Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

3D printed ceramics as solid supports for enzyme immobilization: an automated DoE approach for applications in continuous flow

3D printed ceramics as solid supports for enzyme immobilization: an automated DoE approach for... In recent years, 3D printing has emerged in the field of chemical engineering as a powerful manufacturing technique to rapidly design and produce tailor-made reaction equipment. In fact, reactors with complex internal geometries can be easily fabricated, optimized and interchanged in order to respond to precise process needs, such as improved mixing and increased surface area. These advantages make them interesting especially for catalytic applications, since customized structured bed reactors can be easily produced. 3D printing applications are not limited to reactor design, it is also possible to realize functional low cost alternatives to analytical equipment that can be used to increase the level of process understanding while keeping the investment costs low. In this work, in-house designed ceramic structured inserts printed via vat photopolymerization (VPP) are presented and characterized. The flow behavior inside these inserts was determined with residence time distribution (RTD) experiments enabled by in-house designed and 3D printed inline photometric flow cells. As a proof of concept, these structured inserts were fitted in an HPLC column to serve as solid inorganic supports for the immobilization of the enzyme Phenolic acid Decarboxylase (bsPAD), which catalyzes the decarboxylation of cinnamic acids. The conversion of coumaric acid to vinylphenol was chosen as a model system to prove the implementation of these engineered inserts in a continuous biocatalytic application with high product yield and process stability. The setup was further automated in order to quickly identify the optimum operating conditions via a Design of Experiments (DoE) approach. The use of a systematic optimization, together with the adaptability of 3D printed equipment to the process requirements, render the presented approach highly promising for a more feasible implementation of biocatalysts in continuous industrial processes. . . . . Keywords 3D printing Biocatalysis Automation Continuous flow DoE. Highlights • Design and printing of ceramic inserts engineered for biocatalytic in flow applications. � Characterization of the inserts via Residence Time Distribution experiments, enabled by in house designed and produced 3D printed flow cells. � Immobilization of bsPAD onto the inserts via covalent binding, followed by the application of a DoE approach in an automated setup to quickly identify the optimum operating parameters. * Alessia Valotta CATalytic mechanisms and AppLications of OXidoreductases valotta@tugraz.at (CATALOX), Graz, Austria * Heidrun Gruber-Woelfler Center for Continuous Flow Synthesis and Processing (CCFLOW), woelfler@tugraz.at Research Center Pharmaceutical Engineering GmbH (RCPE), Graz, Austria Institute of Process and Particle Engineering, Graz University of Lithoz GmbH, Vienna, Austria Technology, Graz, Austria 676 J Flow Chem (2021) 11:675–689 Introduction leakage during application. However, the mechanical proper- ties can be improved in the post-processing phase via steps of In recent years, the interest in 3D printing has expanded to drying, heating, or sintering [11]. The resolution for ME many branches of science and engineering, due to the decreas- printed parts is lower, but this technique is still widely used ing cost of desktop printers and the increasing choice of print- due to its high versatility and low costs [11]. ing materials available [1, 2]. In the field of chemical engi- As mentioned above, 3D printed reactors are widely used neering, 3D printing has been identified as a new solution to in flow chemistry, and many reviews can be found on this rapidly design and produce tailor-made reaction ware. This is topic [1, 10, 12–14]. Nevertheless, the use of 3D printing to particularly interesting for the pharmaceutical industry, since realize supports and internals for structured reactors is more active pharmaceutical ingredients (APIs) are structurally com- recent and still limited to few applications [15, 16]. In indus- plex and equipment flexibility is required to produce the target try, packed bed reactors filled with heterogenous catalyst compounds, thereby batch reactors have traditionally been the powder are still preferred due to the facile utilization and the equipment of choice [3]. Nevertheless, several pharmaceutical high surface area available for mass transfer. However, they companies have expressed increasing interest in switching to a pose some operational difficulties, such as high pressure drops continuous production mode, due to the numerous advan- and flow maldistributions, with consequent channeling and tages, such as constant product quality and easier optimization decrease in the overall process performance [8]. Therefore, of energy and resources [4, 5]. However the low flexibility of structured catalysts and packing materials have been devel- continuous equipment still impedes the implementation of oped to overcome these issues. Characterized by a flexible continuous processes into the pharmaceutical sector [5]. 3D design, they can be implemented into commercially available printing of customized reaction ware and equipment might be columns and reactors. Furthermore, they can be easily the solution to this issue. In fact, this manufacturing technique interchanged, which makes them particularly interesting as allows to implement non-conventional and complex reactor catalyst support. The catalyst can be immobilized on the solid geometries, whose parameters can be quickly optimized based matrix as a washcoat or incorporated in the chemical structure on experimental data by iterative design modifications [6]. of the support material [1, 8]. Moreover, by using additive Also, 3D printing in chemical synthesis is not limited to reac- manufacturing technologies, it is possible to design and engi- tor design, it is very often used to produce customized pieces neer the support structures to perfectly meet the reaction re- of equipment, e.g. common laboratory hardware or sensor quirements, such as fast mixing, improved heat transfer and analytic applications [1, 7]. Different additive manufacturing fluid distribution (e.g. reduced channeling) compared to ran- techniques are available, depending on the specific purpose, domly and particle-based packed beds [17]. In addition, rapid but the main working principle is similar: the object, designed optimization of the support structure is straightforward and with computer aided design (CAD) software is sliced in a can be achieved by iteratively improving the design based slicing software in a number of cross sections. The sliced on the collected experimental data. Finally, catalytic supports object is then uploaded to a printer, which shapes the object are much easier to handle and to fill into a reactor compared to by selectively stacking these layers above one another [2]. heterogeneous catalyst powders. How these layers are built depends on the printing principle: In this work, structured ceramic supports printed via vat in this work, we utilized vat photopolymerization (VPP) and photopolymerization and their use for catalytic applications material extrusion (ME). VPP, also known as are presented. In order to describe the microfluidic behavior stereolithography (SLA), generates a 3D object by selective inside the supports, low cost flow cells were designed to easily layer-by-layer solidification of a liquid photopolymer resin perform residence time distribution (RTD) experiments. The using a UV light source. VPP is renowned for its high resolu- RTD setup includes two 3D printed flow cells, common laser tion of the printed parts and for the high quality of surface beams as light source, and an Arduino microcontroller to finish [8, 9]. Moreover, this technique is widely used for print- monitor the change in absorbance when a dye is injected into ing materials with excellent mechanical properties, such as a microfluidic device. The setup was designed to be an effi- ceramics, by suspending solid particles within the resin mix- cient and low cost alternative to flow photometric equipment tures. ME instead is an extrusion-based technique, in which a available on the market and reflects the power of 3D printing thermoplastic polymer is melted and extruded through a hot to design customized analytical tools without having to pur- nozzle. The polymer is then deposited onto a building plate chase expensive and patented equipment [18, 19]. By com- through the nozzle in a layer-by-layer fashion, thereby creat- bining 3D printed parts with low-cost microcontrollers and ing an object by stacking different slices on top of each other electronics, it is possible to extend the scope of in-house de- [8]. After the layer deposition, the polymer hardens on signed laboratory equipment to more complex applications, cooling, which might cause adjacent layers to not bind prop- such as inline and real-time monitoring of defined process erly. In general, this results in low mechanical stability of the parameters [19–21] For rapid prototyping purposes, Arduino printed parts along the z-direction [10] and possible fluid microcontrollers are the most known and commonly used, due J Flow Chem (2021) 11:675–689 677 to the many advantages they offer. Firstly, they are cheap and commonly used in additive manufacturing [23, 24, 27]. This can be coupled with a wide range of sensors and devices. has opened up the possibility to 3D print bioreactors or car- Secondly, the software and the board are user friendly, which riers made from commercially available resins, allowing direct makes it easier for scientists with little programming back- immobilization of the enzymes onto the structures’ surface ground to create their own prototype. Finally, Arduino is an and enabling high performance and recyclability [28–30]. open source project with a wide community of users, which Regarding inorganic materials, enzymes have been very often increases the possibility of sharing ideas and prototypes covalently attached onto silica or ceramic particles and/or among labs and researchers [20]. monoliths [30–33]. However, to the best of our knowledge, As an application-oriented proof of concept, in this work this is the first report of enzyme immobilization onto 3D the model enzyme Phenolic Acid Decarboxylase from printed ceramic supports. Bacillus Subtilis (bsPAD) was covalently immobilized onto In this work, we examined the continuous transformation 3D printed solid supports and tested for the decarboxylation of of para-coumaric acid into vinylphenol catalyzed by bsPAD coumaric acid in a continuous flow setup. The system was (Scheme 1) covalently immobilized onto our 3D printed in- chosen due to the considerable industrial and economical in- serts. Since this enzymatic reaction had already been studied terest in biocatalysis and the ongoing efforts to increase the previously by our group using an encapsulated biocatalyst in feasibility of enzymes by finding innovative and optimal sup- continuous flow [34], in this work we decided to utilize a ports for their immobilization [22]. In fact, biocatalysis has ceramic support material and focus on determining the influ- emerged as a green and highly efficient alternative to metal- ence of different internal geometries and process parameters based catalysis for the manufacturing of APIs and fine on the reaction outcome. Ceramics was preferred to standard chemicals [23]. Enzymes are natural and environmentally 3D printing resins as a support material since it is chemically friendly catalysts, showing high activity in mild conditions inert and does not pose any risks of inactivation to the enzyme (low temperatures and water-based solvents) as well as high in use. In the case of bsPAD, the choice of using ceramics substrate specificity. Moreover, due to past advances in pro- resulted in a great compatibility of the enzyme to the carrier tein expression and purification, the price of enzymes has material, as demonstrated by the long-term activity and stabil- decreased strongly over the last years, making enzymes more ity achieved. Moreover, ceramics is less brittle than standard economically feasible even for large scale processes [24]. resins [1], which makes it easier for an insert to be tightly However, the limited long term stability and reusability of packed in a column without being crushed. Alumina is also enzymes make it very challenging for this technology to be cheap and widely available, therefore it is possible to easily competitive on an industrial level [22, 23]. Hence, enzyme produce objects with great mechanical and chemical proper- immobilization has been proposed as a very efficient solution ties at an affordable cost. to overcome these issues [22]. Various methods for immobi- In order to decrease the time needed for screening the dif- lization are available and differ in the mechanism of protein ferent selected parameters and increase the level of process attachment, such as affinity bonding, physical adsorption, co- understanding, an automated process setup was realized, as valent bonding and encapsulation. Each of these methods has shown in Scheme 3. The possibility of process automation is its advantage and drawbacks, and the choice greatly depends one of the many advantages of continuous manufacturing, as on the specific application [23–26]. Covalent binding is com- it allows for rapid and controlled screening of process param- monly used when enzyme immobilization onto a solid support eters [35–37]. On lab scale, automated reaction platforms are material is targeted, as the support surface can be easily mod- enabled by using standard hardware-connectivity to remotely ified and functional groups for enzyme attachment are facile to control the equipment, and online analytics to monitor the introduce. This technique allows to preserve the enzymatic process parameters in real time [36, 38]. In combination with activity for a longer period of time, reuse of the biocatalyst computer algorithms, it is possible to perform experiments as well as easy enzyme separation from the reaction mixture automatically by controlling the equipment, data acquisition, [26]. Several organic and inorganic materials have been used and performing process optimization based on the recorded as matrix for covalent enzyme attachment including polymers data [36, 38]. Multidimensional systematic optimization Scheme 1 Reaction scheme for the enzymatic decarboxylation of coumaric acid to vinylphenol catalyzed by bsPAD 678 J Flow Chem (2021) 11:675–689 strategies, such as Design of Experiments (DoE), can be easily distribution along the channels, depending on the individual implemented in such automated platforms and are generally channel backpressure, whereby no mixing can take place be- preferred to the one-variable-at-a-time (OVAT) approach tween them. The purpose of this design was to approach a [39–41]. In the DoE approach, process parameters are first plug flow behavior while reducing the extent of backmixing. ‘screened’ to identify the critical factors influencing the reac- However the individual backpressure of the channels can tion outcome (e.g. for a chemical reaction: yield, purity, cost, cause the fluid to not distribute evenly among the channels etc.). Then, an ‘optimization’ step is carried out to determine and might affect a structure’s performance. The second struc- the best settings for the individual variables [40]. DoE also ture features a cubic lattice (CL) design that is repeated along reveals more information on the interactions between the pro- the length of the support (Fig. 1 right). In this way, radial cess parameters and their effect on the outcome of the reac- mixing is introduced, while increasing the available surface tion, and allows to save time and materials by minimizing the area. Both designs are equipped with an outer shell and an O- total number of experiments needed [39]. ring on both sides to fix the support tightly when inserted into a column, thereby preventing bypassing of the fluid. The outer In this work, we have applied a fractional factorial Central diameter of all supports was set to 7.8 mm at the O-ring sec- Composite Design (CCD) [42, 43] for the DoE study to in- tion and kept at this diameter for the lattice structure. The O- vestigate several process parameters and gather as much in- ring prevented adding channels close to the outer radius of the formation as possible on the bsPAD catalyzed reaction using honeycomb as it would have blocked them. As a consequence, our structured inserts. CCD was also used to define the opti- the unused mass in the