High-throughput screening of drug leads

High-throughput experimentation (HTE) chemistry tools have been used to aid in the discovery of new reactions (1–6) and in the scale-up optimization of known reactions (7–9), both areas where substrates for experimentation are plentiful. However, HTE is rarely used in the area where it might have the greatest impact: the synthesis of complex natural products or highly functionalized drug leads. What is needed is a tool that would allow chemists to locate successful reaction conditions (10) on a microgram (nanomole) scale in a high-throughput fashion without depleting precious substrate stores.

In the search for breakthrough medicines in biomedical research, the rapid preparation of new, complex molecules for biological evaluation is of paramount importance, but substrates for the synthesis of such compounds are invariably in short supply. In later stages of chemistry development, substrates are abundant, and state-of-the-art microvial (8, 9) or microfluidic (11) HTE tools can be effective in “turning on” reactions that were otherwise unsuccessful by exploring combinations of catalysts, reagents, and other key reaction variables (12, 13). Such studies require at least milligram (micromole) quantities of substrate per reaction—a prohibitively large amount in early drug discovery, where new molecules are prepared for the first time. Consequently, the set of desirable compounds designed to test a biological hypothesis is often winnowed to the much smaller subset of compounds that can be successfully synthesized using a single set of reaction conditions, with little opportunity to study and improve unsuccessful syntheses (14, 15). Miniaturizing chemistry to the nanomole scale is a potential solution to this problem that has heretofore met with substantial engineering problems, such as accurately bringing together extremely small charges of materials that are often heterogeneous, effectively agitating reaction mixtures, preventing loss of volatile solvents, and incorporating general analytical approaches to assay reaction outcomes. We present the first results of a study aimed at developing general nanomole reaction screening capabilities to support the rapid synthesis of complex, highly functionalized drug leads.

Figure 1A shows highly functionalized molecules from Merck’s compound collection representative of the cores of drug molecules (1 to 8), which in the search for new molecules with optimal biological properties would be coupled to a diversity of polar building blocks (9 to20) such as those in Fig. 1B. Modern organic synthesis (and especially transition metal catalysis) is redefining the rules with which new bonds can be forged; however, it is important to recognize that many “solved” synthetic transformations are far from universal, performing well on simple model substrates yet often failing when applied to complex substrates in real-world synthesis (16). A recent analysis of 2149 metal-catalyzed C-N couplings run in the final steps of the synthesis of highly functionalized drug leads revealed that 55% of reactions failed to deliver any product at all (17). The missed opportunity represented by these unsuccessful syntheses haunts contemporary drug discovery, and there is a growing recognition that the tendency of polar, highly functionalized compounds to fail in catalysis may actually enrich compound sets in hydrophobic molecules that are less likely to become successful drug candidates (15, 16, 18–20).

Soluble superbase coupling conditions for miniaturization from micromole to nanomole scale

Using the catalysis cross-coupling space defined in Fig. 1, A and B, as a test case, this work evaluates the idea that miniaturized chemistry experimentation to inform preparative scale synthesis (Fig. 1C) could be run on the same scale as contemporary biological assays using the high-precision nanoliter robotics commonly used in biochemistry labs (Fig. 1D) in conjunction with high-throughput mass spectrometric analysis (Fig. 1E). The engineering complexities of this task could potentially be overcome by using ambient-temperature chemistry that operates in a solubilizing, low-volatility, plastic-compatible reaction medium such as dimethyl sulfoxide (DMSO) or N-methyl-2-pyrrolidone (NMP). We chose DMSO for the initial proof of concept because it is an environmentally friendly solvent and represents the potential to directly link chemistry with biochemical experimentation. Very few efficient ambient-temperature palladium-catalyzed cross-couplings in DMSO or other polar solvents have been reported (21), in part because of the strong solvent-metal coordination of such solvents and their incompatibility with the strong bases typically used in low-temperature C-N couplings (NaOtBu, LiN(SiMe3)2, Zn(N(SiMe3)2)2 etc.) (22). It seemed possible that recently reported highly hindered, electron-rich ligands (23) may effectively protect Pd from DMSO coordination and that non-nucleophilic organic superbases (24), which are soluble in most solvents and should be strongly basic enough to promote Pd C-N couplings at room temperature, could be compatible with these solvents.

