Removeable reactor inset

Abstract

The present invention relates to removable insets used in model reactor systems, wherein in use said insets reflects the same fluid dynamic characteristics or predefined mixing characteristics in the model reactor as derived from a target reactor, to methods for producing such insets and to their use in ease scale-up, scale-down and process transfer to target reactors.

Description

REMOVEABLE REACTOR INSETField of the Invention

[0001] The present invention relates to the field of biotechnology production plants for producing valuable product, to a method to speed up the scale-up or scale-down of established processes to ensure process stability.Background Art

[0002] The scale-up of established biotechnological processes currently is timeconsuming and resource-intensive. Development of biotechnological processes usually occurs in small-scale systems with a volume of 1 liter or even smaller. For production the developed process needs to run in large-scale reactors with volumes of 2,000 L, 10,000 L or more. Large volumes mix inherently different - a stratification of nutrients, oxygen or CO2occurs - which leads to different environmental conditions for growing cells. These experience starvation or oversaturation which do not appear in small-scale systems at all, wherein the process components usually are mixed homogenous. Therefore, product quality or yield may differ which is only observable very late in process development and poses and immense risk of failure. Various strategies to solve or avoid the problem exist, but all are inadequate and have limits.

[0003] The traditional approach is to have a step-wise iteration with an increased volume on each step - e.g., factor 10 going from 1 L to 10 L, 100 L, to 1,000 L, and so forth. This approach is time-consuming and costly, especially if runs need to be repeated and the process adapted. Keeping the same width-to-height ratio or using power input of the stirrer as well as proportionate aeration helps but does not take any finer fluid dynamics properties into account.

[0004] W02020205611A1 provides a computational method for modeling a bioreactor with kinetics of metabolic fluxes and flux balance analyses to estimate the resulting yield. Since there is no universal metabolic cell model available no complete digital solution is available and existing models are limited in their predictive power.

[0005] Another methodology is to bring the cells into an environment that mimics the large scale production environment. Such approaches use a two vessel small scale system, where one reactor is mixed and supplied with nutrients, and the second reactor is not supplied with nutrients and can be either a mixed vessel, or a tubular reactor (Mayer F., 2023). As a complete system, these two-vessel systemshave the appropriate mixing time of large scale vessels but still fall short in adequately representing the conditions in large scale in terms of nutrient and oxygen distribution and are technically very complex to implement. The newest development is the use of a two vessel in one vessel system, by dividing the culture volume within a single bioreactor with a divider plate into two parts (Gaugler L. et al., 2022). This accomplishes the same as a two-reactor system in terms of mixing times but still is unable to accurately mimic the large scale nutrient and oxygen gradients.

[0006] WO2024133134 discloses a scalable physical reactor representation. It replaces the default vessel of reactor systems with a specially calculated and 3D printed one with wall structures. These wall structures influence the mixing characteristics to replicate the ones from a different reactor system. This approach requires the production of model-reactors exhibiting defined wall structures and the structures cannot be removed from the vessel.

[0007] This approach is limited to single-wall vessels and incompatible with double-jacked systems for heating / cooling, or systems where full vessel replacements are not possible due to other reasons, for instance integration of sensors into the reactor wall, or the use of stainless-steel pressure vessels as bioreactors. Therefore, there is still the need for time and resource effective method with less technical limitations for scaling a developed biotechnological process.Summary of invention

[0008] It is the object of the present invention to provide an inset to a standard reactor system which mimics hydrodynamic phenomena that are derived from a target reactor. The object is solved by the subject matter of the present invention.

[0009] According to the invention, there is provided a method for producing a reactor insert configured to enable specific defined fluid dynamics comprising: a. Defining target criteria based on objective functions, wherein said objective functions consist of values derived from computational methodology; b. Creating an automated algorithm for generating a number of at least 10 different reactor geometries using randomized structures on a predefined inset for a vessel retrieved from a library of structure creation methods (e.g. voxel based addition, plates, etc.); wherein thesestructures can include parametrized geometries (e.g., thickness of a plate, size of a torus); and optionally wherein the parameters of the geometry are part of the randomization. c. Determining the fluid dynamic characteristics using automated computational fluid dynamics for all generated structures; d. Ranking of all generated structures according to the objective function; e. Selecting the best fitting structures, wherein the number of selected structures is smaller than the number of the generated structures; f. Creating an automated algorithm to further randomly retract and add randomized structures to the selected best fitting structures to generate a new number of at least 10 randomized structures; g. Repeating of steps (c) to (f) iteratively until the target criteria of similarity between objective functions and characteristic of randomly generated reactor geometries is reached; and h. Producing the reactor inset with a randomized inner wall structure; and wherein said inset is removeable.

[0010] A further embodiment relates to a removeable inset for a model reactor, wherein said inset consists of a scaffold which exhibits random structures configured to enable specific defined fluid dynamics. The random structures may be comprised of elevations and depressions, or parametrized geometries wherein the parameters defining the geometry are randomly generated and optimized.

[0011] A further embodiment relates to the use of an inset as described herein as a model reactor for the transfer of a bioprocess from large to small scale and vice- versa, or from one large scale to another through two downscale mimics model reactors.

[0012] One embodiment of the invention relates to a method for scale-up of a bioprocess comprising of the following steps: a. Determining the process conditions of a large scale reactor; b. Using an algorithm to generate an inset for a small scale reactor with random structures reflecting the desired mixing behavior of the large scale reactor; c. Manufacturing the inset; d. Inserting the inset into a small scale reactor; e. Developing the bioprocess in the small scale reactor;f. Transferring the established bioprocess from the small scale reactor to the large scale reactor.

