www.sciencemag.org/cgi/content/full/science.1188934/DC1

Supporting Online Material for

Biocatalytic Asymmetric Synthesis of Chiral Amines from Ketones Applied to Sitagliptin Manufacture

Christopher K. Savile,* Jacob M. Janey,* Emily C. Mundorff, Jeffrey C. Moore, Sarena Tam, William R. Jarvis, Jeffrey C. Colbeck, Anke Krebber, Fred J. Fleitz, Jos Brands,

Paul N. Devine, Gjalt W. Huisman, Gregory J. Hughes

*To whom correspondence should be addressed. E-mail: christopher.savile@codexis.com

(C.K.S.); jacob_janey@merck.com (J.M.J.)

Published 17 June 2010 on Science Express

DOI: 10.1126/science.1188934

This PDF file includes:

Materials and Methods Figs. S1 to S6 Tables S1 to S12 References

S1  

Supporting Online Material

Biocatalytic Asymmetric Synthesis of Chiral Amines from Ketones Applied to Sitagliptin Manufacture

Christopher K. Savile,1† Jacob M. Janey,2† Emily C. Mundorff,1 Jeffrey C. Moore,2 Sarena Tam,1 William R. Jarvis,1 Jeffrey C. Colbeck,1 Anke Krebber,1 Fred J. Fleitz,2 Jos Brands,2 Paul N. Devine,2 Gjalt W. Huisman,1 Gregory J. Hughes2

1 Codexis Inc., 200 Penobscot Drive, Redwood City, CA 94063, USA.

2 Department of Process Research, Merck Research Laboratories, Merck & Co., Inc. Rahway, New Jersey 07065, USA.

Email: christopher.savile@codexis.com, jacob_janey@merck.com

Supporting information

Table of contents S1

ATA-117 homology model development S2

Library designs for establishing activity toward sitagliptin ketone S2

Chronology of accumulated mutations and analysis derived from modeling S3

Round 1a screening results for methyl ketone analog S4

Round 1b and 2 screening results for pro-sitagliptin ketone S5

Directed evolution of a process stable transaminase, rounds 3-11 S7

Quantitation of catalyst improvement S9

Development of process for conversion of pro-sitagliptin ketone to sitagliptin S12

Experimental methods S13

Supporting references S27

S2  

ATA-117 homology model development

Molecular modeling. MOE modeling software (Chemical Computing Group, Montréal, Québec, Canada) was used for all structural modeling and analyses. Three potential templates were identified to model ATA-117: 3DAA (D-amino acid aminotransferase from Bacillus sp. YM-1), 1WRV (branched-chain amino acid aminotransferase from T. thermophilus), and 1AGE (branched-chain amino acid aminotransferase from E. coli). Structural alignment of the three structures gave an r.m.s.d. of 1.4 Å for the C-alpha’s for the full protein and 0.9 Å for the C-alpha’s of the structurally conserved binding pocket. The pair-wise sequence identity of ATA-117 with 3DAA, 1WRV and 1AGE was 28%, 27% and 25%, respectively. A homology model of ATA-117 was generated with the AMBER99 force field using the dimer of 3DAA with bound N-(5'-phosphopyridoxyl)-D-alanine in the binding pocket as template – D-alanine was deemed to be an appropriate ligand to be used for modeling, as it is a substrate for ATA-117. The ATA-117 model had no intramolecular steric conflicts and the parent torsion angles were within the allowed regions in the Ramachandaran phi-psi plot. The pro-sitagliptin substrate was modeled as the internal aldimine intermediate and positioned into the binding pocket using the position of the phosphopyridoxyl in the bound ligand of 3DAA as a guide. Individual mutations were explored using the rotamer explorer function in MOE. Structural models of evolved variants were made using the ATA-117 model as a template.

Library designs for establishing activity toward sitagliptin ketone

Increasing activity toward methyl ketone analog. Modeling of the external aldimine intermediate identified two pockets, a large one that fit the tetrahydro-triazolo[4,3-α]pyrazine (THTP) group, and a small pocket for the trifluorophenyl group. Saturation mutagenesis was performed at all positions predicted to interact with the THTP group: H62, G136, E137, W192, L195, V199, E208, A209, S223P, G224, F225, and T282 and the libraries were screened against the methyl ketone analog. The large pocket is at the dimer interface predominantly comprised of one monomer, but with H62, G136 and E137 expected to be donated from the other monomer.

Opening small pocket to accommodate trifluorophenyl group and establish activity toward pro-sitagliptin ketone. Four residues were predicted to line the small pocket of the binding site: V69, F122, T283, and A284. All of these residues and adjacent F70 were subjected to saturation mutagenesis. In addition, a small combinatorial library of V69GA, F122AVLIG, T283GAS, and A284GF was designed. These mutations were chosen as they could potentially carve out enough space in the small pocket to accommodate the F3C-phenyl moiety. For V69, F122 and T283 and A284G, residues of similar type and smaller sizes were chosen. A284F was included as it appeared possible for aromatic stacking to occur with this position and the phenyl moiety of the substrate.

Increasing activity toward pro-sitagliptin ketone by recombining round 1 beneficial mutations. All beneficial mutations identified in Round 1 with the methyl ketone analog were recombined in a combinatorial library: H62AYFT, I122LV, G136Y, E137ATI, V199WI, A209L, F225Y and T282S and the library was screened against pro-sitagliptin ketone.

S3  

Chronology of accumulated mutations and analysis derived from modeling

The directed evolution program started with site-saturation mutagenesis of the large pocket which generated numerous improved variants, the most active being ATA-117: S223P. This mutation could be either increasing the hydrophobic interaction with the THTP group and/or stabilizing a loop in the binding pocket. The next most improved variant from these libraries, G136Y, appears to be increasing the hydrophobic interaction with the THTP group. The variant identified with activity toward pro-sitagliptin ketone was from a small pocket combinatorial library designed to expand the small pocket and accommodate the trifluorophenyl group. This variant had three programmed mutations (V69G, F122I, and A284G) as well as two random mutations (Y26H and V65A). The three programmed mutations appear to be creating sufficient space to accommodate the trifluorophenyl group. The random mutation Y26H was in an unmodeled portion of the enzyme and V65A was at the dimer interface. The next library combined all beneficial mutations from the large pocket site-saturation mutagenesis libraries with beneficial mutations from the small pocket combinatorial library. The most active variant from this library added an additional small pocket mutation, T282S, as well as five mutations in the large pocket: H62T, G136Y, E137I, V199I, and A209L. The mutations G136Y, E137I, V199I, and A209L all appear to be increasing the hydrophobic interactions with the THTP moiety.

At this point in evolution, numerous library strategies were employed and as beneficial mutations were identified they were added into combinatorial libraries. The entire binding pocket was subjected to saturation mutagenesis in round 3. At position 69, mutations TAS and C were improved over G. This is interesting in two aspects. First, V69A was an option in the small pocket combinatorial library, but was less beneficial than V69G. Second, G69T was improved (and found to be the most beneficial in the next round) suggesting that something other than sterics is involved at this position as it was a Val in the starting enzyme. At position 137, Thr was found to be preferred over Ile. Random mutagenesis generated two of the mutations in the round 3 variant: S8P and G215C. S8P was shown to increase expression and G215C is a surface exposed mutation which may be important for stability. Mutations identified from homologous enzymes identified M94I in the dimer interface as a beneficial mutation. In subsequent rounds of evolution the same library strategies were repeated and expanded. Saturation mutagenesis of the secondary sphere identified L61Y, also at the dimer interface, as being beneficial. The repeated saturation mutagenesis of 136 and 137 identified Y136F and T137E as being improved. E137 was mutated to Ile in round 2, Thr in round 3 and then back to Glu in round 4 – as the conditions and context change so to do the preferred residues, sometimes back to their original. Rounds 5 and 6 focused on diversity from homologous enzymes and random mutagenesis. Homologous diversity identified included: I94L, I96L, L269P, S321P, Y60F (highly conserved in the homologous enzymes), A169L, G217N, and L273Y. Random mutations (D81G, T178S and P297S) were also identified in rounds 5 and 6. Round 7 featured saturation mutagenesis of the secondary sphere binding pocket positions and previously-identified influential positions. These libraries identified S124HNT and I122M in the secondary sphere of the binding pocket and Y150S and V152C, which are both at the dimer interface. Random mutation, Q329H, provided a significant performance increase in round 9, but was found to be deleterious in round 10 when screened in 50% DMSO. Recombining S124HNT and I122M with Y150S, and V152C gave the final variant.

S4  

Round 1a screening results for methyl ketone analog

Libraries: Site-saturation mutagenesis at H62, G136, E137, W192, L195, V199, A208, A209, S223, G224, F225, and T282S.

