Follicles, including those from preantral, small antral, and preovulatory stages, were procured from women aged 23 to 36 for messenger ribonucleic acid (mRNA) microarray analysis. These datasets have been previously documented in published studies (8–11), and their summarized details are presented in Table 1 .

Preantral and small antral follicles were harvested from women undergoing fertility preservation who had one ovary surgically excised (9, 11). For fertility preservation purposes, only the ovarian cortex containing primordial follicles was utilized, while the medullary tissue was either discarded or allocated for research, subject to patient consent (12).

In addition, small antral follicles were procured from the medulla tissue of two patients undergoing fertility preservation for subsequent immunohistochemical (IHC) and immunofluorescence (IF) analysis. These patients, aged 30 and 29 years, had been diagnosed with breast cancer and cervical cancer, respectively.

Granulosa cells were collected from preovulatory follicles both before induction of final maturation (pre-OI) and at oocyte pickup (post-OI). These cells were obtained from women undergoing ovarian stimulation for in vitro fertilization (IVF) treatment. Paired samples of GCs were acquired from nine patients: one sample pre-OI and another at oocyte pickup (OPU). These samples were subjected to microarray analysis; the mean age of the patients was 27.9 ± 3.4 (SD) (10).

Furthermore, mural GCs and cumulus cells from preovulatory follicles at the time of OPU were obtained from women undergoing IVF or intracytoplasmic sperm injection for microarray analysis (8). These patients received either 5000 IU hCG (pregnyl, MSD) or 0.5 mg GnRHa (Suprefact, Sanofi-Aventis) for final maturation of follicles.

The utilization of surplus ovarian tissue, encompassing follicles and granulosa cells, obtained in conjunction with ovarian tissue cryopreservation (OTC), received approval from the Scientific Ethical Committee for the Capital Region (Approval No. H-2-2011-044).The collection and use of GCs from both pre- and post-OI follicles, acquired during IVF treatment, were conducted in compliance with the Helsinki Declaration II and were approved by the Danish Scientific Ethical Committee (Approval No. SJ-156). The utilization of paired mural granulosa cells and cumulus cells from preovulatory follicles at OPU in connection with IVF treatment was approved by the Danish Ethical Committee (Approval No. VN2004/61). Informed consent was obtained from all participants prior to their inclusion.

Preantral follicles, ranging in diameter from 40 µm to 200 μm, displaying an intact membrane with granulosa cells and an oocyte, were isolated from the medullary tissue, following established protocols (9, 13). These follicles were rapidly frozen in liquid nitrogen or lysed in RNA lysis buffer and subsequently stored at -80 °C until RNA extraction.

Small antral follicles, measuring 4 to 6 mm in diameter, designated for microarray analysis, were aspirated using a 26 G needle attached to a 1 mL syringe from ovaries acquired for fertility preservation intended for OTC. Following aspiration, the follicle fluid (FF) and GCs were separated through centrifugation at 300g for 2 minutes. The isolated GCs were then subjected to two washes with PBS and preserved at -80 °C until further analysis, as previously described (14).

Additionally, small antral follicles, ranging from 0.5 to 6 mm in diameter and isolated from the medullary tissue for histological analysis, were either fixed in Bouin’s solution for immunohistochemical analysis or in formalin for immunofluorescence analysis. Subsequent, they underwent a series of washes in graded ethanol, were embedded in paraffin, and sectioned into 5 μm sections.

The paired pre- and post-OI follicles were punctured using transvaginal ultrasound guidance at 13 hours (N=4), 37 hours (N=4) and 61 hours (N=1) prior to hCG administration (pre-OI) and at post-OI (10). The isolation of GCs was described previously (15). Mural GCs and cumulus cells post-OI were obtained 34 hours after hCG or gonadotropin-releasing hormone agonist (GnRHa) administration, as previously described (8).

All microarray data sets were conducted at the same core facility (the Microarray Center at Rigshospitalet, Copenhagen) and used the Affymetrix Human Gene ST v1.0 GeneChip array (Affymetrix, Santa Clara, California, USA). The microarray data were validated using RT-PCR and the results are available in previous publications (8–11).