Initially working in glass microvials, which represent the current validated lower limits of scalable batch miniaturization (8, 9), the coupling of amine 21 with bromide 22 (2.5 μmol in 25 μl of DMSO) was evaluated in a 96-reaction array of superbases 24 to 29 with catalysts 30 to45 [10 mole percent (mol %)] (Fig. 1F), and several highly productive room-temperature coupling conditions were identified. All solutions were homogeneous but were stirred nonetheless. This same 96-reaction array was then evaluated on a 1.0-μl scale (100 nmol of substrate) in a plastic 1536-well plate without stirring and with reagent dosing via a nanoliter liquid-handling robot (TTP LabTech Mosquito HTS) inside a nitrogen-filled glovebox, demonstrating the same hits and reaction performance (Fig. 1G). The Mosquito robot allows sequential aspiration from different wells of a source plate into a single pipette tip to create multicomponent mixtures that are dosed as single reaction drops into a 1536-well plate. This ensures proper mixing of the reaction components and obviates the need for stirring. This pipetting mechanism, in conjunction with a material-sparing source plate with minimal dead volume, ensures that almost no material is wasted, and the overage can be recovered if desired. Also, with this batch reaction approach, some degree of sample heterogeneity can be tolerated, unlike for microfluidic technologies where channel clogging can lead to systemic failure.

The coupling of bromide 22 was then evaluated with 16 nitrogen, oxygen, carbon, phosphorus, and sulfur nucleophiles under the same 96 combinations of 24 to 45 in a single miniaturized 1536-reaction experiment (Fig. 2A and table S3). In this experiment, dosing required 30 min (four components per well, 6144 total reagent doses) followed by 22 hours of reaction time, 1 hour for automated reaction quenching and sampling, and 52 hours for analysis by ultra performance liquid chromatography (UPLC). This experiment uncovered extremely mild, soluble coupling conditions for amines, alcohols, amides, sulfonamides, carbamates, amidines, aryl boronates, alkynes, and malonates, most of which have not been reported at ambient temperature with non-nucleophilic bases (table S3). Because the reactions were by design homogeneous and required no heating or cooling, translation to a larger scale was straightforward, delivering 68 to 98% yield when scaled up by a factor of 3000, to 320 μmol of22 (50 mg), now running at 5 mol % catalyst loading and 0.2 M substrate concentration.

Implementation of nanomole screening for complex molecule synthesis

These new coupling conditions were next applied in complex synthesis to the test set of electrophiles 1 to 8 and nucleophiles 9 to 20 (Fig. 1, A and B). Initial synthesis conditions for this array were selected from the best permutation of superbase and catalyst that was previously identified for each nucleophile class when using bromide 22 (table S3). With these conditions, high-purity product samples were successfully prepared for 54 of the 96 functionality-rich substrate combinations (Fig. 1C and table S4) when run on a 500-μmol scale (~25 mg) (25, 26). This chemistry was generally tolerant of polar functional groups and heterocycles; however, nearly half of the 96 reactions failed to provide any product, which is typical for application of a single set of conditions to a complex catalysis problem.

To apply nanomole-scale reaction screening to this problem, we subjected 32 different electrophile-nucleophile substrate combinations, each of which produced little or no material from the 96-reaction array, to 48 reaction conditions (8 organic superbases × 6 ligands; Fig. 2B). The substrate concentration was reduced to 0.05 M and the catalyst loading was doubled to 20 mol % so that running all 48 reactions would consume less than 1 mg of each substrate (50 nmol, ~0.02 mg per reaction) (27). In addition, the reaction time was reduced to 2 hours and a faster liquid chromatography–mass spectrometry (LC-MS) approach was pursued, with a goal of running and analyzing 1536 reactions in less than 24 hours. Of the available label-free techniques (28), MISER (multiple injections in a single experimental run) LC-MS was selected for its ease of data acquisition and analysis (29). Thus, 48 reactions that had the same potential product along a row of the 1536-well plate were analyzed by multiple injections (22 s per sample) in a single isocratic chromatographic run with mass detection settings observing the molecular ion for the desired product (Fig. 2C). In this way, the 1536-well plate could be analyzed in ~9 hours. To reduce the time required for analysis even further, we combined four rows of reaction samples that could give different product masses and simultaneously monitored the four desired products’ m/z values (mass/charge ratios) in these pooled samples (Fig. 2C). By pooling four mass-encoded wells, the analysis of all 1536 reactions could be performed in ~2.5 hours. In addition, a more rigorous approach to this same analysis was pursued using a parallel two-channel LC system connected to a triple-quadrupole mass spectrometer. This instrumentation used a fast gradient to provide improved resolution and additional MS/MS structure confirmation data and required ~2.5 hours for analysis of all 1536 reactions. The MISER analyses located the expected product masses for 21 of 32 substrate combinations, and the best hits were then confirmed with UPLC-MS. These 21 hits were scaled up by a factor of 1000 (25-mg scale), and a pure product sample was obtained for 16 of these reactions (table S6).

