Efficient Modulation of γ-Aminobutyric Acid Type A Receptors by Piperine Derivatives

Piperine activates TRPV1 (transient receptor potential vanilloid type 1 receptor) receptors and modulates γ-aminobutyric acid type A receptors (GABAAR). We have synthesized a library of 76 piperine analogues and analyzed their effects on GABAAR by means of a two-microelectrode voltage-clamp technique. GABAAR were expressed in Xenopus laevis oocytes. Structure–activity relationships (SARs) were established to identify structural elements essential for efficiency and potency. Efficiency of piperine derivatives was significantly increased by exchanging the piperidine moiety with either N,N-dipropyl, N,N-diisopropyl, N,N-dibutyl, p-methylpiperidine, or N,N-bis(trifluoroethyl) groups. Potency was enhanced by replacing the piperidine moiety by N,N-dibutyl, N,N-diisobutyl, or N,N-bistrifluoroethyl groups. Linker modifications did not substantially enhance the effect on GABAAR. Compound 23 [(2E,4E)-5-(1,3-benzodioxol-5-yl)-N,N-dipropyl-2,4-pentadienamide] induced the strongest modulation of GABAA (maximal GABA-induced chloride current modulation (IGABA-max = 1673% ± 146%, EC50 = 51.7 ± 9.5 μM), while 25 [(2E,4E)-5-(1,3-benzodioxol-5-yl)-N,N-dibutyl-2,4-pentadienamide] displayed the highest potency (EC50 = 13.8 ± 1.8 μM, IGABA-max = 760% ± 47%). Compound 23 induced significantly stronger anxiolysis in mice than piperine and thus may serve as a starting point for developing novel GABAAR modulators.


■ INTRODUCTION
γ-Aminobutyric acid type A (GABA A ) receptors are the major inhibitory neurotransmitter receptors in mammalian brain. 1−3 GABA A receptors belong to the superfamily of Cys loop ligandgated ion channels. Five receptor subunits form a central chloride-conducting pore. 4−6 Nineteen genes encoding different subunits have been discovered in the human genome, comprising α 1−6 , β 1−3 , γ 1−3 , δ, ε, θ, π, and ρ 1−3 . 7,8 Different subunit combinations may theoretically form a vast number of receptor subtypes with different pharmacological properties (see ref 9 for review). There is consensus that the most abundantly occurring receptor subtype is formed of two α 1 , two β 2 , and one γ 2 subunits (α 1 β 2 γ 2 receptor). 10−12 Drugs that enhance chloride currents through GABA A receptors play an important role in the treatment of general anxiety, panic disorders, sleep disturbances, and epilepsy. 13−17 The most widely used benzodiazepines induce, however, a variety of side effects including dependence, unwanted sedation, and amnesia, complicating their long-term use. 18−20 Hence, there is high unmet medical need for GABA A receptor modulators lacking these unwanted effects.
Besides their modulation by clinically used drugs such as benzodiazepines, barbiturates, neurosteroids, and anesthetics, 3,9,15,21−27 GABA A receptors are modulated by numerous natural products that may provide lead structures for drug development. 28−30 In this context, we 31 and others 32 have reported that piperine (1-piperoylpiperidine), the pungent component of several pepper species and activator of transient receptor potential vanilloid type 1 receptor (TRPV1), 33 also modulates GABA A receptors. We could establish that replacing the piperidine ring of piperine by a N,N-diisobutyl residue, resulting in (2E,4E)-5-(1,3-benzodioxol-5-yl))-N,N-diisobutyl-2,4-pentadienamide (SCT-66; 34 referred to as 24 in this work), diminishes the interaction with TRPV1 receptors. Furthermore, 24 enhanced chloride currents through GABA A receptors more potently and more efficiently than piperine and displayed, concordantly, a stronger anxiolytic action. 34 Based on these findings, a library of piperine derivatives was synthesized and investigated with respect to modulation of α 1 β 2 γ 2S GABA A receptors expressed in Xenopus laevis oocytes. Within this study we emphasized modifications at the amide functionality and on the diene motif within piperine in order to enhance the modulatory potential of analogue structures. Their structure−activity relationship on GABA A receptors was analyzed by establishing binary classification models.

■ RESULTS AND DISCUSSION
Modification of Amide Nitrogen. Starting with piperine as lead structure from prior biological assessment, the molecule can be structurally divided into three parts: the 1,3benzodioxole or aromatic function, the olefinic linker region comprising four carbon atoms, and the amide function natively constituted by a piperidine ring (Figure 1). Within this study, we investigated modifications at the amide group as well as in the linker region.
