Strong Inhibitory Effect, Low Cytotoxicity and High Plasma Stability of Steroidal Inhibitors of N-Methyl-D-Aspartate Receptors With C-3 Amide Structural Motif

Herein, we report the synthesis, structure-activity relationship study, and biological evaluation of neurosteroid inhibitors of N-methyl-D-aspartate receptors (NMDARs) receptors that employ an amide structural motif, relative to pregnanolone glutamate (PAG) – a compound with neuroprotective properties. All compounds were found to be more potent NMDAR inhibitors (IC50 values varying from 1.4 to 21.7 μM) than PAG (IC50 = 51.7 μM). Selected compound 6 was evaluated for its NMDAR subtype selectivity and its ability to inhibit AMPAR/GABAR responses. Compound 6 inhibits the NMDARs (8.3 receptors (8.3 ± 2.1 μM) more strongly than it does at the GABAR and AMPARs (17.0 receptors (17.0 ± 0.2 μM and 276.4 ± 178.7 μM, respectively). In addition, compound 6 (10 μM) decreases the frequency of action potentials recorded in cultured hippocampal neurons. Next, compounds 3, 5–7, 9, and 10 were not associated with mitotoxicity, hepatotoxicity nor ROS induction. Lastly, we were able to show that all compounds have improved rat and human plasma stability over PAG.

In general, the major structural requirements for neurosteroid NMDAR-inhibitors have been established by several authors: (Irwin et al., 1994;Park-Chung et al., 1997;Weaver et al., 2000) detailing that the inhibitory effect is dependent upon both the 3α-and 5β-stereochemistry of the pregnane skeleton, in association with a negatively charged moiety at C-3, preferably a sulfate or hemiester group. As a part of our continuing interest in the structure-activity relationship (SAR) study of neurosteroid modulators of NMDARs, we have also recently reported on several new structural modifications that generate potent inhibitors of NMDARs: (i) the negatively charged C-3 substituent can be substituted by a positively charged moiety or zwitterion group (e.g., glutamic acid ester); (Borovska et al., 2012) (ii) the C-17 acetyl moiety of the pregnane skeleton is not essential for the inhibitory effect and can be substituted by non-polar substituents (methyl, ethyl, etc.), including full elimination of the C-17 moiety altogether, e.g., androstane 3-sulfate ( Figure 1B); (Kudova et al., 2015) (iii) the steroidal D-ring is not essential for inhibitory effect and can be fully or partially degraded (Slavikova et al., 2016).
Pregnanolone glutamate (PAG, Figure 1C) (Borovska et al., 2012) was synthesized as the synthetic analog of PAS, and its neuroprotective effect was assessed in several biological models in vivo, wherein it was found that: (i) PAG does not induce psychotomimetic symptoms (such as hyperlocomotion and sensorimotor gating deficit), and it actually reduced excitotoxic damage of brain tissue and subsequent behavioral impairment in rats; (Rambousek et al., 2011) (ii) PAG significantly ameliorated neuronal damage in the dentate gyrus and subiculum, and improved behavioral performance in active allothetic place avoidance tasks (AAPA, also known as the carousel maze) after bilateral NMDA-induced lesions to the hippocampus; (Rambousek et al., 2011) (iii) PAG displayed anxiolytic-like and antidepressant-like properties in an elevated plus maze (EPM) and forced swimming models; (Holubova et al., 2014) (iv) PAG displayed a neuroprotective effect in a focal cerebral ischemia model that was induced by endothelin-1 in immature rats (Kleteckova et al., 2014).
From these literature reports, we concluded that the glutamic acid ester moiety could be a promising structural motif at the C-3 position for the development of drug-like molecules. However, we anticipated that the labile ester bond connecting the glutamate moiety of PAG to the steroid skeleton would be susceptible to plasmatic degradation by esterases. Thus, we thought to replace the ester linkage with the more robust amide bond ( Figure 1D). This structural modification was designed to reduce the metabolic liability of new analogs and alter their solubility and permeability profiles. Moreover, the substitution of an ester bond by an amide has already been established as an effective isosteric approach that affords stable analogs while maintaining biological activity (Meanwell, 2011). Based upon our previous SAR on non-polar D-ring modifications, (Kudova et al., 2015) we also decided to incorporate this structural motif into our new library of NMDAR inhibitors. Hence, a series of amide-substituted PAG-like compounds was proposed (compounds 1-12, Figure 2).
From our previous studies, (Borovska et al., 2012;Kudova et al., 2014Kudova et al., , 2015Adla et al., 2017) we proposed that amide PAG-like compounds (1-12) offer a promising target for synthesis, further development and complex evaluation of physicochemical and biological properties including the screening on their stability in plasma. As such, compounds 1-12 were evaluated on HEK293 cells transfected with plasmids encoding GluN1-1a/GluN2B/GFP genes to elucidate the SAR of the amide, amino, and Boc-protected amino moiety in various positions. Considering the recommended guidelines for early stage development of new potent compounds, we have introduced mitotoxicity and hepatotoxicity screening on HepG2 cells as a primary tool to rank our compounds during lead optimization and to select the most promising candidate (Gerets et al., 2009;Van den Hof et al., 2013). From these results, we were able to identify lead compound 6, which we subsequently evaluated on recombinant GluN1/GluN2A-D receptors and native NMDARs α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA)/kainate receptors (AMPAR), and gamma-aminobutyric acid (GABA) receptors (GABARs) expressed in hippocampal neurons. Finally, we assessed the stability of PAG-like compounds vs. PAG in rat and human plasma to demonstrate their likely clinical advantage.