Cell intensity files, commonly known as CEL files, were generated using the GeneChip Command Console software (AGCC, Affymetrix). Subsequently, these CEL files were imported into the R software environment (version 4.2.2) (16, 17) for further analysis. To enhance the quality and reliability of the microarray data, we employed Frozen Robust Multiarray Analysis (fRMA, version 3.16) (18) for microarray preprocessing. This method not only combined data from different batches to mitigate technical variability but also normalized the data to eliminate batch-specific differences, thereby enhancing data reproducibility and comparability. Microarray gene expression data were presented as log₂-transformed expression levels.

We excluded samples that were contaminated with leukocytes. The presence of leukocyte contamination was determined by assessing the expression of the leukocyte-specific marker cluster of differentiation 45 (CD45), also known as the protein tyrosine phosphatase receptor type C (PTPRC) gene. Two samples displaying elevated PTPRC expression were consequently excluded from the study. However, the remaining samples exhibited low PTPRC expression, and these expression levels did not exhibit statistically significant differences between the samples (see Supplementary Table S1 , Supplementary Figure S1 ).

To gauge the level of gene expression noise in the arrays, we utilized the expression of two Y-chromosome-located genes, namely testis-specific protein Y-linked 1 (TSPY1) and deleted in azoospermia 4 (DAZ4). The log₂ expression values of TSPY1 and DAZ4 ranged from 3.2 to 5.4, as documented in Table 2 and Figure 2A . Importantly, no statistically significant differences were observed between the groups, as indicated by q values exceeding 0.05, and the absolute log₂ fold changes remained less than 1. In this study, we considered log₂ gene expression levels below 5.4 as indicative of noise.

Ovarian tissue from seven women were collected in connection with fertility preservation. None of the women received any kind of ovarian stimulation (19). RNA sequencing data (GSE107746) of human oocytes and corresponding GCs from 5 different follicular developmental stages were downloaded from the GEO database (https://www.ncbi.nlm.nih.gov/gds/). Only the data of GCs samples were extracted for the present study. Data normalisation and other pre-processing steps were skipped because the obtained data were already in the log₂ of Fragments Per Kilobase of transcript per Million mapped reads (FPKM) plus one format. The data was first published by Zhang and co-workers (19).

Chromogenic immunohistochemical analysis were conducted following established procedures (20). Briefly, tissue sections were de-paraffinized using tissue-clear solution (Sakura Finetek, Brøndby, Denmark, Cat. No.: 1466) and subsequently rehydrated through graded ethanol. Antigen retrieval was achieved using Tris-EGTA buffer (10 mM Tris, 0.5 mM EGTA, pH 9). To reduce endogenous peroxidase activity, 3% H₂O₂ was applied. Non-specific binding was blocked by incubating sections with 4% bovine serum albumin (BSA) and 5% species-specific normal serum (normal donkey serum, Abcam, Cat. No.: ab7475) before primary antibody application. Sections were incubated with primary antibodies overnight at 4°C in a humid chamber. Primary antibodies used included HSD17B1 (1:100, Abcam, Cambridge, UK, cat. no.: ab51045), CYP17A1 (1:200, Santa Cruz, Dallas, US, Cat. No.: SC-46084), and HSD3B2 (1:2000, Abcam, Cat. No.: 154385). After washing in TBS with Tween20®, sections were incubated for 30 minutes at room temperature with the appropriate HRP-conjugated secondary antibodies (donkey-anti-rabbit, Invitrogen by Thermo Fisher, Roskilde, Denmark, Cat. No.: 31458; donkey-anti-goat, Abcam, cat. No.: ab97110) and washed in TBS with Tween 20®, and counterstained with Meyers haematoxylin for 2 min. Visualization was achieved through peroxidase reaction using 3,3’-diaminobenzidine tetrahydrochloride (DAB substrate kit Abcam, Cat. No. ab64238).

