Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
DOI 10.1007/s00253-005-0003-0
MINI-REVIEW
F. R. Schmidt
Received: 8 March 2005 / Revised: 12 April 2005 / Accepted: 15 April 2005 / Published online: 7 July 2005
# Springer-Verlag 2005
Abstract To increase product yields and to ensure con- up strategies while considering future technologies, with
sistent product quality, key issues of industrial fermen- emphasis on Escherichia coli as one of the most com-
tations, process optimization and scale up are aimed at monly fermented organisms.
maintaining optimum and homogenous reaction conditions
minimizing microbial stress exposure and enhancing met-
abolic accuracy. For each individual product, process and Introduction
facility, suitable strategies have to be elaborated by a com-
prehensive and detailed process characterization, identifi- The scale up of fermentation processes as a central prob-
cation of the most relevant process parameters influencing lem in biotechnology was first recognized and described
product yield and quality and their establishment as scale- during industrial penicillin production at the beginning
up parameters to be kept constant as far as possible. Phys- of the 1940s. At present, several aspects of this subject are
ical variables, which can only be restrictedly kept constant supported as joint research projects by European and na-
as single parameters, may be combined with other pertinent tional research foundations.
parameters to appropriate mathematical groups or dimen- Fermentation scale up is aimed at the manufacture of
sionless terms. Process characterization is preferably based larger product quantities, if at all possible, with a simul-
on real-time or near real-time data collected by in situ and taneous increase or at least consistency of specific yields
on-line measurements and may be facilitated by supportive and product quality. As will be illustrated in the next sec-
approaches and tools like neural network based chemo- tions, the changed geometric and physical conditions in
metric data analysis and modelling, clarification of the larger scales, however, lead to a less favourable mixing
mixing and stream conditions through computational fluid behaviour and to impaired physiological reaction condi-
dynamics and scale-down simulations. However, as fer- tions which in turn may lead to a decreased process con-
mentation facilities usually are not strictly designed ac- stance and reproducibility, to reduced specific yields and to
cording to scale-up criteria and the process conditions in an increase in unwanted side products and thus ultimate-
the culture vessels thus may differ significantly and since ly to a diminished batch-to-batch consistency and product
any strategy and model can only insufficiently consider and quality, key issues in industrial production processes.
reflect the highly complex interdependence and mutual
interaction of fermentation parameters, successful scale up
in most cases is not the result of a conclusive and straight- Scale up related problems
lined experimental strategy, but rather will be the outcome
of a separate process development and optimization on Reduced mixing quality and enhanced stress exposure
each scale. This article gives an overview on the problems
typically coming along with fermentation process optimi- The problem of reduced mixing quality in larger scales is
zation and scale up, and presents currently applied scale- aggravated with increasing vessel sizes: the opposite sub-
strate and oxygen gradients along the vessel height which
are formed as a result of the conventional fermenter design
F. R. Schmidt (*) according to which substrate feed usually occurs from the
Sanofi-Aventis Deutschland, Biocenter H 780, top and aeration from the bottom, are more pronounced in
Industriepark Hchst, larger reactors due to (1) longer distances to be covered
65926 Frankfurt, Germany
e-mail: Frank-Rainer.Schmidt@sanofi-aventis.com leading to larger substrate and oxygen depletion zones, (2)
Tel.: +49-69-3056757 larger volumes of culture broth to be stirred and therewith
Fax: +49-69-30516371 longer mixing times and (3) stronger hydraulic pressure
426
gradients influencing the oxygen transfer rate (OTR; Eq. 14, stead of isoleucin into recombinant E. coli hirudin in po-
15, Table 1) (discussed in more detail by Larsson et al. sitions 29 and 59. A misincorporation of norvalin instead
1996; Bylund et al. 1998, 1999, 2000; Enfors et al. 2001). of leucin and methionine into human recombinant haemo-
Cells at the fermenter top are exposed to excess glucose globin in E.coli has been published by Apostol et al. (1997)
concentrations and simultaneously suffer from oxygen limi- and Kiick et al. (2001). Amino acid misincorporations are
tations, whereas those at the bottom are exposed to glucose thus a cause for an increase in by-products at the expense
starvation. Excess glucose concentrations (threshold value of the desired main product yields in larger scales.
