Directed-evolution ω-transaminase replacing Rh-catalyzed asymmetric hydrogenation for chiral amine API manufacture — the ATA-117 / sitagliptin precedent
Sitagliptin, the active ingredient of Merck's Januvia (a DPP-4 inhibitor for type 2 diabetes), requires a single chiral amine stereocenter installed with very high enantiomeric excess. The original commercial route hydrogenated a prochiral enamide precursor over a rhodium catalyst bearing a chiral Josiphos-type phosphine ligand at elevated hydrogen pressure — a technically sound but operationally and economically heavy process that biocatalysis was ultimately built to replace.
The classical synthesis reduces an unsaturated enamide precursor of sitagliptin using molecular hydrogen across a rhodium center ligated by a chiral bisphosphine:
• Substrate: (Z)-enamide bearing the 2,4,5-trifluorophenyl and triazolopyrazine amide substituents already installed • Catalyst: Rh(I)-Josiphos complex, generated in situ from [Rh(COD)2]BF4 and a proprietary ferrocenyl-phosphine ligand • Conditions: 250–500 psi H2, methanol/toluene solvent, 50°C, catalyst loading ~0.15 mol% • Mechanism: oxidative addition of H2 to Rh(I), followed by migratory insertion into the coordinated prochiral olefin face selected by the chiral ligand pocket, then reductive elimination to release the saturated β-amino amide • Outcome: ~95% ee as the free base, which then required an additional diastereomeric salt recrystallization to reach pharmaceutical-grade enantiopurity
This route is chemically elegant and was, at the time, state-of-the-art industrial asymmetric catalysis — but every element of it (precious metal, high pressure, proprietary ligand, extra purification) adds cost, waste, and regulatory burden.
Three linked problems made the metal-catalyzed route an attractive target for replacement:
1. Precious-metal cost and supply risk: • Rhodium spot price is volatile and historically among the most expensive precious metals (frequently $/oz exceeding platinum and gold) • Catalyst is not perfectly recyclable at manufacturing scale; metal losses during workup are unavoidable
2. Residual metal specification (ICH Q3D elemental impurities): • Rh is a Class 2A metal under ICH Q3D — permitted daily exposure limits are strict • Final API must contain <10 ppm residual Rh, requiring dedicated scavenging resins, activated carbon treatment, or repeated crystallizations solely to strip metal — steps that consume solvent and yield
3. High-pressure hydrogenation infrastructure: • 250–500 psi H2 requires specialized pressure-rated reactors, additional safety systems (H2 is flammable/explosive), and trained operations — capital-intensive and inflexible for multi-product plants • Chiral phosphine ligand synthesis (e.g., Josiphos family) is itself a multi-step, costly process chemistry exercise layered on top of the hydrogenation step
None of these three problems is fixable by better process engineering alone — they are intrinsic to using a precious transition metal under H2 pressure to set a stereocenter. This is precisely the gap that an engineered enzyme, operating in water at ambient pressure with no metal at all, was designed to close.
ω-Transaminases (EC 2.6.1.18-type, PLP-dependent, class III aminotransferase fold) transfer an amino group from a donor amine to a prochiral ketone, installing a stereocenter with near-perfect selectivity in one step. The catalytic chemistry was never the problem — the wild-type enzyme's binding pocket was simply too small to accept the bulky prositagliptin ketone. Codexis, working with Merck, ran an 11-round directed evolution campaign (Savile et al., Science 2010) that is now the textbook case study in industrial protein engineering.
ω-Transaminases naturally act on small, unhindered methyl ketones or aliphatic amino-acid-like substrates. The prositagliptin ketone carries two bulky groups flanking the reacting carbonyl:
• A 2,4,5-trifluorophenyl ring on one face • A triazolopyrazine amide on the other
Homology modeling (built from a related transaminase crystal structure, since no structure of the wild-type enzyme itself was available) showed the small binding pocket — normally sized for a methyl group — sterically clashing with both substituents. Initial wild-type activity on the real substrate was essentially undetectable (a few percent conversion after extended reaction times, if any).
Rather than starting from scratch, the strategy was to enlarge the pocket in stages: computational modeling identified residues lining the small and large binding pockets, and site-saturation mutagenesis libraries were built at those positions.