[0013] One embodiment of the invention relates to an inset being compatible with double jacket heating / cooling systems as well as being compatible with processes and sterilizations run under pressure through integration of the insert into commercially available stainless-steel reactors.

[0014] One embodiment of the invention relates to an inset being compatible with sensor technology that is housed on the reactor wall vessel, by being compatible with commercially available bioreactors equipped with such sensors (e.g., dot technology for the measurement of dissolved oxygen, carbon dioxide, or pH).Brief description of drawings

[0015] Fig. 1 depicts a conventional small scale reactor vessel with a 3D printed inset clamped on the baffles.

[0016] Fig. 2 shows structures created on the inset to change the hydrodynamic properties.

[0017] Fig. 3 depicts a conventional reactor vessel (250 ml) showing the fluid velocity (left) and the resulting flow profile (right).

[0018] Fig. 4 depicts a vessel with added inset with grown structure showing the fluid velocity (left) and the resulting flow profile (right). The flow is redirected and influenced by the structure changing the mixing behavior.

[0019] Fig. 5 depicts an inset without structures for a 3 I reactor fixated with a rubber ring at the top ring.

[0020] Fig. 6 depicts s scheme of a stepwise process upscale from lab-scale to large scale. On the left side, the traditional way is shown whereas on the right side the inset technology according to the invention is depicted.

[0021] Fig. 7 depicts s flow diagram reflecting the differences between traditional digital model based upscaling and the inset technology according to the invention.

[0022] Fig. 8 depicts a detailed decision parameter and iterative approach to generate a new inset design using random structure generation / subtraction and selection. At each generation (iteration) of the process, the reactors including insets are expected to conform more and more to the desired target values, i.e., the objective function.

[0023] Fig. 9 depicts the mixing time of each generation from generation 1 to generation 15 for reaching a mixing time of at least 20 seconds to berepresentative of mixing in large scale. X-Axis: Generation number, Y-Axis: Mixing time in seconds.

[0024] Fig. 10 depicts a reactor simulation with averaged velocity fields showing the flow patterns.

[0025] Fig. 11 depicts a detailed view of the flow patterns resulting from the inset structure (inset and sensors are hidden for better view). The flow patterns are massively changed by the inset in relation to the empty vessel.

[0026] Fig. 12 depicts that tracer distribution is hindered by the inset structure (sensors and empty inset hidden for better view), 1 second after tracer insertion (left), 2 seconds after insertion (middle), 3 seconds after insertion (right).

[0027] Fig. 13 depicts the mixing curves of different reactor geometries offering different mixing times and different mixing behavior in comparison to exponential washout behavior.

[0028] Fig. 14 depicts the generated structure from above. The colors visualize the velocity of the fluid, which is hindered by the created structures. The initial inset is hidden for better view.

[0029] Fig. 15 depicts a workflow starting from a large-scale target reactor followed by determining the fluid dynamic properties via CFD software to the final small lab model reactor using the calculated and manufactured inset exhibiting the same fluid dynamic properties as the large-scale target reactor.

[0030] Fig. 16 depicts the mixing times calculated until generation 35 of the inset creation run (x-Axis: Generation Number, y-Axis: Mixing time in seconds).

[0031] Fig. 17 depicts the rating of the calculated insets in each generation according to residence time in nutrient starved areas. This shows a huge difference to the mixing time diagram, since longer mixing times do not automatically produce longer residence times.

[0032] Fig. 18 depicts the mixing time distribution with an empty vessel (R8653), a pDS inset with 150 seconds mixing time (R8822) and a pDS inset with 240 seconds mixing time (R8884)

[0033] Fig. 19 depicts the maximum cell density for vessels with pDS inset in comparison to control vessels. The same maximum cell density was reached for control vessels or pDS vessels with longer mixing time.

[0034] Fig. 20 depicts lactate production which was more than doubled in vessels with pDS structure in comparison to the control vessels.

[0035] Fig. 21 depicts the reduced product titers in vessels with pDS insets in comparison to the control vessels.

[0036] Fig. 22 depicts examples of randomly generated voxel-based insets integrated into bioreactor vessels.

[0037] Fig. 23 depicts randomly generated variations of parametrized torus geometries influencing hydrodynamic behavior.

[0038] Fig. 24 depicts examples for a hybrid torus and voxel-based structure integrated into scale-down bioreactor insets.Description of Embodiments

[0039] The present invention provides a method for producing a reactor inset geometry with specific defined fluid dynamics comprising: a. Defining target criteria based on objective functions, wherein said objective functions consist of values derived from computational methodology; b. Creating an automated algorithm for generating a number of at least 10 different reactor inset geometries using randomized structures; c. Determining the fluid dynamic characteristics using automated computational fluid dynamics for all generated structures; d. Ranking of all generated structures according to the objective function; e. Selecting the best fitting structures, wherein the number of selected structures is smaller than the number of the generated structures; f. Creating an automated algorithm to further randomly retract and add randomized structures to the selected best fitting structures to generate a new number of at least 10 randomized structures; g. Repeating of steps (c) to (e) iteratively until the target criteria of similarity between objective functions and characteristic of randomly generated reactor geometries is reached; and h. Producing the reactor inset with a randomized inner wall structure; and wherein said inset is removeable.

[0040] The target criteria are based on objective functions retrieved from a target reactor. The target criteria are derived from the fluid dynamics of the target reactor. That means, the target criteria are set to reflect the values derived from an existing reactor.

[0041] As used herein, the term “objective function” refers to any function that formulates the objective. This objective function can have the target criteria of a minimum, maximum, a range or a specific value, and can be comprised of one or more constrains / equations that can be combined to one equation. The objective function can be as simple as a maximization / minimization goal of one parameter, for example the mixing time. The objective function can also include any number of target criteria in any combination, like a minimum mixing time combined with a specific kl_a.

