Mapping the anatomical and transcriptional landscape of early human fetal ovary development

Abstract

The complex genetic mechanisms underlying human ovary development can give rise to clinical phenotypes if disrupted, such as Primary (or Premature) Ovarian Insufficiency and Differences of Sex Development. We combine single-nuclei RNA sequencing, bulk RNA sequencing, and micro-focus computed tomography to elucidate the anatomy and transcriptional landscape of the human fetal ovary across key developmental timepoints (Carnegie Stage 22 until 20 weeks post conception). We show the marked growth and distinct morphological changes within the fetal ovary at the critical timepoint of germ cell expansion and demonstrate that the fetal ovary becomes more transcriptomically distinct from the testis with age. We describe previously uncharacterised ovary developmental pathways, relating to neuroendocrine signalling, energy homeostasis, mitochondrial networks, and inflammasome regulation. We define transcriptional regulators and candidate genes for meiosis within the developing ovary. Together, this work advances our fundamental understanding of human ovary development and has relevance for human ovarian insufficiency phenotypes.

Introduction

In humans, testes or ovaries develop from the bipotential gonad at around 42 days post conception (dpc)¹. It is well established that testis determination is driven by expression of the male-typical testis-determining factor, SRY, at this timepoint, leading to upregulation of SOX9 and a cascade of downstream pro-testis pathways². In contrast to this, an “ovary-determining factor” similar to SRY has not been identified, and for many years ovary development was considered a largely “passive” process occurring in the absence of SRY³. A key differentiating feature between ovary and testis development are the biological processes of germ cell expansion and meiotic entry that occur in the fetal ovary. These primordial follicles – representing an individual’s entire ovarian reserve and the germ cell pool for the next generation – are suspended in meiosis I until puberty many years later. The advent of high throughput transcriptomic approaches has begun to characterize ovary development and its key biological processes in previously unparalleled detail, revealing the process to be multigenic, complex and, still, only partially understood.

Initial microarray studies of gene transcription in developing mouse gonads were first reported at scale in 2007 and introduced the concept that a similar number of genes were differentially expressed in the ovary compared to the testis^4,5. Many of these genes were proposed to antagonize testis development or were involved in meiosis. Analysis of developing human gonad tissue eventually followed. An array-based RNA-sequencing atlas confirmed that the human fetal ovary had a discrete transcriptomic signature with clearly upregulated ovary-enriched genes⁶. Some of these had known roles in the developing ovary, such as NANOG and POU5F1 (OCT3/4), whereas the roles of others remained elusive. Pathway enrichment analysis revealed little further insight with most genes not annotated to known biological functions.

More recently, a bulk RNA-sequencing (bulk RNA-seq) study of human developing gonadal tissue between 6 and 17 weeks post conception (wpc) demonstrated several expected features, such as marked upregulation of meiotic transcripts, but also some novel findings, such as an enrichment of non-coding RNAs and genes with no clear link to ovary development (e.g., neuropeptide TAC1 and neurexin NRXN3)⁷. A limited number of single-cell RNA-sequencing (scRNA-seq) studies have recently built upon these foundations to demonstrate important cell-cell interactions, ovary developmental pathways, and gonadal somatic and germ cell crosstalk (such as FGF9-FGFR2 and PROS1-TYRO3)^8,9. New technologies have facilitated further multi-modal -omic analyses; particularly, the integration of single cell ATAC-sequencing (assay for transposase-accessible chromatin with sequencing) with single-cell data has elucidated novel post-transcriptional regulatory layers in the developing ovary^8,9. Lastly, larger-scale analyses involving multiple species demonstrate human-specific aspects of gonadal development; for example, defining transcription factors driving human germ cell differentiation⁹. These findings highlight the potential limitations of applying animal RNA-seq data to humans and the importance of studying human gonadal tissue where possible. Using bulk and single-cell technologies in isolation to assess transcriptomic expression in tissues has limitations. Single nuclei RNA-seq (snRNA-seq) provides granular transcriptomic detail, localizes gene expression to individual cell populations, and reconstructs cellular differentiation networks in silico. However, sensitivity can be low and only a proportion of the transcriptome may be captured. Bulk RNA-seq captures a higher proportion of transcriptomic activity, but sensitivity is also compromised in that low-level expression from small but important cell populations can be lost¹⁰

Clinically, if the genetic mechanisms that span a female reproductive lifecourse from fetal life to adulthood are disrupted, ovarian dysfunction can result. This manifests in a range of clinical phenotypes including ovarian insufficiency occurring in isolation (Primary Ovarian Insufficiency, POI; also known as Premature Ovarian Insufficiency), as a feature of an extra-ovarian syndrome (e.g., Perrault Syndrome), or as part of a Difference of Sex Development (DSD) (e.g., aromatase [CYP19A1] deficiency). POI affects up to 1% of women and results in infertility, a need for lifelong hormone replacement therapy, and medical co-morbidities including cardiovascular and bone health consequences. DSD is less common and represents several conditions diagnosed at different timepoints across the reproductive lifecourse. Next generation sequencing approaches have increased the genetic diagnostic yield in reproductive disorders – for example, a genetic cause is identified in over 20% of women with POI – although most remain unexplained^11,12. A more granular understanding of the pathogenic mechanisms responsible for the genetically heterogenous condition of ovarian insufficiency is key for an individualized approach to clinical management and reproductive counseling. Further, this knowledge may be used to develop novel fertility preservation techniques, particularly if the pre-symptomatic period is exploited before complete loss of ovary function.

Here, we combine snRNA-seq and bulk RNA-seq datasets, as well as further validation using other established datasets, to provide insight into the anatomy and transcriptional landscape of the developing human fetal ovary across a critical time period. This combinatorial approach has confirmed established pathways and revealed potential novel mechanisms.

Methods

Ethical approval procedures for use of human tissue

Human embryonic and fetal samples used for these studies were obtained from the Human Developmental Biology Resource (HDBR). The HDBR is a Medical Research Council (MRC)/Wellcome Trust-funded tissue bank regulated by the Human Tissue Authority (www.hdbr.org). Samples were collected by the HDBR with appropriate maternal written informed consent and with ethical approval from the NRES London-Fulham Ethics Committee (08/H0712/34 + 5, 18/LO/0822,) and Newcastle Ethics Committee (08/H0906/21 + 5, 18/NE/0290). This specific project has study approval from the HDBR (project numbers 200408; 200481; 200581). All researchers involved in this study underwent appropriate Human Tissue Act training, and all experiments were performed in accordance with relevant guidelines and regulations.

Tissue sample analysis

Ovaries, testes and other relevant 46,XX tissues were identified and isolated by blunt dissection. The age of embryonic and fetal tissue was calculated by experienced HDBR researchers (London and Newcastle) using published staging guidelines, such as Carnegie staging for embryos up to 8wpc and foot length and knee-heel length in relation to standard growth data for older fetuses. Samples were karyotyped by G-banding or quantitative PCR (chromosomes 13, 16, 18, 22, X, Y) to ascertain the sex of the embryo or fetus. Samples were stored at -70 °C or fixative (10% formalin or 4% paraformaldehyde) prior to use.

