The dissociation of anabolic from androgenic effects is proposed to result from tissue-selective actions. A classic method to quantify these is the aforementioned Hershberger assay (5). In the original experiment, 21-day-old rats are castrated and then receive daily administration of the test androgen for 7 days. On day 8, animals are sacrificed and dissected to measure the weights of the levator ani muscle, ventral prostate and seminal vesicles. The ratio between the levator ani and ventral prostate or seminal vesicles weight changes, compared with castrate controls, is calculated to indicate the compound’s relative anabolic versus androgenic activity. The levator ani weight change serves as a surrogate for anabolic action and that of the ventral prostate for androgenic action. These numbers are then compared with those of a reference standard, usually testosterone or testosterone propionate for injectable and 17α-methyltestosterone for oral compounds. We will discuss this assay because it holds historical value and is applied in contemporary SARM research.

While an appealing method, results of the Hershberger assay should be interpreted with caution because of several limitations. First, the levator ani muscle is doubtful to be representative of skeletal muscle. As early as 1957, researchers questioned its validity as an index of anabolic action, arguing that its growth is sex-linked rather than generally myotrophic (22). Despite its common name (a misnomer for the dorsal bulbocavernosus/bulbospongiosus muscle (23, 24)), contraction of the muscle does not lift the anus, is only present in male rats and active only during copulation (24). As a reproductive muscle, it is exquisitely sensitive and dependent on androgens, atrophying significantly after castration (22) and having a severalfold higher AR content than skeletal muscle (25). Second, while rat prostate size dose-dependently increases in response to androgens, human prostate volume remains unchanged even with supraphysiological dosages up to 600 mg testosterone enanthate weekly (approximately 6 times the dosage used for testosterone replacement therapy) for 20 weeks (26). While differences in lifespan and the rate of biological processes between rats and humans limit comparison, it nevertheless seems that rat prostate weight changes in response to androgens are not translatable to humans. Third, there is no empirical support to assume that size changes of the rat ventral prostate or seminal vesicles reflect clinically desirable tissue selectivity (e.g., the assumption that a favorable response of the ventral prostate to an androgen would also signify a favorable response on erythropoiesis, or any other undesired androgenic effect, in relation to its anabolic effect). Fourth, the apparent dissociation of anabolic from androgenic effects depends on the timing of observation, as tissues respond differently across androgen concentration ranges (27).

Despite these drawbacks, pharmaceutical companies continue to use (modifications of) this assay for SARMs screening (28–37). Compounds with ‘favourable’ ratios are further explored and, conversely, compounds with ‘unfavorable’ ratios are discarded, even though the assay holds questionable clinically translational value beyond demonstrating any androgenic action. The failure of translational success from the hundreds, if not thousands, of steroidal compounds evaluated by this bioassay to translate into clinical utility — despite a substantial fraction demonstrating favorable results — might as well serve as an empirical testimony to abandon it for screening purposes.

More recent investigations complement in vivo animal research with in vitro transactivation assays. These involve transfection of cell lines with a luciferase reporter gene under transcriptional control of androgen response elements (AREs), allowing androgen activity to be quantified trough luminescence measurements. Ostrowski et al. applied this method in C2C12 mouse myoblasts and rat prostate epithelial cell lines (38). They evaluated various concentrations of the bicyclic hydantoin SARM BMS-564929 to establish the potency (EC₅₀, calculated as the concentration at which 50% of the maximum stimulatory effect of dihydrotestosterone [DHT] was achieved) compared with testosterone in either cell line. BMS-564929 had an EC₅₀ of 0.44 and 8.66 nM in the myoblasts and prostate cells, respectively, whereas testosterone’s values were 2.81 and 2.17 nM. This suggests a roughly 20-fold selectivity in muscle over prostate cells. The researchers also investigated the androgenic and anabolic effects of BMS-564929 using a model similar to the Hershberger assay. The ED₅₀ (defined as the dose at which tissue wet weight reached 50% of the weight of that of the sham group) was 0.0009 mg/kg in the levator ani and 0.14 mg/kg in the prostate, indicating a 160-fold selectivity for the levator ani over the prostate. This result translates to an 80-fold selectivity over testosterone propionate in the same assay. At first glance, these results look remarkably promising. However, as excellently argued by Goa and Dalton (39), these results (and that of many other SARMs) are likely the result of lack of androgenic amplification by 5α-reductase — unlike with testosterone. The dose-response curves of BMS-564929 vis a vis testosterone with a 5α-reductase inhibitor from a comparable study (40) show considerable overlap.

Comparison of a SARM with testosterone can thus create the misleading impression of relative tissue selectivity as testosterone’s effects are potentiated via conversion to DHT in certain tissues such as the prostate. A similar problem arises when a comparison is made with DHT (as occasionally done (28, 32, 41)) instead of testosterone. Whereas testosterone’s effect is amplified in tissues exhibiting 5α-reductase activity, DHT is inactivated in tissues with 3α/β-dehydrogenase activity yielding the inactive metabolites 3α- and 3β-androstanediol (42). Importantly, irreversible 3α-reduction of DHT occurs in skeletal muscle (43) — both in rodents (including in the levator ani muscle) (44) and humans (45).

Obviously, part of SARMs’ tissue selectivity stems from differences in metabolic fate compared with steroidal androgens such as testosterone and DHT. However, certain decades-old steroidal androgens also exhibit these metabolic features, e.g. oxandrolone (46). Notably, nandrolone, a testosterone derivative lacking the methyl group at the C19 position, is 5α-reduced to dihydronandrolone (DHN). DHN has lower AR binding affinity than nandrolone and thus may result in reduced rather than amplified activity in tissues expressing 5α-reductase (47). Arguably, such steroidal androgens offer a more meaningful basis for comparison in these assays than testosterone and DHT, although other limitations of these assays remain.