Chemistry
Compound 13 was prepared according to the literature:  in brief, commercially available 3β-hydroxy-5β-androstane was treated with phthalimide  and triphenylphosphine, followed by deprotection of the amino group in hydrazine hydrate, to give 3α-amino derivative 13. Compounds 1 and 2 were prepared according to the literature  by treatment of compound 13 with the monoethyl ester of oxalic acid and methyl 3-chloro-3-oxopropionate, respectively, followed by basic hydrolysis. Analogously, compounds 3 and 4 were prepared by treatment of compound 13 with methyl 4-chloro-4-oxobutyrate and methyl 5-chloro-5-oxovalerate, respectively, affording compounds 14 in 79% and 15 in 83% yield. Finally, alkaline hydrolysis of compounds 14 and 15 afforded compounds 3 in 76% and 4 in 32% yield, respectively (Figure 3).
The coupling of compound 13 with Boc-L-glutamic acid 5-benzyl ester and Boc-L-aspartic acid 4-benzyl ester, respectively, gave compounds 16 and 17 in 99% yield (Figure 4). Deprotection of the benzyl ether protecting group was achieved by hydrogenation catalyzed by palladium on carbon (compound 5, 65% yield and 7, 98% yield). Then, treatment of Boc-protected aspartate 5 and glutamate 7 with trifluoroacetic acid gave desired products 6 and 8 in 91% and 92% yield, respectively .
Compounds 10 and 12 were prepared analogously to the synthesis of compounds 6 and 8, using Boc-L-glutamic acid 1-benzyl ester and Boc-L-aspartic acid 1-benzyl ester for coupling reaction with compound 13 ( Figure 5). As such, compounds 18 and 19 were prepared in 93 and 92% yield, respectively. Then, deprotection of the benzyl ether protecting group was achieved by hydrogenation catalyzed by palladium on carbon (compound 9, 93% yield and 11, 96% yield). Finally, treatment of Boc-protected glutamates 9 and 11 with trifluoroacetic acid gave desired products 10 and 12 in 85% and 96% yield, respectively.