Bulk RNA-seq was undertaken using a total of 47 individual embryonic/fetal organs at four developmental stages (CS22-23 (7.5 to 8 wpc); 9-10wpc; 11-12wpc; 15-16wpc). This included 19 ovaries (46,XX; n = 5 biological replicates at each of CS22/23, 9/10wpc, and 11/12wpc; n = 4 replicates at 15/16wpc); 20 testes (46,XY; n = 5 replicates at each of the four developmental stages); and eight 46,XX control tissues (two different tissue samples per development stage: CS22/23 kidney, skin; 9/10wpc muscle, spleen; 11/12wpc lung, stomach; 15/16wpc pancreas, skin (Supplementary Fig. 1). The selection of control tissue followed the same collection, storage, and dissection processes as the gonadal tissue and ensured a mix of endo-, meso-, and ectoderm cell lineages across each developmental stage included in the experimental design. To note, control samples were not pooled together at the dissociation stage; rather, RNA from individual samples was sequenced and these samples were considered together as a control group for bioinformatic analysis. Additionally, control samples and gonadal samples were taken from different individuals where possible.

For single-nuclei sequencing, two 46,XX fetal ovaries (12wpc and 13wpc) were obtained from the HDBR, following ethical approval and consent as described above. The karyotype of the tissue was confirmed on extracted DNA from skin biopsy and array. Samples were stored at -70 °C prior to use.

Fetal ovary measurement

Fetal ovaries between CS23 and 20wpc were obtained from HDBR and fixed in 4% PFA. Ovary dimensions were measured to the nearest mm. Ovary weights were measured using an analytical balance (Pioneer PX, Ohaus).

Micro-focus computed tomography (micro-CT) of fetal ovaries in situ

Human fetuses (14–21 wpc, undergoing micro-CT post-mortem imaging procedures) were prepared for imaging by immersing in 2.5% potassium tri-iodide for between 4 and 16 days dependent on weight, and dried and positioned for scanning, as described previously¹³. Micro-CT scans were carried out using a Nikon XTH225 ST or Med-X scanner (Nikon Metrology, Tring, UK) with the following factors optimized for each fetus, depending on size: Tungsten target, X-ray energy 130 kV, current 134–239 µA (power 17–31 Watts), exposure time 250 ms, two frames per 3141 projections, detector gain 24 dB, scan duration 26 min. Modified Feldkamp filtered back projection algorithms were used for reconstructions within proprietary software (CTPro3D; Nikon Metrology) and post-processed was undertaken using VG StudioMAX (Volume Graphics GmbH, Heidelberg, Germany) to create both axial and coronal images at 30–54 μm isotropic voxel sizes.

Micro-CT of fetal ovary surface morphology

Surface scanning was undertaken of a fetal ovary (20wpc) and attached Fallopian tube stored in 10% formalin at 4 °C. Iodination was performed by immersing the sample in 1.25% potassium tri-iodide (I₂KI) for two days, then rinsing and drying it. The sample was then wrapped in parafilm and secured to a carbon-fibre plate to eliminate any movement or desiccation during the scan. Micro-CT scans were carried out using a Nikon XTH225 ST scanner (Nikon Metrology, Tring, UK) with the following factors: Tungsten target, X-ray energy 100–110 kV, current 46–60 µA (power 6.6 Watts), exposure time 1420–2000 ms, one frame per 3141 projections, detector gain 24 dB, scan duration 75 min. Modified Feldkamp filtered back projection algorithms were used for reconstructions within proprietary software (CTPro3D; Nikon Metrology) and post-processed was undertaken using VG StudioMAX (Volume Graphics GmbH, Heidelberg, Germany) to create the images at 3.78–4.77 μm isotropic voxel sizes.

Bulk RNA-sequencing

RNA extraction

RNA was extracted using the AllPrep DNA/RNA Mini Kit (QIAGEN N.V.) in accordance with the manufacturer’s protocol. In brief, tissues frozen at -70 degrees were homogenized using an electronic pestle (Kimble, New Jersey) and the lysate was transferred to an AllPrep DNA spin column placed in a 2 ml collection tube. This was centrifuged for 30 s at 8000 g and 250 µl 100% ethanol was added to the flow-through and pipette mixed. Following this, 700 µl of the sample was transferred to a RNeasy spin column and centrifuged for 15 s at 8000 g. The flow-through was transferred to a 2 ml collection tube and 700 µl of Buffer RW1 added. A 15 s centrifugation followed at 8000 g to wash the spin column membrane. Next, 500 µl of Buffer RPE was added to the RNeasy spin column and centrifuged for 15 s. The flow-through was discarded and 500 µl of Buffer RPE again added followed by a longer two-minute centrifugation at 8000 g. Extracted RNA was diluted in 30µL of nuclease-free water applied directly to the spin column membrane. RNA concentration and 260/280 ratios were measured using a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Witec, Littau, Switzerland). The minimum amount of RNA required for sequencing was 50ng with a 260/280 ratio of > 2.0. RNA integrity number (RIN) was assessed using an Agilent Bioanalyser (Agilent, Santa Clara, CA); all samples had a RIN > 7.

Library preparation and RNA sequencing

Library preparations were carried out on a Hamilton StarLet (Hamilton, Reno, NV) robotic platform and library qualitative checks performed using the Tapestation 4200 platform (Agilent, California, USA). Libraries were prepared using the KAPA RNA HyperPrep Kit and sequenced on an Illumina HiSeq^® 4000 sequencer (Illumina, San Diego, CA) at a minimum of 25 million paired end reads (75 bp) per sample.

Bioinformatic analysis

Fastq reads underwent QC (FastQC, Babraham Bioinformatics) and were aligned against the genome (GRCh38)to generate BAM files (STAR 2.7)¹⁴. featureCounts (Subread package, v2.0.2), DESeq2 (v1.28.1), and degPatterns (DEGreport package, v1.24.1) were used for gene expression quantification, differential gene expression, and detection of expression patterns, respectively^15,16. The cut-off adjusted p-values and log₂fold changes were 0.05 and 1, 1.5, or 2, respectively. Metascape was used for gene annotation and pathway enrichment analysis¹⁴. The ComplexHeatmap package in R was used to generate heatmaps representing differentially expressed genes¹⁷. All R packages and versions used in the bioinformatic analysis pipeline are listed in Supplementary Methods 1.

Single-nuclei RNA-sequencing (snRNA-seq)

Single-nuclei dissociation

A previously published protocol was used to dissociate tissues into single-nuclei suspensions (Martelotto et al., Broad Institute (https://doi.org/10.17504/protocols.io.bw6qphdw). All samples and reagents were kept on wet ice or at 4 °C throughout. In brief, tissue samples were diced using a scalpel into approximately 1 mm³ pieces. Samples were then dissociated in 300µL of Salty Ez10 Lysis Buffer supplemented with 0.2-0.5U/µL of RNase inhibitor (Sigma-Aldrich) using a 2 ml Kimble douncer (Sigma-Aldrich)(10 strokes with loose pestle and 10 strokes with tight pestle) to achieve single nuclei suspensions. An extra 700 µl of chilled Salty Ez10 Lysis Buffer supplemented with 0.2-0.5U/µl of RNase inhibitor was added to the samples and gently pipette mixed. Samples were incubated on ice for five minutes and slowly pipette mixed on 2–3 occasions during incubation. A MACS 70 μm filter (Miltenyi Biotec) was placed in a 50 ml Falcon tube on ice. The single nuclei suspension was transferred to the filter and then the filtrate transferred to a 1.5 ml LoBind tube (Eppendorf). Nuclei were centrifuged at four degrees at 500 g for five minutes. The supernatant was removed to leave a pellet. Salty Ez10 Lysis Buffer was added (1 ml) and the pellet resuspended gently before a further five-minute incubation on ice and five-minute centrifugation as before. The supernatant was removed, 500 µl of Wash and Resuspension Buffer 2 (WRB2) added without disturbing the pellet, and the sample left to sit on ice for five minutes before being gently resuspended. Samples were counted using the Luna-FL™ Dual Fluorescence Cell Counter (Logos Biosystems) and Acridine Orange/Propidium Iodide (AO/PI) Cell Viability Kit Counter (Logos Biosystems) (9 µl of sample with 1 µl of Acridine Orange). Both samples were within the ideal concentration of cells per µl (ideal range 700-1200cells/µl; 46,XX: 12wpc 1720cells/µl; 46,XX 13wpc: 1170cells/µl). Components for the two buffers, made up to a 100 ml stock, are shown in Supplementary Methods 2.