middle of this design needed to be mal carrier and combination of process parameters. The frac- reduced as it was likely to produce cracks during post-process- tional factorial CCD approach was preferred over the full fac- ing. A final diameter of 5.8 mm was chosen for the middle torial approach and similar approaches such as the Box- section. Both designs feature a length of 39.8 mm in order to Benkhen design due to the lower number of experiments fit perfectly into commercially available empty HPLC col- needed per iteration [35, 44]. umns (ID 8 mm, height 40 mm). All designs were replicated Overall, the goals of this work are to show the power of 3D and printed with 3 different channel sizes. The geometrical printing for designing flexible and easily interchangeable characteristics are summarized in Table 1. structured inserts, as well as for realizing powerful, yet low The structures were printed using a lithography-based ce- cost inline analytical tools to increase process understanding. ramic manufacturing technique. The printer used was a To achieve these goals, two different types of structured in- CeraFab 7500 and the printing setup is shown in the serts were designed, based on two different internal geome- Supporting Information (SI, Figure S3). It features a transpar- tries. To determine the flow pattern inside the inserts, RTD ent vat into which the slurry, made of a mixture of monomer, photoinitiator and alumina, is automatically dispersed and experiments enabled by our in-house designed and printed inline photometric flow cells were used. Then the decarbox- spread. The movable building platform is immersed into the ylation of coumaric acid catalyzed by bsPAD was chosen as a slurry, which is then selectively exposed to visible light from model system to prove that it is possible to use these below the vat. The layer image is created via a digital micro- engineered inserts as structured bed reactors in a continuous mirror device (DMD) coupled with an advanced projection biocatalytic application, with great results in terms of produc- system. By repeating this process, a three-dimensional green tivity and process stability. By automating this setup, it was part can be generated layer-by-layer. A support structure was also feasible to quickly identify the optimum operating condi- used for each column to avoid over polymerization in the tions for the decarboxylation reaction, thereby exploiting the channels. More information on the building parameters can advantages of flow chemistry to the fullest. be found in the SI. After printing, the parts were cleaned with pressurized air and a cleaning solvent. Then, thermal postprocessing was carried out by placing all parts in a furnace Results and discussion and first applying a preconditioning cycle at 120 °C, followed by debinding and sintering at 1500 °C for 2 h. Design and fabrication of the 3D printed inserts Characterization of the inserts– determination of the The supports used for immobilization of the enzyme were mixing behavior/flow pattern designed using Autodesk®’s CAD software Inventor® and printed at Lithoz GmbH. Two different internal structure types In order to assess the flow pattern inside of the columns were realized, by following a similar approach from literature packed with designed inserts, experiments were carried out [17]. The first one consists of a honeycomb-like structure to determine the residence time distribution (RTD) using the (HC) comprising internal straight channels of hexagonal cross step input principle, as reported in literature [45]. The detailed sections (Fig. 1 left). This design was chosen to achieve fluid procedure of the measurements is provided in the SI, and the J Flow Chem (2021) 11:675–689 679 Fig. 1 CAD drawings of the designed and 3D-printed ceramic inserts. Straight honeycomb like structures (HC, left) and cubic lattice (CL, right). From top to bottom, the hydraulic diameter of a respective structure decreases, while the internal surface to vol- ume ratio increases setup is presented in Scheme 2. Two syringe pumps were all experiments and the step itself is not ideal. Another issue used, one filled with ethanol and one with a mixture of lies in the feeding tube or capillary used to deliver the tracer ethanol and methylene blue as tracer. Both were connected to the HPLC column. If the axial dispersion in this capillary to a six-way-valve, controlling the flow through the struc- is high, this might influence the RTD, thereby falsifying the tured insert, which was incorporated in a HPLC column. results. This effect is especially relevant for low Reynolds Two flow cells were implemented in the system, one at the number ranges (as it is the case for microfluidic applica- inlet and one at the outlet of the column. With this setup any tions). One way often proposed in literature to compensate sources of disturbances and deviations occurring not only for deviations in the system is to record the tracer signal at inside of the column, but also at its inlet are taken into the inlet and at the outlet of the device. The results are then account. For example, using a manually switched valve to analyzed by using the approach of the convolution integral inject the tracer can lead to disturbances in the evaluation of theorem, to take into account deviations to the RTD being the step signal, as the time for injection is not univocal for caused by auxiliary equipment [45, 46]. Table 1 Summary of the geometrical characteristics of the designed inserts 3 2 2 Type of insert Number Hydraulic diameter [mm] Internal volume [mm ] Internal area [mm ] Area/Volume [mm /mm³] 1 0.757 351.71 1857.31 5.3 Honeycomb (HC) 2 0.521 314.86 2416.17 7.7 3 0.263 234.65 3572.21 15.2 1 1.564 1044.78 2671.38 2.6 Cubic lattice (CL) 2 0.930 842.98 3627.25 4.3 3 0.433 467.12 4319.53 9.2 The internal volume and area were calculated in Autodesk Inventor 680 J Flow Chem (2021) 11:675–689 Scheme 2 Setup for the evaluation of the Residence Time Distribution (RTD) and thus the flow patterns inside the designed structured inserts For the experiments, 3D printed flow cells mountable on step down was induced by switching the six-way-valve back the outside of transparent 1/16” capillaries were designed (see to the load position. Different flowrates (2, 1, 0.5, 0.2, 0.1, SI Figure S4). Their measurement principle is based on a light 0.05 mL/min) were tested in order to evaluate how the flow intensity measurement passed through a capillary, whereby pattern changes with varying Reynolds number. For each ex- each flow cell represents a photometer on its own. The emitted periment, the results were exported in a .log file and imported light of a red LED is absorbed within a capillary and its fluidic in Microsoft Excel to evaluate the cumulative and exit age content and finally detected on the other side by a photo re- distributions. In order to quantify the extent of axial dispersion sistor. The photo resistor changes its conductivity depending and backmixing into the devices, the Bodenstein number (Bo) on the absorbed light and allows measuring of light intensity was determined for each insert at every investigated flowrate. depending voltages. This voltage is recognized by an analog This dimensionless number is defined as the ratio of the con- to digital converter (ADS1115, 16 bit). To account for differ- vective mass transfer over axial dispersion: ent mounting and manufacturing deviations of the electronic u  L char parts, an adjustable voltage divider was added to tune offset Bo ¼ ax values. Buttons for LED control and reference points are where u is the flow velocity, L is the characteristic length of added as well, to take the needed light and dark reference char values. LEDs and photo resistors are powered by a micro the device andD is the axial dispersion coefficient. Bo can be ax used to estimate how close the fluid behavior inside of a reac- controller (Arduino Nano), which reads the measured voltages of the ADS1115 and provides measured data to a PC via a tor is to an ideal reactor model: for Bo > 100, the fluid pattern serial communication. Beside the LED and photo resistors, all inside the investigated reactor approaches that of a plug flow other electronic parts were soldered on a printed circuit board reactor (PFR), while for Bo < 100, the behavior approaches (PCB) to increase measurement stability. With the developed that of a CSTR [45]. Different methods have been reported set-up, a low cost possibility to measure RTDs is presented. to calculate this number from the RTD curves, based on dif- The experimental procedure started by flushing the column ferent assumptions on the boundary conditions of the system. with pure solvent and saving a light and dark reference spec- In this work, the open-open model was chosen, as it assumes trum. Then, the tracer solution was injected into the column that the flow is dispersed along the length of the whole column and it also provides an analytical solution to calculate Bo after defined time intervals by manually switching the six- way-valve to the inject position. The change in absorbance solely from the variance of the RTD curve [47]. Moreover, since it was possible to place two flow cells at both the inlet was recorded at both the inlet and the outlet of the column and converted into a digital signal via an Arduino Nano and and outlet of the columns, the calculated variance and mean logged into a serial terminal program (Tera Term) on a PC. residence time for each insert could be corrected by taking into When the absorbance of the tracer reached a stable value, a account the influence of the inlet capillary on the output J Flow Chem (2021) 11:675–689 681 response. For this purpose, the additivity approach described pulsating and irregular behavior in this range. Another possi- by Levenspiel [45] was used. bility is that at lower flowrates flow maldistributions might The results for the RTD experiments are summarized in occur, because the connector that ties the capillary to the Fig. 2, by plotting the calculated Bo versus the Reynolds num- HPLC column has a small inlet in the middle, therefore the ber (Re) for each measured point. Re was calculated by iden- fluid might not distribute to the outer channels and flow main- tifying a hydraulic diameter d (as reported in Table 1)for ly through the middle ones. Maldistributions might be more each structured insert, as it allows for the comparison among relevant for HC3 and HC2 due to the lower channel diameter different internal structures. The d was calculated as follows: and the higher number of channels, resulting in the fluid to not be distributed equally among all channels and therefore de- 4  A creasing the mixing efficiency. Since HC1 has bigger chan- d ¼ nels, it is easier for the fluid to be pumped uniformly in all channels, even for low Re. where A is the cross sectional area of the device and P is the The trend instead changes for the square lattice structured wetted perimeter, which were both estimated with the aid of inserts CL 1–3. As visible in Fig. 2b, no trend was recorded Autodesk Inventor®. The Re was then calculated as : for Bo with increasing flowrate, instead it changed around an ρ  u  d average value. A slight increase in Bo could only be detected Re ¼ for CL3, which was expected since it has the smallest hydrau- lic diameter. In fact, the higher density of lattice elements Where ρ and μ are respectively the density and the dynamic present inside CL3 promotes the formation of secondary flow viscosity of the fluid flowing through the channel. structures, thereby increasing the extent of convective mixing. The results show that for all the designed inserts in the For CL1 and CL2 there was almost no difference in Bo for investigated flowrate region, Bo was significantly below Re > 5. However, below that, CL1 seemed to have a similar 100, indicating high backmixing and axial dispersion inside behavior to CL3. As for the HC inserts, this inversion in the the inserts. Considering the results for the honeycomb (HC) trend could be arising from fluctuations in the pumps, or it structured inserts (Fig. 2a), it can be seen that the average Bo could be due to flow maldistributions inside the CL3 insert. was the highest for HC3 and it was rising with the flowrate. In general, for both types of inserts it can be expected that Since the superficial velocity inside a channel is increasing using higher flowrates improves mixing and promotes PFR with decreasing inner diameter, therefore the Re is higher for behavior, however since higher flowrates were not interesting HC3 and the flow is more chaotic, resulting in a lower extent for the chosen application (as they result in too low residence of backmixing. Axial dispersion is instead higher in the case time for the selected reaction), the investigated range was not of HC2 and HC1: in both inserts the change in Bo with the enlarged any further. flowrate was very limited, indicating that the insert was ap- proaching CSTR behavior in the whole investigated region. Also, the values of Bo seemed to increase at lower Re for these Decarboxylation reaction – choice of the solvent inserts, especially for HC1, for which Bo was higher than the values achieved by HC3 at Re below 7. This trend can be Phenolic acid decarboxylase (PAD) is an enzyme belonging explained by the fact that syringe pumps have a more to the family of cofactor-free decarboxylase. It catalyzes the ab Fig. 2 Results of the RTD experiments for the honeycomb (a) and the cubic lattice (b)structured inserts 682 J Flow Chem (2021) 11:675–689 removal of carbon dioxide from hydroxycinnamic acids, such surface modification steps until silanization are the same, but as coumaric acid, to give hydroxystyrene products that are instead of having a surface functionalization step, the enzyme important API precursors [48]. Even though PAD is very ac- was directly immobilized onto the support by introducing the tive towards this reaction, a limit to the industrial application is less harmful linkers N-hydroxysuccinimide (NHS) and N-(3- the low solubility of phenolic acids in water-based solvents. dimethylaminopropyl)-N′-ethylcarbodiimide hydrochloride Therefore, alternative solvent systems have been proposed, in (EDC) into the enzyme solution. These compounds activate order to maintain the high activity of PAD and increasing the the available carboxylic groups on the amino acids, which un- space-time -yield (STY) of the decarboxylation [48]. Deep dergo bond formation with the amine groups on the carriers Eutectic Solvents (DES) have been suggested as an alternative [56]. Preliminary batch tests showed that the latter method is to ionic liquids, due to the lower toxicity and the greater up- preferred in terms of immobilization yield and long-term stabil- take of the CO released during the decarboxylation [34]. ity of the enzyme activity (see SI). Therefore, this method was They consist of a mixture of primarily quaternary ammonium chosen to immobilize bsPAD on the 3D printed inserts for the salts (e.g. choline chloride) with hydrogen-bond-donors purpose of performing the flow experiments. (HBD) (such as glycerol). The heating and mixing of this The immobilization of bsPAD on the structured inserts was solution disrupt the crystalline structure of the ammonium salt, carried out in flow, as shown in the SI. This allowed to avoid generating a viscous liquid with a lower melting point than the material handling in between the different immobilization starting materials [34, 49]. steps and reduced the overall time needed for the procedure. The DES mixture of choline chloride and glycerol in a 1:2 In order to estimate the amount of enzyme immobilized on the molar ratio, diluted 1:1 (v/v) with phosphate buffer at pH 6 supports, the immobilization yield (Y ) was determined via was identified in previous works [34, 48] as an efficient sol- taking 1 mL samples from the enzyme solution before and vent for the decarboxylation, both in terms of CO absorption after immobilization and performing the activity assays in and high substrate solubility, since the properties of the DES batch as described in the SI. By determining the activity of are still relevant at this dilution ratio. Therefore, this solvent the solution before and after immobilization, it was possible to mixture was kept also in this work to investigate the 3D calculate Y as: printed inserts as catalytic supports for the bsPAD enzyme. A  A b a The decarboxylation of coumaric acid to vinylphenol was Y ¼  100 performed first in batch for the preliminary immobilization tests (see the SI) in order to identify an optimum immobiliza- Where A is the activity before and A the activity after b a tion procedure. Finally, automated flow experiments were car- immobilization. The activity of bsPAD was defined as the ried out following the procedure