The following EC model cell lines were used in this study: Ishikawa (RRID: CVCL_2529; #ECACC 99040201, Sigma Aldrich GmbH), HEC1A (RRID: CVCL_0293; #ATCC_HTB-112TM, American Type Culture Collection), RL95-2 (RRID: CVCL_0505; #CRL-1671, American Type Culture Collection), and KLE (RRID: CVCL_1329; #CRL-162, American Type Culture Collection). Ishikawa cells were derived from well differentiated, primary endometrial adenocarcinoma of a 39-year-old patient (10), HEC1A cells from moderately-well differentiated, stage IA, grade II endometrial adenocarcinoma of a 71-year-old patient (11), RL95-2 cells from moderately differentiated, grade II primary endometrial adenosquamous adenocarcinoma of a 65-year-old patient (12), and KLE cells from poorly differentiated, grade III endometrial adenocarcinoma of a 68-year-old patient (13). As a control cell line, HIEEC cells (gift from Dr. Fortier, Laval University, Canada) were used, originally established from a non-neoplastic endometrial biopsy of a 37-year-old woman (14).

The cell lines were cultured in appropriate culture media in a 37°C incubator with 5% CO₂: HIEEC: RPMI-1640 medium (#R5886, Sigma Aldrich GmbH) supplemented with 2 mM L-glutamine (#G7513, Sigma Aldrich GmbH) and 10% fetal bovine serum (FBS, #F7524, Sigma Aldrich GmbH); Ishikawa: Eagle’s Minimum Essential Medium (#M5650, Gibco) with 5% FBS; HEC1A: McCoy’s 5A medium (#M4892, Sigma Aldrich GmbH); RL95-2: DMEM/F12 (#D6421, Sigma Aldrich GmbH) with 10% FBS, 2.5 mM L-glutamine, and 5 µg/mL insulin (#I9278, Sigma Aldrich GmbH); KLE: DMEM/F12 supplemented with 10% FBS and 2.5 mM L-glutamine, following good laboratory practices (15). All cell lines were authenticated by short tandem repeat profiling by the ATCC and routinely tested and confirmed negative for Mycoplasma contamination using the Lonza Mycoalert Mycoplasma Detection Kit. Three independent experiments were performed in cell passages less than 15.

RNA extraction from EC cell lines (from three independent experiments) was carried out using a Macherey-Nagel kit (#740933.5, Macherey-Nagel GmbH&Co, Germany), reverse-transcription with SuperScript^®VILO™ cDNA Synthesis kit (#11754050, Invitrogen, USA). Quantitative gene expression analysis was performed using Taqman chemistry (PAPSS2 (#Hs00989928_m1), CYP11B1 (#Hs01596406_m1), HSD11B2 (#Hs00388669_m1), HSD11B1 (#Hs01547870_m1), H6PD (#Hs00188728_m1), all from Thermo Fisher Scientific, USA), and SYBR Green chemistry (AR-A (forward primer: CCAGGGAAACGAATGCAGAG; reverse primer: AGTCTCCAAACTGTGAAGCC; Sigma Aldrich), AR-B (forward primer: TCATCACAGCCTGTTGAACT; reverse primer: ACTGCACTTCCATCCTTGAG; Sigma Aldrich), GPRC6A (forward: CCGGGACATATCATAATTGGAGG; reverse: CATTGCCACTGTGACTTCTGT, Integrated DNA Technologies, USA). Expression data for other genes relevant to this study were extracted from our previously published data (16–18). Gene expression was compared using one-way ANOVA with Tukey’s HSD post hoc test.

HIEEC, Ishikawa, HEC1A, RL95-2, and KLE cells were seeded in 6-well plates in complete culture media. After 24h, the cells were washed with DPBS (#D8537, Sigma Aldrich), and the medium was replaced with phenol red-free, FBS-free medium. Cells were incubated with 1.6 µM dehydroepiandrosterone sulphate (DHEAS, #D5297, Sigma Aldrich GmbH), 15 nM dehydroepiandrosterone (DHEA, #A8500-00, Sigma Aldrich GmbH), 3 nM androstenedione (A4, #A6030-000, Steraloids), 15 nM 11β-hydroxy-androstenedione (11βOHA4, #A-3009, Sigma Aldrich GmbH), or 3 nM 11-keto-androstenedione (11KA4, #284998, Sigma Aldrich GmbH), for 4, 8, 24, 48, and 72 h. All steroid precursors were prepared in ethanol (#1.11727, Supelco). After each time point, the culture media were collected and stored at -80°C until further analysis. Three independent experiments were performed, each in technical duplicates. One-way ANOVA with Tukey’s HSD post hoc test or Kruskal-Wallis with Dunn’s test with Bonferroni correction were used where appropriate.

