First multifaceted ADME profile of ACP-105 (CAS: 1048998-11-3): a novel non-steroidal selective androgen receptor modulator used as doping in sports—integrative in silico toxicological studies for clinical and forensic toxicology purposes

Abstract

ACP-105 (CAS: 1048998-11-3) is a novel non-steroidal selective androgen receptor modulator (SARM), increasingly detected in anti-doping analyses, yet lacking a comprehensive ADME profile. This study provides the first integrative in silico characterization of ACP-105’s ADME properties using seven independent methods (ADMETlab 3.0, ADMET Predictor 12.0, ACD/Percepta, SwissADME, pkCSM, XenoSite and DruMAP). The compound demonstrated high gastrointestinal absorption (up to 100%), moderate lipophilicity (LogP 3.0–3.52), low solubility (LogS ~ − 4.1 to − 4.4), and Caco-2 permeability ranging from 13.6 to 152 × 10⁻⁶ cm/s. It shows strong plasma protein binding (77–99%), minimal free plasma fraction (< 1%), and variable tissue distribution (Vd 0.18–12 L/kg). Blood–brain barrier penetration was predicted in most models. Metabolic profiling identified six metabolites (M1–M6), primarily formed via CYP3A4, with additional contributions from CYP2C9, CYP2C19, and CYP2D6. ACP-105 is a consistent substrate for CYP3A4 (82–100%) and likely undergoes stable and unstable oxygenation, N-dealkylation, and UGT conjugation. Interactions with DNA/protein and potential cyanide release were also predicted. Clearance predictions varied (7.175–3.86 × 10⁻⁵ mL/min/kg), with a short half-life (~ 1.18 h), and no OCT2-mediated renal excretion expected. These findings provide a foundational ADME profile of ACP-105, essential for interpreting exposure in clinical toxicology and supporting evidence in forensic investigations involving illicit use.

Introduction

ACP-105 ((2-chloro-4-[(3-endo)-3-hydroxy-3-methyl-8-azabicyclo[3.2.1]oct-8-yl]-3-methylbenzonitrile)) belongs to the Selective Androgen Receptor Modulators (SARM) group, i.e., group of non-steroidal molecules designed to selectively stimulate androgen receptors in muscle and bone tissues while minimizing effects on other tissues, such as the prostate (Bradley et al. 2008; Broberg et al. 2021). Hence, this class of substances are banned in sports by the World Anti-Doping Agency (WADA) which have been classified under the “other anabolic agents” category on WADA’s Prohibited List since 2008. It should be noted that despite lacking full clinical approval, SARMs have frequently been detected in doping control samples, underscoring their misuse in competitive sports. A good SARM should have several qualities, namely, have high specificity for androgen receptors, have good oral bioavailability and pharmacological profile, and have tissue-selective pharmacological activity (Gao and Dalton 2007). SARMs are banned by WADA, because they exhibit doping effects and additionally are not fully tested. However, they have quickly gained popularity among athletes, in particular in weightlifters. Despite their non-steroidal structure, these compounds are considered an alternative to steroids, because they exhibit anabolic and androgenic potential and promote the expansion of muscle mass. It should also be mentioned that SARMs are not only seen as doping substances, but as an attempt to treat various diseases including cancer (prostate cancer) (Nyquist et al. 2021), skeletal diseases (osteoporosis) (Miner et al. 2007), or hematological diseases (anemia) (Negro-Vilar 1999).

Due to the fact that SARMs are fairly new substances on the market, mainly in the context of sports as well as doping, a lot of research is needed to assess. First, the safety of individual SARMs and latterly the detection of their use by athletes. Even if they are not used in official competitions subject to doping controls, they are widely available for sale so their safety of use should be thoroughly tested. Considering the very limited toxicological data available, what is known comes almost exclusively from experimental studies in animals and in vitro models. From these experimental findings, ACP-105 has been shown to exert strong anabolic effects in muscle tissue while having reduced activity in androgen-sensitive tissues like the prostate, suggesting tissue selectivity (Bradley et al. 2008). Due to the lack of additional clinical or toxicological data, we initially applied an in silico approach to predict selected toxicological endpoints in previously published study (Fijałkowska and Jurowski 2025) where a comprehensive computational analysis of ACP-105 toxicity was conducted, including acute toxicity, effects on internal organs, genotoxicity based on the Ames test, eye and skin irritation, and cardiotoxicity by testing hERG inhibitors. Since these studies focused on key toxicological parameters of this compound, and there are also scientific reports describing its metabolites found in biological material (including in rats, horses, and humans), the multifaceted Absorption, Distribution, Metabolism, and Excretion (ADME) profile of ACP-105 remains still largely unknown. Specifically, there is a lack of comprehensive data on other ADME parameters. Therefore, the aim of this publication was to address this gap using integrative in silico toxicological studies. The workflow of the conducted study is illustrated schematically in Fig. 1.

Fig. 1.

The workflow as idea of conducted in silico studies designed to predict the multifaceted ADME (absorption, distribution, metabolism, and excretion) profile of ACP-105 (CAS: 1048998-11-3)

In the absorption assessment, key parameters, such as gastrointestinal uptake, water solubility, partition coefficients (logP and logD), Caco-2 cell permeability, and potential interactions with the efflux transporter P-glycoprotein (P-gp), were analyzed. The distribution evaluation included predictions of volume of distribution (Vd), plasma protein binding (PPB), and the likelihood of blood–brain barrier (BBB) penetration. For metabolism, the study focused on interactions with cytochrome P450 enzymes, as well as the identification of potential sites and pathways involved in phase I and II metabolic reactions using tools such as XenoSite and Simulation Plus for prediction of hypothetical metabolites. Finally, the excretion phase involved predicting total body clearance, biological half-life, and possible interaction with the renal transporter OCT2 was conducted.

