All mouse work was performed in accordance with the Animals (Scientific Procedures) Act 1986 under UK law and were bred on a C57Bl/6J or C57Bl6/JCrl background. The Srd5a1^-/- mouse line was obtained from the Jackson Laboratories (https://www.jax.org/strain/002793). The original paper describing the generation of mice with a null allele was published by Mahendroo, Cala, and Russel (20). Due to parturition defects in homozygous mice heterozygote crosses were used to maintain the colony and produce Srd5a1^-/- progeny. Wildtype (WT) mice were either littermates or purchased from Charles River Laboratories (Tranent, Scotland). Expression of 5α-Reductase isozymes (types 1-3) was assessed in uterine tissues by qPCR ( Supplementary Figures 1 ). Srd5a1 was absent in uterine tissues from knockout mice, and consistent with previous reports (20), Srd5a2 was not detected in either wildtype or knockout mice. While Srd5a3 was detected its expression was decreased with decidualization ( Supplementary Figures 1 ).

Mice were ovariectomized to remove endogenous ovarian steroids at day 0 (d0), then administered estradiol (E2, Sigma-Aldrich, Poole) by subcutaneous injection daily on d7-9 (5ug.kg^-1 in 200ul sesame oil) and on d13-15 (0.25ug.kg^-1 in 200ul sesame oil) with a P4 pellet inserted on d13. On day 15 a decidualization stimulus was administered by intrauterine injection of 20ul sesame oil into the uterus via transvaginal delivery using a non-surgical embryo transfer device (NSET). Animals were sacrificed 4 days later and uterine tissues collected ( Figure 1A , (23)). The decidualization response was immediately scored in each horn as outlined in Figure 1B . Specifically a non-decidualized horn had no decidualization reaction whatsoever (Non), whereas a fully decidualized horn had continuous decidualization along the majority (>50%) of its length (Full). Partial decidualization is any level of decidualization between these two extremes ( Figure 1B ). Collectively, samples designated Full or Partial are referred to as decidualized (Dec).

Finasteride (Sigma-Aldrich, Poole) was dissolved in 100% EtOH, then diluted to 5% v/v in sesame oil to a final concentration of 12.5mg.mL^-1 and 100uL (50 mg.kg^-1) was administered via intraperitoneal injection daily from the point of the decidualization stimulus onwards, i.e. d15-18, as show in Figure S1A . Finasteride is a synthetic 4-azasteroid that competitively inhibits both type I and II 5α-reductase enzymes in rodents (24, 25).

DHT (Sigma-Aldrich, Poole) was dissolved in 100% ethanol (EtOH) then diluted to 5% v/v in 0.4% w/v methylcellulose solution for a final DHT concentration of 2mg.mL^-1. 100uL (8mg.kg^-1) was injected subcutaneously on d15 at the point of decidualization stimulus. We have previously shown that this dose is sufficient to elicit changes in endometrial tissues (10, 26). At the end of experiments, mice were terminated by exposure to rising concentrations of CO₂ combined with cervical dislocation.

Uterine tissue for histology was fixed in 4% paraformaldehyde then dehydrated and embedded in paraffin blocks and sectioned according to standard protocols. Protocols for 3,3’-diaminobenzidine (DAB) and immunofluorescence staining were performed as previously reported (10, 27). Briefly, slides were dewaxed, rehydrated, and subjected to antigen retrieval in pH 6.0 citrate buffer using a microwave. Following a wash in phosphate buffered saline (PBS) slides were incubated with 3% H₂O₂ solution in PBS for 15min to remove endogenous peroxidase activity. Following several PBS washes, slides were incubated with normal goat serum solution (NGS, PBS containing 20% goat serum and 0.05% (w/v) bovine serum albumin) to block non-specific interactions. Anti-CD31 primary antibody (Abcam, ab28364, 1:500 dilution) in NGS solution was added and incubated overnight at 4°C. Following several washes, slides were incubated with goat-raised peroxidase-conjugated secondary antibody (Abcam, ab7171, 1:500 dilution) for 1hr at room temperature. Following several washes, slides were incubated with either DAB reagent or Opal red 570 reagent for 10 minutes or until sufficiently developed. After a brief wash, slides were counterstained either with hematoxylin and eosin as per standard procedures, or with DAPI before being mounted with coverslips.

