Buffer salts (Sigma Aldrich), NAD⁺ (Prozomix), carbon black particles (Black Pearls 2000 “BP 2000”, Cabot Corporation), activated charcoal (Darco^®, −100 mesh, Sigma Aldrich), Pd/C powder (5R452, 5% palladium on Activated Carbon Paste, Type 452, Johnson Matthey), Pd/C pellets (1% palladium, Type 783 pellet, 1–3 mm, Johnson Matthey), quinuclidinone-HCl (1, Sigma Aldrich), quinuclidinol (2 as both R and racemic mixture, Sigma Aldrich) were all purchased and used as received. All aqueous solutions were prepared with deionised water. ¹H NMR spectra were obtained at room temperature (298 K) on a Bruker Advance III HD nanobay (400 MHz) using 90 vol% reaction mixture and 10 vol% D₂O. The Bruker proc_1d processing algorithm was used, and a manual phase correction was followed by a multipoint baseline correction. Chiral-phase GC-FID was used as described in S2.3 (Supplementary Information). UV-visible spectra were recorded by a Jenway 7205 spectrophotometer using a quartz cuvette (path length 1 cm). Thin-layer chromatography (TLC) analysis was carried out using silica gel on aluminium plates. Visualisation of the TLC plates was achieved using I₂.

The hydrogenase (E. coli hydrogenase 1, “Hyd1”, molecular weight 100 kDa) was produced and isolated following published protocols (Ramirez et al., 2022). Subsequently, the enzyme was buffer-exchanged into Tris-HCl buffer (20 mM Tris-HCl pH 7.2, 350 mM NaCl, 0.02% Triton X, 1 mM DTT), concentrated to 14.7 mg/mL and stored at −80°C. The NAD⁺ reducing soluble hydrogenase from Ralstonia eutropha (“SH”) was isolated and purified following published protocols (Lauterbach and Lenz, 2013), then stored at −80°C with a concentration of 21.6 mg/mL in Tris-HCl (100 mM, pH 8.0). This enzyme retains its native H₂ oxidation activity but significantly benefits from improved activity and stability via the co-immobilised Hyd1 (Poznansky et al., 2021). The NADH-dependent 3-quinuclidinone reductase from Agrobacterium tumefaciens (AtQR) was heterologously produced in E. coli as previously described (Rowbotham et al., 2020) but was used in this report as the soluble lysate without further purification. The resulting solution of AtQR soluble lysate had a total protein concentration of 143 mg/mL.

Separate slurries (20 mg/mL) of carbon black and activated charcoal were sonicated and dispersed in buffer following the description above. In separate centrifuge tubes, 100 µg NAD⁺ reductase was mixed with 100 µg of the designated carbon material slurry (either carbon black or activated charcoal) in a total volume of 20 µL. The NAD⁺ reductase was immobilised by mixing at 600 r.p.m. in a Thermoblock shaker at 4°C for 1 h, then slurries were centrifuged for 3 min (10,000 × g). The supernatant was removed and analysed by UV-vis to measure the remaining “unabsorbed protein concentration” using the absorbance at 280 nm (see results corresponding with “0” wash cycles in Results; see details in S2.1 in the Supplementary Information for information on calculating the protein adsorption).

From here, the catalysts were subjected to “washing cycles”, by suspending in 10 µL of solution (either 25 mM Tris-HCl, or 10 mM quinuclidinone in 25 mM Tris-HCl) and mixing at 600 r.p.m. on the Thermoblock shaker at 30°C for 1 h. The slurries were then centrifuged for 3 min (10,000 × g). The supernatant was removed and analysed by UV-vis to measure the remaining ‘unabsorbed protein concentration’ using the absorbance at 280 nm. Washing cycles were repeated thrice and analysed for unabsorbed protein in the supernatant, shown in Figure 2.

All continuous hydrogenation reactions were run in the reactor set up depicted in Supplementary Figure S2 (Supplementary Information). Details of the set up are described below.

When planning to translate our hydrogenation biocatalyst into a slurry reactor, we first sought suitable conditions that were likely to lead to high conversion and enable lowered catalyst loading. The ACR/ACX flow reactor has previously handled nanoparticles which had an average particle size of 70 nm (Chuzeville et al., 2022).

