The Prestwick library composed of 1,280 compounds was purchased by the R.E.A.D.S. unit of the University of Geneva. The drugs were dissolved at a stock concentration of 10 mM in dimethyl sulfoxide (DMSO). P4 and PGE1 were purchased from Sigma-Aldrich (Buchs, Switzerland). The fluorescent Ca²⁺ indicator Fluo-4 AM was purchased from Invitrogen and dissolved in DMSO at a final concentration of 2.5 µM (California, United States of America). The nuclear dye Propidium Iodide (PI) was purchased from Thermo Fisher Scientific (MA, United States of America), and diluted in water at a stock concentration of 1 mg/mL. Pisum sativum agglutinin (PSA) staining dye, was purchased from Sigma-Aldrich and dissolved in dissolved in Phosphate Buffered Saline (PBS) at a stock concentration of 1 mg/mL. Hoechst-33342 was purchased from Sigma-Aldrich (Buchs, Switzerland), and dissolved in water at a stock concentration of 1.5 mg/mL (Buchs, Switzerland). Human serum albumin (HSA) was obtained from Polygon Diagnostics (Lucerne, Switzerland). MDL 12330A hydrochloride was purchased from Tocris Bioscience (Bristol, UK). The following steroids were purchased from Sigma-Aldrich (Buchs, Switzerland): P4, dehydroisoandosterone 3-acetate (DHEA), estropipate, stanozolol, epiandrosterone, deoxycorticosterone, dydrogesterone, 3-alpha-Hydroxy-5-beta-androstan-17-one, pregnenolone, medrysone, equilin, estrone, norgestimate, canrenoic acid potassium salt, exemestane, finasteride, lithocholic acid, formestane, exemestane, fluorometholone, estradiol-17 beta, androsterone, medrysone, nandrolone, fluorometholone, oxymetholone, urosiol, oxandrolone, canrenone, fulvestrant, lynestrenol, norethindrone, ethisterone, nomegestrol acetate, gestrinone, chlormadinone acetate, norgestrel-(−)-D, ethynodiol diacetate, mifepristone, budesonide, adrenosterone, ethinylestradiol, megestrol acetate, ethynylestradiol 3-methyl ether, 3α,21-dihydroxy-5α-pregnan-20-one, 5α-pregnan-3α-ol-20-one, 19-norandrosterone, androsterone, 5alpha-pregnan-20-one, 5alpha-androstan-3-one, 5alpha-pregnane-3,20-dione, androstenedione, tanaproget, BB_NC-00182 and dissolved at a stock concentration of 20 mM in DMSO. To implement the pharmacophore model, the steroids were purchased from ChemDIV (California, United States) under the following ID numbers: N017–0003, N039–0025, N050–0010, 3647–1627, N044–0005, 0449–0076, 2361–0084, N039–0043, N050–0013, N039–0021, 0449–0077, N017–0007, 8011–0433, N050–0008, 3529–0011, 3529–0046, 5263–0009, N050-0018, and 3β,21-dihydroxy-5α-pregnan-20-one, 21-hydroxypregnenolone and MolPort-001-728-767 were purchased from MolPort (Riga, Latvia), and promegestrone was purchased from BOC Sciences (NY, United States) and dissolved at a 20 mM stock concentration in DMSO.

The experimental design is summarized in Figure 1. Briefly, motile spermatozoa were selected by swim-up and labeled with the fluorescent calcium indicator Fluo-4 a.m. to image changes in [Ca²⁺]_i. After centrifugation to remove excess dye, Fluo-4 AM loaded cells were distributed in a 384-wells plate to obtain ∼125,000 cells/well. After measuring basal fluorescence, a first injection of compounds was performed in order to evaluate their effect on [Ca²⁺]_i alone. This first screen is therefore considered as a screen to search for ‘activators’. Fluorescence was recorded for 12 min during which sperm cells were continuously exposed to compounds. P4 or PGE1 were then added in the same wells to evaluate whether the tested compounds can reduce the ligand-induced activation of CatSper. This second screen was therefore considered to be a search for CatSper ‘inhibitors’.