It should be noted that the integrative in silico toxicological studies conducted first time in this study was based on the application of a diverse set of computational tools, including ACD/Labs Percepta, ADMETlab 3.0, ADMET Predictor 12.0, DruMAP, pkCSM, SwissADME, Simulation Plus, and XenoSite. Compared to other typical in silico studies focused usually on one tool, this methodological diversity makes the analysis not only more robust but also significantly more insightful. It provides a nuanced view of how ACP-105 might behave toxicokinetically in the human body for the first time—critical information in the context of both clinical safety and forensic toxicology.

Methods

Several programs were used to obtain multifaceted ADME profile for ACP-105. This made it possible to compare the obtained results with each other and discuss where the differences in the obtained results generated by different software between the same parameters might come from. For this purpose, programs, such as ACD/Percepta, ADMETlab 3.0, ADMET_Predictor 12.0, DruMAP 2.0, pkCSM, SwissADME, and XenoSite, were used.

ACD/Percepta

Percepta (ACD/labs Percepta 2023.1.2) is a commercial program that enables the prediction of various toxicological properties using computational techniques. The program is not complicated to use, since to generate results for a given substance, it is necessary, as with other in silico programs, to use databases, enter the SMILES notation in the window designed for this, and the prediction results are obtained immediately. The presence of a reliability indicator makes it easier to discuss the results, as it gives insight into whether the results obtained are reliable. In addition to the above, Percepta provides visualization of chemical structures with color mapping on the analyzed structure. With Percepta, many ADME results were generated, including lipophilic and hydrophilic parts of the ACP-105, where only this program has such a computational model for these parameters. The study was conducted in Percepta version 2023.1.2 with a license purchased by the Institute of Medical Expertise in Lodz, Poland.

ADMETlab 3.0

ADMETlab 3.0 is an open-available software at https://admetlab3.scbdd.com/ that was developed to systematically evaluate ADMET chemicals and has a wide range of prediction. ADMETlab 3.0 uses advanced computational models to make pre-designs highly accurate and reliable, thus providing valuable support for drug design and optimization (Xiong et al. 2021). The 119 ADMET endpoints in ADMET 3.0 consist of 21 physicochemical, 20 medicinal chemistry, 9 absorption, 9 distribution, 14 metabolism, 2 excretion, 36 toxicity, and 8 toxicophores (Fu et al. 2024). Due to use this software, many valuable ADME endpoints has been obtained; quantitative prediction of absorption (logP, logD, logS and P-glycoprotein interaction), distribution (plasma protein binding, fraction unbound in plasma, volume of distribution, BBB interaction), metabolism (enzymatic pathways and substrate specificity) and excretion (total clearance, half-life).

ADMET_Predictor 12.1

ADMET Predictor is a machine learning tool that enables prediction of more than 175 endpoints. This software enables structure–activity modeling (QSAR) of absorption, distribution, metabolism, elimination and toxicity (ADMET). For the purpose of this paper, two of them were used, namely (i) identification of metabolic hotspots and metabolites for ACP-105 and (ii) enzymatic pathways and substrate specificity. The license for this program was obtained by Laboratory of Innovative Toxicological Research and Analyses, University of Rzeszów, from https://www.simulations-plus.com/academic-license-form/.

DruMAP 2.0

DruMAP is a platform that is used to analyze the metabolism and pharmacokinetics of drugs (DMPK). Like other programs, DruMAP predicts parameters based on the chemical structure of a given compound. In addition to parameters that come from public sources, DruMAP also has internal experimental data. The new prediction starts as in the other software excluded based on the structural information of the compounds provided by the user, either in Structural Data File (SDF) format or SMILES notation. For the prediction in this program, 23 prediction models have been contracted (Kawashima et al. 2023). For the purpose of this article, 3 parameters were selected. When performing quantitative prediction of absorption, Caco-2-permeability was chosen, when predicting distribution, fraction unbound in plasma was checked, while as for metabolism, enzymatic pathways and substrate specificity were chosen. DruMAP is a free-available program at https://drumap.nibiohn.go.jp/.

pkCSM

pkCSM represents a new in silico method for enabling pharmacokinetic and toxicity properties of compounds (Pires et al. 2015) and is available at https://biosig.lab.uq.edu.au/deeppk/prediction. pkCSM represents a strategy for predicting and optimizing the pharmacokinetic properties of small molecules based on distance-based graph signatures (Yeni and Rachmania 2022). For the purpose of this paper, many of the results obtained were generated using this particular program. pkCSM provided results for each of the four key elements of ADME. For absorption, five endpoints were obtained, i.e., logP, logS, GI absorption, Caco-2-permeability, and -glycoprotein interaction. As for distribution, we have fraction unbound in plasma, volume of distribution and BBB interaction. Metabolism as a third, no less important element of ADME obtained results regarding enzymatic pathways and substrate specificity. Finally, results were also obtained for excretion, i.e., total clearance and renal OCT2 substrate, which was only generated by the pkCSM program. Regarding the above, it can be seen how many of the results for the endpoints were generated just by this program, which makes it definitely important for in silico research on ADME.

