Jianye Xia,Guan Wang,Meng Fan,Min Chen,Zeyu Wang,Yingping Zhuang*

State Key Laboratory of Bioreactor Engineering,East China University of Science and Technology,Shanghai 200237,China

ABSTRACT The production capability of a fermentation process is predominately determined by individual strains,which ultimately affected ultimately by interactions between the scale-dependent flow field developed within bioreactors and the physiological response of these strains.Interpreting these complicated interactions is key for better understanding the scale-up of the fermentation process.We review these two aspects and address progress in strategies for scaling up fermentation processes.A perspective on how to incorporate the multiomics big data into the scale-up strategy is presented to improve the design and operation of industrial fermentation processes.

Keywords:Fermentation Scale-up Bioreactors Gas–liquid flow Kinetics

1.Introduction

Industrial fermentation utilizes microorganisms to produce various products,including antibiotics [1],organic acids [2],pharmaceutical proteins[3],and industrial enzymes[4],etc.It is estimated that the biotechnology products cover one-third of the worldwide market,with a value of more than 300 million USD [5].Industrial fermentation is advantageous because it involves relatively moderate pressure and temperature,and is more sustainable in terms of the use of natural resources,compared to chemical reaction processes.Thus,it has attracted considerable interest and promoted the emergence of a bio-economy.Furthermore,with the progress in metabolic engineering and synthetic biology,it is relatively easier to obtain engineered cell factories with improved product yield,even producing new products[6,7].High-throughput screening techniques have been developing at great speed.However,process development remains at a lag (3–10 years),and the cost of scaling up the process from lab-scale to industrial production is generally on the order of 100 million USD to 1 billion USD [8].

Scaling up newly developed lab-scale fermentation processes is hampered by a poor understanding of the principles that govern the scale-up requirements [8,9].Empirical approaches based on principle of similarity and dimensional analysis for scale-up are always implemented by the general principle of maintaining specific parameters as constants during the scale-up(e.g.,specific power consumption rate,oxygen transfer coefficient,impeller tip speed,mixing time,etc.) [10].It has been noticed that scaling up a fermentation process from a shake flask to an industrial fermenter consists of challenges [11,12],as non-homogeneous conditions in industrial fermenters may cause negative effects on strains.Even though it is common to keep specific parameters constant during scale-up in practice,it is necessary to find the critical influential parameters,and it is impossible to keep those properties (such as flow dynamics or fluid kinetics) similar or simultaneous at different scales [11].The interaction of two aspects affecting the scaleup process should be studied carefully for successful scaling up.One aspect is the flow field condition in bioreactors,which is determined by many parameters,including geometric configurations,fluid rheology and rotation speed.The other aspect is the physiological response of microbial cells,which is determined by their genetic background(genome)and complicated metabolism regulation systems.We analyzed the relationship between the flow field and the physiology of cells in both nature and in a bioreactor,as summarized in Fig.1.

This review is mainly organized to cover two aspects.First,we discuss flow field issues when scaling up,in which mixing efficiency,mass transfer coefficient and shear forces and their influence on fermentation are discussed.Following that,the cell physiological response to the flow field is reviewed,as arein vivokinetic studies with integration to the flow in bioreactors.This is with a view to understanding the scale-up process.Finally,a short perspective is given on how multiomics data can be used to better understand the scale-up process and what can be discovered with the help of big data on bioprocesses.

Fig.1.Flow field changes have a large impact on the physiology of different life forms (plant,animal or microbe).

2.Flow Field in Different Scale Bioreactors

Stirred tank reactors are the most widely used bioreactors for industrial fermentation,due to their relatively simple applicability and optimized design principles inherited from chemical engineering.However,the flow field generated in a stirred tank bioreactor covers a large range of flow regimes from laminar to transient through to turbulent,depending on the operating conditions.For this kind of bioreactor,lots of work has been done to study the generated flow field.Three issues influence fermentation scale-up:mixing efficiency,gas–liquid mass transfer capacity and shear forces that are generated.The following subsections review these three parts and are presented individually.

2.1.Mixing in the stirred tank bioreactor

The flow field generated by the impeller plays a considerable role on mixing.The mixing performance for various impeller configurations has been studied thoroughly [13–15].In general,the axial flow impellers have higher mixing efficiency while reducing energy consumption compared to radial flow impellers.Despite having a different impeller type,the bioreactor scale greatly affects the mixing performance,the mixing time ranges of different bioreactor scale are summarized in Table 2 of Laraet al.[16].In stirred tank bioreactors,dimensionless mixing time is always used as an important parameter.It is found that the dimensionless mixing time keeps constant when the flow field is under a turbulent flow regime.

