Design and synthesis of a potent, highly selective, orally bioavailable, retinoic acid receptor alpha agonist

Graphical abstract


Introduction
The retinoic acid receptors (RARa, RARb, and RARc) are members of the nuclear receptor superfamily. Compounds which bind to and activate the RARs are termed retinoids and comprise both natural retinol (Vitamin A) metabolites and synthetic analogs. Retinoids regulate a wide variety of biological processes such as vertebrate embryonic morphogenesis and organogenesis, cell growth arrest, differentiation, and apoptosis, as well as their disorders. 1 The RARa isoform is found in the majority of tissues and has been implicated in a number of diseases, most notably acute promyelocytic leukemia (APL). Selective RARa agonists have been shown to inhibit proliferation and induce apoptosis of mammary tumor oncogenesis in murine models (MMTV-neu and MMTV-wnt1 transgenic mice) relevant to human cancer, 2 and to inhibit LPS-induced B-lymphocyte proliferation. 3 Selective RARa agonists have also been shown to prevent neuronal cell death caused by amyloid-b and, when administered orally, can prevent amyloid-b production and Alzheimer's disease progression in a mouse model. 4 It has been shown 5 that selective RARa agonists suppressed allospecific immune response and significantly prolonged the survival of mouse cardiac allografts and can ameliorate nephritis in lupus-prone mice, NZB/NZW F1. 6 This supports the rationale for using RARa agonists as immunosuppressants in human organ transplantation. Thus selective RARa agonists have the therapeutic potential for the treatment of cancer, dermatological diseases, Alzheimer's disease and immunological disorders.
However, although AM 580 (2) and AGN 195183 (4) have moderate and good selectivity respectively for RARa, over RARb and RARc they are quite lipophilic (cLog P 6.3 and 7.2). In addition AM 580 (2) has been shown to be toxic, 11,12 and the more recently discovered compound AGN 195183 (4) 10 which was in Phase I clinical trials for cancer has been discontinued. 13 Our aim was to find a novel, potent, highly selective RARa agonist not based on the bicyclic 5,5,8,8-tetramethyl-5,6,7,8-tetrahydronaphthalene class that was ligand efficient, orally bioavailable and without the lipophilic obesity seen with (2), (3) and (4). We outline here how we discovered our initial hit compound 5 and how this was developed into the orally bioavailable, highly potent and selective RARa agonist 56 (Fig. 1) which exhibits promising drug-like properties.