Gel bead in emulsion (GEM) generation and barcoding

Libraries from single nuclei suspensions were processed using the 10X Chromium Single Cell 3’ kit as per protocol (10X Genomics, Pleasonton, CA). In brief, MasterMix was prepared on ice (1X: 18.8 µl RT Reagent B, 2.4 µl Template Switch Oligo, 2.0 µl Reducing Agent B, 8.7 µl RT Enzyme C). Following this, 31.8 µl of MasterMix was added to an 8-tube PCR strip on ice. The Chromium Next Gel bead in EMulsion (GEM) Chip G was assembled as per instructions. Based on total cells/µl in each sample, the 10X Genomics Cell Suspension Volume Calculator table was used to calculate the ratio of nuclease-free water: single nuclei suspension to add to row 1 of the chip. Gel Beads were dispensed into row 2. Partitioning Oil (45 µl per sample) was dispensed into row 3. A total of 10,000 cells per sample were loaded onto the Chromium Next GEM Chip G (10X Genomics). The Chromium Controller was then run as per protocol for 18 min to form GEMs: a single nucleus, a single Gel Bead, and reagents.

Cleanup and cDNA amplification

After, 100 µl of GEMs were aspirated and incubated in a thermal cycler for 55 min as per protocol (Bio-Rad C1000 Touch). Recovery Agent (125 µl) was added to each sample at room temperature and left to incubate for two minutes resulting in a biphasic mixture of Recovery Agent and Partitioning Oil. From this mixture, 125 µl was discarded from the bottom of the tube. Dynabeads Cleanup Mix was then prepared (1X: 182 µl Cleanup Buffer, 8 µl DynaBeads MyOne SILANE, 5 µl Reducing Agent B, 5 µl Nuclease-Free Water) and 200 µl of this solution added to each sample and pipette mixed ten times followed by a ten-minute incubation. The samples were then placed on a magnetic separator (10X Genomics, Chromium™ Controller Accessory Kit) until the solution cleared. Supernatant was removed and 300 µl of ethanol added to the pellet while on the magnet. After 30 s, this ethanol was removed and the sample centrifuged briefly and air dried for one minute. Elution Solution 1 (35.5 µl) was then added to the sample, pipette mixed, and incubated for two minutes at room temperature. The solution (35 µl) was then transferred to a new PCR strip for cDNA amplification. Here, 65 µl of cDNA Amplification Reaction Mix was added to 35 µl of the sample and incubated inside the thermal cycler for 45 min as per protocol.

Library preparation using the 10X Genomics platform

A 3’ gene expression library was then constructed from the purified cDNA. Fragmentation, end repair, and A-tailing followed standard 10X Genomics workflow: 10 µl of purified cDNA, 25 µl of Buffer EB (1X: 5 µl Fragmentation Buffer, 10 µl Fragmentation Enzyme), and 15 µl of Fragmentation Mix were added to a PCR tube strip and mixed. The sample was then transferred to a pre-cooled thermal cycler (four degrees) and run as per protocol for 35 min. Double-sided size selection followed: 30 µl of SPRIselect (0.6X) reagent was added to each sample, mixed, and incubated for five minutes at room temperature. Samples were placed on the magnet until the solution cleared. Then, 75 µl of the supernatant was transferred to a new PCR strip and 10 µl SPRIselect (0.8X) added to each sample. Samples were incubated at room temperature for five minutes and then again placed on the magnet until the solution cleared. The supernatant was removed (80 µl) and 125 µl of ethanol added to the pellet. After 30 s, ethanol was removed. These size selection steps were repeated twice. Samples were then incubated at room temperature and 50 µl transferred to a new PCR tube strip for adaptor ligation, which involved adding 50 µl of Adaptor Ligation Mix to each sample, mixing, and incubating in the thermal cycler for 15 min. SPRIselect (0.8X) was used for post-ligation cleanup. Sample Index PCR, followed by a final SPRIselect (0.8X) cleanup step, completed the library preparation process.

Single-nuclei sequencing

Single-nuclei libraries were pooled and sequenced on the Illumina NovaSeq S2 v1.5 platform. Sequencing was paired-end using single indexing and a minimum of 20,000 read pairs per cell.

Bioinformatic analysis

The R package Seurat (v4.0.2) was used to generate a single-cell matrix as described previously¹⁸. Cycling cells were included in this analysis. Quality control (QC) filtering retained cells with feature counts > 400, percentage of mitochondrial genes < 1%, and percentage of ribosomal genes < 5% (Supplementary Fig. 5). SoupX (v1.6.2) was used to estimate and remove cell-free mRNA contamination. ParamSweep (v3.0) was used for parameter optimization and doubletFinder (v2.0.3) to remove doublets. The count matrix was normalized and 3000 variable genes were selected. After scaling, dimensionality reduction was performed using the first 30 principal components. Seurat packages FindClusters and RunUMAP were used to identify cell clusters and for Uniform Manifold Approximation and Projection (UMAP) visualization. The clustree R package (v0.5) was used to select a clustering resolution of 0.3 throughout. The 12wpc and 13wpc 46,XX ovaries were integrated for analysis using SCTransform (v0.3.5). Differential gene expression was performed using the FindAllMarkers function (‘min.pct = 0.25, logfc.threshold = 0.25’). Visualization functions within Seurat, including FeaturePlot, VlnPlot, and DotPlot, were used to visualize gene marker expression. All R packages and versions used in the bioinformatic analysis pipeline are described in Supplementary Methods 1. Key novel ovary-specific genes identified in our analysis were validated using the online interactive user tool and resource available at https://www.reproductivecellatlas.org (Roser Vento-Tormo, Wellcome Sanger Institute, Hinxton, UK; data published in Garcia-Alonso, L. et al. Single-cell roadmap of human gonadal development. Nature 607, 540–547 (2022). CC BY 4.0).

Statistical analysis

Statistical analyses were performed using GraphPad Prism v9.1.1 (GraphPad Software) or in R (v4.2.0) and are described in the relevant sections. A p value of less than 0.05 was considered significant. The Benjamini-Hochberg approach was used to adjust for multiple testing with cut-off adjusted p values of 0.05¹⁹. Data are shown as individual data points or as violin plots where appropriate.

Results

The ovary undergoes marked morphological changes during fetal development

In order to characterize changes in human ovary morphology between Carnegie Stage (CS) 22/23 (8wpc) and 20wpc (Fig. 1a), detailed measurements of ovary size were coupled with micro-focused computed tomography (micro-CT) analysis of anatomical location of the ovary in the developing female fetus, and with high resolution micro-CT-generated images of ovarian surface anatomy.