Mixing greatly influences the performance of industrial-scale bioreactors.It influences many aspects of the bioprocess.For example,poor mixing may results in a substrate concentration gradient in the bioreactor for fed-batch fermentation,an oxygen concentration gradient for aerobic fermentation,or a temperature gradient for heat exchange during fermentation,or pH gradients.All these heterogeneous distributions of various parameters may cause undesired effects on the metabolism of the microorganisms,including overflow fermentation,mixed acid fermentation,lower biomass yield or lower product yield and titer.

Mixing is a complicated process.To describe it quantitatively,both circulation time (tc) and mixing time (tm) were defined and measured in experiments [17].Based on both the bulking flow model and the turbulence model,a similar correlation for predicting the mixing time has been proposed as follows:

In this equation,tmis the mixing time(with unit of[s]),Nis the rotation speed of impellers(unit of r·s-1),NPis the power number of the impeller (without unit),T,D,andHLare diameter of tank,diameter of impeller and liquid height (unit of m),respectively,andaandbare model constants,for which different values were given in [18]and [17].It has also been reported that mixing time under aeration is almost the same as that without a gassing condition [18].Following the ‘predicting’ equation,mixing times in different scale stirred tank bioreactors were estimated and summarized in Table 1 of [18],which ranges from 15 s in 10 L and～200 s in 1000 m3.The large discrepancy in mixing time among different bioreactor scale makes the scale-up of the fermentation process difficult.Details of the physiological response of microbes are summarized in Section 3.1.

2.2.Gas-liquid mass transfer in stirred tank bioreactor

Due to the low solubility of oxygen in the fermentation process,it should be supplied continuously during the aerobic fermentation process with limited gas–liquid interface mass transfer capacity.As the bioreactor scale increases,the volumetric power input will decrease,which directly results in a lower liquid mass transfer coefficient.If the total oxygen transfer rate of the bioreactor cannot support the oxygen consumption of microbes cultured in the bioreactor,there will be oxygen limited metabolism responses.Considering this,holding the volumetric oxygen transfer coefficient constant with different scale is always treated as essential in scale-up criteria.This is the case especially for microbes with high oxygen consumption capacities,e.g.,with high cell density cultures ofE.colior yeast.

Garcia-Ochoa and Gomez[19]offered a thorough review of the oxygen transfer rate in bioreactors,to determine the oxygen supply capacity of a bioreactor,in which a key parameter for aerobic fermentation scale-up is the volumetric oxygen transfer coefficient,kLa.The higher thekLa,the higher the bioreactor’s oxygen supply capacity.The volumetric oxygen transfer coefficient,kLaconsists of two aspects:the liquid film transfer coefficientkL,and the specific mass transfer area between bubbles and liquid.It is difficult to separately measure either of these two parameters.Their combination,kLais measured instead.Various methods can be used to measure thekLavalue,including physical[20,21],chemical[22]and the dynamic method [23].Among these methods,a dynamic method is recommended as this method can be applied directly in a fermentation process (see Fig.5 of [19]),the dynamic dissolved oxygen level in the broth was measured along with time by turning off the air for a duration and then turning it back on.The DO profiles contain both oxygen uptake rate of cells and the corresponding in situkLavalue.However,the response time of the DO probe should be taken into consideration when estimatingkLa.The following equation(adapted from Eq.(21)in[19])can be used to measurekLausing the dynamic method:

where DO is the dissolved oxygen level at timet,DO0represents the initial DO value,τpis the response time of the DO probe[s]andkLais the volumetric oxygen transfer coefficient.

There have been many empirical equations derived for predicting thekLavalue in the stirred tank bioreactors.Tables 3 and 4 of[19]summarized available empirical correlations forkLaestimation for both Newtonian and non-Newtonian fluids.ThekLavalue at different bioreactor scales is summarized in Table 1.Computa-tional fluid dynamics has been used to predict the mass transfer coefficient for different scale bioreactors,for example in a shaking flask [27](kLa=166 h-1for an unbaffled flask and 321 h-1for a baffled flask) and in a lab-scale bioreactor [28](kLa=250 h-1in 50L stirred tank bioreactor).