Modifications at the amide function were implemented in a straightforward fashion (Scheme 1). Piperic acid amides (1−16, 20−23, and 25−43) were synthesized by treating piperic acid chloride with the corresponding amine in the presence of triethylamine in tetrahydrofuran (THF). Compounds 17 and 18 were prepared in the same way from benzodioxolyl acryloyl chloride. Treatment of piperine with Lawesson's reagent 35 gave thioamide 44. Reduction of the carbonyl group of piperine with lithium aluminum hydride afforded unsaturated amine 45 (Scheme 2).
N,N-Dibutyl-Substituted Compound 25 Displays the Highest Potency. Figure 2D illustrates I GABA modulation by the most potent N-substituted piperine derivative (EC 50 for 25, 13.8 ± 1.8 μM < EC 50 for 43, 23.1 ± 3.3 μM < EC 50 for piperine, 52.4 ± 9.4 μM 31 ). Based on the modifications at the amide group, it can be concluded that installation of noncyclic substituents bearing 3−4 carbons each at the tertiary amide improves both efficacy and potency of the analogue compounds.
Rigidification of the Linker Region Has No Significant Effect on I GABA Modulation. The influence of linker rigidity on I GABA modulation was studied by means of a library comprising 32 linker derivatives. According to Zaugg et al. 31 and Pedersen et al., 32 a carbon chain containing at least four carbons, a conjugated double bond adjacent to the amide group, and a bulky amine moiety seem to facilitate efficient receptor binding and/or I GABA modulation.
Based on previous reports by Zaugg et al., 31 we hypothesized that rigidification of the linker part of the structure may beneficially affect biological activity. 31 This assessment was based in particular on a decrease in modulatory capacity when partially saturated linkers were installed or when structural flexibility was increased by extending the linker length.
Three major structural modifications were envisaged (Scheme 1). (i) Replacement of the linker by an aryl ring (phenyl, heteroaryl, naphthyl): in this arrangement, both alkene groups of the diene system of the linker would be integrated into the rigid aromatic core. (ii) Integration of one linker double bond into a naphthyl ring: this compound class was expected to render more flexibility but still adopt a more rigidified system compared to the piperine diene structure; moreover, arrangement should allow for different angles of the aryl core relative to the amide anchoring group depending on the substitution site at the naphthyl system. (iii) "Ring closure" of the diene motif with the aryl part, consequently generating a carboxylate-substituted naphthyl lead structure: in this arrangement the double bond adopts a bent geometry, and again different angles of the aryl and amide parts can be obtained depending on the substitution site.
Iridium-catalyzed direct borylation 39 of naphtho[2,3-d]dioxole allowed direct access to the 6-position of the naphtho[2,3-d]dioxole core. Boronic acid ester 68a obtained in this step was converted into the corresponding bromide 40 Figure 1. Piperine molecule can be structurally divided into three moieties: the 1,3-benzodioxole or aromatic function, the linker region comprising four carbon atoms, and the amide function natively constituted by a piperidine ring.
68b and coupled under standard Heck cross-coupling conditions to afford acrylate 68c (Scheme 4). The methyl ester was hydrolyzed, and acid 68d was converted into products 68−70 (Scheme 4).
