Structural Determinants for Transport Across the Intestinal Bile Acid Transporter Using C-24 Bile Acid Conjugates

Abstract

The human apical sodium dependent bile acid transporter (hASBT) re-absorbs gram quantities of bile acid daily and is a potential prodrug target to increase oral drug absorption. In the absence of a high resolution hASBT crystal structure, 3D-QSAR modeling may prove beneficial in designing prodrug targets to hASBT. The objective was to derive a conformationally sampled pharmacophore 3D–QSAR (CSP-SAR) model for the uptake of bile acid conjugates by hASBT. A series of bile acid conjugates of glutamyl chenodeoxycholate were evaluated in terms of K_m and normalized V_max(normV_max) using hASBT-MDCK cells. All mono-anionic conjugates were potent substrates. Dianions, cations and zwitterions, which bound with a high affinity, were not substrates. CSP-SAR models were derived using structural and physicochemical descriptors, and evaluated via cross-validation. The best CSP-SAR model for K_m included two structural and two physiochemical descriptors, where substrate hydrophobicity enhanced affinity. A best CSP-SAR model for K_m/normV_max employed one structural and three physicochemical descriptors, also indicating hydrophobicity enhanced efficiency. Overall, the bile acid C-24 region accommodated a range of substituted anilines, provided a single negative charge was present near C-24. In comparing uptake findings to prior inhibition results, increased hydrophobicity enhanced activity, with dianions and zwitterions hindering activity.

Introduction

The human apical sodium dependent bile acid transporter (SLC10A2, hASBT) is the major transporter involved in uptake and transport of bile acids from the intestine. It also plays an important role in the enterohepatic circulation of bile acids, enabling efficient recycling and conservation of bile acids to maintain the bile acid pool in the body ^1–4. The transport of both conjugated and unconjugated bile acids with over 95% efficacy emphasizes hASBT’s tremendous potential as a prodrug target ⁵. The design of successful prodrug targets to hASBT is impeded by a lack of understanding of the molecular mechanisms of uptake, and the absence of a high resolution crystal structure. Consequently, there is a need to identify the important structural and functional characteristics that underlie the transport mechanisms for successful prodrug targeting of hASBT. Despite the extensive studies on the functional and molecular properties hASBT and its interactions with various bile acid analogues, an understanding of the substrate requirements for hASBT is limited. In particular, the influence of the C-24 chemistry space attributes that enhance or impede interaction of bile acid analogues with hASBT are not clear ⁶. Notably, no 3D-QSAR uptake model has been developed for hASBT substrates.

Prior to cloning of hASBT, Lack et al. demonstrated conjugation improved transport efficacy and proposed that a negative charge around the C-24 region is essential for interaction with hASBT ⁷. Most studies further established that a single negative charge around C-24 was important for transport activity. For instance, Swaan and colleagues examined the transport of four C-24 bile acid conjugates across Caco-2 monolayers, each providing a negative charge near C-24, of which two conjugates were rapidly transported ⁸. Further analysis suggested the length of the C-24 side chain can be 14 Å or longer for translocation, and that large hydrophobic moieties enhanced binding to hASBT. Baringhaus et al. developed a 3D-QSAR inhibition model with a set of 17 inhibitors of the rabbit ileal Na /bile acid cotransporter. The developed pharmacophore included five essential features: one hydrogen bond donor, one hydrogen bond acceptor and three hydrophobic features.⁹ Previously, in our laboratory, native and C-24 modified bile acid conjugates were studied in stably transfected hASBT-MDCK monolayers. These studies suggested the natural bile acid chenodeoxycholic acid (CDCA) can be coupled to glutamic acid as a linker (denoted glu-CDCA) in order to maintain a single negative charge around C-24, to which various probes can be conjugated ^{6, 10}. Additionally, acyclovir bioavailability was increased by conjugating acyclovir to CDCA via a valine linker at the bile acid’s C-24 position, demonstrating negative charge could be replaced by a hydrogen bond accepting group ⁸.

The conformationally sampled pharmacophore (CSP) method is a novel approach ^11–14 that can be combined with physiochemical parameters to yield 3D-QSAR models (denoted CSP-SAR). CSP is well suited for compound classes with high structural flexibility, such as bile acid conjugates. CSP considers all accessible molecular conformations to assure that the bioactive conformer of each compound is included for model development. The inhibition requirements of hASBT using anilinyl conjugates of glu-CDCA ¹⁵ and aminopiperidine conjugates were studied previously using CSP-SAR ¹⁴. In this study, we focus on delineating the uptake requirements of hASBT using a previously reported series of glu-CDCA conjugates of benzene and aminopiperidine analogues ^{14, 15}, where benzene and aminopiperidine represent drug scaffolds. Transport data in these bile acid analogues were used to build a CSP-SAR model that could be of utility to predict potential prodrugs, where CDCA would be conjugated to the drug via a glutamic acid linker. Overall, this work attempts to elucidate hASBT substrate requirements and derive 3D-QSAR uptake models, which have promised to develop a rational prodrug design strategy to target ASBT.

Experimental Section

Materials

[³H]-Taurocholic acid (10 µCi/mmol) was purchased from American Radiolabeled Chemicals, Inc (St. Louis, MO). Taurocholate was obtained from Sigma Aldrich (St. Louis, MO). CDCA was obtained from TCI America (Portland, OR). Geneticin, fetal bovine serum (FBS), trypsin, and DMEM were purchased from Invitrogen (Rockville, MD). All other reagents and chemicals were of the highest purity commercially available.