Table 1 Classical kLa value at different scale bioreactors

2.3.Shear force generated in the stirred tank bioreactor

Shear forces generated in the bioreactor are very important for filamentous microorganisms,as they influence the morphology of the cell.Freely dispersed mycelia,clumps and pellets are the three main morphologies for filamentous microorganisms.Their fractions are mainly determined by the shear force in the flow field.Different morphologies can have direct effects on the productivity of specific products,For example,avermectin production byStreptomyces avermitilisprefers the pellet format [29],which is the format also preferred for citrate fermentation byAspergillus niger[30],while the free mycelia is the preferred format in the production of penicillin fromPenicillium chrysogenum[31].The morphology may influence the mixing and the mass transfer in the flow field by changing the rheology of the broth,e.g.free mycelia with high biomass concentration may cause high viscosity,thus resulting in lower gas–liquid mass transfer and poor mixing properties [32].The pellet form may face another issue,which is that mass transfer in the pellet center with high biomass density will be limited if the diameter of the pellet is larger than 400 μm [33].

To investigate the shear force in the bioreactor,CFD tools were utilized [28,34].Shear forces generated in the bioreactor are evaluated by the shear strain,which is determined by the product of viscosity and shear rate.Quantitative evaluation of critical shear forces that are harmful to shear sensitive cells is an important feature for bioreactor design and scale-up.Different experimental methods and devices have been designed to generate specific shear forces to study their effect on cells [35].The shear force issue is always a matter for scaling up of animal cell cultures because of the high cost and risk in adopting a trial and error method for the scale-up.A novel method based on CFD simulation was proposed for this concern[36].A valid operational 3D space with three shear force parameters(Fig.2)was formed for guiding the scale up process,which reduced the high risk for scaling up significantly.Another study integrated the shear force encountered by plant cellCarthamus tinctorius L.along with its trajectory in a stirred tank bioreactor [37],which was used for its scale up.Pawar researched shear strain rate (SSR) in a 17 L airlift bioreactor by CFD analysis and proposed that microalgae cells were found in suspension for the superficial gas velocities of 0.02–0.04 m·s-1,which experienced an average shear of 23.52–44.56 s-1,far below the critical limit of cell damage [38].

Fig.2.The 3D operation space determined by the novel method,including all successful cases (blue and green circles),and excluding all failed ones (red circles).Taken from Li et al. (2019) [36].

Shear force requirements in the scale up process constraint the rotation speed of the impeller system.This may cause a problem when high gas–liquid mass transfer and ample mixing are preferred,as they need the impeller rotation speed to be as high as possible.The result could be a compromise between these constraints when scaling up.With the help of CFD simulations,the optimal operating conditions can be determined at the design stage.Additionally,integrating the biotic phase kinetics model into the flow field model will give a more precise representation of the real situation in the bioreactor during the scale-up process.However,lack of knowledge of the kinetics of the biotic phase has hindered the application of this method.To fill this gap,work on the physiological response and kinetics models describing the kinetics properties of cells has emerged during the past decade[39,40].The next section will focus on the studies of physiological responses of microorganism to its environment.

3.Physiological Responses of Microbial Cells to Surrounding Environment

One of the key challenges for fermentation scale up is the flow field difference between lab-scale and industrial-scale bioreactors.The constraint of specific power input in large-scale bioreactors always results in heterogeneous environments in the large bioreactors.The relevant properties of the strain cultured in the fluctuating environment are not studied in the lab-scale process development stage because only homogeneous distributed nutrients is formed in lab-scale bioreactor.Biochemical engineers and researchers have noticed this problem since the 1980s so the scale-down concept with relevant devices simulating the heterogeneous environment in the lab-scale bioreactor has been proposed and studied.A new strategy was proposed for carrying out the fermentation scale-up design,namely to begin with the end in mind[8,41].The strategy involved both qualitatively and quantitatively evaluations of the physiological response of microbe cells to fluctuating environments.The following two sections discuss about these two aspects,respectively.

3.1.How the cell response to the fluctuating environment

Microorganism cells are“alive”compared to chemical catalysts.Precise and complicated multi-layer regulation systems coded by the genome take responsibility to respond to the extracellular environment.These above-mentioned heterogeneities can induce multiple physiological responses.Different cells have different responses according to their genetic or metabolic control mechanism.A bioreaction characteristic time is bioreactor scale independent,while the mixing time decreases as bioreactor scale increases[16,42].Thus,oscillating environment in large-scale bioreactors leads to a heterogenetic cell population.This usually causes lower yields and productivities and an increased by-product formation compared to lab-scale bioreactors [43,44].