SwissADME

SwissADME is a web-based tool that is freely available at http://www.swissadme.ch. One can enter the molecules to be estimated for ADME, physicochemical properties, drug similarity, or pharmacokinetics. The models were chosen for their robustness, speed, and ease of interpretation. Some of the models were adapted using open source algorithms, but there are also internally developed and tested models. With this program, predictions have been made for many endpoints mainly in terms of absorption. This program is the only one to provide results for qualitative prediction of absorption parameters bioavailability radar and Boiled-Egg Plot. In addition, predictions were made for logP, logS, GI absorption, and P-glycoprotein interaction. A result was also obtained for BBB interaction, which is a parameter measured in terms of distribution.

XenoSite

XenoSite is a predictor software that is freely available online and was created to model P450 metabolism. This software uses a neural network and allows access to generate a lot of data on P450 isoenzymes, among others, which include 9 main ones: 1A2, 2A6, 2B6, 2C8, 2C9, 2C19, 2D6, 2E1, and 3A4 (Matlock et al. 2015). XenoSite software accepts input in common chemical file formats, including SDF or SMILES. It provides tools for visualizing the likelihood that each atomic site is a site of metabolism for a variety of important P450s (Matlock et al. 2015). XenoSite server is available at http://swami.wustl.edu/xenosite.

Results and discussion

Absorption, distribution, metabolism, and finally excretion (so-called ADME or ADME profile) are four key biological processes that are used to determine the toxicokinetic parameters of investigated xenobiotic. They have a direct impact on the efficacy and adverse reactions to said substances. The first stage is absorption, which determines the movement of a substance that has not been metabolized from the site of its administration into either the general or local circulation (Alagga et al. 2025). The next stage, which refers to the movement of a drug between intravascular and extravascular compartments of the body, is distribution (Mansoor and Mahabadi 2025). The last two stages, i.e., metabolism and excretion, refer to chemical modification and breakdown in the body and removal of residues from the body, respectively, as elimination (Migliorati et al. 2022). The last process, i.e., excretion should not be used interchangeably with elimination, the former being “the irreversible loss of a chemically unchanged compound,” while elimination is “the irreversible loss of a drug from the site of measurement,” i.e., excretion plus metabolism (Doogue and Polasek 2013; Rowland and Tozer 2011). Due to the large number of parameters of the different stages of ADME that were subjected to prediction, the results and their discussion were divided into four stages.

Absorption (A): prediction of absorption parameters for ACP-105

The first step of toxicokinetics is absorption. Given that the absorption of ACP-105 in the human body is a multifactorial process influenced by a range of physicochemical and biological factors, a precise evaluation of its absorption potential necessitates the integration of various parameters, including lipophilicity, solubility, membrane permeability, and interactions with active transport systems. To facilitate a clear and structured interpretation of the complex data estimated for ACP-105, the results have been organized into three complementary sections: qualitative predictions (3.1.1), quantitative assessments (3.1.2), and structural analysis focused on hydrophobic and hydrophilic interaction profiles (3.1.3).

Qualitative prediction of absorption for ACP-105

To determine this step for the ACP-105, several relevant parameters were selected for in silico prediction. The SwissADME program was used to generate qualitative results for bioavailability radar and for BOILED-Egg Plot. The first parameter was bioavailability radar scheme, as shown in Fig. 2.

Fig. 2.

The bioavailability radar for ACP-105 generated from Swiss ADME

Usually, this kind of graph is generated for rapid assessment of drug similarity. The diagram shows six physicochemical properties and an area marked in pink and a red line. The pink area represents the optimal range for each properties (lipophilicity: XLOGP3 between − 0.7 and + 5.0, size: MW between 150 and 500 g/mol, polarity: TPSA between 20 and 130 Å2, solubility: log S not higher than 6, saturation: fraction of carbons in the sp3 hybridization not less than 0.25, and flexibility: no more than 9 rotatable bonds (Daina et al. 2017). Lipophilicity provides information on a molecule's ability, or lack thereof, to penetrate lipid membranes. In other words, it is the affinity of a molecule or grouping for a lipophilic environment (Gold et al. 2025). Polarity is related to the transport of a given molecule across membranes and is measured by the sum of the surface areas of polar fragments in a molecule. Size relates to the number of atoms in a molecule, which affects its binding to proteins and ability to diffuse. Then, solubility and saturation, respectively, determine whether the molecule dissolves well in water and the degree of saturation in the molecule, which ultimately affects its stability and also binding to receptors. The last element is elasticity, which provides information on the stiffness of the molecule and its level of ability to adapt to binding sites. On the chart, in addition to the six parameters described, there is also an area in pink and a red line, which respectively mark the optimal range of values for each of the above-mentioned properties and represent the values for the molecule under analysis. The interpretation of the bioavailability radar is to concentrate on the latter two, because if the red line is entirely contained within the pink area, the molecule follows the rules of drug similarity, and if it goes outside the pink area, there could be either problems with bioavailability or pharmacokinetic properties. In Fig. 2, a bioavailability radar has been generated for ACP-105, where it can be seen that all the red lines are contained within the pink area, so it is predicted that all the physicochemical criteria for drug similarity have been met, and ACP-105 should have good bioavailability. All of the above translates into the possibility of difficulties with the bioavailability of ACP-105.