The procedure for immunofluorescence was carried out essentially as previously described, with few modifications (14). In brief, sections with small antral follicles were de-paraffinized in Neo-clear® (Sigma Aldrich, Soeborg, Denmark, cat. No.: 1.09843.5000)., rehydrated in graded ethanol (99%, 96% and 70%), followed by antigen retrieval for 20 min. in Tris-EGTA buffer in a microwave (10 mM Tris, 0.5 mM EGTA, pH 9). Nonspecific binding was inhibited with Dako Real Antibody Diluent blocking buffer for 30 min at room temperature (Agilent Technologies, Glostrup, Denmark, Cat. No.: S202230-2). Sections were incubated with primary antibodies overnight at 4°C in a humid chamber. Applied antibodies include CYP17A1 (1:200, Santa Cruz, cat.no.: SC-46084) and HSD3B2 (1:200, Abcam, Cat. No.: 154385). Sections were washed in TBS before incubation with secondary antibodies (donkey-anti-rabbit Alexa FluorTM 647, donkey-anti-goat Alexa FluorTM 568; Invitrogen by Thermo Fisher, Cat. No. A31573, A11057, 1:300) at room temperature for 45 min. in dark. After that, the sections were washed in TBS and then incubated with DAPI (1:5000, Invitrogen, Cat. No.: D1306) for 3 minutes in the dark. Sections were washed 3x3 min. in TBS in the dark and transferred to dH₂O for 10 min in the dark before mounting. Antibody dilution buffers were used in place of primary antibodies as negative controls and showed no staining ( Supplementary Figure S2 ). Images were captured on a fully motorized Olympus BX63 upright microscope with an Olympus DP72 color, 12.8-megapixel, 4.140x3.096-resolution camera and edited with the ImageJ software.

For statistical analysis of gene expression differences, microarray data and RNA-seq data were analysed using the limma software package (version 3.16) (21). The expression profiles of samples within various experimental groups were subjected to analysis through the moderated t-test, following linear model fitting. Genes with an adjusted p-value (q-value) below 0.05 and an absolute logarithmic fold change (log FC) greater than one were considered differentially expressed. All box plots in this study were created using the ggplot2 package (version 3.4.0) (22).

We conducted an analysis of microarray gene expression data encompassing a total of 68 samples, categorizing them according to the respective follicular developmental stages. Specifically, we focused on genes associated with steroidogenesis, and the summarized findings are presented in Table 2 .

In early follicle development, we observed predominantly low expression levels of genes related to steroidogenesis. However, as the follicles progressed towards the preovulatory stage, there was a substantial increase in the expression levels of these genes, with notable exceptions being HSD17B1 and FSHR (see Figures 2A, B ). Notably, with the exception of HSD17B1, there were minimal significant differences in gene expression between mural granulosa cells and cumulus cells following ovulation induction, as evidenced by q-values exceeding 0.05 or log FC values less than 1 (see Figure 2A ).

The expression profile of steroidogenic acute regulatory protein (StAR) exhibited a gradual rise in expression with increasing follicle size, culminating around the time of ovulation (see Figures 2A, B ). This elevated expression was consistent in both mural granulosa cells and cumulus cells following ovulation induction, displaying statistical significance when compared to other time points (q-values > 0.05 or log FC < 1).

Furthermore, the expression pattern of cytochrome P450 family 11 subfamily A member 1 (CYP11A1) displayed an initial downregulation, followed by a sharp upregulation with a 48.5-fold increase in pre-ovulation induction follicles (q value < 0.001, log FC = 5.6). These heightened expression levels were maintained post-ovulation induction (see Figures 2A, B ).

CYP17A1 reached expression levels exceeding background levels exclusively in preovulatory follicles before ovulation induction, with all q values < 0.001 and log FC > 2. In contrast, in all other groups, the average CYP17A1 expression was down to background levels (see Figures 2A, B ).