Slides were imaged using a Zeiss Z1 microscope, or a Zeiss AxioScan Slide Scanner using conventional fluorescent setups.

To quantify vessels, regions of non-decidualized endometrium (Non) or decidua (Dec samples) were selected digitally on 20X magnification slide scans of CD31-labelled uterus sections and overlaid with a grid with edge length 250 pixels (~ 28μm). The intersections on the grid were sequentially classified as either vessel or not vessel, and the total proportion of vessel to non-vessel intersections was used as a measure of vessel count. Additionally, each time a vessel was identified, it was given an irregularity score equaling the number of intersections on a tree used to trace out the vessel’s cross-sectional shape. For example, a circular or ‘L’ shaped cross-section would have a score of 0, a ‘T’ shape a score of 1, an ‘H’ shape a score of 2, etc. Irregularity scores were averaged across all observations.

Uterine samples for RNA analysis were collected in RNALater solution (Qiagen, Hilden, Germany) and stored at -80°C until tissue extraction. For decidualized samples, only regions of uterus containing decidualized endometrium were selected. Samples were thawed and homogenized in 1 mL Trizol reagent using a TissueLyser machine (6min, 25Hz). Homogenate was transferred to MaXtract High Density tubes along with 200ul chloroform and 100ul RNase free water, shaken vigorously and centrifuged. The aqueous layer was added to RNeasy spin columns (Qiagen, Hilden, Germany) and processed as per manufacturer’s instructions.

cDNA was synthesized from 100ng.uL^-1 RNA using Superscript VILO cDNA synthesis kit (Thermo Fisher Scientific Life Sciences, Schwerte, Germany) as per the manufacturer’s instructions. Thermal cycler settings were 25°C for 10min, 42°C for 60min, 85°C for 5min. Primers were designed using Universal Probe Library Assay Design Center (Roche Applied Science, Burgess Hill, UK) or, when this was discontinued, using Neoformit (https://primers.neoformit.com/Accessed: 6/4/22; sequences in Supplementary Data 1 ). Primers were synthesized by Eurofins MWG Operon (Ebersberg, Gemany). Reactions were run in duplicate on a Quantstudio 5 384-well PCR machine (Thermo Fisher Scientific Life Sciences, Schwerte, Germany) using the following settings: 95°C for 10min then 40 cycles of 95°C for 15s and 60°C for 1min. A serial dilution of a standard cDNA solution (consisting of a mix of all samples tested) was used to plot a standard curve and expression levels interpolated from this. Data was processed using QuantStudio Design and Analysis Software.

Samples were analyzed on the NanoString nCounter Analysis System using a NanoString Mouse Pancancer Pathways Panel kit (Nanostring, Edinburgh, UK). RNA at 20ng.uL^-1 was used. Processing was performed by the Host and Tumour Profiling Unit Microarray Services, Institute of Genetics and Cancer, University of Edinburgh, as per the manufacturer’s instructions. Briefly, the reporter codeset was mixed with hybridization buffer and added to each of the RNA samples, followed by the capture codeset. This mix was hybridized at 65°C for 18hr. Following hybridization samples were loaded into the provided cartridges on the nCounter prep station and processed using the High Sensitivity protocol. Cartridges were then sealed and read using the digital analyzer on the max setting. No QC flags were registered for any of the samples.

Data was processed using the nSolver analysis software. Data was normalized using a combination of positive controls and housekeeping genes. Firstly, background was subtracted using a background value of the mean plus two standard deviations of the negative control values. Normalization using positive controls and housekeeper genes was performed using the geometric mean to compute the normalization factor, and the manufacturer-provided list of housekeeper genes was used excluding the following genes based on the results of the geNorm algorithm in the Nanostring Advanced Analysis modules: Hprt, Alas1, G6pdx, Gusb, Ppia.

