the endometrium. Our clinical data and results from other molecular data demonstrate similar conclusions. The widely used endometrial receptivity assay (ERA) [27] and implantation window (rsERT) [24] evaluation tools show that the receptivity and implantation window deviation rate of cluster 1 were better than those of cluster 2. Therefore, we speculate that the biological age of the endometrium affects endometrial receptivity and the endometrial environment. The effects of endometrial aging factors were also present in younger women.

Cluster 1 and cluster 2 have their distinct biological processes. Cluster 1 was active in immune process-related entries in the functional enrichment analysis of differentially expressed genes with cluster 2 and the reference group (cluster 3). While cluster 2 showed active substance metabolism. In cluster 1, we found a high abundance of macrophages (Mɸ), Treg cells, and activated Mast cells. Mɸ is the major cytokine producer in the human endometrium, together with NK cells [31]. Mɸ plays a known role in establishing endometrial receptivity by increasing cell surface fucosylated structures by secreting leukemia inhibitory factor (LIF) and IL1B [32]. In our results, LIFR was also significantly higher in cluster 1 than in cluster 2 (Supplement Table 1), evidence of increased Mɸ abundance.

On the other hand, Treg cell-mediated IL-10 production is thought to maintain gut microbial homeostasis by limiting Th1 and Th17 cells [33]. The role of Treg cells in endometrial receptivity has also recently received attention [34]. This seems to explain the better endometrial receptivity of women in cluster 1 than in cluster 2.

We also found that 13 endometrial receptivity markers of ERA were included in the DEGs of cluster 1 and cluster 2. Interestingly, these 13 markers were all significantly highly expressed in cluster 1. The results of GO enrichment analysis showed that 13 markers were mainly related to the cell cycle. This was consistent with the younger biological age of the endometrium in cluster 1.

In the present study, we found a separation between the biological age of the tissue and the individual’s chronological age in the endometrium of RIF women younger than 35 years and constructed a new molecular classification of the endometrium of young women with RIF based on markers of endometrial aging. We found that the endometrium of some young women with RIF tends to age. Both clinical data and examinations suggest that this group of patients has a less receptive endometrium than women with a more “younger” endometrium. To our best knowledge, this is the first study to classify the biological age status of the endometrium in young women with RIF. It is also the largest study on endometrial aging. Endometrial aging was rarely considered in previous treatments for young women with RIF. The new molecular subtyping proposed in this study complemented the existing etiology of RIF. It provides a new direction for RIF clinical treatment decision-making.

Conclusion

In this study, we found a separation between the biological age of the tissue and the individual’s chronological age in the endometrium of RIF women younger than 35 years. We propose a new molecular classification based on markers of endometrial aging in young RIF women: cluster 1: immune active classification and cluster 2: metabolic active classification. The endometrium of cluster 2 exhibits a gene expression pattern that tends to age. The clinical features and examinations of cluster 2 also showed lower endometrial receptivity and lower embryo implantation success rate than cluster 1. Therefore, we proposed that in young women with RIF, some women’s endometrium has an aging trend that is separated from individual age. This should be fully considered in the prescribed treatment plan.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ART

Assisted reproductive technology

RIF

Recurrent implantation failure

NMF

Nonnegative matrix factorization

PCA

Principal component analysis

FDR

False discovery rate

KEGG

Kyoto Encyclopedia of Genes and Genomes

GO

Gene Ontology

DEGs

Differential expressed genes

WOI

Window of implantation

Treg

T regulatory cells

ERA

Endometrial receptivity assay

Author contribution

Cong Fang, Tingting Li, and Peigen Chen carried out the study. Peigen Chen analyzed and interpreted the data and drafted the manuscript. Meng Yang and Yingchun Guo collected the samples. Yanfang Wang and Yun Liu followed up and collected the clinical data. Cong Fang, Tingting Li, and Peigen Chen coordinated the study, participated in the design, and reviewed the manuscript. All authors read and approved the final manuscript.

Funding

Supported by the National Natural Science Foundation of China (grant number 81871214).

Data availability

The transcriptome sequencing for all of endometrium samples have been deposited with CNGB Sequence Archive of China National GeneBank under reference number CNP0003262.

Declarations

Ethics approval and consent to participate

The Ethics Committee approved the study of Sixth Affiliated Hospital of Sun Yat-Sen University (L2021ZSLYEC-280).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.