[0042] The specific defined fluid dynamics may be determined by evaluation of the fluid dynamics of an existing reactor, or by defining optimized parameters by using computational methodologies. The fluid dynamics of an existing reactor may be determined by a Computational Fluid Dynamics (CFD) software, simple engineering equations, or deep learning Al derived CFD equivalent.

[0043] According to one embodiment of the invention, either the specific defined fluid dynamics of an existing large-scale reactor are determined and used or specified defined optimizing parameter are used. The specific defined optimized parameters are selected from either the group of process related parameters like mixing time, power input, etc. or are selected from the group of reactor related parameters like volume, surface area, and the like.

[0044] According to one embodiment of the invention the specific shape of the inset structure is generated based on the results obtained from the CFD software in terms of mixing time or component distribution calculated with the CFD software. According to one example, the terms of mixing time or component distribution of dissolved oxygen and / or nutrients are calculated with the CFD software.

[0045] One embodiment of the invention is directed to a removeable model reactor inset that reproduced the mixing behavior of an existing reactor. The model reactor inset comprises structural features configured to emulate the mixing properties of the existing reactor.

[0046] The model reactor inset may be a small-scale reactor inset or a mediumscale reactor inset, or even a large-scale reactor inset.

[0047] The automated algorithm allows the generation of different structure geometries of the inset by using randomized structures and the original inset retrieved from a library of structure elements (e.g., voxel based addition, plates,etc.) or by defining parametrized geometries (e.g. thickess of plates, size of a torus, etc.)

[0048] The main objective of the present invention is to enable the transfer of bioprocesses between different reactor systems - either from small lab-scale to large-scale during production scaling up, or between two large-scale systems during production scale-out. The invention addresses the challenges caused by different fluid dynamic characteristics of reactor systems by providing an inset for an existing reactor vessel. This inset contains structure produced by an automated design process, enabling the mixing properties of the reactor system to be adapted to those of the target reactor. The inset is manufactured, for example by 3D printing, and placed inside the reactor vessel. As a result, the reactor reproduces substantially the same process environment as the target system.

[0049] The method for producing a fitting inset comprises the following steps:• Simulating the target-reactor and extracting all necessary hydrodynamic or process characteristics or the definition of those characteristics by past experience and / or an educated guess;• Running the optimization loop by providing a base-inset and building on that in each iteration;• Creating a defined number of structure variations by o Removing a defined amount of structure randomly; o Adding a defined amount of structure randomly;• Running a computational fluid dynamics (CFD) simulation to extract the needed characteristics of the vessel including inset;• Ranking the created structure variations of the inset by their distance to the target-characteristics:• Choosing the best fitting ones for the next iteration:• Stopping the optimization loop after the target characteristics are reached.

[0050] A method of adapting the mixing behavior of a reactor comprising an inset with random structures that allow finely changing the resulting mixing properties is a new approach which adds a new degree of freedom to the currently available process parameters. Running a process with this 3D printed inset in a small-scale reactor still may leave the cell as a black box but will mimic the same environment they will experience in large-scale.Computational Fluid Dynamics (CFD)

[0051] Computational fluid dynamics (CFD) is the science of using computers to predict liquid and gas flows based on the governing equations of conservation of mass, momentum, and energy. CFD is also used in bioprocess engineering e.g., as modeling tool. Using CFD for geometric optimization is rare and only limited to parametric modeling without a finer grained approach. These parameters need to be defined and set up by a human and can only model specific properties without having the ability to make fine-grained changes.

[0052] For instance, CN112836444A describes an optimization method of dry anaerobic biogas mixing system based on CFD. The document describes the optimization of CFD modelling by changing different factors within the CFD model itself to obtain a more accurate representation of the CFD modelling of a dry anaerobic biogas mixing system. The use of an optimized high performance CFD modelling to simplify the calculation model and enabling fast evaluation of predefined structures is disclosed therein. Further, the evaluation of different predefined structures is described, but all references to “optimized model” refer to the optimization of the CFD modelling process, not to the inner geometry. The savings mentioned relate to avoiding the construction of small-scale reactors for testing flow properties, by replacing physical testing of flow properties by optimized CFD simulations. The geometries mentioned (e.g., nozzle pipe) or process parameters to be optimized e.g., gas stirring) are predefined and are evaluated, but the geometry is not optimized by automation. The described optimization of the CFD described in CN112836444A by using a lattice structure, different resolution of the lattice structure, boundary conditions and optimization to include multiphase systems through simplifications, is state of the art for modern CFD simulation software, such as used in the commercially available software package M-Star CFD

[0053] Additionally, some work on non-stirred vessels and static mixers was presented before, using CFD as investigation tool without automated random design. For instance, US2019 / 0388859 Al describes the CFD evaluation of static mixers for continuous flow catalytic reactors. The use of pre-defined static mixers comprised of a catalytic material and uses 3D printing to manufacture is disclosed therein but an automated algorithm to generate novel structures is not comprised. Like in CN112836444A, US2019 / 0388859 Al is not concerned with automaticallygenerated new geometries using any algorithm, as the geometries are predefined and CFD is only used for evaluation of the predefined static mixer.

[0054] In summary, the methods described for upscaling use physically generated data on multiple scales, to calibrate a model (either CFD or other models) and / or to digitally predict large scale behavior of processes derived from small scale data. The transfer from a small-scale process to a large-scale process is done by a digital model, trying to estimate the scale dependent changes.

