Influence of Scale-Up on the Quality of
Recombinant Human Growth Hormone
F. Bylund,1 A. Castan,2 R. Mikkola,2 A. Veide,1 G. Larsson1
1
Centre for Bioprocess Technology, Department of Biotechnology, Royal
Institute of Technology (KTH), SE-100 44 Stockholm, Sweden; telephone:
+46-8-790-8738; fax: +46-8-790-9540; e-mail: gen.larsson@biotech.kth.se
2
Pharmacia & Upjohn AB, Process R & D, SE-112 87 Stockholm, Sweden
Received 25 July 1999; accepted 9 January 2000
Abstract: The aerobic fed-batch production of recombi-
nant human growth hormone (rhGH) by Escherichia coli
was studied. The goal was to determine the production
and protein degradation pattern of this product during
fed-batch cultivation and to what extent scale differences
depend on the presence of a fed-batch glucose feed
zone. Results of laboratory bench-scale, scale-down
(SDR), and industrial pilot-scale (3-m3) reactor produc-
tion were compared. In addition to the parameters of
product yield and quality, also cell yield, respiration,
overflow, mixed acid fermentation, glucose concentra-
tion, and cell lysis were studied and compared. The re-
sults show that oxygen limitation following glucose
overflow was the critical parameter and not the glucose
overflow itself. This was verified by the pattern of by-
product formation where formate was the dominating
factor and not acetic acid. A correlation between the ac-
cumulation of formate, the degree of heterogeneity, and
cell lysis was also visualized when recombinant protein
was expressed. The production pattern could be mim-
icked in the SDR reactor for all parameters, except for
product quantity and quality, where 30% fewer rhGH-
degraded forms were present and where about 80%
higher total yield was achieved, resulting in 10% greater
accumulation of properly formed rhGH monomer. 2000
John Wiley & Sons, Inc. Biotechnol Bioeng 69: 119128, 2000.
Keywords: recombinant human growth hormone (rhGH);
product quality; Escherichia coli fed-batch; substrate gra-
dients; scale-up; scale-down reactor (SDR)
INTRODUCTION
Production of proteins by recombinant technology has been
the subject of much industrial interest due to the possibili-
ties of cutting production costs. However, production has
been limited to a few overexpression systems of which
knowledge with respect to process performance is limited.
The complexity of the response of the catalyst to the sub-
strate is also a significant factor. Furthermore, recombinant
protein processes suffer from the same general problems
during choice of production technique and scale-up as all
other biologically and chemically based production sys-
tems.
Correspondence to: G. Larsson
Contract grant sponsors: Swedish Centre for Bioprocess Technology
(CBioPT); Pharmacia & Upjohn AB
2000 John Wiley & Sons, Inc.
In this context, two significant industrial problems with
regard to the cultivation step of biological processes can be
identified: (i) batch variations, especially on the large scale;
and (ii) lack of reproducibility when the scale is changed.
The latter refers to both the scale-up and scale-down situa-
tions and is introduced either when new processes are
implemented or when old processes are being redesigned
and/or optimized. The problems can be caused by a multi-
tude of factors and presently there is no general scientific
method to deal with them. However, some steps in creating
such a methodology have recently been taken.
Although the reasons for batch variation have not been
the topic of much interest in the scientific community there
have been some important results found based on scaling
studies of biological processes during the last 15 years.
Breakthroughs have been made in hydrodynamics and mix-
ing, modeling of two-phase gas/liquid biological processes
(Guillard, 1999), large-scale microenvironment measure-
ments (Bylund et al., 1998, 1999; George, 1997; Larsson et
al., 1996; Manfredini et al., 1983; Oosterhuis and Kossen,
1984), and scale-down methodology studies (Oosterhuis et
al., 1983, 1985). The latter area of study includes the use of
scale-down reactors (SDRs) with various biological model
systems (Bylund et al., 1999; Fowler and Dunlop, 1989;
George et al., 1993, 1998; Larsson and Enfors, 1988;
Moes et al., 1985; Neubauer et al., 1995a,b; Sweere et al.,
1988ac).
In brief summary, these findings point to the existence of
concentration gradients in reactors also on a quite moderate
reactor scale. That substrate gradients exist was to be ex-
pected, because substrates that are fed in limited amounts
(glucose, oxygen) in fed-batch cultivation processes cannot
be immediately mixed to homogeneity in the reactor. Fur-
thermore, substrate concentration is approximately 10,000
times higher near the feed inlet compared with the mean
reactor concentration and the dissipation pattern is a func-
tion of the degree of mixing at the location of the feed point
(Bylund et al., 1999; Larsson et al., 1996).
Presently, however, there are no hydrodynamic models or
translations into engineering parameters of the findings just
referred to that can describe or directly influence the pat-
terns of substrate gradients. An important observation in
this direction was that the frequency pattern of gradient
oscillations was also detectable in the fluctuations of the
turbulent velocity in the same location. This led the present
investigators to conclude that intact packages of substrate
of high concentration travel quite long distances in the re-
actor before dissipation (Larsson et al., 1996).
Patterns of substrate oscillations have thus been identified
as a key to scaling design, and have also resulted in research
in the kinetics of biological reactions toward transitions in
substrate concentration. The scale-up/scale-down research
so far has led to that the biological studies of scaling was
done by making substrate transitions in various set-ups
called scale-down reactors. However, these are not based on
any actual large-scale transitions, but rather are designed to
overemphasize certain phenomena that are randomly pre-
sent in the large reactors, as outlined earlier. Therefore, as
result of the lack of a proper physical model for the impor-
tant microenvironment features in a large reactor, SDR re-
sults can presently provide only a yes/no answer to the
question of whether scaling is likely to create a problem. In
this context, the effect of substrate variations on physiologi-
cal parameters such as cell growth, respiration, substrate
uptake, metabolic byproduct formation, and product forma-
tion is described.
Two distinct efforts have been made with regard to scale-
down-specific processes (to laboratory scale). These strate-
gies are based on actual measurements on a large scale and
use the same general concept, but are designed from the
individual characteristic features of the actual process, re-
actor, and organism.