Ishikawa and RL95-2 cells were cultured in complete media for 24 hours, followed by incubation with 10 nM DHT, 10 nM 11KDHT or ethanol (as control) for 48 hours. RNA extraction was carried out using a Macherey Nagel kit, following the manufacturer’s instructions. mRNA sequencing was performed on an Illumina platform at Novogene Inc. Non-directional poly-A library preparation was used. Quality control of raw reads and read mapping to the reference genome were conducted using fastp and Hisat2 v2.0.5 software, respectively. Gene counts were obtained using featureCounts v1.5.0-p3. Differential gene expression analysis was performed using the DeSeq2 package in R studio (19). Differentially expressed genes were identified based on a fold-change threshold greater than 1.5 (absolute value) and an adjusted p-value (with Benjamini-Hochberg (BH) method) of less than 0.01. Three independent experiments were performed.

The open-access TCGA database of primary endometrial tumors from Kandoth et al. (5), was accessed through the University of California San Francisco Xena browser. Differential gene expression analysis of raw counts of protein-coding genes was performed using the DeSeq2 package in R studio (19). Differentially expressed genes were identified based on a fold-change threshold greater than 2 (absolute value) and an adjusted p-value (BH method) of less than 0.01. Pathway activity scores were inferred on log transformed, Fragments Per Kilobase Milion FPKM-upper quartile normalized (FPKM-uq) values using single-sample gene set enrichment analysis (ssGSEA) implemented in the GSVA R package (20, 21), and hallmark gene sets from the Molecular Signatures Database (MSigDB) (22). Differences in pathway activity scores between groups were analyzed using the Limma R package by moderated t-test with BH correction for multiple testing (23); adjusted p values less than 0.01 were considered significant.

Optimal cutoff points of RNA expression levels in relation to survival were determined with the maxstat package in R studio (24). Survival plots were generated using the Kaplan-Meier method. Uni-and multivariate Cox proportional hazards models were fitted to estimate hazard ratios. P-values were two-sided, confidence intervals were calculated at the 95% level, and significance was defined as <0.05.

Single-cell RNA-seq data from (25) were downloaded from GEO (GSE173682) and involved five endometrioid tumors. The analysis was performed using the Seurat R package (26), and involved filtering of cells with unique feature counts >2,500 and <200, and cells with >25% mitochondrial counts. The filtered count matrices were then normalized and scaled. The top 2,000 most variable genes were summarized by PCA into 50 principal components (PCs). To identify cell clusters, graph-based Louvain clustering was performed with all 50 PCs, and Seurat’s FindClusters function with a resolution of 0.7.

Statistical analysis and visualization were performed using R studio version 4.3.0 or higher. The statistical methods are described in the methods section and figure legends. All p values were two-sided.

We investigated whether classic androgen precursors, including DHEAS, DHEA and A4 can be metabolized to 11-oxyandrogens intra-tumorally using a panel of EC model cell lines representing low-grade, POLE-alt EC - Ishikawa, low-grade, MSI-high EC – HEC1A and RL95-2, high-grade, TP53-alt EC - KLE, and a control cell line HIEEC. For this purpose, we first examined the expression of enzymes involved in the conversion of classic androgen precursors to bioactive classic and 11-oxyandrogens ( Figures 1B, C , Supplementary Figure 2A ). Importantly, CYP11B1, coding for the enzyme that catalyzes 11β-hydroxylation of classic androgens (A4 and T) in the adrenal cortex was not expressed in any of the cell lines ( Supplementary Figure 2A ), indicating that intra-tumoral 11-oxyandrogen formation is not feasible.

