Article pubs.acs.org/jmc
Microscale High-Throughput Experimentation as an Enabling Technology in Drug Discovery: Application in the Discovery of (Piperidinyl)pyridinyl‑1H‑benzimidazole Diacylglycerol Acyltransferase 1 Inhibitors Tim Cernak,* Nathan J. Gesmundo, Kevin Dykstra, Yang Yu, Zhicai Wu, Zhi-Cai Shi, Petr Vachal, Donald Sperbeck, Shuwen He, Beth Ann Murphy, Lisa Sonatore, Steven Williams, Maria Madeira, Andreas Verras, Maud Reiter, Claire Heechoon Lee, James Cuff, Edward C. Sherer, Jeffrey Kuethe, Stephen Goble, Nicholas Perrotto, Shirly Pinto, Dong-Ming Shen, Ravi Nargund, James Balkovec, Robert J. DeVita, and Spencer D. Dreher Department of Discovery Chemistry, MRL, Merck & Co., Inc., Rahway, New Jersey 07065, United States S Supporting Information *
ABSTRACT: Miniaturization and parallel processing play an important role in the evolution of many technologies. We demonstrate the application of miniaturized high-throughput experimentation methods to resolve synthetic chemistry challenges on the frontlines of a lead optimization effort to develop diacylglycerol acyltransferase (DGAT1) inhibitors. Reactions were performed on ∼1 mg scale using glass microvials providing a miniaturized high-throughput experimentation capability that was used to study a challenging SNAr reaction. The availability of robust synthetic chemistry conditions discovered in these miniaturized investigations enabled the development of structure−activity relationships that ultimately led to the discovery of soluble, selective, and potent inhibitors of DGAT1.
■
INTRODUCTION The availability of robust synthetic reactions is an essential component of drug discovery. Hypothesis-driven research in drug discovery relies on a design-make-test learning cycle wherein successful synthesis and purification, the “make” component, is required to test a design hypothesis.1,2 At present, the toolbox of synthetic chemical reactions used to explore medicinally relevant chemical space favors a relatively small subset of robust synthetic procedures.3−6 Surprisingly, despite considerable evolution to enhance substrate scope and experimental ease of use, even preferred reactions from the toolbox such as amide formation, Suzuki coupling, and Narylation can be fraught with experimental challenges. For example, during an effort to generate a full combinatorial 50 × 50 library at GlaxoSmithKline, 566 compounds out of 2500 failed synthesis even though additional efforts were made to optimize individual reaction conditions.7 Meanwhile, a survey of 2149 C−N coupling reactions run over one year in our company revealed that 55% of reactions failed to deliver product on the first attempt, highlighting the capricious nature of popular transformations from the toolbox.8 Moreover, an © XXXX American Chemical Society
analysis by Churcher and colleagues suggests a correlation of synthetic outcome with bulk physicochemical properties such as logP, potentially causing undesirable enrichment of lipophilic compounds from synthetic libraries.9−11 High-throughput experimentation tools developed at our company and by others can rapidly navigate synthetic reaction space.12−16 Such tools are well established in the process chemistry drug development phase where the value of synthetic chemistry optimization can be tied to cost reduction, waste reduction, and other metrics.17 Surprisingly, in the early drug discovery phase in medicinal chemistry, where the focus is on robust access to diverse chemical space rather than high-yielding access to a single target, high-throughput reaction screening tools are rarely utilized despite being well positioned to address synthetic chemistry challenges.18 Described herein is the application of micromole-scale high-throughput experimentation in the early drug discovery phase to enable the lead identification and Received: October 24, 2016
A
DOI: 10.1021/acs.jmedchem.6b01543 J. Med. Chem. XXXX, XXX, XXX−XXX
Journal of Medicinal Chemistry
Article