Human semen samples were collected from healthy donors by masturbation after a recommended sexual abstinence period of minimum 2 days. Semen samples were allowed to liquefy at 37°C for 15–30 min. Motile spermatozoa were selected by a ‘swim-up’ procedure as previously described (Strünker et al., 2011). Briefly, 1 mL of semen was deposited in a 45°-inclined 50 mL Falcon tube with 4 mL of Human Tubular Fluid (HTF) medium containing: 93.8 mM NaCl, 4.7 mM KCl, 0.2 mM MgSO₄, 0.37 mM KH₂PO₄, 2.04 mM CaCl₂, 20.98 mM HEPES, 2.78 mM glucose, 21.4 mM Na-lactate, 4 mM NaHCO₃, 0.33 mM Na-pyruvate, and the pH was adjusted to 7.35. Motile sperm were allowed to swim at 37°C for 1 hour. The supernatant was collected in 15 mL Falcon tubes and washed twice by centrifugation at 700 x g for 20 min. Sperm were supplemented with 3 mg/mL HSA. Written informed consent was obtained from all donors prior to donation and approved by the Cantonal Ethics Committee of the State of Geneva (CCER #14-147).

Cells were loaded with the fluorescent Ca²⁺ indicator Fluo-4 AM at a final concentration of 2.5 µM for 30 min at 37°C. Excess dye was washed via centrifugation at 700 x g for 20 min and sperm were resuspended in HTF medium to obtain a final concentration of 6.25 x 10⁶ cells/mL. 20 μL of cells (corresponding to ∼125,000 cells) loaded with Fluo-4 a.m. were dispensed in a 384-wells plate (Corning, NY, United States) to allow fluorescence measurement. The dye was excited at 480 nm, and emission was recorded at 530 nm. The change in [Ca²⁺]_i was assessed simultaneously in all wells using a Functional Drug Screening System (FDSS/μCELL) (Hamamatsu Phototonics K.K., Yokohama, Japan) at 37°C. Basal fluorescence was first measured for 30 s. A volume of 5 µL of compounds was then added to obtain a final concentration of 10 µM and P4 or PGE1 were used as positive control at a final concentration of 2 µM corresponding to a value close to saturation (EC₈₀). P4 and PGE1 were tested at a near-saturating concentration in the HTS in order to evaluate a potential additive or synergistic effect induced by the compounds. Non-specific CatSper inhibitor MDL 12330A was used as a negative control at final concentration of 20 µM while DMSO was used as a vehicle control. For alkalization-induced Ca²⁺ measurement, NH₄Cl was used as a positive control at a final concentration of 10 mM. Fluorescence was recorded for a total of 16.5 min: 30 s before injection, 12 min after the first injection of the tested compounds, and 4 min after the second injection of either P4 or PGE1. The Prestwick library was tested in technical duplicate. Ca²⁺ response curves were normalized using the following formula F − F 0 F 0 × 100 , where F ₀ is the average of the basal fluorescence prior to the addition of the compound of interest and F corresponds to the fluorescence value at each time point. The ∆F/F₀ of the vehicle control (DMSO) was subtracted from all the other values to adjust for dilution and pipetting artifacts. The values of all compounds were normalized to their respective controls.