These data support the previous observation that when the cyclic piperidine residue is replaced by N,N-dialkyl moieties such as N,N-dipropyl (23), N,N-diisopropyl (24), 34 or N,Ndibutyl (25), efficiency and potency can be significantly enhanced. However, while 24 34 lost its ability to distinguish between the β-subunit isoforms, preferential modulation of β 2/3 receptors by 23 was comparable to piperine, and it was even more pronounced for 25 (see Figure 4 B,D and Tables 4−6). Thus, 23 and 25 displaycompared to classical GABA A receptor modulators such as benzodiazepinesa distinct subunit selectivity profile. Unlike benzodiazepines, 23 and 25 also modulate GABA A receptors containing α 4 subunits with high efficiency and are not dependent on the presence of a γ 2S subunit (data not shown). Whether this subunit selectivity profile has any pharmacological relevance has to be clarified in further studies. Structure−Activity Relationships: General Trends. When the whole data set was analyzed, several distinct SARs could be deduced. They are mostly related to the substitution pattern at the amide nitrogen atom, as this was the main point of variation in the data set. Thus, concerning N,N-dialkylsubstituted amides, there is evidence that I GABA enhancement is related in a nonlinear (parabolic) function to the number of carbon atoms ( Figure 5), with the optimum being dipropyl (23). This type of parabolic relationship is quite common, especially when it refers to a parameter that is linked to lipophilicity of the compounds and activity data obtained in a cellular assay. It has, for example, also been observed for a series of capsaicin analogues with respect to their TRPV1 activation. 43 Interestingly, whether the alkyl chains are linear or branched does not reverse the order: 20 (dimethyl) < 21 (diethyl) < 23 (dipropyl)/22 (diisopropyl) < 25 (dibutyl)/24 34 (diisobutyl) < 26 (dihexyl)/27 (dicyclohexyl). With respect to compounds where the amide nitrogen atom is part of a ring, methylpiperines 33, 34, and 35 induced the strongest I GABA potentiation, followed by azepane amide 28 and piperine. Interestingly, the dimethylpiperine 38 was comparably active to the parent compound. Introduction of a second heteroatom Replacement of the tertiary nitrogen atom for a secondary one, irrespective of alkyl or aryl substitution, led to a complete loss of activity (aryl-substituted N, 1−3, 5−7, 9, and 10; alkylsubstituted N, 4, 8, and 11−16). Reducing the H-bond acceptor strength of the amide by synthesizing the respective thioamide (44) abolished the modulatory activity. Reduction of the amide to the analogous amine changed the profile of the compound from potentiation (piperine at 100 μM, 226% ± 26%) 31 to inactive (45 at 100 μM, −16% ± 14%; Table 1).
With respect to the linker region, shortening the distance by removing one vinylene unit significantly reduced I GABA enhancement (piperine vs 17 and 22 vs 18). All the other modifications, such as rigidification by inserting benzene, thiophene, or naphthalene moieties, reduced I GABA potentiation by at least a factor of 5 compared to 23. Interestingly, the modulatory activity did not seem to be related to distance of pharmacophoric substructures, such as the benzodioxole and the amide moiety. For naphthalene analogues 72 and 65, an increase in distance led to a decrease of activity, whereas in the case of 22 and 18, a decrease of distance led to a decrease of activity. Comparing 23 and 70, which show identical distance of these two moieties, 70 completely lacks activity (32% ± 12%, Table 1). In conclusion, the best compounds achieved in terms of efficiency were the piperine analogues 22 and 23.
Computational Analysis. In order to rationalize the trends observed in the SAR with respect to physicochemical properties and chemical substructures, we explored the possibility to apply quantitative structure−activity relationship (QSAR) methods. As I GABA potentiation does not allow classical QSAR analysis, binary classification models were built from five methods and three descriptor sets. For these studies, all 76 piperine derivatives described above were employed. Sixteen compounds showing ≥200% I GABA potentiation were assigned to an active class, since they were at least as active as the lead compound piperine. The remaining 60 ligands were assigned to an inactive class. Classification methods comprised instancebased classifier (IBk), J48 decision tree (J48), naïve-Bayes classifier (NB), random forest (RF), and support vector machine (SMO) implemented in the software package WEKA. 44 The software package Molecular Operating Environment (MOE) was used for calculation of 2D descriptors and fingerprints. The three descriptor sets used comprised six 2D descriptors obtained after applying a feature selection algorithm on the whole panel of 125 2D MOE descriptors (6D), 11 physical chemical properties (PHYSCHEM), and MACCS fingerprints (MACCS).
The statistical parameters obtained for the 15 best classification models are listed in Table 7. Most of the models possess reliable quality (except models 11 and 13); that is, values of the Matthews correlation coefficient (MCC) are higher than 0.4 and total accuracy varies from 0.7 to 0.9. Models 3 and 4, although possessing the best statistical performance parameters, are not discussed further, as they are difficult to interpret. Instead, models 7 and 12 are discussed in more detail, because these models (i) show almost equal performance, (ii) were built using descriptors of physical chemical properties and MACCS fingerprints, (iii) provide clear separation between active and inactive instances, and (iv) allow us to trace back the decisive chemical and structural descriptors for the data set.