The next parameter of qualitative prediction was BOILED-Egg Plot, estimated also by SwissADME—as shown in Fig. 3.

Fig. 3.

The BOILED-Egg plot of ACP-105 generated from Swiss ADME

Two predictions are based on this model, which relate to both (1) passive absorption through the human intestinal tract (HIA)—the white area as well as (2) penetration through the blood–brain barrier (BBB)—the yellow area. In addition, a visualization of two points is available on the chart. They are color-coded to indicate active P-gp transport (blue) and predicted non-substrate for P-gp (red). In the analyzed graph, the blue point is located in the yellow area, which may suggest its strong potential to penetrate the blood–brain barrier. On the other hand, the blue point is clearly visible, indicating active P-gp transport. Referring to this situation, it is necessary to explain the main function of this protein, which is to prevent the absorption of toxins and protect major organs including the brain (Marchetti et al. 2007). P-gp significantly affects the pharmacological properties of drugs and also their metabolites, modifying their bioavailability after administration. High levels of P-glycoprotein expression have been observed in endothelial cells of the blood/brain barrier, among others (Aryal et al. 2017). P-gp furthermore affects the clearance of drugs and substances by transporting them outside the cell against a concentration gradient using ATP. In addition to the parameters described above, quantitative prediction of logP, logD, logS, GI absorption, Caco-2-permeability, and P-glycoprotein interactions was also performed. First, a graphical representation of the LogD-pH—Fig. 4 as dependence diagram for ACP-105 was obtained using ACD/Percept.

Fig. 4.

The graphical representation of the LogD at a given pH for ACP-105

The diagram pre-dates the plot of this dependence at different pH levels, so that one can have insight into its solubility and behavior under different physiological conditions. This is essential for understanding the distribution of the relationship between the aqueous and lipid phases in biological systems with different pH environments.

When it comes to quantitative prediction of absorption for ACP-105, more parameters have been used than in qualitative prediction. Starting with a discussion of the first three, i.e., logP, logD, and logS. The definition of partition coefficient is the ratio of the distribution of a molecule in two immiscible solvents. The logarithm of partition coefficient (logP) between octanol and water is a property that is related to the physiochemistry and physiology of the test molecule, namely absorption, distribution, metabolism, excretion, and toxicity (ADMET) (Sun et al. 2023). The logarithm of D (logD), on the other hand, is the logarithm of the distribution coefficient (D) and is also one of the important parameters used to assess the drug susceptibility of a molecule in pharmaceutical formulations (Krishnamoorthy et al. 2018).

To ensure bioavailability and absorption after oral administration, a substance (drug) must have strong water solubility and for this reason logP is determined, which is used to estimate the water solubility (Vázquez-Tato et al. 2021). Firstly, LogP = 3.00 is the result obtained using Consensus Log_o/w in SwissADME software. This descriptor, which is the arithmetic mean of the values predicted by the five proposed methods (iLOGP, XLOGP3, WLOGP, MLOGP, and SILICOS-IT), was used for analysis. This value indicates moderate lipophilicity and an optimal value for oral bioavailability. Percepta predicts a score for LogP of 3.52, while Admetlab 3.0 predicts a score of 3.478, and the last program pkCSM predicts a score of 3.4023. The results obtained, especially from the last three programs, are very close to each other. The score generated by SwissADME is also not very different from these results, but it differs, which may be due to the fact that it was averaged from five individual descriptors. The LogP_o/w descriptor gave a score closest to the other programs at 3.43. The second parameter LogD was checked using Percept and Admetlab 3.0 and the results were 3.514 and 3.393, respectively. On the other hand, the value of the LogS parameter was calculated by SwissADME at − 4.09, which indicates low solubility in water (about 8.1 µM/L), which may cause problems with absorption and bioavailability. Admetlab 3.0 reported a similar result equal to − 4.404, Percepta − 4.34 and pkCSM − 4.118. The results obtained using the various programs for logP, logD, and logS agree with each other, confirming their reliability. The similar values suggest, first of all, the consistency of the predictive models and, moreover, indicate the high predictability of the properties of the tested compound ACP-105. Thus, the obtained results can be considered well founded and of practical significance in ADME analysis. Another parameter calculated during the absorption analysis was gastrointestinal absorption (GI absorption). Predictions were made using SwissADME, pkCSM, and Percept. The results obtained are similar and indicate a high level of absorption. Determining the value of this parameter is very important, mainly because the tested SARM is administered orally, so its bioavailability is an important issue.

Results for the Caco-2 permeability parameter, which measures the permeability of compounds across the human intestinal epithelial cell barrier, were made available by Percept, pkCSM and DruMap and standardized to the same result form. The first program predicts the highest permeability, while the results generated by the other two programs are similar to each other and indicate several times lower permeability of ACP-105.

On the other hand, the interpretation of P-glycoprotein interaction (P-glycoprotein substrate) is the most difficult parameter to interpret, as the results generated by the programs obtained, differ significantly from each other. Only the ADMETlab 3.0 and pkCSM programs reported similar values to each other, which suggest in both cases that ACP-105 is not a P-gp substrate.

The results for ACP-105 absorption are summarized in Table 1.

Table 1.