Normalized counts data was then analyzed for differential gene expression using R, specifically the edgeR analysis workflow. Dispersions were estimated with the estimateDisp() command and fit to the design model using glmQLFit(), before testing for differential expression using glmQLFTest(). Graphs were produced using the ggplot2 package. Finally, to perform gene ontogeny (GO) analysis we utilized the clusterProfiler analysis workflow. Ensembl IDs were mapped from the Bioconductor org.Mm.eg.db genome annotation and enrichment assessed using the enrichGO() command with the ‘universe’ set as the genes included in the Nanostring Pancancer Pathways panel. Graphs were produced using the dotplot () and cnetplot () commands from the enrichplot R package.

Normalized counts and differential gene expression data was further analyzed in the context of publicly available single cell RNA sequencing (scRNAseq) data, accessed from NCBI’s Gene Expression Omnibus (28) through GEO Series accession numbers GSE160772 and GSE198556. These scRNAseq data defined the transcriptomic profile of over 20,000 perivascular, fibroblast and epithelial cell types present in the mouse uterus. The AverageExpression() command (Seurat) was used to calculate the mean expression of Nanostring DEgenes in perivascular, fibroblast and epithelial cell clusters from scRNAseq data. Gene expression matrices were normalized and clustered heatmaps produced using the pheatmap package. Expression values shown are scaled but not centered. This comparison allowed us to attribute the expression of genes that were found to be differentially expressed between Srd5a1^-/- and WT genotypes to certain cell fractions in the mouse endometrium.

Resin casts were prepared using a method previously validated in male mice (29). Briefly, female mice (Non and Dec) were culled using a terminal dose of sodium pentobarbital (150 mg/kg, ip) and perfusion fixation of the vasculature was achieved via the left ventricle. Heparinized PBS (heparin, 20 U/mL) was infused at 6 mL/min for 2 minutes. Low-viscosity resin (10 mL; Microfil MV-122; Flow Tech Inc) was prepared according to the manufacturer’s instructions and then infused via the left ventricle. Uterine tissues were recovered, trimmed, fixed in paraformaldehyde and embedded in 1.5% low melting point agarose (Invitrogen, UK), dehydrated in methanol (100%; 24 hours) and optically cleared in benzyl alcohol:benzyl benzoate (1∶2 v/v; 24 hours). Imaging was carried out as described in Rebourcet et al. with a positive control tissue being a previously prepared mouse testis (29).

Evans Blue (EB) dye binds to albumin in the blood so its presence in tissues, which can be quantified spectrophotometrically, implies a breakdown in the endothelial barrier. The Evans Blue dye (Sigma-Aldrich, Poole) was dissolved in sterile saline solution to a concentration of 0.5% w/v, and 200uL was injected i.v. (via tail vein) 30-90 minutes prior to cull via cervical dislocation. The interval between EB injection and cull was recorded and found not to associate with dye retention in tissues.

To quantify the amount of Evans blue in the tissues, following sacrifice a portion of the uterus was taken, precisely weighed, and transferred to 500μl of formamide. The gross appearance of the tissue was recorded using a digital camera. For decidualized samples, only sections of uterus containing decidualized endometrium were selected. Samples were incubated for 24h at 55°C to extract the dye, then tissue pieces were discarded, the solution centrifuged to remove debris, and absorbance read in triplicate at 610nm.

Unless otherwise stated, data was processed using Microsoft Excel and Python. Statistics were performed using either Python or Graphpad Prism 5.

In the analysis of Nanostring data, two-way ANOVAs were performed on all genes found to be expressed and ANOVA results corrected for multiple comparisons using the Benjamini and Hochberg method (30). The false discovery rate (FDR) was 0.05.

In the analysis of Evans Blue absorbance data, absorbance readings are shown as a ratio relative to the mean value of WT or vehicle-treated WT data. In the case of data from Figure 2 , ratios have been calculated separately for partially and fully decidualized horns and the results combined for the subsequent statistical analysis.

Statistical tests performed for other comparisons are stated in figure legends and the text.