Next, we incubated the cell lines with physiologically relevant concentrations of classic androgen precursors: 1.6 µM DHEAS, 15 nM DHEA, and 3 nM A4 over a 72-hour period ( Figure 1A , Supplementary Figure 1D ), followed by LC-MS/MS profiling of formed metabolites. Indeed, we confirmed the absence of 11-oxyandrogen formation from DHEAS, DHEA and A4 by LC-MS/MS. Moreover, we observed that EC cell lines had different potential to metabolize classic precursors to bioactive androgens, which was not related to tumor grade or molecular phenotype.

In terms of DHEAS metabolism, we observed RL95-2 cells to metabolize a higher percentage of this precursor to downstream metabolites compared to the rest of cell lines ( Figure 1D ). This could be explained by significantly higher STS expression in this cell line ( Figure 1B ). Consequently, the levels of DHEA, the first downstream metabolite of DHEAS, as well as those of A4, T and DHT were highest in RL95-2 compared to the control cell line, HIEEC, and the cancer cell lines, Ishikawa, HEC1A and KLE ( Figures 1E–H ). Here, it should be noted that the levels of DHEA, A4 and those of the bioactive androgens T and DHT that formed from 1.6 µM DHEAS in RL95-2 in 72 hours were relatively low (DHEA ≈ 150 nM, A4 <10 nM, T <1 nM, DHT <0.1 nM), and did not account for the whole DHEAS that was metabolized. This might be explained by the low levels of HSD3B1/2 ( Supplementary Figure 2A ) and high expression of AKR1C3, leading to DHEA being shunted towards 5-androstenediol (5-Adiol), which was not profiled in our assay. Of note, the catalytic efficiency was reported to be only two-fold lower for DHEA conversion to 5-Adiol (k_cat/Km: 12 ± 1.9 min^-1 µM^-1) as compared with A4 conversion to T (k_cat/Km: 23 ± 3.1 min^-1 µM^-1) (27).

Moreover, DHEA can be hydroxylated at C16 position by CYP3A4/7 (28) and also 16α-DHEA was not measured in our assay. In addition, conjugation of metabolites formed from DHEAS, primarily glucuronidation, mediated by UGT2B isoforms 7, 15 and 17 (29), can be also involved and account for the rest of DHEAS metabolites. However, the expression of CYP3A4, CYP3A7, UGT2B7, UGT2B15 and UGT2B17 in RL95-2, and the other EC cell lines included in our study, is very low, as seen in the Cancer Cell Line Encyclopedia (CCLE) data [(30), https://depmap.org/portal/]. Finally, the cancer cell lines Ishikawa, HEC1A and KLE and the control cell line, HIEEC, metabolized DHEAS to low amounts of DHEA whereas A4, T or DHT practically did not form ( Figures 1E–H ).

In terms of DHEA metabolism, we observed RL95-2 cells to metabolize a higher percentage of this precursor to downstream metabolites compared to other cell lines ( Supplementary Figure 1A ). In this cell line, less than 10% DHEA proceeded to A4, and subsequently to low levels of T ( Supplementary Figure 1B ). This can be explained by expression of HSD3B1/HSD3B2 and AKR1C3 ( Supplementary Figure 2A ). Apart from RL95-2, HEC1A cell line also expressed comparable levels of HSD3B2 to RL95-2, however, did not form A4 from DHEA. This could be due to higher expression of AKR1C3 in this cell line, which converts A4 to T and further conjugation with glucuronic acid or sulfate. Alternatively, it can involve DHEA conversion to 5-Adiol via AKR1C3 (27).

The formation of bioactive T from A4, on the other hand, was highest in low-grade, AKR1C3-high cell line, Ishikawa but not in low-grade, AKR1C3-high model, HEC1A ( Supplementary Figures 1E, F ). This could be explained by the high SRD5A1 levels in HEC1A, ( Supplementary Figure 2A ), which potentially shunts A4 to 5α-androstenedione (not profiled in our LC-MS/MS method) instead of T. The levels of DHT from DHEA and A4 were below detection in all cell lines. Altogether, the low-grade, MSI-high cell line, RL95-2 formed higher levels of bioactive androgens from classic androgen precursors compared to the rest ( Figure 1I ).

We next wondered whether endometrial tumors could metabolize 11-oxyandrogen precursors, including 11βOHA4 and 11KA4 to AR-activating 11-oxyandrogens, such as 11KT and 11KDHT. To explore this, we incubated the same panel of EC cell lines with physiologically relevant concentrations of 11βOHA4 (15 nM) and 11KA4 (3 nM), for a total of 72 h ( Figure 2A ).