explained in the dedicated amount of μmoles of coumaric acid being consumed per min- paragraph. ute, taking the first 30 min of reaction for the calculation, as in this region the conversion of coumaric acid was linear. The Enzyme immobilization calculation was carried out as follows: Covalent binding was chosen as an immobilization strategy, C  C  V CA;0 CA;30 assay U ¼ since it has been widely used as a method for protein attachment on ceramic supports [50–53]. Two different methods (1 and 2) Where C and C are the concentration of coumaric CA;0 CA;30 were selected and first tested in batch by using the particles acid at the start and after 30 min into the reaction (given in obtained from grinding the ceramic objects that failed during μmol/L, determined by HPLC),V is the total reaction assay printing. All steps for both methods are described in detail in the volume and t is 30 min. The activity is therefore is given in SI. Method 1 was taken and adapted with slight modifications units U [μmol/min]. from literature [50]. It involved a first step of surface activation The theoretical activity of the enzyme immobilized on by acid treatment, followed by a silanization step with 3- each insert was calculated by assuming that this is equal aminopropyltriethoxysilane (APTES) to introduce amino to the difference between the activity in the starting im- groups onto the surface, which are functionalized with glutar- mobilization solution and the one measured at the end of aldehyde. This compound comprises two reactive aldehyde the immobilization. This value was only used as an esti- groups and it is used to link two amine functionalities, one on mation, mainly as it was not possible to access the small the surface of the support and one on the enzyme itself, by internal channels in order to determine the exact concen- forming amide bonds [54]. This method has the advantage that tration of the bound enzyme via standard assays (e.g. pro- it ensures stable protein attachment. However, glutaraldehyde is tein labelling with fluorescent molecules [57]). Also it toxic and might even deactivate the enzyme [55]. Therefore, a does not take into account whether the enzyme undergoes different immobilization procedure (method 2) was chosen, al- conformational changes upon covalent binding to the so slightly modified from literature [52, 53]. In this case, initial J Flow Chem (2021) 11:675–689 683 Table 2 Overview of the measure BET surface area and the achieved immobilization of bsPAD onto the structured inserts Type of insert Number BET surface area [m /g] Weight of insert [g] Immobilization yield [%] Estimated immobilized activity [U/g] Honeycomb (HC) 1 0.254 2.32 19 1.24 2 0.310 2.5 3 0.543 2.86 58 6.02 Cubic lattice (CL) 1 0.484 1.98 14 1.09 2 0.522 2.52 3 0.536 3.77 50 2.62 carrier, which might affect its structural properties com- from the BET results). The calculated values for Y and the pared to the free enzyme [23, 25]. estimated activity showed that a higher amount of enzyme The exact surface area of each uncoated insert was deter- bounded to HC3 and CL3, most probably due to the higher mined via Brunauer–Emmett–Teller (BET) measurement, as surface area available for the immobilization. described in the SI. The results are summarized in Table 2. Only insert types 1 and 3 were tested in the continuous setup, Automated flow setup as the inserts with the middle channel size did not show to be interesting both in terms of their fluidic behavior (as shown After being coated with bsPAD, the inserts are ready to be from the RTD results) nor available surface area (as shown used in the continuous flow setup shown in Scheme 3.The Scheme 3 Setup for the automated screening and optimization of the decarboxylation of coumaric acid catalyzed by bsPAD immobilized onto 3D printed inserts. DES = Deep eutectic solvent (ChCl/Gly 1:2 mol) 684 J Flow Chem (2021) 11:675–689 scope of this setup was to allow the automated evaluation of having performed all experimental runs on the first col- the different coated carriers and the effect that changing tem- umn, the algorithm prompted the valve to switch to the perature, flowrate and substrate concentration have on the second column, and the same optimization algorithm was decarboxylation reaction, according to a DoE approach. repeated. Once both inserts had been evaluated, the col- The structured inserts were again enclosed into HPLC col- umns were removed from the thermostat and the inserts umns and constituted the structured bed reactors at the heart of were exchanged manually to the other design type. Then, the process. The setup featured one HPLC pump (Knauer the columns were reinstalled in the setup and the DoE Azura P 4.1 S, pump B in Scheme 3) delivering solvent, and evaluation was started again by following the same proce- a syringe pump (Lambda VIT-FIT, pump A in Scheme 3) dure as before. equipped with a 50 mL stainless steel syringe filled with a 10 mM stock solution of coumaric acid. The outlets of the Design of experiments (DoE) pumps were then connected to a static mixer, in order to pro- vide a feed stream with uniform concentration before it Fractional CCD was chosen for the DoE study due to its effi- reaches the column inlet. As static mixer, a 3D printed stain- ciency for screening a high number of reaction parameters less steel microfluidic device was used, which was designed in using a low number of experiments. Regarding the particular our group and characterized in a previous work [58]. This enzymatic reaction studied in this work, it was highly desired chosen device was the so-called AP04 with an internal diam- to reduce the amount of experiments to not waste solvent or eter of 0.6 mm, into which the solvent and the substrate solu- reagent, thereby reducing the amount of waste and containing tion were mixed according to a split-and-recombine principle. the costs. Three factors were chosen for optimization: feed The AP04 was chosen as static mixer since it was proved in a flowrate, reaction temperature and substrate concentration. previous publication [58] that it provides reasonably good These values were evaluated within a limited design space degrees of mixing while remaining a compact device. The reported in Table 3. The maximization of the yield and feed was then pumped into one of the columns, each connect- space-time-yield (STY) of vinylphenol were chosen as the ed to a six-way-valve (Knauer Azura VU 4.1), which dictates target response of the DoE. The STY was calculated accord- the column into which the feed is pumped to. Two inserts of ing to the following equation [59], assuming that the enzyme the same type (e.g. HC1 and HC3, or CL1 and CL3) were is forming a monolayer on the inner surface of the insert: evaluated at a time, by being fitted in HPLC columns and then c  FR prod connected to ports 1 and 2 of the six-way-valve. The columns STY ¼ were immersed in a temperature-controlled water bath to keep i the temperature at the desired value. The pumps, six-way- where c is the concentration of product formed (in g prod valve and thermostat were connected to a PC and controlled vinylphenol/L) at steady state at the chosen experimental con- via a Python-based script. In this way, the automated control ditions, FR is the flowrate (in L/h) and V is the internal and change of temperature, flow rate and substrate concentra- volume of the reactor (in L). tion for each experiment were facilitated. The upper and lower limits for the flow rate were chosen Before starting the automated experimental sequence, considering a sufficient residence time within the column as the whole setup including the columns was flushed for at well as concerning the structural limits of the UV-vis flow cell least 1 h in order to remove loosely bound enzyme and take (it can only withstand a pressure up to 10 bar). For the tem- a blank reference for the UV-vis measurement. Then perature, the limits were set around the optimal temperature starting with column one, the DoE algorithm performed defined in a previous work [34]. In terms of the substrate different experimental runs in order to screen the reaction concentration, the low solubility of coumaric acid in the reac- space as determined by the fractional CCD method, which tion solvent and the flow cell path length were the limiting is described in detail in the next section. During each run, factors. The starting substrate concentration for each the reaction was monitored via an Avantes Micro Flow Cell with 1.5 mm path length connected to a spectrometer (AvaLight-DS-DUV equipped with a deuterium lamp and Table 3 Boundaries of the design space investigated in the DoE AvaSpec-ULS2048 detector), allowing the real-time track- experiments ing of the rate of vinylphenol formation. The product was Parameter Lower limit Upper limit identified by a characteristic peak at the wavelengths be- tween 269 and 263 nm, whereas the reagent was measured Flowrate 0.2 mL/min 2 mL/min at wavelengths between 325 and 328 nm, where Temperature 25 °C 35 °C vinylphenol does not absorb. From this data and previously Substrate concentration 0.25 mM 2 mM determined calibration curves, the conversion of coumaric Dilution ratio 4 40 acid and the yield of vinylphenol were calculated. After J Flow Chem (2021) 11:675–689 685 experiment was set in the platform by changing the flowrates this work. The new experimental points lie at the corners of of the solvent and the substrate pumps in order to achieve the the cube, and are therefore equally distant from the center. desired dilution ratio. Therefore, dilution ratio was given as an Since a fractional factorial design was applied, only 4 of the input parameter to the DoE algorithm. corner points were added to the total experiments, resulting A scheme summarizing the steps for the chosen experi- in a final amount of 11 experimental runs (Fig. 3b). Then, mental design is shown in Fig. 3. For each 3D printed struc- each experiment is evaluated singularly and after all runs are tured insert, two DoE runs were performed. In the first run, a completed, the best point of the first iteration is chosen in subsection of 70 % of the total design space was screened to terms of highest STY (Fig. 3c). In the second run, the reac- identify an optimum combination of parameters (each rep- tion is further optimized by investigating the experimental resented as a point) and narrow down the experimental space near the best point found in the first iteration, which space. Based on the boundaries given by the user, the algo- was taken as center point (Fig. 3d). In this phase, the geo- rithm defined the center point and the high and low levels metrical size of the CCD was set to 40 % of the first itera- for each factor, which were connected by a cross on each tion, therefore ensuring that no point would lie outside the plane in the 3D design space (Fig. 3a) and constituted the initial boundaries of the design space. The overall amount of axial experimental points, resulting in 7 experiments. Then, experiments needed is 21, of which 11 were carried out in further points were added to the starting ones, and were the first run and 10 in the second. The combined size of the chosen by the algorithm based on a spherical composite two iterations ensured, that 98 % of the reaction space can design approach [43]. According to this approach, a cube be explored via this method. is drawn, which is centered at the center point of the design The DoE was executed via the automated reaction platform and has a side length of α. This distance is defined as the shown in Scheme 3. A Python-based algorithm with an integrat- 1/2 square root of the total amount of factors and was 3 for ed graphic user interface (GUI) was developed to automatically Fig. 3 Graphical representation of the steps for the chosen Fractional from the center point (b). However not all corner points are selected as Factorial CCD approach. For each structured reactor, two DoE runs experimental runs, but only 4, as this is a fractional design. Once all points were performed. First, the algorithm choses the center point of the DoE are defined, the algorithm performs all 11 experiments automatically, and according to the boundaries of the design space given by the user. Then, choses the best point based on the highest STY (c). A new iteration is then the high and low levels for each factor are also identified and combined to started (d) by choosing 10 points around the best point from the first give the axial experimental points (a). Since this is a CCD approach, iteration, which becomes the new center point, following the same further points are added to the experiments, which lie at the corners of a procedure as in the first iteration but by scanning a smaller area of the cube that has a fixed length (identified as α) and are at the same distance design space 686 J Flow Chem (2021) 11:675–689 choose the DoE points, according to the experimental space The results for the optimization experiments are plotted in limits defined by the user, and to set the thermostat and the Fig. 4 andreported inTable 4. At the end of the second itera- pumps at the desired level for each experimental run. The same tion, the best combination of factors for HC1, HC3 and CL1 algorithm saved and accessed the UV-Vis spectra recorded by was 30 °C, 1.1 mL/min and a dilution ratio of 4.36, which the spectrometer in order to calculate the compound concentra- correspondstoastartingconcentrationof1.86mM. Forthese tions, yield and STY of each experiment. These variables were inserts, it was apparent that the optimum temperature lays chosen as response parameters to assess the result of each exper- around 30 °C, which is in line with what was already suggested iment and determine the optimum combination of factors. The in previous works [34, 48]. Moreover, for all of them a combi- final output parameter that was to be maximized via the DoE was nation of a medium flowrate and lowest dilution ratio was pre- STY, as this parameter can take into account not only the yield of ferred, as this gives the highest productivity, thereby optimizing the reaction but also the influence of flowrate and free volume in the STY. The picture is instead different for CL3; for this insert the column. Once all of the experiments were performed on one the best point was 28 °C, 1.464 ml/min and dilution ratio of column, the algorithm prompted the six-way-valve to switch the 9.676 which corresponds to a starting concentration of 0.93 feed to the second column, onto which the DoE approach was mM. So in this case the algorithm went down another path repeated as before. After having evaluated both inserts, the algo- for optimization, but still preferred a compromise between rithm returns the results of the DoE, and chooses the best insert flowrate and dilution ratio to increase productivity. The reason and set of process parameters. The columns could then be man- for which in this case a different optimum was found is clear ually removed from the setup and the inserts exchanged to the when looking at Fig. 4 (CL3): in the first iteration, the points at other design type for further experiments, which were carried out 28 °C,1.464 mL/min and 0.63 mM (dilution ratio of 14.71) and as explained above. at 30 °C, 1.1 mL/min and 0.96 mM (dilution ratio of 9.4) Fig. 4 Graphical summary of the DoE results for the 4 structured inserts insert depending on the internal volume. The best point in the first investigated. The cross points for the first iteration are connected via blue iteration for HC1, HC3 and CL1 was 30 °C,1.1 mL/min and a dilution dashed lines, while for the second iteration they are connected in red. The of 9.4. For CL3 instead it was 28 °C, 1.464 mL/min and 14.71 dilution. color map to the right of each graph gives an indication of the level of For the second iteration, the best combination of factors was 30 °C, 1.1 STY reached for each experimental point. The results are plotted in terms mL/min and a dilution ratio of 4.36 for HC1, HC3 and CL1. For CL3 the of Residence time rather than flowrate, as the first differs among each best point was 28 °C, 1.464 mL/min and ad dilution ratio of 9.676 J Flow Chem (2021) 11:675–689 687 Table 4 Summary of the DoE First iteration Second iteration results for each structured insert used in this work Type of insert Yield [%] STY [g/L·h] Yield [%] STY [g/L·h] Residence time [s] HC1 62.6 13.59 64.7 27.27 19.15 HC3 70.9 23.11 62.2 39.21 12.8 CL1 59.2 4.32 67.3 9.53 56.95 CL3 77.6 11.16 61.9 13.11 19.14 resulted in a very close STY, but since the algorithm goes for pieces of equipment for many applications in flow. First, the best point and does not explore the space around the second novel 3D printed ceramic structured inserts have been de- best, the second iteration went in a different direction. A more signed and 3D printed via VPP. In order