for Escherichia coli at 30 mg/l) result in acetate overpro-
duction (overflow metabolism), a simultaneous oxygen limi-
tation further induces the formation of ethanol, hydrogen, Plasmid stability
formiate, lactate and succinate (mixed acid fermentation,
Bylund et al. 2000; Xu et al. 1999; Castan and Enfors 2001). An essential prerequisite for high product yields particu-
The produced acids can become reassimilated in oxygen- larly in larger scales, in which cultures pass a higher num-
rich zones, but in any case first lead to a temporary acid- ber of generations due to larger culture broth volumes and
ification of the microenvironment, later eventually also of longer inoculation chains from the cell bank to the produc-
larger regions. Combined with a decreased transportation tion stage, is the stable propagation of plasmids to daugh-
and elimination of carbon dioxide, detrimental metabolites ter cells. Plasmid stability is influenced by the plasmid
and surplus heat generated by agitation and metabolic properties, including size and nucleotide sequence, by the
processes, resulting in zonal overheating, the lower mixing genetic background of the host (details discussed by Friehs
rates in large scales thus lead to the formation of zones with 2004; Thiry and Cingolani 2002) as well as by process
enhanced stress conditions. The subsequent activation of parameters like temperature, growth rates and substrate
stress genes (a survey on bacterial stress proteins is given by concentrations. Lin and Neubauer (2000) show that rapid
Kwint et al. 2003) only partially protects the cell against glucose oscillations favour plasmid stability and recom-
detrimental stress effects. For instance, despite activation of binant protein production rate, whereas high glucose con-
heat stress genes like E. coli dnaK and clpB, reducing centrations diminish plasmid stability (Neubauer et al.
misfolding and aggregation of heat-sensitive proteins (Mogk 2003). Plasmid stability and plasmid numbers are thus neg-
et al. 1999), metabolic changes and damages such as trans- atively influenced, more difficult to control and less easy to
location of proteins and membranes have been observed be maintained in larger scales. A concept to render plasmid
upon heat exposure (Umakoshi et al. 1998). The evidently stability and expression rates more independent from such
unavoidable detrimental effects of the repeated and cyclic physiological influences is the application of runaway plas-
passing of different stress zones and the subsequent con- mids, the replication of which can be induced separately
tinuous activation and shut down of the corresponding from growth in the desired fermentation phase (e.g. Ansorge
stress genes are believed to lead to a completely altered and Kula 2000; Nordstrom and Uhlin 1992), enabling copy
physiology (Schweder et al. 1999; Enfors et al. 2001) with numbers of up to 1,000 in the expression phase and, due
constant metabolic shifts, which ultimately reduce growth to the separation of their replication and expression from
and productivity and increase by-product formation. the growth phase, simultaneously a more precise replica-
tion and amino acid incorporation. Further possible mea-
sures to enhance the accuracy of metabolic processes are (1)
Metabolic shifts a reduction the generation time and generation numbers
throughout the process, (2) a deceleration of the metabolic
One of the most important consequences of stress-induced speed, e.g. through reduction of temperature, and (3) the
metabolic shifts affecting product quality is the misincor- development more stress-resistant microbial strains.
poration of amino acids into native and recombinant pro- A prioritized goal of process optimization and scale up
teins, the reasons for which are complex and not yet fully thus consists of an appropriate process design which im-
clarified: according to current knowledge, stress-triggered proves the physiological conditions and the metabolic ac-
changes in glycolysis, citrate cycle and of pathways in- curacy by minimizing microbial stress exposure.
volved in anaerobic and mixed acid fermentation lead to
shifts in the coupled amino acid biosynthetic pathways
(Fenton et al. 1997). Even though translation processes Process characterization
are considered to be generally performed quite precisely
(Weickert and Apostol 1998), the resulting shifts in the To identify process-specific stress factors and to understand
amino acid pool in turn lead, due to the inherent inaccu- the physiological responses to the vessel specific physical
racy of the translation machinery (unspecificity of amino- conditions, the mutual influences and interactions of the
acyl-tRNA synthases, wobbling of tRNA), to amino acid various physical and physiological parameters have to be
misincorporations in freshly synthesized proteins, partic- analysed in detail. An overview on analytical methods
ularly in phases of high protein production rates which are currently applied and yet in development is given in the
characterized by amino acid shortages. Muramatsu et al. following subsections.