The evaluation of acrosome reaction in live human sperm cells was performed using a modified version of a previously described method (Zoppino et al., 2012). Briefly, swim-up recovered motile spermatozoa were allowed to capacitate for 4 h at 37°C in capacitating media (HTF+) containing: 93.8 mM NaCl, 4.7 mM KCl, 0.2 mM MgSO₄, 0.369 mM KH₂PO₄, 2.04 mM CaCl₂, 20.98 mM HEPES, 2.78 mM glucose, 21.4 mM Na-lactate, 25 mM NaHCO₃, 6.6 µM Na-pyruvate, and the pH was adjusted to 7.35. Capacitated sperm were then incubated for 30 min with 5 μg/mL Pisum sativum agglutinin conjugated with fluorescein isothiocyanate (PSA-FITC) and 10 μg/mL Hoechst-33342. Samples were centrifuged at 700 x g for 10 min at room temperature to remove excess dye and resuspended in HTF+. Sperm were then incubated for 45 min in the absence (DMSO) or presence of the steroids at 10 μM, or positive control P4 at 10 µM. The ionophore A23187 was also used as a positive control and incubated with sperm at 2 µM for 30 min only. PI (0.5 μg/mL) was then added to all samples which were analyzed using BDFACSAria. FITC-PSA fluorescence was detected by excitation at 475 nm and emission at 560/35 nm, PI was detected by excitation at 530 nm and emission at 675/75 nm and Hoechst was detected by excitation at 346 nm and emission at 460 nm. Hoechst-positive and PI-negative cells were classified as live cells. Doublet exclusion was performed by two-dimensional dot plot analysis of forward-scatter width (FSC-W) versus forward-scatter height (FSC-H), and side-scatter width (SSC-W) versus side-scatter height (SSC-H) on the pre-selected singlets. Data were collected from 40,000 PSA-FITC negative events per sample to define the sperm population. Based on the selected population, PI-negative and PSA-FITC-positive cells were selected as live acrosome reacted cells, whereas PI-negative and PSA-FITC-negative cells were selected as live acrosome intact cells.

The ability of sperm to penetrate in viscous media similar in viscosity to that found in the female reproductive tract was assessed using the modified Kremer test, as previously described (Rahban et al., 2021). Briefly, swim-up recovered and capacitated sperm cells at 3 x 10⁶/mL were incubated in the absence (DMSO) or presence of steroids (10 µM), or P4 (5 µM) as a positive control for 1 h at 37°C. A glass capillary (0.2 × 4.0 × 50 mm, CM scientific, UK) was filled with 1% (w/v) methylcellulose (MC, 4000 centipoises cP) prepared with HTF media supplemented with 3 mg/mL HSA, respective compounds (10 µM), P4 (5 µM), or DMSO. The glass capillary tube was then sealed on one end with wax (Vitrex, UK) and added to the sperm cells on the open end. Sperm penetration was assessed after 1 h of incubation at 37°C by counting sperm at 1 cm using a microscope with a ×10 objective.

Data are shown as mean ± standard deviation (SD) with “n” referring to the number of independent experiments performed using sperm samples from ≥3 different donors. Statistical analysis and fitting of dose-response relations were performed using GraphPad Prism 8.1.1 (Prism, La Jolla, United States). To generate sigmoidal curves, the concentrations were converted to their respective log values, and the data were normalized using the following formula: 100 ⋅ x F max , where F_max is the maximum value of the positive control. A four-parameter Hill equation with a variable Hill coefficient was fitted to generate the EC_02, EC_50, and IC₅₀. Statistical significance between control and treated conditions was evaluated using a one-way analysis of variance (ANOVA), followed by Dunnett´s test. A p-value <0.05 was considered significant.