Prediction of absorption for ACP-105 using in silico software

Parameter	Software	Result
LogP	ADMETlab 3.0 Percepta pkCSM SwissADME	3.478 3.52 3.4023 3.00
LogD	ADMETlab 3.0 Percepta	3.393 3.514
LogS	ADMETlab 3.0 Percepta pkCSM SwissADME	− 4.404 − 4.34 − 4.118 − 4.09
GI absorption	Percepta pkCSM SwissADME	100% 94.412% High
Caco-2 permeability	DruMAP Percepta pkCSM	13.553 × 10–6 cm/s 152 × 10–6 cm/s 23.55 × 10–6 cm/s
P-glycoprotein interaction (P-glycoprotein substrate)	ADMETlab 3.0 Percepta pkCSM SwissADME	0–10% 49% No Yes

Visualization of hydrophobic and hydrophilic domains relevant to absorption of ACP-105

Predictions of structural relativity to the hydrophobic and hydrophilic properties of the SARM under study were also made. An evaluation by Global, Adjusted Locally According to Similarity (GALAS) modeling was performed and the ACD/LogP algorithm was used. The Percepta program enabled the calculation of LogP and LogD by comparing the ACP-105 molecule with structurally similar molecules whose experimental data are known. Hydrophilic groups were marked in red, while hydrophobic groups were marked in green (Fig. 5).

Fig. 5.

Prediction of property distribution indicating lipophilic and hydrophilic parts of the ACP-105 provided by the ACD/LogP GALAS algorithm

Distribution (D): prediction of distribution parameters for ACP-105

Distribution, as the next step occurring after absorption, describes how a substance administered to the body spreads in the body. Distribution varies depending on certain properties, such as the biochemistry of the drug or the physiology of the person who takes the substance (Slørdal and Spigset 2005). To be able to analyze the distribution of ACP-105, several parameters were chosen. The results summarizing these parameters are shown in Table 2.

Table 2.

Prediction of distribution for ACP-105 using in silico software

Parameter	Software	Result
Plasma protein binding	ADMETlab 3.0 Percepta	99% 77%
Fraction unbound in plasma	ADMETlab 3.0 DruMAP Percepta pkCSM	0.7% 1.6% 0.23 0.226
Volume of distribution	ADMETlab 3.0 Percepta pkCSM	0.18 12 L/kg 0.437
Blood–brain barrier interaction	ADMETlab 3.0 Percepta pkCSM SwissADME	90–100% 0.78 0.267 Permeant
Central nervous system permeability	Percepta pkCSM	− 1.79 − 2.77

Plasma protein binding (PPB) is a parameter that determines whether a substance binds to plasma proteins. This is an important issue, as it affects first the distribution of the substance/drug and also its biological activity. If a substance, after absorption into the systemic circulation, is bound to plasma proteins, these bound compounds will not cross the biological membrane and will not induce a pharmacological response as a result. Using the Percept program, a prediction was made with a probability of 77% that ACP-105 would bind to plasma proteins. The ADMETlab program predicts an even higher probability of as much as 99%. These results may suggest that ACP-105 will not be biologically active. In addition, any reduction in the binding process of a compound to plasma proteins increases the amount of drug available to act on receptors, which may result in a greater effect or increased probability of toxicity (Grogan and Preuss 2025). Fraction unbound in plasma is another parameter checked in distribution analysis. This is another important factor that is checked during the efficacy of the compound/drug in pharmacokinetic and pharmacodynamic studies. This is because only the free fraction (unbound) can interact with target proteins in the drug data, e.g., receptors, channels, and enzymes (Bohnert and Gan 2013).

The ADMETlab and Percepta programs provided very similar values, 0.7% and 1.6%, respectively, which is practically zero probability, for the occurrence of the free fraction in plasma. In other words, these results suggest that ACP-105 is almost completely bound to plasma proteins (about 98%) which is largely in line with the results obtained for the parameter discussed above—plasma protein binding. Larger values and also close to each other were presented by Percept and pkCSM of 22.74% and 22.6%, respectively. These results thus represent a quite different profile from the results obtained by the two previous softwares. They suggest that a much larger portion of the drug remains in the free form which in effect allows it to access the action. These differences may be due to a number of reasons. Note that the programs use different predictive models, databases, or computational methods. In addition to the above parameters for distribution, predictions were also made for volume of distribution (Vd) and blood–brain barrier (BBB) interaction. Volume of distribution is a parameter generally used to describe the spread of a drug. This volume is defined as the amount of drug present in the body divided by the plasma concentration of that drug (Nancarrow and Mather 1983). This parameter depends on certain factors, because when the molecule is small and hydrophilic it has a higher Vd level. On the other hand, when there is a very large molecule that can be bound to proteins in the circulation and remain inside the vessels, it is unable to diffuse and as a result has a low level of Vd (Grogan and Preuss 2025). Obtained results (0.18–0.437 L/kg) were generated by ADMETlab and pkCSM, and suggest limited distribution. In contrast, the result generated by Percept differs drastically from those above. The result obtained by the latter program equal to 12L/kg indicates a very high level of Vd. Such a value may also suggest that the substance is lipophilic, which in effect means that it can easily penetrate biological membranes. ACP-105 can therefore, for example, accumulate in the body, e.g., in adipose tissue. As for the BBB, SwissADME predicts that ACP-105 can penetrate the blood–brain barrier, also ADMETlab has set such probability at 90–100% which is very high. Another program, Percepta, also at a fairly high level predicts the possibility of penetration through the blood–brain barrier; 78%. Only the last of the programs, pkCSM, generated a score for this parameter of 26.7%, which is a significant departure from the results obtained above. However, despite this discrepancy between pkCSM and the other three programs, one can conclude that ACP-105 does indeed have the ability to penetrate the BBB, since in three out of four programs such a result was obtained.