Biological Activity
To investigate the activity of PAG and its amide analogs (1-12) on NMDARs, cDNA encoding for the rat GluN1-1a and GluN2B subunits were co-transfected into HEK293 cells. As the   amphipathic character of compounds 1-12 is similar to our previously published D-modified steroids (Kudova et al., 2015), we have used an identical approach for the assessment of the obtained data and calculation of the IC 50 values. In brief, the IC 50 for the newly synthesized steroids was determined from a single dose of the steroid using the following formula: where I I is the relative degree of inhibition, [compound] is the steroid concentration used, and h stands for the Hill coefficient (fixed at 1.2). The IC 50 value was determined for a minimum of two steroid doses, differing twofold in the concentration range. If the difference in the IC 50 values was <10%, then the mean steroid IC 50 was calculated from the steroid concentration most proximal to that inducing 50% inhibition. If the difference in the IC 50 values determined for two steroid doses was >10%, then the dose-response analysis was determined at lower steroid concentrations to reach the formal criterion. In accordance with previous results, IC 50 calculations were made assuming 100% inhibition at saturating steroid concentration (Petrovic et al., 2005;Borovska et al., 2012). Compounds 1-4, bearing an alkylcarboxylic moiety of various chain lengths on the C-3 position, displayed relatively higher IC 50 values (8.9-21.7 µM) as compared with the C-3 glutamate and aspartate derivatives 5-11 (1.4-5.4 µM). We also found that no significant difference in IC 50 values was obtained for analogs with the transposed locations of the N-Boc and NH 2 groups on the glutamate and aspartate substituent (3 vs. 2 for compounds 6 vs. 10 and 5 vs. 9). Lastly, we found that the introduction of a Boc-protecting group did not significantly alter the inhibitory effect of our compounds. Taken together, we can conclude that amide structural motif positively affects the ability of compounds 1-12 to modulate NMDAR currents.
The Computational Estimate of Thermodynamic Properties of Compounds 1-12 The relevant physicochemical properties (Faassen et al., 2003) of the studied steroidal inhibitors were estimated by quantum mechanics computational methods and by a physicochemical properties predictor. The computational results are summarized in Table 2. We investigated the lipophilic qualities and solvation free energy ( G solv ) of the inhibitors, as these properties are inherent characteristics of neuroactive compounds and influence their interactions with NMDAR (Kudova et al., 2015). The G solv values describe the behavior of single-molecules in water and in n-octanol, which acts as a proxy for the membrane environment. Lipophilicity of the compounds was estimated by the logP and logD coefficients.
The results in Table 2 show that the G solv , logP and logD values of the investigated compounds are in the similar range to the earlier studied inhibitors (Kudova et al., 2015;Slavikova et al., 2016). Compounds 1-4 have more simple structures as compared to compounds 5-12 and evince a poorer inhibitory effect relative to the other studied compounds. For example, compound 1 with the simplest structure among the studied inhibitors (1-12) has the least inhibitory activity. The action of neuroactive compounds at the NMDAR is induced by a combination of many factors which determine the resulting inhibitory effect. Steroids interact with the entire non-polar inner surface of the NMDAR channel, mostly through attractive van der Waals interactions which compete with the repulsive effects (e.g., the partial desolvation at the site of the interaction, or the behavior of the charged and polar substituents of the steroid in the non-polar surroundings . Previously, we demonstrated that the behavior of steroidal substances can be significantly influenced by conformational change inside the narrow region of the NMDAR channel, where the inhibitory effect occurs within proximity of     the threonine ring through non-specific interactions . Compounds 5, 7, 9, 11 (N-Boc protected) and 6, 8, 10, 12 (free NH 2 group) have similar inhibitory effects, as well as similar G solv , logP and logD values. It is evident that N-Boc addition does not improve the inhibitory effect. For instance, it can be caused by steric effect.

In vitro Cytotoxicity of Compounds 1-12
In the present study, HepG2 cells were exposed to compounds 1-12 for 72 h. Then, the cell viability (XTT assay) and reactive oxygen species (ROS) induction were evaluated. The results of cytotoxic effect for PAG-like compounds (1-12) are summarized in Table 3. The hepatic effect of compounds 1-12 was compared with amiodarone (4.9 ± 0.2 µM) and nimesulide (2.2 ± 0.3 µM) -marketed drugs which cause hepatotoxicity. Notably, compounds 3, 5-7, 9, and 10 displayed no adverse hepatic effect (>200 µM), whereas compounds PAG and 8 had a comparable hepatic effect (IC 50 = 62 and 65 µM, respectively). Compound 4, bearing a five-carbon chain, had a strong hepatic effect (IC 50 = 4.0 µM). Moreover, this compound showed Glu/Gal index higher than 3, which indicates its potential mitochondrial toxicity (Figures 6A,B). Interestingly, the adverse hepatic effect seen in compound 8 was abated by the inclusion of a Boc-protecting group on the glutamate moiety, as is reflected in matched compounds 7 (IC 50 > 200 µM). Nevertheless, we strongly caution that a glutamate moiety at C-3 should be regarded as a red flag for cytotoxicity, warranting further research.
Contrary to the glutamate moiety, the aspartate moiety has been demonstrated as an "allowed" structural feature. Indeed, compounds 6 and 10, as well as their Boc-protected analogs 5 and 9, showed no adverse hepatic effect (>200 µM). Furthermore, compound 3, which has an analogous four-carbon moiety at C-3, also did not display any adverse hepatic effect (>200 µM). Therefore, we have established the aspartate moiety as a pharmacophore of the C-3 moiety to be further researched.
Decrease in cell viability was accompanied by concentrationdependent ROS induction ( Figure 6C and Table 3). We hypothesize that the ROS mediated cytotoxicity can be associated with the type of side chain. Glutamate moiety, the source glutamate, has been reported to induce lipid peroxidation, decrease reduced glutathione and increase activities of catalase and superoxide dismutase in the liver of animals (Onyema et al., 2006). The hemioxalate moiety has been connected with lipid peroxidation (Sevam and Bijikurien, 1987). Shortening of chain from glutamate to aspartate, and extension of chain from oxalate to malonate did lead to loss of both ROS and cytotoxicity increase without decrease of inhibitory activity.