The first investigation involves the scale-down of aerobic
batch cultivation in a continuous stirred tank reactor
(CSTR) of 25 m3 with Gluconobacter suboxydans (Ooster-
huis et al., 1983, 1985). This system is based on batch
technique characteristics; that is, all substrates are in excess
except for the oxygen that is fed to the process over time.
From oxygen gradient patterns, geometry, and modeling of
the large-scale reactor, an SDR was designed consisting of
two CSTRs connected in a loop. The large-scale reactor was
thus approximated by one smaller reactor representing the
stirring zone of comparatively high oxygen concentration
and one comparatively larger reactor to mimic the bulk zone
of lower concentration. It was shown that gluconic acid
production could be influenced by a variation of parameters
in the SDR set-up. However, no large-scale parameter
changes or production optimization were based on these
findings.
The second example involves the aerobic fed-batch cul-
tivation of Saccharomyces cerevisiae in a bubble-column
reactor of 215 m3 (George et al., 1998) and the aerobic
fed-batch CSTR (12 m3) cultivation of Escherichia coli pro-
ducing a recombinant protein (Bylund et al., 1999). To dis-
tinguish the fed-batch effects, large-scale glucose concen-
tration patterns and reactor hydrodynamic/kinetics model-
ing were used to design the SDR. This reactor consisted of
one CSTR and one PFR (plug-flow reactor) in which the
PFR mimics the feed zone (in the grams-per-liter range) of
120 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 69, NO. 2, JULY 20, 2000
the large reactor and the CSTR mimics the bulk zone of a
comparatively low concentration (in the milligrams-per-liter
range). In this context, the feed zone is the sum of the
volume of the glucose packages existing in the large re-
actor (Larsson et al., 1996). Both reactors were aerated. This
system has been successful in studies of bakers yeast pro-
duction with regard to bubble-column performance, even up
to the point of describing changes in the gassing power (a
quality parameter). However, the SDR set-up could only
mimic some parameters of the fed-batch CSTR producing
recombinant protein by cells of E. coli.
The major optimization parameters in recombinant pro-
tein production is productivity (g/gh), yield (g/g), and prod-
uct quality. To the latter category inclusion-body formation,
protein aggregation, and product proteolysis can be in-
cluded. Proteolysis is often a major problem and has been
shown to vary drastically between shake-flask and bioreac-
tor cultivation (Yang, 1995). It can be assumed that uncon-
trolled synthesis of a recombinant product protein under
conditions in which the microenvironment varies can lead to
considerable stress on the microorganism and can lead to,
for example, high proteolytic protein turnover.
The goal of this work was to detect variations in product
formation and protein quality (mainly in degradation pat-
tern) during cultivation. Furthermore, we wanted to deter-
mine whether this pattern changed with scale and also if the
fed-batch feed zone, forming glucose gradients, could be a
reason for the change in product quality during scaling.
A further aim was to verify the hypothesis that the gra-
dient effect in CSTRs, if present, is caused by a secondary
effect of high glucose concentration, namely oxygen limi-
tation, that would follow due to the increased respiration.
This would be seen in the pattern of metabolic byproduct
accumulation; that is, for glucose the acetic acid accumula-
tion, and for oxygen the formic acid accumulation. The goal
was to modify existing SDRs according to this assumption.
We chose to study the production of recombinant human
growth hormone (rhGH) in Escherichia coli and the degra-
dation products of this protein during aerobic fed-batch cul-
tivation on glucose in CSTRs of different scale.
The hypothesis of oxygen limitation as a main control
factor in process performance, in spite of use of the concept
of control by glucose limitation in fed-batch operation, was
verified. Formate, not acetic acid, was present already in the
small-scale laboratory reactor, and increased with increased
scale and was also detected much earlier the larger the scale.
This pattern could be mimicked by the SDR set-up. The
results also showed that proteolytic degradation occurred in
all scales and reactor configurations and increased through-
out the cultivation. Even though the total amount of product
increased, the degradation products accumulated at the ex-
pense of the properly formed rhGH monomer. However, the
degree of proteolysis was considerably lower in the SDR,
providing more of the correct rhGH form and less of the
proteolytically degraded forms. The latter finding indicates
the possibility of designing a new production technology for
recombinant production processes.
MATERIALS AND METHODS
Organism and Medium
Escherichia coli W3110, with a pBR322-derived plasmid
coding for recombinant human growth hormone (rhGH) and
tetracycline and ampicillin antibiotic resistance, was used in
the cultivations performed. In these experiments tetracy-
cline was added. This rhGH form is a 22-kDa protein con-
sisting of 191 amino acids with two intramolecular disulfide
bridges. Its production is controlled by an inducible pro-
moter, and the protein is secreted to the periplasm where the
disulfide bridges are formed. As a reference, a pilot-scale
experiment was performed with the same construct as noted
earlier, but where the gene coding for rhGH had been re-
moved.
Cultivations were performed in a glucose mineral salt
medium as described by Forsberg et al. (1997) with minor
modifications of the components of the medium and trace
elements. Medium components together with distilled water
were sterilized in the bioreactor at 121C. Sterilized mag-
nesium sulfate, trace elements, and antibiotics were added
separately after sterilization. Breox (International Speciality
Chemicals, Ltd., Southampton, UK), was added as needed
for foam control. The feed solution had a glucose concen-
tration of 590 g/L.
Cultivation Procedure and Bioreactors
Exponentially growing cells in batch mode were transferred
from the seed fermentor to the production bioreactor. On the
laboratory scale, the cell suspension was transferred via a
sterile flask. In pilot-scale cultivations, the cell suspension
was pressed through a pipe connecting the bioreactors. The
inoculum volume was about 10% of the final total volume.