(2002) report the incorporation of -methylnorleucin in-
427
each other or other process relevant variables, preferably but not necessarily, as dimensionless groups as described and discussed, e.g. by Hubbard (1987) and Wang and Cooney (1997)
Parameter/coefficient Mathematical characterisation Symbol explanation
Power input (P); volumetric 1 P 2nM NPo n3 dI5 kgm2 s2 W P=power input; n=stirrer speed; M=momentum; NPo=dimensionless power
power input (P/V) number (impeller specific); =density of the medium; n=stirrer speed;
Dimensionless power 2 NPo P=n3 dI5 dI=impeller diameter
number (NPo)
Impeller tip speed (vtip) 3 vtip 2 n dI m=sec n=stirrer speed; dI=impeller diameter
Reynolds number (Re) 4 Re ndI2 = Re=Reynolds number; n=stirrer speed; dI=impeller diameter; =density of the
medium; =dynamic viscosity
Mixing time (Tm) 5 Tm f n;dI ;V =QV = Nfl n dI3 sec Tm=mixing time; n=stirrer speed; dI=impeller diameter; =kinematic viscosity;
V=volume; Q=volumetric flow rate; Nfl=pumping number
Dimensionless mixing time 6 Tm nf Re; NPo Tm=mixing time; n=stirrer speed; Tmn=dimensionless mixing time; Re=Rey-
(Tmn) nolds number NPo=dimensionless power number
Modified dimensionless 0 NPo=modified dimensionless power number Tmn=modified dimensionless
7 NPo NPo Re3 dF =dI PdF 2 =3
power number mixing time; NPo=dimensionless power number; Re=Reynolds number;
.h i
Modified dimensionless 8 T0m n nTm dF =dI 2 Re Tm =dF 2 dF=inside vessel diameter; dI=impeller diameter; P=power input; n=stirrer
mixing time speed; tm=mixing time; =fluid density; =dynamic viscosity
Aeration rate (volume per 9 AR FG =VR m3 =m3 min AR=aeration rate; FG=volumetric gas flow rate; VR=fermenter reaction volume
volume per minute, vvm)
Superficial gas velocity (vs) 10 vs FG =A m=sec vs=superficial gas velocity; FG=volumetric gas flow rate; A=fermenter cross
section
Gas hold up () 11 VR =FG =gas hold up; VR=fermenter reaction volume; FG=volumetric gas flow rate;
Fractional gas hold up () 12 " VG =VR =fractional gas hold up; VG=dispersed gas volume
Gassing number (NQg) 13 NQg f Fr;Re;S; g =0 ; g =0 qg =ndI3 NQg=gassing number; Fr=Froude number; Re=Reynolds number; S=fluid
constant; =density; =dynamic viscosity; g=aerated; 0=not aerated; qg=gas
throughput; n=stirrer speed; dI=impeller diameter
Oxygen transfer rate (OTR) 14 OTR kL aCG CL kL a LO2 pO2G pO2L kg O2 =m3 h with OTR=oxygen transfer rate from gas to liquid phase; kL=mass transfer
15 CG 0; 526pi =36 T mg=1 coefficient; a=specific interfacial surface area; CG=oxygen saturation
concentration in the gas phase; CL=measured oxygen saturation concentration
in the liquid phase; LO2=oxygen solubility in the liquid phase; pO2G=partial
pressure of oxygen in the gas phase; pO2L=partial pressure of oxygen in the
liquid phase; pi=vessel back pressure [bar]; T=temperature [C]
Volumetric oxygen mass 16 kL aa0 P=VR b vcs s1 kLa=oxygen transfer coefficient; P=power input; VR=fermenter reaction
transfer volume; vs=superficial gas velocity; a, b and c as specific fluid constants to be
coefficient (kLa) determined experimentally
a
kLa Scale-dependent 17 kL a k 0 Pg =VR vs b B=60;8 j=dI 0;3 s1 k, a, b=vessel specific coefficients; Pg=power input in aerated vessel;
according to V=volume of culture broth; Vs=superficial gas velocity; B=number of stirrers;
WangCooney j=buffle width; dI=impeller diameter
Respiratory quotient (RQ) 18 RQ CER=OUR pCO2E pCO2i =pO2i pO2E mol CO2 =mol O2 RQ=respiratory quotient; CER=carbon dioxide emission rate; OUR=oxygen
uptake rate; pCO2=partial pressure of carbon dioxide; pO2=partial pressure of
oxygen; i=air feed; E=exhaust air
429
For instance, the kLa valueas the most frequently applied physical scale-up variablehas been combined with (1) the vessel backpressure p by maintaining the product of kLap constant
through variation of pressure, agitation and aeration for scale up of Bacillus thuringiensis fermentations (Flores et al. 1997), or (2) the aeration rate vvm under variation of power input,
Qt=total energy; Qmet=energy generated by metabolic activities; Qag=energy
perature difference
21
Process design tects the scattered and reflected light pattern (Doppler shift)
generated by floating particles passing a dual laser wave
The realization of concepts aimed at the improvement of interference field. For the mathematical calculation and
mixing quality in larger scales to ensure homogenous re- description of turbulent flows [e.g. through direct numeri-
action conditions and to reduce both, the size of stress cal simulation (DNS)], the reaction volume is discretized