The Inhibitory Effect of Compound 6 on GluN1/GluN2A-D Receptors
Considering the effect of compounds 1-12 on current responses of GluN1/GluN2B receptors and their cytotoxicity profile, compound 6 (Figure 2) emerged as the lead structure and it was chosen for further biological evaluation. Comparison of the IC 50 values of the steroid 6 at GluN1/GluN2A-D receptors shows no significant differences (one-way ANOVA; P > 0.05) (Figure 7 and Table 4) between NMDAR subtypes. This low subunit selectivity is strikingly different from previously published IC 50 dependency of naturally occurring neurosteroid PAS on NMDAR subunit composition (Petrovic et al., 2005). PAS was found to inhibit GluN1/GluN2A-B (IC 50 = 50.0 and 44.4 µM, respectively) receptors with lower potency than GluN1/GluN2C-D receptors (IC 50 = 25.6 and 30.1 µM, respectively) (Petrovic et al., 2005). On the other hand, similar effect of compound 6 on NMDAR subunit dependency was found when compared to 17β-methyl analog of pregnanolone sulfate -17β-methyl-5β-androstane 3α-yl-sulfate, which afforded comparable potency to all GluN1/GluN2A-D receptors (IC 50 values varying from 0.4 to 0.7 µM). The reason for this phenomenon remains unknown.
The Effect of Compound 6 on Native NMDARs, AMPARs, and GABARs We have shown earlier that certain steroids preferentially inhibit tonically over phasically activated NMDARs with receptors with implications for synaptic transmission and excitotoxicity . Figure 8 shows the analysis of the effect of compound 6 on the peak and steady-state response induced in cultured hippocampl neurons by fast application of NMDA, AMPA, and GABA. The data indicate that compound 6 present continuously was a more efficient inhibitor of tonically activated receptors at the steady-state of the response induced by coapplication of the steroid with NMDA (100 µM) or GABA (5 µM) than of the peak response (Paired t-test; P < 0.001 and P = 0.009 for NMDARs and GABARs, respectively) ( Figure 8B). The amplitude of responses induced by AMPA (100 µM) and recorded in the presence of cyclothiazide (10 µM) to block the receptor desensitization were only little affected by the steroid and no differences in the steroid effect at the peak and steady-state response were observed ( Figure 8B).
We have also found differences in the potency of compound 6 to inhibit native NMDAR and GABAR (Figure 9). Concentration-response analysis of the inhibitory effect of compound 6 at native NMDAR responses showed IC 50 = 8.3 ± 2.1 µM and h = 1.3 ± 0.2 (n = 6). In contrast, compound 6 inhibits the responses to 5 µM GABA with IC 50 = 17.0 ± 0.2 µM and h = 1.2 ± 0.1 (n = 6). The differences  50 and logHill values (one-way ANOVA); P ≤ 0.05 was used for the determination of significance. in IC 50 of compound 6 at NMDARs and GABARs was significantly different (P < 0.001; Student's t-test). We have reported an opposite effect for other neurosteroid analogs where steroids and steroid-like compounds inhibited the GABARs with the same or higher affinity than NMDARs (Kudova et al., 2015;Slavikova et al., 2016). AMPAR responses were only little affected by compound 6 and estimates of IC 50 from the steroid effect at a single dose (30 µM) and with a Hill coefficient fixed at 1.2 indicate value of 276.4 ± 178.7 µM (n = 6). Such a weak inhibitory steroid effect on AMPARs was not observed earlier for PAS and its synthetic analogs (Kudova et al., 2015;Vyklicky et al., 2015;Slavikova et al., 2016).