Chemistry
The phenyl carboxamido-benzoic acids (Schemes 1, 3, 4 and 5) and phenyl carbamoyl-benzoic acid 26 (Scheme 2) were prepared by coupling the appropriately substituted aniline with a substituted benzoic acid using a variety of standard methods for the formation of an amide bond. The 3,5-dichloro-4-alkoxy compounds 12-15 and 17-21 (Scheme 1) were prepared by alkylation of the phenolic group of methyl 3,5-dichloro-4-hydroxybenzoate 6 followed by hydrolysis of the benzoate ester. Coupling the resultant acid 7 via the acid chloride by reaction with oxalyl chloride or directly with HATU, with the appropriate methyl 4-aminobenzoate 8 followed by hydrolysis with lithium hydroxide gave the required acids 12-15 and 17-21.
For compound 16 the initial alkylation of 6 was carried out with benzyl bromide, and the resulting benzyloxy compound was hydrolyzed, coupled with the aniline 8 (R 2 = H) and the benzyl group was removed using boron trichloride to result in compound 11 (R 2 = H). This material was then alkylated using 1,1-di-tertbutoxy-N,N-dimethylmethanamine in toluene at 80°C. A final hydrolysis using lithium hydroxide in a mixture of tetrahydrofuran and water gave the tertiary butoxy compound 16.
A similar sequence (Scheme 2) coupling the aniline 24 and acid chloride 23 (obtained from acid 22) followed by hydrolysis gave the phenolic acid 25 which upon alkylation with ethyl iodide followed by hydrolysis gave the reverse amide analog 26.
For the symmetrical tri-alkoxy compound 31, treatment of 27 with sodium hydrogen carbonate and ethyl iodide gave mainly compound 28 (R 1 = Et) where alkylation had only occurred in the 4-position of the substrate. After purification, this compound on treatment with potassium carbonate and 2-bromopropane gave an intermediate compound where both remaining hydroxyl groups had reacted with the alkylating reagent. Hydrolysis resulted in the fully alkylated benzoic acid 29 (R 1 = Et, R 2 = R 3 = iPr) which was coupled via the acid chloride to give the methyl ester of compound 31. A final hydrolysis using lithium hydroxide yielded compound 31. The other tri-alkoxy derivatives 32-34 were similarly prepared (Scheme 3).
For the derivatives 39, 40, 42, 43, and 44, (Scheme 4) where the alkoxy groups are the same, both hydroxyl groups in 36 were alkylated by using potassium carbonate and the appropriate alkyl halide in N, N-dimethylformamide heated to 70°C.
Hydrolysis gave rise to the fully substituted benzoic acids 37 (R = iPr), 37 (R = cyclobutyl) and 37 (R = cyclopentyl. These were then coupled to the aniline 38 via the acid chloride generated by treatment of the benzoic acid with oxalyl chloride. The di-tert-butoxy derivatives 41 and 45 were synthesized from the acid 37 (R = t Bu), which was prepared by reacting the two hydroxyl groups in 36 with N, N-dimethylformamide di-tert-butyl acetal followed by hydrolysis, and then coupling the product directly with aniline 38 using HATU. A final treatment of the coupled products with lithium hydroxide in aqueous 1,4-dioxane gave the required acids.
On treatment of 36 with potassium carbonate and benzyl bromide, the 4-benzyloxy methyl ester 46 was produced. For 49 this was then alkylated with isopropyl bromide and base to give the 3-isopropoxy-4-benzyloxy compound 47 (R 2 = iPr) which was hydrogenated, alkylated with ethyl iodide and base and hydrolyzed to give rise to the benzoic acid 48 (R 2 = iPr, R 3 = Et). This benzoic acid 48 was then coupled to the aniline 38 (R 1 = H) using T3P in ethyl acetate and triethylamine as a base, followed by hydrolysis with lithium hydroxide to provide the final compound 49. The other non-identical di-alkoxy compounds 50-57 and 59 were similarly prepared via their corresponding benzoic acids 48 (Scheme 5).

Results and discussion
A ligand-based virtual screening approach, which ranks compounds by their similarity towards known active ligands, was adopted in a search for a novel chemical series of small molecule RARa agonists. The extended electron density representation offered by the Cresset XED force-field provides a way to characterize the calculated field around a molecule. 14  negative charge, steric shape and hydrophobicity, and allows a complete 3D conformational analysis of compounds to be performed. 15, 16 The crystal structure of the selective RARa antagonist BMS 195614 (1) in the human RARa active site 17 was overlaid with AM 580 (2), the antagonist removed and the resulting complete assembly minimized to give the putative bioactive conformation of AM 580 (2). This procedure was also performed for AGN 193836 (3) to get its bioactive conformation. Molecular fields were added to each of these bioactive conformations (Fig. 2).
These unique molecular field patterns were used to search Cresset's database of 2.5 M commercially available molecules, and the results ranked in similarity to the initial bioactive conformations (see Supplementary data for further details).
This methodology identified 3000 commercially available compounds as possible hit compounds. The 200 compounds that had the highest field overlays, Lipinski likeness, and synthetic tractability, were purchased. These were tested in transactivation assays at the RARa, b and c receptors. Full dose-response curves were generated for each active agonist, and the potency of each compound was expressed as a ratio of its EC 50 compared to that of reference ATRA EC 50 value generated on each 96 well plate. This produced several potent hits, including the lead compound 5 ( Table 1). The 3,5-dichloro-4-ethoxy derivative 5 was considered to be one of the better starting points for a lead optimization exercise, not only because of its potency as an RARa agonist but also because of its good selectivity over the RARb and RARc receptors, with moderate lipophilicity (cLog P = 4.4) compared to AGN 195183 (4) (cLog P = 7.2). In addition, it had no systematic Cyp450 liability (inactive at 25 lM at Cyp1A, 2C19, 2C9, 2D6 and 3A4 isoforms), and was not cytotoxic in COS-7 cells (i.e. showed <20% cell death @ 50 Â EC 50 at the RAR alpha receptors).
Our aim was to increase the RARa potency and selectivity over RARb while retaining the excellent selectivity over RARc shown by 5 and achieve oral bioavailability in the rat. The target profile was RARa potency (RARa EC 50 /ATRA EC 50 < 10) with a selectivity of 2 orders of magnitude over RARb and 3 orders of magnitude over RARc with an oral bioavailability of >35% in the rat.
Initial SAR showed that the three aromatic substituents in 5 seemed important for potency as the disubstituted, 3,5-dichloro derivative 60 was less potent at RARa and also less selective than the 3,5-dichloro-4-ethoxy derivative 5 at RARb and RARc. This helped focus our SAR on derivatives with a 3,4,5 substituted aromatic ring.