Synthetic Procedures

A series of substituted anilinyl ¹⁵ and aminopiperidine conjugates glu-CDCA were previously synthesized ¹⁴. Briefly, almost all conjugates employed coupling α-benzyl glutamic acid to CDCA via either N-hydroxysuccinimide (OSU) ester or benzotriazole (OBT) ester. Various substituted probes were then coupled to bile acid via glutamic acid by stirring either at RT or 60°C using O-benztriazol-1-yloxytris-1,1,3,3 tetra methyl uranium hexaflourophosphate (HBTU) as the activating agent and triethylamine (TEA) as the base. The resulting neutral compounds were then subjected to hydrogenation in a parr shaker for 1–2 h in ethanol and 10% palladium to remove the α- benzyl group, yielding the mono and dianionic targets. Neutral compound intermediates were purified by column chromatography using a gradient of hexane and ethyl acetate. Final target compounds were obtained as solids after deprotection. Identity and purity were confirmed by TLC, MS, NMR, and elemental analysis ¹⁵ .

Cell Culture

Stably transfected hASBT-MDCK cells were cultured as described previously ^{10, 16}. Briefly, cells were grown at 37°C, 90% relative humidity, and 5% CO₂ atmosphere and fed every 2 days. The media for cell culture comprised DMEM supplemented with 10% FBS, 50 U/mL penicillin, and 50 µg/mL streptomycin. Geneticin at 1 mg/mL was used to maintain selection pressure. Cells were passaged every 4 days or after reaching 90% confluence.

Conjugate uptake into hASBT-MDCK cells

Uptake studies of 32 conjugates of glu-CDCA were performed in 12-well cell culture polystyrene plates (Corning Costar; Corning, NY). Of the 32 conjugates, 29 were anilinyl conjugates of glu-CDCA ¹⁵, that were either mono- or di-anionic, while three were aminopiperidinyl conjugates of glu-CDCA ¹⁴. Aminopiperidinyl conjugates were either cationic or zwitterionic. The 12-well plates were seeded with hASBT-MDCK cells at a density of 1.5×10⁶ cells/well and induced on day 4 with 10 mM sodium butyrate for 12–15 h prior to uptake experiments. Conjugate concentration varied from 0–100 µM for most monoanions and up to 5000 µM for dianions. Uptake studies were initiated by washing the cells three times with warm Hank’s balanced salt solution (HBSS) [pH 6.8] and exposing cells to donor solution for 10 min at 37°C and 50 rpm in an orbital shaker. Cells were washed three times with chilled sodium free buffer. Parallel identical studies were performed without sodium, in order to measure passive conjugate uptake. Additionally, taurocholate V_max was measured on each occasion as previously described ¹⁴, as taurocholate is a reference hASBT substrate. For each conjugate, conjugate V_max was normalized to observed taurocholate V_max to yield substrate normalized V_max, denoted normV_max. normV_max accommodates variable hASBT protein expression across occasions.

Quantification of conjugates by LC/MS/MS

Cells were lysed and subjected to extraction as described previously ¹⁷. Briefly, at the end of uptake experiment, cells were lysed by addition of 300 µL of acetonitrile. Acetonitrile was evaporated at RT over 2–3 hr. Extraction of conjugate from cell lysate was performed by adding 1mL of 50:50 acetonitrile:water spiked with 500 nM internal standard; plates were then sealed with parafilm. After 30 min, extract was placed into silanized vials and stored in −80°C until analyzed. Recovery experiments using the extraction protocol have been previously described using ³[H]-taurocholate as a standard substrate for hASBT ¹⁷. Taurocholate uptake studies were performed as a standard control; taurocholate kinetic parameters were determined from LC/MS/MS analysis using the methods described below except detection was achieved under negative ion electrospray tandem MS using the [M-H]⁻, peaks and compound 1 as the internal standard. The multiple reaction monitoring (MRM) transition used for taurocholate was 514.4> 123.66 and run time was 8 min.

Conjugate uptake was quantified by LC/MS/MS using a Finnigan Surveyor HPLC system equipped with an autosampler with a Finnigan TSQ quantum Discovery Max mass spectrometer (Thermo Fisher Scientific Inc., Waltham MA). The column was a Phenomenex Luna C8, 50 × 2 mm, 5µ column, heated to 40°C. The flow rate was 0.4 mL/min. The gradient of 5–95% acetonitrile was employed over 0.2–4 min and the overall run time for most conjugates was 6–8 min. The starting mobile phase gradient for most conjugates was 50:50 acetonitrile: water (Table S1a in supplemental data) up to 0.2 min, followed by increase in organic phase to 95% up to 3–4 min. It was maintained at 95% organic phase for 1 min and then reduced back to 50% organic phase by 6–8 min. The mobile phase included 0.1% formic acid as a modifier for all monoionic and dianionic conjugates. For zwitterionic and cationic conjugates, 0.1% ammonium hydroxide was used as the modifier. Injection volume was 10 µL. Detection was achieved under positive ion electrospray tandem MS using the [M+H]⁺ peak, as positive mode provided the greatest sensitivity, except for the standard taurocholate where detection was achieved under negative ion mode. The unsubstituted aniline conjugate, which shares a structure similar to all studied conjugates, was used as the internal standard. MRM transitions for the conjugates are listed in Table S1a.

Kinetic Analysis

Conjugate kinetic parameters V_max, K_m and P_p were obtained by fitting uptake data to modified the Michaelis-Menten model as previously described using WinNonlin 5.2 (Pharsight, Mountain View, CA) ¹⁷. Eqn 1 is a modified version of the classical Michaelis-Menten which accounts for the presence of an apical boundary layer, P_ABL was 1.5 × 10⁻⁴ cm²/s ¹⁸. Data from sodium free studies were fitted to eqn. 2 which represents the passive transport.

$$J=\frac{P_{\mathit{\text{ABL}}}\left(\frac{V_{\text{max}}}{K_m+S}+P_p\right)}{P_{\mathit{\text{ABL}}}+\frac{V_{\text{max}}}{K_m+S}+P_p}S$$ (1)

$$J=\frac{P_{\mathit{\text{ABL}}}{P}_{p}S}{P_{\mathit{\text{ABL}}}+P_p}$$ (2)

Computational Methods

The twenty-seven anilinyl conjugates of glu-CDCA that were substrates were used to develop CSP-SAR substrate models, individually for K_m and K_m /normV_max. None of the aminopiperidine conjugates were substrates, and hence not used in CSP-SAR analysis. K_m reflects substrate affinity for the transporter. K_m values less than 0.1 µM were assigned a value of 0.1 µM, due to the limited ability to accurately estimate potent K_m values ¹⁸. V_max is the transporter capacity for the substrate. normV_max is V_max normalized against observed taurocholate V_max in order to accommodate variable hASBT protein expression across occasions. K_m/normV_max describes the reciprocal of the transporter efficiency. Model development for normV_max and normV_max/K_m were not successful, as described below.