The Effect of Compound 6 on the Frequency of Action Potentials
Given that steroids typically have multiple effects at the level of postsynaptic receptors, it is important to examine the net influence of a given compound on network activity. Compound 6, for example, potently inhibits NMDA receptors, but also shows inhibitory activity at GABA receptors. Therefore, we examined the effect of compound 6 (10 µM) on action potential firing frequency of neurons grown in primary hippocampal culture. Extracellular solution contained physiological concentrations of Ca 2+ (2 mM), Mg 2+ (1 mM) and glycine (10 µM). As shown in Figure 10, compound 6 consistently and robustly decreased the action potential frequency to 35 ± 8% of control values (Paired ttest, P < 0.001), and this effect was reversible. The observation that compound 6 has an overall inhibitory effect on network activity is consistent with its higher inhibitory potency at NMDA receptors compared to GABA receptors.

Plasma Stability
As discussed above, PAG had not been expected to afford sufficient metabolic stability, as the glutamate moiety could  undergo hydrolysis by carboxylesterase as described in the literature (Schötller and Krisch, 1974). Indeed, we have demonstrated that all compounds 1-11 have improved stability in rat and human plasma compared to PAG (Table 5) which confirmed our hypothesis of plasma-stable isosteric effect of amide structural modification.

CONCLUSION
In this study, we have examined a structural feature, the C-3 amide bond, which exhibits an inhibitory effect on NMDARs. Compounds 1-12 were evaluated for their ability to modulate NMDARs, wherein we demonstrated that the C-3 amide bond of these compounds is an allowed structural modification for maintaining inhibitory activity. Moreover, we have also shown that these new ligands are more potent than the endogenous ligand -PAS. In addition, we found that aspartate structural modification did not lead to an adverse hepatic effect while giving rise to improved plasma stability. Taken together, this new structural motif offers new prospects for the further modification and optimization of the pharmacological and pharmacokinetic properties of these neuroactive steroids.

EXPERIMENTAL SECTION Chemistry
Melting points were determined on a micromelting point apparatus Hund/Wetzlar (Germany) and are uncorrected. Optical rotations were measured in chloroform using an Autopol IV (Rudolf Research Analytical, Flanders, NJ, United States).

Preparation of Structures
The studied compounds were manually built in PyMOL (version 1.5.0.4.) (De Lano and Lam, 2010) using the geometry of the molecule taken from the crystal structure (3CAV PDB code), (Faucher et al., 2008) and were relaxed by the RI-DFT/B-LYP/SVP method with the Turbomole program (version 6.1) (Ahlrichs et al., 1989). The empirical dispersion correction (D) (Jurecka et al., 2007) and COSMO continuum solvation model (Klamt and Schuurmann, 1993) were applied on the gradient optimization. The most stable local minima of the compounds were generated by the molecular dynamics simulation with the general AMBER empirical force field (the simulation was run 30 ns; the constant temperature was 400 K) (Wang et al., 2004). The AMBER 14 MD package was used (Case et al., 2014). The partial charges of the molecules were calculated using the RESP procedure (Bayly et al., 1993)

Computational Methods
The solvation free energy ( G solv ) of the compounds was calculated in the SMD continuum solvation model (Marenich et al., 2009) (the transfer from vacuum to water and from n-octanol to water) at the HF/6-31G * level with the Gaussian program (version 09) (Frisch et al., 2009). In this method, the single-point energies were computed at the identical molecular geometry both for the transfer from vacuum to water and from n-octanol to water. The partition coefficient (P) is defined as the ratio of concentrations of a neutral solute in n-octanol and water, and it represents the solute lipophilicity. It is usually reported as common logarithm: The calculated logP was obtained via equation logP = G ow / −RTln (10) , where G ow is transfer free energy, R is molar gas constant and T is temperature (298.15 K) (Kolar et al., 2013;Bannan et al., 2016).
The G ow was calculated on the basis of the change in the molecular conformation related to the transfer between n-octanol and water. The compounds were optimized at the M06-2X/6-31G * level in SMD with the Gaussian, and G ow was expressed as the difference between the total energies in water and in n-octanol, because this energy includes the internal energy of the molecule (Kolar et al., 2013).
The distribution-coefficient (D), which is also presented as common logarithm, takes into account both neutral and ionized form of the solute in both phases and is used for estimation of lipophilicity of ionizable species (Kah and Brown, 2008). logD = log c ionized,octanol + c neutral,octanol c ionized,water + c neutral,water The logD values were predicted at pH = 7.4, which is the physiological pH of blood serum, using the MarvinSketch program (ChemAxon, 2015).