To study the response of cells under oscillation,environmental gradients need to be simulated in the lab-scale.Various kinds of scale-down bioreactors therefore have been designed to improve the understanding of the cell response in heterogeneous environments[45,46].The predominant approach of scaling down is based on compartmented bioreactor devices.Cells are cultured in a stirred tank reactor(STR)linked with another STR or plug flow reactor(PFR),which is called a two-compartment system [47].To more accurately simulate multiple heterogeneous regions in the fermentation process,a novel three-compartment scale-down bioreactor has been designed by adding a non-aerated PFR on the basis of a two-compartment system [48].Pulse or stimulation experiments are also used to investigate the response to the environment fluctuation in a short time frame of seconds or minutes.Microorganisms show a quick response to sudden change of extracellular substrate and different cells have various dynamic response to similar stimulation [49].Pulse response experimentation is an effective quantitative tool to study the response to the fluctuation.

Exposure to concentration gradients can cause certain changes in cell metabolism.Some microorganisms show their robustness in the bioreactor.Corynebacterium glutamicumis significantly robust to oxygen and substrate oscillations during the fed batch process [50].The determination of this key mechanism of robustness is a future goal for robust strain development.A twocompartment system(STR-PFR)was used to simulate the gradients of dissolved oxygen and substrate concentration ofBacillus subtilisin industrial scale fermentation [51].The results show that the decrease in amino acid synthesis is largely due to the metabolic shift to ethanol formation.Interestingly,Escherichia colifed-batch cultivations in a dynamic environment shows reduced biomass but a low death rate [52].This means that such fluctuations in large bioreactors may be beneficial for the cell.We recently investigated the fast response of a glucoamylase producing strainAspergillus nigerto high and low glucose pulsing [53],which showed that fast response of the upper glycolysis and higher flux through the hexose monophosphate (HMP) pathway may be the reason for the drop of glucoamylase productivity in fluctuating environment.

All of the above studies on the physiological response of microbe cells to fluctuating environments show the variety of responses of microorganisms,depending on their genotype.However,this should be taken considered when developing an industrial fermentation process,even as early as the beginning of the design phase.

3.2.In vivo kinetics of cells to quantitatively describes the relationship between environment and cell metabolism

Thus far,we showed that the fluctuating environment in a large-scale bioreactor has a profound effect on the strain’s metabolism.This can lead to failures during the scale-up of a developed fermentation process from homogeneous lab-scale bioreactors.However,to evaluate and predict the output of microorganisms under large-scale bioreactors before implementing a real scale-up process,we need a kinetics model that describes the relationship between nutrient concentration and their reaction rates.In addition,the different environments,whetherin vivoorin vitro,limit the kinetics model derived within vitrodata.

The applications of kinetic models of cell factories can improve the understanding of cell metabolism and identify genetic engineering goals to achieve high productivity.A successful kinetics model can describe the relationship between environment and cell metabolism.Furthermore,it could predict how metabolic phenotypes are dynamically shaped by gene content and environmental conditions [54].Almquistet al.offers an excellent review of the basic workflow for establishing kinetic models [55].The kinetic model is also convenient to apply in CFD modelling [11].

Pulse response experiment is a useful and effective tool to estimate parameters of enzyme kinetics for investigatingin vivometabolism.Generally,the external stimulation is addressed on a steady-state chemostat system,and then rapid sampling and quenching protocols are established to obtain perturbed metabolome quantitatively [56].Multiple-pulse response experiments can obtain morein vivoinformation than single-pulse experiments.Useless or uncertain information can then be eliminated [39].Microorganisms in general have the properties of robustness to confront external environmental perturbation.Thus,pulse response experiments are proposed to study the stability of cells in the bioreactor and provide research directions for metabolic engineering.Structuredin vivokinetics model development relies on high qualityin vivometabolite data and an elaborately designed experiment.The construction of thein vivokinetics model can be established by stimulus-pulse experiments with a fast sampling technique and high-resolution measurements of the intracellular metabolite concentrations.A diagram showing how to undertake thein vivokinetics measurement is given in Fig.3.

4.Perspective on Rational Fermentation Scale-up with the Help of Big Multiomics Data-based Models

Tight and precise regulation mechanisms endow the microorganism with the ability to dynamically shift their metabolism to the extracellular fluctuating environment.How to make full use of multiomics big data in the fermentation process for better understanding the regulation mechanisms during process scale up is very important for rational design of a large-scale industrial fermentation process.Omics data itself does not help for solving the scale-up issue,however systems biological models based on multiomics data integration are helpful for better industrial fermentation scale up [57].Even though it is challenging,systems biological analysis has been proposed in recent years to make big data modeling possible.A classic method first proposed in 1999,known as a genomic-scale metabolic models (GEM),integrates genomic,transcriptome,proteome,metabolome,and fluxome data to provide a comprehensive insight into cell metabolism and predict cell behaviour [58,59].

