Silicon constitutes almost 30% of the mass of Earth’s crust, yet no life form is known to have the ability to forge carbon–silicon bonds (1). Despite the absence of organosilicon compounds in the biological world, synthetic chemistry has enabled us to appreciate the distinctive and desirable properties that have led to their broad applications in chemistry and material science (2,3). As a biocompatible carbon isostere, silicon can also be used to optimize and repurpose the pharmaceutical properties of bioactive molecules (4,5).
The natural supply of silicon may be abundant, but sustainable methods for synthesizing organosilicon compounds are not (6–8). Carbon–silicon bond-forming methods that introduce silicon motifs to organic molecules enantioselectively rely on multistep synthetic campaigns to prepare and optimize chiral reagents or catalysts; precious metals are also sometimes needed to achieve the desired activity (9–19). Synthetic methodologies such as carbene insertion into silanes can be rendered enantioselective using chiral transition metal complexes based on rhodium (11,12), iridium (13), and copper (14,15). These catalysts can provide optically pure products, but not without limitations: They require halogenated solvents and sometimes low temperatures to function optimally and have limited turnovers (<100) (16).
Because of their ability to accelerate chemical transformations with exquisite specificity and selectivity, enzymes are increasingly sought-after complements to, or even replacements for, chemical synthesis methods (17,18). Biocatalysts that are fully genetically encoded and assembled inside of cells are readily tunable with molecular biology techniques. They can be produced at low cost from renewable resources in microbial systems and perform catalysis under mild conditions. Although nature does not use enzymes to form carbon–silicon bonds, the protein machineries of living systems are often “promiscuous”—that is, capable of catalyzing reactions distinct from their biological functions. Evolution, natural or in the laboratory, can use these promiscuous functions to generate catalytic novelty (19–21). For example, heme proteins can catalyze a variety of non-natural carbene-transfer reactions in aqueous media, including N–H and S–H insertions, which can be greatly enhanced and made exquisitely selective by directed evolution (22–24).
We hypothesized that heme proteins might also catalyze carbene insertion into silicon–hydrogen bonds. Because iron is not known to catalyze this transformation (25), we first examined whether free heme could function as a catalyst in aqueous media. Initial experiments showed that the reaction between phenyldimethylsilane and ethyl 2-diazopropanoate (Me-EDA) in neutral buffer (M9-N minimal medium, pH 7.4) at room temperature gave racemic organosilicon product3at very low levels, a total turnover number (TTN) of 4 (Fig. 1A). No product formation was observed in the absence of heme, and the organosilicon product was stable under the reaction conditions.
We next investigated whether heme proteins could catalyze the same carbon–silicon bond-forming reaction. Screening a panel of cytochrome P450 and myoglobin variants, we observed product formation with more turnovers compared to the hemin and hemin with bovine serum albumin (BSA) controls, but with negligible enantioinduction (table S4). Cytochrome c fromRhodothermus marinus(Rmacyt c), a Gram-negative, thermohalophilic bacterium from submarine hot springs in Iceland (26), catalyzed the reaction with 97% enantiomeric excess (ee), indicating that the reaction took place in an environment where the protein exerted excellent stereocontrol. Bacterial cytochromes c are well-studied, functionally conserved electron-transfer proteins that are not known to have any catalytic function in living systems (27). Other bacterial and eukaryotic cytochrome c proteins also catalyzed the reaction, but with lower selectivities. We thus choseRmacyt c as the platform for evolving a carbon–silicon bond-forming enzyme.
The crystal structure of wild-typeRmacyt c [Protein Data Bank (PDB) ID: 3CP5; (26)] reveals that the heme prosthetic group resides in a hydrophobic pocket, with the iron axially coordinated to a proximal His (H49) and a distal Met (M100), the latter of which is located on a loop (Fig. 1, B and C). The distal Met, common in cytochrome c proteins, is coordinatively labile (28,29). We hypothesized that M100 must be displaced upon iron-carbenoid formation, and that mutation of this amino acid could facilitate formation of this adventitious “active site” and yield an improved carbon–silicon bond-forming biocatalyst. Therefore, a variant library made by site-saturation mutagenesis of M100 was cloned and recombinantly expressed inEscherichia coli.After protein expression, the bacterial cells were heat-treated (75°C for 10 min) before screening in the presence of phenyldimethylsilane (10 mM), Me-EDA (10 mM), and sodium dithionite (Na2S2O410 mM) as a reducing agent, at room temperature under anaerobic conditions. The M100D mutation stood out as highly activating: This first-generation mutant provided chiral organosilicon3as a single enantiomer in 550 TTN, a 12-fold improvement over the wild-type protein (Fig. 1D).