[0055] This technologic field is well described by “Scale down simulators for metabolic analysis of large-scale bioprocesses” (Neubauer P, and Junne S., 2010) wherein state of the art imitation of large-scale reactors is described. The presented systems are limited to modelling or two-compartment reactor systems (Mayer et.al, 2023). These systems can either be composed of a two-vessel stirred system, or of a combination of one vessel and one plug-flow-reactor. Such systems do rely on two vessels and can therefore only be defined through a mixing time for the whole two-vessel system. The disadvantage of these systems is that they only have two zones, with limited mixing to each other which are not representative of true large-scale conditions and stratification in large scale bioreactors. Such a system has only two compartments, which both are individually well mixed, which means that no gradient of nutrients and / or dissolved gas is achievable with such systems. Additionally, it is a technical challenge to use, calibrate and develop these two-vessel systems.

[0056] Fig. 1 shows an example of a clamped inset in a 250 ml_ reactor vessel without inner wall structures. Fig. 2 shows an example of a clamped inset in a 250 mL reactor vessel with inner wall structures. Fig. 3 shows the corresponding velocity field and resulting flow profile. The same inset with grown structure in Fig. 2 and corresponding velocity field and flow profile is shown in Fig. 4.

[0057] The specific defined fluid dynamics may be determined by evaluation of the fluid dynamics of an existing reactor, or by defining optimized parameters. The fluid dynamics of an existing reactor may be determined by a Computational Fluid Dynamics (CFD) software.

[0058] According to one embodiment of the invention, the specific defined fluid dynamics of an existing large-scale reactor are determined and used or specified defined optimizing parameter are used. The specific defined optimized parameters are selected from either the group of process related parameters like mixing time,power input, residence times, or the like, or are selected from the group of reactor related parameters like volume, surface area, and the like.

[0059] The inset consists of a scaffold which exhibits random configured structures to enable specific defined fluid dynamics. The scaffold may be comprised of a cylindrical base body formed from rods, plates or other simple geometry. The randomly formed structures are molded onto the scaffold.

[0060] The specific created structures of the inset can be adapted to CFD characteristics like mixing time, residence time in nutrient depleted zones, dissolved oxygen, nutrient distribution, pH-Value, or the like. The inset can be created for small, medium or even large reactor vessels.

[0061] The inset exhibiting the specific structures may be produced by suitable means. For example, the inset may be produced by injection molding or by 3D printing.

[0062] Thus, one embodiment of the invention relates to an inset for a model reactor, wherein said inset exhibits random structures. The model reactor may be any reactor suitable to set up a process which than can be immediately transferred to the target reactor. Thus, the model reactor may be a small-scale reactor, a medium scale reactor, or even a large scale reactor.

[0063] The structure of the inset is configured to simulate the fluid dynamic characteristic of a given model reactor. This simulation allows the set-up of a bioprocess in a model reactor which upon meeting the target criteria is transferred to the target reactor.

[0064] The structure of the inset as described herein may exhibit elevations and depressions and does not comprise fully enclosed spaces and / or exhibit parts without elevation and depression structures for assembling inlets and outlets. Specifically, the resulting three-dimensional structure exhibits varying solid and void distributions, thereby creating distinct hydrodynamic patterns in the small- scale reactor.

[0065] The inset may be fixed to the reactor vessel by suitable means. For example, the inset may be fixed by clamping it on existing reactor vessel baffles or by a rubber-ring already incorporated in the inset design or even by being screwed to the reactor vessels.

[0066] The inset is specifically useful for scale-up or scale-down a process as well as for a transfer of an existing process to another reactor different in size.According to one embodiment, an established process running in one reactor is adapted in a way that after the transfer, the process also is smoothly running in the target reactor.

[0067] Using the inset in small-scale system for process scale-up may include the following steps• Calculating the target large scale hydrodynamic or process conditions;• Calculating the inner structures of an inset for the lab-scale system;• 3D printing of the inset having said inner structures;• Inserting the inset into the lab-scale vessel;• Running the process in the lab-scale system with mixing behavior adapted to the mixing behavior of the target large scale process.

[0068] The target large scale process conditions may either be derived from CFD or obtained from existing real runs.

[0069] The translation and evolution of the model reactor inset may be conducted by Computational Fluid Dynamics (CFD) software or the like.

[0070] According to one embodiment of the invention, the determined process condition is one or more selected from mixing time, mixing of nutrient feeds, shear rates, and mass transfer. Additionally, residence-times - e.g., in nutrient or oxygen depleted zones - CO2distribution or pH and temperature distribution are possible targets.

[0071] A further embodiment relates to the inset as described, wherein the inset may be placed into the target reactor and, once the bioprocess has commenced, may be withdrawn and subsequently reused in a different standard reactor.

[0072] The removable inset has several advantages over a full reactor replacement as described in WO2024133134 which has defined structures directly attached to the reactor wall. The inset according to the invention can be removed after a process cycle and reused in another reactor, reducing material and operational costs. The removable inset can be sterilized separately or replaced entirely, lowering the risk of cross-contamination between process transfer. The inset according to the invention are designed to fit in standard reactor geometries having the required additional equipment such as for example, stirring means, valves, in- and outlet, and the like.

[0073] The inset is compatible with double-jacket heating / cooling systems and does not influence the heat transfer into the vessel. The inset is also compatiblewith processes or sterilization procedures that use pressure, as it can be placed in commercially available stainless steel bioreactors. The inset is also compatible with sensor technologies that measure properties through the wall of the bioreactor (e.g., sensor dot technologies for the measurement of 02, C02and pH, or turbidity sensors). The inset technology therefore offers a number of technical advantageous features that are not achievable with complete reactor replacement technology as described in WO2024133134.