A variety of carbon materials have been proven as suitable supports for the biocatalytic hydrogenation system (Reeve et al., 2015; Thompson et al., 2020b), which enables us to take advantage of the different properties (e.g., dispersion, suspension, enzyme adsorption). For example, enzymes immobilised on carbon black nanopowder offer excellent dispersion when mixed in batch. In a previous report, enzymes adsorbed onto −100 mesh activated charcoal particles are suitable in packed bed flow reactors without any carbon leaching (Poznansky et al., 2021), whereas nanomaterials can leak through the packed bed frits. An advantage of dynamically-mixed slurry processes is the possibility of using high surface area nanoparticles as a (this report) compared to packed bed reactors that typically require confined mm sized materials. Advantageously the reactor used in this work allows operation with catalyst in slurry or confined mode.

Initially, we examined the ability of the two carbon types for immobilisation of hydrogenase and NAD⁺ reductase via simple absorption (Table 1, with details described in Section 2.1 in the Supplementary Information). In the case of activated charcoal, when the mass ratio of enzyme to carbon was 2:1, only 52% of the enzyme adsorbed, whereas when the mass ratio of enzyme to carbon was decreased to 0.5:1, 65% was adsorbed (entries 1–2). Carbon black had a superior ability to adsorb the enzymes, even at high (3:1) mass ratio of enzyme to carbon black, resulting in 72% enzyme adsorption, and a moderate 1:1 mass ratio of enzyme to carbon resulted in 96%–98% adsorption (tested at two different scales, entries 3–5). This high enzyme loading on the nano sized carbon material suggests that translation into continuous reactor with high activity should be prioritised as a slurry process.

We next sought to determine if the NAD⁺ reductase leaches from the carbon after immobilisation (see procedure details in Section 2.3). The NAD⁺ reductase was first immobilised onto the two carbon materials. The slurry was then centrifuged and the supernatant was analysed for protein in solution, which showed NAD⁺ reductase immobilisation that in line with those in Table 1 (shown as 0 wash cycles, Figure 2). Subsequently, the immobilised enzyme on carbon aliquots were subjected to washing cycles with Tris-HCl buffer with or without presence of quinuclidinone solution (to mimic reaction conditions, procedure described in detail in Section 2.3). The supernatants were analysed for NAD⁺ reductase leaching between washes. This information was used to calculate the percent of enzyme applied that was retained on the carbon materials.

The results in Figure 2 show minimal leaching, significantly, with negligible (<0.5%) NAD⁺ reductase lost in the absence quinuclidinone. In the presence of 10 mM quinuclidinone and 25 mM Tris-HCl, only 5% (activated charcoal) or 2% (carbon black) of the NAD⁺ reductase had leached. The quinuclidinone may interfere with some of the interactions that allow the enzyme to stay adsorbed to the carbon materials, however the quantity lost over the three wash cycles supported moving forwards with a carbon black slurry process.

The heterogeneous biocatalysts were next used for the hydrogenation of 3-quinuclidinone (1) in scale-down batch (Figure 3). Here, we sought to achieve key reaction parameters required for translation into continuous flow slurry process, including short reaction times and low (0.2–0.6 mg/mL) catalyst loading.

The activated charcoal and carbon black biocatalyst systems were compared in parallel for the ability to drive the biocatalytic hydrogenation of 1 to 2 following the procedure detailed in Section 2.4. Solutions containing 3-quinuclidinone-HCl (5 mM) and 1 mM NAD⁺ (1 mM) in Tris-HCl (25 mM, pH 8.0, 25 °C) were mixed with heterogeneous biocatalyst (0.2 mg/mL) and AtQR (0.05 mg/mL) under a steady flow of H₂ (1 bar) at 25°C for 30 min.

The reactions were analysed for conversion to 2 using ¹H NMR spectroscopy (see results in Table 2). Entries 1 and 2 show that the catalyst formulated using carbon black had marginally higher (84%) conversion per mg biocatalyst compared with that which was formulated using activated charcoal (79%), although due to the catalyst formulation, a higher enzyme turnover frequency (TOF) was observed using activated charcoal. We also observed improved mixing using the carbon black particles, which did not settle as rapidly as the activated charcoal. The particle size is likely a contributing factor to this, where the carbon black “BP 2000” particle size is on the nanometer scale (Zhao et al., 2021) and the −100 mesh activated charcoal is on the micron scale. Conversion to 2 in 30 min was further improved to 90% at elevated reaction temperature (35°C, entry 3, Table 2). Although performed at modest reactant loadings, previous reports on this biocatalyst system in batch have used 24 h reaction times and this is an initial step to achieving high conversion in a timeframe suitable for translation into flow, whilst maintaining relatively low carbon loadings that are suitable for slurry processes.