Compound descriptors for CSP-SAR modeling were previously generated ¹⁵. Briefly, in silico models of conjugates were built using the program CHARMM ¹⁹ with the all-atom CHARMM general force field (CGenFF) ²⁰ . Each molecule was subjected to 1000 steps of steepest descent (SD) and 500 steps of Newton-Raphson (NRAP) energy minimization in the gas phase to a gradient of 10⁻⁴ kcal/mol/Å. Structures minimized using CHARMM were subjected to replica exchange molecular dynamic (MD) simulations involving 20 ns simulations with four replicas of each molecule between 300K and 400K, using an exponential scale (300K, 330K, 363K, and 400K) ^21–23. Exchange of replicas was attempted after every 250 MD steps. MD simulations were performed using Langevin dynamics ²⁴ with an integration time step of 0.002 ps and the aqueous solvation was modeled implicitly via the Generalized Born Continuum Solvent Model (GBMV) ^{25, 26}. SHAKE was applied to all covalent bonds involving hydrogens ²⁷. Conformations saved every 20 ps and obtained from all 4 replicas were used for the analysis. The protonation states of the ionizable groups present in the molecules were determined based on the experimental pH of 6.8. All free acids were assumed to be protonated.

As previously reported, several physico-chemical descriptors were calculated for each conjugate by averaging over the full trajectories ¹⁵. Those descriptors that manifested here in models are briefly described here. The free energy of solvation of each compound in water (ΔG_w) was calculated using the semi-empirical quantum chemical program AMSOL6.8 by invoking the AM1-SM5.42R continuum solvent model ²⁸. ΔG_w was calculated by taking the difference between the heat of formation plus solvation energy of the solvated system in its relaxed electronic state and that for the gas-phase system. Average solvent accessible surface area (SASA) and polar surface area (PSA) were calculated using the Lee and Richards method as implemented in CHARMM ²⁹. The radius of the solvent molecule was taken as 1.4 Å, which approximates the radius of a water molecule ³⁰. PSA for each compound was calculated by adding the contribution of polar atoms (N, O, F, Cl, Br and hydrogen atoms covalently attached to any of these atoms). For calculation of PSA, halogens were included as they could potentially take part in polar interactions with other polar atoms ^31–34. SASA and PSA were also calculated exclusively for the region beyond C-24 of the bile acid conjugates (SASA_side and PSA_side) to access the contribution of the substituents alone towards the interaction with hASBT. Surface area terms were obtained as averages over the conformations saved from the MD simulations. Molar refractivity (MR) and flexibility parameter (KierFlex) were calculated using the program MOE ³⁵. The number of hydrogen bond donors (HD) for each molecule was calculated using Discovery Studio (Accelrys Inc.) ³⁶.

Conformational properties of molecules required for CSP model development included the 20 pharmacophore distances and 26 pharmacophore angles reported previously ¹⁵ as shown in Figure 1. Overlap coefficients (OC) for each compound were calculated with respect to the most potent substrate in the data set (Compound 13, K_m = 0.1 µM). Greater OC values reflect greater structural similarity to the most potent compound.

Figure 1.

Structural features in anilinyl conjugates of glu-CDCA used for pharmacophore development. Notation is as follows: O3, position of 3-OH; O7, position of 7-OH; BC, centroid of B and C rings of steroidal nucleus; C18, location of C-18; C19, location of C-19; C20, location of C-20; C-24 is the location of the charge on unconjugated native bile acids; AA, centroid of carboxylic acid oxygens on a-acid; GS, centroid of γ-substituent (R1); R2 is the substituent on the a-acid (-H or -CH2C6H5).

Regression Analysis

Using the statistical program R, regression analysis was applied to each K_m and K_m/normV_max to develop CSP-SAR substrate models. Molecular descriptors were subjected to single variable as well as multivariable regression against each of the dependent variables. The structural and physico-chemical descriptors were then combined to make a complete set of molecular descriptors for model development. All molecular descriptors were subjected to multivariable regression analysis against both K_m and K_m/normV_max separately.

Molecular descriptors were subjected to multivariable regression in all possible combinations of two. However, any combination having correlation between each other (r) greater than 0.8 were not considered. Table S2 (supplemental data) lists correlation values between independent descriptors. All combinations of descriptors with p-values of the independent variables, as well as the intercept, less than 0.05 were selected for further analysis. Additional descriptors were sequentially added to each selected pair of descriptors. This approach was applied until all possible combinations of descriptors yielded p-values greater than 0.05. All combinations of descriptors with p-values less than 0.05 formed the set of candidate models. The standardized coefficients of the independent variables ³⁷ of the candidate models were also calculated ³⁸, as previously described ¹⁵. Model evaluation was performed using the modified Akaike Information Criterion (AIC_C) ^{39, 40} analysis and leave-one-out cross validation method, as described previously ^{14, 15}. In calculating AIC_C, the residual sum of squares (RSS) obtained from fitting all compounds.

Results and Discussion

Results from uptake experiments into hASBT-MDCK cells are presented, followed by CSP-SAR substrate models. CSP-SAR models are for substrate affinity (i.e. K_m) and for reciprocal of efficiency (i.e. K_m/normV_max).