4-Substituted derivatives
We initially concentrated on the 4-substituent (Table 1). Increasing the length of the 4-alkoxy substituent to n-propoxy 12 and n-butoxy 13 resulted in a loss of selectivity at RARb and RARc.  Increasing the bulk of the 4-alkoxy substituent to isopropoxy 14 and tert-butoxy 16 resulted in an increase in potency at RARa and an increase in selectivity over RARb but a loss of selectivity at RARc. In contrast, the cyclopentoxy compound, 15 was less selective than 5 at both RARb and RARc.
We also explored the reverse amide 26 of 5 which lost significant selectivity against RARb when compared to 5 and hence further work on the reverse amides was curtailed.
We next investigated the PK profile of these 3,5-dichloro-4alkoxy derivatives. We used intrinsic clearance figures in mouse and human microsomes as a simple in vitro screen to minimize the risk of Phase 1 metabolism, before progressing to in vivo studies. The PK profile of the 3,5-dichloro-4-alkoxy series of compounds was poor. The ethoxy 5, tert-butoxy 16 and cyclopentoxy 15 derivatives all had a high mouse, and moderate human intrinsic clearance and 15 was poorly orally absorbed with very low oral bioavailability in the rat ( Table 2).

3,5-Disubstituted derivatives
To overcome these difficulties we turned our attention to the 3,5-sustituents in 5. The patent analysis in this class of compounds showed that non-alkyl substituents in the 3,4,5-substituted aromatic ring of 5 appeared novel. With this in mind we analysed the medicinal chemistry parameters of the 3,5-substituents of our initial 4-OEt derivatives containing non-alkyl 3,5-substituents 5, 62, 63 and 3,5-dialkyl substituents 64 (Table 3). Ranking these four derivatives in terms of RARa potency against the properties of the 3,5 substituents in the second aromatic ring, such as size (MR), lipophilicity (p) and electronic resonance (r) ( Table 3), shows that potency only increases with the lipophilicity p of the 3,5-sustituents (and not with the size or resonance effects of these substituents).
A search of possible aromatic substituents showed that the isopropoxy group has a similar lipophilicity to a chlorine/bromine atom found in 5/63 and a similar size to a tert-butyl found in the   more potent derivative 64. This suggested that the 3,5-diisopropoxy derivative 31 should be at least as active as the chloro and bromo derivatives 5 and 63, and why the 3,5-diethoxy analog 62 which is the least lipophilic, is the least active.