The gene expression of key enzymes of 11-oxyandrogen metabolism differed significantly between cell lines. More specifically, HSD11B2, coding for the enzyme that catalyzes 11βOHA4 oxidation to 11KA4, was highest in low-grade Ishikawa and RL95-2 cells ( Figure 2B ) but not in low-grade HEC1A cells. In contrast, the control cell line HIEEC expressed highest levels of HSD11B1, which encodes the enzyme that catalyzes the reverse reaction (Figure 2C; Supplementary Figure 2A ).

In accordance with HSD11B2 expression levels RL95-2 formed highest levels of 11KA4 from 11βOHA4, followed by Ishikawa and KLE ( Figure 2E ), ( Figure 2B ). The levels of 11KT formed from 11KA4 via AKR1C3 were greatest in RL95-2 and Ishikawa, followed by HEC1A ( Figure 2F ). AKR1C3 levels were highest in Ishikawa and HECA1, followed by RL95-2 ( Supplementary Figure 2 ). Of note, the low levels of 11KA4 observed in HEC1A suggest fast conversion of this metabolite to 11KT, which could be explained by high AKR1C3 expression in this cell line ( Supplementary Figure 2A ). Low levels of 11βOHT were detected in HEC1A and RL95-2 ( Figure 2G ). Notably, the high-grade, TP53-alt cell line KLE metabolized 11βOHA4 only to 11KA4, whereas the control cell line HIEEC practically did not metabolize the precursor at all, due to very low HSD11B2 levels ( Figures 2D–G ).

We also incubated cells with 3 nM 11KA4, which has similar levels to A4 in the systemic circulation. AKR1C3-high cell lines Ishikawa and HEC1A formed highest levels of 11KT from 11KA4 ( Figures 2H, I ). RL95-2 formed low 11KT levels from 11KA4, probably due to due to high expression of HSD17B2, which catalyzes 11KT conversion back to 11KA4 (31) ( Supplementary Figure 2A ). Of note, 11KDHT levels were below the detection limit upon incubation with 15 nM 11βOHA4 or 3 nM 11KA4.

Altogether, cell lines of low-grade, non-TP53-alt EC formed higher 11KT levels from 11-oxyandrogen precursors than the high-grade, TP53-alt cell line, and the control cell line ( Figure 2J ). High HSD11B2 and AKR1C3 expression conferred high metabolizing potential of 11-oxyandrogen precursors. Notably, the amount of 11KT formed from 11-oxyandrogen precursors was several folds higher than T formed from classic androgen precursors. This highlights intra-tumoral 11-oxyandrogen metabolism as an important source of AR-activating hormones in endometrial tumors.

To investigate the transcriptomic effect of androgen and 11-oxyandrogen signaling on endometrial cancer cells, we performed mRNA sequencing upon 48h incubation with 10 nM DHT, 10 nM 11KDHT or ethanol as control of two model cell lines, namely Ishikawa, which expressed AR ( Supplementary Figure 2A ) and RL95-2, which was the most efficient in metabolizing androgen and 11-oxyandrogen precursors and expressed low AR levels ( Figures 1 , 2 ). In both cell lines, the transcriptomic effects of DHT and 11KDHT were mild, without significantly altered signaling pathways. The lists of differentially expressed genes are given in Supplementary Tables 2A–D . No significantly expressed genes were observed in RL95-2 cells upon incubation with either DHT or 11KDHT (Figures 3C, D)

In the AR-expressing, low-grade EC model, Ishikawa, both DHT and 11KDHT induced upregulation of MYO1D, coding for unconventional myosin ID protein ( Figures 3A, B ). Furthermore, incubation with 10 nM DHT in Ishikawa cells also caused upregulation of LAMC3, coding for laminin subunit gamma 3 ( Figure 3A ). Incubation with 10 nM 11KDHT, apart from MYO1D upregulation, also induced changes in MAGEA2 gene, coding for melanoma-associated antigen 2 in Ishikawa ( Figure 3B ).