In addition to that, in Percepta, ACP-105 was determined to have sufficient brain penetration to detect central nervous system (CNS) activity; see Fig. 6.

Fig. 6.

A scatter plot to compare the brain penetration characteristics of ACP-105 with a set of well-known CNS and peripheral drugs analyzed by ACD/Percepta

Metabolism (M): prediction of metabolism parameters for ACP-105

The process of transforming a drug into successive compounds in the body is called metabolism (Grogan and Preuss 2025). It is also essential for the bioavailability of orally administered drugs, and what's more, it can produce active metabolites (Doogue and Polasek 2013). The entire process of metabolism takes place in many areas in the body, including the gastrointestinal tract, skin, plasma, kidneys, and lungs. However, most of the metabolism takes place through phase I reactions, i.e., CYP450 (P450) and phase II. As for the former, it should be mentioned that these reactions mostly turn substances into polar metabolites by oxidation, in effect allowing phase II reactions to occur (Starkey and Sammons 2015). To be able to predict the metabolism of ACP-105, a prediction was made using in silico tools of parameters such as P450 inhibitors and substrates. Moreover, due to in silico tool XenoSite the metabolic hotspots were identified. For better understanding this key of ADME, the metabolic results are presented in two sections. The section “Identification of high-reactivity metabolic regions and metabolites for ACP-105” includes the identification of high-reactivity metabolic regions and metabolites for ACP-105 generated using neural network-based tool Xenosite (Table 3) and ADMET_Predictor 12.0 (Fig. 7). The section “Enzymatic pathways and substrate specificity” discusses the enzymatic pathways and substrate specificity, i.e., predicted metabolites using ADMET_Predictor 12.0 (Table 5), metabolism parameters using various in silico methods (Table 6), and finally heatmap of ACP-105 substrate (Fig. 8).

Table 3.

Prediction of high-reactivity metabolic region(s) for ACP-105 using XenoSite

Fig. 7.

Predicted high-reactivity metabolic regions for ACP-105 obtained from ADMET_Predictor 12.0

Table 5.

Predicted metabolites (n = 6; M1–M6) of ACP-105 obtained with ADMET_Predictor 12.0

Table 6.

Prediction of cytochrome P450 (CYP) substrates for ACP-105 using in silico tools

	Software
Enzymes	ADMETlab 3.0	DruMap	Percepta	ADMET_Predictor 12.0	pkCSM
	Probably [%]			Categorical (Yes/No)
CYP2D6	0–10	71.2	8	Yes (54%)	No
CYP3A4	90–100	82.5	89	Yes (86%)	Yes
CYP1A2	90–100	61.7	4	Yes (91)	NA
CYP2C19	90–100	NA	25	Yes (81%)	NA
CYP2C9	0–10	60.0	17	Yes (65%)	NA

Fig. 8.

Heatmap of ACP-105 substrate predictions across CYP isoenzymes

Identification of high-reactivity metabolic regions and metabolites for ACP-105

Metabolism testing is a key point in evaluating the metabolism profiles of substances with medical applications. An in silico computational approach provides the ability to predict potential points of metabolism mediated by P450 group proteins (Andrade et al. 2014).

Analyzing the results from the XenoSite website, a variety of algorithms related to predicting metabolic modifications linked mainly to P450 proteins are available. Processes such as epoxidation or N-dealkylation have been analyzed, and there are methods for simultaneously labeling metabolic sites and reaction types, which have been classified into five key reaction classes: stable and unstable oxidation, dehydrogenation, hydrolysis, and reduction. Dehydrogenation as the process of hydrogen removal, hydrolysis indicating the process of bond cleavage involving water, and reduction as the process of reduction by enzymes mainly other than CYP. In addition to those mentioned above, uridine diphosphate glucuronosyltransferase (UGT) conjugation process and algorithms to check potential toxicity between the substance/drug and biological macromolecules, such as DNA, proteins, etc., are also available. In the results obtained for predicting metabolic modifications associated with the action of ACP-105, many hot spots were obtained depending on the type of metabolic transformation. As for epoxidation, it should be noted that epoxides are a group of reactive metabolites, acting mainly on aromatic and double bonds. In the results obtained for ACP-105, these are sites within just the aromatic ring, the so-called Sites of Epoxidation (SOE). The software provides identification of these sites with an accuracy of 94.9% while separating epoxidized and non-epoxidized molecules with an accuracy of 78.6%, which provides reliable results. The second result obtained is the prediction of the N-dealkylation site, a process catalyzed by cytochrome P450 enzymes, and is mainly focused on the nitrogen atom site in the bicyclic moiety.

Discussing further parameters, hot spots in the cyclopentane bicyclic moiety for stable oxidation were also highlighted. On the other hand, for unstable oxidation, the indicated sites are fewer than in the above case, they are mainly concentrated around the nitrogen atom, and they indicate metabolic sites that may result in unstable and potentially toxic metabolites. Regarding the hydrogen elimination process, XenoSite predicts, several potential hot spots, but in the case of hydrolysis and reduction, no potential sites of modification associated with these processes were visualized. In contrast, parameters related to reactivity, i.e., potential sites that could cause toxicity between ACP-105 and biological macromolecules, have been visualized at multiple points on the molecule. These sites are the methyl group, the nitrile group, or the carbon atoms at the ring nitrogen atom that binds to the bicyclic moiety.