to characterize the complex optimization algorithm could identify and evaluate fluidic behavior inside the inserts, RTD measurements also different local optima, however for the purpose of this were carried out with the aid of self-made 3D printed pho- work the chosen DoE approach was enough to identify a trend tometric flow cells. These tools were used to determine the to optimize the overall productivity. exit age distribution for each insert via recording the step The STY achieved at steady state for the best combination of input of a tracer at different operating conditions. By re- parameters in each insert is summarizedinTable 4. For all the cording the tracer input at both the inlet and the outlet of a investigated inserts, the final STY was higher than the value of reactor, it was possible to remove any disturbances caused 4.8 g/L·h ,which was achieved in a previous work [34]. As by the feeding capillary, thereby increasing the quality of showninTable 4, HC3 gave the best results in terms of STY: the results and proving to be a valid low-cost alternative to at the optimum conditions, a value of 39.21 g/L·h was reached, expensive inline analytical equipment. The evaluation of which represents a 8-fold increase compared to what was the RTD results further allowed to conclude that all inserts achieved previously. This result can be explained by the fact approach the fluidic behavior of theoretical CSTR reactors. that HC3 has the lowest internal volume and the highest surface Then, as an application-oriented proof of concept, the in- to volume ratio, thus the highest enzyme coverage. Therefore, serts were enclosed in HPLC columns to be used as solid HC3 was deemed as the best choice to perform the decarbox- supports onto which bsPAD was immobilized for the cat- ylationofcoumaric acidincontinuousflow, as it resultsina alytic conversion of coumaric acid to vinylphenol in con- higher amount of substrate being converted in time and also tinuous flow. To identify the optimal conditions for the offered the best mixing properties compared to all designed selected system, a fractional factorial CCD was chosen as inserts. Moreover, compared to the previous work, the applica- a DoE approach to systematically investigate the effect of tion of an automated flow setup together with a systematic DoE different process parameters (temperature, flowrate and di- strategy proved to be a successful approach to rapidly screen the lution ratio) on the STY. The DoE experiments were suc- effect of different parameters on the outcome of a reaction by cessfully carried out in an automated platform controlled minimizing the need of human intervention and the time needed by a Python-based algorithm, which automatically chose for evaluation. Also, the reaction could be followed in real time the DoE points based on design space given by the user via implementing an online measurement, which is more reli- and run the experiments by remotely controlling the equip- able compared to offline methods and does not require sample ment, reducing the need for human intervention and saving handling/manipulation [60], thus saving time and resources. time and materials. As a result, the HC3 insert proved to be Finally, the possibility of easily designing and manufacturing the best choice for the production of vinylphenol, as it gave interchangeable inserts via means of 3D printing, allows to aSTY of 39.21g/L·h,representingan8-foldincreasecom- rapidly adapt to the reaction needs to achieve optimal process pared to the value of 4.8 g/L·h obtained previously. These conditions. These advantages, coupled with the improved pro- results showed that it is possible to combine different ad- ductivity, represent definitely a step forward towards a more vantages of flow chemistry and 3D printing to biocatalytic industrially feasible enzymatic conversion of coumaric acid to processes, with great results in terms of productivity. In the vinylphenol. near future, the presented approach combining 3D printing and process automation will be applied to further enzymat- ic reaction systems, in order to increase the feasibility of Conclusions biocatalysts for their application in industrial processes. Moreover, the presented inserts will also be implemented In conclusion, we have proved in this work how 3D printing as supports for other active species and tested in different catalytic applications. can be used as a powerful tool to design and produce advanced 688 J Flow Chem (2021) 11:675–689 Supplementary Information The online version contains supplementary ingredients and finished dosage forms: an industry perspective. Org material available at https://doi.org/10.1007/s41981-021-00163-4. Process Res Dev 16:1586–1590 5. Poechlauer P et al (2013) Pharmaceutical roundtable study demon- strates the value of continuous manufacturing in the design of greener processes. Org Process Res Dev 17(12):1472–1478 Abbreviations API, Active Pharmaceutical Ingredient; BET, Brunauer– 6. Kitson PJ, Rosnes MH, Sans V, Dragone V, Cronin L (2012) Emmett–Teller; Bo,Bodenstein number; bsPAD, Phenolic acid decar- Configurable 3D-printed millifluidic and microfluidic ‘lab on a boxylase from Bacillus Subtilis; CAD, Computer Aided Design; CSTR, chip’ reactionware devices. Lab Chip 12(18):3267–3271 Continuous Stirred Tank Reactor; DES, Deep Eutectic Solvent; DoE, 7. Maierhofer M, Maier MC, Gruber-Woelfler H, Mayr T (2021) Design of Experiments; CCD, Central Composite Design; CL, Cubic Inline monitoring of high ammonia concentrations in methanol Lattice; GUI, Graphical User Interface; HC, Honeycomb; ID, internal with a customized 3D printed flow cell. J Flow Chem. https://doi. diameter; ME, material extrusion; OD, outer diameter; PFR, Plug Flow org/10.1007/s41981-021-00141-w Reactor; Re, Reynolds number; RTD, Residence Time Distribution; SLA, 8. Parra-Cabrera C, Achille C, Kuhn S, Ameloot R (2018) 3D printing Stereolithography; STY, Space-Time-Yield; VPP, vat in chemical engineering and catalytic technology: Structured cata- photopolymerization lysts, mixers and reactors. Chem Soc Rev 47(1):209–230. Royal Society of Chemistry Acknowledgements The authors kindly acknowledge the funding by the 9. Michel FM, Rimstidt JD, Kletetschka K (2018) 3D printed mixed CC FLOW project (Austrian Research Promotion Agency FFG No. flow reactor for geochemical rate measurements. Appl 862766), which is funded through the Austrian COMET Program by Geochemistry 89(November 2017):86–91 the Austrian Federal Ministry of Transport, Innovation and Technology 10. Bhattacharjee N, Urrios A, Kang S, Folch A (2016) The upcoming (BMVIT), the Austrian Federal Ministry of Science, Research and 3D-printing revolution in microfluidics. Lab Chip 16(10):1720– Economy (BMWFW) and by the State of Styria (Styrian Funding 1742. Royal Society of Chemistry Agency SFG). The authors also acknowledge the funding by the CATALOX (CATalytic mechanisms and AppLications of 11. Ambrosi A, Pumera M (2016) 3D-printing technologies for elec- OXidoreductases) project, doc.fund program funded by the Austrian trochemical applications. Chem Soc Rev 45(10): 2740– Science Fund (FWF). 2755. Royal Society of Chemistry 12. Au AK, Huynh W, Horowitz LF, Folch A (2016) 3D-Printed Microfluidics. Angew Chemie - Int Ed 55(12):3862–3881 Funding Open access funding provided by Graz University of 13. Parra-Cabrera C, Achille C, Kuhn S, Ameloot R (2018) 3D printing Technology. in chemical engineering and catalytic technology: structured cata- lysts, mixers and reactors. Chem Soc Rev 47(1):209–230 Declarations 14. Capel AJ, Edmondson S, Christie SDR, Goodridge RD, Bibb RJ, Thurstans M (2013) Design and additive manufacture for flow Conflict of interest There are no conflicts of interest to declare. chemistry. Lab Chip 13(23):4583–4590 15. Avril A et al (2017) Continuous flow hydrogenations using novel catalytic static mixers inside a tubular reactor. React Chem Eng Open Access This article is licensed under a Creative Commons 2(2):180–188 Attribution 4.0 International License, which permits use, sharing, 16. Kundra M, Grall T, Ng D, Xie Z, Hornung CH (2021) Continuous adaptation, distribution and reproduction in any medium or format, as flow hydrogenation of flavorings and fragrances using 3D-printed long as you give appropriate credit to the original author(s) and the catalytic static mixers. Ind Eng Chem Res 60(5):1989–2002 source, provide a link to the Creative Commons licence, and indicate if 17. Hock S, Rose M (2020) 3D-structured monoliths of nanoporous changes were made. The images or other third party material in this article polymers by additive manufacturing. Chem Ing Tech 92(5):525– are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the 18. Prikryl J, Foret F (2014) Fluorescence detector for capillary sepa- article's Creative Commons licence and your intended use is not rations fabricated by 3D printing. Anal Chem 86(24):11951–11956 permitted by statutory regulation or exceeds the permitted use, you will 19. Prabhu GRD, Urban PL (2020) Elevating chemistry research with a need to obtain permission directly from the copyright holder. To view a modern electronics toolkit. Chem Rev 120(17):9482– copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. 9553. American Chemical Society 20. Nguyen T, Zoëga Andreasen S, Wolff A, Bang DD (2018) From Lab on a Chip to point of care devices: the role of open source microcontrollers. Micromachines 9(8):403 21. Baden T, Chagas AM, Gage G, Marzullo T, Prieto-Godino LL, References Euler T (2015) Open labware: 3-D printing your own lab equip- ment. PLoS Biol 13(3):1–12 1. Capel AJ, Rimington RP, Lewis MP, Christie SDR (2018) 3D 22. Woodley JM (2019) Accelerating the implementation of biocataly- printing for chemical, pharmaceutical and biological sis in industry. Appl Microbiol Biotechnol 103(12):4733–4739 applications. Nat Rev Chem 2(12):422–436. Springer US, New 23. Zhu Y, Chen Q, Shao L, Jia Y, Zhang X (2020) Microfluidic York immobilized enzyme reactors for continuous biocatalysis. React 2. Dragone V, Sans V, Rosnes MH, Kitson PJ, Cronin L (2013) 3D- Chem Eng 5(1):9–32 printed devices for continuous-flow organic chemistry. Beilstein J 24. Faber K (2018) Biotransformations in Organic Chemistry. Springer Org Chem 9:951–959 International Publishing, Berlin 3. Gutmann B, Cantillo D, Kappe CO (2015) Continuous-flow tech- 25. Hanefeld U, Gardossi L, Magner E (2009) Understanding enzyme nology — a tool for the safe manufacturing of active pharmaceuti- immobilisation. Chem Soc Rev 38(2):453–468 cal ingredients. Angew Chemie - Int Ed 54:6688–6728 26. Guisan JM, Bolivar JM, López-Gallego F, Rocha-Martín J (eds) 4. Poechlauer P, Manley J, Broxterman R, Ridemark M (2012) (2020) Immobilization of enzymes and cells, vol 2100. Springer Continuous processing in the manufacture of active pharmaceutical US, New York J Flow Chem (2021) 11:675–689 689 27. Marques MPC, Fernandes P (2011) Microfluidic devices: Useful 44. Rakić T, Kasagić-Vujanović I, Jovanović M, Jančić-Stojanović B, Ivanović D (2014) Comparison of full factorial design, central com- tools for bioprocess intensification. Molecules 16(10):8368–8401 28. Ye J, Chu T, Chu J, Gao B, He B (2019) A versatile approach for posite design, and box-behnken design in chromatographic method enzyme immobilization using chemically modified 3D-printed development for the determination of fluconazole and its impurities. scaffolds. ACS Sustain Chem Eng 7(21):18048–18054 Anal Lett 47(8):1334–1347 29. Li R, Greenchem / (2017) Tuneable 3D printed bioreactors for 45. Levenspiel O (1999) Chemical reaction engineering, 3rd edn. transaminations under continuous-flow. Green Chem 19(22): Wiley, New York 5259–5516 46. Bošković D, Loebbecke S (2007) Modelling of the residence time 30. De Santis P, Meyer L-E, Kara S (2020) The rise of continuous flow distribution in micromixers. Chem Eng J 135(Suppl. 1):138–146 biocatalysis – fundamentals, very recent developments and future 47. Levenspiel O, Smith WK (1957) Notes on the diffusion-type model perspectives. React Chem Eng 5(12):2155–2184 for the longitudinal mixing of fluids in flow. Chem Eng Sci 6:227– 31. Datta S, Christena LR, Rajaram YRS (2013) Enzyme immobiliza- tion: an overview on techniques and support materials. 3 Biotech 48. Schweiger AK et al (2019) Using deep eutectic solvents to over- 3(1):1–9 come limited substrate solubility in the enzymatic decarboxylation 32. Plagemann R, Jonas L, Kragl U (2011) Ceramic honeycomb as of bio-based phenolic acids. ACS Sustain Chem Eng 7(19):16364– support for covalent immobilization of laccase from Trametes versicolor and transformation of nuclear fast red. Appl Microbiol 49. Müller CR, Meiners I, Domínguez De P, María (2014) Highly Biotechnol 90(1):313–320 enantioselective tandem enzyme-organocatalyst crossed aldol reac- 33. Stroock Abraham SKWAA, Dertinger D, Mezic I, Stone HA, tions with acetaldehyde in deep-eutectic-solvents. RSC Adv 4(86): Whitesides GM (2002) Chaotic mixer for microchannels. Science 46097–46101 295(5555):647–652 50. Thomsen MS, Nidetzky B (2009) Coated-wall microreactor for 34. Grabner B, Schweiger AK, Gavric K, Kourist R, Gruber-Woelfler continuous biocatalytic transformations using immobilized en- H (2020) A chemo-enzymatic tandem reaction in a mixture of deep zymes. Biotechnol J 4(1):98–107 eutectic solvent and water in continuous flow. React Chem Eng 51. de Lathouder KM et al (2008) Polyethyleneimine (PEI) functional- 5(2):263–269 ized ceramic monoliths as enzyme carriers: Preparation and perfor- 35. Fath V, Kockmann N, Otto J, Röder T (2020) Self-optimising pro- mance. J Mol Catal B Enzym 50(1):20–27 cesses and real-time-optimisation of organic syntheses in a 52. Hollermann G, Dhekane R, Kroll S, Rezwan K (2017) microreactor system using Nelder-Mead and design of experiments. Functionalized porous ceramic microbeads as carriers in enzymatic React Chem Eng 5(7):1281–1299 tandem systems. Biochem Eng J 126:30–39 36. Sans V, Cronin L (2016) Towards dial-a-molecule by integrating 53. Halfer T, Rei A, Ciacchi LC, Treccani L, Rezwan K (2014) continuous flow, analytics and self-optimisation. Chem Soc Rev Selective covalent immobilization of ferritin on alumina. 45(8):2032–2043. Royal Society of Chemistry Biointerphases 9(3):1–9 37. Waldron C, Pankajakshan A, Quaglio M, Cao E, Galvanin F, Gavriilidis A (2019) An autonomous microreactor platform for 54. Kim D, Herr AE (2013) Protein immobilization techniques for the rapid identification of kinetic models. React Chem Eng 4(9): microfluidic assays. Biomicrofluidics 7:1–47 1623–1636 55. Ma H et al (2018) Glutaraldehyde inactivation of enveloped DNA 38. Fabry DC, Sugiono E, Rueping M (2016) Online monitoring and viruses in the preparation of haemoglobin-based oxygen carriers. analysis for autonomous continuous flow self-optimizing reactor Artif Cells Nanomedicine Biotechnol 46(1):33–38 systems. React Chem Eng 1(2):129–133. Royal Society of 56. Jia F, Narasimhan B, Mallapragada S (2014) Materials-based strat- Chemistry egies for multi-enzyme immobilization and co-localization: A re- 39. MurrayPM, BellanyF,BenhamouL,Bučar DK, Tabor AB, view. Biotechnol Bioeng 111(2):209–222 Sheppard TD (2016) The application of design of experiments 57. Chen YX, Triola G, Waldmann H (2011) Bioorthogonal chemistry (DoE) reaction optimisation and solvent selection in the develop- for site-specific labeling and surface immobilization of proteins. ment of new synthetic chemistry. Org Biomol Chem 14(8):2373– Acc Chem Res 44(9):762–773 2384. Royal Society of Chemistry 58. Maier MC et al (2020) 3D printed reactors for synthesis of active 40. Gooding OW (2004) Process optimization using combinatorial de- pharmaceutical ingredients in continuous flow. Org Process Res sign principles: Parallel synthesis and design of experiment Dev 24(10):2197–2207 methods. Curr Opin Chem Biol 8(3):297–304. Elsevier Current 59. Bommarius AS, Riebel BR (2004) Biocatalysis. Wiley, Hoboken Trends 60. Kessler W (2006) Prozessanalytik- Strategien und Fallbeispiele aus 41. Weissman SA, Anderson NG (2014) Design of Experiments (DoE) der industriellen Praxis. Wiley- VCH Verlag GmbH & Co. KGaA, and process optimization. A review of recent publications. Org Weinheim Process Res Dev 19(11):1605–1633 42. Box GEP, Wilson KB (1951) On the experimental attainment of Publisher’snote Springer Nature remains neutral with regard to jurisdic- optimum conditions. J R Stat Soc Ser B 13(1):1–38 tional claims in published maps and institutional affiliations. 43. Borkowski JJ (1995) Spherical prediction-variance properties of central composite and Box—Behnken designs. Technometrics 37(4):399–410 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Flow Chemistry Springer Journals

3D printed ceramics as solid supports for enzyme immobilization: an automated DoE approach for applications in continuous flow

Loading next page...