In order to build a ligand-based model that could be used to screen in silico for additional steroid-based compounds with an agonist effect, we started by constructing an initial pharmacophore model and a semi-parametric zero-inflated beta regression model using the 90 steroids identified through the screening of the Prestwick library (see Supplementary Figure S1 for the details on the construction of the models). Given the known importance of different physicochemical properties of compounds for their binding to charged porins in bacteria, including their polarity (Acosta-Gutierrez et al., 2018), we developed a physicochemical property-based filtering criterion relying on a statistical regression model. This model was then used in combination with our pharmacophore model to increase the probability of the in silico screening. The regression model assumes a zero-inflated beta distribution to model the agonist effect, which is the dependent variable, while it considers the original physicochemical properties and their second and third-order interactions between them as independent variables. The QikProp tool of the Schrödinger suite (QikProp, Schrödinger, LLC, New York, NY, 2021) was used to calculate the physicochemical properties of each compound. Of the 10 physicochemical properties selected using the Adaptive Lasso procedure (Zou, 2006) and included in the model, seven were identified as significant when considering a confidence level of 1 − α = 0.95 , most of which were the result of interaction between variables. Out of these variables, only one was an original variable, while the rest of the significant variables were interactions between two or three initially measured properties. To select the most optimal pharmacophore and statistical model combination, we used a blind dataset of 18 compounds whose agonist effect was predicted by the pharmacophore and statistical model prior to their experimental testing. Among the different pharmacophore models considered, the one with the two hydrophobic sites and one acceptor site, combined with a beta zero-inflated model with adaptive-Lasso selected covariates was considered the most optimal (Supplementary Figure S1). In particular, the combination of the two models was able to identify five true positives and one false positive out of the six compounds. The results of the blind test demonstrated that the combination of the two models was crucial to achieving a good prediction ratio, as considering either model alone led to several false positives. Using this combination of models, we screened in silico the GeDB (n = 187), ChemDiv (n = 498), and ZINC (n = 98) libraries after querying only for compounds with a steroidal core. For all the compounds screened virtually, 30 additional steroids were tested: the top 30 with the highest pharmacophore fitting score and predicted agonist effect, and five that were expected to have no effect but were included for further validation of the model were purchased and tested experimentally.

To evaluate the potential effect of synthetic and physiological compounds including steroids on Ca²⁺ influx, we screened the Prestwick library which consists of 1,280 compounds from various therapeutic classes. Our HTS assay was designed to screen for compounds that can induce a Ca²⁺ influx (screen 1, activators) and/or can alter the P4- and PGE1-induced Ca²⁺ influx (screen 2, inhibitors) (Figure 1A). Compound plates were validated with a Z′-factor >0.4 (Figure 1B). Hit selection was based on an arbitrary cutoff of 30% activation and/or inhibition compared to the controls (Figure 1C). We identified 106 (8.3%) hits, of which 31 were putative activators, 45 were putative inhibitors, and 30 had a dual effect (Figure 2A). Interestingly, the compounds with dual activating and inhibiting effects were mostly steroids. All 26 of them reduced the P4- but not PGE1-induced Ca²⁺ influx (Figure 2B, C). The hit steroidal compounds were validated by dose-response curves (DRC) in technical duplicates and biological triplicates and their EC₅₀s as well as their IC₅₀s for the P4-induced response were measured (Supplementary Table S1). Interestingly, of the 90 steroids in the Prestwick library, more than half (53%) were able to both activate and inhibit P4-induced Ca²⁺ influx in human sperm (Figure 2B). In contrast, inhibition of PGE1-induced Ca²⁺ influx was rarely observed with these steroids (Figure 2C). For the next part of this study, the 10 steroids with the strongest dual activation and inhibition potential were selected for further investigation. These were the synthetic steroids stanozolol, estropipate, dydrogesterone, medrysone, and the natural steroids dehydroepiandrosterone acetate (DHEA), testosterone, epiandrosterone, nandrolone, pregnenolone, deoxycorticosterone.

