1038 Biotechnol. Prog.

2005, 21, 1038−1047

Development of a Robust, Versatile, and Scalable Inoculum Train

for the Production of a DNA Vaccine
J. Okonkowski, L. Kizer-Bentley, K. Listner, D. Robinson, and M. Chartrain*
Bioprocess R&D, Merck Research Laboratories, PO Box 2000, Rahway, New Jersey 07065

For many microbial fermentation processes, the inoculum train can have a substantial
impact on process performance in terms of productivity, profitability, and process
control. In general, it is understood that a well-characterized and flexible inoculum
train is essential for future scale-up and implementation of the process in a pilot plant
or manufacturing setting. A fermentation process utilizing E. coli DH5 for the
production of plasmid DNA carrying the HIV gag gene for use as a vaccine is currently
under development in our laboratory. As part of the development effort, we evaluated
inoculum train schemes that incorporate one, two, or three stages. In addition, we
investigated the effect of inoculum viable-cell concentrations, either thawed or actively
growing, over a wide range (from 2.5 × 104 to 1.0 × 108 viable cells/mL or approximately
0.001% to 4% of final working volume). The various inoculum trains were evaluated
in terms of final plasmid yield, process time, reproducibility, robustness, and feasibility
at large scale. The results of these studies show that final plasmid yield remained in
the desired range, despite the number of stages or inoculation viable-cell concentrations
comprising the inoculum train. On the basis of these observations and because it
established a large database, the first part of these investigations supports an
exceptional flexibility in the design of scalable inoculum trains for this DNA vaccine
process. This work also highlighted that a slightly higher level of process reproduc-
ibility, as measured by the time for the culture to reach mid-exponential growth, was
observed when using actively growing versus frozen cells. It also demonstrated the
existence of a viable-cell concentration threshold for the one-stage process, since we
observed that inoculation of the production stage with very low amounts of viable
cells from a frozen source could lead to increased process sensitivity to external factors
such as variation in the quality of the raw materials used in the medium formulation.
However, our analysis indicates that, despite this slight disadvantage, a one-stage
inoculum train was a viable option in many situations, especially if the inoculation
viable-cell concentration was kept above 4.8 × 106 viable cells/mL. Because it leads to
a reduction in process steps and eliminates some capital investments (i.e., inoculum
fermenter), when feasible a one-stage process configuration will positively impact
process economics.