[0074] The inset achieves key compatibility features with existing bioreactors that the current state of the art are lacking: a. being compatible with existing double jacked heating / cooling systems; b. being compatible with processes that use pressure for sterilization or during bioprocessing by the possibility to integrate inserts into stainless steel pressure vessels; and / or c. being compatible with wall-mounted sensors of existing reactors e.g, dot technology for measuring O2, pH or CO2)Description of Embodiments

[0075] The Embodiments which follow are set forth to aid in the understanding of the invention but are not intended to and should not be construed to limit the scope of the invention in any way. The Examples do not include detailed descriptions of conventional methods. Such methods are well known to those of ordinary skill in the art.Small Scale Reactor - Mixing Time

[0076] Moving an established biotechnological process from one bioreactor to another is always a challenge. The process is heavily depending on the mixing characteristics of the reactor since this may result in inhomogeneity.Homogeneities are important environment properties for the growing cells, like nutrient or oxygen distribution. A new bioprocess is usually established in lab-scale of about 1 L to 3 L or even smaller. An established and optimized process needs to be transferred to large production reactors of about several hundreds of liters. Large volumes have physically different hydrodynamics - they have longer mixing times and produce stratifications for e.g, nutrients, oxygen, pH, or temperature. Even different large scall production reactors, for example at different manufacturing sites, may have different mixing capabilities due to different stirrer setups or width-to-height ratios. Transferring a process is tricky and often needsadaptions during scale-up and sometimes results in loss of product quality, lower yield or even failed scale-up.

[0077] The mixing characteristics of a small volume present in a small lab-scale reactor can be changed through an inset in the small lab-scale reactor. Mixing time can be increased, zones with different flow-speeds could be established and residence time in nutrient or oxygen oversaturated or depleted areas may be created. The stratifications that usually occur from top to bottom in a large-scale production reactor can be reproduced by the structure of the reactor inset. Thus, later-stage problems during scale-up can be observed already during process development in the laboratory scale and actively targeted.

[0078] Such speed up and de-risked scale-ups help to reach a number of sustainable development goals besides reducing time and cost. The traditional stepwise scale-up, as shown in Fig. 6, with ever increasing reactor volumes is replaced by a number of lab-scale runs with the inset to replace the in between steps and then straight jump to the large production volume. This reduces the used resources immensely and avoids having large amounts of genetically modified micro-organisms and chemical waste. This is especially true in the areas of agrifood, biopharmaceutical, cosmetics or chemical industries.

[0079] Another approach that tries to avoid the stepwise scale-up is the use of prediction models as discussed before. The main issue of these models is that they are host- and product-specific. This means that for each process data from the large scale is necessary to recalibrate the model, which limits the time and cost savings. An overview of the state of the art procedure and the present invention is shown in Fig. 7.

[0080] Direct upscaling from lab scale to production scale needs therefore to be host and product independent to be usable independent of the used organism and exact process conditions. The inset according to the invention allows to downscale or upscale the physical characteristics of the target reactor, wherein the cells used are put it into the same environment as in the target reactor, independent from the cell type. That means, the type of cells used in the reactor is not taken into account when designing the inset. Purely, the physical parameters are taken into account for designing the random structures of the inset.

[0081] Scale-up of fermentation processes in the biotech industry is a crucial step in bringing bioprocess and bioproduct innovations to commercialization and lifescience will be the cornerstone of a future cyclic economy. This includes CO2capture technologies, bioplastics, and novel foods as well as biopharmaceuticals like cancer therapy, vaccines, and gene therapy. As mentioned above, the upscaling step is notorious for performance losses and delays in process development. The scale-down approach has been advocated as a smart way to minimize these issues but current used technologies are technically complex (2-vessel systems) or very limited by human design instead of automated design. The presented approach consists of interconnected steps to overcome the shortcomings of the state of the art:• Detailed analysis of the hydrodynamic or process characteristics in a (large-scale) target reactor;• Computational representation of the dynamic fluid characteristic;• Translation of this dynamic fluid characteristic to a (laboratory-scale) reactor using the presented automated structure generation of an inset for a (small-scale) bioreactor vessel (see Fig. 2; the main generation flow is shown in Fig. 8);• Production of said (small-scale) model vessel inset exhibiting complex inner wall structures ;• Developing the process under conditions representative for the (large- scale) target reactor; and finally• Translation of the successful findings to the (large-scale) target reactor directly without any necessary intermediate steps.

[0082] The method provided by the present invention avoids the time consuming and resource demanding upscaling in biotechnological manufacturing wherein a new process is first established in a small experimental reactor, then transferred to larger reactors in several sequential volume increasing steps to final production volumes of up to several cubic meters.

[0083] The target reactor may be a large-scale production reactor used in a plant to produce various desired biotechnological products. The large-scale production reactor may have a volume of 10 L and more than 500 L up to several cubic meters. The model reactor may be a small-scale lab reactor. The small-scale model reactor may have a volume of 0.1 L, 0.5 L, 1.0 L 1.5 L, 2.0 L, 5.0, and up to 10.0 L or larger.

[0084] Specifically, the flow-physical properties differ significantly in reactors of different volume and shape. For example, the distribution of nutrients and themixing time are mainly depending on the volume and shape of the reactor used. Every time a process developed in one scale (e.g., laboratory scale) is transferred to a reactor of different size (e.g, production scale) plenty of time and resources are required to redevelop the process in the target reactor with the goal of achieving the same results, which however are not always being met.

[0085] This embodiment targets a mixing time representative for a large-scale reactor running a microbial process (>12 m3) which typically have mixing times of 10-50 seconds. The general decision scheme for random structure generation to yield a specific hydrodynamic behavior (in this case a mixing time > 15 seconds) is shown in Fig. 4. The loop of structure generation and removal is performed as long as required to yield the specific target goal. In this case, the algorithm will automatically generate new inset geometries based on the last iteration, and automatically evaluate them by CFD (e.g., determining their mixing time) until an inset structure is generated that exhibits a mixing time >15 seconds.