Finally, an overarching issue in SARM development for many clinical applications is their lack of estrogenic effects. Unlike some steroidal androgens (in particular testosterone), current SARMs are not aromatized into estrogens and do not exhibit intrinsic estrogenic activity. They therefore cannot replace or supplement estrogen. When SARMs suppress gonadotropin release, testosterone and estradiol levels fall, potentially causing a relative or absolute estrogen deficiency. In men, experimentally induced estrogen deficiency increases body fat and contributes to declines in sexual function (48). Over time, longstanding inadequate estrogen levels may increase the risk of osteoporosis (49, 50). Even when SARMs achieve substantial tissue selectivity, their inability to preserve or supplement estrogenic action presents an additional clinical challenge. Nevertheless, because negative feedback on the hypothalamo-pituitary gonadal axis (HPGA) largely relies on estrogenic feedback, gonadotropin suppression may be mild in therapeutic dosages, as also is the case for nonaromatizable steroidal androgens (51–54).

Early SERMs, such as tamoxifen, were initially viewed as antiestrogens. Their surreptitious tissue selectivity was unknown at the time and identified only later (55). The molecular basis for this selectivity remained elusive until the discovery of a second estrogen receptor isoform (ERβ; leading to renaming of the classical ER to ERα) in 1996 provided some insight (56, 57). Future developments in the field could exploit this biological ‘gift’ for potential selectivity. As the AR lacks multiple receptor isoforms with distinct biological action, this relatively easy vector to establish tissue selectivity is unavailable in SARM development. Instead, tissue selectivity needs to rely on other mechanisms, such as the interaction with the AR and tissue-specific differences in AR coregulators and their recruitment.

While not purposely engineered for it, the estrogenic action of tamoxifen in the uterus relies on high expression of steroid receptor coactivator 1 (SRC-1) in endometrial cells (58). This highlights the critical role of coregulators that are differentially expressed in tissues and suggests that a similar approach could apply to SARMs (59). The underlying idea is that different ligands induce a different conformational change in the AR ligand-binding pocket (LBP), which, in turn, induces a conformational shape at the AR surface that differentially binds coregulators and thereby exhibits a novel profile of coregulator interaction that favors a desired tissue selectivity. This might also include altered post-translational modifications of the AR, as the receptor contains many phosphorylation, and to a lesser extent acetylation, SUMOylation, ubiquitination and methylation sites across its N-terminal, DNA-binding, hinge, and ligand binding domains (60). SARM research on post-translational modifications remains unexplored, however.

Ostrowski et al. provided structural evidence for the concept of adopting a distinct conformational shape by resolving the cocrystal structure of the AR ligand-binding domain (LBD) bound to the SARM BMS-564929 at 3 Å resolution, and compared this with the known structure of the DHT-bound AR LBD at 2 Å resolution (38). Unlike DHT, BMS-564929 interacted with the side chain of F764 and lacks the hydrogen bond with T877. This forms a unique LBP which likely determines distinct surface features for differential coregulator action. These findings establish credibility for the concept of differential coregulator recruitment, echoing the mechanism of SERMs (61).

Additional support comes from a study by Hikichi et al. (34). By using a mammalian two-hybrid (M2H) assay expressing the AR and 112 cofactors in human embryonic kidney cells (293T), the authors determined the cofactor recruitment profile of the steroidal ‘SARM’ TSAA-291 and compared that with DHT. Of the 112 cofactors, 55 bound to AR, with 12 demonstrating differences in recruitment by TSAA-291 compared with DHT. Among these, the protein inhibitor of activated STAT 1 (PIAS1) was notable for its differential recruitment. The authors further showed that PIAS1 was more highly expressed in prostate tissue than in skeletal muscle, and that PIAS1 functions as a coactivator of AR transcriptional activity. PIAS1 overexpression increased AR transcriptional activity, while silencing of PIAS1 by siRNA reduced DHT-induced prostate-specific antigen (PSA) production. This indicates that PIAS1 plays a functional role in AR signaling and suggest that differential recruitment of cofactors like PIAS1 may contribute to tissue-selective activity of TSAA-291. While suggestive, these findings do not conclusively prove that coregulator differences explain clinically meaningful tissue selectivity.

Importantly, coregulator research followed observations of selective activity in bioassays rather than that it preceded or guided the rational design of SARMs.

Nongenomic pathways may also play a role in SARM tissue selectivity. Narayanan et al. investigated this using the arylpropionamide SARM S-22 (62). Xenopus oocyte maturation, a purely nongenomic effect that depends on the AR, was induced by both S-22 and testosterone. Previous research demonstrated that some steroidal androgens, such as DHT and R1881, do not induce Xenopus oocyte maturation, whereas others, such as testosterone and androstenedione, do (63). Additionally, Narayanan et al. showed that LNCaP cells treated with S-22 or DHT showed differences in phosphorylation of various kinases. S-22 phosphorylated p38 MAPK, whereas DHT phosphorylated extracellular signal-regulated kinase (ERK). In U2OS osteoblasts transfected with AR, both DHT and S-22 phosphorylated ERK to a similar degree. These findings suggest a shared action on bone but divergent on the prostate resulting from differences in nongenomic signaling pathways. Importantly, modulation of these pathways affected recruitment of the ligand-AR complex to a PSA enhancer region in LNCaP cells, underscoring that these differences can also affect genomic action. These findings support the notion that nongenomic pathways may also provide a potential mechanism for tissue selectivity, although its clinical relevance remains unclear.