Conjugate uptake into hASBT-MDCK cells

A congeneric series of 32 aniline and aminopiperidine bile acid conjugates previously synthesized and tested for hASBT inhibition ¹⁵ were evaluated for uptake by hASBT. Figure 2 shows the uptake profiles of taurocholate (K_m= 4.84 µM and V_max= 0.699 pmol/cm²/s) which is a known potent substrate, and the unsubstituted anilinyl conjugate of glu-CDCA (compound 1). Compound 1 kinetics was similar to taurocholate. Like taurocholate, compound 1 was a potent hASBT substrate (K_m= 0.823 µM and normV_max= 1.19). Total uptake (in presence of sodium) was markedly higher than passive uptake (absence of sodium).

Figure 2.

Concentration-dependent uptake profiles. (A) Taurocholate and (B) unsubstituted aniline conjugate were actively translocated by hASBT. Uptake was greater in the presence of sodium (●) than in the absence of sodium (○), indicating them to be hASBT substrates.

Kinetics of hASBT-mediated transport was assessed for each conjugate from transport studies employing a range of donor concentrations for each conjugate. Table 1 lists 27 compounds (compounds 1–27) that were hASBT substrates. Table S1b (supplemental data) provides stereochemical SMILES of the compounds used in this study. All were monoanionic, except one dianionic compound. Table 2 lists five compounds (compounds 28–32) that were not hASBT substrates, and includes two dianions, one cation, and two zwitterionic compounds. Tables 1 and 2 indicate compound kinetic parameters K_m, normV_max, and P_p. Mono-anionic conjugates were potent substrates of hASBT with activity ranging from 0.1µM to 17 µM. This K_m range is the same as the range for native bile acids, which is from <0.1 to 16 µM ⁴¹. These results suggest the potential of further exploiting C-24 conjugation to design prodrugs that can be targeted to hASBT. Among the substrates, ring substituents varied in electronic character, size, and hydrophobicity. Increased polarity led to a slight decrease in affinity (i.e. increase in K_m) as seen by esters, amine and boc group substituents on the benzene ring. Electronic character did not significantly impact affinity, as electron donating (e.g. methyl, methoxy) and electron withdrawing substituents (e.g. chloro, flouro) were potent substrates.

Table 1.

Structures, K_m, and normalized V_max values of anilinyl conjugates of glu-CDCA. This table only includes conjugates that were substrates.

Compound	R1	R2	Charge	Km (µM)a	Normalized Vmax	Pp × 106 (cm/s)b
1		H	−1	0.823 ± 0.320	1.19±0.05	1.40±0.11
2		H	−1	0.100a ± 0.773	1.51±0.18	4.51±0.05
3		H	−1	2.22 ± 1.71	1.30±0.18	6.03+0.72
4		H	−1	1.11 ± 0.35	1.01±0.03	0.643±.040
5		H	−1	0.450 ± 0.611	0.686±0.075	2.54±0.16
6		H	−1	0.100a ± 0.330	1.09±0.01	2.08+0.15
7		H	−1	1.69 ± 0.71	1.72±0.14	2.74±0.19
8		H	−1	0.919 ± 0.622	1.04±0.11	1.62±0.26
9		H	−1	0.100a ± 0.500	1.48±0.17	2.44±0.10
10		H	−1	3.87 ± 0.55	0.978±0.087	1.21±0.08
11		H	−1	1.21 ± 0.58	0.553±0.058	2.43±0.13
12		H	−1	6.56 ± 0.98	0.903±0.105	0.666±0.086
13		H	−1	0.100a ± 5.302	2.32±0.887	9.98±0.09
14		H	−1	3.42 ± 1.23	0.375±0.070	1.71±0.01
15		H	−1	7.27 ± 2.80	1.47±0.31	4.08±0.28
16		H	−1	2.54 ± 0.69	0.800±0.110	3.39±0.12
17		H	−1	10.3 ± 3.1	0.203±0.017	1.59±0.07
18		H	−1	17.5 ± 4.7	0.250±0.022	1.73±0.07
19		H	−1	9.89 ± 1.15	0.762±0.241	0.667±0.031
20		H	−1	14.9 ± 1.2	0.535±0.081	0.610±0.029
21		H	−1	7.91 ± 1.94	0.336±0.043	0.586±0.045
22		H	−1	0.549 ± 0.240	1.07±0.06	1.00+0.05
23		H	−1	7.82 ± 2.06	0.546±0.057	1.45+0.06
24		H	−1	2.53 ± 0.48	0.989±0.066	1.42+0.02
25		H	−1	11.4 ± 1.5	0.441±0.047	0.451+0.017
26			−1	495 ± 222	1.34±0.412	1.32+0.39
27		H	−2	607 ± 476	0.639±0.190	0.183±0.008

^aK_m was assigned a value of 0.1 µM when regressed value was less than 0.1 µM; regressed K_m values for compound 2, 6, 9, and 13 were 0.010±0.773 µM, 0.011±0.331 µM, 0.007±0.551 µM, and 0.008±5.301 µM, respectively.

^bP _p is the passive permeability of the conjugates.

Table 2.

Conjugates that were non-substrate for ASBT. R₁ and R₂ refer to substituents on the same general structure in Table 1.

Compound	R1	R2	Charge	Pp × 106 (cm/s)
28		H	−2 (Dianion)	0.216±0.012
29		H	−2 (Dianion)	0.104±0.005
30		CH2C6H5	+1 (Cation)	4.71±0.09
31		H	zwitterion	0.204±0.023
32		H	zwitterion	0.336±0.017

Regarding compound charge, substrate results differed from inhibition results. For inhibition, a single negative charge was not an essential requirement ⁶. However, charge played the dominant role in translocation across hASBT. Only monoanions were potent substrates. Compared to other monoanions, the location of the negative charge was moved distal from C-24 to the benzene ring by protecting the α-acid on glutamic acid, resulting in loss of activity (compound 26, K_m = 495µM). This result suggests that the negative charge should preferentially be within the vicinity of C-24 for active hASBT uptake. None of the three aminopiperidine conjugates (one cation and two zwitterions) were substrates. Only one of the three dianionic conjugates was a substrate (compound 27), but was a very weak substrate. Previous studies have also shown that dianions are weak inhibitors of hASBT ¹⁵, but that cations and zwitterions bound with moderate to good affinity ^{6, 14}. Translocation is a more elaborate and complex phenomenon than binding, such that compounds which were strong inhibitors of hASBT were not found to be strongly translocated if not bearing a single negative charge.

