Discordant effects of maternal age on the human MII oocyte transcriptome

Abstract

While advanced maternal age is associated with significant changes in oocyte gene expression, these are not global changes but limited to a fraction of the transcriptome. However, there is little consensus on the specific genes affected, and on the transcriptomic signatures of age-related declines in oocyte quality. To characterize the effects of age on the human MII oocyte transcriptome, here we take a two-part approach. We first generated single-oocyte Smart-seq2 datasets from 10 younger (21-29 years) and 10 older (37-43 years) donors, identifying genes differentially expressed between the two groups, then cross-referenced our results with those of 12 studies (9 human, 3 mouse) performing equivalent analyses using a variety of single-cell transcriptomic or microarray platforms. Technical differences notwithstanding, we found considerable discordance between the datasets, suggesting that age-related signatures of differential gene expression are not easily reproducible. Independent corroboration of age-associated changes in expression was limited to few genes, with the vast majority only supported by one of the 13 datasets, including our own. Nevertheless, we identified 40 genes whose expression significantly altered with age in multiple studies, highlighting common processes underlying ageing, including dysregulated proteostasis. As human Smart-seq2 oocyte libraries are challenging to procure and rare in public archives, we next implemented a meta-analytic method for their re-use, combining our 20 oocytes with 130 pre-existing libraries sourced from 12 different studies and representing a continuous age range of 18-43 years. We identified 25 genes whose expression level significantly correlated with age and corroborated 14 of these genes with RT-PCR, including the proteasomal subunits PSMA1 and PSMA2, both of which were downregulated in older oocytes. Overall, our findings are consistent with both pronounced inter-oocyte heterogeneity in transcription and with oocyte ageing being a multifactorial process to which bona fide transcriptomic changes may only play a restricted role, while proteomic changes play more pronounced roles.

Keywords: differential gene expression; meta-analysis; oocyte ageing; oocyte quality; proteasome.

Smart-seq2 profiling and meta-analysis of human MII oocytes reveal age-associated transcriptomic changes, uncovering candidate genes linked to reproductive ageing.

Figure 1.

Transcriptomic characteristics of MII oocyte libraries. (A) Principal component biplot distinguishing younger (red) and older (blue) oocytes. (B) Heatmap representing Pearson’s correlation coefficients between the expression profiles of each sample. (C) Volcano plot illustrating the distribution of differentially expressed genes. (D) Top 10 enriched gene ontology (GO) terms among a set of 1005 genes significantly up-regulated in older oocytes. (E) Top 10 enriched GO terms among a set of 201 genes significantly down-regulated in older oocytes. Raw data for this figure are given in Supplementary Table S5 (differentially expressed genes) and Supplementary Table S6 (enriched GO terms). AMII, aged MII oocyte; YMII, younger MII oocyte; PC, principal component; BP, biological process; CC, cellular component; MF, molecular function.

Figure 2.

Genes differentially up- or down-regulated in ageing human and mouse MII oocytes, both from 12 previous studies and the present work. (A) Genes differentially up-regulated. (B) Genes differentially down-regulated. Data are presented as upset plots, as an alternative to Venn diagrams when visualizing large numbers of sets. For each panel, every possible intersection between the 13 sets is represented by black dots on the bottom plot, with dots connected by lines when a gene appears in more than one set. The total number of genes for each intersection, and in each set, are shown in the top and bottom-left bar plots of each panel, respectively. The analytic methodologies and criteria for significance for each study are briefly described in Supplementary Table S7, with the identities of each gene given in Supplementary Table S8.

Figure 3.

RT-PCR results showing the relative mRNA levels of 25 genes, the expression of which was considered strongly correlated with age. Ten MII oocytes were collected for each sample, with human 18S rRNA serving as an internal control. Bars display the fold change in gene expression in older oocytes (AMII, aged 37–43 years) relative to younger oocytes (YMII, aged 21–29 years). Genes up- and down-regulated in AMII oocytes are shown in red and green, respectively. For ease of visualization, the relative mRNA level of the younger (YMII) was set to 1.0, indicated by the dashed blue line. Error bars represent the mean ± SD. Statistical significance was determined using two-tailed Student’s t-tests: *P < 0.05, **P < 0.01, ***P < 0.001. AMII, aged MII oocyte; YMII, younger MII oocyte.

Figure 4.

Protein–protein interaction network of genes whose expression level significantly correlates with age in a meta-dataset of oocyte ageing. Panel (A) shows a Cytoscape network comprising 63 nodes (genes) connected by 192 edges (interactions between their respective protein products). Genes are those whose expression correlates with age in multiple subsets of meta-dataset, and after correction for multiple testing (Supplementary Tables S14 and S15). The clustering coefficient is 0.312 and the average number of neighbours per node is 7.111, indicating a highly interconnected structure (only nine nodes are isolated, representing genes having no protein–protein interaction with others; of these, four do not produce protein anyway, namely three processed pseudogenes [EIF3FP3, HNRNPA1P8, ENSG00000227948] and one lncRNA [ENSG00000262898]). Colour intensity represents the number of interacting proteins, with darker shaded nodes having more. Panel (B) shows a highly interconnected sub-network (6 nodes, 11 edges, clustering coefficient 0.711) identified by an MCODE (Molecular Complex Detection) analysis of the wider network. Panel (C) shows a UniProt keyword enrichment analysis for this sub-network, functionally associating it with proteasomal/proteolytic activity. On the x-axis, ‘signal’ represents the strength of enrichment, with higher values indicating a stronger degree of association between the UniProt keyword and the genes in the network. FDR, false discovery rate.