To investigate the functional requirement for 5α-reductase during decidualization we induced decidualization in WT and Srd5a1^-/- mice using a previously validated protocol ( Figure 1A ; (23)). We used a ternary system to classify the extent of decidualization. Uterine horns were classified as non-, partially- or fully-decidualized (Non, Partial, Full, respectively; Figure 1B ). For the purposes of analysis individual uterine horns were treated as separate biological entities. In instances where only the decidualized regions of the uterus are analyzed Partial and Full horns were grouped together: referred to as ‘decidualized’ (Dec). Decidualization was associated with a significant increase in Srd5a1 mRNA expression in wildtype mice which was absent in knockout mice ( Supplementary Figure 1 ). Decidualization was observed in both genotypes, however Srd5a1^-/- uteri had a noticeably impaired decidualization response compared with those of WT. Specifically uteri from Srd5a1^-/- mice had fewer ‘Full’ horns and decidualized areas in ‘Partial’ horns were smaller ( Figure 1C ). When quantified and compared statistically we were able to reject the hypothesis that the number in each of the decidualization response groups is the same between genotypes, which supports our interpretation that decidualization is impaired in Srd5a1^-/- mice ( Figure 1D ).

In line with expectations, uterine weight increased significantly in Dec horns compared to non-decidualized horns in both WT and Srd5a1^-/- mice ( Figure 1E ). Uterine horn weights of partially decidualized horns from WT mice were significantly heavier than those from Srd5a1^-/- females ( Figure 1F ). The expression of mRNAs encoded by Prl8a2, a marker of decidualization, was significantly increased in Dec compared to Non horns in WT mice. A similar, non-significant, trend for increased Prl8a2 mRNA expression was observed in Srd5a1^-/- mice ( Figure 1G ).

To complement these findings we tested the impact of transiently blocking 5α-reductase activity in WT mice using the pharmacologic 5α-reductase inhibitor finasteride (for 4 days from days 15-18; Supplementary Figure 2A ). Notably this treatment regime did not significantly affect the decidualization response, uterine horn weights, nor expression of Prl8a2 ( Supplementary Figures 2B, C, D ). Thus transient inhibition of 5α-reductase under this paradigm was insufficient to replicate the effects of the global Srd5a1 knock-out on decidualization responses.

To investigate if impaired decidualization in Srd5a1^-/- mice was also associated with deficient vascular morphology and function we stained blood vessels with the endothelial cell marker CD31. Observational analysis of tissue sections suggested blood vessels in mutant mice appeared less numerous and more dilated compared to WT ( Figure 2A ). To quantify the vessels, counts were performed on samples fluorescently labelled for CD31 ( Supplementary Figures 3A ). This analysis revealed a significant increase in vessel frequency in decidualized compared to non-decidualized uterine horns ( Figure 2B ). The mean vessel frequency in Srd5a1^-/- decidualized uteri was lower than in WT but no statistically significant difference was detected ( Figure 2B ). Vessel irregularity was also quantified, revealing a significant increase with decidualization, but no distinct effect of genotype ( Supplementary Figures 3B ).

We attempted to map the three-dimensional architecture of vessels using the resin cast method, however accurate casts could not be obtained which we believe is due to the high permeability of vessels in the decidualized tissues ( Supplementary Figures 4 ). This finding and observations from histological slides of dilated vessels in the decidual tissue of Srd5a1^-/- mice led us to consider whether vessel permeability might be altered as a consequence of ablation of the gene in mice.

To quantify whether these morphological changes lead to functional deficiency in endothelial barrier integrity, we investigated vascular permeability using Evans Blue assay. We observed a significant increase in permeability of decidualized uterine horns from Srd5a1^-/- compared to WT mice ( Figure 2C ). In mice treated with finasteride, similar trends were observed, and there was a significant increase in permeability in decidualized horns following finasteride treatment ( Figure 2D ).

To investigate the potential molecular basis for the vascular phenotype observed in Srd5a1^-/- uteri, we quantified the expression of key angiogenic genes in decidualized horns from WT and Srd5a1^-/- mice. A significant decrease in the expression of Pecam1 and Flt1 was detected in Srd5a1^-/- uteri compared to WT ( Figure 2E ). The mean expression of Tek (TIE2/Angiopoietin-1 receptor) and Kdr (VEGFR2) was also decreased but this was not statistically significant ( Figure 2E ). When gene expression was analyzed in WT mice treated with finasteride there was no detectable change in the expression of the same panel of angiogenic genes ( Figure 2F ).