After the transfer procedure the glucose feed was immedi-
ately started. The feed profile consisted of three different
phases: (i) an exponential feed; (ii) a constant feed when the
estimated maximum oxygen transfer capacity of the biore-
actor was approached; and finally (iii) the feed rate was
decreased to 80% of its maximum value at the time of
induction. The feed was constant thereafter. The feed profile
was allowed to vary by 5%.
Temperature was controlled near the maximal value for
E. coli growth and the pH was controlled at close to neu-
trality by the addition of 25% NH3(aq). Dissolved oxygen
tension (DOT) in the bioreactor was monitored by a polaro-
graphic oxygen electrode and was controlled by varying the
stirrer speed up to 1000 rpm and 205 rpm in the lab-scale
and pilot-scale bioreactors, respectively. The aeration rate
was 1.4 vvm (relative to the start volume) and the pressure
at 0.4 bar over pressure. One exception was in the scale-
down experiments where increased pressure could not be
BYLUND ET AL.: SCALE-UP EFFECTS ON RECOMBINANT HUMAN GROWTH HORMONE PRODUCTION
maintained due to sensitive equipment. As a consequence,
the stirrer speed was increased to 1200 rpm to avoid oxygen
limitation in the CSTR.
All homogeneous control and inoculum lab-scale culti-
vations were performed in a 15-L CF 3000 Chemap-
Fermentor (Switzerland) CSTR with a start volume of 7.0 L.
The volume increased during the feed phase up to 9.0 L.
A SDR consisting of two compartments with exchange
flow was used for simulation of large-scale conditions. This
reactor set-up is based on the view of a large-scale fed-batch
bioreactor with enlarged glucose concentrations with limit-
ing amounts of oxygen in the feed zone or close to it and the
remaining space of a low bulk concentration (Bylund et al.,
1999). The two-compartment system consists of one CSTR
(Chemap Fermentor) in series with a PFR. The flow char-
acteristics of the PFR have been described elsewhere
(George et al., 1993). In the SDR experiments, the glucose
feed was added to the PFR inlet, thus giving a zone of local
high glucose concentration. Because no additional air was
supplemented to this reactor, only low amounts of oxygen
were present at high cell concentrations. The circulation of
medium (1.1 L/min) was started 3 h after start of feed. DOT
was measured by a galvanic oxygen electrode at the top of
the PFR and showed a value of 0 at 3 to 4 h thereafter. The
total volume in the two-compartment system was the same
as in homogeneous lab-scale cultivations (7.0 L), and the
volume of the PFR was 440 mL. Consequently, the volume
fraction of medium in this reactor was 5% to 6% and the
residence time 24 sec in the zone with combined high-
glucose and low-oxygen concentration.
In pilot-scale experiments, two geometrically similar pro-
duction lines were used, each consisting of one inoculum
and one production bioreactor, and one substrate reactor for
ammonia and one for glucose feed. The inoculum reactor
had a total volume of 300 L, and the production bioreactor
3 m3 (1.4-m3 start volume). The tank had a height-to-
diameter ratio of 1.7 and was equipped with two Rushton
turbines, four baffles, and a ring sparger at the bottom. The
oxygen transfer rate (OTR) was 0.13 mol/Lh, a value simi-
lar to that measured for the lab-scale reactors during the
conditions studied. Glucose feed was added to the bioreac-
tor through a pipe in the top that had the orifice positioned
at the bottom. Media samples were taken from the bottom
through a pipe and collected at the top. The bottom position
of this sampling pipe differed in location between pilot-
scale vessels. In one of the reactors, this pipe was positioned
at the same level as the feed inlet but on the opposite side,
and in the second reactor the sampling pipe was located next
to the feed inlet. Pilot-scale cultivations were volumetrically
scaled up from control cultivations by a factor of 200.
Thereby, the same amount of glucose was added per reactor
volume and hour in the different experiments.
Analytical Procedures
Growth was followed by biomass concentration measure-
ments. The biomass concentration was determined as cell
121
dry weight (CDW). This was measured by centrifugation of
3 5 mL of cell suspension in preweighed test tubes for 10
min at 4500 rpm, washing of the pellet with 5 mL of dis-
tilled water, recentrifuging, and drying overnight at 105C
before weighing.
The supernatant from the first centrifugation was sterile-
filtered and used for acetate, formate, and DNA analysis
(cell lysis). Glucose samples were taken by rapid sampling
of cell suspension from the bioreactor into preweighed test
tubes containing an equal volume of perchloric acid (0.13
M) as described by Larsson and Tornkvist (1996). This was
done to inactivate substrate uptake by reducing the pH, but
avoiding cell lysis. Duplicate samples were taken from the
pilot-scale bioreactor. Concentrations of glucose, acetate,
and formate were determined with Boehringer-Mannheim
kits (glucose: Kit No. 716 251; acetate: Kit No. 148 261;
formate: Kit No. 979 732). The DNA content of media
samples was determined with Pico-Green.
Samples for product analyses were taken from the pellet
fraction of 5 mL of centrifuged cell suspension. The extrac-
tion method used releases the periplasmic fraction of pro-
teins. Centrifugation was performed at 5000 rpm for 15 min
at +4C. Extraction buffer consisted of (per liter): Tris-HCl,
0.9837 g; Tris-base, 0.4551 g; and ethylene-diamine tetra-
acetic acid (EDTA) 2H2O, 0.372 g. Purity and quantity of
rhGH were determined by hydrophobic interaction chroma-
tography (HIC) and by isocratic reverse-phase high-
performance liquid chromatography (RP-HPLC). HIC was
performed on a TSK phenyl 5PW column (Tosoh Haas) and
RP-HPLC on a polymeric column (Ashipac ODP-50), at pH
8.5, with ultraviolet (UV) detection.
Four different variants of rhGH were separated by HIC
with a decreasing salt gradient and detected with UV de-
tection at 230 nm: (des-Phe1)-rhGH (abbreviated LMW);
rhGH (correctly formed rhGH); (trisulfide Cys182
Cys189)-rhGH (abbreviated trisulfide); and (clipped 142/
143)-rhGH (abbreviated clip-2), as described by Gellerfors
et al. (1990).