Biological Activity
Electrophysiological experiments were performed on HEK293 cells transfected with plasmids encoding rat GluN1-1a/GluN2A-D and GFP genes as described previously, (Petrovic et al., 2005;Cais et al., 2008;Korinek et al., 2010) or on primary hippocampal neurons prepared from P0-P1 rats, plated on collagen/poly-D-lysine-coated glass coverslips at a density of ∼50,000 cells/cm 2 , and grown in Neurobasal A medium with B27 Supplement (Gibco). Agonist-induced responses were voltage-clamped at a holding potential of −60 mV. Whole-cell voltage clamp recordings were made with a patch-clamp amplifier (Axopatch 200B; Axon Instruments Inc., Foster City, CA, United States) after a serial resistance (<10 M ) and capacitance compensation of 80-90%. For the application of test and control solutions, a microprocessor controlled multibarrel fast-perfusion system was used, with a time constant of solution exchange around cells of ∼10 ms. Agonist-induced responses were low-pass filtered at 2 kHz, digitally sampled at 5 kHz, and analyzed with pClamp software version 10.5 (Molecular Devices). Patch pipettes (3-5 M ) pulled from borosilicate glass were filled with Cs + -based intracellular solution (ICS) containing the following (in mM): 120 gluconic acid, 15 CsCl, 10 BAPTA, 10 HEPES, 3 MgCl 2 , 1 CaCl 2 , and 2 ATP-Mg salt (pH-adjusted to 7.2 with CsOH). The extracellular solution (ECS) contained the following (in mM): 160 NaCl, 2.5 KCl, 10 HEPES, 10 glucose, 0.7 CaCl 2 , 0.5 EDTA, and 0.01 glycine (pH-adjusted to 7.3 with NaOH). NMDAR responses were induced by 1 mM glutamate (in recombinant receptors) and 100 µM NMDA (native receptors). The ECS used for native receptors had the same composition as the ECS used for recombinant receptors. Native NMDAR and AMPAR recordings were performed on primary hippocampal mass-culture neurons at 8 days in vitro. For NMDAR recordings, ECS contained additionally 10 µM CNQX, 10 µM bicuculline and 0.5 µM TTX. AMPAR responses were induced by 100 µM AMPA and the ECS contained additionally 50 µM D-AP5, 10 µM bicuculline, 0.5 µM TTX, and 10 µM cyclothiazide. GABA receptor responses were induced in primary hippocampal mass-culture neurons at 12 days in vitro by 5 µM GABA and the ECS contained additionally 50 µM D-AP5, 10 µM CNQX, and 0.5 µM TTX. Action potentials were recorded in the current-clamp mode from primary hippocampal mass-culture neurons at 14-15 days in vitro. The ICS contained (in mM): 125 gluconic acid, 15 KCl, 5 EGTA, 10 HEPES, 0.5 CaCl 2 , 2 ATP-Mg salt, 0.3 GTP-Na salt, and 10 creatine phosphate (pH adjusted to 7.2 with KOH). The ECS contained (in mM): 160 NaCl, 2.5 KCl, 10 HEPES, 10 glucose, 2 CaCl 2 , 1 MgCl 2 , and 0.01 glycine (pH-adjusted to 7.3 with NaOH). Junction potential correction was applied post hoc. Steroid solutions were prepared fresh as a stock solution of either 5 or 20 mM in dimethyl sulfoxide (DMSO) before each experiment (1% DMSO final concentration). The same concentration of DMSO was added in all extracellular solutions. Experiments were performed at room temperature (21-25 • C).