N

O O

NN

N

F3C

N

O NH2

NN

N

F3C

Transaminase+ +

NH2 O

PLP0.5 M

Table S1. Hits from round 1a Screening. Conditions: 2 g/L substrate, 0.5 M i-PrNH2, 100 µM PLP, 5% DMSO in 100 mM triethanolamine, pH 7.5 at 22°C for 24 h.

Amino acid position and mutation Maximum fold improvement

over ATA-117 observed in screening

H62AFTY 4-fold

G136Y 6-fold

E137AIT 2-fold

W192 --

L195 --

V199WI 5-fold

A208 --

A209L 2-fold

S223P* 11-fold

G224 --

F225Y 4-fold

T282S 3-fold

*Selected as round 1b parent.

Table S2. Retesting of round 1a hits as fermentation powders. Conditions: 10 g/L enzyme, 2 g/L substrate, 0.5 M i-PrNH2, 100 µM PLP, 5% DMSO in 100 mM triethanolamine, pH 7.5 at 22°C for 24 h.

S5  

Variant % Conversion

ATA-117 4

G136Y 18

S223P 38

Round 1b and 2 screening results for pro-sitagliptin ketone

Library: Combinatorial of V69GA, A284GF, F122AVLIG, and T283GAS.

Table S3. Top 10 hits from screening. Conditions: 2 g/L substrate, 0.5 M i-PrNH2, 100 µM PLP, 5% DMSO in 100 mM triethanolamine, pH 7.5 at 22°C for 24 h.

Variant % Conversion

V65A; V69G; F122I; S223P 0.76

Y26H; V65A; V69G; F122I; S223P; A284G 0.71

A31T; V65A; V69G; F122I; S223P 0.52

F122L; S174A; S223P; A284G; 0.49

F122L; S223P; I254V; A284G; 0.43

V69G; F122V; S223P; A284G; 0.37

F122I; S223P; A284G 0.34

F122L; S223P; A284G 0.29

V65G; V69A; F122I; S223P; A284G 0.26

Table S4. Retesting of round 1b hits as fermentation powders. Conditions: 10 g/L

S6  

enzyme, 2 g/L substrate, 0.5 M i-PrNH2, 100 µM PLP, 5% DMSO in 100 mM triethanolamine, pH 7.5 at 22°C for 24 h.

Variant % Conversion

Y26H; V65A; V69G; F122I; S223P; A284G*

0.7

A31T; V65A; V69G; F122I; S223P 0.6

V65A; V69G; F122I; S223P 0.6

*Selected as Round 2 parent.

Library: Combinatorial of H62AYFT, I122LV, G136Y, E137ATI, V199WI, A209L, F225Y and T282S.

Table S5. Top 5 variants from round 2 screening. Conditions: 2 g/L substrate, 0.5 M i-PrNH2, 100 µM PLP, 5% MeOH in 100 mM triethanolamine, pH 7.5 at 22°C for 24 h.

Variant Fold Improvement over Round 2 parent

Y26H; H62T; V65A; V69G; F122I; G136Y; E137I; V199I; A209L; S223P; T282S; A284G 75-fold

Y26H; H62T; V65A; V69G; F122I; G136Y; E137T; V199I; A209L; S223P; F225Y; T282S; A284G 51-fold

Y26H; V65A; V69G; F122V; G136Y; E137I; S174A; V199I; A209L; S223P; I230V; A284G 50-fold

Y26H; H62T; V65A; V69G; F122V; G136Y; E137T; V199I; A209L; S223P; T282S; A284G 48-fold

Y26H; V65A; V69G; E117G; F122V; G136Y; E137I; V199I; A209L; S223P; A284G 31-fold

S7  

Table S6. Retesting of round 2 hits as fermentation powders. Conditions: 10 g/L enzyme, 2 g/L substrate, 1 M i-PrNH2, 100 µM PLP, 5% MeOH in 100 mM triethanolamine, pH 7.5 at 22°C for 24 h.

Variant % Conversion

Y26H; H62T; V65A; V69G; F122V; G136Y; E137T; V199I; A209L; S223P; T282S; A284G

39

Y26H; H62T; V65A; V69G; F122I; G136Y; E137I; V199I; A209L; S223P; T282S; A284G

38

Y26H; V65A; V69G; F122V; G136Y; E137I; S174A; V199I; A209L; S223P; I230V; A284G

34

Directed evolution of a process stable transaminase, rounds 3-11

Evolution rounds 3-11 focused on increasing enzyme activity and in-process stability. Mutations were generated, sorted (ProSAR)(S1), and recombined mutations across the whole protein. Successive rounds were screened under increasingly challenging conditions: increased substrate loading, i-PrNH2 concentration, co-solvent concentration, pH and temperature to meet the demands of the process. Biocatalyst evolution and process development proceeded in parallel where process changes were reflected in biocatalyst screening conditions as described below.

pH: The pH was increased from 7.5 to 8.5 in Round 4, as this change to the process resulted in a 2.5-fold rate enhancement.

Temperature: The temperature was increased incrementally from 22°C to 45°C over three rounds of evolution (rounds 2-5) to increase reaction rate. While many variants were stable at >60°C, the substrate decomposed above 50°C.

Solvent: Methanol and DMSO were initially identified as the optimal solvents for the process, Table S7. However, high concentrations of substrate were poorly soluble in methanol, and significant substrate-product adduct formation was observed, so co-solvent was switched to DMSO and its concentration was increased as needed.

Table S7. Solvent screening experiments with top hit from Round 1b. Conditions: 10 g/L enzyme, 2 g/L substrate, 500 mM i-PrNH2, 100 µM PLP, 5% solvent in 100 mM triethanolamine, pH 7.5 at 22°C for 24 h.

S8  

Solvent % Conversion

DMSO 0.9

Methanol 0.7

Ethanol 0.6

Isopropanol 0.3

Isopropylamine: Initial screening identified 1 M i-PrNH2 as optimal concentration for enzyme rate, but unfavorable for enzyme stability. The i-PrNH2 concentration was maintained at 500 mM in rounds 1 and 2 to ensure that active (but possibly unstable) variants were identified. This was increased to 1 M once sufficient enzyme activity/stability was established. Employing a large excess of the amine donor i-PrNH2 is also important for shifting equilibrium toward product sitagliptin, so higher concentrations were desirable.

Table S8. Screening conditions for each round of evolution.

Round # 1 and 2 3 4 5 6 7-9 10-11

Substrate, g/L 2 5 10 40 100 100 50

[i-PrNH2], M 0.5 0.5 1.0 1.0 1.0 1.0 1.0

Cosolvent 5% DMSO

5% MeOH

5% MeOH

10% MeOH

20% MeOH

25% DMSO

50% DMSO

pH 7.5 7.5 8.5 8.5 8.5 8.5 8.5

Temp, °C 22 30 30 45 45 45 45

Table S9. Library types and number of variants screened (Round 3-11). The total number of variants screened in each assay over the course of the evolution is given for the different library types. The homology, random mutagenesis, rational design and site saturation mutagenesis libraries were used to generate diversity. The ProSAR libraries were used to sift through existing diversity as well as generate diversity in the form of random mutations.

S9  

Library type Rounds Utilized Variants Screened

Homology 3,5,11 4536

Random Mutagenesis 3,4,5,10,11 7476

ProSAR 3,4,5,6,7,8,9,10,11 11088

Rational design 5,8,9,11 1536

Site saturation mutagenesis 3,4,5,6,7,8,9,11 11844

Total 36480

Quantitation of catalyst improvement

We estimate that the final variant is around four orders-of-magnitude improved over the first pro-sitagliptin ketone active variant (Fig. S1). Over the first 4 rounds of evolution, the best variants from each round were 75-fold, 4-fold, and 9-fold improved over their respective parent, for a theoretical cumulative 2700-fold improvement. In subsequent rounds of evolution, the highest fold improvement per round was ~2-fold and the overall increase during round 5 through 11 was ~11-fold. Under the successively more challenging conditions the final catalyst is estimated to be ~27,000-fold improved as calculated by comparing each rounds’ best variant with its parent.

We evaluated the best variants of each round under identical (but not final) process conditions to assess their relative activity (Fig. S2, Table S10). The round 1 and 2 best variants showed no activity under these conditions and the overall improvement of the final catalyst cannot be determined. This data underscores the need for a stepwise increase in the challenge exerted by the screening conditions to identify hits that ultimately perform under process conditions. Under these conditions the best hit from round 4 gave 35% conversion after 24 hrs, whereas the final variant reached similar conversion after ~ 1 hr representing a 24-fold improvement.