Three peaks (A to C) were identified and characterized
with RP-HPLC: peak A contained (Met(O)125)-rhGH and
(des-Phe1, des-Pro2)-rhGH; peak B contained (Met(O)14)-
rhGH, (des-Phe1)-rhGH (LMW), (clipped 142/143)-rhGH
(clip-2), and deamidated rhGH; and peak C contained rhGH
and (trisulfide Cys182Cys189)-rhGH.
The reactor volume and total amount of biomass were
calculated at each sampling occasion as described by By-
lund et al. (1998).
The oxygen and carbon dioxide content in the outlet gas
flow was simultaneously measured with a combined oxygen
and carbon dioxide analyzer (Monitor Type 1308, Bruel &
Kjr), which uses magnetoacoustic and photoacoustic in-
frared spectroscopy.
RESULTS
The cultivation parameters are generally shown in a set-up
of figures describing the three different experiments: (a)
122 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 69, NO. 2, JULY 20, 2000
homogeneous control; (b) heterogeneous SDR; and (c) in-
dustrial pilot-scale cultivations.
All cultivations were performed in the same principal
manner, passing through four different phases; that is, one
batch phase and three fed-batch phases. The initial batch
phase varied between 2 and 5 h due to differences in inocu-
lum biomass concentration or viability. After depletion of
excess glucose, the exponential glucose-limited feed phase
was started, resulting in the corresponding exponential
growth ( < max). After this, the first constant feed phase
started and the specific growth rate declined successively, as
expected, whereas the biomass increased linearly. The
growth rate further declined after induction (25 h), when the
feed rate was reduced to its final value, and the biomass
increased only slightly thereafter and eventually stopped
some hours thereafter. The corresponding biomass con-
centration in the different experiments was of the order of
40 g/L.
Biomass Growth
Cell growth is represented in Figure 1 as total amount of
biomass, yield of biomass, and specific growth rate. As a
comparison to further experiments with the SDR and pilot-
scale reactors, a homogeneous control cultivation is in-
cluded in Figure 1. This is supported by a multitude of
process data based on optical density measurements (results
not shown).
The yield coefficient was 0.4 g/g in all cultivations from
the start of the process to approximately 20 h. After this
point, the yield decreased and, at the time of induction, was
about 0.3 g/g in the pilot-scale and control cultivations. In
SDR experiments, the decrease was more pronounced and
dropped to 0.25 g/g during the same time period. In all
experiments, the yield coefficient was further reduced dur-
ing the production phase. The same principal observation
was made in the cultivation where the recombinant gene
was removed from the plasmid, although a slightly higher
final yield of biomass was obtained compared with the other
cultivations. This means that the reason for the generally
decreased yield was due to the comparatively higher main-
tenance requirements at the end of cultivation and not due to
product formation.
Substrate Uptake and Respiration
The glucose concentration in a fed-batch process is ex-
pected to be around the saturation constant for the organism
(about 5 mg/L for E. coli). This was also true for the present
cultivations as shown in Figure 2. However, the results ob-
tained on a pilot scale (Fig. 2c) are more difficult to interpret
because the glucose concentration varied. The close posi-
tion of sampling relative to feedpoint (see Materials and
Methods) is the most likely explanation for the fluctuations
between samples and the occasionally higher concentrations
(up to 100 mg/L) measured in the pilot-scale bioreactor.

Figure 1. Cell growth shown as biomass (au), yield coefficient of bio-
mass per glucose added (YX/S, g/g), and specific growth rate (/max). (a)
Bench-scale control experiment, (b) scale-down experiments, and (c) pilot-
scale experiments. Repeated cultivations in the different reactors are num-
bered (1), (2), and (3) and marked with different symbols. Biomass marked
as (1) , (2) , and (3) , respectively. Yield coefficient marked as (1) ,
(2) , and (3) +, respectively. Specific growth rate marked as (1) , (2)
, and (3) X, respectively. Pilot-scale cultivation (3) corresponds to the
experiment in which the recombinant gene was removed from the plasmid.
This is a further verification of gradient formation in bio-
reactors at the 3-m3 scale.
Respiration is shown in Figure 2ac as the yield coeffi-
cients of oxygen and carbon dioxide per added glucose
(YO2/S and YCO2/S, g/g). The general trend was a relatively
constant value during the growth phase that increased dras-
tically after induction. This indicates higher respiration dur-
ing the production of the target protein and when growth
declines.
Glycolytic Byproducts
Acetate is produced when E. coli is grown under anaerobic
or oxygen-limiting conditions (mixed acid fermentation) as
well as when growing aerobically in the presence of excess
glucose (overflow metabolism) (Lee, 1996). Formate, on
BYLUND ET AL.: SCALE-UP EFFECTS ON RECOMBINANT HUMAN GROWTH HORMONE PRODUCTION
Figure 2. Glucose concentration (milligrams per liter) and respiration
data shown as yield coefficient of oxygen and carbon dioxide per glucose
added (YO2/S and YCO2/S g/g, respectively). (a) Bench-scale control experi-
ment, (b) scale-down experiments, and (c) pilot-scale experiments. Dupli-
cate samples are shown for the glucose concentration in pilot-scale experi-
ments. Repeated cultivations in the different reactors are numbered (1), (2),
and (3) and marked with different symbols. Glucose concentration as well
as yield coefficient of oxygen marked as: (1) , (2) , and (3) , respec-
tively. Yield coefficient of carbon dioxide marked as: (1) , (2) , and (3)
, respectively. Respiration data for pilot-scale experiments (1) and (3) are
not shown due to error in the analysis.
the other hand, is only produced during mixed acid fermen-
tation. Furthermore, it has been shown that formate is not
consumed as rapidly as acetate when returning to a region
with a high level of oxygen (Cleland, 1988). Of these me-
tabolites, acetate was shown by Vollbrecht (1982) to be
detected before formate in a batch phase when exposed to
increasing oxygen-limiting conditions. However, the con-
centration of the latter increased first after the DOT showed
a value of 0, which was never measured during the cultiva-
tions in the present investigation.