The expression of cytochrome P450 family 19 subfamily A member 1 (CYP19A1) remained consistently low in both preantral and small antral follicles, with no statistically significant difference between them (q value > 0.05). However, a marked increase in CYP19A1 expression was observed in preovulatory follicles compared to all other developmental stages, evident by all q values < 0.001 and log FC > 1.5 (see Figures 2A, B ). Furthermore, after ovulation induction, the gene expression levels in both mural granulosa cells and cumulus cells surpassed those in the early stages of folliculogenesis, with significant statistical differences (all q values < 0.001 and log FC > 1) (see Figures 2A, B ).

HSD3B2 displayed low expression levels during the early stages of folliculogenesis but exhibited significant upregulation both pre-OI and post-OI (all q values < 0.001 and log FC > 2) (see Figures 2A, B ). There was no statistically significant difference in expression between the pre-OI and post-OI periods (q values > 0.05 or log FC < 1) (see Figure 2A ).

Hydroxysteroid 17-beta dehydrogenase 1 (HSD17B1) exhibited relatively high expression levels consistently throughout follicular development when compared to other genes ( Figure 2A ). The lowest average expression levels were observed in post-OI cumulus cells ( Figure 2B ), mirroring the levels seen in the preantral follicle group (q value = 0.008, log FC = 4.99) ( Figure 2A ). However, both of these were lower (all q values < 0.05 and absolute log FC > 1) than the levels observed in small antral follicles and preovulatory follicles, where no significant expression differences were evident ( Figure 2A ). Samples from the preantral follicle and mural granulosa cells (post-OI) group displayed substantial variability in HSD17B1 expression ( Figure 2B ).

Expression of AKR1C3 (HSD17B5) was close to or below the background level of noise in all follicle classes.

The expression patterns of FSHR and the LHCGR exhibited opposing trajectories. FSHR expression was significantly higher in preantral, small antral, and pre-OI follicles when compared to post-OI follicles, where expression was nearly absent (all q values < 0.001 and absolute log FC > 2) ( Figures 2A, B ). In contrast, LHCGR exhibited negligible expression during the early stages of folliculogenesis but peaked just before the induction of final follicular maturation (both q values < 0.001 and absolute log FC > 5) ( Figures 2A, B ). Subsequently, LHCGR expression decreased but remained at a high-level post-OI, albeit with some variability among samples ( Figures 2A, B ).

The expression of cytochrome b5 type A (CYB5A) was relatively highly expressed in preantral, small antral follicles and in follicles pre-OI. Notably, CYB5A expression significantly increased GCs as follicles advanced from the antral stage to the preovulatory stage. Furthermore, CYB5A expression in GCs collected at the time of oocyte aspiration displayed a significant increase compared to other follicle classes ( Figures 2A, B ).

The expression patterns of POR, HMGCR, and LDLR share a similar trajectory. These genes exhibited relatively low expression levels in preantral – and small antral follicles, but their expression significantly increased in GCs collected preovulatory follicles pre-OI. Subsequently, there was a slight reduction in their expression levels in mural and cumulus cells collected post-OI ( Figures 2A, B ).

Expression of DHCR24 share the pattern of POR, HMGCR, and LDLR in preantral and small antral follicles with significant increase in follicles pre-OI, to a level that remained high in GCs and cumulus cells at oocyte pickup ( Figures 2A, B ).

Data from a total of 71 samples were subjected to analysis, stratified into groups based on the stage of follicular development. The expression levels of genes associated with steroidogenesis are presented in Table 3 and Figure 3 . The expression patterns of the majority of steroidogenesis-related genes exhibited a progressive increase as follicles advanced in growth, ultimately peaking in preovulatory follicles just prior to ovulation induction (as illustrated in Figures 3A, B ). Nevertheless, a noteworthy subset of genes displayed variability in their relative expression levels across samples collected from various stages of follicle development (as indicated in Figure 3B ). It is important to note that in results presented here the groups of follicles are different from use used in microarray data. In both instances the background noise has been subtracted.

StAR expression was consistently observed in all five distinct follicular stages, with no statistically significant differences in expression levels between primordial, primary, and secondary follicles. However, a notable and significant increase in StAR expression was seen in preovulatory follicles when compared to the other stages (all q values < 0.001 and log FC > 3), as illustrated in Figure 3A . This trend aligns with the findings from our microarray data analysis of comparable follicle classes.