The overall fold improvement can also be calculated by comparing initial turnover with the turnover achieved with the final catalyst. The round 1 best variant showed 0.7% conversion of 2 g/l substrate in 24 hrs using 10 g/L enzyme for a turnover of 0.0014 g/gTA.day (Table S4). The final catalyst tested under various conditions (Table S11) shows turnovers ranging from 24.8 to 48 g/gTA.day (entries 9 and 10) corresponding to improvements that range from 17,700 to 34,300 with an averaging of ~25,000.

S10  

Figure S1. Compounded fold improvements identified in high-throughput screening.

Figure S2. Head-to-head comparison of the top variants from each round of evolution under process-like conditions. The top variants from each round of evolution were compared under identical reaction conditions: 5 g/L enzyme, 50 g/L substrate, 1 M i-PrNH2, 1 mM PLP, 50% DMSO in 100 mM triethanolamine, pH 8.5 at 45°C for 24 h without acetone removal or additional pH control.

Table S10. Evaluation of top variants from each round of evolution in vial reactions under the same reaction conditions. Conditions: 5 g/L enzyme, 50 g/L substrate, 1 M i-PrNH2, 1 mM PLP, 50% DMSO in 100 mM triethanolamine, pH 8.5 at 45°C for 24 h. *Reference amino acid refers to ATA-117.

rounds 1‐4, log scale

1

10

100

1000

10000

1st 2nd 3rd 4th

compounded fold 

improvement identical mild 

conditions

increasingly challenging conditions

round # best variant

75X 9X 4X

rounds 1‐4, log scale

round # best variant

rounds 4‐9, linear scaletentative “final” 

conditions

compounded fold 

improvement increasingly challenging 

conditions

rounds 4‐11, linear scale

0

5000

10000

15000

20000

25000

30000

35000

4th 5th 6th 7th 8th 9th 10th 11th

0

10

20

30

40

50

60

70

80

90

100

0 5 10 15 20 25

%Conversion

Time (h)

1st

2nd

3rd

4th

5th

6th

7th

8th

9th

10th

11th

S11  

Round Mutations from ATA-117* Conversion

1.5 h 5 h 24 h

1 Y26H; V65A; V69G; F122I; S223P; A284G 0 0 0

2 Y26H; H62T; V65A; V69G; F122I; G136Y; E137I; V199I; A209L; S223P; T282S; A284G 0 0 0

3 S8P; H62T; V65A; V69C; M94I; F122I; G136Y; E137T; V199I; A209L; G215C; S223P; T282S; A284G 1 2 6

4 S8P; L61Y; H62T; V65A; V69T; M94I; F122I; G136F; V199I; A209L; G215C; S223P; T282S; A284G 7 14 39

5 S8P; L61Y; H62T; V65A; V69T; D81G; M94L; I96L; F122I; G136F; T178S; V199I; A209L; G215C; S223P; L269P; T282S; A284G; P297S; S321P 8 16 44

6 S8P; Y60F; L61Y; H62T; V65A; V69T; D81G; M94I; I96L; F122I; G136F; A169L; V199I; A209L; G215C; G217N; S223P; L269P; L273Y; T282S; A284G; P297S; S321P 17 30 61

7 S8P; Y60F; L61Y; H62T; V65A; V69T; D81G; M94I; I96L; F122I; S124H; G136F; A169L; V199I; A209L; G215C; G217N; S223P; L269P; L273Y; T282S; A284G; P297S; S321P 14 27 63

8 S8P; Y60F; L61Y; H62T; V65A; V69T; D81G; M94I; I96L; F122M; S124N; G136F; A169L; V199I; A209L; G215C; G217N; S223P; L269P; L273Y; T282S; A284G; P297S; S321P 17 30 66

9 S8P; Y60F; L61Y; H62T; V65A; V69T; D81G; M94I; I96L; F122M; S124N; G136F; A169L; V199I; A209L; G215C; G217N; S223P; L269P; L273Y; T282S; A284G; P297S; S321P; Q329H 28 47 84

10 S8P; Y60F; L61Y; H62T; V65A; V69T; D81G; M94I; I96L; F122M; S124T; G136F; Y150S; V152C; A169L; V199I; A209L; G215C; G217N; S223P; L269P; L273Y; T282S; A284G; P297S; S321P; 37 56 85

11 S8P; Y60F; L61Y; H62T; V65A; V69T; D81G; M94I; I96L; F122M; S124T; S126T; G136F; Y150S; V152C; A169L; V199I; A209L; G215C; G217N; S223P; L269P; L273Y; T282S; A284G; P297S; S321P; 48 67 89

S12  

Development of process for conversion of pro-sitagliptin ketone to sitagliptin

An additional process challenge not addressed during the enzyme optimization program was "oiling out", or crystallization, of the pro-sitagliptin ketone during the reaction and imine dimer formation between the ketone substrate and amine product. A slow feed of the substrate as a DMSO solution over 2-3 h readily addressed these issues and minimized imine dimer formation to 2-5%. The pH also needed to be controlled throughout the reaction, as the i-PrNH2 that was converted to acetone and lost to atmosphere caused the pH to decrease from the pH optimum of 8.5 – i-PrNH2 is more basic than sitagliptin (pKa ~ 8-9). We found it convenient to use a 0.2 M triethanolamine buffer and to control the pH by monitoring and feeding in 4 M i-PrNH2.

If the reaction is performed in a sealed vessel, the equilibrium with 4-10 equivalents of i-PrNH2 gives an 87:13 mixture of product:pro-sitagliptin ketone with ~1.8 vol% acetone generated. Using in situ IR monitoring with GC calibration, we were able to track the build-up of acetone and product during the reaction. We found that by simply running the pilot plant vessel under a mild 300-350 torr vacuum with a 2 fps nitrogen sweep of the head-space removed ~70% of the acetone generated with minimal volume loss and thus increased the reaction rate and shifted to equilibrium to >90% conversion. On small scale, it is sufficient to remove acetone by running the reaction in open air, or with a mild nitrogen sweep rather than vacuum.

The final process runs in 50% DMSO at 45°C (~8 g/L pro-sitagliptin ketone solubility) using up to 250 g/L substrate. Although the enzyme can tolerate temperatures up to 60°C, the pro-sitagliptin ketone was prone to slow decompositions at temperatures above 50°C. The enzyme tolerates a variety of pro-sitagliptin ketone concentrations (Table S11, entries 1-7) by charging 3 wt.% enzyme relative to the substrate loading. Entry 8 shows the optimal conditions of 200 g/L pro-sitagliptin ketone using 3 wt.% enzyme in 50% DMSO gives a 92% assay yield. Increasing the enzyme loading to 4-6 wt.% (entries 9, 11-16) gives a decrease in the reaction times from 24 hours down to as little as 12 hours (entry 13). This robust operating space allows for flexibility in terms of reaction times and concentrations, thus allowing the reaction to be tailored to the ideal time-cycles and through-put without compromising yields. It should be noted that in no instance has the minor enantiomer been detected by chiral HPLC analysis (>99.95% ee).

Table S11. Reaction parameters.

Entry Substrate (g/L)

Enzyme(wt.%)

Addition Time (h)

DMSOvol%

ReactionTime (h)

Assay Yield (%)

1 100 3 4 40 17 92 2 125 3 5 40 25 91 3 150 3 6 40 25 88 4 175 3 7 40 23 91 5 200 3 8 40 23 89 6 225 3 9 40 23 86 7 250 3 10 40 23 83 8 200 3 8 50 24 92 9 200 6 3 50 15 93

10 150 3 2 50 15 90 11 150 4 2 50 14 92 12 150 4 2 50 13 93

S13  

13 150 4 2 50 12 90 14 160 4 2.5 50 15 91 15 170 4 2.5 50 15 92 16 180 4 3 50 15 88

Oftentimes, enzymatic transformations are difficult to run on a factory scale due to emulsions during the work-up. For sitagliptin, we also wanted a method to completely reject and eliminate all detectable enzyme during the work-up to ensure purity in the downstream processing. We found two convenient work-up methods which readily eliminated the enzyme residue and prevented emulsions. At the end of reaction, the batch is acidified to pH 2 with concentrated HCl at 45°C. This serves to not only to denature enzyme to an insoluble precipitate, but it also hydrolyzes the imine dimer impurity to pro-sitagliptin ketone and sitagliptin. Filtration of this mixture then removes the majority of the enzyme precipitate. Adjusting the pH to 10.5 with NaOH and extraction with isopropylacetate allows for efficient product recovery with minimal emulsion and complete enzyme rejection in the product containing organic phase. A simpler procedure is to skip the filtration step and simply extract after adjustment to pH 10.5 using a 1/1 mixture of isopropyl alcohol and isopropyl acetate. No emulsion is generated and the enzyme is sequestered in the bottom aqueous layer.

Experimental methods

General methods. All reagents, buffers, starting materials and solvents were purchased from Sigma-Aldrich (Milwaukee, USA) and used without purification.