The decision tree obtained in model 7 with PHYSCHEM descriptors ( Figure 6) uses as a first criterion for separation of active and inactive piperine derivatives: the topological polar surface area. By applying a threshold of 39, 25 inactive ligands exhibiting polar substituents at the amide nitrogen were filtered out. These include compounds 1−16 with monosubstituted amide function and compounds 29, 31, 32, 36, 37, 40−42, and 44 containing several heteroatoms (e.g., OH groups or an additional nitrogen as in piperazines or both). Thus, application of a single filter decreased the number of inactive ligands in the Table 1.
a All data are given as mean ± SEM. Asterisks indicate statistically significant differences from zero: *p < 0.05, **p < 0.01. a All data are given as mean ± SEM. Asterisks indicate significant differences from zero: *p < 0.05, **p < 0.01. data set almost by half, from 60 to 35 compounds. In the next branch of the decision tree, 10 compounds with less than four rotatable bonds were excluded from the data set. These included highly rigid piperine derivatives with linker regions modified to either a single double bond (17) or to an aromatic system (46, 50, 54, 58, 62, 65, 68, 71, and 75). Furthermore, 11 compounds with high lipophilicity (log P > 5.2) were filtered out: 26 and 27 with n-hexyl and cyclohexyl sustituents at the amide nitrogen, as well as 48, 52, 56, 60, 64, 67, 70, 77, and 63, which have dibutyl and dipropyl substituents in the same region. The fact that the top-ranked compounds are either N,N-dipropyl-, N,N-dibutyl-, or N,N-diisobutyl-substituted is reflected in the next leaf, which assigns five compounds (23, 24, 34 25, 43, and 73) with more than seven rotatable bonds to the active class. The last two branches of the decision tree filter out compounds on the basis of their molecular weight and refractivity. The decision tree obtained for model 12 with MACCS fingerprints (Figure 7) is fully in line with the one based on the PHYSCHEM descriptor set. The first filtering criterion was presence or absence of an NH group. It filtered 21 derivatives (1−16, 31, 32, 40, 42, and 45), most of which were those showing high polar surface area (TPSA). The next branching filter was presence of a sulfur atom, which removes six inactive ligands (30,44, and 58−61) from the data set. The next leaf separates compounds that do not have a six-membered ring as in piperidinyl, cyclohexyl, and morpholinyl, which led to seven correctly classified active ligands (21−23, 24, 34 25, 28, and 43) and three missclassified inactives (18, 20, and 26). This criterion is in line with the filter "b_rotN > 7" for active compounds in the PHYSCHEM model.
To summarize, active piperine analogues are mainly characterized by a topological polar surface smaller than 39, have at least three rotatable bonds (better more than 7), and show a log P value smaller than 5.2.
Compounds 25 and 23 Induce Anxiolysis in Mice. Activation of TRPV1 by piperine and its derivatives may cause unwanted side effects, including changes in pain sensation and body temperature and induction of fear that would interfere with GABA A -mediated effects 45,46 (for review see ref 47). In order to rule out potential activation of TRPV1, selected compounds were studied in X. laevis oocytes for interaction with TRPV1 prior to in vivo characterization. The most potent (25) and most efficient (23) piperine analogues (Table 3, Figure 2C,D) did not activate TRPV1 expressed in Xenopus oocytes (upon application of 100 μM, data not shown). Both compounds were further characterized concerning their anxiolytic activity (see also ref 34).
As illustrated in Figure 8A, male C57BL/6N mice treated with 23 at doses ≥0.3 mg/kg body weight spent significantly more time in the open arms (OA) of the elevated plus maze (EPM) test compared to a saline-treated control group (control, 28.7% ± 2.7% for n = 41; 23 at 0.3 mg/kg, 45.6% ± 3.2% for n = 17; p < 0.01). This effect was dose-dependent and reached its maximum at a dose of 3 mg/kg body weight, indicating strong anxiolytic effects of 23. Similarly, mice treated with 25 also spent significantly more time in the OA of the EPM test at doses ≥0.3 mg/kg body weight compared to saline-treated control littermates (control, 28.7% ± 2.7% for n = 41; 25 at 0.3 mg/kg, 39.8% ± 4.1% for n = 23; p < 0.05; Figure 8B). The anxiolytic effect of 25 reaching its maximum at a dose of 3 mg/kg body weight (25 at 3 mg/kg, 43.9% ± 4.3% for n = 12), however, was less pronounced compared to 23.
Application of doses ≥10 mg/kg of 23 or 25 did not further increase the anxiolytic effect in the EPM, which is presumably due to the concomitant occurring/developing of reduced Scheme 3. Synthesis of Piperine Analogues Containing an Aryl Spacer a locomotor activity (see Figure 8C,D for sedative effects in the open field test). Compared to piperine and the previously studied 24 34 ( Figure 8A, shaded bars taken from ref 34), anxiolysis induced by 23 was significantly (p < 0.05) more enhanced, which might reflect the stronger I GABA potentiation by 23 and/or the higher potency of 23 on receptors containing α 2/3 and β 3 subunits. Interestingly, the anxiolytic effect of the most potent and also more efficient derivative 25 did not differ from that of piperine and 24. 34 It has, thus, to be clarified in further studies to what extent derivatization of the amide moiety affects the anxiolytic properties of piperine derivatives and whether receptors/channels other than GABA A receptors are targeted in vivo by these compounds.