zones and the zonal residence times, requires the char- and meshed into as many zones and cells as possible and
acterization of the large-scale hydrodynamic and reaction subsequently, on the basis of given geometric and rhe-
conditions. As supportive approaches, the conditions gov- ological data, the physical conditions governing these cells
erning in large vessels and their relevant zones are simu- are described by solving the corresponding NavierStokes
lated and depicted through high-performance computing equations under consideration of the conservation laws of
[computational fluid dynamics (CFD); e.g. Cant 2002] and mass, momentum and energy. The generation and distri-
through transfer into small reactors (scale down). bution of the resulting field variables enable the vectorial
representation of physical conditions at any point and
therewith detailed information with respect to trajectories,
Scale down approach mixing time distributions, mixing rates and profiles, hy-
drodynamic conditions, flow behaviour, mass transports
Scale down designs, like unstirred bypass loops to simulate and shear stress (Bezzo et al. 2003) in dependence of al-
and investigate effects of oxygen limitations, excess glu- ready existing or still to be developed reactor and stirrer
cose concentrations or pH fluctuations, are illustrated by design (Kelly and Humphrey 1998), properties of the cul-
Amanullah (2001); Enfors et al. (2001); Hewitt et al. (2000) ture broth like density and viscosity as well as the adjust-
and Onyeaka et al. (2003). The characterization of the ed process parameters. In this way, also scale-up influences
governing stress conditions by tracking of the correspond- are predictable. Williams et al. (2002) report a standard-
ing stress gene expression (see above) can be performed ization of hydrodynamic and mixing conditions as well
on-line by means of bioluminescence, e.g. by the green as the avoidance of shearing stress and oxygen limita-
fluorescing protein (GFP signaling) and the detection by tions even in critical reactor zones for the scale up of car-
appropriate probes (Reischer et al. 2004). The sort and tilage cell fermentations. Broadly applied in the field of
amount of parameters which can be simulated in small numerical flow simulations are the flow modelling software
scales, however, appear to be restricted (see Bylund et al. tools of Fluent Inc. (USA) (Fluent 4, 5, 6 and MixSim). A
2000). On one side, the viability of cells is impaired in simulation of the trajectories and distributions of gas bub-
smaller scales (Hewitt et al. 2000; Enfors et al. 2001) and bles and mass transports enables the determination of zon-
physiological parameters evidently are not straight-lined al pO2 values and oxygen transfer rates. The combination
transferable. When comparing E. coli fermentations in dif- with parameters like substrate concentrations and gradients,
ferent scales with similar and identical courses of biomass, the residence times in the respective zones, population
respiration, cell lysis, mixed acid fermentation, overflow dynamics and metabolic fluxes (structured metabolic mod-
metabolism, glucose and oxygen concentrations, Bylund els) leads to integrative models [integrated fluid dynamics
et al. (2000) found that the yields of human growth hor- (IFD)], which permit the prediction of physiological ef-
mone were 80% higher and the concentration of pro- fects and reactions. Even though it is emphasized that the
teolytically digested side products 30% lower in the small integration of CFD and structured biokinetics at present
scale. Evidently, knowledge about the conditions to be requires a deeper understanding of the dynamics of meta-
transferred is not yet sufficiently comprehensive. Current bolic and regulatory networks and cascades of signal trans-
efforts thus try to complete the picture by a detailed char- duction triggered by microenvironmental fluctuations and
acterization of the hydrodynamic conditions, as performed thus further research (Schmalzriedt et al. 2003), it will
by Enfors et al. (2001) for investigation and description most likely contribute at long terms to a realistic model-
of gradients and oscillations of substrate concentrations ling of the interplay between physics and physiology and
through computational fluid dynamics (CFD). thus will facilitate the identification of key parameters
influencing product yield and quality the most and there-
with of suitable scale-up parameters and strategies. Prob-
Computational fluid dynamics lems related to the scale up of physical parameters are
illustrated in the next section.