The campaign combined rational, structure-guided design with classical directed evolution screening:
Rounds 1–4 (pocket-opening, structure-guided): • Homology-model-guided site-saturation mutagenesis at residues lining the small binding pocket • Selection criterion: any detectable activity on a simplified surrogate methyl-ketone analog of the real substrate (activity too low to screen on the real bulky ketone yet) • Result: first variants showing low but measurable activity on the surrogate, establishing a foothold
Rounds 5–8 (activity on the real substrate, combined libraries): • Once surrogate activity was established, screening shifted to the actual prositagliptin ketone • Random mutagenesis (error-prone PCR) layered on top of the pocket mutations to pick up second-shell and remote stabilizing mutations • Combinatorial recombination of beneficial mutations from parallel lineages (DNA shuffling-like approach)
Rounds 9–11 (activity + stability + solvent tolerance co-optimization): • Selection pressure expanded beyond raw activity to include tolerance of high substrate/product concentration, DMSO co-solvent, and elevated temperature (process-relevant conditions) • Final evolved variant: 27 total mutations relative to the wild-type-derived starting point, distributed across the small pocket, large pocket, and surface/stability positions
The net result was better than a simple "hit or miss" screen — each round's winning variant became the template for the next round's library, compounding small gains into an overall >25,000-fold improvement in activity on the real, bulky pharmaceutical substrate.
Crystallographic and modeling analysis of the evolved enzyme, compared to the wild-type-derived starting point, showed the accumulated mutations acting through several coordinated mechanisms:
• Small-pocket enlargement: bulky residues lining the pocket that normally accepted only a methyl group were replaced with smaller side chains, opening room for the trifluorophenyl ring • Large-pocket reshaping: complementary substitutions accommodated the triazolopyrazine amide on the opposite face • Loop remodeling: flexible loop segments near the active-site entrance repositioned to funnel the larger substrate toward the PLP cofactor • Distal stabilizing mutations: several mutations far from the active site had no direct catalytic role but rescued protein stability lost by the pocket-opening substitutions — a common pattern in directed evolution, where "second-order" compensatory mutations are essential for a viable final enzyme
The ATA-117 campaign is frequently cited as proof that directed evolution can convert an enzyme with no meaningful activity on an industrially relevant, structurally demanding substrate into a manufacturing-ready biocatalyst — without ever solving a crystal structure of the starting enzyme. Homology modeling plus iterative screening was sufficient.
Transamination is thermodynamically reversible: the same PLP-dependent chemistry that converts ketone + amine donor → chiral amine + co-product ketone can run backward just as readily. Achieving high, practical conversion of the prositagliptin ketone therefore required reaction engineering as much as enzyme engineering — driving an intrinsically balanced equilibrium decisively toward the desired product.
ω-Transaminases operate through a "ping-pong bi-bi" kinetic mechanism built around the PLP cofactor:
Half-reaction 1: PLP (aldehyde form) + isopropylamine → pyridoxamine-5'-phosphate (PMP) + acetone (released) Half-reaction 2: PMP + prositagliptin ketone → PLP (regenerated) + chiral sitagliptin amine (released)
Each half-reaction is reversible. The overall equilibrium constant for transamination between a simple amine donor and a ketone substrate is often close to 1, meaning a naive 1:1 stoichiometric reaction would stall near 50% conversion — commercially unacceptable for an API intermediate that must reach near-quantitative, high-purity conversion.
Two independent levers are used to displace the equilibrium toward the desired chiral amine product: donor excess and co-product removal.
Isopropylamine (IPA) was selected as the amine donor for several practical reasons:
• Low cost and ready availability at manufacturing scale • Small, symmetric molecule (2-propanamine) whose co-product, acetone, is volatile and easy to remove from the reaction mixture • No stereocenter of its own, so no risk of introducing an unwanted diastereomer
By dosing IPA in substantial molar excess relative to the ketone substrate (multi-fold excess, tunable against cost and downstream removal burden), the forward half-reaction is kinetically and thermodynamically favored: mass action pushes PMP formation, which in turn feeds the second half-reaction toward the chiral amine product. The trade-off is that excess unreacted IPA and its salts must be efficiently recovered or removed downstream, an engineering balance between conversion, cost, and waste (E-factor) — favoring the smallest excess that still reaches target conversion.