[0086] Target lab-scale system is a 3 L reactor running a microbial setup. The vessel without structures but including necessary sensors and tubes has a calculated mixing time of 2.0-2.6 seconds. This creates a much more homogeneous environment than the large reactor, which has a multiple of the small scale mixing time.

[0087] Using this methodology, the mixing time necessarily will increase with each generation, if suitable parameters for the amount and kind of structure generated and deleted in each iteration are selected. For this specific embodiment of the invention, an inset for a 3 L bioreactor was designed using the described methodology, with the target goal of at least 20 seconds mixing time. Bioreactors of this size (3 L) usually have mixing times of 2-3 seconds, which is by way too fast to be representative of large-scale reactors. Fig. 9 shows the mixing times of each generation of randomly generated structures using the described methodology and the successful increase of mixing time with each iteration is shown. These progressively longer mixing times are achieved by the random structure addition and subtraction to the 3D structure of the bioreactor vessel inset, and with each generation, the best performing structures (according to the objective function) are taken as the base for the next generation of structure addition / subtraction.

[0088] The mixing time increase depends on the addition of structures, hindering the mixing of any added substance (e.g, nutrients) as they break the efficientmixing of small-scale reactors. The nature and design of inset structures in this embodiment is selected to be purely random, and the mixing behavior can be easily visualized by the addition of a scalar into the CFD simulation from the top of the reactor, simulating the addition of nutrients by a feeding line at the top of the reactor.

[0089] Fig. 11 shows the created inset structure, whereas Fig. 11 shows the detailed flow pattern. The resulting flow characteristic is massively changed by the inset. Fig. 12 shows the distribution of a tracer in one of the reactor structures generated in this embodiment that showed more than 20 seconds mixing time. This behavior is not possible in conventional bioreactors, as the reactor would be homogenously mixed after 2-3 seconds in such small-scale reactors. Only the iterative random structure approach can offer this kind of tracer distribution representative of large-scale reactors.

[0090] The difference between a single plate separating one vessel as described in the background (Gaugler L. et al., 2022), or the two-vessel system can also be visualized by the resulting standard deviation of tracer in the bioreactor over time. For a two-vessel system as well as for a bioreactor separated in two parts by a plate with holes, the resulting curve will always resemble an exponential washout curve. This exponential washout curve is not what is seen in a large bioreactor, as this bioreactor is comprised of different zones being differently mixed with the main bulk of the liquid in the bioreactor, resulting in a more complex nutrient mixing over time (Fig. 11). The mixing curves represented by the standard deviation of tracer over time shown offer behavior from quasi-exponential washout behavior (lower curves) to a more complex mixing behavior that is potentially more suitable as scale-down bioreactor for process development and more representative of how mixing curves look like in large scale bioreactors. The presented methodology can either use the mixing time as an objective function, or it can use the resulting mixing curve as presented in Fig. 13 as objective function to produce a physical small scale reactor that more accurately represents the conditions in large scale bioreactors.

[0091] In this embodiment of the invention, the reactor was computationally evaluated using M-Star CFD, but this evaluation can of course be done with any suitable CFD modelling software, such as for examples Siemens StarCCM+ or open-source solutions like OpenFoam. The basic idea of CFD is to solve theequations of motion of the fluid in the reactor (Ze., the Navier-Stokes equations). These equations describe conservation of mass and the balance of momentum in the fluid. In particular, the momentum balance is of a nonlinear nature, which prohibits analytical solutions for virtually all practical flows. This requires discretizing the mathematical equations and solving them numerically. When scaling up, the flow becomes more turbulent and a feature of turbulent flows is a wide span of scales both in space and in time leading to significant mixing differences in small and large-scale reactors.

[0092] Histograms of well / badly mixed areas as well as all other values or series that are calculated by the CFD can be used as objective function according to the invention for determining if large and small-scale reactors exhibit the same flow characteristics. According to the invention any target-property may be used if a rating of the CFD target property can be calculated. Even combinations of properties with rated properties could be used, such as a mixing time >15 seconds, while simultaneously targeting a certain kl_a gas transfer coefficient.

[0093] The created inset structures can leave out parts that need to be free for inlets and outlets such as sensors or pipes. These are fed into the optimization algorithm, whereas these areas are left out (see Fig. 13). Additionally, only structures that are valid and can actually be produced, e.g., by 3D printing, are created without any closed areas and closed in bubbles. For one embodiment according to the invention, the inset was created for a 3 L volume bioreactor vessel excluding sensors, probes, tubes, sparger and stirrer.

[0094] According to one embodiment of the invention the vessel inset is a physical downscale model reflecting large scale hydrodynamic behavior. Such reactor insets therefore are host and product independent. The problem of a missing suitably detailed model for cell behavior that is the road block for digital model predictive power for the traditional approach (Fig. 6, left side) is not applicable for the present invention, as no modelling of cells is done. The cells used are transferred into a physical environment representative of large-scale production reactors, allowing the direct upscaling of process parameters developed in small scale, to large scale production without the loss of performance (Fig. 6, right side)

[0095] Finally, after the generation of the inner wall geometry of the small-scale reactor inset, representative of large scale mixing behavior, the reactor vessel inset may be produced, e.g, by 3D printing, as the digital file to do so is alreadyavailable. 3D printing is also the only manufacturing technology that will be capable of accurate and fast manufacture of the complex geometries resulting from the methodology as described herein. The production of the vessel inset can be easily done by 3D printing. Since the algorithm can ensure an inset ready for 3D printing, the physical downscale reactor can be realized by any 3D printing technology, e.g., be it laser sintering, FDM, DLP, LCA or any other methodology, as well as any available material, like any thermoplastics, (e.g., PLA, ABS) but also resins, medical resins, glass and ceramic resins and any sintering material from plastics to metals or the like may be used. The interconnection of manufacturing with the methodology of generating the geometry of the reactor is depicted in Fig. 15.

