Breeding Smart Enzymes

FREAKY FUTURE Is silicon-based life (illustrated above) out there in the universe? The jury is still out — but with a little help from humans, life on Earth might be able to sprinkle a little silicon in with its carbon.  LEI CHEN AND YAN LIANG (BEAUTYOFSCIENCE.COM) FOR CALTECH

Editor's Introduction

Directed evolution of cytochrome c for carbon–silicon bond formation: Bringing silicon to life

annotated by

Can we improve the function of enzymes? Can we choose the fittest enzyme to carry out reactions not seen in nature? Can we assign new functions to enzymes? Arnold’s team uses directed evolution to mutate the wild-type enzyme, Rma cyt c. The fourth-generation variant shows exclusive chemoselectivity toward carbon-silicon bond formation, a reaction not known to occur in nature. The variant is very selective and forms only one product with a high turnover number. A mechanism is proposed to explain the enzyme’s catalytic activity. This study offers an environmentally green route to forming organosilicon compounds.

Paper Details

Original title
Directed evolution of cytochrome c for carbon–silicon bond formation: Bringing silicon to life
Original publication date
Vol. 354, Issue 6315, pp. 1048-1051
Issue name


Enzymes that catalyze carbon–silicon bond formation are unknown in nature, despite the natural abundance of both elements. Such enzymes would expand the catalytic repertoire of biology, enabling living systems to access chemical space previously only open to synthetic chemistry. We have discovered that heme proteins catalyze the formation of organosilicon compounds under physiological conditions via carbene insertion into silicon–hydrogen bonds. The reaction proceeds both in vitro and in vivo, accommodating a broad range of substrates with high chemo- and enantioselectivity. Using directed evolution, we enhanced the catalytic function of cytochrome c from Rhodothermus marinus to achieve more than 15-fold higher turnover than state-of-the-art synthetic catalysts. This carbon–silicon bond-forming biocatalyst offers an environmentally friendly and highly efficient route to producing enantiopure organosilicon molecules.


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 (23). As a biocompatible carbon isostere, silicon can also be used to optimize and repurpose the pharmaceutical properties of bioactive molecules (45).

The natural supply of silicon may be abundant, but sustainable methods for synthesizing organosilicon compounds are not (68). 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 (919). Synthetic methodologies such as carbene insertion into silanes can be rendered enantioselective using chiral transition metal complexes based on rhodium (1112), iridium (13), and copper (1415). 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 (1718). 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 (1921). 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 (2224).

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 product 3 at 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.



Fig. 1 Heme protein–catalyzed carbon–silicon bond formation. (A) Carbon–silicon bond formation catalyzed by heme and purified heme proteins. (B) Surface representation of the heme-binding pocket of wild-type Rma cyt c (PDB ID: 3CP5). (C) “Active site” structure of wild-type Rma cyt c showing a covalently bound heme cofactor ligated by axial ligands H49 and M100. Amino acid residues M100, V75, and M103 residing close to the heme iron were subjected to site-saturation mutagenesis. (D) Directed evolution of Rma cyt c for carbon–silicon bond formation [reaction shown in (A)]. Experiments were performed using lysates of E. coli expressing Rma cyt c variant (OD600 = 15; heat-treated at 75°C for 10 min), 10 mM silane, 10 mM diazo ester, 10 mM Na2S2O4, 5 vol % MeCN, M9-N buffer (pH 7.4) at room temperature under anaerobic conditions for 1.5 hours. Reactions were done in triplicate. (E) Carbon–silicon bond forming rates over four generations of Rma cyt c. Single-letter abbreviations for the amino acid residues are as follows: D, Asp; E, Glu; M, Met; T, Thr; and V, Val.

Figure 1A (left)  

epresents the reaction between phenyldimethylsilane and Me-EDA catalyzed by heme protein to form the silicon-carbon bond.

Figure 1A (right)

The table shows the results from the preliminary experiments with heme and heme-purified proteins. It was observed that the reaction catalyzed by R. marinus cyt c resulted in 97% ee and a TTN of 34. Bovine heart cyt c showed the highest TTN of 54 but a very poor 6% ee.

Figure 1B  

The figure shows the active site structure for wild-type Rma cyt c and amino acid residues V75, M100, and M103.

Figure 1C  

The figure shows the enzyme’s active site and the possible approach of the silane molecule.