The second, complementary lever removes acetone as it forms, preventing the reverse half-reaction from re-consuming product amine:
• Reduced-pressure operation / vacuum stripping: acetone (bp 56°C) is volatile enough to be continuously stripped from the aqueous reaction mixture under mild vacuum, especially as reaction temperature and mixing are optimized • Enzymatic acetone scavenging: an auxiliary enzyme system (e.g., a ketone-consuming or acetone-reducing coupled enzyme) can be used in some transaminase process designs to continuously destroy the co-product, making the overall transformation effectively irreversible • Combined effect: with both donor excess and active acetone removal, single-pass conversions of >90–99% are routinely achievable, versus the ~50% ceiling implied by a naive equilibrium
The PLP cofactor itself is used sub-stoichiometrically (catalytic loading, typically micromolar to low-millimolar) because it is continuously regenerated within the ping-pong cycle — it is a shuttle, not a stoichiometric reagent, which keeps cofactor cost a minor contributor to overall process economics.
Reaction engineering — not just enzyme engineering — was essential to making the biocatalytic route commercially viable. An enzyme evolved for activity and stability still needs a reactor strategy (donor excess + co-product removal) to convert a reversible equilibrium into a near-quantitative, one-directional manufacturing process.
An evolved enzyme with excellent activity in a small-scale test tube still has to survive the realities of plant-scale manufacturing: high substrate concentration, defined temperature/pH windows, practical enzyme dosing, and a reactor and workflow that a regulated pharmaceutical manufacturing site can run reliably, batch after batch, at multi-ton scale.
Unlike many industrial enzyme applications that require costly chromatographic purification, the evolved transaminase is deployed as a lyophilized cell-free extract (CFE) directly from the fermentation host (typically an engineered E. coli production strain):
• Fermentation: the evolved transaminase gene is expressed under a strong inducible promoter in a high-cell-density E. coli fermentation, yielding enzyme as a substantial fraction of soluble cell protein • Cell lysis and clarification: cells are lysed, cell debris removed by centrifugation/filtration, and the clarified extract is lyophilized (freeze-dried) directly — no chromatography step • Dosing: the dried powder is weighed and dosed directly into the aqueous reaction at a defined enzyme loading (g of CFE per liter of reaction), analogous to dosing a fine chemical catalyst • Cost driver: because purification is skipped, enzyme production cost per kg of API scales favorably with fermentation titer improvements — a lever process engineers can keep pulling even after the protein sequence itself is fixed
This "formulate as crude extract" strategy is a deliberate economic choice: purification would add cost without adding selectivity, since the crude extract's residual host-cell protein is removed downstream in normal API workup, not by enzyme purification.
The manufacturing reaction is run as a stirred-tank batch (or fed-batch) process in standard glass-lined or stainless steel vessels — no specialized high-pressure hydrogenation equipment is required:
• Temperature: optimized in the 40–50°C range, balancing enzyme activity (higher temperature generally faster) against thermal stability (the evolved variant's later rounds specifically selected for elevated-temperature tolerance) • pH: maintained near neutral-to-mildly-alkaline (typical PLP-enzyme optimum, pH ~7.5–8.5), controlled by base addition as the reaction proceeds • Substrate feed strategy: fed-batch addition of the ketone substrate avoids high initial substrate inhibition and keeps the enzyme working in its optimal concentration range throughout the batch • Co-solvent tolerance: process-relevant DMSO or other water-miscible co-solvent may be used to keep the poorly water-soluble ketone substrate dissolved at high loading — a tolerance trait specifically selected for during the later evolution rounds
The achieved substrate loading of ~200 g/L (~13 wt%) is exceptionally high for a biocatalytic process — most enzymatic reactions are run at far lower substrate concentration because of solubility and inhibition limits, another testament to how thoroughly the enzyme was engineered for real manufacturing conditions, not just for a single-turnover activity assay.
The key manufacturing metric process chemists track when comparing routes is volumetric (space-time) productivity — how much product a given reactor volume produces per unit time, since reactor capacity is often the binding capital constraint at a manufacturing site.