Libraries. Mutational diversity was generated at the beginning of the project and throughout the evolution program, as necessary, using random mutagenesis via error-prone PCR, semisynthetic shuffling approaches (S2), using sequence diversity (S3) from the nearest family members; and information for specific structural regions, such as the active site binding pockets, and structural loops. Saturation mutagenesis was also carried out at regions of apparent functional importance. Random mutations present in the ProSAR libraries served as another source of diversity. The number of variants screened for each library type is given in Table S9.

Transaminase library construction and expression. The gene for ATA-117 transaminase – a closely-related nonnatural homolog of transaminase from Arthrobacter sp. KNK168 (S4) – was codon-optimized for expression in E. coli based on the amino acid sequence of the transaminase (S5). The gene was cloned into expression vector pCK110700 or pCK110900 (S6) under the control of a lac promoter. This expression vector also contains the P15a origin of replication and the chloramphenicol resistance gene. The vector was transformed into E. coli W3110 from which plasmid DNA was prepared using standard methods.

Transaminase libraries were constructed according to previously described methods (S2,S7) cloned into vector pCK110700 or pCK110900, transformed and expressed in E. coli W3110.

Oligonucleotides designed to unlink all programmed mutations were spiked into semisynthetic libraries to give 50% incorporation for single oligonucleotides (lower in the case of multiple oligos at one

S14  

position). Such libraries contained a random mutation rate of 1/variant with an s.d. of 1. Unlinking of mutations was accomplished by designing oligos in such as way as to ensure each programmed mutation or combination of mutations was present on oligos within a defined region. In cases where only one position was targeted, either separate oligos were designed for each mutation or a single oligo was designed to have ambiguity at the targeted position corresponding to the different desired codons. Similarly, for two or more mutations near each other, one or more oligos were designed in such a way that all combinations of mutations were made possible by using zero or more ambiguous codons at zero or more positions.

High-throughput screening of transaminase variants. The gene encoding transaminase, constructed as described in above, was mutagenized using methods described above and the population of altered DNA molecules was used to transform E. coli W3110.

Recombinant E. coli colonies carrying a gene encoding transaminase were picked using a Q-Bot® robotic colony picker (Genetix USA, Inc., Boston, MA) into 96-well shallow well microtiter plates containing 180 μL LB Broth, 1% glucose and 30 µg/mL chloramphenicol (CAM). Cells were grown overnight at 30°C with shaking at 200 rpm. A 10 μL aliquot of this culture was then transferred into 96-

deep well plates containing 390 μL M9YE broth, 100 M pyridoxine and 30 μg/mL CAM. After incubation of the deep-well plates at 30°C with shaking at 250 rpm for 2-3 h, recombinant gene expression within the cultured cells was induced by addition of IPTG to a final concentration of 1 mM. The plates were then incubated at 30°C with shaking at 250 rpm for 18 h.

Cells were pelleted by centrifugation (4000 RPM, 10 min, 4°C), resuspended in 200 μL lysis buffer and lysed by shaking at room temperature for 2 hours. The lysis buffer contained 100 mM triethanolamine (chloride) buffer, pH 7.5 or 8.5, 1 mg/mL lysozyme, 500 μg/mL polymixin B sulfate

(PMBS) and 250-1000 M PLP. After sealing the plates with aluminum/polypropylene laminate heat seal tape (Velocity 11, Menlo Park, CA, Cat# 06643-001), they were shaken vigorously for 2 hours at room temperature. Cell debris was collected by centrifugation (4000 RPM, 10 min., 4°C) and the clear supernatant assayed directly or stored at 4°C until use.

For screening in methanol or DMSO at pH 7.5 with early-stage evolvants, a 10 L aliquot of a solution of methyl ketone analog or pro-sitagliptin ketone (40 mg/mL) in methanol or DMSO was added

to each well of a Costar® deep well plate, followed by addition of 90 L of 1.1 M i-PrNH2 hydrochloride using a Biomek NXp robotic instrument (Beckman Coulter, Fullerton, CA). This was then followed by addition of 100 μL of the recovered lysate supernatant, also performed using the Biomek NXp, to provide a reaction mixture comprising of 2 mg/ml pro-sitagliptin ketone, 500 mM i-PrNH2 hydrochloride, 50 mM triethanolamine pH 7.5, and 5% methanol or DMSO (v/v). The plates were heat-sealed with aluminum/polypropylene laminate heat seal tape (Velocity 11, Menlo Park, CA, Cat# 06643-001) at

175°C for 2.5 seconds and then shaken 24 h at 30C. Reactions were quenched by the addition of 1 mL acetonitrile using a Phoenix Liquid Handling System (Art Robbins Instruments, Sunnyvale, CA). Plates were resealed, shaken for 5 min, and then centrifuged at 4000 rpm for 10 min. A 200 μL aliquot of the cleared reaction mixture was transferred to a new shallow well polypropylene plate (Costar #3365), sealed and analyzed as described below.

For screening in DMSO at pH 8.5 with late-stage evolvants, a 100 L aliquot of a solution of pro-

S15  

sitagliptin ketone (100 mg/mL) in dimethyl sulfoxide (DMSO) was added to each well of a Costar® deep

well plate, followed by addition of 50 L of 4 M i-PrNH2 hydrochloride using a Biomek NX robotic instrument (Beckman Coulter, Fullerton, CA). This was then followed by addition of 50 μL of the recovered lysate supernatant, also performed using the Biomek NX, to provide a reaction comprising of 50 mg/ml pro-sitagliptin ketone, 1 M i-PrNH2 hydrochloride, 25 mM triethanolamine pH 8.5, and 50% DMSO (v/v). The plates were heat-sealed with aluminum/polypropylene laminate heat seal tape (Velocity 11, Menlo Park, CA, Cat# 06643-001) at 175°C for 2.5 seconds and then shaken overnight (at

least 16 hours) at 45C. Reactions were quenched by the addition of 1 mL acetonitrile using a Phoenix Liquid Handling System (Art Robbins Instruments, Sunnyvale, CA). Plates were resealed, shaken for 5 min, and then centrifuged at 4000 rpm for 10 min. A 10 μL aliquot of the cleared reaction mixture was

transferred to a new shallow well polypropylene plate (Costar #3365) containing 190 L acetonitrile, sealed and analyzed as described below.

Production of transaminase powders – shake flask procedure. A single microbial colony of E. coli. containing a plasmid encoding a transaminase of interest was inoculated into 50 mL Luria Bertani broth containing 30 µg/mL chloramphenicol and 1% glucose. Cells were grown overnight (at least 16 h) in an incubator at 30°C with shaking at 250 rpm. The culture was diluted into 250 mL M9YE (1.0 g/L ammonium chloride, 0.5 g/L of sodium chloride, 6.0 g/L of disodium monohydrogen phosphate, 3.0 g/L of potassium dihydrogen phosphate, 2.0 g/L of Tastone-154 yeast extract, 1 L/L de-ionized water) containing 30 µg/mL chloramphenicol and 100 µM pyridoxine, in a 1 L flask to an optical density at 600 nm (OD600) of 0.2 and allowed to grow at 30°C. Expression of the transaminase gene was induced by addition of iso propyl β D-thiogalactoside (IPTG) to a final concentration of 1 mM when the OD600 of the culture is 0.6 to 0.8 and incubation was then continued overnight (at least 16 h). Cells were harvested by centrifugation (5000 rpm, 15 min, 4°C) and the supernatant discarded. The cell pellet was resuspended with an equal volume of cold (4°C) 100 mM triethanolamine (chloride) buffer, pH 7.5 containing 100-500 µM pyridoxal 5'-phosphate (PLP), and harvested by centrifugation as above. The washed cells were resuspended in two volumes of the cold triethanolamine (chloride) buffer containing PLP and passed through a French Press twice at 12,000 psi while maintained at 4°C. Cell debris was removed by centrifugation (9000 rpm, 45 min., 4°C). The clear lysate supernatant was collected and stored at -20°C. Lyophilization of frozen clear lysate provides a dry powder of crude transaminase enzyme. Alternatively, the cell pellet (before or after washing) may be stored at 4°C or -80°C.

Production of transaminase powders – fermentation procedure. A single microbial colony of E. coli. containing a plasmid with the transaminase gene of interest was inoculated into 2 mL M9YE broth (1.0 g/L ammonium chloride, 0.5 g/L of sodium chloride, 6.0 g/L of disodium monohydrogen phosphate, 3.0 g/L of potassium dihydrogen phosphate, 2.0 g/L of Tastone-154 yeast extract, 1 L/L de-ionized water) containing 30 μg/ml chloramphenicol and 1% glucose. Cells were grown overnight (at least 12 h) in an incubator at 37°C with shaking at 250 rpm. After overnight growth, 0.5 mL of this culture was diluted into 250 ml M9YE Broth, containing 30 μg/ml chloramphenicol and 1% glucose in 1 liter flask and allowed to grow at 37°C with shaking at 250 rpm. When the OD600 of the culture is 0.5 to 1.0, the cells were removed from the incubator and either used immediately, or stored at 4°C.