For GSH conjugation and protein binding, similar sites in the molecule have been visualized in blue. In the former case, these are potential glutathione attachment sites that may be involved in the detoxification of metabolites, which may ultimately facilitate their excretion. As for proteins, on the other hand, the highlighted sites are analyzed as potential protein interaction sites. Such interactions may affect toxic effects mainly in terms of toxicokinetics. In the case of DNA, sites highlighted in blue in the molecule, may indicate a high probability of DNA adduct formation. In the case of cyanide formation, many sites of low-to-moderate reactivity are predicted, mainly, these sites are in the cyclopentane bicyclic moiety. Thus, it can be inferred that under certain conditions, the metabolic decomposition process of ACP-105 can lead to biochemical pathways resulting in cyanide release, which can translate into toxic effects.

Thus, it can be clearly seen that there are many predicted potential sites in the ACP-105 molecule that can interact toxicly with biological molecules. All of these processes affect the activity of the molecule and could potentially contribute to toxic effects. Table 3 summarizes the parameters discussed above for ACP-105 generated with XenoSite.

In addition to the results obtained above for the prediction of metabolism mainly related to P450, a very important point is the prediction of the final products of ACP-105 metabolism. In this way, it is possible to gain a deeper understanding of how the compound under study works, what pathways enter into its biotransformation and how this may translate into possible toxicity. Using ADMET Predictor 12.0, a prediction of the likely end metabolites of ACP-105 was made and the results are summarized in Fig. 7 and the metabolic hotspots are also predicted and presented in Table 4.

Table 4.

Predicted ‘metabolic hotspots’ for ACP-105 obtained from ADMET_Predictor 12.0

Enzymatic pathways and substrate specificity

The first step was using ADMET_Predictor 12.0—a tool for predicting metabolites that can be formed by enzymes belonging to the cytochrome P450 (CYP) enzyme family. program provided prediction of six metabolites (M1–M6) (Table 5).

Drug metabolism as the process of metabolic breakdown of drugs by enzyme systems (De Groot 2006) reduces their therapeutic effect (Tao et al. 2020). Regarding drug metabolism, CYPs are the best-known drug-metabolizing enzymes (Almazroo et al. 2017) and are involved in more than 90% of reported enzymatic reactions (Rendic and Guengerich 2015). In silico tools predict CYP activity as substrates. In the available in silico methods, predictions are made mainly for CYP isoforms belonging to the CYP1, CYP2, and CYP3 families probably because they are responsible for the metabolism of about 80% of drugs. Predictions were made for CYP1A2, CYP3A4, CYP2C9, CYP2C19, CYP2D6, and CYP3A4.

For CYP3A4 do the results overlap at a high level. ADMETlab predicts in the range of 90–100%, DruMap at 82.5%, percepta 89%, and moreover, pkCSM and ADMET_Predictor 12.0 generated a result of yes, for a positive answer that ACP-105 is metabolized by this particular isoform. When it comes to CYP1A2 and CYP2C19 ADMELlab 3.0 and ADMET_Predictor 12.0 predicts that this proteins are substrates with average score 80–100%. while Percepta for these two enzymes predicts a much lower probability of a positive prediction as if they were substrates, at 4% and 25%, respectively. In the rest of the cases, the discussion of the results is difficult due to the differences that occur between the data obtained. The results are summarized in Table 6.

The final stage in the metabolic analysis for ACP-105 was the preparation of a heatmap summary, which produced results for ACP-105 substrate prediction for the above-discussed CYP isoenzymes (Fig. 8).

Comparative analysis of in silico predictions of ACP-105 interactions with key cytochrome P450 (CYP) isoenzymes reveals significant differences between software platforms. The results obtained for CYP3A4 are unambiguous, as all of the software that generated results, including ADMETlab 3.0, DruMap, Percepta, ADMET Predictor 12.0, and pkCSM consistently, predict that ACP-105 is a substrate for this isoenzyme. Considering the results obtained for the CYP2C19 isoenzyme, two programs, namely, ADMETlab 3.0 and ADMET Predictor 12.0, overlap and oscillate in the 81–95% range, but the result generated by Percept predicts only a 25% probability that ACP-105 is a substrate for this isoenzyme. It is noteworthy that the predictions for CYP1A2 differ significantly—while ADMETlab 3.0 and ADMET Predictor 12.0 indicate a high substrate potential, Percepta estimates a low probability of only 4%, highlighting methodological discrepancies. This points to possible limitations in algorithmic coverage or representation of the dataset. Also discussing the results for CYP2D6, we see significant variability between in silico methods, as results of 54–71% were obtained for CYP2D6 using DruMap and ADMET Predictor 12.0 and 5–8% using ADMETlab 3.0 and Percept. For this reason, discussion of these results is hampered by inconsistencies in the data obtained. The last of those analyzed—CYP2C9, like CYP2D6 also generated results for the average probability that the tested compound is a substrate for this isoenzyme at 60–65% using DruMap and ADMET Predictor 12.0, but results were also obtained at 5–17% using ADMETlab 3.0 and Percept, respectively, indicating a very low possibility for ACP-105 to be a substrate for this isoenzyme. This therefore suggests minimal participation in the metabolism of ACP-105.