Amino acid residues V75 and M103 reside close (within 7Å) to the iron heme center in wild-typeRmacyt c. Sequential site-saturation mutagenesis at these positions in the M100D mutant led to the discovery of triple-mutant V75T M100D M103E, which catalyzed carbon–silicon bond formation in >1500 turnovers and >99% ee. This level of activity is more than 15 times the total turnovers reported for the best synthetic catalysts for this class of reaction (16). As standalone mutations, both V75T and M103E are activating for wild-typeRmacyt c, and the beneficial effects increase with each combination (table S5). Comparison of the initial reaction rates established that each round of evolution enhanced the rate: Relative to the wild-type protein, the evolved triple mutant catalyzes the reaction more than seven times faster, with turnover frequency (TOF) of 46 min–1(Fig. 1E).
Assaying the new enzyme against a panel of silicon and diazo reagents, we found that the mutations were broadly activating for enantioselective carbon–silicon bond formation. The reaction substrate scope was surveyed with the use of heat-treated lysates ofE. coli–expressingRmacyt c V75T M100D M103E under saturating conditions for both silane and diazo ester to determine TTN. Whereas many natural enzymes excel at catalyzing reactions on only their native substrates and little else (especially primary metabolic enzymes), the triple mutant catalyzed the formation of 20 silicon-containing products, most of which were obtained cleanly as single enantiomers, demonstrating the broad substrate scope of this reaction with just a single variant of the enzyme (Fig. 2). The reaction accepts both electron-rich and electron-deficient silicon reagents, accommodating a variety of functional groups including ethers, aryl halides, alkyl halides, esters, and amides (5to10). Silicon reagents based on naphthalenes or heteroarenes (11to13), as well as vinyldialkyl- and trialkylsilanes, could also serve as silicon donors (14,15,18). In addition, diazo compounds other than Me-EDA could be used for carbon–silicon bond formation (16,17) (16).
The evolvedRmacyt c exhibits high specificity for carbon–silicon bond formation. Even in the presence of functional groups that could compete in carbene-transfer reactions, enzymatic carbon–silicon bond formation proceeded with excellent chemoselectivity. For example, styrenyl olefins, electron-rich double bonds, and terminal alkynes that are prime reaction handles for synthetic derivatization are preserved under the reaction conditions, with no competing cyclopropanation or cyclopropenation activity observed. As a result, organosilicon products12and13and18to20were afforded with 210 to 5010 turnovers and excellent stereoselectivities (98 to >99% ee). Preferential carbon–silicon bond formation could also be achieved with substrates bearing free alcohols and primary amines, yielding silicon-containing phenol21(910 TTN, >99% ee) and aniline22(8210 TTN, >99% ee). This capability removes the need for functional-group protection and/or manipulation, offering a streamlined alternative to transition-metal catalysis for incorporating silicon into small molecules. Indeed, when the same reactants were subjected to rhodium catalysis [1 mol % Rh2(OAc)4], O–H and N–H insertions were the predominant reaction pathways, and copper catalysis [10 mol % Cu(OTf)2] gave complex mixtures of products (table S7). Tolerance of these highly versatile functionalities in enzymatic carbon–silicon bond-forming reactions provides opportunities for their downstream processing through metabolic engineering, bioorthogonal chemistry, and other synthetic endeavors.
We next asked whether allRmacyt c variants would catalyze carbon–silicon bond formation selectively over insertion of the carbene into an N–H bond in the same substrate. We reexamined the evolutionary lineage and tested all four generations ofRmacyt c (wild-type, M100D, V75T M100D, and V75T M100D M103E) with Me-EDA and 4-(dimethylsilyl)aniline (23), a reagent that could serve as both nitrogen and silicon donor, to probe the proteins’ bond-forming preferences. The wild-type cytochrome c exhibited a slight preference for forming amination product24over organosilicon product22. Even though silane23was not used for screening, and theRmacyt c, therefore, did not undergo direct selection for chemoselectivity, each round of evolution effected a distinct shift from amination to carbon–silicon bond-forming activity (Fig. 3A). This evolutionary path that focused solely on increasing desired product formation culminated in a catalyst that channeled most of the reactants (97%) through carbon–silicon bond formation (>30-fold improved with respect to the wild type), presumably by improving the orientation and binding of the silicon donor.
Some fungi, bacteria, and algae have demonstrated promiscuous capacities to derivatize organosilicon molecules when these substances were made available to them (1). The possibility of ultimately establishing silicon-based biosynthetic pathways led us to investigate whether the evolvedRmacyt c could produce organosilicon products in vivo.E. coliwhole cells [optical density at 600 nm (OD600) = 15] expressingRmacyt c V75T M100D M103E in glucose-supplemented M9-N buffer were given silane23(0.1 mmol) and Me-EDA (0.12 mmol) as neat reagents. The enzyme in this whole-cell system catalyzed carbon–silicon bond formation with 3410 turnovers, yielding organosilicon product22in 70% isolated yield (>95% yield based on recovered silane23) and 98% ee (Fig. 3B). These in vitro and in vivo examples of carbon–silicon bond formation using an enzyme and Earth-abundant iron affirm the notion that nature’s protein repertoire is highly evolvable and poised for adaptation: With only a few mutations, existing proteins can be repurposed to efficiently forge chemical bonds not found in biology and grant access to areas of chemical space that living systems have not explored.