In addition,unlike the traditional random mutation pattern,GEMs can identify new specific target genes through combining numerous omics data with specific algorithms,providing information for the second round of genetic engineering of the strain[57,60].For example,Broet al.[61]used the GEMs to select engineering target for ethanol production withS.cerevisiae,which resulted in 40% lower glycerol yield and 3% higher ethanol yield.More examples can be referred to [60].Currently,it has been applied to industrial strains (e.g.Saccharomyces cerevisiae[62],Escherichia coli[63],andBacillus subtilis[64],etc.),giving constructive guidance for target genes aimed at improving productivity.For example,GEMs has been applied successfully for metabolic engineering of budding yeast to improve ethanol[61]and succinic acid yields [65].

Fig.3.Schematic diagram for in vivo kinetics model construction by using stimulus-response experiment with fast sampling and high-quality measurements of in vivo dynamic metabolite concentration.

Fig.4.Scale-up strategy taking industrial-scale heterogeneous environment in mind at the biginning.The heterogeneous environment is then scaled down to the level of a lab.This is followed by stimulus-response experiments,and multiomics studies of the cell,providing the system with big data on different omics levels.These multiomics big data are then used to construct a reliable constrained GEM model,which is then integrated to the flow-field model to predict the target gene for the cell or the optimal operation condition in a large-scale bioreactor,untill the fermentation process is scaled up.

To be applied for rational design and scale-up of bioprocess,the integration simulation approach coupling fluid dynamics in bioreactor and cellular kinetics based on both the Euler-Euler and the Euler-Lagrange shows attractive potential [11].A structured nine pool kinetics model of penicillin production byPenicillium chrysogenumhas been constructed[66]and used to integrate with a fluid dynamics model in a large-scale stirred tank bioreactor (150 m3)using the Lagrangian method [41,67].The Lagrangian simulation method gives the population dynamics as it simulates millions of individual cell’s trajectories within the chaotic flow in the largescale bioreactor.Currently,the integrated model only takes the simple kinetics model with only 10 kinetics equations.If the multiomics-constrained GEM model can be applied in the Lagrangian model,more detailed regulation process in the metabolic reaction network can be predicted and provide possible target gene identifications candidates that will benefit the scale-up process.Thus,a new comprehensive rational scale-up strategy for the fermentation process is proposed,as shown in Fig.4.

5.Conclusions

Fermentation performance is mainly determined by the capacity of the strain.However,the environment around the cell inside a bioreactor may hinder the ability of the strain.The interrelationship between the environment in the bioreactor and the cell physiological properties is the key that determines the outcome of bioreactors at different scales.The flow field structure in a bioreactor is complex,especially under a chaotic transient and turbulent flow regime.This makes it challenging to predict the performance of the fermentation process carried out at different scales by solely investigating the flow field through computational fluid dynamics.Integrating fluid dynamics models with physiological kinetics models is thought to be the most promising route to investigate and understand the complicated relationship between these two aspects.This concept has been known and studied since 1996 when the EU launched their project,“Bioprocess scale-up strategy based on integration of microbial physiology and fluid dynamics”with grant number BIO4950028 [68].In addition,the study for investigating the physiological response of microbe cells to fluctuating environments was boosted with various scale-down systems,which received more interest from both academics and industrial researchers [47,69–71].

To develop an industrial fermentation process,the design must incorporate considerations for the ultimate scaled-up size[8,41].These considerations are as follows:inclusion of the heterogeneous flow field in the industrial scale bioreactor,awareness of process efficiency,the ability to integrate both flow field in the bioreactor and various scale biokinetics to predict the population dynamics.All this is conducted at during research and development in the lab.Also,it is necessary to mimic the heterogeneous flow field in the lab by conducting a thorough investigation to find the optimal process with respect to the heterogeneous condition.Anin vivokinetics representation can be formed in the lab by conducting stimulus–response experiments.A detailed genome-scale metabolic model with omics data as a constraint can be used to form a more precise biotic kinetic model.These models can be integrated into the fluid dynamics model in the production-scale bioreactor to predict the outcome before actually deploying the final facility.A more robust and efficient industrial fermentation process with a high-quality product can hence be expected.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

The authors would like to acknowledge the Projects 21776082 and 21978085 supported by National Natural Science Foundation of China,and Project 22221818014 supported by the Fundamental Research Funds for the Central Universities.