Summarizing the above discussed, it is clear that Percepta is the program that provided the results closest to the probability that ACP-105 is not a substrate for most CYP isoenzymes. In contrast, the categorical nature of pkCSM, which only provides binary output, supports key predictions (e.g., CYP3A4) but lacks qualitative interpretation. By highlighting the discrepancies in results obtained with different in silico tools, one can see the importance of using different prediction software to predict and discuss the metabolic profile of a substance as accurately as possible.

Excretion (E): prediction of excretion parameters for ACP-105

Excretion is the last of the four key ADME processes by which a drug is removed from the body. Most excretion occurs through the kidneys, but it can also happen through the lungs, skin or gastrointestinal tract (Grogan and Preuss 2025).

The clearance (CL) is one of the main parameters used to describe excretion (Pantaleão et al. 2022). Total clearance (CL tot) determines with what efficiency the drug was removed from the body. The parameter has important applications, for example, in determining at what interval a drug should be dosed (Yap et al. 2006). The CLplasma and total clearance parameters were checked using two in silico tools, i.e., AdmetLab 3.0 and pkCSM, and although the former predicts moderate clearance (7.175 mL/min/kg), the latter program generated a result for total clearance that indicates that ACP-105 is not removed from the plasma volume at all (− 4.413 log/mL/min/kg which in conversion yields 3.86 × 10⁻⁵ mL/min/kg). Thus, these results are completely discrepant, indicating that the predictive models of these programs do not match.

The second parameter checked during the prediction of the last stage of ADME, i.e., excretion, is the final half-life (t_1/2 in plasma) obtained with the AdmetLab 3.0 program. The obtained result of 1.182 shows that ACP-105 is classified as a short half-life drug, which in practice translates into the prediction that the compound lasts a short time in the in the body.

The last parameter—Renal OCT2 substrate—was generated by pkCSM. OCT2 is an organic cation transport protein that is expressed mainly in the kidney. According to the prediction, ACP-105 is not a substrate of OCT2 that is, it is not actively transported. The results are summarized in Table 7.

Table 7.

Parameters of excretion for ACP-105 obtained with various in silico methods

Parameter	Software	Result
Total clearance	ADMETlab 3.0 pkCSM	7.175 mL/min/kg 3.86 × 10–5 mL/min/kg
Half-life T1/2	ADMETlab 3.0	1.182 h
Renal OCT2 substrate	pkCSM	No

Conclusions

This study provides the first multifaceted in silico ADME profile of ACP-105, a novel non-steroidal selective androgen receptor modulator (SARM) increasingly detected in the context of sports doping. By utilizing a wide array of complementary computational tools, including ADMETlab 3.0, ADMET Predictor 12.0, ACD/Percepta, DruMap, pkCSM, SwissADME, and XenoSite, the research offers a robust and integrative analysis of key toxicokinetic properties. The results indicate that ACP-105 demonstrates favorable gastrointestinal absorption, moderate lipophilicity, and a high probability of blood–brain barrier penetration. The compound is strongly bound to plasma proteins and likely exhibits limited renal excretion, with a relatively short biological half-life. On the other hand, the metabolism section reveals that CYP3A4 is consistently predicted across all platforms as a primary metabolic pathway, underscoring the importance of this isoenzyme in ACP-105 biotransformation. Other CYPs such as CYP2C19 and CYP1A2 show more variable predictions, reflecting inter-tool differences in algorithmic structure and training datasets.

Abbreviations

3R

Replacement, reduction, and refinement

ACP

Androgen receptor modulator compound (in the context of ACP-105)

ADME

Absorption, distribution, metabolism, and excretion

ADMET

Absorption, distribution, metabolism, excretion, and toxicity

ADMETlab

Platform for ADMET evaluation

admetSAR

Platform for ADMET evaluation

ADT

Acute dermal toxicity

AIT

Acute inhalation toxicity

AOT

Acute oral toxicity

CAS

Chemical Abstracts Service

CYP

Cytochrome P450

DMPK

Drug metabolism and pharmacokinetic

DNN

Deep neural networks

FDP

Fixed dose procedure

GHS

Globally harmonized system

hERG

Human ether-à-go-go related gene

LD₅₀

Lethal dose 50%

NAMs

New approach methodologies

NPS

New psychoactive substances

OECD

Organisation for Economic Co-operation and Development

PA

Prediction accuracy

QSAR

Quantitative structure–activity relationship

RAD

Testolone (w kontekście RAD140, powiązane z SARM)

RI

Reliability Index

SARM

Selective androgen receptor modulator

SL-Tox

Super learner-toxicity

SMILES

Simplified Molecular Input Line Entry System

STopTox

Systemic and Topical chemical toxicity

TTC

Threshold of Toxicological Concern

UDP

Up-and-Down Procedure

UGT

UDP-glucuronosyltransferase

VEGA

Virtual evaluation of genotoxicity and acute toxicity

WADA

World Anti-Doping Agency

Author contributions

OF: data curation, writing—original draft preparation, and visualization; KJ: writing—original draft preparation, data curation, supervision, writing—review and editing, formal analysis and investigation, writing—original draft preparation, and visualization.

Funding

Not applicable.

Data availability

All data analyzed in this study will be provided to the corresponding author upon request.

Declarations

Conflict of interests

The authors declare no competing financial or non-financial interest.

Ethical approval and consent to participate

Not applicable.