Generally, hydrodynamic model development can be per-
formed mathematically abstract as well as data driven on
the basis of experimental flow and circulation measure- Physical scale-up parameters
ments (Davidson et al. 2003), during which the trajectory
and velocity of tracer particles (fluorescence dyes, magnet- Suitable for employment as physical scale-up parameters
ic or radio pills) is pursued and detected optically or by are all known process parameters and coefficients exerting
means of antenna or induction coils wound around the ves- known physiological effects, particularly those affecting
sel (flow follower techniques). A further optical detection oxygen supply, heat transportation and mixing, such as
procedure, the application of which however is restricted power input (Eq. 1; Table 1), aeration and agitation rate
to glass vessels, is the laser Doppler method, which de- (Eq. 3, 9, 10; Alves and Vasconcelos 1996), mixing time
431
(Eq. 5; Oniscu et al. 2001) and dimensionless mixing time groups and characteristic curves (examples given, e.g. by
(Eq. 6), gas hold up and fractional gas hold up (Eq. 11, 12), Wang and Cooney 1997).
pO2, OTR (Eq. 14, 15; Diaz and Acevedo 1999; Russell This relates also to the kLa value (Eq. 16), which is
et al. 1995), oxygen mass transfer coefficient (kLa, Eq. 16, currently the mostly applied physical scale-up variable
17), and biocalorimetric variables like heat fluxes and heat since it includes the relevant parameters influencing oxy-
transfer coefficients (Eq. 19, 20, 21). Biocalorimetric mea- gen supply like agitation (via energy input) and aeration as
surements were found to be in definite correlation with superficial gas velocity (Eq. 10) and which, as a component
metabolic activities and with OTR and are employable as of dimensionless terms, is also frequently employed for
growth phase indicators (Trker 2004; Anderson et al. reactor scale up (see Yawalkar 2002).
2002); however, the calculations require a precise measure- Taking into account the limitation of a kLa-oriented scale
ment and knowledge of heat fluxes and sources, energy up in the form of a limited power input, OTR and tolerable
inputs and of the temperature distribution in the vessel. In shear stress, Flores et al. (1997) integrated the backpressure
most cases, however, it is, depending on the scale-up factor, p by maintaining the product of kLap constant through
not or only restrictedly possible from the physical view- variation of pressure, agitation and aeration for scale up of
point to keep physical parameters constant throughout the Bacillus thuringiensis fermentations. While the specific
scale up. Constance of even a single specific parameter growth was reduced and the biomass production and spor-
mostly leads to an uncontrolled and unpredictable change ulation efficiency remained constant, fermentation time
of other variables into dimensions that are technically not could be shortened and toxin yields were increased. Wong
realizable. Classical examples are (1) the mixing time, et al. (2002) successfully scaled up E. coli fermentations
which inevitably increases in larger vessels due to the from 5 to 200 l by keeping constant the product of kLa
larger volumes to be stirred and which cannot be un- and aeration rate (vvm) under variation of power input,
limitedly compensated and kept constant by increasing working volume and aeration according to the Wang
stirrer speeds and energy inputs and (2) the volumetric Cooney equation (Eq. 17; Wang and Cooney 1997), which
energy input P/V itself. A constant volumetric power input reflects these parameters in dependence of the fermenta-
has indeed been successfully applied as scale-up parameter tion scale. Diaz and Acevedo (1999) argue that the oxygen
for the early industrial penicillin fermentations (1 hp per transfer capacity, indicated by the kLa value, is not the most
gallon, equivalent to 1.8 kW per 1 m3) and in fermentations process-relevant parameter, but the effective oxygen trans-
with low energy inputs (see Kim et al. 2003), but is limited fer rate as a product of kLa and mass transfer potential, i.e.