Bench-scale fermentations were carried out at 30°C in an aerated, agitated 15 L fermentor using 6.0 L of growth medium (0.88 g/L ammonium sulfate, 0.98 g/L of sodium citrate; 12.5 g/L of dipotassium hydrogen phosphate trihydrate, 6.25g/L of potassium dihydrogen phosphate, 3.3 g/L of

S16  

Tastone-154 yeast extract, 0.083 g/L ferric ammonium citrate, and 8.3 ml/L of a trace element solution containing 2 g/L of calcium chloride dihydrate, 2.2 g/L of zinc sulfate septahydrate, 0.5 g/L manganese sulfate monohydrate, 1 g/L cuprous sulfate heptahydrate, 0.1 g/L ammonium molybdate tetrahydrate and 0.02 g/L sodium tetraborate. The vessel was sterilized at 121°C and 15 PSI for 30 minutes, and 100 µM pyridoxine was added post sterilization. The fermentor was inoculated with a late exponential culture of E. coli W3110 containing a plasmid encoding the transaminase gene of interest (grown in a shake flask as described above to a starting OD600 of 0.5 to 1.0. The fermentor was agitated at 250-1250 rpm and air was supplied to the fermentation vessel at 0.6-25 L/min to maintain a dissolved oxygen level of 50% saturation or greater. The pH of the culture was maintained at 7.0 by addition of 20% v/v ammonium hydroxide. Growth of the culture was maintained by addition of a feed solution containing 500 g/L cerelose, 12 g/L ammonium chloride and 5.1 g/L magnesium sulfate heptahydrate. After the culture reached an OD600 of 70 +/-10, expression of transaminase was induced by addition of isopropyl-β-D-thiogalactoside (IPTG) to a final concentration of 1 mM and fermentation is continued for another 18 hours. The culture was then chilled to 4°C and maintained at that temperature until harvested. Cells were collected by centrifugation at 5000 G for 40 minutes in a Sorval RC12BP centrifuge at 4°C. Harvested cells were used directly in the following downstream recovery process or they may be stored at 4°C or frozen at -80°C until such use.

The cell pellet was resuspended in 2 volumes of 100 mM triethanolamine (chloride) buffer, pH 7.5 containing 100 µM or 500 µM pyridoxal 5'-phosphate (PLP), at 4°C to each volume of wet cell paste. The intracellular transaminase was released from the cells by passing the suspension through a homogenizer fitted with a two-stage homogenizing valve assembly using a pressure of 12000 psig. The cell homogenate was cooled to -20°C immediately after disruption. A solution of 11% w/v polyethyleneimine pH 7.2 was added to the lysate to a final concentration of 0.5% w/v. A solution of 1 M Na2SO4 was added to the lysate to a final concentration of 100 mM. The lysate was then stirred for 30 min. The resulting suspension was clarified by centrifugation at 5000G in a Sorval RC12BP centrifuge at 4°C for 30 min. The clear supernatant was decanted and concentrated ten-fold using a cellulose ultrafiltration membrane with a molecular weight cut off of 30 kD. The final concentrate was dispensed into shallow containers, frozen at -20°C and lyophilized to powder. The transaminase powder was stored at -80°C.

Evaluation of hits. Round 1. A 500 μL solution of transaminase variant (20 mg/mL) in 100 mM triethanolamine-chloride buffer pH 7.5 with 250 µM pyridoxal 5'-phosphate was added to 5 mL reaction vial equipped with a stir bar. Subsequently, 450 μL of 1.1 M i-PrNH2 hydrochloride, followed by 50 μL of a solution of pro-sitagliptin ketone (40 mg/mL) in DMSO was added to the transaminase solution. The

reaction was stirred at 22C and monitored by HPLC analysis of samples taken periodically from the reaction mixture.

Round 2. A 500 μL solution of transaminase variant (20 mg/mL) in 100 mM triethanolamine-chloride buffer pH 7.5 with 250 µM pyridoxal 5'-phosphate was added to 5 mL reaction vial equipped with a stir bar. Subsequently, 450 μL of 1.1 M i-PrNH2 hydrochloride, followed by 50 μL of a solution of pro-sitagliptin ketone (40 mg/mL) in methanol was added to the transaminase solution. The reaction was

stirred at 22C and monitored by HPLC analysis of samples taken periodically from the reaction mixture.

Round 3. A 500 μL solution of transaminase variant (20 mg/mL) in 100 mM triethanolamine-

S17  

chloride buffer pH 7.5 with 250 µM pyridoxal 5'-phosphate was added to 5 mL reaction vial equipped with a stir bar. Subsequently, 450 μL of 2.2 M i-PrNH2 hydrochloride, followed by 50 μL of a solution of pro-sitagliptin ketone (100 or 200 mg/mL) in methanol was added to the transaminase solution. The

reaction was stirred at 30C and monitored by HPLC analysis of samples taken periodically from the reaction mixture.

Round 4. A 500 μL solution of transaminase variant (20 mg/mL) in 100 mM triethanolamine-chloride buffer pH 8.5 with 250 µM pyridoxal 5'-phosphate was added to 5 mL reaction vial equipped with a stir bar. Subsequently, 400 μL of 2.5 M i-PrNH2 hydrochloride, followed by 100 μL of a solution of pro-sitagliptin ketone (200 mg/mL) in methanol was added to the transaminase solution. The reaction

was stirred at 45C and monitored by HPLC analysis of samples taken periodically from the reaction mixture.

Rounds 5-9. A 250 μL solution of transaminase variant (20 mg/mL) in 100 mM triethanolamine-chloride buffer pH 8.5 with 250 µM pyridoxal 5'-phosphate was added to 5 mL reaction vial equipped with a stir bar. Subsequently, 500 μL of 2 M i-PrNH2 hydrochloride, followed by 250 μL of a solution of pro-sitagliptin ketone (200 mg/mL) in DMSO was added to the transaminase solution. The reaction is

stirred at 45C and monitored by HPLC analysis of samples taken periodically from the reaction mixture.

Rounds 10-11. A 250 μL solution of transaminase variant (20 mg/mL) in 100 mM triethanolamine-chloride buffer pH 8.5 with 1000 µM pyridoxal 5'-phosphate was added to 5 mL reaction vial equipped with a stir bar. Subsequently, 250 μL of 4 M i-PrNH2 hydrochloride, followed by 500 μL of a solution of pro-sitagliptin ketone (100 mg/mL) in DMSO was added to the transaminase solution. The

reaction is stirred at 45C and monitored by HPLC analysis of samples taken periodically from the reaction mixture.

Analytical Methods.

High throughput screening achiral HPLC method to determine conversion of pro-sitagliptin ketone to sitagliptin. Enzymatic conversion of pro-sitagliptin ketone to sitagliptin was determined using an Agilent 1200 HPLC equipped with an Agilent Eclipse XDB-C8 column (4.6 x 150

mm, 5 m), using 45:55 10 mM NH4Ac/MeCN as eluent at a flow rate of 1.5 ml/min and a column temperature 40°C. Retention times: sitagliptin: 1.4 min; pro-sitagliptin ketone: 1.7 min. The pro-sitagliptin ketone and product were determined as the peak area at 210 nm or 268 nm. Using these conditions, the limit of detection for sitagliptin was 5 µg/mL.

Chiral HPLC method to determine stereopurity of sitagliptin. Stereomeric purity of sitagliptin was determined using an Agilent 1200 HPLC equipped with a Daicel Chiralpak AD-H column

(4.6 x 150 mm, 5 m) using 60:40:0.1:0.1 EtOH/Heptane/diethylamine/water as the eluent at a flow rate of 0.8 ml/min and a column temperature of 35°C. Retention times: pro-sitagliptin ketone: 6.3 min; (S)-enantiomer: 8.4 min; sitagliptin: 10.8 min. The pro-sitagliptin ketone and product were determined as the peak area at 210 nm or 268 nm.

Liquid chromatography-mass spectroscopy (LC/MS) method for detecting low-level

S18  

conversion of pro-sitagliptin ketone to sitagliptin. Low-level enzymatic conversion of pro-sitagliptin ketone to sitagliptin was determined using an LC/MS/MS method. Five microliters of sample was loaded into an Eclipse XDB-C8 HPLC column (4.6 x 150 mm) and eluted isocratically with a 40:60 mobile phase of 0.2% ammonium formate and methanol at 1.0 mL/min. The retention time of sitagliptin was 1.5

minutes at 35C. Mass spectrometry was used for detection on a Waters Quattro triple quadruple. Q1 was set to pass the M+H ion at 408.1 AMU and Q3 was set to pass the 235.1 daughter ion. The collision cell (Q2) had a collision energy of 17.0 and Argon gas flow of 0.3 mL/min. Ionization was by APCI with a corona discharge of 5 µA, source temperature of 130˚C and probe temperature of 600˚C. Desolvation gas flow was 100 L/minute and the cone gas was set to 50 L/minute.