A comparison Pd/C-catalysed hydrogenation (entry 4) was run using analogous reaction conditions (to entry 3, 0.2 mg/mL catalyst loading, 35°C, 1 bar H₂), though deionised water was used in place of buffer in order to avoid additives that could poison the Pd. After 30 min, a 7% conversion was determined using ¹H NMR spectroscopy, and this equates to a Pd TOF of 0.1 min^-1. This low conversion and catalyst activity could be explained by the likelihood that the Pd requires higher H₂ availability usually achieved via elevated temperature and pressure, improved mixing, or a different solvent. For example, Pd/C has previously been used in tandem with a chiral-rhodium catalyst to hydrogenate 1 (84% yield, 58% ee) using intensified conditions: 75°C and 10–14 bar H₂ in methanol (Brieden, 1996). It’s worth noting that such conditions requiring elevated pressures complicate translation into a continuous slurry process.

High conversions to 2 using the biocatalytic hydrogenation strategy in an aqueous, ambient batch reactions provided the groundwork to move the system into a continuous slurry flow process. Here the focus is on understanding if the biocatalyst system can simply “slot-in” without significant chemical or process engineering.

In comparing the protocols for biocatalytic hydrogenation of 1 compared to typically metal-catalysed hydrogenations, a key difference to be addressed is in the solvent, moving from organic solvents for metals to aqueous media for biocatalytic with implications for H₂ solubility. For example, the solubility of hydrogen at 35°C is ca. 8 times lower (when comparing by molality) in water than for methanol under comparable pressures (Descamps et al., 2005). We expected that using the Coflore ACR would help overcome this rate limitation due to the three-fold increase regarding mass transfer coefficients (Toftgaard Pedersen et al., 2017), opening up opportunities to use aqueous solvent. The use of water in place of methanol or other toxic, flammable organic solvents is desirable because it lowers overall health and safety risks of a hydrogenation process (Prat et al., 2014; Byrne et al., 2016).

To test the Coflore ACR with an aqueous hydrogenation reaction, we took the biocatalyst formulated on the carbon black nanopowder and the 35°C reaction temperature forward to translate to continuous flow. With a typical particle size range of 20–50 nm as measured through transmission electron microscopy (Weingarth et al., 2014), and a specific surface area of 1272 m²/g as measured through nitrogen gas sorption and BET analysis (or 1949 m²/g using the Langmuir model [Zhao et al., 2021)], risk reduction in sedimentation, as well as a higher activity per catalyst mass were expected when compared to larger, lower surface area carbon catalyst materials (see comparable data for a variety of Pd/C catalysts in Zhao et al., 2021 supporting information). Reliable slurry feed at flow rates as low as 0.4 mL/min was possible using a Masterflex peristaltic pump when using Norprene LS14 flexible tubing.

The continuous biocatalytic hydrogenation was set up as shown in Figure 4 (see details in Section 2.5), where a slurry feed that contained 1 (5 mM), NAD⁺ (1 mM), heterogeneous biocatalyst (0.15 mg/mL), and AtQR (0.05–0.40 mg/mL) in Tris-HCl buffer (75 mM, pH 8.0 at 25°C) was stirred in a jacketed vessel. This slurry was pumped into the ACR where it as mixed with H₂ at 35°C at various residence times (tRes) achieving 2 bar overall system pressure. Reaction slurries were then collected in a vessel, with samples taken for NMR spectroscopy analysis after every reactor volume. Additional details for process is shown in Supplementary Table S2 (Supplementary Information).

The results in Table 3 show conversions to 2 during the steady state using different continuous flow parameters. For the reaction summarised in entry 1, a Coflore ACR block was initially used in conjunction with 10 high shear agitators resulting in a 90 mL reactor assembly. The suspension feed contained 5 mM 1, 0.15 mg/mL heterogeneous biocatalyst and 0.05 mg/mL AtQR. The slurry flow rate was set such that the liquid phase tRes in the agitated reactor was 15 min. Under these conditions, a 35% conversion to 2 was reached during steady state, with catalyst TOF (286 min⁻¹) similar to that achieved in batch suggesting that the biocatalyst system had good compatibility with the slurry continuous mode.