in fermentations requiring high energy inputs, like recom- the difference between the pO2 values in the gas and the
binant E. coli cultures. liquid phase and suggest an OTR based scale-up strategy
For this reason, mathematically driven approaches are (Eq. 14, 15).
pursued both for process and reactor scale up by form- A common simple and robust method is the maintenance
ing dimensionless coefficients (e.g. by the Buckingham of a constant pO2 in the culture broth by variation of stirrer
Pi method as cited by Hubbard 1987), which are kept speed and aeration rate. To avoid shear stress and to keep
constant by an appropriate choice and adjustment of the the energy input (P) at the lowest possible level, the pO2
relevant influencing process and fermenter parameters. is steered at the minimum limit. Riesenberg et al. (1990)
Well-established examples are, among many others, the employ a stirrer speed and glucose feed steered pO2 of 20%
dimensionless power number (Eq. 2), Reynolds number for interferon E. coli fermentations in a 30- and 450-l
(Eq. 4), gassing number (Eq. 13) and the modified di- scale. These examples demonstrate that, for each specific
mensionless power number (Eq. 7) and the modified di- process, appropriate variables and coefficients and there-
mensionless mixing time (Eq. 8). The comparison of the with scale-up strategies have to be identified individually.
latter two coefficients and of the corresponding curves
enable the identification of appropriate types of stirrers
capable of exerting the desired mixing performance at a Conclusion
given stirrer speed with a minimum of energy consump-
tion and therewith of compensating the above-mentioned Despite the central role of the scale-up issue in biotech-
limitations of volumetric energy inputs at larger scales. nology and the comparably large body of literature, there
(A quantum leap in stirrer technology in this regard seems seems to be no common, generally applicable strategy es-
to be the Visco-Jet of Inotec, which possesses cone formed tablished. For each individual product, process and facility,
short tubes instead of blades and is reported to exhibit best a suitable scale-up strategy has to be elaborated.
mixing performance with a minimum of power input and A wholistic scale-up strategy consists of a comprehen-
shear stress without foam generation.) The performance of sive and detailed process characterization to identify key
further various types and designs of stirrers is presented and stress factors and key parameters influencing product yield
discussed by Junker et al. (1998). Stirrer performance co- and quality the most, and of an appropriate process control
efficients and curves are thus crucial scale-up aids. and process design ensuring optimum mixing and reaction
In addition, any of these parameters and coefficients can conditions, supported by appropriate knowledge and data
be combined with other variables to set up and create new, driven models (Berkholz and Guthke 2001; Berkholz et al.
process-specific parameter correlations, coefficients, terms, 2000) as well as computational tools (Fig. 1).
432
Process characterisation
Scale down
Chemometric modelling Model development Computational
Neural networks
Process optimisation Fluid Dynamics
Identification of
Scale up parameters
SCALE UP
Fig. 1 Scheme roughly illustrating the theoretical steps and stages of logical application like electric tongues, artificial noses or the lab-on-a-
fermentation process optimization and scale up, characterised by chip technology. Process characterisation as well as the subsequent
gliding transitions. Process development usually starts with a detailed process optimisation, which aim at the improvement physiological
process characterisation through elucidation of the mutual influences reaction conditions and reduction of stress exposure and which enable
of physiological parameters (e.g. productivity, metabolite and sub- the identification of suitable scale-up parameters and strategies, are
strate concentrations, cell mass and viability, respiration) and of phys- usually performed on the basis of experimentally developed data
ical process parameters (e.g. aeration, agitation, pressure, temperature, and knowledge driven assumptions and models. Supportive approaches
fluid dynamic conditions), eventually supported by neural networks and tools for model development are scale down experiments, (neu-
and integrated fluid dynamics. Employable analytical tools for phys- ral network based) chemometric modelling and computational fluid
iological parameter characterisation are the established physical and dynamics. Examples of established physical scale-up parameters are
chemical procedures like flow cytometry, spectroscopy, chromatog- given in Table 1
raphy, etc. as well as techniques yet in development for biotechno-