Optimization of continuous variables (for example, the stoichiometry of reagents) is another potential path to increase product formation in HTE. Returning to the conditions originally identified for bromide 22, six of the remaining failed reactions were evaluated under 48 combinations of catalyst loading, nucleophile charge, and base charge, again using 1 mg of substrate per 48-reaction evaluation (fig. S22). Markedly better conditions for five of the six compounds were found in this way, which allowed their isolation upon scale-up by a factor of 1000 (table S8). Hence, in these two miniaturized screening experiments, looking first at catalyst-base combinations and then at reaction stoichiometries, 21 complex new compounds could be rapidly prepared. These solutions to 21 complex synthesis problems represent the opportunity to answer biological hypotheses that would have been passed over in previous chemistry approaches because of lack of time and material.

Nanomole reaction optimization to inform gram-scale synthesis

An important consideration for this nanomole-scale approach toward discovery synthesis is the optimization of reactions for application to larger-scale synthesis. Compounds that are deemed interesting in initial biological evaluation or are synthetic intermediates in a multistep route must rapidly be prepared in larger quantities once conditions are identified. Statistical tools such as DOE (design of experiments) are powerful tools for synthesis (30–32), yet they typically focus on performing minimal numbers of experiments, as large numbers of experiments are often resource- and time-intensive to conduct. However, because large numbers of experiments are readily feasible with this nanomole-scale chemistry platform, we were able to construct a three-factorial, four-level response surface modeling experiment to study the loading of catalyst against varying stoichiometries of base and nucleophile in the reaction of chloride 6 with amine 10. In this DOE experiment, each condition was repeated twice, resulting in 128 total reactions with <3 mg of 6. Indeed, a high-quality response surface model was generated with the nanomole-scale chemistry approach (Fig. 2D), which helped to define the critical charges of nucleophile and base for optimal reaction performance. The optimized conditions used 15 mol % 42 at 0.05 M concentration; by translating to more practical conditions of 5 mol % 42 and 0.24 M concentration, we obtained full conversion and a 79% isolated yield of 46 on a 25-mg scale, which was reproduced to obtain a 76% isolated yield on a 1-g scale (Fig. 2D). This result shows that advanced statistical reaction analysis, which is typically reserved for chemistry opportunities where material is plentiful, can be applied to reactions in the material-limited front lines of drug discovery or natural product synthesis.

In biomedical research, chemical synthesis should not limit access to any molecule that is designed to answer a biological question. This work demonstrates an example of how conditions for complex Pd-catalyzed C-O, C-N, and C-C cross-coupling reactions can be evolved into a powerful, substrate-focused approach to chemistry miniaturization to overcome limited access to complex products. With innovative research, other high-value modern chemistry reactions could be similarly designed into this paradigm to improve synthesis in material-limited environments by evolution of catalysts and reagents to perform in DMSO, NMP, or other high-boiling solvents at ambient temperature.

Supplementary Materials

www.sciencemag.org/content/347/6217/49/suppl/DC1

Materials and Methods

Supplementary Text

Figs. S1 to S32

Tables S1 to S11

References (33–40)