series. Initial forays into the development of structure−activity relationships (SAR) for the lead class involved replacement of 4 with other piperidine and piperazine nucleophiles targeting diverse analogues such as 6−12 (Figure 1). However, the productivity of the SNAr reaction to generate these first eight targets was low, resulting in complete synthetic failure for half of the designed targets (7−9 and 11), thus leaving gaps in the targeted SAR data. This sampling of compounds, with respective yields and inhibitory activities, exemplifies the reality of research at the early stages of lead identification. There are some steep SAR trends (e.g., 6 versus 5) yet subtle structural changes, such as incorporation of nitrogen (10) and removal of the lactone carbonyl (12), are tolerated. These changes affect only modest losses in potency and present an opportunity to alter physicochemical properties in compound design. However, the lack of synthetic efficiency for the series, with only 50% of targeted compounds prepared on the first pass, hampered further SAR development. We desired a modular route wherein both the benzimidazole and piperidine moieties could be studied via parallel synthesis; thus, it was important to locate robust conditions for each transformation.4 Early attempts at the benzimidazole condensation used air as the oxidant,29 which was satisfactory for early exploration, but it quickly became apparent that the spurious nature of the air-mediated reaction would complicate advancement of the series. Fortunately, potassium peroxymonosulfate-mediated benzimidazole condensation, as reported by Beaulieu and co-workers,30 could be used in place of the airmediated process. These conditions proved more reliable across multiple substrates, and multiple parallel synthesis libraries were successfully executed on the program with the potassium peroxymonosulfate conditions. Conversely, optimization of the SNAr displacement used to make the 2-aminopyridine products was a challenge that required significantly more effort to resolve. Initially, compounds were synthesized as singletons, and the SNAr reactions were executed on a pyridyl chloride by heating at 150 °C with microwave irradiation overnight in the presence of piperidine nucleophiles (13) using DIPEA as base and DMA as solvent. Five analogues were produced from ten attempts31 in an average overall yield of 8% (Figure 2A, condition A). The poor performance of these conditions led us to try potassium carbonate in DMSO at 150 °C as an alternative but these conditions were even less productive, producing six analogues from ten attempts in an average overall yield of 3% (Figure 2A, condition B). With either protocol, reactions typically took days to complete and yields were low, owing to both poor conversion and material decomposition due to the harsh conditions. We next investigated 2-fluoropyridines as alternate electrophiles, and significant improvement in reaction performance was realized with an average overall yield of 32% with nine analogues produced from nine attempts (Figure 2A, condition C). Two of the more potent analogues (14, 15) that were successfully isolated and assayed from these studies demonstrated low nanomolar inhibition of human DGAT1 but had at least one off-target selectivity issue, such as CYP2C9 or hERG, and were poorly soluble. Therefore, further synthesis and SAR development were required for lead optimization. Switching from 2-chloropyridine to 2-fluoropyridine substrates vastly improved overall synthetic chemistry performance; however, three issues remained. First, reaction performance was still capricious such that some analogues were prepared in very low yields. Second, undesired byproduct 17
optimization of diacylglycerol acyltransferase 1 (DGAT1) inhibitors.
■
RESULTS AND DISCUSSION DGAT1 is a transmembrane protein that effects the transformation of diglycerides into triglycerides for energy storage and is a potential target for metabolic disease.19,20 Multiple small molecule inhibitors of DGAT1 have been reported,21 and clinical studies have demonstrated significant lowering of postprandial plasma triglycerides by small molecule DGAT1 inhibition,22−24 although gastrointestinal side-effects have so far complicated the development of drugs for this target.25 During our investigation of DGAT1 inhibitors,26−28 we became interested in a series of benzimidazoles (Figure 1). An
Figure 1. A modular synthetic approach to DGAT1 inhibitors allowed initial development of structure−activity relationships but reaction failures meant many targeted analogues were not tested for DGAT1 activity.