Figure 3 (panel A) illustrates the correlation between normV_max and substrate K_m for compounds 1–25. Compounds 26 and 27 were weak substrates and not included. An inverse relationship was observed where higher capacity was weakly associated with substrate affinity (i.e. lower K_m) (linear r² = 0.391). This association points towards high binding affinity to generally promote transport capacity.

Figure 3.

Plotted relationships between substrate and inhibition parameters for conjugates 1–25 (i.e. Set 1). (A) The relationship between normalized V_max and K_m is plotted. Compounds with higher affinity (i.e. lower K_m) were associated with higher capacity. (B) The relationship between K_m and K_i is plotted. Substrate affinity and inhibitory potency were generally similar, implicating binding as the rate limiting step in overall conjugate translocation. (C) The relationship between normalized V_max and K_i is plotted. Inhibitor potency and capacity for each conjugate were generally positively associated, as lower K_i (i.e. higher binding) was associated with higher translocation capacity. The trend between K_i and normalized V_max was similar to the trend between K_m and normalized V_max in panel A.

Figure 3 also plots the relationships of K_m and normV_max versus inhibitory K_i for compounds 1–25 (panels B and C, respectively). K_i values were reported previously ¹⁵ and reflect inhibition potency. Each K_m and normV_max were weakly associated with K_i (linear r² = 0.360 for K_m and linear r² = 0.310 for normV_max). These monoanionic conjugates were potent inhibitors (K_i ranged from 0.889 to 16.3 µM) and potent substrates (K_m ranged from 0.1 to 17.5 µM). K_i and K_m values were qualitatively similar, reflecting high inhibition potency and high substrate affinity. Like native bile acids ¹⁰, these data indicate substrate binding was the rate limiting step for translocation of monoanions.

Computational Modeling

The compounds were subjected to CSP-SAR analysis. K_m, normV_max, normV_max/K_m, and K_m/normV_max were separately considered, with the aim to identify structural and physicochemical descriptors that explain the biological activity. Results for normV_max and normV_max/K_m did not yield predictive models. The single variable regression analysis yielded poor correlation for normV_max and normV_max/K_m (r² ≤ 0.3), and further fitting was not attempted. However, successful models were obtained for K_m and K_m/normV_max and these results are discussed below.

CSP-SAR hASBT Substrate Model Development: Quantitative Model

Two data sets were examined to obtain substrate CSP-SAR models. Set 1 included all compounds that were substrates, while Set 2 excluded compounds 26 and 27. The K_m values for 26 and 27 were around 400 – 600 fold higher than that of other compounds, indicating that specific features of 26 and 27 lead to significantly lower activities. 27 was the only dianionic conjugate that showed any measurable translocation. 26 was similar to 27 except the a-acid on the glutamic acid linker was protected by the benzyl group, providing a distal charge to C-24. As shown below, these compounds dominated model fitting when included in data set (i.e. Set 1), such that the sensitivity of the model to properties of the other compounds was diminished. Table S3 (supplemental data) shows various molecular descriptor values used to describe K_m and K_m/normV_max of the anilinyl conjugates of glu-CDCA. Aminopiperidine conjugates were not substrates and were not considered in the model development.

CSP-SAR Model for K_m: Single Variable Regression

For each data Set 1 and Set 2, two best CSP-SAR models for K_m were obtained. Table S4 (supplemental data) presents intermediate analysis results from single variable regression of molecular descriptors against K_m for Set 1. Table S5 (supplemental data) represents the same results for Set 2. Molecular descriptors included both structural (e.g. angle, distance) as well as physico-chemical descriptors where physico-chemical descriptors provided higher correlations to K_m for single variable regression.

In the single regression analysis, the physio-chemical descriptors provided higher correlations to K_m than the structural terms. For Set 1, ΔG_w yielded the highest correlation against K_m amongst all the molecular descriptors (Table S4), while HD yielded the highest correlation with K_m for Set 2 (Table S5). LogP and SASA_side yielded the second highest correlation with K_m for Set 1 and Set 2, respectively. The negative coefficients of ΔG_w and logP for both sets indicated that increased compound hydrophobicity enhanced affinity (i.e. decreased K_m). Consistently, HD yielded a positive coefficient for Set 2, further indicating that polar character reduced affinity.

Amongst the structural descriptors, OC of the angles C19-GS-O3 and O7-AA-C18 yielded the two highest correlations for Set 1. Figure 4 (panel A and B) shows the probability distributions of these two angles for compounds 13 (most potent substrate among all compounds) and 27 (weakest substrate in Set 1). All conformational descriptors, except angles GS-O3-AA and GS-O7-AA, exhibited negative linear coefficients for the OCs in Set 1, indicating that conformational similarity to 13 promoted lower K_m value. Angles GS-O3-AA and GS-O7-AA had positive coefficients. However, further analysis showed almost no change in the angle distributions by compounds (Table S4) and poor r², suggesting these descriptors were not predictive of K_m.

Figure 4.

Probability distributions of structural descriptors for K_m modeling of Set 1. (A) Angle C19-GS-O3 and (B) angle O7-AA-C18 displayed the two highest single variable correlations amongst all structural descriptors. (C) Angle O7-AA-BC represents spatial orientation of 7-OH, α-acid and the steroidal nucleus. The red line represents the distribution of 13 (most potent substrate of all compounds) and the blue line for 27 (weakest substrate in Set 1). Panels B and C depict the conformational distributions of the topological descriptors in the top two multivariable CSP-SAR models for Set 1.