Acetate and formate were produced in all cultivations, but
at different rates and at different times (Fig. 3). Generally,
acetate is formed in the beginning of the process, later re-
placed by formate formation.
Acetate was accumulated during the batch and the expo-
nential feed phases. Acetate level was dependent on process
123
 process. However, the overall result of the SDR cultivations
gave the same byproduct pattern as for the other cultivations
(Fig. 4). In pilot-scale cultivations, the results were inter-
mediate but accumulation was more rapid after induction up
to the levels obtained at the laboratory scale. As can be seen
from Figure 4, the differences in biomass between cultiva-
tions cannot account for the differences in formate accumu-
lation.

Cell Lysis
The amount of DNA in the medium, which is comparable to
cell lysis, in relation to induction is shown in Figure 5.
These results show that the amount of DNA was highest in
scale-down cultivations, followed by pilot-scale experi-
ments, and was lowest in the control experiment. The ac-
cumulation rate increased after induction in all cultivations.
Notably, the pilot-scale experiment in which the recombi-
nant gene had been removed was similar to the lab-scale
control experiment, which shows that production of rhGH is
an additional source for cell lysis. The higher cell lysis in
scale-down experiments may partially explain the lower
biomass in these cultivations.

Product Formation
The total yield of product, including correct GH monomer
and its variants, is shown in Figure 6a. The general trend is

Figure 3. Formate and acetate accumulation during cultivations. (a)

Bench-scale control experiment, (b) scale-down experiments, and (c) pilot-
scale experiments. Repeated cultivations in the different reactors are num-
bered (1), (2), and (3) and marked with different symbols. Formate con-
centration marked as (1) , (2) , and (3) , respectively. Acetate con-
centration marked as (1) , (2) , and (3) , respectively. Pilot-scale
cultivation (3) corresponds to the experiment in which the recombinant
gene was removed from the plasmid.

time in the batch phase, where the higher the concentration

measured, the longer the period in the phase. The reason
why acetate was also produced during the late exponential
feed phase is probably due to the high specific growth rate.
It has been shown in the literature that overflow metabolism
occurs at substrate uptake rates comparable to a growth rate
of 0.30 to 0.35 h1 (in minimal medium), and to avoid
acetate formation the flux should be kept below this value
(Lee, 1996). Acetate was mainly consumed during the first
hours of the subsequent constant feeding phase.
Formate was produced after induction on a laboratory
scale. This byproduct accumulated rapidly up to a concen-
tration of 500 mg/L. In the SDR set-up, both metabolites
were produced in the PFR at all times (results not shown), Figure 4. Comparison of the different reactors with respect to (a) cell
growth shown as biomass (au), and (b) formate concentration (milligrams
indicating that acetate was more rapidly taken up by the
per liter). Cultivations marked as: control ; SDR ; and repeated pilot-
cells compared with formate, because acetate accumulated scale cultivations (1) , (2) , and (3) , respectively. Pilot-scale culti-
only in the beginning of the cultivation. Formate increased vation (3) corresponds to the experiment in which the recombinant gene
up to a value of 1000 mg/L in the latter part of the SDR was removed from the plasmid.

124 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 69, NO. 2, JULY 20, 2000
Figure 5. DNA concentration in the medium (micrograms per milliliter).
The amount of DNA is correlated with the degree of cell lysis in the
experiments. Cultivations marked as: control ; SDR and ; and re-
peated pilot-scale cultivations (1) , (2) , and (3) , respectively. Pilot-
scale cultivation (3) corresponds to the experiment in which the recombi-
nant gene was removed from the plasmid.

a linear increase of product but at a lower level and rate in

the pilot-scale cultivations. This may be partially attribut-
able to the quality of the inoculum, and further points to the
difficulties of the scale-up procedure. The inoculum prepa-
ration is comparatively simpler in the laboratory, where
these experiments showed comparable results. The final
product mean values per cell dry weight in lab-scale (con-
trol and SDR) and pilot-scale experiments were 64.6 9.4,
75.8 11.1, and 41.7 8.2 au, respectively. The amount of
correct monomer (Fig. 6c) was, at the end of the cultiva-
tions, 70.0 1.8, 76.2 1.9, and 72.0 1.8% in the control,
scale-down, and pilot-scale experiments, respectively (val-
ues expressed with 95% confidence interval).
The amounts of proteolytically degraded forms of the
product were different in the experiments. Figure 6b shows
that the amount of B, which includes LMW, clip-2,
(Met(O)14)-rhGH, and deamidated rhGH, was lowest in
scale-down experiments, with about 30% less B in the
final sample. The dominating form here was clip-2 and the
differences in the amount of trisulfide, for instance, were
smallonly about 1% higher during pilot-scale cultivations
(result not shown). There were no apparent differences in
proteolysis between the laboratory-scale control and the pi-
lot-scale productions.
DISCUSSION
The comparison of pilot-scale and scale-down two-
compartment experiments shows that the ruling regime on a
pilot scale is mimicked in the SDR. This was concluded
from the findings that the pattern of formate and DNA ac-
cumulation could be simulated. However, there was a dif-
ference in the absolute value, where the highest values were
found in the SDR, followed by the pilot-scale and the lab-
Figure 6. (a) Total amount of rhGH per biomass (au) that includes cor-
rect GH monomer and its variants; (b) amount of B (% of total GH),
which includes LMW, clip-2, (Met(O)14)-rhGH, and deamidated rhGH;
and (c) amount of correct GH monomer (% of total GH). Cultivations are
marked: control, (1) ; SDR, (1) , (2) and (3) ; and pilot scale, (1)
and (2) . Values are shown with 95% confidence interval.
scale control experiments, respectively. The pilot-scale re-
actor was mimicked in the SDR, but was exaggerated. No-
tably, the amounts of formate were detected even though the
dissolved oxygen showed a value of 30% air in the stirred
tank reactor. Comparisons of protein production show that
the substrate oscillations decreased the accumulation of pro-
teolytically degraded forms of rhGH.