We next assessed the expression of key enzymes of the 11-oxyandrogen metabolism in primary endometrial tumors and tumor-adjacent tissues from the TCGA UCEC cohort (5). A list of differentially expressed genes between tumor-adjacent endometrium and endometrial tumors of low-grade and high-grade EC can be found in Supplementary Tables 3A, B .

We found that endometrial tumors of both low- and high-grade have significantly higher HSD11B2 levels than tumor-adjacent endometrium ( Figure 4A ), which, like the control cell line HIEEC, displayed low HSD11B2/HSD11B1 ratio compared to EC of any grade ( Figure 4C ). Furthermore, HSD11B2 expression differed between grades and molecular subtypes, being most pronounced in low-grade endometrioid tumors ( Figure 4A ) and in MSI-high and NSMP molecular subtypes ( Figure 4B ). Importantly, HSD11B2 expression was also associated with better disease-specific survival (DSS) (Univariate Cox proportional hazard model adjusted for grade: Hazard Ratio [HR] 0.39, 95% CI, 0.12-0.85, p=0.02) ( Figure 4D ).

The downstream utilization of circulating or locally formed 11KA4 is also dependent on the expression of relevant enzymes, including AKR1C3, which catalyzes the bioactivation of (11-oxy)-A4 to 11KT, among others. AKR1C3 was slightly upregulated in endometrial tumors compared to tumor-adjacent tissue ( Supplementary Figures 3A, B ), suggesting greater 11KA4 metabolism potential in tumors than tumor-adjacent endometrium. Of note, not all endometrial tumors from the TCGA dataset had a matching tumor-adjacent tissue, therefore, the latter comparison may not be optimal. Finally, expression of SRD5A2, coding for a key enzyme that catalyzes the formation of the most potent androgen, DHT, and its 11-oxyhomologue, 11KDHT, was associated with lower pro-tumoral cellular pathway activity ( Supplementary Figures 3C, D ), and better DSS vs. patients with SRD5A2-low tumors (Univariate Cox proportional hazard model adjusted for grade: Hazard Ratio [HR] 0.34, 95% CI, 0.12-0.94, p=0.04) ( Figure 4E ). SRD5A1, on the other hand, remained unchanged in low-grade EC compared to tumor-adjacent endometrium. However, it was upregulated in high-grade tumors compared to tumor-adjacent endometrium; however, the change was below the set threshold of an absolute 2-fold change (Fold change: 1.6; adjusted p-value: 3.82 × 10^-8). Altogether, the data on EC tumor tissue expression levels as well as our in vitro data suggest that in situ 11-oxyandrogen metabolism is characteristic for tumors of lower grade and clinically more favorable molecular subtypes of EC.

Based on the mild transcriptomic effects that we observed upon incubation of Ishikawa and RL95-2, we suspected that (11-oxy)-androgen signaling might not primarily affect the epithelial cancer cell population. In continuation we examined AR expression and activity across EC grades and molecular subtypes using transcriptomics data from (5) and scRNA-seq data from Regner et al. (25).

AR expression was highest in tumor-adjacent endometrium and lowest in high-grade tumors, G3 EEC and USC ( Figure 5A ). In terms of molecular subtype, NSMP tumors displayed the highest AR expression compared to the rest ( Figure 5B ). In terms of cell populations, we found AR expression to be low in epithelial cells and immune cell populations, but prominent in stromal populations, which comprise a great portion of the tumor mass and might be the main target of bioactive (11-oxy)-androgens ( Figure 5C ).

Finally, we inferred the responsiveness of endometrial tumors to androgens using bulk transcriptomics data of the TCGA UCEC cohort. Unsurprisingly, low-grade tumors, which expressed highest AR levels were more responsive to androgens than high-grade tumors ( Figure 5D ). Moreover, within the high-grade subset, those with a TP53-alt molecular profile were less responsive to androgens compared to high-grade tumors with unaltered TP53 ( Figure 5D ). Finally, AR expression was associated with better DSS (Univariate Cox proportional hazard model adjusted for grade: HR 0.41, 95% CI, 0.18-0.95, p=0.04) ( Figure 5E ), whereas patients with AR-enriched, non-TP53-alt tumors had better disease-free interval (DFI) comparing to AR-low counterparts (Univariate Cox proportional hazard model adjusted for grade: HR 0.29, 95% CI, 0.09-0.95, p=0.04) ( Figure 5F ).