early entry in the benzimidazole series (5) was synthesized via oxidative cyclization of diamine 1 with 2-chloronicotinaldehyde (2) to generate 3. Next, SNAr displacement of the chloride with amine spiropiperidine 4, by heating to 140 °C overnight in N,N-dimethylacetamide (DMA) with N,N-diisopropylethylamine (DIPEA), gave piperidinylbenzimidazole 5, which inhibited DGAT1-catalyzed triglyceride formation with an IC50 of 52 nM. The potency and synthetic modularity of 5 encouraged further exploration of this piperidinylbenzimidazole B
DOI: 10.1021/acs.jmedchem.6b01543 J. Med. Chem. XXXX, XXX, XXX−XXX
Journal of Medicinal Chemistry
Article
Figure 3. High-throughput reaction screening to generate improved SNAr conditions. (A) Reaction scheme to produce 19. (B) Glass vials used to facilitate high-throughput screening. The current studies used 4 mm × 21 mm microvials with 20 μL and ∼1 mg per reaction. (C) Results of solvent-based screen for the reaction shown in (A), represented by UVproduct/UVinternal standard.
∼0.02 mg of substrate per reaction in plastic 1,536-well microtiter plates.8 At our company, these high-throughput technologies have been successfully applied in reaction optimization,33−38 catalyst design, 39−41 reaction discovery,13,42,43 late-stage functionalization,44 reaction informatics studies,45 analytical development,46 and synthetic route scouting.40 In the present medicinal chemistry campaign, we used our glass microvial platform, wherein reaction arrays were run in 24-well aluminum reactor blocks fitted with modified glass HPLC insert vials (4 mm × 21 mm). Each reaction contained 20 μL of solvent and ∼1 mg of starting material (3 μmol per reaction). Reactions were set up by manual pipetting in a nitrogen-filled glovebox and agitated with small magnetic stir bars, and reaction analysis was achieved by UPLC-MS.47 We first investigated the reaction of spiropiperidine 18 with 2-fluoropyridine 16 to produce 19 (Figure 3A). We expected that use of a mild base and lower reaction temperature would improve substrate scope, so a survey of DBU, DIPEA, K2CO3, Cs2CO3, KOAc, and NaHCO3 was performed at 110 °C as an array against four diverse solvents, DMSO, NMP, CPME, and t AmOH, which would solubilize druglike substrates. NMP was selected as a surrogate to DMA in an effort to mitigate the formation of dimethylamine adducts 17. The best single conditions for coupling 16 to 18 used CPME as solvent and DIPEA as base (Figure 3C). Although it was tempting to carry out the remainder of our studies with these conditions alone, we noted that NMP, DMSO, and tAmOH outperformed CPME as a solvent when considering the average conversion across all six bases (Figure 3C). Similarly, from a
Figure 2. (A) Initial structure−activity relationships were explored using a suboptimal SNAr reaction. Only 11 out of 28 targeted analogues were prepared in >10% yield using the first three reaction protocols explored. (B) Compounds 14 and 15 isolated from these explorations were potent DGAT1 inhibitors but required further optimization. Solubility was measured at pH 2. (C) Byproduct 17 was observed in these reactions.