Fig. 1

Human fetal ovary growth and morphology during development. (a) Graphic depicting the antagonistic differentiation pathways of gonads arising from the adrenogonadal primordium at approximately CS17 and resulting in the formation of the fetal ovary or testis. Red arrows indicate experimental stages for the bulk RNA sequencing study; blue arrows indicate stages included for the single-cell RNA sequencing study; green arrows indicate microCT imaging. CS, Carnegie stage; wpc, weeks post conception. (b) Growth curves of the ovary between CS22 and 20wpc (left panel: weight (mg); right panel: length (mm)). Data for single glands (n = 27) are shown. (c) Micro-CT of the fetal ovary in situ at 12wpc, 14wpc, 16wpc, and 19wpc demonstrating growth and anterolateral migration of the ovary over time. Ovaries are indicated with red arrows. Upper panels show coronal views and lower panels show axial views. (d) Micro-CT images of a fetal ovary and associated Fallopian tube (FT) structures at 20wpc.

Fetal ovary growth curves (n = 27 samples) were generated that showed substantial changes in size across this critical time period, with the most marked increase in weight and length from 17wpc (Fig. 1b; Supplementary Table 1). The fetal ovary also undergoes distinct anatomical changes, as shown by coronal and axial micro-CT images of four ovaries in situ between 12wpc and 19wpc (Fig. 1c). These images demonstrate the anterior and lateral migration of both ovaries within the abdomen in early development, from their early location on the posterior abdominal wall near the adrenogonadal primordium to a more lateral position in the developing pelvic cavity. By 20wpc, micro-CT clearly shows that the ovary has a distinct blood supply entering the hilum and is in close apposition to the developing fimbriae of the Fallopian tube, as is typical of adult anatomy (Fig. 1d; Supplementary Video 1).

The fetal ovary and testis become transcriptomically distinct with age

As described in Methods, to define transcriptomic changes over time, bulk RNA-seq of ovary, testis, and control tissue was undertaken using a total of 47 individual embryonic/fetal organs at four developmental stages (CS22-23 (7.5 to 8 wpc); 9-10wpc; 11-12wpc; 15-16wpc; see Supplementary Fig. 1).

Gene expression profiles between the three groups – ovary, testis, and control – were compared using principal component analysis (PCA) and by generating heatmaps and cluster dendrograms. Samples clearly clustered together according to developmental stage and tissue type (Fig. 2a). PCA analysis demonstrated that the testis and ovary are more similar to one another than to corresponding control tissue, particularly at earlier developmental stages, and adopt a divergent transcriptomic profile with time (Fig. 2a, Supplementary Fig. 2).

Fig. 2

Transcriptome analysis during human fetal ovary development. (a) Principal component analysis of all 47 tissue samples included in this study (19 ovaries, 20 testes, 8 controls). The plot shows individual samples positioned according to the first and second principal components. RNA was extracted from each individual sample and RNA sequencing undertaken without pooling. Individual samples were grouped together as described for bioinformatic analysis. CS, Carnegie stage; PC, principal component; wpc, weeks post conception. (b) Overlay between three differential gene expression analyses: ovary v control (data from all stages combined); testis v control (data from all stages combined); ovary v testis (data from all stages combined)(log₂FC > 2, p.adj < 0.05). The total numbers of positively differentially-expressed genes (DEGs) are shown above the Venn diagrams. Total numbers and names of the top ten genes within the ovary-specific (red), testis-specific (green), gonad-specific (violet), and highly ovary-specific (pink) DEGsets are shown within the Venn diagram. “Highly ovary-specific” refers to DEGset in both the ovary v control and ovary v testes datasets. The top ten differentially expressed genes within the gonad-specific and highly ovary-specific genesets are shown in boxes. (c) Volcano plots demonstrating four differential gene expression analyses comparing ovary v testis samples at CS22/23 (both n = 5), 9/10wpc (both n = 5), 11/12wpc (both n = 5), 15/16wpc (ovary n = 4, testis n = 5). Each dot represents a single gene. Genes that were differentially expressed with a log₂FC > 2 and p.adj < 0.05 are coloured cyan; those coloured red did not meet these significance thresholds. A max value of log₁₀(p.adj) is displayed on the y-axis of the graph. Total numbers of positively and negatively DEGs are shown in violet. Gene names of the top ten positively and negatively DEGs across each developmental timepoint are displayed, based on log₂FC. (d) Heatmap showing normalized gene expression values for the top 30 differentially expressed highly ovary-specific genes identified in (b) across the entire count matrix. Genes are ordered according to descending log₂FC values. Samples are ordered according to tissue and stage (supervised). A color scale represents gene expression intensity (violet, lowest; yellow, neutral; red, highest). A p.adj value of < 0.05 was used throughout this analysis.

Differential gene expression analyses across all developmental timepoints were performed to identify positively and negatively differentially expressed genes (DEGs) between all (1) ovary versus (v) testis samples; (2) ovary v control samples; (3) testis v ovary samples; and (4) testis v control samples (Supplementary Tables 2–5). Overlaying highly DEGs (log₂FC > 2, p.adj < 0.05) within these analyses revealed ovary-specific (positively DEGs in the ovary v control or ovary v testis datasets); testis-specific (positively DEGs in the testis v control or testis v ovary datasets); and gonad-specific (positively DEGs in the ovary and testis within the four analyses) factors (Fig. 2b). A subset of “highly ovary-specific” genes (n = 228) was compiled from genes differentially expressed in both the ovary v control and ovary v testis analyses (Fig. 2b, Supplementary Table 6). Four time-specific differential gene expression analyses were also performed between ovary v testis samples at CS22/23, 9/10wpc, 11/12wpc, and 15/16wpc, and between ovary samples at CS22/23 and 15/16wpc (Supplementary Tables 7–11).

Globally, the ovary and testis had distinct transcriptomic signatures with increasing divergence in gene expression patterns over time: there were more differentially expressed genes between the later ovary and testis samples compared to between earlier ovary and testis samples (Fig. 2c, d). Indeed, more DEGs were seen in the developing ovary compared to testis both globally and at individual developmental stages, as well as a complex developmental trajectory, firmly rejecting the concept of ovary development being a “default” pathway (Supplementary Table 12). Transcripts of long non-coding RNA genes (lncRNAs) were also captured. Notably, there were significantly more differentially expressed non-coding RNA transcripts in the ovary compared to both testis and control samples (log₂FC > 2, p.adj < 0.05) (Supplementary Fig. 3). Of the 200 non-coding transcripts positively differentially expressed in the ovary compared to either control or testis, 65.7% (n = 135) were ovary-specific and 19.5% (n = 39) were highly ovary-specific (Supplementary Fig. 3, Supplementary Table 13).