3,4,5-Trialkoxy and 3,4,-dialkoxy derivatives
Encouragingly 31 proved to have good RARa potency (Table 3). In addition, 31 has high selectivity over RARb and RARc (Table 4), and low mouse and human intrinsic clearance with excellent oral absorption and bioavailability (81%) in the rat (Table 2), although it was shown to be only a partial RARa agonist. The close profile of 5 and 31 in terms of RARa potency, as well as RARb and RARc selectivity, shows that in this case, the iPrO group is a good bioisostere of the Cl group. This led the project away from the 3,5-dichloro template and enabled exploration of the alkoxy derivatives at these positions which give a lipophilic surface without the high lipophilicity of the similar sized tertiary butyl group seen in 64, making the template more drug-like. Further analogs of this trialkoxy template 31 were investigated in an attempt to increase its alpha potency while maintaining the excellent beta and gamma selectivity as well as its good PK profile. Increasing the size of the 3,5-substituents in 31 to give the di-cyclopentoxy derivative 32 or increasing the size of the 4-substituents to give 33 maintained the good RARa potency and RARb selectivity but lost selectivity against RARc (Table 4). Decreasing the size of both the 3-and 5-isopropoxy  groups to give the 3,4,5-triethoxy derivative 62, resulted in a substantial loss of RARa potency (Table 3). In addition 31, 32 and 33 all exhibited some partial agonist activity at RARa. However a close analog the 3,4-diethoxy-5-isopropoxy derivative 34 showed that it was possible to have full RARa agonist properties with trialkoxy derivatives (Table 4). This unsymmetrical derivative was further exploited by the investigation of a series of 3,4-alkoxy derivatives (Table 4). Replac-ing one of the isopropoxy groups in the lead 31 with a chloro atom gave the chloro-dialkoxy derivative 49 which had increased potency at RARa and also maintained the excellent selectivity at RARb and RARc. However, this compound was also only a partial agonist at RARa.
Increasing the size of the 3-isopropoxy in 49to 3-cyclobutyl in 50 gave no change in profile. However, increasing the 4-ethoxy group in 49 to the 4-isopropoxy in 39 gave a similar level of  potency at RARa as a full agonist. The molecule was also an order of magnitude more potent than 31 at RARa while maintaining excellent selectivity at RARc with moderate selectivity at RARb. In addition, the di-isopropoxy derivative 39 was orally well absorbed in the rat with a bioavailability of 39% (Table 2). Thus 39 satisfied our target profile except for selectivity at RARb. Increasing the size of the alkoxy groups to the di-cyclobutyl in 40, di-tert-butyl in 41 and di-cyclopentyl in 42 maintained potency at RARa, but decreased selectivity at RARb and RARc. Interestingly reducing the size of the 4-ethoxy in 49 to 4-methoxy in 51 gave a full agonist with good RARa potency and selectivity at RARb and RARc. However, it had a low oral bioavailability (13%) in the rat (Table 2).

Substitution of the benzoic acid ring
Ortho-fluoro substitution in the benzoic acid ring of the bicyclic 5,5,8,8-tetramethyl-5,6,7,8-tetrahydronaphthalene analogs which lead to AGN 195183 (4) 10 has been shown to increase RARa binding potency and increase selectivity over RARb and RARc in the transactivation assay.
Based on this precedent, a series of ortho-substituted benzoic acid derivatives of our initial lead template 5 were prepared (Table 5). While the ortho-fluoro substitution product 17 maintained potency and selectivity, the ortho-methyl substitution product, 18 improved RARa potency 20-fold and maintained good RARb and RARc selectivity. In addition, 18 had a lower mouse and human intrinsic clearance, as well as a somewhat improved bioavailability (12%) in the rat (Table 2), compared to the unsubstituted benzoic acid derivative 5. Compounds 19, 20 and 21 with larger substituent groups, either lost RARa potency or RARb/RARc selectivity compared to 5. As a result of these findings, a series ortho-methyl and ortho-fluoro substituted benzoic acid derivatives of the 3,4-dialkoxy-5-chloro template were prepared ( Table 6). The initial trend from the 5 series (Table 5) was also seen in the 3,4-dialkoxy-5-chloro series ( Table 6).
The ortho-fluoro substituted derivatives 53, 54 and 43 (Table 6) maintained RARa potency and selectivity compared to their corresponding unsubstituted derivatives 51, 49 and 39 and, in addition, had a lower mouse and human intrinsic clearance with the latter being in single figures. However, although both 53 and 54 met our target profile in terms of high RARa potency with a selectivity of 2 orders of magnitude over RARb and 3 orders of magnitude over RARc, they both had low oral bioavailability (15% and 12% respectively) in the rat. The ortho-methyl substituted derivative 55 had a similar mouse and lower human intrinsic clearance (Table 6) compared to the unsubstituted derivative 51 (Table 2). However, it had only partial RARa agonist activity. Both the ortho-methyl derivatives 56 and 44 had good bioavailability (!40%) in the rat and lower mouse intrinsic clearance (Table 6) compared to the unsubstituted derivatives 49 and 39 (Table 2), with 56 having the lowest (single figure) human intrinsic clearance of these four derivatives.
While both derivatives 56 and 44 had high RARa potency and good bioavailability 56 was superior in terms of selectivity at RARb (2 orders of magnitude) and RARc (4 orders of magnitude) and possessed a better overall potency, selectivity and PK profile than the other analogs 45, 57-59 shown in Table 6. a,b and c see Table 1. Table 6 Ortho-fluoro and ortho-methyl (3-chloro-4,5-dialkoxybenzamido)benzoic acids.
The 3-OEt, 4-OiPr geometrical isomer 57 was less potent and less selective at RARc than 56 which is analogous to the trend seen with compounds 51 and 49 in the unsubstituted benzoic acid series. This emphasizes the need for a more lipophilic group than OEt in the 3-and 5-position in this template which was initially seen in Table 3. Thus the 3-OiPr, 4-OEt derivative 56 reached our target profile in terms of potency, selectivity, and oral bioavailability.
The excellent RARa potency, good RARb and RARc selectivity and PK profile of the full agonist 56 suggested further investigations to see if it had sufficient drug-like properties to be an orally bioavailable, highly potent and selective RARa agonist with therapeutic potential.