Data Files S1 to S5

References and Notes

1. M. R. Friedfeld et al., Science 342, 1076–1080 (2013).

2. D. A. DiRocco et al., Angew. Chem. Int. Ed. 53, 4802–4806 (2014).

3. D. W. Robbins, J. F. Hartwig, Science 333, 1423–1427 (2011).

4. A. McNally, C. K. Prier, D. W. C. MacMillan, Science 334, 1114–1117 (2011).

5. K. D. Collins, T. Gensch, F. Glorius, Nat. Chem. 6, 859–871 (2014).

6. R. Moreira, M. Havranek, D. Sames, J. Am. Chem. Soc. 123, 3927–3921 (2001).

7. S. M. Preshlock et al., J. Am. Chem. Soc. 135, 7572–7582 (2013).

8. A. Bellomo et al., Angew. Chem. Int. Ed. 51, 6912–6915 (2012).

9. J. R. Schmink, A. Bellomo, S. Berritt, Aldrichim. Acta 46, 71–80 (2013).

10. M. Peplow, Nature 512, 20–22 (2014).

11. T. Rodrigues, P. Schneider, G. Schneider, Angew. Chem. Int. Ed. 53, 5750–5758 (2014).

12. S. Monfette, J. M. Blacquiere, D. E. Fogg, Organometallics 30, 36–42 (2011).

13. P. M. Murray, S. N. G. Tyler, J. D. Moseley, Org. Process Res. Dev. 17, 40–46 (2013).

14. S. D. Roughley, A. M. Jordan, J. Med. Chem. 54, 3451–3479 (2011).

15. T. W. J. Cooper, I. B. Campbell, S. J. F. Macdonald, Angew. Chem. Int. Ed. 49, 8082–8091 (2010).

16. A. Nadin, C. Hattotuwagama, I. Churcher, Angew. Chem. Int. Ed. 51, 1114–1122 (2012).

17. Merck internal study of electronic notebooks.

18. M. M. Hann, G. M. Keserü, Nat. Rev. Drug Discov. 11, 355–365 (2012).

19. F. Lovering, J. Bikker, C. Humblet, J. Med. Chem. 52, 6752–6756 (2009).

20. H. A. Malik et al., Chem. Sci. 5, 2352–2361 (2014).

21. R. E. Tundel, K. W. Anderson, S. L. Buchwald, J. Org. Chem. 71, 430–433 (2006).

22. D. S. Surry, S. L. Buchwald, Chem. Sci. 2, 27−50 (2011).

23. N. C. Bruno, M. T. Tudge, S. L. Buchwald, Chem. Sci. 4, 916–920 (2013).

24. T. Ishikawa, Y. Kondo, H. Kotsuki, T. Kumamoto, D. Margetic, K. Nagasawa, W. Nakanishi, in Superbases for Organic Synthesis: Guanidines, Amidines, Phosphazenes and Related Organocatalysts, T. Ishikawa, Ed. (Wiley, West Sussex, UK, ed. 1, 2009), pp. 1–326.

25. Compounds were generally purified by MS-directed purification. Isolated yields ranged from 1 to 100%, but we made no attempt to maximize the isolated yields in these reactions and instead focused on obtaining high-purity compounds as quickly as possible, which is typical in most medicinal chemistry campaigns. Some reactions showed product formation by UPLC-MS analysis but were either insufficiently pure or too low in yield for purification.

26. M. Liu et al., ACS Comb. Sci. 14, 51–59 (2012).

27. Some electrophiles were not fully soluble in DMSO, so NMP was used instead. Even though three stock solutions in NMP still displayed mild insolubility, the TTP Mosquito operates on positive-displacement pipetting, so viscous solutions or suspensions of small particulates are easily transferred.

28. W. Schafer, X. Bu, X. Gong, L. A. Joyce, C. J. Welch, in Comprehensive Organic Synthesis, C. J. Welch, Ed. (Elsevier, Oxford, ed. 2, 2014), vol. 9, pp. 28−53.

29. C. J. Welch et al., Tetrahedron Asymmetry 21, 1674–1681 (2010).

30. J. C. Ianni, V. Annamalai, P.-W. Phuan, M. Panda, M. C. Kozlowski, Angew. Chem. Int. Ed. 45, 5502–5505 (2006).

31. S. E. Denmark, C. R. Butler, J. Am. Chem. Soc. 130, 3690–3704 (2008).

32. K. C. Harper, M. S. Sigman, Science 333, 1875–1878 (2011).

Acknowledgements: We thank S. Krska, M. Tudge, G. Hughes, and E. Parmee for helpful discussions; M. Liu, E. Streckfuss, T. Meng, N. Pissarniski, and W. Li for assistance in purification of compounds; M. Christensen and J. Voigt for experimental assistance; and S. M. O’Brien and M. McColgan for graphic design. S.B. was supported by an NSF GOALI Grant associated with the University of Pennsylvania. Supported by the MRL Postdoctoral Research Fellows Program (A.B.S. and E.L.R.).