[0096] Scaling up a biotechnological process that is developed in lab-scale e.g, 1 L or even smaller) to a large production volume of 2,000 L and up to 50,000 L is a hard task. Several approaches mentioned earlier above like a stepwise progression with ever increasing volume while adapting process parameters is one of them, but very tedious and time and resource consuming. The perfect downscaled reactor having an inset exhibiting a scaffolding exhibiting specific structures allows a simulation of the mixing characteristic of a large-scale reactor in said small scale reactor. The mixing time was previously established as a good indicator of the reactor characteristic. Another approach is to look at the lifelines and environment that cells would perceive. The cells are not directly affected by mixing time, but by the inhomogeneous distribution of nutrients or oxygen. Especially the length of time cells stay in over or undersaturated areas have an effect on the organisms’ metabolism. Residence in nutrient depleted zones is generally longer in larger volumes and shorter in small volumes due to their better homogeneity through mixing time. Microbial processes usually run nutrient limited which always creates depleted zones. However, in small reactors cells have shorter stays in such zones.

[0097] An optimization of the vessel inset according to residence time in undersaturated area is done by using 1,000 probes in the simulation the bioreactor that track time in nutrient depleted zones. The nutrient itself is modeled by using a fluid tracer that is continuously fed near the top of the surface. The uptake by the cells is modeled by using a constant value (simulating a homogeneous distribution of the cells). The amount of nutrient inserted is 2 / 3 of the total possible consumption of the cells, which is a common setting. For evaluation, a large-scalereactor with a volume of 54 m3is used, representing a typical setup for a microbial process. A small-scale target system with 3 L volume is used as starting point.

[0098] By running the large reactor simulation a mixing time of 65 seconds is established. An average residence of 3.9 seconds in totally nutrient depleted regions is calculated with a standard deviation of 5.69 seconds.

[0099] The small-scale system has a calculated mixing time of 9 seconds with an average residence time of 0.96 seconds in totally nutrient depleted regions with a standard deviation of 0.934 seconds showing that there exist nutrient depleted zones through the nutrient limitation, but the average stay is much shorter in the small-scale reactor. The longer residence time in the large reactor may introduce metabolism changes, product quality or yield reduction. To get a prediction of the problems arising in the large reactor already in small scale reactor, an inset creation for the small-scale reactor is done by optimizing the starvation time residence in the large-scale reactor. The base-insert is created manually and structure is added / removed randomly by the mutation algorithm to the inner wall of the modulated inset.

[0100] Fig. 16 shows the optimization of 35 generations. The mixing time is not used as rating, whereas Fig. 17 shows the rating by residence time in nutrient starved areas. The difference between those two is clearly visible - a longer mixing time does not automatically create a longer starvation time. The best fitting residence time was achieved with an inset generating a mixing time of 56 seconds, a residence time in totally starved areas of 3.0 seconds and a standard deviation of 3.8 seconds. This created much longer residence times in comparison to the original small reactor with an average of 0.9 seconds much nearer to the target. This resembles the conditions for the cells in the large-scale reactor much better, running a process with this insert allows to see the large-scale effects on the metabolism already in the small scale.Example 1 - Scaling Down a 12,000 L Cell Culture Process

[0101] The effects of large volumes on CHO cells are well documented in publications. A good target is a production setup with 12,000 L - CHO large-scale effects are published by Gaugler L. et al., 2024re.

[0102] The issues arising areFormation of gradients for oxygen / CO2- these cannot be measured directly;Prolonged lag phase for cells to adjust to different conditions;Increased cell specific lactate production due to hypoxia and cell stress, andLower product titer than in a homogeneous small-scale environment.

[0103] To calculate the necessary perfect downscale insets the target 12.000 L reactor with a cell culture setup was simulated, resulting in a mixing time of 150- 180 seconds.

[0104] To achieve a good setup, a lab scale bioreactor system with a vessel working volume of II was used that allows to have 4 runs at the same time allowing to run exactly the same process using- The original glass vessel as control run;- A vessel made from a polymer with the same geometry as the original glass vessel;- A glass vessel with a perfect DownScale inset made from PA12 Nylon with structure resulting in a mixing time of 150 seconds; and- A glass vessel with a perfect DownScale inset made from PA12 Nylon with structure resulting in a mixing time of 240 seconds.

[0105] The setup included the use of the standardized NISTCHO cell line (Test Material 10197). The cultivation was according to recommendations with a standard fed-batch. The runs were done with 1 L model reactor volume. As analytics was chosen:Metabolites (Lactate, Glucose);- Cell number and viabilities; andProduct concentration (mAb)

[0106] Regarding cell growth and lag phase the results wereNo effects of the material: There was no difference between the glass vessel and the control vessel with the glass vessel geometry made with a polymer;No cell loss during lag phase;- A clearly prolonged lag phase of about 36 hours for the two pDS inset vessels;Little to no impact on the highest viable cell density; andDependence of the lag phase on the mixing time of the pDS inset - the 240 seconds inset produced a clearly longer lag phase than the 150 seconds inset.

[0107] The effects of the inset on lactate production, similar to the target large scale was clear.In vessels with the structured insets, the lactate production was doubled, and it correlated with the mixing time.- There was no impact on the lactate production due to the material - the control vessels behaved the same.