Predevelopment studies of 4-(3-chloro-4-ethoxy-5isopropoxybenzamido)-2-methylbenzoic acid 56
3.5.1. ADME profiling Predevelopment ADME studies revealed that 56 has a good Cyp 450 profile with no significant inhibition IC 50 > 25 lM against five Cyp 450 isozymes (1A2, 2C9, 2C19, 2D6, 3A4), and has a human and mouse plasma protein binding of 93% and 91% respectively. 20 Compound 56 has also been examined by CEREP in a panel of 120 other receptors, channels and enzymes. The compound at 10 lM demonstrated no significant interactions with any of the sites examined leaving a window of some 4 orders of magnitude between its actions at RAR and any non-RAR site. 21 The highest inhibition of 25% was found for the 5HT2B site. To exclude potential cardiovascular side effects, compound 56 was tested in vitro on the cardiac hERG channel and did not show any significant binding to hERG up to the concentration of 10 lM. 20

Hepatocyte stability
We initially used a microsomes assay as a screen to rank order compounds of interest in terms of their metabolic stability. As microsomes only contain phase I metabolising enzymes it was of interest to screen our lead compound 56 in a secondary screen using hepatocytes which contain the full complement of drug metabolising enzymes present in the liver.
The metabolic stability of compound 56 was tested at two concentrations (1 lM and 30 lM) in mouse, rat, dog, Cynomolgus monkey and human cryopreserved hepatocytes. The compound was shown to be stable, with a long t½ and low clearance in all species (Table 7), which correlates with the available PK data ( Table 8).

PK profile in mice and dogs
The PK profile of 56 was also studied in mice and dogs (Table 8). Compound 56 showed low plasma clearance (Cl) and low volume of distribution (Vss), resulting in sustained plasma half-lives in each species (iv t 1/2 : mice, 1.9 h; dog, 9.2 h). In addition, oral administration of 56 exhibited high bioavailabilities >80% in both mice and dogs. These results encouraged us to investigate 56 further as a predevelopment candidate.