Significant amounts of 23 and 25 were detected in mouse plasma after intraperitoneal (ip) application (see Table 8). The estimated plasma concentrations were below the micromolar concentrations required for significant I GABA potentiation of GABA A receptors expressed in Xenopus oocytes. However, drugs are commonly less potent on ion channels expressed in Xenopus oocytes as compared to channels expressed in either mammalian cells or even native tissues. 48 The metabolite formation of 23 and 25 is currently unknown. At the current stage of our research, we cannot exclude that the observed anxiolytic and sedative effects are induced by more active metabolites. Furthermore, the currently unknown brain-barrier penetration of 23 and 25 and possible tissue accumulation warrants further research.

■ CONCLUSIONS
Piperine analogues modulating GABA A receptor with the highest efficiency show a tertiary amide nitrogen, substituted with flexible alkyl chains with a total of 6−8 carbon atoms. Polar substituents as well as rigid substituents give rise to a decrease of activity. Modifications of the linker region that lead to rigidification of the molecules also did not improve efficacy.
Chloride currents through GABA A receptors (I GABA ) were measured at room temperature (+21 ± 1°C) by means of a two-microelectrode voltage clamp technique making use of a Turbo TEC-05X amplifier (npi electronic, Tamm, Germany). I GABA were elicited at a holding potential of −70 mV. Data acquisition was carried out by means of an Axon Digidata 1322A interface using pCLAMP v.10 (Molecular Devices, Sunnyvale, CA). The modified ND96 solution was used as bath solution. Microelectrodes were filled with 2 M KCl and had resistances between 1 and 3 MΩ.
Fast Perfusion System. GABA and the studied derivatives were applied by means of the ScreeningTool (npi electronic, Tamm, Germany) fast perfusion system as described previously. 50 To elicit I GABA , the chamber was perfused with 120 μL of GABA-or compoundcontaining solution at a volume rate of 300 μL/s. 34 Care was taken to account for possible slow recovery from increasing levels of desensitization in the presence of high drug concentrations. The duration of washout periods was therefore extended from 1.5 min (<10 μM compounds) to 30 min (≥10 μM compounds). Oocytes with maximal current amplitudes >3 μA were discarded to exclude voltage clamp errors.
Data Analysis: GABA A Receptors. Stimulation of chloride currents by modulators of the GABA A receptor was measured at a GABA concentration eliciting between 3% and 7% of the maximal current amplitude (EC 3−7 ). The GABA EC 3−7 was determined for each oocyte individually. Enhancement of the chloride current was defined as (I GABA+compd /I GABA ) − 1, where I GABA+compd is the current response in the presence of a given compound and I GABA is the control GABA current. I GABA-max reflects the maximal I GABA enhancement. Concentration−response curves were generated and the data were fitted by nonlinear regression analysis using Origin Software (OriginLab Corp.). Data were fitted to the equation 1/(1 + (EC 50 / [compound]) nH ), where n H is the Hill coefficient. Each data point represents the mean ± SEM from at least three oocytes and ≥2 oocyte batches. Statistical significance was calculated by paired Student t-test with a confidence interval of <0.05.
Molecular Modeling and Quantitative Structure−Activity Relationships. Data Set. The 2D structures of 76 piperine derivatives and piperine were drawn in the InstantJChem package for Excel (www.chemaxon.com/products/jchem-for-excel) and exported in sdf format. The LigPrep tool provided by Schrodinger in the Maestro package (Maestro, version 9.2; Schrodinger LLC, New York, 2011) was used to generate low-energy 3D structures and protonated states. All possible stereoisomers per ligand were computed and one low-energy conformation was generated per each stereoisomer in MMFF force field. The protonated states were determined at pH 7.4 (pH used in the experiments). For compounds 33, 34, 36, 38, and 39, several stereisomers were determined. Since these structures were not ionizable at this pH, the stereoisomers were considered equal in terms of 2D structure and duplicates were removed. Subsequently, the structures were imported into MOE, where partial atomic charges were calculated in the MMFF94 force field. Piperine (obtained from Sigma−Aldrich, Vienna, Austria) was used as a reference compound to determine the class labels of its derivatives. Potentiation of GABA current by piperine was 226% ± 26%; 31 therefore, compounds with potentiation ≥200% were assigned to the active class, otherwise to the inactive. This led to an unbalanced data set with 17 "active" and 60 "inactive" compounds.