 
/lp/springer-journals/3d-printed-ceramics-as-solid-supports-for-enzyme-immobilization-an-LnxYw2mNaY
Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2021
ISSN
2062-249X
eISSN
2063-0212
DOI
10.1007/s41981-021-00163-4
Publisher site
See Article on Publisher Site

Abstract

In recent years, 3D printing has emerged in the field of chemical engineering as a powerful manufacturing technique to rapidly design and produce tailor-made reaction equipment. In fact, reactors with complex internal geometries can be easily fabricated, optimized and interchanged in order to respond to precise process needs, such as improved mixing and increased surface area. These advantages make them interesting especially for catalytic applications, since customized structured bed reactors can be easily produced. 3D printing applications are not limited to reactor design, it is also possible to realize functional low cost alternatives to analytical equipment that can be used to increase the level of process understanding while keeping the investment costs low. In this work, in-house designed ceramic structured inserts printed via vat photopolymerization (VPP) are presented and characterized. The flow behavior inside these inserts was determined with residence time distribution (RTD) experiments enabled by in-house designed and 3D printed inline photometric flow cells. As a proof of concept, these structured inserts were fitted in an HPLC column to serve as solid inorganic supports for the immobilization of the enzyme Phenolic acid Decarboxylase (bsPAD), which catalyzes the decarboxylation of cinnamic acids. The conversion of coumaric acid to vinylphenol was chosen as a model system to prove the implementation of these engineered inserts in a continuous biocatalytic application with high product yield and process stability. The setup was further automated in order to quickly identify the optimum operating conditions via a Design of Experiments (DoE) approach. The use of a systematic optimization, together with the adaptability of 3D printed equipment to the process requirements, render the presented approach highly promising for a more feasible implementation of biocatalysts in continuous industrial processes. . . . . Keywords 3D printing Biocatalysis Automation Continuous flow DoE. Highlights • Design and printing of ceramic inserts engineered for biocatalytic in flow applications. � Characterization of the inserts via Residence Time Distribution experiments, enabled by in house designed and produced 3D printed flow cells. � Immobilization of bsPAD onto the inserts via covalent binding, followed by the application of a DoE approach in an automated setup to quickly identify the optimum operating parameters. * Alessia Valotta CATalytic mechanisms and AppLications of OXidoreductases valotta@tugraz.at (CATALOX), Graz, Austria * Heidrun Gruber-Woelfler Center for Continuous Flow Synthesis and Processing (CCFLOW), woelfler@tugraz.at Research Center Pharmaceutical Engineering GmbH (RCPE), Graz, Austria Institute of Process and Particle Engineering, Graz University of Lithoz GmbH, Vienna, Austria Technology, Graz, Austria 676 J Flow Chem (2021) 11:675–689 Introduction leakage during application. However, the mechanical proper- ties can be improved in the post-processing phase via steps of In recent years, the interest in 3D printing has expanded to drying, heating, or sintering [11]. The resolution for ME many branches of science and engineering, due to the decreas- printed parts is lower, but this technique is still widely used ing cost of desktop printers and the increasing choice of print- due to its high versatility and low costs [11]. ing materials available [1, 2]. In the field of chemical engi- As mentioned above, 3D printed reactors are widely used neering, 3D printing has been identified as a new solution to in flow chemistry, and many reviews can be found on this rapidly design and produce tailor-made reaction ware. This is topic [1, 10, 12–14]. Nevertheless, the use of 3D printing to particularly interesting for the pharmaceutical industry, since realize supports and internals for structured reactors is more active pharmaceutical ingredients (APIs) are structurally com- recent and still limited to few applications [15, 16]. In indus- plex and equipment flexibility is required to produce the target try, packed bed reactors filled with heterogenous catalyst compounds, thereby batch reactors have traditionally been the powder are still preferred due to the facile utilization and the equipment of choice [3]. Nevertheless, several pharmaceutical high surface area available for mass transfer. However, they companies have expressed increasing interest in switching to a pose some operational difficulties, such as high pressure drops continuous production mode, due to the numerous advan- and flow maldistributions, with consequent channeling and tages, such as constant product quality and easier optimization decrease in the overall process performance [8]. Therefore, of energy and resources [4, 5]. However the low flexibility of structured catalysts and packing materials have been devel- continuous equipment still impedes the implementation of oped to overcome these issues. Characterized by a flexible continuous processes into the pharmaceutical sector [5]. 3D design, they can be implemented into commercially available printing of customized reaction ware and equipment might be columns and reactors. Furthermore, they can be easily the solution to this issue. In fact, this manufacturing technique interchanged, which makes them particularly interesting as allows to implement non-conventional and complex reactor catalyst support. The catalyst can be immobilized on the solid geometries, whose parameters can be quickly optimized based matrix as a washcoat or incorporated in the chemical structure on experimental data by iterative design modifications [6]. of the support material [1, 8]. Moreover, by using additive Also, 3D printing in chemical synthesis is not limited to reac- manufacturing technologies, it is possible to design and engi- tor design, it is very often used to produce customized pieces neer the support structures to perfectly meet the reaction re- of equipment, e.g. common laboratory hardware or sensor quirements, such as fast mixing, improved heat transfer and analytic applications [1, 7]. Different additive manufacturing fluid distribution (e.g. reduced channeling) compared to ran- techniques are available, depending on the specific purpose, domly and particle-based packed beds [17]. In addition, rapid but the main working principle is similar: the object, designed optimization of the support structure is straightforward and with computer aided design (CAD) software is sliced in a can be achieved by iteratively improving the design based slicing software in a number of cross sections. The sliced on the collected experimental data. Finally, catalytic supports object is then uploaded to a printer, which shapes the object are much easier to handle and to fill into a reactor compared to by selectively stacking these layers above one another [2]. heterogeneous catalyst powders. How these layers are built depends on the printing principle: In this work, structured ceramic supports printed via vat in this work, we utilized vat photopolymerization (VPP) and photopolymerization and their use for catalytic applications material extrusion (ME). VPP, also known as are presented. In order to describe the microfluidic behavior stereolithography (SLA), generates a 3D object by selective inside the supports, low cost flow cells were designed to easily layer-by-layer solidification of a liquid photopolymer resin perform residence time distribution (RTD) experiments. The using a UV light source. VPP is renowned for its high resolu- RTD setup includes two 3D printed flow cells, common laser tion of the printed parts and for the high quality of surface beams as light source, and an Arduino microcontroller to finish [8, 9]. Moreover, this technique is widely used for print- monitor the change in absorbance when a dye is injected into ing materials with excellent mechanical properties, such as a microfluidic device. The setup was designed to be an effi- ceramics, by suspending solid particles within the resin mix- cient and low cost alternative to flow photometric equipment tures. ME instead is an extrusion-based technique, in which a available on the market and reflects the power of 3D printing thermoplastic polymer is melted and extruded through a hot to design customized analytical tools without having to pur- nozzle. The polymer is then deposited onto a building plate chase expensive and patented equipment [18, 19]. By com- through the nozzle in a layer-by-layer fashion, thereby creat- bining 3D printed parts with low-cost microcontrollers and ing an object by stacking different slices on top of each other electronics, it is possible to extend the scope of in-house de- [8]. After the layer deposition, the polymer hardens on signed laboratory equipment to more complex applications, cooling, which might cause adjacent layers to not bind prop- such as inline and real-time monitoring of defined process erly. In general, this results in low mechanical stability of the parameters [19–21] For rapid prototyping purposes, Arduino printed parts along the z-direction [10] and possible fluid microcontrollers are the most known and commonly used, due J Flow Chem (2021) 11:675–689 677 to the many advantages they offer. Firstly, they are cheap and commonly used in additive manufacturing [23, 24, 27]. This can be coupled with a wide range of sensors and devices. has opened up the possibility to 3D print bioreactors or car- Secondly, the software and the board are user friendly, which riers made from commercially available resins, allowing direct makes it easier for scientists with little programming back- immobilization of the enzymes onto the structures’ surface ground to create their own prototype. Finally, Arduino is an and enabling high performance and recyclability [28–30]. open source project with a wide community of users, which Regarding inorganic materials, enzymes have been very often increases the possibility of sharing ideas and prototypes covalently attached onto silica or ceramic particles and/or among labs and researchers [20]. monoliths [30–33]. However, to the best of our knowledge, As an application-oriented proof of concept, in this work this is the first report of enzyme immobilization onto 3D the model enzyme Phenolic Acid Decarboxylase from printed ceramic supports. Bacillus Subtilis (bsPAD) was covalently immobilized onto In this work, we examined the continuous transformation 3D printed solid supports and tested for the decarboxylation of of para-coumaric acid into vinylphenol catalyzed by bsPAD coumaric acid in a continuous flow setup. The system was (Scheme 1) covalently immobilized onto our 3D printed in- chosen due to the considerable industrial and economical in- serts. Since this enzymatic reaction had already been studied terest in biocatalysis and the ongoing efforts to increase the previously by our group using an encapsulated biocatalyst in feasibility of enzymes by finding innovative and optimal sup- continuous flow [34], in this work we decided to utilize a ports for their immobilization [22]. In fact, biocatalysis has ceramic support material and focus on determining the influ- emerged as a green and highly efficient alternative to metal- ence of different internal geometries and process parameters based catalysis for the manufacturing of APIs and fine on the reaction outcome. Ceramics was preferred to standard chemicals [23]. Enzymes are natural and environmentally 3D printing resins as a support material since it is chemically friendly catalysts, showing high activity in mild conditions inert and does not pose any risks of inactivation to the enzyme (low temperatures and water-based solvents) as well as high in use. In the case of bsPAD, the choice of using ceramics substrate specificity. Moreover, due to past advances in pro- resulted in a great compatibility of the enzyme to the carrier tein expression and purification, the price of enzymes has material, as demonstrated by the long-term activity and stabil- decreased strongly over the last years, making enzymes more ity achieved. Moreover, ceramics is less brittle than standard economically feasible even for large scale processes [24]. resins [1], which makes it easier for an insert to be tightly However, the limited long term stability and reusability of packed in a column without being crushed. Alumina is also enzymes make it very challenging for this technology to be cheap and widely available, therefore it is possible to easily competitive on an industrial level [22, 23]. Hence, enzyme produce objects with great mechanical and chemical proper- immobilization has been proposed as a very efficient solution ties at an affordable cost. to overcome these issues [22]. Various methods for immobi- In order to decrease the time needed for screening the dif- lization are available and differ in the mechanism of protein ferent selected parameters and increase the level of process attachment, such as affinity bonding, physical adsorption, co- understanding, an automated process setup was realized, as valent bonding and encapsulation. Each of these methods has shown in Scheme 3. The possibility of process automation is its advantage and drawbacks, and the choice greatly depends one of the many advantages of continuous manufacturing, as on the specific application [23–26]. Covalent binding is com- it allows for rapid and controlled screening of process param- monly used when enzyme immobilization onto a solid support eters [35–37]. On lab scale, automated reaction platforms are material is targeted, as the support surface can be easily mod- enabled by using standard hardware-connectivity to remotely ified and functional groups for enzyme attachment are facile to control the equipment, and online analytics to monitor the introduce. This technique allows to preserve the enzymatic process parameters in real time [36, 38]. In combination with activity for a longer period of time, reuse of the biocatalyst computer algorithms, it is possible to perform experiments as well as easy enzyme separation from the reaction mixture automatically by controlling the equipment, data acquisition, [26]. Several organic and inorganic materials have been used and performing process optimization based on the recorded as matrix for covalent enzyme attachment including polymers data [36, 38]. Multidimensional systematic optimization Scheme 1 Reaction scheme for the enzymatic decarboxylation of coumaric acid to vinylphenol catalyzed by bsPAD 678 J Flow Chem (2021) 11:675–689 strategies, such as Design of Experiments (DoE), can be easily distribution along the channels, depending on the individual implemented in such automated platforms and are generally channel backpressure, whereby no mixing can take place be- preferred to the one-variable-at-a-time (OVAT) approach tween them. The purpose of this design was to approach a [39–41]. In the DoE approach, process parameters are first plug flow behavior while reducing the extent of backmixing. ‘screened’ to identify the critical factors influencing the reac- However the individual backpressure of the channels can tion outcome (e.g. for a chemical reaction: yield, purity, cost, cause the fluid to not distribute evenly among the channels etc.). Then, an ‘optimization’ step is carried out to determine and might affect a structure’s performance. The second struc- the best settings for the individual variables [40]. DoE also ture features a cubic lattice (CL) design that is repeated along reveals more information on the interactions between the pro- the length of the support (Fig. 1 right). In this way, radial cess parameters and their effect on the outcome of the reac- mixing is introduced, while increasing the available surface tion, and allows to save time and materials by minimizing the area. Both designs are equipped with an outer shell and an O- total number of experiments needed [39]. ring on both sides to fix the support tightly when inserted into a column, thereby preventing bypassing of the fluid. The outer In this work, we have applied a fractional factorial Central diameter of all supports was set to 7.8 mm at the O-ring sec- Composite Design (CCD) [42, 43] for the DoE study to in- tion and kept at this diameter for the lattice structure. The O- vestigate several process parameters and gather as much in- ring prevented adding channels close to the outer radius of the formation as possible on the bsPAD catalyzed reaction using honeycomb as it would have blocked them. As a consequence, our structured inserts. CCD was also used to define the opti- the