Structural descriptors did not yield high correlation (r² ≤ 0.114) with K_m for Set 2 as shown in Table S5. Figure S1 (supplemental data) shows the probability distributions of angles O7-C18-GS and O7-C19-GS for 13 (most potent substrate among all compounds) and 18 (weakest substrate in Set 2). These angles were the structural descriptors in single variable regression for Set 2 that exhibited the two best correlations with K_m, but correlations are still relatively low.

CSP-SAR Model for K_m: Multivariable Regression

For Set 1 and Set 2, 860 and 168 candidate models, respectively, were identified via multivariable regression. Tables 3 and 4 presents the two top models for Set 1 and 2, respectively. AIC was used to rank model fit quality. The best model for Set 1 yielded r² = 0.926 and Q² = 0.630, and showed 10.8% probability of being the best model. The best model Set 2 yielded r² = 0.821 and Q² = 0.683, and had 13.5% probability of being the best model. The regression plots of the predicted vs. observed K_m values are drawn in Figure 5 for Set 1 and Set 2.

Table 3.

Results from multivariable regression and AIC_C analysis for the selected CSP-SAR models for K_m for Set 1. ^a

CSP-SAR Models	r2	Q2	F	Wi	K (µM)	A (µM/ kcal/mol)	B (µM)	C (µM/m3 /mol)	D (µM)	Standardized A (µM/ kcal/mol)	Standardized B (µM)	Standardized C (µM/ m3/mol)	Standardized D (µM)
ΔGw, O7-AA-C18, MR, BC-GSmin	0.926	0.630	66.1	10.8	− 1460 ± 217	−6.59 ± 0.58	− 630 ± 137	72.4 ± 11	327 ± 104	−0.780	−0.456	0.390	0.329
ΔGw, O7-AA-BC, MR, O7-GSmin	0.926	0.633	60.3	10.4	− 1488 ± 214	− 6.89 ± 0.60	− 591 ± 129	73.1 ± 11	295 ± 103	−0.816	−0.413	0.394	0.288

^aA, B, C and D represent the coefficient of the independent variables (molecular descriptors); K represents the coefficient of the intercept; Q² represents the cross validated r²; F is the test statistic value from the F-test regarding the significance of the model; W_i is percent probability of the model being the best model; Set 1 includes all compounds that were substrates.

Table 4.

Results from multivariable regression and AIC_C analysis for the selected CSP-SAR models for K_m for Set 2. ^a

CSP-SAR Model	r2	Q2	F	Wi	K (µM)	A (µM)	B (µM)	C	D	Standardized A (µM)	Standardized B (µM)	Standardized C	Standardized D
HD, KierFlex, ΔGw, GS-O3-C18	0.821	0.683	21.7	13.5	− 99.3 ± 17	8.03 ± 1.3	5.10 ± 1.1	− 0.556 ± 0.184 (µM/kcal /mol)	− 23.2 ± 6.7 (µM)	0.662	0.457	−0.397 (µM/kcal /mol)	−0.438 (µM)
HD, GS-O7-C18, KierFlex, ΔGw	0.809	0.673	20.1	6.35	− 96.7 ± 17	7.74 ± 1.4	− 18.6 ± 5.8	5.10 ± 1.2 (µM)	− 0.483 ± 0.18 (µM/kcal /mol)	0.639	−0.401	0.456 (µM)	−0.344 (µM/kcal /mol)

^aA, B, C and D represent the coefficient of the independent variables (molecular descriptors); K represents the coefficient of the intercept. Set 2 includes all compounds that were substrates, except the 26 and 27, which were weak substrates.

Figure 5.

Regression plots of predicted vs. observed K_m values for anilinyl conjugates of glu-CDCA using CSP-SAR. (A) Using Set 1, the model yielded r² = 0.926 and Q² = 0.630. (B) Using Set 2, which excludes compounds 26 and 27 which were the two least potent substrates, the model yielded r² = 0.821 and Q² = 0.683.

ΔG_w and MR appeared in the two top models for Set 1 with negative and positive coefficients, respectively, indicating hydrophobicity favors transport affinity (Table 3). In these models, coefficients of angles O7-AA-C18 and O7-AA-BC were negative, indicating conformational similarity to 13, promotes activity. Angle O7-AA-C18 is the spatial relationship between 7-OH, α-acid, and C18; angle O7-AA-BC characterizes the spatial orientation of 7-OH, α-acid, and the steroidal nucleus. Figure 4 (panel B and C) depicts the conformational distribution of these two structural descriptors for 13 and 27 (weakest substrate in Set 1). The standardized coefficients for distances BC-GS_min or O7-GS_min, indicated these distances had the least effect on substrate activity.

ΔG_w appeared in 395 of the 860 candidate models for Set 1, in part due to the dianionic compound 27. This compound’s weak affinity is explained by its extreme aqueous solvation free energy. For the top two models, the standardized coefficients of ΔG_w indicate this descriptor to be the most influential. The role of ΔG_w in determining K_m is consistent with inhibition results for these conjugates, where ΔG_w was also the most significant factor for binding affinity (i.e. K_i).

The two top models for Set 2 included HD, KierFlex and ΔG_w (Table 4). The positive coefficient of HD indicated that increased polar character diminished activity. Consistently, the negative coefficient of logP indicated hydrophobicity favors activity. The positive coefficient of KierFlex indicated that substrate flexibility disfavored affinity. The models also contained angle GS-O3-C18 or angle GS-O7-C18. Each showed negative coefficients, indicating conformational similarity to 13 promoted activity. GS-O3-C18 is the angle between the γ-substituent, 3-OH, and C-18, while GS-O7-C18 represents the spatial relationship between the γ-substituent, 7-OH, and C-18. Figure 6 (panels A and B) depicts the conformational distributions for 13 and 18 (weakest substrate in Set 2). While ΔG_w was perhaps the dominant parameter for Set 1, in part due to compound 27 all descriptors in Set 2 contributed more equally to explain the K_m data, as judged by the similar magnitude coefficients. The predicted K_m values of the conjugates in both sets using their respective best model are shown in Table S8 (supplemental data).