Physical characterization of round 11 transaminase.

Molecular weight and impurity profile by size exclusion chromatography (SEC).

A. Reagents

1. Water (H2O), HPLC grade

2. Tris(hydroxymethyl)aminomethane (also Tris, C4H11NO3), ACS Reagent Grade

3. Methanol (MeOH), HPLC grade

4. Sodium Chloride (NaCl), ACS Reagent Grade

5. Protein reference standards, see Table S12:

Table S12. Molecular weights of protein standards.

Protein Molecular weight

Thymoglobulin 669000

Gamma globulin 149000

BSA 66400

Ovalbumin 44300

Myoglobin 17000

B. Solutions

1. Mobile phase: Tris buffer (25 mM final concentration) solution pH 8, was prepared by dissolving 3.03 g Tris into 700 mL water in a 1 L beaker, followed by pH adjustment with 1

S19  

N HCl. Then, 11.7 g of NaCl was added to the solution (200 mM final concentration), followed by 50 mL of methanol. The solution was transferred into a 1 L volumetric flask and diluted to 1 L with water. The mobile phase is stable for one month under ambient conditions.

2. Diluent: The mobile phase is used as the diluent.

C. Chromatographic Conditions

Column:

Zenix-SEC 300, 4.6 x 300 mm, 3 µm, 300Å

(Sepax Catalog # 213300-4630), or equivalent

UltraShield Precolumn Filter (MMUS-1505, MacMod)

Flow rate: 0.35 mL/min, isocratic

Injection volume: 5 µL

Column temperature: Ambient

Detection: UV absorbance at 214 nm

D.  Standard Preparation

Prepare a protein standard solution at 0.5 mg each / mL in mobile phase, which contains Thymoglobulin, Gamma globulin, BSA, Ovalbumin, and Myoglobin. For example, weigh approximately 5 mg of each protein standard into a 10 mL volumetric flask and dilute to volume with the diluent. Transfer 300 µL of each protein solution into a 2 mL vial and mix well. Filter the standard solution through a Millex-PVDF syringe filter (0.45 µm, 4 mm diameter, Millipore, Part#: SLHV004SL). Inject sample (Fig. S3).

E.  Sample Preparation

Prepare a 0.5 mg/mL transaminase solution in mobile phase. For example, weigh 50 mg of sample into a 100 mL volumetric flask. Dissolve in and dilute to volume with diluent and mix well. Filter the sample solution through a Millex-PVDF syringe filter (0.45 µm, 4 mm diameter, Millipore, Part#: SLHV004SL). Inject sample (Fig. S4).

F.  Calculation

The curve of logarithmic retention time vs. molecular weight of the five proteins is plotted using Excel software and a linear regression equation can be obtained.

Molecular weight = 10^(a×tu + b)

where tu is the retention time of a protein and a and b are the regression constants, the slope and y-intercept respectively. Note that the retention time should be reported in three decimals.

S20  

The molecular weight for an unknown transaminase sample is calculated by substituting the retention time (tu) for the transaminase in minutes into the regression equation.

Figure S3. SEC chromatogram for protein standard preparation.

Figure S4. SEC chromatogram for standard preparation of round 11 transaminase.

Molecular weight MALDI-TOF. All MALDI-TOF mass spectra were acquired on a Bruker Autoflex III MALDI-TOF mass spectrometer (Bruker Daltonics, Billerika, MA) equipped with a nitrogen laser operating at 337 nm and an ion source operated in delayed extraction mode. The MALDI target used in this study is a 384 regular stainless steel sample support (Bruker Daltonics, Billerika, MA). The total accelerating voltage was 20 kV and the extraction delay is typically 150 ns. Spectra were acquired with 80% to 90% laser energy and 70% laser attenuation. An average of 4000 shots was accumulated for each spectrum at the mass range of m/z 20k to 150k. Linear positive mode was found to generate the best signal in terms of both resolution and intensity. The TOF analyzer was calibrated using cytochromC (10 nM/mL, MW: 12361.96 Da).

S21  

Balance: B-5

Sample prep: round 11 transaminase, 0.12 mg/ml in water (3 nM/mL)

Diluent: 50:50/0.05, v/v/v, water/acetonitrile/TFA

Matrix: 10 mg/ml sinapinic acid in diuent

MALDI sample prep: mix 1:1 sample with matrix, vortex and then spot 1 µl of mixture onto the target and let sample dry at ambient temperature.

Figure S5. MALDI-TOF of monomer of round 11 transaminase.

Process for conversion of pro-sitagliptin ketone to sitagliptin.

CO2H

F

F

F

O O

O O

+

2,4,5-trif luoro acetic acid Meldrum's acid

+HCl•HN

NN

N

CF3

O

N

O

NN

N

CF3

F

F

F

triazolehydrochloride pro-sitagliptin ketone

Preparation of 4-oxo-4-[3-(trifluoromethyl)-5,6-dihydro[1,2,4]triazolo[4,3-a]pyrazin-7(8H)-yl]-1-(2,4,5-trifluorophenyl)butan-2-one. 2,4,5-Trifluorophenylacetic acid (2.5 kg, 13.15 mol),

S22  

Meldrum's acid (2.09 kg, 14.46 mol), DMAP (128.5 g, 1.052 mol), and acetonitrile (7.5 L) were charged into a 50 L three-neck flask. N,N-Diisopropylamine (4.92 L, 28.27 mol) was added in portions at room temperature while maintaining the internal temperature below 50 °C. Pivaloyl chloride (1.78 L, 14.46 mol) was added dropwise over 1-2 h while maintaining the temperature below 55°C. The reaction was aged at 45-50°C for 2-3 h. Triazole hydrochloride (3.01 kg, 13.2 mol) was added in one portion at 40-50°C. Trifluoroacetic acid (303 mL, 3.95 mol) was added dropwise, and the reaction solution was aged at 50-55°C for 6 h. The resulting acetonitrile reaction solution assayed at 90% yield of product ketoamide (11.8 mol, 4.81 kg).

Isolation: To a stirring solution of 13 L of the crude reaction above (259 g/L ketoamide) at room temperature was added 6.5 L of 0.5 N NaOH slowly over 1 h. The batch was then seeded with 3 wt. % seed (101 g) and stirred 30 min. at room temperature. Next, 13 L of 0.5 N NaOH was added over 2 h via addition funnel. The batch was then cooled to 0 °C over 2 h and aged a further 3 h at this temperature. The resulting slurry was filtered and the wet-cake washed three times with 5.2 L of 20% acetonitrile in water. The cake was dried overnight under a nitrogen atmosphere, transferred to drying trays, and further dried in a vacuum oven at 50 °C, 300 torr, and with a nitrogen sweep until KF was 2.1% water (hemihydrate form). 94% isolated yield (3.35 kg, 7.92 mol) at 96 wt.% (2.1% water present, remainder are various impurities).

Transaminase reaction at 40 mL reaction Scale.

For small scale runs (40 mL reaction volume), a Mettler-Toledo MultiMax RB04-50 reaction system was utilized which was fitted with pH probes and a base dispenser. Each reaction vessel was fitted with the instrument's mechanical stirrer, temperature probe, pH probe, base addition line (4 M isopropylamine in water), and a nitrogen sweep line. A stock buffer solution of 0.2 M triethanolamine in deionized water was prepared and adjusted to pH 9 with neat isopropylamine. To a reaction vessel 3.80 g (39.8 mmol, 2.10 equiv.) of isopropyl amine hydrochloride followed by 20 mg of PLP (pyridoxal 5'-phosphate) was added. The 0.2 M TEA buffer was then added (20 g or 20 mL) into the multimax vessel. After brief agitation, 240 mg (3wt.% relative to ketone) of round 11 transaminase was added and gently dissolved. The stirrer head is then attached to the vessel along with the temperature probe, stirrer motor, and a 4 M isopropylamine addition line (through the tallest port with the addition line near the top of the actual reaction vessel and along the wall such that the base will be added in above surface). A nitrogen sweep line was also installed through the multimax's tallest port, but only partway down. The nitrogen sweep rate is set at ~1 fps. The remaining reactor ports were left open to vent. At this point, the stirrer was set to 400 RPM and temperature to 25°C (pH control loop off). DMSO (12 mL) was then added to the multimax vessel over 5 min. The vessel was then heated to 45°C and stirrer was set to 600 RPM. When the temperature stabilized, the multimax's pH probe was installed in one port and a calibrated handheld pH probe was installed in another port. After 5 min, the handheld pH probe was removed, quickly rinsed