unused mass in the middle of this design needed to be mal carrier and combination of process parameters. The frac- reduced as it was likely to produce cracks during post-process- tional factorial CCD approach was preferred over the full fac- ing. A final diameter of 5.8 mm was chosen for the middle torial approach and similar approaches such as the Box- section. Both designs feature a length of 39.8 mm in order to Benkhen design due to the lower number of experiments fit perfectly into commercially available empty HPLC col- needed per iteration [35, 44]. umns (ID 8 mm, height 40 mm). All designs were replicated Overall, the goals of this work are to show the power of 3D and printed with 3 different channel sizes. The geometrical printing for designing flexible and easily interchangeable characteristics are summarized in Table 1. structured inserts, as well as for realizing powerful, yet low The structures were printed using a lithography-based ce- cost inline analytical tools to increase process understanding. ramic manufacturing technique. The printer used was a To achieve these goals, two different types of structured in- CeraFab 7500 and the printing setup is shown in the serts were designed, based on two different internal geome- Supporting Information (SI, Figure S3). It features a transpar- tries. To determine the flow pattern inside the inserts, RTD ent vat into which the slurry, made of a mixture of monomer, photoinitiator and alumina, is automatically dispersed and experiments enabled by our in-house designed and printed inline photometric flow cells were used. Then the decarbox- spread. The movable building platform is immersed into the ylation of coumaric acid catalyzed by bsPAD was chosen as a slurry, which is then selectively exposed to visible light from model system to prove that it is possible to use these below the vat. The layer image is created via a digital micro- engineered inserts as structured bed reactors in a continuous mirror device (DMD) coupled with an advanced projection biocatalytic application, with great results in terms of produc- system. By repeating this process, a three-dimensional green tivity and process stability. By automating this setup, it was part can be generated layer-by-layer. A support structure was also feasible to quickly identify the optimum operating condi- used for each column to avoid over polymerization in the tions for the decarboxylation reaction, thereby exploiting the channels. More information on the building parameters can advantages of flow chemistry to the fullest. be found in the SI. After printing, the parts were cleaned with pressurized air and a cleaning solvent. Then, thermal postprocessing was carried out by placing all parts in a furnace Results and discussion and first applying a preconditioning cycle at 120 °C, followed by debinding and sintering at 1500 °C for 2 h. Design and fabrication of the 3D printed inserts Characterization of the inserts– determination of the The supports used for immobilization of the enzyme were mixing behavior/flow pattern designed using Autodesk®’s CAD software Inventor® and printed at Lithoz GmbH. Two different internal structure types In order to assess the flow pattern inside of the columns were realized, by following a similar approach from literature packed with designed inserts, experiments were carried out [17]. The first one consists of a honeycomb-like structure to determine the residence time distribution (RTD) using the (HC) comprising internal straight channels of hexagonal cross step input principle, as reported in literature [45]. The detailed sections (Fig. 1 left). This design was chosen to achieve fluid procedure of the measurements is provided in the SI, and the J Flow Chem (2021) 11:675–689 679 Fig. 1 CAD drawings of the designed and 3D-printed ceramic inserts. Straight honeycomb like structures (HC, left) and cubic lattice (CL, right). From top to bottom, the hydraulic diameter of a respective structure decreases, while the internal surface to vol- ume ratio increases setup is presented in Scheme 2. Two syringe pumps were all experiments and the step itself is not ideal. Another issue used, one filled with ethanol and one with a mixture of lies in the feeding tube or capillary used to deliver the tracer ethanol and methylene blue as tracer. Both were connected to the HPLC column. If the axial dispersion in this capillary to a six-way-valve, controlling the flow through the struc- is high, this might influence the RTD, thereby falsifying the tured insert, which was incorporated in a HPLC column. results. This effect is especially relevant for low Reynolds Two flow cells were implemented in the system, one at the number ranges (as it is the case for microfluidic applica- inlet and one at the outlet of the column. With this setup any tions). One way often proposed in literature to compensate sources of disturbances and deviations occurring not only for deviations in the system is to record the tracer signal at inside of the column, but also at its inlet are taken into the inlet and at the outlet of the device. The results are then account. For example, using a manually switched valve to analyzed by using the approach of the convolution integral inject the tracer can lead to disturbances in the evaluation of theorem, to take into account deviations to the RTD being the step signal, as the time for injection is not univocal for caused by auxiliary equipment [45, 46]. Table 1 Summary of the geometrical characteristics of the designed inserts 3 2 2 Type of insert Number Hydraulic diameter [mm] Internal volume [mm ] Internal area [mm ] Area/Volume [mm /mm³] 1 0.757 351.71 1857.31 5.3 Honeycomb (HC) 2 0.521 314.86 2416.17 7.7 3 0.263 234.65 3572.21 15.2 1 1.564 1044.78 2671.38 2.6 Cubic lattice (CL) 2 0.930 842.98 3627.25 4.3 3 0.433 467.12 4319.53 9.2 The internal volume and area were calculated in Autodesk Inventor 680 J Flow Chem (2021) 11:675–689 Scheme 2 Setup for the evaluation of the Residence Time Distribution (RTD) and thus the flow patterns inside the designed structured inserts For the experiments, 3D printed flow cells mountable on step down was induced by switching the six-way-valve back the outside of transparent 1/16” capillaries were designed (see to the load position. Different flowrates (2, 1, 0.5, 0.2, 0.1, SI Figure S4). Their measurement principle is based on a light 0.05 mL/min) were tested in order to evaluate how the flow intensity measurement passed through a capillary, whereby pattern changes with varying Reynolds number. For each ex- each flow cell represents a photometer on its own. The emitted periment, the results were exported in a .log file and imported light of a red LED is absorbed within a capillary and its fluidic in Microsoft Excel to evaluate the cumulative and exit age content and finally detected on the other side by a photo re- distributions. In order to quantify the extent of axial dispersion sistor. The photo resistor changes its conductivity depending and backmixing into the devices, the Bodenstein number (Bo) on the absorbed light and allows measuring of light intensity was determined for each insert at every investigated flowrate. depending voltages. This voltage is recognized by an analog This dimensionless number is defined as the ratio of the con- to digital converter (ADS1115, 16 bit). To account for differ- vective mass transfer over axial dispersion: ent mounting and manufacturing deviations of the electronic u  L char parts, an adjustable voltage divider was added to tune offset Bo ¼ ax values. Buttons for LED control and reference points are where u is the flow velocity, L is the characteristic length of added as well, to take the needed light and dark reference char values. LEDs and photo resistors are powered by a micro the device andD is the axial dispersion coefficient. Bo can be ax used to estimate how close the fluid behavior inside of a reac- controller (Arduino Nano), which reads the measured voltages of the ADS1115 and provides measured data to a PC via a tor is to an ideal reactor model: for Bo > 100, the fluid pattern serial communication. Beside the LED and photo resistors, all inside the investigated reactor approaches that of a plug flow other electronic parts were soldered on a printed circuit board reactor (PFR), while for Bo < 100, the behavior approaches (PCB) to increase measurement stability. With the developed that of a CSTR [45]. Different methods have been reported set-up, a low cost possibility to measure RTDs is presented. to calculate this number from the RTD curves, based on dif- The experimental procedure started by flushing the column ferent assumptions on the boundary conditions of the system. with pure solvent and saving a light and dark reference spec- In this work, the open-open model was chosen, as it assumes trum. Then, the tracer solution was injected into the column that the flow is dispersed along the length of the whole column and it also provides an analytical solution to calculate Bo after defined time intervals by manually switching the six- way-valve to the inject position. The change in absorbance solely from the variance of the RTD curve [47]. Moreover, since it was possible to place two flow cells at both the inlet was recorded at both the inlet and the outlet of the column and converted into a digital signal via an Arduino Nano and and outlet of the columns, the calculated variance and mean logged into a serial terminal program (Tera Term) on a PC. residence time for each insert could be corrected by taking into When the absorbance of the tracer reached a stable value, a account the influence of the inlet capillary on the output J Flow Chem (2021) 11:675–689 681 response. For this purpose, the additivity approach described pulsating and irregular behavior in this range. Another possi- by Levenspiel [45] was used. bility is that at lower flowrates flow maldistributions might The results for the RTD experiments are summarized in occur, because the connector that ties the capillary to the Fig. 2, by plotting the calculated Bo versus the Reynolds num- HPLC column has a small inlet in the middle, therefore the ber (Re) for each measured point. Re was calculated by iden- fluid might not distribute to the outer channels and flow main- tifying a hydraulic diameter d (as reported in Table 1)for ly through the middle ones. Maldistributions might be more each structured insert, as it allows for the comparison among relevant for HC3 and HC2 due to the lower channel diameter different internal structures. The d was calculated as follows: and the higher number of channels, resulting in the fluid to not be distributed equally among all channels and therefore de- 4  A creasing the mixing efficiency. Since HC1 has bigger chan- d ¼ nels, it is easier for the fluid to be pumped uniformly in all channels, even for low Re. where A is the cross sectional area of the device and P is the The trend instead changes for the square lattice structured wetted perimeter, which were both estimated with the aid of inserts CL 1–3. As visible in Fig. 2b, no trend was recorded Autodesk Inventor®. The Re was then calculated as : for Bo with increasing flowrate, instead it changed around an ρ  u  d average value. A slight increase in Bo could only be detected Re ¼ for CL3, which was expected since it has the smallest hydrau- lic diameter. In fact, the higher density of lattice elements Where ρ and μ are respectively the density and the dynamic present inside CL3 promotes the formation of secondary flow viscosity of the fluid flowing through the channel. structures, thereby increasing the extent of convective mixing. The results show that for all the designed inserts in the For CL1 and CL2 there was almost no difference in Bo for investigated flowrate region, Bo was significantly below Re > 5. However, below that, CL1 seemed to have a similar 100, indicating high backmixing and axial dispersion inside behavior to CL3. As for the HC inserts, this inversion in the the inserts. Considering the results for the honeycomb (HC) trend could be arising from fluctuations in the pumps, or it structured inserts (Fig. 2a), it can be seen that the average Bo could be due to flow maldistributions inside the CL3 insert. was the highest for HC3 and it was rising with the flowrate. In general, for both types of inserts it can be expected that Since the superficial velocity inside a channel is increasing using higher flowrates improves mixing and promotes PFR with decreasing inner diameter, therefore the Re is higher for behavior, however since higher flowrates were not interesting HC3 and the flow is more chaotic, resulting in a lower extent for the chosen application (as they result in too low residence of backmixing. Axial dispersion is instead higher in the case time for the selected reaction), the investigated range was not of HC2 and HC1: in both inserts the change in Bo with the enlarged any further. flowrate was very limited, indicating that the insert was ap- proaching CSTR behavior in the whole investigated region. Also, the values of Bo seemed to increase at lower Re for these Decarboxylation reaction – choice of the solvent inserts, especially for HC1, for which Bo was higher than the values achieved by HC3 at Re below 7. This trend can be Phenolic acid decarboxylase (PAD) is an enzyme belonging explained by the fact that syringe pumps have a more to the family of cofactor-free decarboxylase. It catalyzes the ab Fig. 2 Results of the RTD experiments for the honeycomb (a) and the cubic lattice (b)structured inserts 682 J Flow Chem (2021) 11:675–689 removal of carbon dioxide from hydroxycinnamic acids, such surface modification steps until silanization are the same, but as coumaric acid, to give hydroxystyrene products that are instead of having a surface functionalization step, the enzyme important API precursors [48]. Even though PAD is very ac- was directly immobilized onto the support by introducing the tive towards this reaction, a limit to the industrial application is less harmful linkers N-hydroxysuccinimide (NHS) and N-(3- the low solubility of phenolic acids in water-based solvents. dimethylaminopropyl)-N′-ethylcarbodiimide hydrochloride Therefore, alternative solvent systems have been proposed, in (EDC) into the enzyme solution. These compounds activate order to maintain the high activity of PAD and increasing the the available carboxylic groups on the amino acids, which un- space-time -yield (STY) of the decarboxylation [48]. Deep dergo bond formation with the amine groups on the carriers Eutectic Solvents (DES) have been suggested as an alternative [56]. Preliminary batch tests showed that the latter method is to ionic liquids, due to the lower toxicity and the greater up- preferred in terms of immobilization yield and long-term stabil- take of the CO released during the decarboxylation [34]. ity of the enzyme activity (see SI). Therefore, this method was They consist of a mixture of primarily quaternary ammonium chosen to immobilize bsPAD on the 3D printed inserts for the salts (e.g. choline chloride) with hydrogen-bond-donors purpose of performing the flow experiments. (HBD) (such as glycerol). The heating and mixing of this The immobilization of bsPAD on the structured inserts was solution disrupt the crystalline structure of the ammonium salt, carried out in flow, as shown in the SI. This allowed to avoid generating a viscous liquid with a lower melting point than the material handling in between the different immobilization starting materials [34, 49]. steps and reduced the overall time needed for the procedure. The DES mixture of choline chloride and glycerol in a 1:2 In order to estimate the amount of enzyme immobilized on the molar ratio, diluted 1:1 (v/v) with phosphate buffer at pH 6 supports, the immobilization yield (Y ) was determined via was identified in previous works [34, 48] as an efficient sol- taking 1 mL samples from the enzyme solution before and vent for the decarboxylation, both in terms of CO absorption after immobilization and performing the activity assays in and high substrate solubility, since the properties of the DES batch as described in the SI. By determining the activity of are still relevant at this dilution ratio. Therefore, this solvent the solution before and after immobilization, it was possible to mixture was kept also in this work to investigate the 3D calculate Y as: printed inserts as catalytic supports for the bsPAD enzyme. A  A b a The decarboxylation of coumaric acid to vinylphenol was Y ¼  100 performed first in batch for the preliminary immobilization tests (see the SI) in order to identify an optimum immobiliza- Where A is the activity before and A the activity after b a tion procedure. Finally, automated flow experiments were car- immobilization. The activity of bsPAD was defined as