SUPPLEMENTARY MATERIALS
www.sciencemag.org/content/354/6315/1048/suppl/DC1
Materials and Methods
Figs. S1 to S6
Tables S1 to S7
REFERENCES AND NOTES
1. M. B. Frampton, P. M. Zelisko, Silicon 1, 147–163 (2009).
2. Z. Rappoport, Y. Apeloig, Eds., The Chemistry of Organic Silicon Compounds (Wiley, 2003), vol. 3.
3. S. A. Ponomarenko, S. Kirchmeyer, Adv. Polym. Sci. 235, 33–110 (2011).
4. G. A. Showell, J. S. Mills, Drug Discov. Today 8, 551–556 (2003).
5. A. K. Franz, S. O. Wilson, J. Med. Chem. 56, 388–405 (2013).
6. P. T. Anastas, J. Warner, Green Chemistry: Theory and Practice (Oxford Univ. Press, New York, 1998).
7. A. M. Tondreau et al., Science 335, 567–570 (2012).
8. A. A. Toutov et al., Nature 518, 80–84 (2015).
9. B. Marciniec, Ed., Hydrosilylation: A Comprehensive Review on Recent Advances (Springer, Netherlands, 2009).
10. T. Lee, J. F. Hartwig, Angew. Chem. Int. Ed. 55, 8723–8727 (2016) and references therein.
11. R. Sambasivan, Z. T. Ball, J. Am. Chem. Soc. 132, 9289–9291 (2010).
12. D. Chen, D.-X. Zhu, M.-H. Xu, J. Am. Chem. Soc. 138, 1498–1501 (2016).
13. Y. Yasutomi, H. Suematsu, T. Katsuki, J. Am. Chem. Soc. 132, 4510–4511 (2010).
14. Y.-Z. Zhang, S.-F. Zhu, L.-X. Wang, Q.-L. Zhou, Angew. Chem. Int. Ed. 47, 8496–8498 (2008).
15. S. Hyde et al., Angew. Chem. Int. Ed. 55, 3785–3789 (2016).
16. See supplementary materials for details.
17. U. T. Bornscheuer et al., Nature 485, 185–194 (2012).
18. J. L. Tucker, M. M. Faul, Nature 534, 27–29 (2016).
19. P. J. O’Brien, D. Herschlag, Chem. Biol. 6, R91–R105 (1999).
20. S. D. Copley, Curr. Opin. Chem. Biol. 7, 265–272 (2003).
21. O. Khersonsky, D. S. Tawfik, Annu. Rev. Biochem. 79, 471–505 (2010).
22. P. S. Coelho, E. M. Brustad, A. Kannan, F. H. Arnold, Science 339, 307–310 (2013).
23. Z. J. Wang, N. E. Peck, H. Renata, F. H. Arnold, Chem. Sci. 5, 598–601 (2014).
24. V. Tyagi, R. B. Bonn, R. Fasan, Chem. Sci. 6, 2488–2494 (2015).
25. Only stoichiometric iron carbenoid insertion into Si–H bonds has been reported (30).
26. M. Stelter et al., Biochemistry 47, 11953–11963 (2008).
27. J. G. Kleingardner, K. L. Bren, Acc. Chem. Res. 48, 1845–1852 (2015).
28. B. D. Levin, K. A. Walsh, K. K. Sullivan, K. L. Bren, S. J. Elliott, Inorg. Chem. 54, 38–46 (2015).
29. S. Zaidi, M. I. Hassan, A. Islam, F. Ahmad, Cell. Mol. Life Sci. 71, 229–255 (2014).
30. E. Scharrer, M. Brookhart, J. Organomet. Chem. 497, 61–71 (1995).
ACKNOWLEDGMENTS
This work was supported in part by the National Science Foundation, Office of Chemical, Bioengineering, Environmental and Transport Systems SusChEM Initiative (grant CBET-1403077); the Caltech Innovation Initiative (CI2) Program; and the Jacobs Institute for Molecular Medicine at Caltech. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the funding organizations. R.D.L. is supported by a NIH–National Research Service Award training grant (5 T32 GM07616). We thank A. Buller, S. Dodani, S. Hammer, and C. Prier for helpful discussions and comments on the manuscript and N. Peck for screening P450 variants. We are grateful to S. Virgil and the Caltech Center for Catalysis and Chemical Synthesis and to N. Torian and the Caltech Mass Spectrometry Laboratory for generous analytical support; the Beckman Institute Laser Resource Center (BILRC) at Caltech for use of their CD spectrometer; B. Stoltz for use of the polarimeter; and H. Gray for providing the pEC86 plasmid. A provisional patent application has been filed through the California Institute of Technology based on the results presented here. All data necessary to support this paper’s conclusions are available in the supplementary materials