S23  

with water, submerged into pH 7 phosphate buffer standard solution and then re-calibrated. The handheld probe was then removed from the standard buffer, rinsed with water, and put back into the multimax vessel. This external handheld probe was then used to calibrate the reading on the internal multimax probe, after which time the handheld probe was removed. The pH control loop on the multimax was then turned on with a 20 sec ramp and a setting of pH 8.5. The 4 M isopropylamine base solution was then set to add when the pH dropped below 8.4. Next, a solution of 8.00 g ketoamide (96wt.%, 18.9 mmol, 1.0 equiv.) was dissolved in 8 mL of DMSO. The ketoamide solution was taken up into a 20 mL syringe with long, flexible Teflon tubing attached to the luer lock. This ketoamide addition line was fed through the same port as the base addition line, along the wall, above surface. The syringe was placed in a syringe pump and set for 14 mL volume (total volume including air bubble) at 3.5 mL/h (4 h addition). The syringe pump was tilted such that the air bubble in the syringe will push out any remaining ketoamide solution at the end of addition. After 20-24 h, the conversion was typically 93% with 5.6A% dimer and 1.4 equivalents (6.6 mL) of 4 M isopropylamine base used.

Filtration work-up: The pH control loop was turned off and 12 M HCl was added until pH 2-3 was reached. The reaction was then aged 1-2h at 45°C and 1000 RPM. A 30mL medium glass fritted filter funnel was fitted with two ply cheese cloth or other loose weave cotton towel. The reaction solution was then filtered through the cheese cloth followed by rinsing of the vessel and filter cake (precipitated enzyme) with 0.01 N HCl. To this aqueous acidic filtrate was added 20 mL of IPAc (isopropyl acetate) and the pH of the aqueous phase was adjusted to pH 11 with 10 N NaOH. The layers were agitated with stirring, and then allowed to settle and separated. This was repeated twice with IPAc and the combined organics are then washed with 15 mL of brine. The resulting IPAc solution of the sitagliptin free base was then assayed for yield (typically 88-92% assay yield, 6.78-7.08 g), and the solvent was switched to IPA for downstream processing to sitagliptin phosphate monohydrate.

Direct extraction work-up: The pH control loop was turned off and 12 M HCl was then added until pH 2-3. The reaction was then aged 1-2 h at 45°C and 1000 RPM. The batch was cooled to RT and then 20mL of isopropylalcohol (IPA) was added, followed by 20 mL IPAc. The pH of the aqueous layer was then adjusted to 11 with 19 N NaOH. The mixture was then agitated at 20-45oC (heat may be used to break the emulsion), and then allowed to settle and separate. The IPAc/IPA layer was set aside and the aqueous layer was then extracted with 20 mL of 80/20 (vol/vol) IPAc/IPA. The combined IPAc/IPA extracts were then washed with 20 mL of brine. The resulting IPAc/IPA solution of the sitagliptin free base was then assayed for yield (typically 87-90% assay yield, 6.70-6.93 g) and the solvent was switched to IPA for downstream processing to sitagliptin phosphate monohydrate.

Transaminase reaction at 1 kg reaction scale. The reaction was run in a vessel fitted with a mechanical stirrer, temperature probe, pH probe, and base addition line. The base addition line was used to control pH between 8.4 and 8.6 using a feed of 4 M isopropylamine free base in water. To the vessel was added 1.92 L water, followed by 109 mL (0.82mol, 0.33 equiv.) triethanolamine and 1.64 L (6.56 mol, 2.67 equiv.) of 4 M isopropyl amine solution. The pH was then adjusted to 8.5 using 12 N HCl (424 mL). The reactor was then charged with 6.7 g (0.027 mol, 0.011 equiv.) of PLP followed by 40 g of CDX-017 and the mixture was carefully dissolved with gentle agitation. The vessel was placed on the reactor block with the temperature probe, base addition line, pH probe, and stirrer set to 400 RPM (Note: pH control loop is off at this point). Next, 2.22 L of DMSO was added into the stirring solution and the reactor was heated to 45°C. When the temperature stabilized, the pH control loop was turned on and

S24  

adjusted to pH to 8.5 (pH controlled with 4 M iPr-amine in water). At this point, stirring was increased to 600 RPM, but tip speed is kept below 2 m/s to avoid vortexing. Then, 1.0 kg (corrected weight is 1 kg as received ketoamide is typically 96-98wt.% as a hemi-hydrate; 2.46 mol, 1.00 equiv.) of ketoamide was dissolved into 1.11 L of DMSO. This DMSO/ketoamide solution was then added to the reactor over 2-3 h. The reactor was then stirred at 45°C and with the pH maintained between 8.6-8.4 for another ~13 h with acetone removal being accomplished with 300 torr vacuum and 2 fps nitrogen sweep. After ~15 h total reaction time (1.3-2.0 equiv. isopropylamine uptake), the reaction was at 90-95% conv. as judged by reverse phase HPLC analysis.

Filtration work-up: The pH control loop was turned off and 13 g of solka-floc was added to the vessel followed by 12 M HCl until pH 2-3. The reaction was then aged 1-2 h at 45°C and 1000 RPM. The slurry was then passed through a filter (e.g., fritted plastic Buchner with filter paper on 1 kg scale, or sparkle filter with no recycle loop on pilot plant scale). The vessel and filter was rinsed with 1 L of 0.01 N HCl. To this aqueous acidic filtrate was then added 3 L of IPAc and the pH of the aqueous phase was then adjusted to pH 11 with 19 N NaOH. The layers were agitated with stirring, and then allowed to settle and separated (mild heat or vacuum accelerates phase separation). This was repeated twice more with 3 L of IPAc and the combined organics were then washed with 3 L of brine (at pH 11). The resulting IPAc solution of the sitagliptin free base was then assayed for yield (typically 88-92% assay yield; 882-922 g) and solvent switched to IPA for downstream processing to sitagliptin phosphate monohydrate.

Direct extraction work-up: The pH control loop was turned off and 12 M HCl was added until pH 2-3. The reaction was then aged 1-2 h at 45°C and 1000 RPM. The batch was cooled to RT and then 3 L of IPA was added, followed by 3 L of IPAc. The pH of the aqueous layer was then adjusted to 11 with 19 N NaOH. The mixture was agitated at 20-45oC (heat may be used to break the emulsion), and then allowed to settle and separate. The IPAc/IPA layer was set aside and the aqueous layer was extracted with 3 L of 80/20 (vol/vol) IPAc/IPA. The combined IPAc/IPA extracts were then washed with 3 L of brine. The resulting IPAc/IPA solution of the sitagliptin free base was then assayed for yield (typically 87-90% assay yield, 872-902 g) and solvent switched to IPA for downstream processing to sitagliptin phosphate monohydrate.

General reaction conditions for other ketones.

Figure S6. Substrate scope of ATA-117 (8) and round 11 transaminase.

S25  

A stock solution containing 125 g/L i-PrNH2 hydrochloride and 0.2 M triethanolamine was prepared and adjusted to pH 9.5 with neat i-PrNH2. PLP (10 mg) and round 11 transaminase (100 mg) were added to a 20 mL vial equipped with a stir bar. The stock i-PrNH2/ triethanolamine (7 mL) was added and the mixture was stirred at 700 RPM at RT for 5 min. Then, a solution of 100 mg ketone dissolved in 3 mL DMSO was added slowly. The resulting reaction solution was then heated to 45°C and stirred for 1-3 days in a loosely capped vial. When complete (conversion checked by reverse phase HPLC), 3 g of K2CO3 was added and dissolved at 45°C for 30 min. Then, the reaction was cooled and 50 µL of 5 N NaOH was added, followed by 5 mL of EtOAc. The extraction was stirred at 800 RPM for 1 h, and then allowed to settle and separated. The resulting aqueous layer was extracted twice more with 5 mL EtOAc. The combined organics were then washed with 2 mL of brine, dried with Na2SO4, and filtered. The organic solution was then analyzed by GC to determine conversion.

GC analysis method: Restek RTX-1701 (30 m x 0.32 mm), 2 µL injection (in MTBE), 250°C inlet, 10:1

split, 20.6 psi (2.8 mL/min), 50 °C for 2 min then 20°C/min to 280°C and hold 1.5 min., 15 min. total

runtime.