(Figure 2C) was commonly observed, likely the result of dimethylamine being generated by heating DMA at high temperatures for prolonged periods.32 Finally, four excess equivalents of DIPEA were used as base, and reactions were typically purified by direct injection of the homogeneous crude reaction onto a reverse-phase HPLC; occasionally, the large excess of soluble base complicated successful product isolation. For this reason, an alternate base that could be removed by filtration was desired. We therefore applied high-throughput experimentation techniques to optimize reaction conditions toward these attributes and enable library synthesis with diverse electrophile and nucleophile coupling partners. High-throughput experimentation (HTE) allows for facile and systematic evaluation of reaction space. Our efforts to continually explore high-throughput experimentation in drug discovery have led to high-throughput protocols that use ∼4 mg of substrate per reaction in small glass vials,12 ∼1 mg of substrate per reaction in glass microvials (Figure 3B),33 and C
DOI: 10.1021/acs.jmedchem.6b01543 J. Med. Chem. XXXX, XXX, XXX−XXX
Journal of Medicinal Chemistry
Article
holistic standpoint, DIPEA, KOAc, and NaHCO3 were all productive bases showing good conversion to product in at least three different solvents. Our objective was to locate leadoriented conditions capable of delivering diverse products.9,10 Thus, rather than moving forward with CPME-DIPEA, the best performing conditions from a single substrate pair, we executed a parallel-in-parallel evaluation of multiple reaction conditions against multiple substrate pairs aiming to learn if one protocol would stand out for its overall performance across multiple points in chemical space.48 An array of four reaction conditions was targeted comprised of two solvents and two bases. We selected NMP as a solvent to test alongside CPME as high conversion was observed across a range of bases, and for solubility, and we selected NaHCO3 to test alongside DIPEA due to high conversion to product observed (in DMSO, t AmOH, and NMP) and for its previously mentioned desirable modest solubility, which allows for removal of excess base by filtration. Thus, 12 diverse SNAr transformations, namely (16 × 20−25) and (4 × 26−30,16), were subjected to four reaction conditions, CPME-DIPEA, CPME-NaHCO3, NMP-DIPEA, and NMP-NaHCO3, selected from the previous screen. The forty-eight reactions were set up in an inert atmosphere glovebox in glass microvials at 3 μmol scale (∼1 mg per reaction) and analyzed by UPLC-MS (Figure 4). Because the UV extinction coefficients cannot be practically measured in high-throughput, the analytical method used offers only a semiquantitative readout but nonetheless provides a quick estimate of relative reaction performance. In reactions with electrophile 16, diversely functionalized nucleophiles were investigated including two pyrrolidines (20, 21), a piperidinium chloride salt (22), a primary amine (23), an azetidinium chloride salt (24), and a spiropiperidine bearing a cyclic carbamate (25). Likewise, in reactions with nucleophile 4, diverse pyridyl fluorides were selected bearing ketone (26), acetamide (27), or iodide (29) functionality as well as different heterocycles (16, 28, 30). Both free bases and hydrochloride reagent salt forms (22, 24) were included to better approximate the salt diversity of our compound collection, and we reasoned that the excess of base used would neutralize any acidic salts. Surprisingly, the top-performing conditions located in the previous screen using 16 and 18 as substrates, CPME-DIPEA (Figure 3C), were not the “best” conditions and produced only six out of 12 possible products in any appreciable amount. Meanwhile, both DIPEA and NaHCO3 in NMP produced 10 out of 12 products with acceptable conversion (Figure 4). NMP was an attractive solvent for its ability to solubilize many druglike compounds and was selected as the preferred solvent for future studies. NaHCO3 and DIPEA produced compounds with comparable conversion to product, so NaHCO3 was selected as the preferred base for future studies because it is cheap, mild, safe, and is only partially soluble in NMP. The excess base could therefore be readily removed by filtration and the filtrate submitted directly to preparative HPLC purification. Ultimately, our optimal reaction conditions for the preparation of new piperidinylbenzimidazole analogues involved heating overnight at 110 °C in NMP with an excess of NaHCO3 as the base. Importantly, compounds 8 and a saponified analogue of 11, which failed synthesis under the previous conditions shown in Figure 1, were successfully prepared using this protocol (see Table 1 and Supporting Information). With good reaction conditions in hand, a library was constructed using diverse amine building blocks available commercially or from our compound collection. Modern
Figure 4. Micromole-scale parallel-in-parallel reaction optimization. Screening enabled the discovery of robust reaction conditions for library synthesis.
parallel synthesis tools were used for both synthesis and purification,49 30 mg of 16 was used per reaction, and 35 out of 43 library members were isolated in high purity and sufficient yield to deliver to the assay (31−65, Table 1). From the library, 12 compounds exhibited potency against DGAT1 with IC50 values below 100 nM (31−36, 40, 43, 58, 62, 63, 65). The most potent compound, 35, was attractive for its low molecular weight (472 g/mol) and thus presented a good starting point for lead optimization. Upon further in vitro profiling, we learned of several attributes of 35 that required optimization (Table 2). First, the solubility of the compound in acidic media (