Figure 1D  

Three trials of experiments were run and the average TTNs are reported. Wild-type (WT) showed a TTN of 44, M100D showed a TTN of 549, V75T M100D showed a TTN of 892, and V75T M100D M103E showed a TTN of 1518. With each evolution, the turnover number increased significantly while still maintaining a high 99% ee.

Figure 1E 

Turnover frequency (TOF) is a measure of the efficiency of the catalyst. The initial rate of the reaction and the turnover frequency with the WT and each variant was determined. The WT had a TOF of 6.4, M100D had a TOF of 17.5, V75T M100D had a TOF of 29.5, and V75T M100D M103E had a TOF of 45.5. The most evolved variant had seven times greater TOF than the wild type.

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 from Rhodothermus marinus (Rma cyt 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 chose Rmacyt c as the platform for evolving a carbon–silicon bond-forming enzyme.

The crystal structure of wild-type Rma cyt 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 (2829). 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 in Escherichia 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 (Na2S2O4 10 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 organosilicon 3 as 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-type Rma cyt 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-type Rma cyt 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 of E. coli–expressing Rma cyt 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 (5 to 10). Silicon reagents based on naphthalenes or heteroarenes (11 to 13), as well as vinyldialkyl- and trialkylsilanes, could also serve as silicon donors (141518). In addition, diazo compounds other than Me-EDA could be used for carbon–silicon bond formation (1617) (16).



Fig. 2 Scope of Rma cyt c V75T M100D M103E-catalyzed carbon–silicon bond formation. Standard reaction conditions: lysate of E. coli expressing Rma cyt c V75T M100D M103E (OD600 = 1.5; heat-treated at 75°C for 10 min), 20 mM silane, 10 mM diazo ester, 10 mM Na2S2O4, 5 vol % MeCN, M9-N buffer (pH 7.4) at room temperature under anaerobic conditions. Reactions performed in triplicate. [a] OD600 = 5 lysate. [b] OD600 = 0.5 lysate. [c] OD600 = 15 lysate. [d] 10 mM silane. [e] OD600 = 0.15 lysate.

Figure 2  

Rma cyt c V75T M100D M103E catalyzed reactions were run with a range of silanes and diazo compounds. The broad range scope of the method is evident in the TTN and enantiomeric excess data. All reactions proceeded with >99% ee and moderate to high TTN.


Reaction a  

Reactions 3-5 show...


Reaction b  

Reactions 8-13 show


Reaction c 

Reactions 14 and 15 show


Reaction d  

Reactions 16 and 17 show


Reaction e  

Reactions 18-22 show

The evolved Rma cyt 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 products 12 and 13 and 18 to 20 were 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 phenol 21 (910 TTN, >99% ee) and aniline 22 (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 all Rma cyt 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 of Rma cyt 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 product 24 over organosilicon product 22. Even though silane 23 was not used for screening, and the Rma cyt 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.



Fig. 3 Chemoselectivity and in vivo activity of evolved Rma cyt c. (A) Chemoselectivity for carbene Si–H insertion over N–H insertion increased markedly during directed evolution of Rma cyt c. Standard reaction conditions as described in Fig. 2. Reactions were performed in duplicate using heat-treated lysates of E. coli expressing Rma cyt c with protein concentration normalized across variants. Product distribution was quantified after 2 hours of reaction time (before complete conversion, no double insertion product was observed under these conditions). (B) In vivo synthesis of organosilicon compound 22.

Figure 3A  

The WT and the variants of Rma cyt c were tested for their chemical preference toward Si-H and N-H bonds. As the enzymes evolved, selectivity toward Si-H insertion increased while that of N-H decreased.

Figure 3B 

The reaction scheme corresponds to a whole-cell biocatalytic reaction. The fourth-generation enzyme was tested for both chemoselectivity and enantiospecificity. The organosilicon product 22 was isolated and purified by column chromatography with a 70% yield. The unreacted silane 23 (starting compound) was also recovered at 26% yield. The enantiomeric excess was determined to be 98% ee by chiral SFC. The TTN for this reaction was 3410.

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 evolved Rma cyt c could produce organosilicon products in vivo. E. coli whole cells [optical density at 600 nm (OD600) = 15] expressing Rma cyt c V75T M100D M103E in glucose-supplemented M9-N buffer were given silane 23 (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 product 22 in 70% isolated yield (>95% yield based on recovered silane 23) 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.



Materials and Methods

Figs. S1 to S6

Tables S1 to S7

References (3167)



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).



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