To discover genes and signaling pathways that are altered by deletion of Srd5a1 in decidualized endometrium we analyzed the transcriptome of Non and Dec uterine tissue from WT and Srd5a1^-/- mice using the Nanostring Pancancer Pathways panel. To visualize the general patterns of gene expression, count values were subjected to unsupervised hierarchical clustering and visualized on a heat map ( Figure 3A ). Clustering of samples forms two major arms, one containing exclusively Dec samples and the other mostly Non. These findings are consistent with decidualization being a major driver of changes in gene expression in these samples. Notably, within the Dec arm Srd5a1^-/- samples largely clustered together and closer to the Non arm, and several decidualized Srd5a1^-/- samples are found within the Non arm, suggesting a transcriptional phenotype in Srd5a1^-/- Dec uterus which was not the same as in the WT Dec tissue.

To interrogate the differential expression of genes between the two genotypes, we performed four ratio comparisons: Srd5a1^-/- Non vs. WT Non; Srd5a1^-/- Dec vs. WT Dec; WT Dec vs. WT Non; and Srd5a1^-/- Dec vs. Srd5a1^-/- Non ( Figure 3B ). This analysis revealed no significantly differentially expressed (DE) genes between genotypes in non-decidualized tissue (Srd5a1^-/- Non vs. WT Non), suggesting a minimal role for 5α-reductase in regulating uterine gene expression in the absence of decidualization. In stark contrast, decidualization in WT mice induced significant changes in the expression of many genes: 507 or 66% of the total analyzed (WT Dec vs. WT Non; Table 1 ). Interestingly, the majority (71%) of the DE genes in this panel were down-regulated by decidualization. In comparison with the WT data, gene expression changes in Srd5a1^-/- decidualized tissues (Srd5a1^-/- Dec vs. Srd5a1^-/- Non) were greatly suppressed, with fewer DE genes (299, 39%) and smaller changes in expression level for both up- and down-regulated genes ( Table 1 ). When the expression of genes in decidualized uteri of WT and Srd5a1^-/- mice was directly compared (Srd5a1^-/- Dec vs. WT Dec), many significantly DE genes are observed (254, 33%), most of which are up-regulated in Srd5a1^-/- compared to WT (80%). Notably when significantly DE genes were considered 95% of the up-regulated genes in this comparison are downregulated in the WT Dec vs. WT Non comparison.

To identify genes differentially regulated by the Srd5a1^-/- genotype, we performed two-way ANOVA on Nanostring data using the count number of all genes and the variables decidualization (Non/Dec) and genotype (WT/Srd5a1^-/-). Results are shown in Table 1 . We identified 47 genes significantly regulated by either genotype or the interaction of genotype and decidualization, and gene lists can be found in Supplementary Data 2 . We further validated key differentially expressed genes from this set by qPCR in a cohort containing additional samples and these tissues showed gene expression patterns that matched those seen in the Nanostring counts or followed similar trends ( Figure 3C ).

To gain insight into which gene expression pathways are driving decidualization, and how they are affected by deletion of Srd5a1, we performed gene ontology (GO) analysis based on DE genes identified from our Nanostring data analysis. Upregulated and downregulated genes from WT Dec vs. WT non and Srd5a1^-/- Dec vs. WT Dec comparisons were analyzed separately.

No significant GO terms were identified based on genes upregulated by decidualization (WT Dec vs. WT Non), however in the downregulated genes there were several terms related to transmembrane transporter activity ( Supplementary Figures 5A ). Category netplot (cnet) analysis of the enriched genes showed that all the GO terms were related to the same set of genes ( Supplementary Figures 5B ). Many of these genes were subunits of the L-type voltage-dependent Ca²⁺ channel (VDCC) which has been shown to be downregulated with decidualization in endometrial stromal cells (31). In agreement with these findings, many significant GO terms were identified for genes upregulated in Srd5a1^-/- decidua compared to WT (Srd5a1^-/- Dec Vs. WT Dec) all of which related to transmembrane ion transport ( Supplementary Figures 5C, D ).