The 10 selected steroids were classified into 4 classes: androgens, estrogens, progestagens, and ‘others’, and they all induced an increase in [Ca²⁺]_i similar to P4 with estropipate inducing the highest increase in [Ca²⁺]_i (Figures 3A, D). All 10 steroids were tested in dose-response experiments and revealed a dose-dependent increase in [Ca²⁺]_i (Supplementary Figure S2). More specifically, estropipate, pregnenolone, and dydrogesterone induced a dose-dependent increase in [Ca²⁺]_i with effective concentrations as low as 10 nM. Stanozolol and nandrolone also induced a 30% increase in [Ca²⁺]_i at low concentrations in the nanomolar range (Supplementary Figure S2). EC₅₀ values ranged from 0.073 µM for pregnenolone to 0.5 µM for DHEA (Table 1). To determine the lowest activating concentration, EC₀₂ was calculated for all steroids, and compared with the reported maximum concentration found in blood serum (C_max) and the reported average concentration of steroids found in the follicular fluid (FF) (Figure 4). This comparison highlighted that the effective concentration at which progesterone, pregnenolone, dydrogesterone, epiandrosterone, nandrolone, and DHEA activate CatSper in vitro is within the biological range of their concentration measured in vivo. In the case of nandrolone, progesterone, and dydrogesterone, the EC₀₂ values were more than 100 times lower than the C_max reported value in the literature (Table 1; Figure 4). All steroids tested significantly reduced the P4 response with IC₅₀ values ranging from 0.19 µM to 1.75 µM (Supplementary Figure S2; Table 1). This indicates that P4 is the most potent steroid tested. Interestingly, a strong correlation was observed between the steroids inducing the highest increase in [Ca²⁺]_i and the ones that reduce the P4 response the most potently (Supplementary Figure S3). This decrease was not class-specific, suggesting that different functional groups can be accommodated in the binding site. Steroids did not alter the PGE1-induced increase in [Ca²⁺]_i suggesting that steroids act on the same binding site as that of P4 but not that of PGE1 (Figures 3C, F).

To determine whether the strong inhibitory effect of the 10 steroids on P4-induced Ca²⁺ influx was due to competitive binding, we performed cross-desensitization experiments by challenging pre-incubated sperm cells with steroids at 10 µM with increasing doses of P4 (Figure 5A). The P4-induced increase in [Ca²⁺]_i was reduced significantly by the presence of the 10 steroids tested. In contrast, the same experiment performed with increasing doses of PGE1 revealed a slight shift in the EC₅₀ without a decrease in maximum response, indicating that steroids reduce the potency of PGE1 but not its efficacy. (Figure 5B).

Besides ligand activation, CatSper can be weakly activated by changes in membrane voltage or alkalization of intracellular pH (pH_i) (Brown et al., 2019; Wang et al., 2021). Therefore, the 10 selected steroids were evaluated for their ability to affect CatSper-mediated Ca²⁺ influx induced by intracellular alkalization via the weak base NH₄Cl. All tested steroids almost completely suppressed the Ca²⁺ signals evoked by NH₄Cl at 10 mM (Figure 6). We also measured changes in pH_i in the presence of the 10 steroids. None of the steroids induced an increase in intracellular pH (Supplementary Figure S4), indicating that CatSper activation by steroids is not triggered by the increase in pH_i.