[0108] An additional factor is the decreased product titer by about 30% using the inset. The product concentration correlates strongly with the mixing time provided. Other performance indicators like highest achieved viable cell density and viability where in range. Process parameters for all 4 reactor runs were equal.Example 2 - Random Voxel-based Insert Structures

[0109] An inset structure for a scale-down bioreactor was generated by randomized voxel occupation within a defined volume envelope. The resulting three-dimensional structures exhibited varying solid and void distributions, thereby creating distinct hydrodynamic patterns in the small-scale reactor. A plurality of such structures was produced and subsequently tested by computational fluid dynamics (CFD) simulations to assess mixing performance, shear distribution, and flow heterogeneity. Optimized variants were selected according to predetermined hydrodynamic criteria for each generation, in this example, for maximization of mixing time. Figure 22 shows the generational variations from one generation to the next and progressively finer and more elaborate structures “growing” on the initial surface of the inset geometry.Example 3 - Parametrized Torus Geometry Around Stirring Element as reactor inset

[0110] In another embodiment, the randomly varied element was a parametrized torus fused to the inset in proximity to the stirring element. The torus geometry was defined by parameters including major radius, minor radius, axial position, height, and the number and size of cut-outs. A set of randomized parameter combinations was generated, and each geometry was automatically evaluated byCFD. Hydrodynamic optimization led to selection of torus geometries that achieved high mixing times and high kLA, while providing openings for sensor and sparger insertion, stirring element insertion (through a two-part assembly) as well as clearance for stirrer and temperature well (see Fig. 23).Example 4 - Mixing of parametrized geometries and random voxel-based geometries

[0111] Parametrized geometrical elements were combined with voxel-based random structures within the same inset. The torus geometry was defined by parameters including major radius, minor radius, axial position and height. Within and around the torus, a voxelized design space was populated with randomly generated solid and void regions. The combined structures were subjected to automated CFD evaluation, and variants were selected that provided the desirable mixing conditions in terms of global mixing time and mixing time distribution. Fig. 24 shows such final combined insets with both features of parametrized geometry and random voxelated geometry.ReferencesMayer F. et al., Computational fluid dynamics simulation improves the design and characterization of a plug-flow-type scale-down reactor for microbial cultivation processes. Biotechnol J. 2023 Jan;18(l)Gaugler L. et al., Mast Y, Fitschen J, Hofmann S, Schluter M, Takers R. Scalingdown biopharmaceutical production processes via a single multi-compartment bioreactor (SMCB). Eng Life Sci. 2022 Mar 14;23(1)Neubauer P, and Junne S. Scale-down simulators for metabolic analysis of large- scale bioprocesses. Curr Opin Biotechnol. 2010 Feb;21(l):114-21 Gaugler L, et al., Mimicking CHO large-scale effects in the single multicompartment bioreactor: A new approach to access scale-up behavior.Biotechnol Bioeng. 2024 Apr;121(4):1244-1256

Claims

Claims1. A method for producing a reactor inset configured to enable specific defined fluid dynamics comprising:(a) Defining target criteria based on objective functions, wherein said objective functions consist of values derived from computational methodology;(b) Creating an automated algorithm for generating a number of at least 10 different structured insets using randomized structures retrieved from a library of structure elements;(c) Determining the fluid dynamic characteristics using automated computational fluid dynamics for all generated structures;(d) Ranking of all generated structures according to the objective function;(e) Selecting the best fitting structures, wherein the number of selected structures is smaller than the number of the generated structures;(f) Creating an automated algorithm to further randomly retract and add randomized structures to the selected best fitting structures to generate a new number of at least 10 different structured insets;(g) Repeating of steps (c) to (f) iteratively until the target criteria of similarity between objective functions and characteristic of randomly generated structured insets and their characteristics in the simulation is reached; and(h) Producing the reactor inset with said randomized structures; and wherein said inset is removeable.

2. The method of claim 1, wherein the target criteria comprise fluid dynamic data derived from a target reactor.

3. The method of claim 1 or 2, wherein the computational methodology is any one of computational fluid dynamics, simple engineering equations, and deep learning Al derived CFD equivalent.

4. The method of any one of claims 1 to 3, wherein at least one objective function is a target value derived from a target reactor.

5. The method of claim 4, wherein the target value is any one of kl_a, mixing time, and / or nutrient distribution.

6. The method of claim 4, wherein the objective function is an optimum in any of the hydrodynamic characteristics.

7. The method of any one of claims 1 to 6, wherein the generated structured inset is produced by 3D printing.

8. An inset for a reactor system, wherein said inset consists of a scaffold exhibiting randomized structures which mimic predefined fluid dynamics, and wherein said inset is removable attached to the reactor system.

9. The inset of claim 8, wherein the randomized structures of the inset are configered to simulate the fluid dynamic characteristic of a target reactor.

10. The inset of claim 8 or 9, wherein the randomized structures of the inset does not comprise fully enclosed spaces and / or exhibit parts without random structures for assembling inlets and outlets.

11. The inset of any one of claims 8 to 10, wherein said inset is 3D printed.

12. Use of an inset of any one of claims 8 to 11 in a model reactor for the establishing a bioprocess to be transferred to production unit.

13. The use to claim 12, wherein the process transfer is intended from a small scale reactor to a large scale reactor, or from a large scale to a large scale reactor.

14. A method for scale-up of a bioprocess comprising of the following steps:(a) Determining the process conditions of a large scale bioreactor;(b) Using an algorithm to generate an inset for a small scale reactor with random structures reflecting the desired process conditions of the large scale reactor;(c) Manufacturing the inset;(d) Inserting the inset into a small scale reactor;(e) Establishing the bioprocess in the small scale reactor;(f) Transferring the established bioprocess from the small scale reactor to the large scale reactor.

15. The method of claim 14, wherein the determined process condition is one or more selected from mixing time, mixing of nutrient feeds, shear rates, oxygen distribution, nutrient distribution, and mass transfer.

Images