For the reactions summarised in entries 2–4 (Table 3), the tRes was increased to 54 min in order to achieve higher conversions. The high shear agitators were substituted for 50% volume agitators, which decreased the reactor volume from 90 mL to 50 mL. The Coflore ACR block was substituted for an ACX block to minimise the gas phase holdup, whilst maintaining a constant liquid phase volume, and minimisation of working volumes. At steady state, a 67%–71% conversion was achieved (entry 2). We estimate the NAD⁺ reductase TOF of 161 min^-1 based on concentration of 2 generated, although a build up of NADH was observed, suggesting that the reaction was limited here by AtQR loading. Regardless, this catalyst activity is still on the same order of magnitude as those achieved in batch (see values in Table 2).

For comparison, a slurry feed that contained 1 (5 mM) and 5% Pd/C (0.15 mg/mL) in deionised water was pumped through the ACX reactor such that tRes was 54 min at 35°C and under 2 bar H_2. Under these conditions, there was no detection of 2 (entry 4, Table 3). Similar to the low conversion when Pd/C was used these mild temperature and pressure conditions in batch, it is likely that harsher conditions or higher catalyst loading would be required for Pd-catalysed hydrogenation of 1. The Pd mass loading was 7.5 μg/mL (ca. 0.8 mmol/mL), whereas the immobilised enzyme loading was six orders of magnitude smaller per presumed ‘active site’ (ca. 5 × 10⁻⁷ mmol/mL).

Results in entry three show that on doubling tRes to 108 min and increasing AtQR loading to 0.3 mg/mL, 100% conversion to 2 was achieved at steady state. The sample collected at 8 h (4 reactor volumes) was also analysed by chiral-phase GC-FID, and only the R-enantiomer of 2 was detected (>99% ee).

Although these results are promising, the substrate loading is low. Therefore the reaction was next intensified to 50 mM 1 with increased heterogenous biocatalyst (0.45 mg/mL) and AtQR (0.40 mg/mL) loading (entry 1, Table 4). Continuous process was carried out with 108 min slurry tRes, 35°C and 2 bar H₂. During steady state, a 18%–20% conversion to 2 was achieved (entry 5), showing a step improvement to provide 10 mM 2 in continuous flow.

A comparable batch reaction was also carried out (entry 2), leading to higher conversion (53%, 26.5 mM 2) and catalyst activity (341 min⁻¹). Given the low TOF for entry 1 (65 min⁻¹), the slurry mode continuous process is likely to benefit from improved H₂ availability which could be achieved via improved mixing, use of solvent or increased pressure.

For operation of continuous hydrogenation in Coflore reactors, metal/C catalysts can be handled in packed basket agitators. This prevents a need to pump slurries under higher pressure conditions, and advantageously retains the catalyst in the reactor system.

To demonstrate the Pd/C hydrogenation of 1 under increased pressure more typically used for metal-catalysed hydrogenations, the reactor set up was adjusted. The conditions (10 bar H₂) required that an Elados EMP II E60 solenoid driven diaphragm pump be used in place of the peristaltic pump that was used with the more mild slurry reactions. This in turn necessitated using Pd/C in packed catalyst basket agitators rather than pumping through the ACX as a slurry, as the Pd/C particles increase the risk of blockages in the diaphragm pump check valves or settling in the diaphragm chamber.

Catalyst basket agitators (ACR-CB-Ha) were packed with Pd/C pellets (1 wt% Pd) and fitted into the ACX (see Supplementary Figure S4 in the Supplementary Information for diagram). A solution of 1 (5 mM) in deionised water was then pumped through the agitated (5 Hz) ACX such that a 56 min tRes was achieved at 35°C under 10 bar H₂ (results summarised in entry 1, Table 5). When the catalyst basket agitators were completely filled with Pd/C pellets (5.34 g total catalyst, 68 mL reactor volume), full conversion to 2 was determined during the steady state. Despite the excellent conversion, we note that this equates to <3 total Pd turnover numbers owing to the high catalyst loading.

In a second experiment the overall Pd/C loading to 0.96 g (78 mL reactor volume) and increased the [1] to 63.7 mM, in order to overall lower the mass ratio of catalyst: substrate. As catalytic activity is commonly surface area-limited (mass transfer is limited to the number of catalyst active sites), equalising catalyst loading based on this metric allows a more direct comparison of activity, cost, and performance. In this case, the conversion to 2 averaged to 17% during steady state (entry 2). This equates to 10.8 mM 2 generated and 37 total Pd turnovers (Pd TOF = 0.16 min⁻¹). This Pd activity is similar to the Pd activity achieved during the batch reaction that was performed at 35°C and 1 bar H₂ (see entry 4, Table 2). A sample from the fourth reactor volume (4 h) was also analysed using chiral-phase GC-FID, which revealed that a mixture of both enantiomers were generated by the metal catalyst.