Cytotoxicity and Mitotoxicity Evaluation
The cytotoxicity of the tested compounds was assessed with use of XTT cell proliferation kit II (Roche Diagnostics GmbH, Mannheim, Germany) according to the manufacturer's instructions. Briefly HepG2 cells were seeded in a 96-well plate at a density of 22,000 cells per well. After 24 h, the tested neurosteroids in a concentration range from 0.5 to 200 µM were added to the culture media and incubated for 72 h before the XTT dye was added. Mitochondrial toxicity was assessed in HepG2 cells using glucose and galactose conditioned media and XTT assay as a read out for cell viability. In brief, cells were seeded onto 96 well, clear bottom tissue culture plates at a density of 22,000 cells/well and left overnight to attach. Six hours prior to the neurosteroids treatment the media was replaced with 100 µL DMEM media containing either glucose (25 mM glucose; 1 mM pyruvate; 2 mM glutamine, 10% FBS) or galactose (10 mM galactose, 1 mM pyruvate, 6 mM glutamine, 10% FBS). After that, the tested neurosteroids in a concentration range from 0.5 to 200 µM were added in the fresh appropriate assay medium (glucose or galactose) and incubated for 24 h before the XTT dye was added. The XTT absorbance was recorded at 495 nm after 1 h incubation with the dye. Compounds were dissolved in DMSO to produce the stock solutions at the concentration of 10 mM, with exception of PAG, 1, 6 (5 mM), and 11 (2.5 mM). Results were expressed as percentages of change in viability compared to appropriate DMSO control. IC 50 values were determined by GraphPad Prism version 5.00 for Windows (GraphPad Software, La Jolla, CA, United States). All experiments were done in triplicates in three independent experiments to check the reproducibility.

Detection of Reactive Oxygen Species
The effect of neurosteroids on oxidative stress was determined using general oxidative stress indicator CM-H2DCFDA (

Plasma Stability
Compounds were prepared as 1.6 mg/mL stock solutions in methanol. Then, 37.8 µL of stock solution was added to 1400 µL of human plasma, maintained at 37 • C. Aliquots (50 µL) withdrawn at 0, 1, 4, 8, and 24 h were analyzed by HPLC. An aliquot of plasma was extracted with methanol (450 µL) containing internal standard of deuterated pregnanolone glutamate (Rambousek et al., 2011) (5.3 µg/mL) and the solution was vortexed (20 s) and centrifuged at 13500 rpm for 10 min. The supernatant was transferred to an autosampler vial and 1 µL was injected onto LC-MS system. The samples were analyzed with an Agilent 1260 HPLC (Agilent Technologies) coupled to ESI-TOF Agilent 6530 (Agilent Technologies) with Agilent Jet Stream Technology. Samples were separated on a Waters ACQUITY UPLC CSH Phenyl-Hexyl (100 × 2.1, 130 Å, 1.7 µm) at a flow rate 0.3 ml/min. The concentration of mobile phase B (0.1% formic acid in acetonitrile) was gradually increased from 10 to 100% in mobile phase A (0.1% formic acid in water) over 7 min. The mass spectrometry instrument was operated in a negative ion mode with a voltage of +3.00 kV applied to the capillary. The temperature, the flow rate of the nitrogen drying gas, the pressure of the nitrogen nebulizing gas, and the flow rate of the sheath gas were set at 325 • C, 10 l/min, 40 psi, 390 • C, and 11 l/min, respectively. Results are represented as a percentage of a compound remaining in spiked human plasma.

General Procedure II -Hydrogenolysis of Benzyl Ether Protecting Group
Boc-Bn-conjugate with steroid was dissolved in ethanol (8 mL per 100 mg of steroid-conjugate). To this, Pd/C (10%, 10 mg per 100 mg of steroid-conjugate) was added. The reaction mixture was hydrogenated under slight pressure at room temperature for 10 h. Then, the reaction mixture was filtered through a short column of silica gel to remove the catalyst by washing with chloroform. Solvents were removed under vacuo.

General Procedure III -Boc-Group Deprotection
Trifluoroacetic acid (4 mL per 150 mg of Boc-protected compound) was added dropwise at room temperature to an ice-cooled solution of Boc-protected compound in dichloromethane (10 mL per 150 mg of Boc-protected compound). The reaction mixture was allowed to stir at room temperature until starting material was consumed (TLC monitoring). Then, the excess of dichloromethane and TFA was removed by flushing out with nitrogen. The resultant brownish oil was dissolved in pyridine/MeOH (4.5 mL/0.5 mL) and the solution was added dropwise to ice cold water (50 mL). The aqueous solution was kept in the fridge at 5 • C for 24 h. The precipitate was filtered and dried in high vacuum to give the desired compound.