Gene enrichment analysis of DEGs in the ovary compared to controls (n = 1229; log₂FC > 2; p.adj < 0.05) demonstrated an enrichment of gene ontology terms related to meiosis, gametogenesis, and/or oogenesis (n = 122 genes; 9.9% of total DEGs) as well as an enrichment of genes relating to neuroactive signaling (n = 55 genes; 4.5% of total; Table 1); steroid hormone metabolism (n = 55 genes; 4.5% of total); and piwi RNA (piRNA) processing (n = 28 genes; 2.3% of total) (Supplementary Fig. 4). Enrichment analysis of DEGs in the ovary compared to testis showed a similar enrichment of genes relating to meiosis (n = 61 genes; 9.7% of total DEGs) and neuroactive signalling, neurotransmitter development, and/or neural development (n = 101; 16.0% of total DEGs) (Supplementary Fig. 4). The highly ovary-specific geneset (n = 288) was analysed separately and was found to be particularly enriched for meiosis genes (n = 29; 10.1% of total geneset); G-protein coupled receptor (GPCR) signalling (n = 18; 6.3% of total DEGs); neuroactive signalling (n = 9; 4.1% of total DEGs); and steroid hormone biosynthesis (n = 8; 2.8% of total DEGs).

Table 1 Neuroactive genes in the fetal ovary.

Combining bulk RNA-seq and snRNA-seq provides novel insight into ovary development

To link the findings in bulk RNA-seq time series analysis to individual cell types, snRNA-seq was undertaken at a critical stage of ovary development: the onset of meiosis. This allowed key differentially expressed genes to be localized to specific cell populations within the developing human ovary, providing further insight into their function.

For this analysis, snRNA-seq profiled a total of 10,291 cells from one 46,XX 12wpc and one 46,XX 13wpc fetal ovary (quality control data: Supplementary Fig. 5) and yielded a UMAP projection of the snRNA-seq transcriptome of the peri-meiotic ovary (Fig. 3a). Cell annotation was assigned using established and recently identified markers identified in this study (e.g., GABRG1) and on markers identified from previous single-cell sequencing analyses of gonadal tissue (Garcia-Alonso et al., Nature, 2022) (Fig. 3b, Supplementary Table 14)^20,21. The bulk RNA sequencing and single-nuclei data were then used to localize key DEGs of interest to cell population and developmental stage. To validate this approach, the datasets were first used to explore genes with established function in ovarian development. Germ cell markers (e.g., DDX4, SYCP1, STRA8, MEIOC, DAZL) and somatic cell markers (e.g., GATA4, KITLG, NR2F2, IRX3) were expressed in the expected cell subpopulations and at predicted developmental timepoints (Fig. 3b). Next, the single-nuclei data were used to explore highly differentially expressed ovary-specific genes discovered in the bulk RNA-seq dataset (log₂FC > 2.5) that had no, or limited, reported associations with ovary development.

Fig. 3

snRNA-seq of two 46,XX human fetal ovaries. (a) UMAP (uniform manifold approximation and projection) of cell lineages across the merged dataset, which represents 10,291 cells from two 46,XX ovaries. Individual cell populations are annotated. FGC, fetal germ cells; OSE, ovarian surface epithelium; PGC, primordial germ cells. Markers were based on prior knowledge, markers identified in this study (e.g., GABRG1), and on previously published markers (Garcia-Alonso et al., Nature, 2022). (b) Upper panel: Dot plot showing the relative average expression and percentage of cells expressing germ cell markers used for snRNA-seq analysis of the developing ovary. Lower panel: Dot plot showing the relative average expression and percentage of cells expressing somatic cell markers. FGC, fetal germ cells; OSE, ovarian surface epithelium; PGC, primordial germ cells. (c) Violin plots showing key differentially expressed genes in the fetal ovary localized to the merged 11/12wpc 46,XX snRNA-seq dataset (left panel; see Fig. 3a for cluster annotation) and to the bulk RNAseq data at four developmental stages. (d) Dot plot demonstrating expression within different single-cell populations of potentially upregulated novel ovary genes, some of which are related to metabolic function or cell proliferation. e Dot plot demonstrating expression within different single-cell populations of key MAGE family genes.

Several potentially novel patterns emerged. Highly differentially expressed genes with no known role in gonadal development were identified. Throughout, key differentially expressed genes were validated using the Garcia-Alonso single-cell RNA-sequencing gene expression data (Supplementary Fig. 6–8)⁸. For example, CRYM, a highly-specific ovary gene here (ovary v testis log₂FC 4.93; ovary v control log₂FC 4.91), has no previously described function in ovarian development but in these data²²CRYM expression localized to the pre-granulosa cell clusters and was highly expressed from 9/10 weeks onwards (Figs. 2d and 3c; Supplementary Fig. 6)^22,23. CYP19A1, encoding aromatase (an enzyme essential for estrogen biosynthesis and folliculogenesis), was another highly ovary-specific gene in our dataset (ovary v testis log₂FC 4.13; ovary v control log₂FC 4.05). CYP19A1 is known to be expressed from granulosa cells; interestingly, here, it was also expressed in oogonia cell populations (Fig. 3c). Another positively differentially expressed gene in the ovary across multiple analyses was NLRP7, a modulator of cellular inflammation with no recognised role in human ovary development (ovary v testis log₂FC 3.16; ovary v control log₂FC 9.56; Supplementary Fig. 6). Expression localized to all primordial germ cell (PGC), FGC, and oogonia cell clusters (Fig. 3d). C15orf48, another highly expressed ovary gene (ovary v testis log₂FC 4.17; ovary v control log₂FC 4.70), is also known to modulate the intracellular inflammatory response²⁴. C15orf48 expression localized to pre-granulosa cells as well as oogonia and had increased expression at meiosis (Fig. 3d; Supplementary Fig. 6).

There was also high expression within the developing germ cells of a gene subset associated with mitochondrial metabolism. These included the highly ovary-specific genes SLC25A31 (ovary v control log₂FC 5.39; ovary v testis log₂FC 3.22) and OTUD6A (ovary v control log₂FC 7.56; ovary v testis log₂FC 3.29)(Supplementary Fig. 6). Mice with Slc25a31 mutations cannot effectively utilize mitochondrial ATP and their spermatogonia undergo arrest at the leptotene stage and subsequent apoptosis²⁵. In the context of human colorectal cancer, low expression of OTUD6A leads to suppressed mitochondrial fission and reduced cell growth of proliferating cells²⁶. The expression of both OTUD6A and SLC25A31 in oogonia is consistent with a role in proliferating cells (Fig. 3d; Supplementary Fig. 6). DNM1L facilitates mitochondrial fission by increasing the expression of OTUD6A; DNM1L also had marked increase at the time of meiosis in this dataset²⁶.

Additionally, several DEGs with acknowledged roles in male gonadal and germ cell development, but with little previously described roles in the ovary, were identified. For example, MAGE genes (melanoma antigen genes) are members of the cancer-testis family classically associated with spermatogonia development where they protect the male germline against environmental stress²⁷. MAGE genes have little known function in female tissues and are not expressed in the adult ovary in the Human Protein Atlas²⁸. However, MAGEA1 has been associated with proliferating germ cells in the developing ovary^29,30. In our dataset, MAGEA1 was differentially expressed in the ovary compared to control tissue (log₂FC 2.77). Other MAGE genes were very highly differentially expressed in the ovary; for example, MAGEB2, MAGEC1, MAGEA10, and MAGEB1, all X chromosome genes, were among the top 50 differentially expressed genes in the ovary compared to controls (log₂FC 9.82, log₂FC 9.70, and log₂FC 8.66, respectively) and MAGEA10 was one of the most highly ovary-specific genes (Figs. 2d and 3e, Supplementary Table 15). Although overall counts were modest, the expression of these genes clustered to germ cell subpopulations in snRNA-seq data (Fig. 3e). These data suggest an unexplored function for the MAGE gene family during germ cell development in the fetal ovary.