Significant GO terms were also identified for genes downregulated in Srd5a1^-/- decidualized uterus compared to WT (Srd5a1^-/- Dec Vs. WT Dec; Figure 4A ). The majority related to cell migration, however there was also a large proportion of GO terms related to the regulation of the vasculature, endothelial cells, or angiogenesis. Cnet analysis highlighted multiple highly connected genes from canonical angiogenesis pathways such as the VEGF pathway ( Figure 4B ; Flt1 (VEGFR1), Kdr (VEGFR2), Vegfa, Pgf). This Cnet also shows that many of the genes related to migration are also linked to angiogenesis pathways. This analysis highlights that gene regulatory pathways that promote angiogenesis are significantly downregulated in decidualized uteri from Srd5a1^-/- mice.

To identify which cells in the uterus were likely to have altered gene expression as a result of deletion of Srd5a1, we compared the results of our Nanostring analysis with our previously published single cell RNA sequencing (scRNAseq) data from cycling mouse uterus which included mesenchymal cells (i.e. fibroblasts and perivascular cells), which are AR⁺ (10, 32), and epithelial cells (27). Expression of significantly DE genes from the Srd5a1^-/- Dec vs. WT Dec ratio analysis were sorted by hierarchical clustering ( Figure 4C ). All genes had detectable expression in at least one cell population and AR was expressed in stromal (fibroblast>perivascular), but not epithelial, populations. Together, stromal fibroblasts and perivascular cells have detectable expression of the majority of the angiogenic and migratory genes whose expression is regulated by 5α-reductase deficiency identified in the cnet plot ( Figure 4B ). Perivascular cells play a key role in supporting the growth, stability, and integrity of blood vessels, as well as playing a major role in angiogenesis. This analysis highlights AR⁺ perivascular cells and fibroblasts as plausible cell types mediating the angiogenesis related effects of SRD5A1 activity.

Given that AR+ cell types were associated with differential gene expression in Srd5a1^-/- mice, we postulated that a lack of intracrine androgen may have altered cell function during decidualization in the presence of 5α-reductase deficiency. To test whether impaired decidualization in Srd5a1^-/- mice is due to a reduction in intracrine/paracrine DHT, we performed a rescue experiment by administering exogenous DHT coincident with the decidualization stimulus ( Figure 5A ). Following this treatment, the rate of decidualization in DHT-treated mutants (Srd5a1^-/–DHT) significantly increased compared to vehicle-treated mutants (Srd5a1^-/–VC) and was statistically indistinguishable from vehicle-treated WT animals (WT-VC; Figure 5B ). Furthermore, the weight of decidualized uterine horns was also the same between WT-VC and Srd5a1^-/–DHT groups ( Figure 5C ).

We also tested whether the vessel permeability phenotype seen in Srd5a1^-/- animals could be rescued by exogenous DHT and found that Evans Blue dye detection in extracts from decidualized Srd5a1^-/–DHT was equivalent to that from WT-VC horns, demonstrating that the vascular permeability phenotype is also rescued by DHT administration (Figure 5D).

Finally, we tested whether altered gene expression identified in the Srd5a1^-/- uterus was affected by administration of DHT. Notably, DHT restored mRNA expression levels to WT-VC levels (Brca1, Fgfr1, Mapk8ip1, Runx1, Vegfa; Figure 5E ), and others showed the same trend (Flt1, Pecam, Il11, Rasgrp1; Figure 5E , Supplementary Figure 6A ). For some mRNAs there was no evidence of a rescue (Itga3, Pik3cb, Pik3r2; Supplementary Figure 6B ). The most significant change in gene expression in response to DHT supplementation occurred in genes related to angiogenesis signaling and downstream signal transduction pathways such as Vegfa, Fgfr1, and Mapk8ip1, suggesting that these stromal/vascular associated pathways are androgen-regulated during decidualization.

Overall this shows that gross tissue changes and many of the transcriptomic changes associated with impaired decidualization in Srd5a1^-/- mice can be restored by administration of DHT.