Processes for recombinant protein production are often
difficult to control due to the number of different phases,
such as growth phase (inoculum procedure, exponential and
constant growth as well as declining growth), production
phase (time of induction), and harvest. In the present inves-
tigation, the cultivations passed through different control
regimes. Initially, variations in biomass concentration of the
inoculum were found to have a pronounced effect on a

BYLUND ET AL.: SCALE-UP EFFECTS ON RECOMBINANT HUMAN GROWTH HORMONE PRODUCTION 125
number of parameters, including biomass growth and ac-
etate formation. This was seen in pilot-scale cultivations as
well as on a laboratory scale. Before induction, the cultiva-
tion was shifted to feed-controlled growth where the effect
of overflow metabolism (acetate) was most dominant, but
oxygen limitation due to scale effects was also clearly ob-
served. At the point of induction, the increase in cell lysis
was severely interfering with the process performance, giv-
ing several effects: (i) oxygen limitation, although the DOT
was >30% air saturation, which was seen from the increas-
ing amounts of formate; and (ii) lower product formation
than expected. After this point the cultivations were
strongly influenced by the feed-controlled regime, because
any overfeed of glucose might result in a rapid accumula-
tion of the overflow metabolite acetate to lethal amounts,
and limitation of oxygen because of the increased demand.
Furthermore, during the production phase, several param-
eters may interfere: DNA concentration reaches a stable
value, although different depending on scale; increased pro-
teolytical breakdown of the product; increasing mainte-
nance effects (reduced biomass); and a 100% increase in
oxygen demand and carbon dioxide production per glucose
added. The results point to the dynamics of recombinant
processes and the need for efficient feedback control where
critical parameters need to be known and measured in a
real-time scale.
Despite the fact that recombinant cultivations performed
on a pilot scale and in a control experiment showed a dif-
ferent pattern with regard to glucose concentration, respira-
tion (YO2/S and YCO2/S), acetate, and formate accumulation,
similar biomass growth was generally obtained. The bio-
mass yield (YX/S) was 0.4 g/g in the cultivations during the
main part of the growth phase, but this value was succes-
sively decreased some hours before induction; therefore, the
reduction took place mainly during the production phase.
This decrease occurred due to increased maintenance when
growth declined (Fig. 1). A similar maintenance effect has
been shown during IPTG (isopropyl--D-thiogalactoside)
and temperature induction of other recombinant proteins
(Bylund et al., 1998; Strandberg et al., 1994). The decrease
in biomass yield was larger in the scale-down experiments,
resulting in a 6% to 10% lower final biomass. The reason
for this was probably the stress introduced due to the fluc-
tuations of glucose and oxygen concentrations and the fol-
lowing turn on/off of metabolism, which caused the forma-
tion of formate and also higher cell lysis (Figs. 3b and 5).
However, this did not lead to increased degradation of prod-
uct. The explanation why the biomass yield coefficient was
lower in all the cultivations than the normal value for E. coli
grown on glucose (0.5 g/g) is probably the high plasmid
burden under uninduced conditions. It has been shown by
Seo and Bailey (1985) that carrying an increased number of
plasmid copies results in a decreased cell yield and specific
growth rate. Furthermore, removing the recombinant gene
from the plasmid gave a similar low-yield coefficient of 0.4
g/g. After induction, the growth declined in this cultivation
as in the others, but at a lower rate (Fig. 1c). Therefore, the
126 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 69, NO. 2, JULY 20, 2000
production of the recombinant protein had, in this specific
case, a minor contribution to the decrease in yield coeffi-
cient.
The acetate concentration in the medium showed an ex-
pected pattern for most experiments; that is, an increased
concentration during the batch and exponential feed phase
followed by consumption during the constant feed (Fig. 3).
The formation of acetate during the exponential feed phase
can, however, be avoided by choosing a specific growth rate
lower than 0.3 to 0.35 h1 (Lee, 1996). The acetate forma-
tion pattern differed in control and one pilot-scale cultiva-
tion compared with the others. In these cultivations, acetate
was not consumed at the same level during the early con-
stant feed. There was, however, a difference in that the
batch phase was longer in these experiments. The acetate
was consumed immediately after induction and decrease in
feed rate. This indicates that the cell physiology differed in
these cultivations as a result of the longer batch phase.
The most likely reason for the accumulation of formate
seen in all cultivations is the appearance of local low oxy-
gen concentrations. This byproduct formation occurred de-
spite the fact that dissolved oxygen tension was >30% air
saturation. The increased level of formate was obtained ear-
lier in pilot-scale cultivations when compared with homo-
geneous lab-scale experiments (Fig. 4b) in which the for-
mate concentration increased first after induction. On the
other hand, when the recombinant gene was removed from
the plasmid, the pilot-scale cultivation was similar to con-
trol experiments in this respect. This indicates that the de-
gree of oxygen limitation was influenced by both the induc-
tion as well as its scale dependence.
The formate concentration in scale-down experiments
showed the highest value and the concentration also in-
creased well before induction. These results clearly show
that the accumulation of formate in large-scale bioreactors
can be predicted by introducing an oscillating environment
for substrates fed to the process. The presence of substrate
gradients and oscillating conditions has been shown in other
large-scale bioreactors by several investigators (Bylund et
al., 1998, 1999; George et al., 1998), and in this study was
indicated by the variance of the glucose values in the pilot
reactor (Fig. 2). Furthermore, it has been shown in batch
experiments that the synthesis of specific proteins during a
shift from aerobic conditions to anaerobiosis is in part simi-
lar to the response of cells shifted to glucose starvation
(Nystrom, 1994). However, no increase in formate or ac-
etate concentrations was obtained during glucose starvation
conditions, even though the enzymes for the corresponding
pathways were present. The rate of protein synthesis also
differed for a shift to anaerobiosis and glucose starvation
conditions and was on the order of minutes to hours, re-
spectively. Therefore, the low growth rate and the reduced
feed rate in our experiments, which probably result in severe
glucose-limiting conditions, were most likely not the rea-
sons for the production of formate.