In the absence of a resolved structure for human CatSper, we used pharmacophore modelling and machine learning to search in silico for additional steroid compounds with a strong activating effect as well as define the steroid structure-activity relationship (SAR) responsible for the CatSper activation. We first constructed a pharmacophore model for the steroids with agonist activity that we identified in our initial screening (Figures 2B, C), and then combined them with a statistical regression model capable of relating their physicochemical properties to their activity profile. After validating the ability of our combined model to identify compounds that could increase the Ca²⁺ influx against a blind dataset, we used it to screen 783 compounds in silico. Of the 20 top compounds that we selected and tested experimentally, 11 showed activity in the high nanomolar range (Supplementary Table S1). Of particular interest are the synthetic steroids S46 and S65 (Figure 7A). These two compounds potently increased [Ca²⁺]_i in human sperm (EC₅₀ of 32 nM and 73 nM, respectively) at concentrations close to P4 (EC₅₀ of 2.5 nM) (Figures 7B, C). These two compounds also significantly decreased the response to P4 (IC₅₀ of 46 nM and 138 nM, respectively) (Figures 7D, E). Among the different physicochemical properties that we considered for the construction of the statistical regression model to predict the agonist effect of new steroids, we found a significant positive correlation between the predicted IC₅₀ value for human Ether-à-go-go-Related Gene (hERG) K⁺ channel block (QPlogHERG) and the number of amide groups and the square of the dipole moment divided by the molecular volume (dip²/V). This finding is not unexpected as dip²/V is a crucial term in the free energy of solvation of a dipole with volume V. Additionally, dip²/V and the octanol/water partition coefficient (QPlogP_o/w), as well as dip²/V and the y-component of the dipole moment, were found to be significantly positively correlated with the dependent variable, in this case, the agonist effect. The coefficients of the properties used in the estimated model can be found in Supplementary Table S2. From our model, it is clear that the desolvation of the steroids and their dipole moment in the direction of the electric field inside the channel with respect to their volume are key properties that need to be fine-tuned when designing new steroid analogues that target CatSper. To investigate the SAR of steroid activation of the CatSper channel, we evaluated a wide range of steroids with different functional groups in each of the steroid rings for their ability to induce Ca²⁺ influx in human sperm. Our results showed that the CatSper channel was activated by multiple steroids with various structural modifications. Structural analysis of the steroids provides useful insights into substitutions that can improve the potency of P4 analogues (Figure 8). In particular, various substitutions on the A-ring could be tolerated, including a pyrazole ring (stanozolol, Supplementary Table S1), resulting in moderate CatSper activators. Interestingly, the removal of the double bond present on the A-ring of P4 also increased the agonist activity of some compounds such as pregnenolone, S67, S85, and S45 (Figures 7B, C; Supplementary Table S1). This suggests that some flexibility in this ring may facilitate stronger interactions with the channel residues. In agreement with previous reports (Carlson et al., 2022b), methylation at the C19 position on the B-ring does not appear to be important for the activity, whereas our results indicate that methylation at the C7 position on the same ring can be tolerated. Any significant substitution on the C-ring rendered the steroids inactive or unable to bind at all. Only steroids with a hydroxyl group at the C13 or C14 position seemed tolerable in our dataset. Finally, bulky aliphatic substitutions at the C17 position of the D-ring led to a dramatic decrease in agonist activity, but small substitutions with hydrogen bond donors had a modest activating effect (deoxycorticosterone, stanozolol, norgestimate, Supplementary Table S1), suggesting that this part of the steroid may play a key role in stabilizing binding and thus activating CatSper. Notably, structural modifications on the D-ring have been reported to convert agonists to antagonists (Carlson et al., 2022b) and, therefore, this region can be exploited in the design of potent steroidal antagonists of CatSper (Supplementary Figure S1).

We finally investigated the effect of the 10 selected steroids on the acrosome reaction (AR) in human sperm (Figure 9A). All steroids showed similar results to negative control DMSO except stanozolol, epiandrosterone, and pregnenolone. These steroids induced acrosome reaction in more than 10% of the cells, but this increase was not significant because of the large variation in response among donors. P4 only slightly increased the AR in consistence with previously published data (Baldi et al., 1991; Prajapati, et al., 2022). We also assessed sperm penetration in a viscous medium using a modified Kremer’s test (Figure 9B). At a distance of 1 cm, the majority of the steroids tested increased the number of sperms that can penetrate in viscous media, although only stanozolol and estropipate induced a significant increase in cell number similar to P4. In addition, we evaluated the sperm motility in the presence and absence of the selected steroids at 10 µM using Computer Assisted Sperm Analysis (CASA) (Supplementary Figure S5). Incubation of spermatozoa with the selected steroids for 1 hour did not alter total, progressive, and hyperactive motility (Supplementary Figure S5). Cell viability was also assessed by flow cytometry of PI incubated cells (Supplementary Figure S6). Our results show that steroids at 10 µM were not cytotoxic.