The fetal ovary is enriched for a distinct subset of transcription factors

A focused analysis of transcription factor (TF) expression in the developing ovary was conducted to identify potentially novel regulators of gene expression (n = 1,638 TFs included, 6.8% of coding genes; geneset from AnimalTFDB v4.0; accessed 15/3/23)³¹. Broadly, there was an enrichment of TFs within DEGs in the developing ovary compared to testis across most developmental stages (Supplementary Table 16). There was a similar enrichment of TFs within DEGs in the ovary compared to control (Supplementary Table 16).

The set of differentially expressed TFs in the ovary compared to testis (n = 77, log₂FC > 2, padj < 0.05) was investigated further to identify ovary-specific transcriptional regulation networks. Globally, most TFs belonged to Homeobox (n = 31; 40.3%), zf-C2H2 (n = 14; 18.2%), helix-loop-helix (bHLH) (n = 7; 9.1%), or Fork-head (n = 6; 7.8%) TF families. Individual stage-specific analyses demonstrated an increase of Homeobox family TFs from 9/10wpc onwards (e.g., LHX2, IRX3, IRX5, NOBOX); a corresponding decrease in zf-C2H2 TFs from CS22/23 (e.g., ZNF727, ZNF729) (Fig. 4a); and a subset of TFs highly expressed at meiosis (15/16wpc) (Fig. 4b). Overlaying differentially expressed TFs (log₂FC > 2, padj < 0.05) from the ovary v control, ovary v testis, testis v control, and testis v ovary differential gene expression analyses yielded gonad-specific, ovary-specific, and testis-specific TFs (Fig. 4c and d). Gene enrichment of ovary-specific TFs revealed gene ontology terms including sex differentiation, gametogenesis, central nervous system differentiation, and olfactory bulb development (Fig. 4e). Several of the differentially expressed TF-encoding genes identified above have established functions in the developing ovary, such as FOXL2, LHX8, NOBOX, LIN28A, POU5F1, FIGLA, and PRDM14, and gonad-specific roles for others were described in a recent single-cell atlas²⁰. Some, such as ZIC1, have plausible biological links to gametogenesis or gonadal development but unknown roles in the developing ovary.

Fig. 4

Ovary-, testis-, and gonad-specific transcription factors in early development. (a) Relative proportions of differentially expressed transcription factors (TFs) in the ovary from the most represented TF families are shown across four developmental stages (log₂FC > 2, p.adj < 0.05). bHLH, Basic Helix-Loop-Helix; DEGs, differentially expressed genes; HMG, high mobility group; TF, transcription factor; zf-C2H2, Cys2–His2 zinc finger. (b) Heatmap showing top ten differentially expressed TFs at the 15/16wpc peri-meiosis stage (log₂FC > 2, p.adj < 0.05). (c) Heatmap showing expression of the top 15 ovary-specific TFs across the dataset (differentially expressed in both ovary v testis and ovary v control differential gene expression analyses). (d) Venn diagram demonstrating overlap between four differential gene expression analyses (log₂FC > 2, p.adj < 0.05; ovary v testis, ovary v control, testis v control, and testis v ovary) in order to identify ovary-, gonad-, and testis-specific TFs. All TFs are shown where possible, or else the total number (n) and a subset of most highly expressed in that category. (e) Gene enrichment analysis (Metascape) of ovary-specific TFs. (f) Dot plot localizing the expression of key TFs of interest to individual cell clusters on snRNA-seq analysis.

The bulk and snRNA-seq data were used to localize the expression of TFs of interest to cell subpopulations and developmental timepoints (Fig. 4c and f). Again, expression data were cross-validated using the Garcia-Alonso single-cell gonadal atlas (Supplementary Fig. 7)⁸. LHX2, a highly ovary-specific TF (ovary v testis log₂FC 5.92, ovary v control log₂FC 4.86) proposed to suppress endothelial cell migration and vascularization in the developing ovary³², localized to pre-granulosa cells and particularly to the ovarian surface epithelium (OSE) cluster (Fig. 4c and f; Supplementary Fig. 7)²⁰. The expression of SALL1 was higher in the ovary v testis (log₂FC 3.82) and localized to the pre-granulosa and ovary interstitial populations, albeit to small numbers of cells (Fig. 4f; Supplementary Fig. 7). Expression of DMRTC2 and DMRTB1 within developing oogonia has been previously shown in mice, macaque, and humans^8,33,34. Here, expression of these genes localized to oogonia populations and was highest at meiosis, providing further evidence for a meiotic role in humans (Supplementary Fig. 7). ARX, a gene important for mammalian germ cell development and associated with X-linked lissencephaly and testis dysgenesis, was also identified as a key differentially expressed TF in this dataset and localized to gonadal interstitial and OSE populations³⁵. BARX1, a Homeobox TF involved in Wnt signalling, has no known ovary role but was differentially expressed in the ovary compared to testis in our data (log₂FC 2.13). Its expression increased at meiosis and localized to PGCs and FGCs. TFs with roles in meiotic progression were also identified: MAEL, a piRNA gene governing microtubule organisation within the oocyte; TERB1, required for telomere attachment during meiotic homologous recombination; and ESX1, associated with retinoic-acid responsive FGC development. These three genes are required for normal meiosis in mice and monkeys^36,37,38,39; in humans, TERB1 and ESX1 pathogenic variants lead to spermatogenic failure and male infertility^39,40,41. In our dataset, MAEL, TERB1, and ESX1 were strongly expressed in the ovary, localized to germ cell populations, and exhibited peri-meiotic expression (Fig. 4f; Supplementary Fig. 7). Consistent with meiotic roles, MAEL and TERB peaked at 15/16wpc and clustered to later germ cell clusters (Fig. 4c and f). ESX1 had modest expression in FGCs, with highest expression at 11/12wpc.

A subset of nuclear receptors is expressed in the developing ovary

Nuclear receptors (NRs) are a key subfamily of transcription factors which often have roles in gonadal development and disease (e.g., NR5A1/steroidogenic factor-1, NR2F2/COUP-TFII)⁴². Given this, the ovary v control, ovary v testis, testis v control, and testis v ovary datasets were systematically analyzed for differential expression (log₂FC > 0.5, p.adj < 0.05) of the 48 known human NRs. As before, datasets were conflated to identify gonad-specific, ovary-specific, and testis-specific NRs (Fig. 5a and b). This approach revealed differential expression of NRs with established gonadal roles, such as gonad-specific expression of ESR1 (encoding the estrogen receptor-alpha), ESR2 (estrogen receptor-beta), and NR6A1 (encoding GCNF, germ cell nuclear factor), as well as NRs with less well-established roles in the developing ovary. For example, NR1H4, also known as the farnesoid X/bile acid receptor, was highly expressed at later ovary stages and, as seen previously, mostly localized to pre-granulosa IIb cells (Fig. 5b, c)⁸. PPARG (peroxisome proliferator activated receptor gamma) was an ovary-specific NR with expression at all developmental timepoints localizing to germ cell populations (Fig. 5b, c). NR2E1 (encoding TLX) and NR2E3 (encoding PNR), two NRs more typically associated with retinal development, showed ovary-specific expression that increased with age, most marked at meiosis⁴². The two key NRs previously linked to reproductive development, NR5A1 (encoding steroidogenic factor-1) and NR0B1 (encoding DAX-1), were both identified in the gonad v control genesets but more positively differentially expressed (i.e., higher log₂FC) in the developing testis (Supplementary Table 17).