Figure 6.

Probability distributions of structural descriptors for K_m modeling of Set 2. (A) Angle GS-O3-C18 and (B) angle GS-O7-C18 were the structural descriptors in the best and second best model for Set 2, respectively. Compounds 13 and 18 were the most and least potent substrates in Set 2, respectively.

Figure 7 presents representative conformers for compounds 13, 18, and 27, (red, blue, and green, respectively). The relative orientation of 27 (weakest substrate) is distinctly different from 13 (most potent substrate). Compound 18 is the least potent substrate in Set 2 and was more similar to 13 than was 27.

Figure 7.

Representative conformations of compounds 13, 18, and 27. The most potent substrate (13) is shown in red. The least potent substrate (27) of all compounds is shown in green. Compound 18 is shown in blue and is the least potent substrate in Set 2. The γ-substituent of Compound 27 sample distinctive conformations compared to other compounds. Regarding structural conformation, compound 18 was more similar to 13 than was 27.

CSP-SAR Model for K_m/normV_max: Single Variable Regression

K_m/normV_max describes the reciprocal of the transporter efficiency. The more familiar form normV_max/K_m reflects transporter efficiency, but was not successfully modeled.Compared to K_m/normV_max, normV_max/K_m was not successful due to the regression effects of assigned K_m = 0.1 µM to the most potent substrates, as K_m is in the denominator of normV_max/K_m but not K_m/normV_max.

Interestingly, single variable regression results for K_m/normV_max were similar to such results for K_m. ΔG_w yielded the highest correlation in Set 1 (Table S6 in supplemental data), while HD yielded the highest correlation for Set 2 (Table S7). logP and SASA_side yielded the second highest correlation with K_m/normV_max for Set 1 and 2, respectively. For each ΔG_w, logP, and HD, compound hydrophobicity favored efficiency. The positive coefficient of SASA_side indicates that bulky groups beyond C-24 region disfavors activity, in agreement with prior findings ⁶. OCs of the angles C19-GS- O3 and O7-C19-GS yielded the two highest correlations amongst the structural descriptors for Set 1. Figure 4 (panel A) and Figure 8 presents these angle distributions for 13 and 27. Similar to the results for K_m, all conformational descriptors except angles GS-O3-AA and GS-O7-AA exhibited negative linear coefficients for the OC, as expected since similarity to 13 promoted lower K_m/normV_max value. Angles GS-O3-AA and GS-O7-AA had positive coefficients in Set 1, but showed almost no change in compound sampling patterns (Table S6) and poor r².

Figure 8.

Probability distribution of angle O7-C19-GS. O7-C19-GS was the best structural description from single variable analysis for K_m/normV_max for Set 1. The red line represents the distribution of 13; the blue line represents the same for 27.

Amongst the structural descriptors, OCs of angles O7-C19-GS and O7-C18-GS yielded the two highest correlations in single variable regression analysis for Set 2. Angle distributions for 13 and 18 are shown in Figure S1. All structural descriptors, except for C19-GS-O7, AA-GS-O3 and GS-O3-AA, yielded negative coefficient, indicating structural similarity with 13 favors activity (Table S7). The poor r² (< 0.005) of C19-GS-O7, AA-GS-O3 and GS-O3-AA indicated that these angles did not influence conjugate uptake efficiency.

CSP-SAR Model for K_m/normV_max: Multivariable Regression

Multivariable regression identified 500 candidate models for Set 1 and 465 candidate models for Set 2. In Table 5, the best model for Set 1 yielded r² = 0.975 and Q² = 0.895, and showed 28.9% probability of being the best model. In Table 6, the best model for Set 2 yielded r² = 0.937 and Q² = 0.877, and showed 24.7% probability of being the best model. The regression plots of the predicted vs. observed K_m/normV_max values are shown in Figure 9 for Set 1 and Set 2.

Table 5.

Results from multivariable regression and AIC_C analysis for the selected CSP-SAR models for K_m/normV_max for Set 1. ^a

CSP-SAR Model	r2	Q2	F	Wi	K (µM)	A (µM/kcal/mol)	B (µM)	C (µM/Å2)	D (µM/Å2)	Standardized A (µM/kcal/mol)	Standardized B (µM)	Standardized C (µM/Å2)	Standardized D (µM/Å2)
ΔGw, O7-AA-C18, PSAside, SASA	0.975	0.895	207	28.9	− 1210 ± 140	− 9.61 ± 0.42	− 535 ± 77	0.600 ± 0.25	0.835 ± 0.13	−0.868	−0.296	0.087	0.249
ΔGw, O7-AA-C18, SASA, PSA	0.974	0.867	197	15.1	− 1218 ± 148	− 9.62 ± 0.44	− 556 ± 82	0.828 ± 0.14	0.554 ± 0.26	−0.869	−0.307	0.247	0.081

^aA, B, C and D represent the coefficient of the independent variables (molecular descriptors); K represents the coefficient of the intercept. Set 1 includes all compounds that were substrates.

Table 6.