Descriptor Sets. One hundred forty-three 2D descriptors implemented in MOE were calculated. The full list is provided in Supporting Information (Table S1A). Descriptors showing no variance were removed from the data set, and the remaining 125 descriptors (Supporting Information, Table S1B) underwent feature selection by the BestFirst algorithm implemented in the software package WEKA version 3.7.9. Consequently, the six descriptors left (set 6D) were used for further classification studies (Table 9). Additionally, as a reference descriptor set, we used 11 descriptors of physicochemical properties (set PHYSCHEM) from the list of 125 descriptors described above (Table 10). These descriptors allow us to trace molecular features important for biological activity and have previously shown good performance in application to ligand-based studies. 51 As an attempt to trace the structural features relevant to the activity of piperine derivatives, MACCS fingerprints (MACCS Keys; MDL Information Systems, Inc., San Leandro, CA) were computed in MOE. MACCS are a set of structural keys, where each key describes a small substructure consisting of up to 10 non-hydrogen atoms. A Python script (Supporting Information) was applied to divide the fingerprints into bit strings. The latter were further used in the classification studies as descriptor set "MACCS".  a Asterisks indicate significant differences from α 1 β 2 γ 2S receptor subtype as follows: *p < 0.05; **p < 0.01. Computational Methods. As classification methods, instance-based classifier (IBk), J48 decision tree (J48), naïve-Bayes classifier (NB), random forest (RF), and Support vector machine (SMO) were used as implemented in Weka. All methods were used with the default parameter settings. Nevertheless, different costs were associated with misclassified compounds since the data set was unbalanced. The costs were evaluated by use of an in-house script (Supporting Information), which consequently built models with different costs of the false positive (FP) and false negative (FN) compounds (from 1 to 200 with step of 1 for FN and from 0 to 20 with step of 0.1 for FP). Moreover, inside the script the 10-fold cross-validation was applied and statistical parameters were computed. Subsequently, one model per method and descriptor set was selected on the basis of highest values of MCC, accuracy, sensitivity, and specificity and was taken for visual inspection and possible interpretation. The cost-sensitive parameters obtained for the best 15 models are listed in Table 11.

Journal of Medicinal Chemistry
Statistical Parameters. The statistical parameters of every model were calculated on the basis of values from confusion matrix (for details see ref 52), where TP and TN stand for correctly classified active and inactive compounds and FP and FN for misclassified inactive and active ligands. The true-positive rates of active (sensitivity) and inactive (specificity) classes were calculated by the following formulas: The accuracy of the model was defined as the ratio of correctly predicted compounds to the total amount of compounds.

= + accuracy TP TN total
Additionally, the Matthews correlation coefficient (MCC) was used to assess the quality of the obtained models. It was calculated from the formula

(TP FP)(TP FN)(TN FP)(TN FN)
MCC is independent of the class sizes and therefore gives a rational evaluation of prediction in our case. It can return values from −1 to +1, where +1 determines perfect prediction, 0 means random classification, and −1 represents a total misclassification. The value of 0.4 was taken as a threshold to filter the best-performing models. Behavioral Studies. Male mice (C57BL/6N) were obtained from Charles River Laboratories (Sulzfeld, Germany). For maintenance, mice were group-housed (maximum five mice per type IIL cage) with free access to food and water. At least 24 h before the commencement of experiments, mice were transferred to the testing facility, where they were given free access to food and water. The temperature in the maintenance and testing facilities was 23 ± 1°C; the humidity was 40−60%; a 12 h light−dark cycle was in operation (lights on from 07:00 to 19:00). Only male mice aged 3−6 months were tested.
Compounds were applied by intraperitoneal (ip) injection 30 min before each test. Testing solutions were prepared in a solvent composed of 0.9% NaCl solution with 10% dimethyl sulfoxide (DMSO) and 3% Tween 80. Application of the solvent alone did not influence animal behavior. All doses are indicated as milligrams per kilogram of body weight of the animal.