The amount of DNA in the medium, which is released
due to cell lysis, showed a strong correlation with the ac-
cumulation of formate (Figs. 3 and 5). On the one hand,
oxygen limitation caused by the increased respiratory de-
mand might have caused the formate formation and cell
lysis, which in turn released DNA. On the other hand, the
presence of DNA due to induction could decrease the dif-
fusion rate of oxygen to the cell surface, thus causing an
increased apparent saturation constant for this substrate. In
any case, the results show that the concentration of oxygen
in the reactor can be rate limiting even at higher values; that
is, a higher DOT is needed to avoid mixed acid fermentation
products such as formate, acetate, ethanol, succinate, and
lactate.
One may observe, in Figure 5, that the amount of cell
lysis follows the degree of exposure to heterogeneity in
recombinant protein-producing cultivations. The most het-
erogeneous environment is thought to be in the SDR, fol-
lowed by the pilot-scale reactor and the homogeneous lab-
scale reactor (control). However, it seems to be the higher
rate of release of DNA that causes the earlier accumulation
of formate rather than its absolute value. This is since the
final formate concentration is similar in control and pilote-
scale experiments while the final DNA concentration dif-
fers. A reason why the DNA concentration is not propor-
tional to the formate accumulation might be that the size of
the DNA change with cultivation time due to shear damage.
During the process, there was a significant difference in
product accumulation and product quality of the rhGH be-
tween experiments. It was found that the amount of correct
GH monomer increased by approximately 10% when the
cells were exposed to an oscillating high-glucose, low- oxy-
gen concentration in the two-compartment reactor. The in-
creased quantity of GH monomer was comparable to a
lower occurrence of proteolytically cleaved forms (i.e., the
amount of B, with emphasis on clip-2). This cleavage site
has been identified to amino acid residues Thr(142)
Tyr(143), which is an exposed domain easily accessible to
proteolytic enzymes (Gellerfors, 1990). It may therefore be
a connection between the degree of substrate gradients and
proteolysis of a specific target protein when accumulated in
the periplasm. It is known that ATP-dependent proteolysis
of staphylococcal protein A (SPA) is reduced when glucose
concentration is kept low in the medium (Yang, 1995).
However, most likely, growth hormone is not cleaved by an
energy-dependent protease because the product is accumu-
lated in the periplasm. The proteolytic cleavage could, on
the other hand, occur in the cytoplasm before or during the
translocation to the periplasm and the formation of intact
disulfide bridges. The concentration of some recombinant
proteins in batch cultivations has been shown to reach a
maximum value just before the glucose is exhausted and
declines thereafter. This degradation is assumed to be due to
glucose starvation. Carbon-starved E. coli is, for instance,
known to increase the degradation of the bulk protein at an
average rate from about 1% to 2% to 4% to 5% (Reeve et
al., 1984). Therefore, the conclusion is that, in spite of the
increased heterogeneity in the microenvironment, the cells
BYLUND ET AL.: SCALE-UP EFFECTS ON RECOMBINANT HUMAN GROWTH HORMONE PRODUCTION
are more productive. However, this is at the expense of
growth, because the cell mass is decreased and the cell lysis
is increased by the oscillations. The reason could be that
cells have a longer relaxation time when going from high to
low glucose concentration than vice versa. This means that
the cells imagine being in a richer environment for a
longer time and hence no signals of severe limitation call for
degradation of high-energy internal products. An alternative
is that the oxygen limitation following the high glucose
concentration reduces the amounts of otherwise formed pro-
teases that degrade the product. At first sight, the latter
seems more likely, because the greatest difference between
cultivations was in the accumulation of formate and not in
acetate. However, it should be noted that, even though ac-
etate did not accumulate, it might have been formed but also
consumed again when not exposed to zones with high-
glucose, low-oxygen concentrations.
CONCLUSIONS
Precultivation stages must be reproducible to give reliable
process results and thereby batch variations should most
likely be avoided. The amount of glucose added per reactor
volume and time must be adapted to the actual amount of
biomass at feed start. It is also critical to follow and control
the substrate feed with the same accuracy in all scales.
Better control will result in the same cell physiology at the
start of the production phase. The complex dynamics of
recombinant processes, especially after induction, points to
the need for efficient real-time feedback control of the glu-
cose feed.
The oscillations in the scale-down reactor result in better
production and product quality compared with homoge-
neous laboratory-scale and pilot-scale cultivations. How-
ever, the SDR mimics patterns of substrate and metabolite
accumulation of the pilot-scale process for growth rate, sub-
strate uptake rate, production of acetate/formate and the
accumulation interdependency, and cell lysis (i.e., a quali-
tative resemblance). This was not found in absolute values.
But, there was a trend in the experiments, especially for
formate concentration and cell lysis, of increasing numbers
from laboratory control, to pilot-scale, to SDR production.
This means that oxygen limitation periods are probably
longer and/or relaxation times shorter in the SDR. A future
design might, therefore, be based on an SDR with a broader
residence time distribution.
The authors thank Anders Persson (Pharmacia & Upjohn) who
performed all of the small-scale experiments.
NOMENCLATURE
DOT dissolved oxygen tension (%)
OTR oxygen transfer rate (mol/Lh)
PFR plug-flow reactor
SDR scale-down reactor
CSTR continuous stirred tank reactor
vvm volume air per reactor volume and minute (L/Lmin)
127
 YCO2/S yield coefficient of carbon dioxide per glucose added (g/g)
YO2/S yield coefficient of oxygen per glucose added (g/g)
YX/S yield coefficient of biomass per glucose added (g/g)
Greek symbols
specific growth rate (h1)
References
Bylund F, Collet E, Enfors S-O, Larsson G. 1998. Substrate gradient for-
mation in the large-scale bioreactor lowers cell yield and increases
by-product formation. Bioproc Eng 18:171180.