the ried out following the procedure explained in the dedicated amount of μmoles of coumaric acid being consumed per min- paragraph. ute, taking the first 30 min of reaction for the calculation, as in this region the conversion of coumaric acid was linear. The Enzyme immobilization calculation was carried out as follows: Covalent binding was chosen as an immobilization strategy, C  C  V CA;0 CA;30 assay U ¼ since it has been widely used as a method for protein attachment on ceramic supports [50–53]. Two different methods (1 and 2) Where C and C are the concentration of coumaric CA;0 CA;30 were selected and first tested in batch by using the particles acid at the start and after 30 min into the reaction (given in obtained from grinding the ceramic objects that failed during μmol/L, determined by HPLC),V is the total reaction assay printing. All steps for both methods are described in detail in the volume and t is 30 min. The activity is therefore is given in SI. Method 1 was taken and adapted with slight modifications units U [μmol/min]. from literature [50]. It involved a first step of surface activation The theoretical activity of the enzyme immobilized on by acid treatment, followed by a silanization step with 3- each insert was calculated by assuming that this is equal aminopropyltriethoxysilane (APTES) to introduce amino to the difference between the activity in the starting im- groups onto the surface, which are functionalized with glutar- mobilization solution and the one measured at the end of aldehyde. This compound comprises two reactive aldehyde the immobilization. This value was only used as an esti- groups and it is used to link two amine functionalities, one on mation, mainly as it was not possible to access the small the surface of the support and one on the enzyme itself, by internal channels in order to determine the exact concen- forming amide bonds [54]. This method has the advantage that tration of the bound enzyme via standard assays (e.g. pro- it ensures stable protein attachment. However, glutaraldehyde is tein labelling with fluorescent molecules [57]). Also it toxic and might even deactivate the enzyme [55]. Therefore, a does not take into account whether the enzyme undergoes different immobilization procedure (method 2) was chosen, al- conformational changes upon covalent binding to the so slightly modified from literature [52, 53]. In this case, initial J Flow Chem (2021) 11:675–689 683 Table 2 Overview of the measure BET surface area and the achieved immobilization of bsPAD onto the structured inserts Type of insert Number BET surface area [m /g] Weight of insert [g] Immobilization yield [%] Estimated immobilized activity [U/g] Honeycomb (HC) 1 0.254 2.32 19 1.24 2 0.310 2.5 3 0.543 2.86 58 6.02 Cubic lattice (CL) 1 0.484 1.98 14 1.09 2 0.522 2.52 3 0.536 3.77 50 2.62 carrier, which might affect its structural properties com- from the BET results). The calculated values for Y and the pared to the free enzyme [23, 25]. estimated activity showed that a higher amount of enzyme The exact surface area of each uncoated insert was deter- bounded to HC3 and CL3, most probably due to the higher mined via Brunauer–Emmett–Teller (BET) measurement, as surface area available for the immobilization. described in the SI. The results are summarized in Table 2. Only insert types 1 and 3 were tested in the continuous setup, Automated flow setup as the inserts with the middle channel size did not show to be interesting both in terms of their fluidic behavior (as shown After being coated with bsPAD, the inserts are ready to be from the RTD results) nor available surface area (as shown used in the continuous flow setup shown in Scheme 3.The Scheme 3 Setup for the automated screening and optimization of the decarboxylation of coumaric acid catalyzed by bsPAD immobilized onto 3D printed inserts. DES = Deep eutectic solvent (ChCl/Gly 1:2 mol) 684 J Flow Chem (2021) 11:675–689 scope of this setup was to allow the automated evaluation of having performed all experimental runs on the first col- the different coated carriers and the effect that changing tem- umn, the algorithm prompted the valve to switch to the perature, flowrate and substrate concentration have on the second column, and the same optimization algorithm was decarboxylation reaction, according to a DoE approach. repeated. Once both inserts had been evaluated, the col- The structured inserts were again enclosed into HPLC col- umns were removed from the thermostat and the inserts umns and constituted the structured bed reactors at the heart of were exchanged manually to the other design type. Then, the process. The setup featured one HPLC pump (Knauer the columns were reinstalled in the setup and the DoE Azura P 4.1 S, pump B in Scheme 3) delivering solvent, and evaluation was started again by following the same proce- a syringe pump (Lambda VIT-FIT, pump A in Scheme 3) dure as before. equipped with a 50 mL stainless steel syringe filled with a 10 mM stock solution of coumaric acid. The outlets of the Design of experiments (DoE) pumps were then connected to a static mixer, in order to pro- vide a feed stream with uniform concentration before it Fractional CCD was chosen for the DoE study due to its effi- reaches the column inlet. As static mixer, a 3D printed stain- ciency for screening a high number of reaction parameters less steel microfluidic device was used, which was designed in using a low number of experiments. Regarding the particular our group and characterized in a previous work [58]. This enzymatic reaction studied in this work, it was highly desired chosen device was the so-called AP04 with an internal diam- to reduce the amount of experiments to not waste solvent or eter of 0.6 mm, into which the solvent and the substrate solu- reagent, thereby reducing the amount of waste and containing tion were mixed according to a split-and-recombine principle. the costs. Three factors were chosen for optimization: feed The AP04 was chosen as static mixer since it was proved in a flowrate, reaction temperature and substrate concentration. previous publication [58] that it provides reasonably good These values were evaluated within a limited design space degrees of mixing while remaining a compact device. The reported in Table 3. The maximization of the yield and feed was then pumped into one of the columns, each connect- space-time-yield (STY) of vinylphenol were chosen as the ed to a six-way-valve (Knauer Azura VU 4.1), which dictates target response of the DoE. The STY was calculated accord- the column into which the feed is pumped to. Two inserts of ing to the following equation [59], assuming that the enzyme the same type (e.g. HC1 and HC3, or CL1 and CL3) were is forming a monolayer on the inner surface of the insert: evaluated at a time, by being fitted in HPLC columns and then c  FR prod connected to ports 1 and 2 of the six-way-valve. The columns STY ¼ were immersed in a temperature-controlled water bath to keep i the temperature at the desired value. The pumps, six-way- where c is the concentration of product formed (in g prod valve and thermostat were connected to a PC and controlled vinylphenol/L) at steady state at the chosen experimental con- via a Python-based script. In this way, the automated control ditions, FR is the flowrate (in L/h) and V is the internal and change of temperature, flow rate and substrate concentra- volume of the reactor (in L). tion for each experiment were facilitated. The upper and lower limits for the flow rate were chosen Before starting the automated experimental sequence, considering a sufficient residence time within the column as the whole setup including the columns was flushed for at well as concerning the structural limits of the UV-vis flow cell least 1 h in order to remove loosely bound enzyme and take (it can only withstand a pressure up to 10 bar). For the tem- a blank reference for the UV-vis measurement. Then perature, the limits were set around the optimal temperature starting with column one, the DoE algorithm performed defined in a previous work [34]. In terms of the substrate different experimental runs in order to screen the reaction concentration, the low solubility of coumaric acid in the reac- space as determined by the fractional CCD method, which tion solvent and the flow cell path length were the limiting is described in detail in the next section. During each run, factors. The starting substrate concentration for each the reaction was monitored via an Avantes Micro Flow Cell with 1.5 mm path length connected to a spectrometer (AvaLight-DS-DUV equipped with a deuterium lamp and Table 3 Boundaries of the design space investigated in the DoE AvaSpec-ULS2048 detector), allowing the real-time track- experiments ing of the rate of vinylphenol formation. The product was Parameter Lower limit Upper limit identified by a characteristic peak at the wavelengths be- tween 269 and 263 nm, whereas the reagent was measured Flowrate 0.2 mL/min 2 mL/min at wavelengths between 325 and 328 nm, where Temperature 25 °C 35 °C vinylphenol does not absorb. From this data and previously Substrate concentration 0.25 mM 2 mM determined calibration curves, the conversion of coumaric Dilution ratio 4 40 acid and the yield of vinylphenol were calculated. After J Flow Chem (2021) 11:675–689 685 experiment was set in the platform by changing the flowrates this work. The new experimental points lie at the corners of of the solvent and the substrate pumps in order to achieve the the cube, and are therefore equally distant from the center. desired dilution ratio. Therefore, dilution ratio was given as an Since a fractional factorial design was applied, only 4 of the input parameter to the DoE algorithm. corner points were added to the total experiments, resulting A scheme summarizing the steps for the chosen experi- in a final amount of 11 experimental runs (Fig. 3b). Then, mental design is shown in Fig. 3. For each 3D printed struc- each experiment is evaluated singularly and after all runs are tured insert, two DoE runs were performed. In the first run, a completed, the best point of the first iteration is chosen in subsection of 70 % of the total design space was screened to terms of highest STY (Fig. 3c). In the second run, the reac- identify an optimum combination of parameters (each rep- tion is further optimized by investigating the experimental resented as a point) and narrow down the experimental space near the best point found in the first iteration, which space. Based on the boundaries given by the user, the algo- was taken as center point (Fig. 3d). In this phase, the geo- rithm defined the center point and the high and low levels metrical size of the CCD was set to 40 % of the first itera- for each factor, which were connected by a cross on each tion, therefore ensuring that no point would lie outside the plane in the 3D design space (Fig. 3a) and constituted the initial boundaries of the design space. The overall amount of axial experimental points, resulting in 7 experiments. Then, experiments needed is 21, of which 11 were carried out in further points were added to the starting ones, and were the first run and 10 in the second. The combined size of the chosen by the algorithm based on a spherical composite two iterations ensured, that 98 % of the reaction space can design approach [43]. According to this approach, a cube be explored via this method. is drawn, which is centered at the center point of the design The DoE was executed via the automated reaction platform and has a side length of α. This distance is defined as the shown in Scheme 3. A Python-based algorithm with an integrat- 1/2 square root of the total amount of factors and was 3 for ed graphic user interface (GUI) was developed to automatically Fig. 3 Graphical representation of the steps for the chosen Fractional from the center point (b). However not all corner points are selected as Factorial CCD approach. For each structured reactor, two DoE runs experimental runs, but only 4, as this is a fractional design. Once all points were performed. First, the algorithm choses the center point of the DoE are defined, the algorithm performs all 11 experiments automatically, and according to the boundaries of the design space given by the user. Then, choses the best point based on the highest STY (c). A new iteration is then the high and low levels for each factor are also identified and combined to started (d) by choosing 10 points around the best point from the first give the axial experimental points (a). Since this is a CCD approach, iteration, which becomes the new center point, following the same further points are added to the experiments, which lie at the corners of a procedure as in the first iteration but by scanning a smaller area of the cube that has a fixed length (identified as α) and are at the same distance design space 686 J Flow Chem (2021) 11:675–689 choose the DoE points, according to the experimental space The results for the optimization experiments are plotted in limits defined by the user, and to set the thermostat and the Fig. 4 andreported inTable 4. At the end of the second itera- pumps at the desired level for each experimental run. The same tion, the best combination of factors for HC1, HC3 and CL1 algorithm saved and accessed the UV-Vis spectra recorded by was 30 °C, 1.1 mL/min and a dilution ratio of 4.36, which the spectrometer in order to calculate the compound concentra- correspondstoastartingconcentrationof1.86mM. Forthese tions, yield and STY of each experiment. These variables were inserts, it was apparent that the optimum temperature lays chosen as response parameters to assess the result of each exper- around 30 °C, which is in line with what was already suggested iment and determine the optimum combination of factors. The in previous works [34, 48]. Moreover, for all of them a combi- final output parameter that was to be maximized via the DoE was nation of a medium flowrate and lowest dilution ratio was pre- STY, as this parameter can take into account not only the yield of ferred, as this gives the highest productivity, thereby optimizing the reaction but also the influence of flowrate and free volume in the STY. The picture is instead different for CL3; for this insert the column. Once all of the experiments were performed on one the best point was 28 °C, 1.464 ml/min and dilution ratio of column, the algorithm prompted the six-way-valve to switch the 9.676 which corresponds to a starting concentration of 0.93 feed to the second column, onto which the DoE approach was mM. So in this case the algorithm went down another path repeated as before. After having evaluated both inserts, the algo- for optimization, but still preferred a compromise between rithm returns the results of the DoE, and chooses the best insert flowrate and dilution ratio to increase productivity. The reason and set of process parameters. The columns could then be man- for which in this case a different optimum was found is clear ually removed from the setup and the inserts exchanged to the when looking at Fig. 4 (CL3): in the first iteration, the points at other design type for further experiments, which were carried out 28 °C,1.464 mL/min and 0.63 mM (dilution ratio of 14.71) and as explained above. at 30 °C, 1.1 mL/min and 0.96 mM (dilution ratio of 9.4) Fig. 4 Graphical summary of the DoE results for the 4 structured inserts insert depending on the internal volume. The best point in the first investigated. The cross points for the first iteration are connected via blue iteration for HC1, HC3 and CL1 was 30 °C,1.1 mL/min and a dilution dashed lines, while for the second iteration they are connected in red. The of 9.4. For CL3 instead it was 28 °C, 1.464 mL/min and 14.71 dilution. color map to the right of each graph gives an indication of the level of For the second iteration, the best combination of factors was 30 °C, 1.1 STY reached for each experimental point. The results are plotted in terms mL/min and a dilution ratio of 4.36 for HC1, HC3 and CL1. For CL3 the of Residence time rather than flowrate, as the first differs among each best point was 28 °C, 1.464 mL/min and ad dilution ratio of 9.676 J Flow Chem (2021) 11:675–689 687 Table 4 Summary of the DoE First iteration Second iteration results for each structured insert used in this work Type of insert Yield [%] STY [g/L·h] Yield [%] STY [g/L·h] Residence time [s] HC1 62.6 13.59 64.7 27.27 19.15 HC3 70.9 23.11 62.2 39.21 12.8 CL1 59.2 4.32 67.3 9.53 56.95 CL3 77.6 11.16 61.9 13.11 19.14 resulted in a very close STY, but since the algorithm goes for pieces of equipment for many applications in flow. First, the best point and does not explore the space around the second novel 3D printed ceramic structured inserts have been de- best, the second iteration went in a different