Elevated Plus Maze Test. The animals' behavior was tested over 5 min on an elevated plus maze 1 m above ground consisting of two closed and two open arms, each 50 × 5 cm in size. The test instrument was built from gray PVC; the height of closed arm walls was 20 cm.  34 Open Field Test. Ambulation was tested over 10 min in a 50 × 50 cm flexfield box equipped with infrared rearing detection. Illumination was set to 150 Lux. The animals' explorative behavior was analyzed by use of the ActiMot 2 equipment and software (TSE-systems, Bad Homburg, Germany). Arenas were subdivided into border (up to 8 cm from wall), center (20 × 20 cm, 16% of total area), and intermediate area according to the recommendations of EMPRESS (European a Potency (P), expressed as EC 50 , and efficiency (E), expressed as I GABA-max , are compared. Asterisks indicate statistical significance as follows: *p < 0.05, **p < 0.01.
a Potency (P), expressed as EC 50 , and efficiency (E), expressed as I GABA-max , are compared. Asterisks indicate statistical significance as follows: *p < 0.05, **p < 0.01. Estimation of Plasma Levels. Trunk blood from male C57BL/6N (6 months) was taken 15, 30, and 60 min after ip application of 23 and 25 (doses 1, 3, and 10 mg/kg body weight; injection solutions were prepared as described for behavioral analysis). At each time point, mice were euthanized and blood samples (500−800 μL) were collected and compiled into ethylenediaminetetraacetic acid (EDTA)-coated microtubes (1.6 mg of EDTA/sample) and centrifuged at 12 000 rpm for 5 min at 4°C. Plasma samples were transferred into 1.5 mL tubes and stored at −80°C until analysis.
Materials. All solvents used were of UPLC grade. Acetonitrile and dimethyl sulfoxide (DMSO) were supplied by Scharlau (Barcelona, Spain). Methanol was from Lab-Scan (Gliwice, Poland). Ammonium formate, formic acid and trifluoroacetic acid (TFA) were purchased from BioSolve (Valkenswaard, Netherlands), and HPLC-grade water was obtained from an EASYpure II (Barnstead, Dubuque, IA) water purification system. Blank K 3 EDTA C57BL/6N mouse plasma was collected for generating plasma calibrators and quality controls (QC).
Preparations of Calibrators and Quality Control Samples. Two separate sets of 23 and 25 stock solutions were prepared in DMSO for making calibrators and quality control (QC) samples. Plasma calibrators were prepared by spiking corresponding stock solutions into a blank plasma sample. The following 23 and 25 concentrations were added: 20, 50, 100, 250, 500, 1000, and 2000 ng/mL. The same blank plasma and both stock solutions (for QC) were used to generate three level plasma QC samples at 60, 1000, and 1600 ng/mL for both 23 and 25.   LC-MS/MS Analyses. Quantification was performed on a 1290 Infinity LC system coupled with a 6460 triple quadrupole mass spectrometer with Jet Stream Technology, and data was processed with a MassHunter Workstation Software version B.06.00 (Agilent; Waldbronn, Germany). The 1290 Infinity LC system was equipped with a binary capillary pump, degasser, autosampler, autosampler thermostat, thermostated column compartment, and FlexCube. Separation was performed at 55°C on a Kinetex XB-C18 column, 100 × 2.1 mm, 1.7 μm particle size (Phenomenex; Torrance, CA); mobile phase of (A) 0.05% formic acid in 10 mM ammonium formate     Table S2 (Supporting Information).
Syntheses. Details of synthesis and characterization of selected products 25, 51, and 62 and key intermediates 65a, 68a−c, 71a, and 75a,b are described below. Synthetic procedures and characterization data for all other compounds are included in Supporting Information. Purity was determined either by elemental analysis or by HPLC and was >95%. Unless otherwise noted, chemicals were purchased from commercial suppliers and used without further purification. Microwave reactions were performed on a Biotage Initiator Sixty microwave unit (Biotage AB, Uppsala, Sweden). Flash column chromatography was performed on silica gel 60 from Merck (40−63 mm), whereas most separations were performed by using a Buchi Sepacore mediumpressure liquid chromatography (MPLC) system with a 9g column (Buchi Labortechnik AG, Flawil, Switzerland). For thin-layer chromatography (TLC), aluminum-backed silica gel was used. Melting points were determined by using a Kofler-type Leica Galen III micro hot stage microscope (Aigner-Unilab Laborfachhandel GmbH, Vienna, Austria) and are uncorrected. For compounds unknown in the literature, either high-resolution mass spectrometry (HR-MS) or combustion analysis was performed. HR-MS was performed by E. 3-(Benzo[d] [1,3]dioxol-5-yl)-N,N-dipropylbenzamide (51). Benzodioxol-5-boronic acid (138 mg, 0.83 mmol, 1 equiv), 51a (237 mg, 0.83 mmol, 1 equiv), Pd(PPh 3 ) 4 (19 mg, 2 mol %), and sodium carbonate (615 mg, 5.81 mmol, 7 equiv) were charged into a microwave vial. Then a mixture of dimethyl ether (DME)/EtOH 5:1 (6.4 mL) and water (1.8 mL) was added, and the resulting suspension was degassed by passing through argon for 5 min. The vial was sealed and heated to 140°C for 1 h in the microwave. After cooling to rt, the reaction mixture was extracted with dichloromethane (DCM), the solvent was evaporated, and the crude product was directly subjected to column chromatography with light petroleum (LP)/EtOAc mixture as eluent.