Human RAR alpha receptor
As we planned to perform PK and further in vivo evaluation in rodents, we initially used the corresponding in vitro transcriptional transactivation assays with gal4 fusion receptor constructs, created using each of the mouse RAR ligand-binding domains. Although the percentage identity of amino acid sequences between the mouse and human RAR ligand-binding domains of all three RAR types (a,b or c) is 99-100%, 23 we thought it prudent to confirm the activity and selectivity of our lead compound 56 against the human RAR ligand-binding domains in a transcriptional transactivation assay before further predevelopment studies were investigated. We also tested an earlier less active analog 15 from the 3,5-dichloro template, and AM 580 (2) for comparison (Table 9).
There is a good correspondence for RARa potency between human vs mouse for 56 and 15 with the human being slightly more potent, in contrast to the RARa potency for AM 580 (2) where the human is less potent than the mouse (Table 9). Similarly, the a vs b selectivity comparison for 56 and 15, shows that the human is more selective than the mouse, while for AM 580 (2) the human is less selective than the mouse. Also, a vs c selectivity for 56 is 4 orders of magnitude compared to AM 580 (2) where it is only 2 orders of magnitude for both human and mouse.

In vitro toxicology
In common with most of the other compounds in the series, the lead compound 56 showed no cytotoxicity in COS-7 cells at a 50fold multiple of its EC 50 . 20 When examined in a high content cell toxicity screen in HEPG2 cells (Cyprotex), 56 was found to have no effect at concentrations up to 50 lM on cell or mitochondrial viability markers. 20 This is in contrast to the more lipophilic molecule AM 580 (2) which caused a significant increase in cell membrane permeability and a significant decrease in mitochondrial membrane potential at concentrations between 10 and 30 lM.
When 56 was examined for genetic toxicity, it was negative in bacterial cytotoxicity tests up to 100 lM, negative in an Ames test  in three bacterial strains and in an in vitro micronucleus test in CHO-K1 cells, in all cases in both the presence and absence of S9. 21 In the absence of S9 it should be noted that AM 580 (2), the reference RARa agonist has been shown by others to be a mutagen in vitro. 11,12 3.5.6. Ease of synthesis The 4-(3-chloro-4-ethoxy-5-isopropoxybenzamido)-2-methylbenzoic acid 56 can be synthesized in 9 high yielding reaction steps from 3-chloro-4-hydroxy-5-methoxybenzoic acid (35) (Scheme 5). It is available as a stable highly crystalline, non-hygroscopic, white powder with a melting point of 186°C, and with a solubility of >5 mg/mL, as the sodium salt in water at 35°C.

Profile of lead compound 56
The 3-OiPr, 4-OEt, 5-Cl ortho methyl benzoic acid derivative 56 met our target profile in terms of high RARa agonist potency with a high degree of selectivity over RARb (of 2 orders of magnitude) and excellent selectivity over RARc (4 orders of magnitude) at both the mouse and human receptors. It has high levels of potency in the RARa binding assay (IC 50 ) showing that the transactivation activity observed was being mediated through the alpha receptor (Table 10). As expected 56 was also selective vs RXR (IC 50 > 10 lM in human RXR a and b binding assays). 22 It also possesses good drug-like properties, a low human intrinsic clearance (5.3 mL/min/ mg protein) in microsomes and a measured Log D = 1.8, which resulted in good oral exposure with low clearance and good bioavailability (40%) in the rat (Table 10). In contrast, both 15 and 2 have human intrinsic clearance in double figures and a higher Log D = 2.8, which resulted in low oral exposure in the rat with low bioavailability (0.3%) for 15. Compound 56 was also shown to be metabolically stable to hepatocytes with a long t½ and low clearance in human and 4 animal species (Table 7) together with a high bioavailability (>80%) in both mice and dogs with low plasma clearance (CL) and a sustained plasma half-live (iv t 1/2 : mice, 1.9 h; dog, 9.2 h) ( Table 8). In addition 56 has a solubility of >5 mg/mL as the sodium salt, no systematic Cyp 450 liability against five isoforms (1A2, 2C9, 2C19, 2D6, 3A4) and demonstrated no inhibition (at 10 lM) in a binding assay for hERG channels. It was not cytotoxic in COS-7 cells and was negative for genetic toxicity in the Ames test and micronucleus test in CHO-K1 cells.