S26  

(S)-1-phenyl-2,2,2-trifluoroethylamine. Isopropyl amine hydrochloride (1.4 g) was added to 14 mL of 0.1 M triethanolamine buffer at pH 8.5. After dissolving the isopropylamine hydrochloride, 20 mg of PLP and 100 mg of round 6 transaminase were added and dissolved with gentle agitation at 400 rpm. The reactor was heated to 60 °C and the pH of the solution was adjusted to pH 8.5 with 4 M i-PrNH2. Trifluoromethyl ketone (400 mg) was dissolved in 6 mL of DMSO and added dropwise to the solution over 2 h. The reaction was stirred at 500 RPM at 60°C for 24 h with pH control set to pH 8.5. After 24 h, the reaction was at 99% conversion. The temperature of the reaction was decreased to 45°C and 12 M HCl was added dropwise to decrease the pH of the reaction to pH 2. The reaction was allowed to stir for 1 h and then the enzyme precipitate was filtered through a glass fritted funnel equipped with cotton towel. The enzyme precipitate was washed three times with 0.1 N HCl (10 mL). The aqueous filtrates were combined and the pH was increased to pH 11 with 5 N NaOH, and then extracted with 2 x 100 mL of IPAc. The combined IPAc layers were washed with 25 mL brine, dried with MgSO4, filtered, and concentrated to give an oil with 67% isolated yield with 99% ee of S-enantiomer. All spectral data matched literature values (S9).

HPLC analysis method: Chiralcel OD-H (250 x 4.6 mm), 25°C, 5 uL inj., 1 mL/min, 10% 2-propanol in heptane. Major (S) = 8.6 min., Minor (R) = 10.0 min.

(R)-(+)-1-(2-naphthyl)ethylamine. A Mettler-Toledo MultiMax RB04-50 reaction system fitted with pH probes and a base dispenser was used to perform each reaction. Each reaction vessel was fitted with a mechanical stirrer, temperature probe, pH probe, base addition line (4 M isopropylamine in water), and a nitrogen sweep line. A stock buffer solution of 0.2 M triethanolamine in deionized water was prepared and adjusted to pH 9 with neat isopropylamine. Isopropylamine hydrochloride (3.70 g, 38.8 mmol, 6.67 equiv.) and PLP (40 mg) were dissolved in 0.2 M triethanolamine buffer (24 g or 24 mL) in the multimax reaction vessel. After brief agitation, round 11 transaminase (140 mg; 14wt.% relative to ketone) was added and gently dissolved. The stirrer head was then attached to the vessel along with the temperature probe, stirrer motor, and 4 M isopropylamine addition line (through the tallest port with the addition line near the top of the actual reaction vessel and along the wall such that the base will be added in above surface). A nitrogen sweep line was also installed through the multimax's tallest port and the sweep rate was set at ~1 fps. The remaining reactor ports are left open to vent. At this point, the stirrer was set for 400 RPM, temperature at 25°C and the pH control loop was off. DMSO (14 mL) was then added to the multimax vessel over 5 min. The vessel was then heated to 50°C and the stirrer was set for 600 RPM. When the temperature stabilized, the Multimax pH probe was installed in one port with a calibrated handheld pH probe installed in another port. After 5 min, the handheld pH probe was removed, quickly rinsed with water, submerged into the pH 7 phosphate buffer standard solution and then re-calibrated. The handheld probe was then removed from the standard buffer, rinsed with water, and put back into the multimax vessel. This external handheld probe was used to calibrate the reading on the internal multimax probe, after which time the handheld probe was then removed. The pH control loop on

S27  

the multimax was turned on with a 20 sec ramp and a setting of pH 8.5. The 4 M isopropylamine base solution was then set to add if the pH dropped below pH 8.4. Next, a solution of 2-acetonaphthone (1.0 g, 5.82 mmol, 1.0 equiv.; 99wt.%) was dissolved in 6 mL of DMSO and the solution was taken up into a 10 mL syringe with long, flexible Teflon tubing attached to the luer lock. This solution addition line was fed through the same port as the base addition line, along the wall and above the surface. The syringe was placed in a syringe pump and set for 7.5 mL volume (total volume including air bubble) at 2.0 mL/h (3 h addition). The syringe pump was tilted such that the air bubble in the syringe would push out any remaining ketone solution at the end of addition. After 4 d, the conversion was typically 95% with 9.1 equivalents (13.2 mL) of 4 M isopropylamine. All spectral data matched authentic sample of (R)-(+)-1-(2-naphthyl)ethylamine purchased from Sigma-Aldrich.

HPLC analysis method: Chiralcel OJ (250 x 4.6mm), 25°C, 5 µL inj., 1mL/min, 10% ethanol and 0.1% diethylamine in heptane. Major (R) = 10.2 min., Minor (S) = 11.8 min.

(R)-2-(2-Fluorophenyl)pyrrolidine. Isopropylamine-hydrochloride (5.63 g) was dissolved with 45 mL of 0.1 M triethanolamine buffer, pH = 8.65 in a 50 mL Falcon tube. PLP (45 mg) was added, dissolved with vortexing and the pH of the solution was adjusted pH = 8.6 with 5 N NaOH. This solution (40 mL) was added to a 100 mL RB equipped with stir bar, septum, and 21 gauge needle, and round 5 transaminase (500 mg) was then added with stirring. A solution of 4-chloro-1-(2-fluorophenyl)-1-oxobutane (500 mg; Rieke Fine Chemicals) in 10 mL DMSO was then added and the flask was stirred at 45°C for 24 h. After 24 h, the mixture was was cooled to RT, the pH was adjust to pH = 3.5 using 6 N HCl and then the aqueous layer was washed with ethylacetate (2 x 25 mL). The pH of the aqueous layer was adjusted to pH = 12.5 using 50 wt% NaOH and then extracted with ethylacetate (50 mL). This organic extract was washed with 1 M aqueous K2CO3 (2 x 5 mL) and then concentrated to give 357 mg of crude product. Purification via chromatography (2.5 x 16.5 cm silica gel column) using 90:10:2 EtOAc:MeOH:NH4OH (aq) as eluent provided 237 mg (57%) of product as colorless oil in 98% ee. All spectral data (with the exception of ee) matched authentic sample of 2-(2-Fluorophenyl)pyrrolidine (CGeneTech, Inc).

Chiral SFC analysis method: Chiralpak AD-H column (250 x 4.6 mm), 35°C, 200 bar, gradient elution from 4% MeOH to 32% MeOH at a rate of 4% per minute with a flow rate of 2.0 mL/min, 10 µL injection, UV detection at 215 nm. Minor (S) = 5.9 min, Major (R) = 6.2 min.

Supporting references

                                                            S1. R. J. Fox, S. C. Davis, E. C. Mundorff, L. M. Newman, V. Gavrilovic, S. K. Ma, L. M. Chung, C. Ching, S. Tam, S. Muley, J. Grate, J. Gruber, J. C. Whitman, R. A. Sheldon, G. W. Huisman, Nature Biotechnol. 25, 338 (2007).

S28  

                                                                                                                                                                                                S2. K. Stutzman-Engwall, S. Conlon, R. Fedechko, H. McArthur, K. Pekrun, Y. Chen, S. Jenne, C. La, N. Trinh, S. Kim, Y.-X. Zhang, R. Fox, C. Gustafsson, A. Krebber. Metab. Eng. 7, 27 (2005).

S3. J. Liao, M. K. Warmuth, S. Govindarajan, J. E. Ness, R. P. Wang, C. Gustafsson, J. Minshull, BMC Biotechnology 7, 16 (2007).

S4. A. Iwasaki, Y. Yamada, N. Kizaki, Y. Ikenaka, Appl. Microbiol. Biotechnol. 69, 499 (2006).

S5. ATA-117 protein sequence: MAFSADTSEIVYTHDTGLDYITYSDYELDPANPLAGGAAWIEGAFVPPSEARISIFDQGYLHSDVTYTVFHVWNGNAFRLDDHIERLFSNAESMRIIPPLTQDEVKEIALELVAKTELREAFVSVSITRGYSSTPGERDITKHRPQVYMYAVPYQWIVPFDRIRDGVHAMVAQSVRRTPRSSIDPQVKNFQWGDLIRAVQETHDRGFEAPLLLDGDGLLAEGSGFNVVVIKDGVVRSPGRAALPGITRKTVLEIAESLGHEAILADITLAELLDADEVLGCTTAGGVWPFVSVDGNPISDGVPGPVTQSIIRRYWELNVESSSLLTPVQY

S6. S. C. Davis, J. H. Grate, D. R. Gray, J. M. Gruber, G. W. Huisman, S. K. Ma, L. M. Newman, R. Sheldon, L. A. Wang, US patent 7,132,267 (2006).

S7. W. P. C. Stemmer, Nature 370, 389 (1994).

S8. M. D. Truppo, Ph.D. Thesis, University of Manchester (2009). Reaction conditions: 4 g/L ATA-117, 5 g/L ketone using GDH/Glucose/NADH recycle system.

S9. V. Y. Truong, M. S. Menard, I. Dion Org. Lett. 9, 683 (2007).