Because the biocatalytic route: • Eliminates the specialized high-pressure hydrogenation train (freeing reactor availability for other uses) • Runs at high substrate loading (200 g/L) in standard vessels • Avoids the multi-step metal-scavenging purification train that added cycle time to the Rh route
…the overall reported volumetric productivity of the biocatalytic process was approximately 53% higher than the Rh-catalyzed hydrogenation route it replaced, while simultaneously removing an entire class of hazardous, high-pressure operations from the plant.
A common misconception is that switching to biocatalysis trades yield/throughput for "greenness." In the sitagliptin case, the opposite was true: the engineered enzyme process was both greener AND more productive per unit reactor volume than the metal-catalyzed route it replaced — a rare win-win that made the business case for the multi-year engineering campaign straightforward to justify.
The last mile of any API process — isolation, purification, and final quality control — is where the cumulative benefits of the biocatalytic route become concrete, auditable numbers: enantiomeric excess, isolated yield, waste generated per kilogram of product (E-factor), and process mass intensity (PMI). The transaminase route to sitagliptin became a widely cited industrial case study precisely because these numbers were so favorable, and because the process won the U.S. EPA Presidential Green Chemistry Challenge Award (Merck & Codexis, 2010).
Because the biocatalytic reaction contains no transition metal at any point, the entire downstream purification train dedicated to stripping residual rhodium to <10 ppm is eliminated outright:
• No activated-carbon treatment cycles targeting metal residues • No dedicated metal-scavenging resin columns • No repeated recrystallizations solely to chase the ICH Q3D elemental-impurity limit
Instead, downstream processing is a comparatively simple sequence: reaction quench/pH adjustment, extraction of the chiral amine product into an organic phase, removal of residual host-cell protein and excess IPA/salts by standard aqueous workup, and crystallization of the sitagliptin free base (or a downstream salt form) directly to pharmaceutical-grade purity — no additional chiral enrichment step needed, because the enzyme already delivers >99.95% ee in the reaction itself.
Comparing the full biocatalytic process to the Rh-catalyzed hydrogenation route it replaced, the reported improvements (Savile et al., Science 2010; Desai, Angew. Chem. Int. Ed. 2011) span both economic and environmental dimensions:
• Overall isolated yield: increased by approximately 10–13% relative to the metal-catalyzed route • Total waste generated (E-factor basis): reduced by approximately 19% • Volumetric productivity (space-time yield): increased by approximately 53% • Heavy/precious-metal usage: reduced to zero — no rhodium, no chiral phosphine ligand, and no associated metal-recovery/recycling operations • High-pressure hydrogenation infrastructure: no longer required, freeing capital-intensive equipment for other manufacturing campaigns and removing an entire category of process safety hazard (flammable H2 at 250–500 psi)
These are not marginal, single-digit-percent improvements typical of routine process optimization — they represent a genuine step-change enabled by replacing the chemistry itself, not merely tuning conditions around an existing chemical mechanism.
The ATA-117/sitagliptin transaminase project (Merck and Codexis, published in Science, 2010) is cited across the pharmaceutical and green-chemistry literature for several reasons that generalize well beyond this one molecule:
1. It demonstrated that directed evolution could take an enzyme from essentially zero activity on an industrially relevant, sterically demanding substrate to a manufacturing-ready biocatalyst within a feasible, multi-year but tractable engineering timeline 2. It showed that biocatalytic routes, properly engineered at both the protein and reactor-process level, can out-produce (not just "green-wash") an established, highly optimized metal-catalyzed route on volumetric productivity 3. It provided a template — structure-guided pocket engineering, followed by iterative combinatorial screening, followed by process-conditions co-optimization (stability, solvent tolerance, high substrate loading) — that has since been reapplied to numerous other transaminase, ketoreductase, and other biocatalytic API intermediate projects across the industry
The project won the 2010 U.S. EPA Presidential Green Chemistry Challenge Award, cementing its role as the go-to teaching case for how enzyme engineering can directly displace precious-metal asymmetric catalysis in pharmaceutical manufacturing.
The single number that best captures the achievement: an enzyme with immeasurably low activity on the real substrate was evolved, over 11 rounds and 27 mutations, into a biocatalyst more than 25,000 times more active — active enough to run at 200 g/L substrate loading, >99.95% ee, and higher volumetric productivity than the rhodium catalysis it replaced.