Results from multivariable regression and AIC_C analysis for the selected CSP-SAR models for K_m/normV_max for Set 2. ^a

CSP-SAR Model	r2	Q2	F	Wi	K (µM)	A (µM)	B (µM/Å2)	C (µM/Å2)	D (µM)	Standardized A (µM)	Standardized B (µM/Å2)	Standardized C (µM/Å2)	Standardized D (µM)
HD, SASAside, SASA, O7-AA-C19	0.937	0.877	71.1	24.7	− 71.9 ± 29	26.7 ± 2.6	0.454 ± 0.061	− 0.346 ± 0.077	70.7 ± 23	0.629	1.24	−0.976	0.432
HD, SASAside, SASA, C19-AA	0.936	0.884	69.6	19.4	− 83.2 ± 28	27.8 ± 2.7	0.422 ± 0.054	− 0.313 ± 0.069	63.5 ± 21	0.656	1.15	−0.884	0.379

^aA, B, C and D represent the coefficient of the independent variables (molecular descriptors); K represents the coefficient of the intercept. Set 2 includes all compounds that were substrates, except the 26 and 27, which were weak substrates

Figure 9.

Regression plots of predicted vs. observed K_m/normV_max values for anilinyl conjugates of glu-CDCA using CSP-SAR. (A) Using Set 1, the model yielded r² = 0.975 and Q² = 0.895. (B) Using Set 2, the model yielded r² = 0.937 and Q² = 0.877.

Each ΔG_w and SASA appeared in the two top models for Set 1 with negative and positive coefficients, respectively. ΔG_w indicated that lipophilicity favored transport efficiency. SASA suggested that bulky ligands disfavored efficiency. Both models also included angle O7-AA-C18, which is the angle involving 7-OH, α-acid and C18. Its negative coefficients indicated conformational similarity to 13 promoted efficiency. Figure 4 (panel B) depicts O7-AA-C18 distribution for 13 and 27. The two top models also included either PSA_side and PSA (Table 5). Higher conjugate polar surface area disfavored efficiency. Similar to K_m models for Set 1, the standardized coefficients of both models suggested that ΔG_w predominated over other descriptors to determine transport efficiency.

The two top models for Set 2 included HD, SASA, and SASA_side (Table 6). The positive coefficient of HD indicated that increase in polar character diminished efficiency. The positive coefficient of SASA_side indicated bulker C-24 side chain disfavored efficiency. In contrast to SASA_side, the negative coefficient of SASA indicated that overall solvent accessible surface area favored efficiency. However, the standardized coefficients indicate that SASA did not impact efficiency as much as SASA_side. Each model included either angle O7-AA-C19 or distance C19-AA. Each exhibited a positive coefficient, indicating that similarity with 13 disfavored efficiency, although amongst all the descriptors had the least amount of contribution to the models indicating that activity was mainly explained by physicochemical properties. The predicted K_m/normV_max values of the conjugates in both sets using their respective best model are shown in Table S9 (supplemental data).

Logarithmic transformed data was also analyzed [i.e. logK_m and log(K_m/normV_max)]. Model regression results for logK_m for Set 1 and Set 2 are presented in Table S10 and Table S11, respectively (supplemental data). Model regression results for log(K_m/normV_max) for Set 1 and Set 2 are presented in Table S12 and Table S13, respectively (supplemental data). Results from logarithmic transformed data agreed with non-logarithmic transformed data analysis, in that increased hydrophobicity enhanced activity. Figure S2 (supplemental data) shows predicted versus observed logK_m for Set 1 and Set 2. Figure S3 (supplemental data) shows predicted versus observed log(K_m/normV_max) for Set 1 and Set 2.

Conclusion

hASBT is a prodrug target to increase the absorption of drug candidates that otherwise are insufficiently orally absorbed. The objective was to derive a CSP-SAR model for hASBT uptake, since 3D-QSAR models for ASBT substrates are lacking. A congeneric series of 32 aniline and aminopiperidine bile acid conjugates were evaluated. Twenty-seven compounds were substrates. Five were non-substrates. Mono-anionic conjugates were potent substrates, with K_m in the same range as K_m of native bile acids, and thus supporting C-24 conjugation as a prodrug approach. Among the substrates, ring substituents that increased polarity led to a slight decrease in affinity. Electronic character did not significantly impact affinity. Only monoanions were potent substrates, with preference for the negative charge to be proximal to C-24 for active hASBT uptake. None of the three aminopiperidine conjugates (one cation and two zwitterions) were substrates, although were potent inhibitors. Only one of the three dianionic conjugates was a substrate, but was a weak substrate. Translocation across hASBT is a more elaborate and complex phenomenon than binding to hASBT, such that compounds which were strong inhibitors of ASBT were not found to be strongly translocated if not bearing a single negative charge.

CSP-SAR models successfully explained substrate affinity (i.e. K_m) and K_m/normV_max, which is the inverse of substrate efficiency. Since K_m data for compounds 26 and 27 were around 400–600 fold higher than the rest of the compounds, two parallel sets of analysis were performed; one set included all substrates (Set 1), while the other set excluded 26 and 27 (Set 2). Single variable linear regression analysis showed that physicochemical properties of these compounds yielded better correlation with the various kinetic parameters than the conformational descriptors. Both K_m and K_m/normV_max yielded successful CSP-SAR models for both sets of compounds. The best CSP-SAR model for K_m included two structural and two physiochemical descriptors, where substrate hydrophobicity enhanced affinity. The best CSP-SAR model for K_m/normV_max in both Set 1 and Set 2 employed one structural and three physicochemical descriptors, indicating hydrophobicity enhanced efficiency. Notably, the difference in the quality of model fit and the predictive power of the best models between Set 1 and Set 2 was less than 0.105 showing that CSP-SAR method was able to develop robust models for both set of compounds. CSP-SAR models for the substrate requirements were overall guided by conjugate hydrophobic and polar characteristics.

Abbreviations

hASBT

Human apical sodium-dependent bile acid transporter

CDCA

Chenodeoxycholic acid

CSP

Conformationally sampled pharmacophore

3D–QSAR

three dimensional structural acivity relationship

MD

Molecular dynamics

PSA

Polar surface area

HD

Hydrogen bond donor

OC

Overlap coefficient

SASA

Solvent accessible surface area

MR

Molar refractivity

MDCK

Madin Darby canine kidney