Fig. 5

Nuclear receptor expression in the developing human fetal ovary. (a) Venn diagram showing overlapping differential gene expression analyses of nuclear receptors (NRs): ovary v control, ovary v testis, testis v control, testis v ovary (log₂FC > 0.5, p.adj < 0.05). Ovary-, testis-, and gonad-specific NRs are identified. (b) Heatmap showing differentially expressed NRs across the geneset (log₂FC > 0.5, p.adj < 0.05). (c) Selected highly expressed NRs localized to UMAPs from snRNA-seq data (upper panels) and as violin plots over time in the bulk RNAseq data (lower panels). See Fig. 3 for cluster annotation.

Complex neuroendocrine signalling exists in the fetal ovary

An enrichment of terms relating to neurotransmitter signaling, neuroendocrine networks, and neural development was a consistent feature of genes highly expressed in the developing ovary across multiple analyses (Fig. 6a, b and c; Supplementary Fig. 4). This finding was more marked at certain timepoints; for example, at 9/10wpc, 25.4% of all differentially expressed genes in the ovary v control (log₂FC > 2; p.adj < 0.05; n = 149/587 genes) related to these functions (Fig. 6c). Specific genes of interest emerged consistently from these analyses (Table 1). For example, NPY (encoding neuropeptide Y), a neuropeptide expressed by interneurons of the sympathetic nervous system, was differentially expressed in the ovary v testis (log₂FC 3.92), the ovary v control (log₂FC 4.07), and particularly highly expressed at earlier developmental timepoints (9/10wpc ovary v testis log₂FC 4.59). Receptors NPY1R (Y₁), NPY4R (Y₄), and NPY4R2 (paralogue NPY4R, Y_4 − 2) were also highly expressed in the developing ovary. NPY was highly expressed from large numbers of OSE cells as well as some pre-granulosa cell populations; NPY4R from small numbers of ovary interstitial, oogonia, and pre-granulosa cell populations; and NPY4R from small numbers of FGC populations (Fig. 6d). GABRG1, GABRA2, and GABRA4, coding for three of the six alpha subunits of the GABAergic receptor, were also highly expressed alongside NPY; both GABAergic genes and NPY expression localized to OSE clusters (Fig. 6a and b). Genes involved in neural differentiation, such as IRX3, IRX5, CRHR1, ASCL1, TAC3R, and TAC1, were also highly expressed in the early ovary (Fig. 6d). Together, these data suggest complex, ovary-specific neuroendocrine programmes exist in the human developing fetal ovary, particularly expressed from somatic cell populations.

Fig. 6

The fetal ovary is enriched for neuroendocrine and neurotransmitter genes. (a) Violin plots demonstrating the increased expression of selected neuroactive genes in the ovary at four developmental timepoints. Green, ovary; orange, testis. (b) UMAP representation of 46,XX ovary snRNA-seq data. Expression of NAV3 and GABRA2 localizes to the ovarian surface epithelial (OSE) cell cluster and to interstitial cell populations. See Fig. 3b for cluster annotation. (c) Gene enrichment analysis (Metascape) of differentially expressed genes (log₂FC > 2, padj < 0.05) in the 9/10wpc ovary demonstrating an enrichment of terms relating to neuroendocrine and neurotransmitter signalling. (d) Dot plot demonstrating expression of selected neuroactive genes within individual cell clusters in snRNA-seq analysis. FGCs, fetal germ cells; OSE, ovarian surface epithelium; PGCs, primordial germ cells.

A focused peri-meiotic analysis reveals new meiosis candidate genes

Meiosis begins at approximately 11/12wpc and peaks at 15/16wpc. To identify genes with convincing roles in meiosis, separate analyses of differentially expressed genes in the ovary v testis at 15/16wpc and in the 15/16wpc ovary v CS22/23 ovary were conducted (Supplementary Tables 6 and 10). Gene enrichment analysis (Metascape) annotated a high proportion of differentially expressed ovary genes in either of these two analyses with meiotic gene ontology terms (n = 209), including piRNA processing, meiotic cell cycle, homologous recombination and DNA repair, and cilium movement (Fig. 7a). To date, several pathogenic variants in meiosis genes have been associated with the pathogenesis of primary ovarian insufficiency (POI), many related to meiosis and/or DNA repair^43,44,45,46. Given this known association, we examined the overlap between differentially expressed genes in the developing ovary at meiosis with genes included on the “PanelApp” POI panel (a panel curated by Genomics England experts (v 1.67); accessed November 2024; doi: https://9.genomicsengland.co.uk/panels/155/); linked to the UK 100,000 Genome Study)(Fig. 7b, Supplementary Table 18)⁴⁷. In total, 25.0% (n = 17) of the 69 genes on this panel were differentially expressed (log₂FC > 2, p.adj < 0.05) across distinct developmental timepoints and 70.6% of these (n = 13 of 17) differentially expressed PanelApp genes were included in the 209 genes annotated with meiotic terms by enrichment analysis. This demonstrates that a high proportion of PanelApp POI genes have proven or possible meiotic functions.

Fig. 7

Identifying novel meiosis candidate genes. (a) Gene enrichment analysis (Metascape) of differentially expressed genes (log₂FC > 2, p.adj < 0.05) at 15/16wpc, confirming an enrichment of genes related to meiosis and germ cell competence. (b) Venn diagram overlapping the PanelApp geneset; differentially expressed PanelApp genes in the ovary v testis (log₂FC > 2, p.adj < 0.05); and meiosis-annotated genes on enrichment analysis of genes differentially expressed in the 15/16wpc ovary v 15/16wpc testis and 15/16wpc ovary v ovary CS22/23. (c) Genes differentially expressed in the 15/16wpc ovary v 15/16wpc testis and 15/16wpc ovary v ovary CS22/23 which were not annotated on enrichment analysis as meiosis genes were identified (n = 622) and overlapped with markers of the oogonia cluster (n = 33) to extract potentially novel candidate meiosis genes. (d) Dot plot showing expression of 33 potentially novel meiosis candidate genes across cell clusters.

We then combined transcriptomic datasets to uncover potential novel candidate meiosis genes. First, DEGs in both the 15/16wpc ovary v CS22/23 ovary and the 15/16pcw ovary v testis analyses that were not included in the list of 209 meiotic genes were identified (Fig. 7c). Next, these genes were overlaid with differentially expressed genes from the oogonia cluster in the snRNA-seq dataset. This approach led to the identification of 33 novel candidate genes for roles in meiosis and, accordingly, potential genetic causes of POI (Fig. 7c; Table 2; data validated in the previously described Garcia-Alonso gonadal atlas scRNAseq expression data⁸ (Supplementary Fig. 8)). Each of these candidate genes had increased expression at meiosis in the bulk RNAseq data and localized to meiotic germ cell clusters (oogonia +/- synaptic oogonia) in the snRNA-seq data (Fig. 7d).

Table 2 Novel meiosis candidate genes.

Discussion

We present an integrated multimodal study of human fetal ovary development during a critical early stage of growth and differentiation, extending from CS22/23 (8wpc) up until the second trimester. We use complementary multiomic and imaging approaches with the aim of defining the anatomical and molecular landscape during this critical period.