Bylund F, Guillard F, Enfors S-O, Tragrdh C, Larsson G. 1999. Scale
down of recombinant production: a comparative study of scaling per-
formance. Bioproc Eng 20:377389.
Cleland N. 1988. Sensors for bioprocess control. PhD thesis, Royal Insti-
tute of Technology (KTH), Stockholm.
Forsberg G, Forsgren M, Jaki M, Norin M, Sterky C, Enhorning , Larsson
K, Ericsson M, Bjork P. 1997. Identification of framework residues in
a secreted recombinant antibody fragment that control production level
and localization in Escherichia coli. J Biol Chem 272:1243012436.
Fowler JD, Dunlop EH. 1989. Effects of reactant heterogeneity and mixing
on catabolite repression in cultures of Saccharomyces cerevisiae. Bio-
technol Bioeng 33:10391046.
Gellerfors P, Pavlu B, Axelsson K, Nyhlen C, Johansson S. 1990. Sepa-
ration and identification of growth hormone variants with high perfor-
mance liquid chromatography techniques. Acta Paediatr Scand 370:
93100.
George S, Larsson G, Enfors S-O. 1993. A scale-down two-compartment
reactor with controlled substrate oscillations: Metabolic response of
Saccharomyces cerevisiae. Bioproc Eng 9:249257.
George S. 1997. Effects of high sugar concentrations in a production scale
bioreactor: a scale-down study of S. cerevisiae. PhD thesis, Royal
Institute of Technology (KTH), Stockholm.
George S, Larsson G, Olsson K, Enfors S-O. 1998. Comparison of the
bakers yeast process performance in laboratory and production scale.
Bioproc Eng 18:135142.
Guillard F. 1999. Mixing patterns in Rushton-turbine-agitated reactors.
PhD thesis, Lund University, Lund.
Larsson G, Enfors S-O. 1988. Studies of insufficient mixing in bioreactors:
effect of limiting oxygen concentrations and short term oxygen star-
vation on Pencillium chrysogenum. Bioproc Eng 3:123127.
Larsson G, Tornkvist M. 1996. Rapid sampling, cell inactivation and
evaluation of low extracellular glucose concentrations during fed-
batch cultivation. J Biotechnol 49:6982.
Larsson G, Tornkvist M, Sthl Wernersson E, Tragrdh C, Noorman H, Vollbrecht D. 1982. Restricted oxygen supply and excretion of metabolites.
Enfors S-O. 1996. Substrate gradients in bioreactors: origin and con-
sequences. Bioproc Eng 14:281289.
Lee SY. 1996. High cell-density culture of Escherichia coli. Trends Bio-
technol 14:98105.
128 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 69, NO. 2, JULY 20, 2000
Manfredini R, Cavallera V, Marini L, Donati G. 1983. Mixing and oxygen
transfer in conventional stirred fermentors. Biotechnol Bioeng 25:
31153131.
Moes J, Griot M, Keller J, Heinzle E, Dunn IJ, Bourne JR. 1985. A
microbial culture with oxygen-sensitive product distribution as a po-
tential tool for characterizing bioreactor oxygen transport. Biotechnol
Bioeng 27:482489.
Neubauer P, hman M, Tornkvist M, Larsson G, Enfors S-O. 1995a. Re-
sponse of guanosine tetraphosphate to glucose fluctuations in fed-
batch cultivations of Escherichia coli. J Biotechnol 43:195204.
Neubauer P, Haggstrom L, Enfors S-O. 1995b. Influence of substrate os-
cillations on acetate formation and growth yield in Escherichia coli
glucose limited fed-batch cultivations. Biotechnol Bioeng 47:139146.
Nystrom T. 1994. The glucose-starvation stimulon of Escherichia coli:
induced and repressed synthesis of enzymes of central metabolic path-
ways and role of acetyl phosphate in gene expression and starvation
survival. Mol Microbiol 12:833843.
Oosterhuis NMG, Groebeek NM, Olivier APC, Kossen NWF. 1983. Scale-
down aspects of the gluconic acid fermentation. Biotechnol Lett 5:
141146.
Oosterhuis NMG, Kossen NWF. 1984. Dissolved oxygen concentration
profiles in a production-scale bioreactor. Biotechnol Bioeng 26:
546550.
Oosterhuis NMG, Kossen NWF, Olivier APC, Schenk ES. 1985. Scale-
down and optimization studies of the gluconic acid fermentation by
Gluconobacter oxydans. Biotechnol Bioeng 27:711720.
Reeve CA, Bockman AT, Matin A. 1984. Role of protein degradation in
the survival of carbon-starved Escherichia coli and Salmonella typhi-
murium. J Bacteriol 157:758763.
Seo J-H, Bailey E. 1985. Effects of recombinant plasmid content on growth
properties and cloned gene product formation in Escherichia coli.
Biotechnol Bioeng 27:16681674.
Strandberg L, Andersson L, Enfors S-O. 1994. The use of fed batch cul-
tivation for achieving high cell densities in the production of a recom-
binant protein in Escherichia coli. FEMS Microbiol Rev 14:5356.
Sweere APJ, Mesters JR, Janse L, Luyben KCAM. 1988a. Experimental
simulation of oxygen profiles and their influence on bakers yeast
production. I. One-fermentor system. Biotechnol Bioeng 31:567578.
Sweere APJ, Janse L, Luyben KCAM. 1988b. Experimental simulation of
oxygen profiles and their influence on bakers yeast production. II.
Two-fermentor system. Biotechnol Bioeng 31:579586.
Sweere APJ, Matla YA, Zandvliet J, Luyben KCAM, Kossen NWF. 1988c.
Experimental simulation of glucose fluctuation. Appl Microbiol Bio-
tech 28:109115.
Eur J Appl Microbiol Biotechnol 15:111116.
Yang S. 1995. Influence of proteolysis on production of recombinant pro-
teins in Escherichia coli. PhD thesis, Royal Institute of Technology
(KTH), Stockholm.