Conclusions
We have used a ligand-based virtual screening exercise based on the bioactive conformation of AM 580 (2) and AGN 193836 (3) to identify the novel, less lipophilic RARa agonist 4-(3,5dichloro-4-ethoxybenzamido) benzoic acid 5, which has good selectivity over the RARb, and RARc receptors. Analysis of the   Table 1. c hu = human receptor, see Table 9. d Human microsomes Cl int (mL/min/mg protein). e AUC po ngÁmin mL À1 ,Cl mL/kg/min. f Log D see Table 2.
medicinal chemistry parameters of the 3,5-substituents of derivatives of template 5 showed that RARa potency is driven by the lipophilicity of these substituents. It showed that the iPrO group is a good bioisostere of the Cl group in this case and that the 4 0 -(3,5-diisopropoxy-4-ethoxybenzamido)benzoic acid derivative 31 has a close profile to 5 in terms of RARa potency as well as RARb and RARc selectivity. The low mouse and human intrinsic clearance with excellent oral absorption and bioavailability (81%) in the rat shown by 31 led to the exploration of the more drug-like branched dialkoxy derivatives, the best of which was the 4-(3chloro-4,5-diisopropoxybenzamido)benzoic acid derivative 39 which was an order of magnitude more potent than 31 at RARa, while maintaining excellent selectivity over RARc with moderate selectivity at RARb and was orally well absorbed in the rat with a bioavailability of 39%. Substitution at the ortho-position of benzoic acid 5, with a range of groups, has shown that methyl groups are the best at increasing potency while maintaining good RARb and RARc selectivity. Methyl substitution at the ortho-position of the 4 0 -benzoic acid ring of a series of 4 0 -(3-chloro-4,5-dialkoxybenzamido)benzoic acid derivatives gave the novel RARa agonist 4-(3chloro-4-ethoxy-5-isopropoxybenzamido)-2-methylbenzoic acid 56 as the best in terms of RARa agonist potency and selectivity versus RARb (2 orders of magnitude) and RARc (4 orders of magnitude) at both the human and mouse RAR receptors. This potent RARa-specific agonist with improved physicochemical properties also has high bioavailability (>80%) in both mice and dogs with a good PK profile and drug-like properties and was shown to be negative in the cytotoxicity and genotoxicity screens warranting further consideration as a potential therapeutic agent.

Experimental procedures
All starting materials and solvents, as well as compounds 5, 60, 61 and 62, were obtained from commercial sources. Hydrogenations were performed either on a Thales H-cube flow reactor or with a suspension of the catalyst under a balloon of hydrogen. Microwave reactions were carried out on a Personal Chemistry Smith Synthesizer Workstation with a 300 W single mode microwave cavity. Ion exchange chromatography was performed using strong cation exchange resin (SCX) cartridges purchased from Sigma-Aldrich and washed with methanol prior to use. The reaction mixture to be purified was first dissolved in methanol and then loaded directly onto the SCX and washed with methanol. The desired material was then eluted by washing with 1% NH 3 in methanol. Silica gel column chromatography was performed using Silicycle pre-packed silica (230-400 mesh, 40-63 lM) cartridges.
Preparative HPLC was carried out using a Gilson HPLC and an Agilent 5 mm Prep-C18 21.2 Â 50 mm column. Detection was achieved using a UV detector at 254 nm. Mobile phase A: 0.1% aqueous formic acid, Mobile phase B: 0.1% formic acid in methanol. A flow rate of 40 mL/min was used and a gradient employed as follows; 0.0-0.8 min 5% B; 0.8-7.3 min 5-95% B; 7.3-8.3 min 95% B; 8.3-8.4 min 95-5% B. Analytical LCMS was performed using an Agilent 1200 HPLC and mass spectrometer system with a Scalar 5 mm C18 4.6 Â 50 mm column and peaks detected by positive or negative ion electrospray ionization and a UV detector at 254 nm. All tested compounds were found to be of !95% purity using analytical LCMS. 1 H and 13 C NMR spectra were recorded using a Bruker Avance III TM 400 spectrometer at 400 and 110 MHz respectively, using either residual non-deuterated solvent or tetramethylsilane as a reference in the various solvents specified. All animal studies were ethically reviewed and carried out in accordance with the United Kingdom Animals (Scientific Procedures) Act 1986 by CXR Biosciences Ltd, James Lindsay Place, Dundee Technopole, Dundee DD 5JJ.
A mixture of 3,4-di-tert-butoxy-5-chlorobenzoic acid (37: R = t Bu) (250 mg, 0.831 mmol) and methyl 4-aminobenzoate (38: R 1 = H) was converted to themethyl 4-(3,4-di-tert-butoxy-5chlorobenzamido)benzoate (185 mg, 50%) using the procedure described for compound 18  Step (iii): Methyl 4-benzyloxy-3-chloro-5-hydroxybenzoate (46). Methyl 3-chloro-4,5-dihydroxybenzoate 36 (14.19 g, 70 mmol) was dissolved in N,N-dimethylformamide (210 mL) and treated with potassium carbonate (8.71 g, 63 mmol). After stirring for 5 min, benzyl bromide (8.32 mL, 70 mmol) was added and the mixture was heated to 60°C for 0.75 h. The reaction mixture was diluted with diethyl ether (500 mL) and washed successively with 1 M hydrochloric acid (500 mL) and with brine (2 Â 500 mL). The aqueous phase was re-extracted with diethyl ether (500 mL) and the luciferase that is driven by a gal4UAS. Briefly, on day one, 96 well plates were seeded with 8000 cells per well then left to recover overnight. On day two, the cells were co-transfected with 100 ng of reporter plasmid and 10 ng of the appropriate receptor plasmid per well using lipofectamine (Invitrogen). On day three, the lipofectamine containing media was replaced by a DMEM without phenol red, followed by the addition of novel compounds dissolved in 1 mL of DMSO to each well's 100 mL total volume. Finally, on day four, the cells were lysed and their luciferase substrate was provided by the BrightGlo reagent (Promega), the plates were then read on the MicroBeta TriLux (Perkin Elmer). In each experiment, an 8 point dose response curve of ATRA was run in duplicate, and the various compounds tested were compared to these values.