Using a novel approach of micro-CT, we have demonstrated, three-dimensionally, the significant morphological change occurring within the fetal ovary within a relatively narrow time window. We show the antero-lateral movement of ovaries in the fetal abdomen throughout the second trimester and, using high resolution surface micro-CT imaging, demonstrate that by 20wpc the human ovary has a complex structure with hilar vascularization and close apposition to the fimbriae of the oviduct (Fallopian tube), which has already adopted mature anatomical features by this stage. The ovary demonstrates marked growth and an increase in weight from 16wpc onwards, corresponding with the massive germ cell expansion and meiosis occurring at this time. By 20-22wpc, the ovaries of 46,XX fetuses have the maximum number of oogonia (approximately 7,000,000) they will hold during their entire reproductive lifetime, representative of their lifetime ovarian reserve. Disruption of this critical timepoint results in insufficient germ cell expansion and/or accelerated apoptosis of this ovarian reserve, resulting in the fewer germ cells at birth, smaller ovaries macroscopically, and the clinical phenotype of Primary Ovarian Insufficiency (POI).

A series of transcriptomic approaches were then used to show that the fetal ovary has a clear transcriptomic blueprint across all developmental timepoints defined by thousands of differentially expressed protein-coding genes as well as non-coding transcripts. Indeed, the fetal ovary has more differentially expressed genes at the developmental timepoints studied than the testes, once again confirming that ovary development is not a “default” pathway occurring in the absence of genes promoting testis development. No “ovary-determining factor” was identified, and more detailed investigation at earlier time points (from 6 wpc, coinciding with the onset of SRY expression in the testis) would be required to address this. However, sets of genes known to prevent testis development, such as FOXL2, were highly expressed, confirming its role in a major ovary developmental pathway^48,49. However, these data show that the fetal ovary has important and unique features beyond testis antagonism, including anatomical and vascular organization, which are linked temporally to the key biological ovary developmental processes of germ cell expansion, meiosis onset, and meiotic arrest.

Integrating bulk- and snRNA-seq throughout this work allowed several unique features of the developing ovary to be elucidated. Genetic networks relating to meiosis were highly expressed, with a particular enrichment of genes related to DNA repair. This highlights the vulnerability of the growing follicle pool to DNA damage and the implications of this for transgenerational epigenetics. Indeed, pathogenic variants in DNA repair genes are recognised genetic causes of POI – such as YTHDC2 and ZSWIM7, which we recently described^44,45.

We also describe a complex network of neuroendocrine signalling, mitochondrial pathways, and energy metabolism within the fetal ovary that begins as early in ovary development as CS22. NPY is highly expressed in the developing ovary, a gene which is classically expressed in the arcuate nucleus (ARC) of the hypothalamus where it dampens inhibitory GABAergic activity (modulated by GABRG1, also highly expressed in the developing ovary). NPY has known neuroendocrine functions, including food intake regulation and energy storage, but a potential role within a neuroendocrine signalling pathway in the developing ovary has not been described. Other neuroactive genes with roles in appetite and fat regulation in adult life – IRX3, IRX5, NHLH2 – were also highly expressed; IRX3 and IRX5 have postulated roles in obesity regulation as well as ovary folliculogenesis^50,51,52,] and mice require Nhlh2 for meiosis and fertility as well as for energy metabolism⁵³. PPARG, encoding for a top differentially expressed nuclear receptor (NR) in the ovary compared to testis, regulates immune cell activation as well as adipogenesis and has been detected in the adult human ovary previously, specifically within granulosa cells⁵⁴. Excessive intra-ovarian PPARG production within granulosa cells disrupts steroidogenesis and contributes to PCOS, obesity, and insulin resistance⁵⁴. However, we localise fetal PPARG expression to germ cell populations, suggesting a role in the developing ovary other than steroidogenesis – perhaps relating to energy homeostasis within germ cells or to inflammation regulation (specifically macrophage activity) within the ovary^55,56. The high expression of Nod-like receptors (NLRP7, NLRP4, NLRP14) within germ cell populations in our data also aligns with the concept of a role for “inflammasome” regulation in the developing ovary. The clinical phenotypes of obesity, central hypogonadotropic hypogonadism (CHH), and PCOS have previously linked fertility with energy metabolism and its neuroendocrine control – for example, excessive GABAergic activation and disrupted weight homeostasis are implicated in the pathogenesis of PCOS⁵⁷. Analysis of testis and control tissue data did not reveal an enrichment for neuroendocrine components. Thus, our data introduce the paradigm that these pathways are required within the ovary from early points in its development in fetal life. The specific roles of ovary-expressed neuroendocrine genes are unclear, especially the high expression in the OSE cluster of cells. Possibly, these findings relate to later activation of ovarian function at puberty, where both sufficient peripheral fat and muscle stores are required for pubertal onset^58,59, but further work is required to explore this in detail.

In addition to those with postulated neuroendocrine function, other neuroactive genes were highly expressed in the ovary from the CS22/23 timepoint. There was a marked expression of genes involved in sympathetic nervous system development, such as ASCL1, which interestingly localized to the ovarian surface epithelium (OSE) clusters – cells on the epithelial surface of the ovary – indicating an as yet unexplored role of sympathetic innervation to the ovarian capsule required for ovarian development. Other neurotransmitters, such as NAV3, also localized to OSEs and sometimes gonadal interstitium; notably, epitranscriptomic downregulation of this gene has been implicated in chemotherapy resistance in ovarian cancer, suggesting a possible role of these primitive neuroactive pathways within oncogenesis in later life⁶⁰. There was also a striking enrichment of genes related to sensorineural function, including retinal nuclear receptor encoding genes (TLX, PNR), as well as upregulation of other neurotransmitters (TAC1, TAC3R). Disruption of these pathways have been previously linked to CHH (olfactory dysfunction (anosmia) and hypogonadotropic hypogonadism in Kallmann Syndrome; TAC3R variants in CHH⁶¹) but not to ovary insufficiency phenotypes. Expression of olfactory receptors outside of olfactory sensory neurons is increasingly recognized and olfactory receptors have been previously identified as highly expressed in the fetal ovary⁶. Postulated non-olfactory roles for these receptors include cell-cell recognition, migration, proliferation, and apoptosis⁶², although any potential role for olfactory receptors and their respective ligands in human ovary development remains unexplored. Taken together, we propose that the fetal ovary has specific neuroendocrine axes and energy homeostasis programmes which are required for its development and lay down the infrastructure required for ovarian function much later in adult life.

This study has certain limitations, including the relatively few samples included for snRNA-seq which necessitated a focus on the key meiotic timepoint of 12wpc. Bulk RNA-seq has inherent limitations, including the risk of overlooking small but important signals from smaller cell populations, which was mitigated by cross-referencing with snRNA-seq data and by designing a well-powered study with appropriate control samples. Key strengths of this study are its integrated investigative approach and its focus on clinically relevant biological timepoints.

We describe several novel candidate genes, genetic pathways, and signaling networks with potential roles in ovary development which remain unexplored and require interrogation in future studies. Particularly, the ovary-specific neuroendocrine networks, mitochondrial pathways, and novel meiosis genes identified here warrant further investigation. We show that the fetal ovary has a complex genetic programme which underlies a developmental trajectory characterized by critical biological processes which are essential for ovarian function and may hold relevance for clinical human ovarian insufficiency phenotypes.