Yield 94% (555 mg, 2.21 mmol), colorless solid, mp 135−138°C. Naphtho[2,3-d] [1,3]dioxole-5-carboxylic acid (71a). For synthesis of 71a, a modification of a published procedure 41 was used. In a twonecked flask equipped with magnetic stirrer, septum, and balloon, naphtho[2,3-d] [1,3]dioxol-5-yl trifluoromethanesulfonate 42 (96 mg, 0.3 mmol, 1 equiv), 1,3-bis(diphenylphosphino)propane (dppp; 7 mg, 0.018 mmol, 6 mol %), and palladium(II) acetate (2 mg, 0.009 mmol, 3 mol %) were suspended in DMF/water 3:1 (1 mL). A steel cannula reaching to the bottom of the flask was used to bubble carbon monoxide through the solution for 10 min; after that, the balloon was filled with CO gas in order to maintain its supply throughout the reaction time. Hunig's base (102 μL, 0.6 mmol, 2 equiv) was added via syringe and the reaction mixture was heated to 70°C. After 3 h, reaction control with TLC indicated complete consumption of the starting material. The reaction mixture was diluted with ethyl acetate (10 mL) and extracted with 3 × 5 mL of saturated NaHCO 3 . The combined aqueous extracts were acidified to pH = 2 with 2 N HCl and extracted with 3 × 10 mL of ethyl acetate. The combined organic extracts were washed with 10 mL each water and brine and dried with sodium sulfate. Evaporation of the solvent gave the pure product.  [1,3]dioxole-6-carboxylate (75a). For synthesis of 75a, a modification of a published method 42 was used. A three-necked flask with magnetic stirrer, septum, reflux condenser, and balloon was charged with 5,6bis(bromomethyl)benzo[d] [1,3]dioxole (2.0 g, 6.5 mmol, 1 equiv), methyl acrylate (2.94 mL, 32.5 mmol, 5 equiv), and anhydrous DMF (50 mL) and was flushed with argon. Sodium iodide (3.9 g, 26 mmol, 4 equiv) was added and the reaction was heated to 90°C overnight (in previous experiments on a smaller scale, full conversion had been reached after 2 h). Above 70°C the reaction mixture began to turn red. The reaction was quenched with 200 mL of water, and then, sodium thiosulfate 5% was added until the mixture became colorless. The aqueous mixture was extracted with 5 × 50 mL of methyl tertbutyl ether (MTBE). The organic phase was washed with 50 mL each water and brine, dried with anhydrous sodium sulfate, and evaporated.
Yield 89% (1. Methyl Naphtho[2,3-d] [1,3]dioxole-6-carboxylate (75b). Compound 75a (100 mg, 0.43 mmol, 1 equiv) was dissolved in benzene (3 mL, p.a.) under argon. DDQ (242 mg, 1.07 mmol, 2.5 equiv) was added and the reaction mixture was heated to 80°C for 2 h. TLC analysis was inconclusive due to very similar R f values of starting material and product. Staining with cerium molybdophosphoric acid dip reagent indicated full conversion (The starting material is readily stained; the product only weakly). The reaction was quenched with 20 mL of 2 N NaOH and changed color to brown. The reaction was extracted with 3 × 10 mL of EtOAc. The organic phase was washed with water until the washings were colorless (5 × 10 mL) and subsequently washed with brine, dried over sodium sulfate, and evaporated.
Yield 73% (72 g, 0.31 mmol), colorless solid, mp 130−132°C, sublimation above 105°C. TLC 0.15 (LP/EE 30:1). 1  Synthetic procedures and characterization data for piperine analogues; shell script for evaluation of costs; Python script to divide MACCS fingerprints into bite strings; parameters obtained for model 3; full composition of 10 trees in model 4; and a table with the full list of descriptors calculated. This material is available free of charge via the Internet at http:// pubs.acs.org.