In contrast, the expression of CYP11A1 did not exhibit significant variation during the early follicular stages, with q values exceeding 0.05 and log FC remaining below 1. Notably, its expression was at its lowest point in secondary follicles, followed by a rapid and substantial increase in preovulatory follicles, where it reached its highest expression levels (all q values < 0.001, log FC > 3), as depicted in Figures 3A, B . This pattern resembles the observations made in our microarray study for corresponding follicle classes.

The expression of CYP19A1 did not significantly differ among the primordial, primary, secondary, and antral follicle stages, with q values exceeding 0.05 and log FC remaining below 1. Nevertheless, there was a slight increase in expression observed in preovulatory follicles, indicating a marginal advantage over the primordial, primary, and secondary stages (all q values < 0.05, log FC > 1), as shown in Figure 3A . This observation contrasts with our microarray data, which exhibited a more pronounced up-regulation of CYP19A1 from small antral to pre-ovulatory follicles.

The HSD3B2 expression exhibited a clear upward trend with follicle maturation, with significantly higher expression levels observed in antral and preovulatory follicles compared to the less mature stages (all q values < 0.001, log FC > 3, and q = 0.0003, log FC = -2.36, respectively), as illustrated in Figures 3A, B . This pattern closely mirrors the results obtained from our microarray data analysis.

Gene expression of HSD17B1 showed a trajectory like that of the microarray data with no significant difference detected between antral- and pre-OI follicles (q value = 0.76, log FC = - 0.35) ( Figures 3A, B ).

CYP17A1 and LHCGR remained silent throughout the entire follicular development while FSHR showed relatively high expression only in the antral follicle group (q values < 0.001, log FC > 2) ( Figure 3A ). The expression of LHCGR and FSHR differ from that of the microarray data.

Expression of DHCR24 was most notably significantly increased from the antral stage to the pre-OI stage of follicular development ( Figures 3A, B ).

The expression patterns of HMGCR and POR shared a similar trajectory but in contrast to the microarray data levels were similar between the antral and pre-OI stage, while that of LDLR became significantly upregulated in pre-OI follicles in contrast to microarray data ( Figures 3A, B ).

The expression of CYB5A increased significantly from antral to pre-OI follicles with expression being present in preantral follicles showing a significant drop in secondary follicles ( Figures 3A, B ). These data resemble those of the microarray data for comparable follicle sizes.

The dataset comprised 39 samples, which were categorized and analysed based on cell type and the type of ovulation induction trigger (GnRHa or hCG). The influence of the ovulation trigger type on the expression of steroidogenesis-related genes was limited, as shown in Figure 4 and Table 4 . The only statistically significant differences observed between the two protocols were in the upregulation of HSD3B2 and LHCGR genes in cumulus cells following GnRHa treatment (both q values < 0.001, absolute log fold change > 1) when compared to the hCG protocol.

Expression of HSD17B1 was exclusively detected in GCs from human small antral follicles ( Figures 5A, B ), CYP17A1 was exclusively detected in theca cells ( Figures 5C, D ). Expression of HSD3B2 was detected in the theca interna of follicles with a diameter of 5-6 mm, with weak staining observed in granulosa cells facing the lumen of the follicle ( Figures 5E, F ). In smaller follicles with a diameter of 1.5 mm, only theca externa and stroma cells expressed HSD3B2 ( Figure 5G ). The broader expression of HSD3B2 in theca and stroma cells surrounding the follicle was even more pronounced in small antral follicles with a diameter of 0.5 mm ( Figure 5H ).

The proteins HSD3B2 and CYP17A1 were detected exclusively in TCs in small antral follicles with a diameter of 5-6 mm ( Figures 6A, B ). DAPI was utilized to label DNA ( Figure 6C ) and was merged with images of HSD3B2 and CYP17A1 ( Figures 6A–C ), illustrating the co-localization of HSD3B2 and CYP17A1 to TCs ( Figure 6D ).