FlashPlate Ò Scintillation Proximity Binding Assay (SPA)
In the FlashPlate Ò Scintillation Proximity Assay (SPA) wells of a 96-well plate are coated with scintillant and capture antibody (or similar) for tagged proteins. This requires just 100 ng of RAR proteins and 2 nM [ 3 H]-retinoic acid per well. This enables competition of specifically bound [ 3 H]-retinoic acid by unlabelled retinoid compounds. As only radioligand specifically bound to the captured protein is sufficiently close to the scintillant to produce a signal, separation of bound and free radioactivity is not required. Binding of the tritiated retinoid to biotinylated RARa is specific, saturable, time dependent and reversible. We have successfully applied our assay to a screen of known retinoid standards and novel compounds and it is both rapid and reproducible (see Supplementary data file for details).

Intrinsic clearance Cl int
In this in vitro model of hepatic clearance mouse or human liver microsomes were incubated with the test compound at 37°C in the presence of the co-factor, NADPH, which initiates the reaction. The reaction is terminated by the addition of methanol. Following centrifugation, the supernatant is analyzed on the LC-MS/MS. The disappearance of the test compound is monitored over a 45 min time period. The data is the mean on 5 separate experiments. SEM is less than 10% of the mean values.
The ln peak area ratio (compound peak area/internal standard peak area) is plotted against time and the gradient of the line determined.

PK studies in rats
Test compounds were administered orally and intravenously to groups of 4 male Sprague-Dawley rats. Oral dosing solutions of each Test Item were prepared at a concentration of 1 mg/mL in 8% ethanol and 92% PEG-400. The Test Items were orally administered at a dose of 10 mg/kg and a dosing volume of 10 mL/kg. Intravenous dosing solutions of each Test Item were prepared at a concentration of 0.25 mg/mL in 8% ethanol, 92% PEG-400. The Test Items were intravenously administered at a dose of 0.5 mg/kg and a dosing volume of 2 mL/kg. Approximately eight blood samples were collected from each animal at appropriate intervals up to 6 h post dosing for the iv groups and up to 24 h for the oral groups. Whole blood concentrations of the Test Items were measured using LC-MS/MS and selected pharmacokinetic parameters calculated using Pharsight WinNonLin software. For furthers details see Supplementary data file.