direction. A more signed and 3D printed via VPP. In order to characterize the complex optimization algorithm could identify and evaluate fluidic behavior inside the inserts, RTD measurements also different local optima, however for the purpose of this were carried out with the aid of self-made 3D printed pho- work the chosen DoE approach was enough to identify a trend tometric flow cells. These tools were used to determine the to optimize the overall productivity. exit age distribution for each insert via recording the step The STY achieved at steady state for the best combination of input of a tracer at different operating conditions. By re- parameters in each insert is summarizedinTable 4. For all the cording the tracer input at both the inlet and the outlet of a investigated inserts, the final STY was higher than the value of reactor, it was possible to remove any disturbances caused 4.8 g/L·h ,which was achieved in a previous work [34]. As by the feeding capillary, thereby increasing the quality of showninTable 4, HC3 gave the best results in terms of STY: the results and proving to be a valid low-cost alternative to at the optimum conditions, a value of 39.21 g/L·h was reached, expensive inline analytical equipment. The evaluation of which represents a 8-fold increase compared to what was the RTD results further allowed to conclude that all inserts achieved previously. This result can be explained by the fact approach the fluidic behavior of theoretical CSTR reactors. that HC3 has the lowest internal volume and the highest surface Then, as an application-oriented proof of concept, the in- to volume ratio, thus the highest enzyme coverage. Therefore, serts were enclosed in HPLC columns to be used as solid HC3 was deemed as the best choice to perform the decarbox- supports onto which bsPAD was immobilized for the cat- ylationofcoumaric acidincontinuousflow, as it resultsina alytic conversion of coumaric acid to vinylphenol in con- higher amount of substrate being converted in time and also tinuous flow. To identify the optimal conditions for the offered the best mixing properties compared to all designed selected system, a fractional factorial CCD was chosen as inserts. Moreover, compared to the previous work, the applica- a DoE approach to systematically investigate the effect of tion of an automated flow setup together with a systematic DoE different process parameters (temperature, flowrate and di- strategy proved to be a successful approach to rapidly screen the lution ratio) on the STY. The DoE experiments were suc- effect of different parameters on the outcome of a reaction by cessfully carried out in an automated platform controlled minimizing the need of human intervention and the time needed by a Python-based algorithm, which automatically chose for evaluation. Also, the reaction could be followed in real time the DoE points based on design space given by the user via implementing an online measurement, which is more reli- and run the experiments by remotely controlling the equip- able compared to offline methods and does not require sample ment, reducing the need for human intervention and saving handling/manipulation [60], thus saving time and resources. time and materials. As a result, the HC3 insert proved to be Finally, the possibility of easily designing and manufacturing the best choice for the production of vinylphenol, as it gave interchangeable inserts via means of 3D printing, allows to aSTY of 39.21g/L·h,representingan8-foldincreasecom- rapidly adapt to the reaction needs to achieve optimal process pared to the value of 4.8 g/L·h obtained previously. These conditions. These advantages, coupled with the improved pro- results showed that it is possible to combine different ad- ductivity, represent definitely a step forward towards a more vantages of flow chemistry and 3D printing to biocatalytic industrially feasible enzymatic conversion of coumaric acid to processes, with great results in terms of productivity. In the vinylphenol. near future, the presented approach combining 3D printing and process automation will be applied to further enzymat- ic reaction systems, in order to increase the feasibility of Conclusions biocatalysts for their application in industrial processes. Moreover, the presented inserts will also be implemented In conclusion, we have proved in this work how 3D printing as supports for other active species and tested in different catalytic applications. can be used as a powerful tool to design and produce advanced 688 J Flow Chem (2021) 11:675–689 Supplementary Information The online version contains supplementary ingredients and finished dosage forms: an industry perspective. Org material available at https://doi.org/10.1007/s41981-021-00163-4. Process Res Dev 16:1586–1590 5. Poechlauer P et al (2013) Pharmaceutical roundtable study demon- strates the value of continuous manufacturing in the design of greener processes. Org Process Res Dev 17(12):1472–1478 Abbreviations API, Active Pharmaceutical Ingredient; BET, Brunauer– 6. Kitson PJ, Rosnes MH, Sans V, Dragone V, Cronin L (2012) Emmett–Teller; Bo,Bodenstein number; bsPAD, Phenolic acid decar- Configurable 3D-printed millifluidic and microfluidic ‘lab on a boxylase from Bacillus Subtilis; CAD, Computer Aided Design; CSTR, chip’ reactionware devices. Lab Chip 12(18):3267–3271 Continuous Stirred Tank Reactor; DES, Deep Eutectic Solvent; DoE, 7. Maierhofer M, Maier MC, Gruber-Woelfler H, Mayr T (2021) Design of Experiments; CCD, Central Composite Design; CL, Cubic Inline monitoring of high ammonia concentrations in methanol Lattice; GUI, Graphical User Interface; HC, Honeycomb; ID, internal with a customized 3D printed flow cell. J Flow Chem. https://doi. diameter; ME, material extrusion; OD, outer diameter; PFR, Plug Flow org/10.1007/s41981-021-00141-w Reactor; Re, Reynolds number; RTD, Residence Time Distribution; SLA, 8. Parra-Cabrera C, Achille C, Kuhn S, Ameloot R (2018) 3D printing Stereolithography; STY, Space-Time-Yield; VPP, vat in chemical engineering and catalytic technology: Structured cata- photopolymerization lysts, mixers and reactors. Chem Soc Rev 47(1):209–230. Royal Society of Chemistry Acknowledgements The authors kindly acknowledge the funding by the 9. Michel FM, Rimstidt JD, Kletetschka K (2018) 3D printed mixed CC FLOW project (Austrian Research Promotion Agency FFG No. flow reactor for geochemical rate measurements. Appl 862766), which is funded through the Austrian COMET Program by Geochemistry 89(November 2017):86–91 the Austrian Federal Ministry of Transport, Innovation and Technology 10. Bhattacharjee N, Urrios A, Kang S, Folch A (2016) The upcoming (BMVIT), the Austrian Federal Ministry of Science, Research and 3D-printing revolution in microfluidics. Lab Chip 16(10):1720– Economy (BMWFW) and by the State of Styria (Styrian Funding 1742. Royal Society of Chemistry Agency SFG). The authors also acknowledge the funding by the CATALOX (CATalytic mechanisms and AppLications of 11. Ambrosi A, Pumera M (2016) 3D-printing technologies for elec- OXidoreductases) project, doc.fund program funded by the Austrian trochemical applications. Chem Soc Rev 45(10): 2740– Science Fund (FWF). 2755. Royal Society of Chemistry 12. Au AK, Huynh W, Horowitz LF, Folch A (2016) 3D-Printed Microfluidics. Angew Chemie - Int Ed 55(12):3862–3881 Funding Open access funding provided by Graz University of 13. Parra-Cabrera C, Achille C, Kuhn S, Ameloot R (2018) 3D printing Technology. in chemical engineering and catalytic technology: structured cata- lysts, mixers and reactors. Chem Soc Rev 47(1):209–230 Declarations 14. Capel AJ, Edmondson S, Christie SDR, Goodridge RD, Bibb RJ, Thurstans M (2013) Design and additive manufacture for flow Conflict of interest There are no conflicts of interest to declare. chemistry. Lab Chip 13(23):4583–4590 15. Avril A et al (2017) Continuous flow hydrogenations using novel catalytic static mixers inside a tubular reactor. React Chem Eng Open Access This article is licensed under a Creative Commons 2(2):180–188 Attribution 4.0 International License, which permits use, sharing, 16. Kundra M, Grall T, Ng D, Xie Z, Hornung CH (2021) Continuous adaptation, distribution and reproduction in any medium or format, as flow hydrogenation of flavorings and fragrances using 3D-printed long as you give appropriate credit to the original author(s) and the catalytic static mixers. Ind Eng Chem Res 60(5):1989–2002 source, provide a link to the Creative Commons licence, and indicate if 17. Hock S, Rose M (2020) 3D-structured monoliths of nanoporous changes were made. The images or other third party material in this article polymers by additive manufacturing. Chem Ing Tech 92(5):525– are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the 18. Prikryl J, Foret F (2014) Fluorescence detector for capillary sepa- article's Creative Commons licence and your intended use is not rations fabricated by 3D printing. Anal Chem 86(24):11951–11956 permitted by statutory regulation or exceeds the permitted use, you will 19. Prabhu GRD, Urban PL (2020) Elevating chemistry research with a need to obtain permission directly from the copyright holder. To view a modern electronics toolkit. Chem Rev 120(17):9482– copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. 9553. American Chemical Society 20. Nguyen T, Zoëga Andreasen S, Wolff A, Bang DD (2018) From Lab on a Chip to point of care devices: the role of open source microcontrollers. Micromachines 9(8):403 21. Baden T, Chagas AM, Gage G, Marzullo T, Prieto-Godino LL, References Euler T (2015) Open labware: 3-D printing your own lab equip- ment. PLoS Biol 13(3):1–12 1. Capel AJ, Rimington RP, Lewis MP, Christie SDR (2018) 3D 22. Woodley JM (2019) Accelerating the implementation of biocataly- printing for chemical, pharmaceutical and biological sis in industry. Appl Microbiol Biotechnol 103(12):4733–4739 applications. Nat Rev Chem 2(12):422–436. Springer US, New 23. Zhu Y, Chen Q, Shao L, Jia Y, Zhang X (2020) Microfluidic York immobilized enzyme reactors for continuous biocatalysis. React 2. Dragone V, Sans V, Rosnes MH, Kitson PJ, Cronin L (2013) 3D- Chem Eng 5(1):9–32 printed devices for continuous-flow organic chemistry. Beilstein J 24. Faber K (2018) Biotransformations in Organic Chemistry. Springer Org Chem 9:951–959 International Publishing, Berlin 3. Gutmann B, Cantillo D, Kappe CO (2015) Continuous-flow tech- 25. Hanefeld U, Gardossi L, Magner E (2009) Understanding enzyme nology — a tool for the safe manufacturing of active pharmaceuti- immobilisation. Chem Soc Rev 38(2):453–468 cal ingredients. Angew Chemie - Int Ed 54:6688–6728 26. Guisan JM, Bolivar JM, López-Gallego F, Rocha-Martín J (eds) 4. Poechlauer P, Manley J, Broxterman R, Ridemark M (2012) (2020) Immobilization of enzymes and cells, vol 2100. Springer Continuous processing in the manufacture of active pharmaceutical US, New York J Flow Chem (2021) 11:675–689 689 27. Marques MPC, Fernandes P (2011) Microfluidic devices: Useful 44. Rakić T, Kasagić-Vujanović I, Jovanović M, Jančić-Stojanović B, Ivanović D (2014) Comparison of full factorial design, central com- tools for bioprocess intensification. Molecules 16(10):8368–8401 28. Ye J, Chu T, Chu J, Gao B, He B (2019) A versatile approach for posite design, and box-behnken design in chromatographic method enzyme immobilization using chemically modified 3D-printed development for the determination of fluconazole and its impurities. scaffolds. ACS Sustain Chem Eng 7(21):18048–18054 Anal Lett 47(8):1334–1347 29. Li R, Greenchem / (2017) Tuneable 3D printed bioreactors for 45. Levenspiel O (1999) Chemical reaction engineering, 3rd edn. transaminations under continuous-flow. Green Chem 19(22): Wiley, New York 5259–5516 46. Bošković D, Loebbecke S (2007) Modelling of the residence time 30. De Santis P, Meyer L-E, Kara S (2020) The rise of continuous flow distribution in micromixers. Chem Eng J 135(Suppl. 1):138–146 biocatalysis – fundamentals, very recent developments and future 47. Levenspiel O, Smith WK (1957) Notes on the diffusion-type model perspectives. React Chem Eng 5(12):2155–2184 for the longitudinal mixing of fluids in flow. Chem Eng Sci 6:227– 31. Datta S, Christena LR, Rajaram YRS (2013) Enzyme immobiliza- tion: an overview on techniques and support materials. 3 Biotech 48. Schweiger AK et al (2019) Using deep eutectic solvents to over- 3(1):1–9 come limited substrate solubility in the enzymatic decarboxylation 32. Plagemann R, Jonas L, Kragl U (2011) Ceramic honeycomb as of bio-based phenolic acids. ACS Sustain Chem Eng 7(19):16364– support for covalent immobilization of laccase from Trametes versicolor and transformation of nuclear fast red. Appl Microbiol 49. Müller CR, Meiners I, Domínguez De P, María (2014) Highly Biotechnol 90(1):313–320 enantioselective tandem enzyme-organocatalyst crossed aldol reac- 33. Stroock Abraham SKWAA, Dertinger D, Mezic I, Stone HA, tions with acetaldehyde in deep-eutectic-solvents. RSC Adv 4(86): Whitesides GM (2002) Chaotic mixer for microchannels. Science 46097–46101 295(5555):647–652 50. Thomsen MS, Nidetzky B (2009) Coated-wall microreactor for 34. Grabner B, Schweiger AK, Gavric K, Kourist R, Gruber-Woelfler continuous biocatalytic transformations using immobilized en- H (2020) A chemo-enzymatic tandem reaction in a mixture of deep zymes. Biotechnol J 4(1):98–107 eutectic solvent and water in continuous flow. React Chem Eng 51. de Lathouder KM et al (2008) Polyethyleneimine (PEI) functional- 5(2):263–269 ized ceramic monoliths as enzyme carriers: Preparation and perfor- 35. Fath V, Kockmann N, Otto J, Röder T (2020) Self-optimising pro- mance. J Mol Catal B Enzym 50(1):20–27 cesses and real-time-optimisation of organic syntheses in a 52. Hollermann G, Dhekane R, Kroll S, Rezwan K (2017) microreactor system using Nelder-Mead and design of experiments. Functionalized porous ceramic microbeads as carriers in enzymatic React Chem Eng 5(7):1281–1299 tandem systems. Biochem Eng J 126:30–39 36. Sans V, Cronin L (2016) Towards dial-a-molecule by integrating 53. Halfer T, Rei A, Ciacchi LC, Treccani L, Rezwan K (2014) continuous flow, analytics and self-optimisation. Chem Soc Rev Selective covalent immobilization of ferritin on alumina. 45(8):2032–2043. Royal Society of Chemistry Biointerphases 9(3):1–9 37. Waldron C, Pankajakshan A, Quaglio M, Cao E, Galvanin F, Gavriilidis A (2019) An autonomous microreactor platform for 54. Kim D, Herr AE (2013) Protein immobilization techniques for the rapid identification of kinetic models. React Chem Eng 4(9): microfluidic assays. Biomicrofluidics 7:1–47 1623–1636 55. Ma H et al (2018) Glutaraldehyde inactivation of enveloped DNA 38. Fabry DC, Sugiono E, Rueping M (2016) Online monitoring and viruses in the preparation of haemoglobin-based oxygen carriers. analysis for autonomous continuous flow self-optimizing reactor Artif Cells Nanomedicine Biotechnol 46(1):33–38 systems. React Chem Eng 1(2):129–133. Royal Society of 56. Jia F, Narasimhan B, Mallapragada S (2014) Materials-based strat- Chemistry egies for multi-enzyme immobilization and co-localization: A re- 39. MurrayPM, BellanyF,BenhamouL,Bučar DK, Tabor AB, view. Biotechnol Bioeng 111(2):209–222 Sheppard TD (2016) The application of design of experiments 57. Chen YX, Triola G, Waldmann H (2011) Bioorthogonal chemistry (DoE) reaction optimisation and solvent selection in the develop- for site-specific labeling and surface immobilization of proteins. ment of new synthetic chemistry. Org Biomol Chem 14(8):2373– Acc Chem Res 44(9):762–773 2384. Royal Society of Chemistry 58. Maier MC et al (2020) 3D printed reactors for synthesis of active 40. Gooding OW (2004) Process optimization using combinatorial de- pharmaceutical ingredients in continuous flow. Org Process Res sign principles: Parallel synthesis and design of experiment Dev 24(10):2197–2207 methods. Curr Opin Chem Biol 8(3):297–304. Elsevier Current 59. Bommarius AS, Riebel BR (2004) Biocatalysis. Wiley, Hoboken Trends 60. Kessler W (2006) Prozessanalytik- Strategien und Fallbeispiele aus 41. Weissman SA, Anderson NG (2014) Design of Experiments (DoE) der industriellen Praxis. Wiley- VCH Verlag GmbH & Co. KGaA, and process optimization. A review of recent publications. Org Weinheim Process Res Dev 19(11):1605–1633 42. Box GEP, Wilson KB (1951) On the experimental attainment of Publisher’snote Springer Nature remains neutral with regard to jurisdic- optimum conditions. J R Stat Soc Ser B 13(1):1–38 tional claims in published maps and institutional affiliations. 43. Borkowski JJ (1995) Spherical prediction-variance properties of central composite and Box—Behnken designs. Technometrics 37(4):399–410

Journal

Journal of Flow ChemistrySpringer Journals

Published: Sep 1, 2021

Keywords: 3D printing; Biocatalysis; Automation; Continuous flow; DoE.

References