Introduction al., 2001). Unlike other vaccines, DNA vaccine dosages

DNA vaccines, also referred to as genetic vaccines, are
a novel method of vaccination that has the potential to
elicit both cellular and humoral immune responses. The
applications of DNA vaccinations are explored in protec-
tion against intracellular pathogens (Shroff et al., 1999;
Gurunathan et al., 2000a,b; Berzofsky et al., 2000;
Leitner et al., 2000; Bergmann-Leitner and Leitner, 2004;
Powell, 2004) and gene therapy and cancer treatment
(Mountain, 2000; Ferber, 2001; Leitner and Thalhamer,
2003; Berzofsky et al., 2004). Briefly, plasmids coding for
a specific protein are transformed into and mass pro-
duced in the bacterial host Escherichia coli (Prather-
Jones et al., 2003). Upon purification, the plasmids are
formulated and injected into the muscle of the patient.
DNA vaccination is one of the strategies investigated at
Merck Research Laboratories in the development of a
vaccine against HIV (Barouch et al., 2000; Barouch et
* To whom correspondence should be addressed. Email:
chartrain@merck.com.
10.1021/bp040041p CCC: $30.25 © 2005 American Chemical Society and American Institute of Chemical Engineers
Published on Web 07/07/2005
are large, to the order of several milligrams per dose
(Babuik et al., 2000; Donnelly et al., 2003), and therefore
require extensive production efforts. To support positive
economics, production of large amounts of plasmid re-
quires that the entire cultivation process be efficient and
consistent while containing operating costs.
The development of a microbial inoculum has the
potential to greatly impact process performance in terms
of productivity, reproducibility, and economics (Mar-
tinkova et al., 1991; Warr et al., 1996; Ignova et al., 1999).
Although rarely individually reported in the literature,
inoculum trains are an intrinsic and crucial part of any
large-scale industrial bioprocess. Not only is the develop-
ment of a reliable and well-characterized inoculum train
a sine qua non condition for any well-designed com-
mercial process, in addition, an appropriately docu-
mented and reproducible inoculum supports the regula-
tory requirements necessary for any pharmaceutical
bioprocess to secure regulatory approval for commercial-
ization (FDA, 1985; Falk and Ball, 2001).
Biotechnol. Prog., 2005, Vol. 21, No. 4
Traditionally, microbial inocula originate from a frozen
source, typically a small vial that is expanded via
successive transfers into shake flasks of increasing
volume. The shake-flask culture is subsequently used to
inoculate a small bioreactor, thereby initiating a chain
of additional bioreactor-based transfers that continues
until inoculation of the production fermentor, in sizes
reaching up to 100 000 L or more. In a production
environment regulated by current Good Manufacturing
Practices (cGMP), inoculum trains can be procedurally
complex. For example, unlike transfers made from biore-
actor to bioreactor via steam-sterilized lines, transfers
involving the use of shake flasks require that the vessels
be opened several times during the cultivation of the
microorganisms. These operations are cumbersome, in-
crease the risk of contamination, and require extensive
testing of the cultures and of the surrounding environ-
ment when performed under cGMP conditions (FIP, 1990;
PDA, 2001). Inoculum train configurations, including the
volumes transferred and the number of stages, can
significantly impact process flexibility, ease of implemen-
tation, and ultimately economics. For example, when
facing the challenge to meet increasing program material
requirements while progressing from clinical phases to
full-scale production, the configuration of the available
facility and reactor size is likely to change. Understand-
ing the limits of inoculum train flexibility and robustness
carries the potential to lead to increased process efficiency
and productivity by limiting capital investment and
reducing process cycle time.
The motivation for the studies presented here was two-
fold. First, it was to establish a database detailing the
influence of inoculation viable-cell concentration and
inoculum train configuration on process performance.
Once this database was established, the second motivat-
ing factor was to study the effect of key process param-
eters on the performance of the most desirable strategy
and evaluate implementation with respect to the size of
the operations. Relevant data along with their scientific
and operational analyses, supporting the implementation
of flexible and scaleable inoculum train strategies under
various operational constraints, are presented here.
Materials and Methods
Chemicals. All chemicals used were of reagent grade
and were purchased from the following companies: Fisher
Scientific (Springfield, NJ), Sigma Chemical Company
(St. Louis, MO), Aldrich Chemical Co. (Milwaukee, WI),
JT Baker (Phillipsburg, NJ), Gibco BRL (Grand Island,
NY), Mediatech, Inc. (Herndon, VA), Mallinckrodt (Paris,
KY), EM Science (Cherry Hill, NJ), ICN (Costa Mesa,
CA), Difco Becton Dickinson (Sparks, MD), and Strat-
agene (La Jolla, CA).
Microorganisms. The construct used was E. coli
strain DH5 (Gibco BRL) carrying the plasmid V1Jns-FL-
opt-gag (6393 bp, constructed in-house), which contains
a ColEI origin, a cytomegalovirus promoter (CMV), a
kanamycin/neomycin resistance gene, and the HIV gag
gene (Barouch et al., 2000; Barouch et al., 2001). Through-
out this publication, this construct will be referred to as
HIV-FL-gag.
Solution Composition. The following solutions were
used during the course of the investigations. Pre-DME-
P4 contained (per L of distilled H2O) KH2PO4, 7.0 g; K2-
HPO4, 7.0 g; (NH4)2SO4, 6.0 g; monosodium glutamate
(MSG), 5.0 g; glycerol, 10.0 g. Flask supplement solution
(FSS) contained (per L of distilled H2O) MgSO4‚7H2O, 60
g; thiamine HCl, 6 g; neomycin sulfate, 2.4 g. Trace
1039
elements solution (TES) contained (per L of 1.2 N
hydrochloric acid) FeCl3‚6H2O, 27 g; ZnCl2, 2 g; CoCl2‚
6H2O, 2 g; Na2MoO4‚2H2O, 2 g; CaCl2‚2H2O, 1 g; CuCl2‚
2H2O, 1.27 g; H3BO3, 0.5 g. FSS and the TES were each
0.22 µm vacuum-filter sterilized and stored at -70 and
4 °C, respectively. Seed supplement solution (SSS) con-
tained (per L of distilled H2O) MgSO4‚7H2O, 240 g;
thiamine HCl, 24 g; neomycin sulfate, 9.6 g. DME-P4
medium was made up by combining 1 mL of TES and
8.3 mL of SSS to 1 L of pre-DME-P4. STET buffer
contained (per L distilled H2O) Tris EDTA buffer (Sigma),
50 mL; 0.5 M EDTA pH8 (Sigma), 190 mL; sucrose, 80
g; Triton X-100 (Sigma), 20 g. Lysozyme solution con-
tained (per L STET buffer) lysozyme (Sigma), 0.4 g. CM
80 buffer contained (per L of water) NaCl, 7 g; K2HPO4,
0.674 g; KH2PO4, 0.2 g.
Preparation of Small-Scale Working Cell Banks.
An aliquot of frozen cells was used to inoculate three 2-L
baffled shake flasks (each flask received 200 µL) contain-
ing 200 mL of DME-P4 medium. The flasks were incu-
bated at 37 ( 0.5 °C with shaking (180 rpm on an orbital
shaker with a 50 mm shaking diameter). Two of the
flasks were used for monitoring purposes, and the third
flask was used to prepare the frozen cell suspension.
When the optical density (at 600 nm) of the culture in
the monitoring flasks reached a value in the range of
5-9, the optical density of the source flask was then
checked to ensure that it had also reached a similar
value. The source flask was then chilled on wet ice. An
equal volume of chilled glycerol solution (40% v/v in
water) was added to the contents of the flask. The mixed
suspension was dispensed (ca. 1 mL) in cryovials that
were immediately flash frozen on dry ice and stored at
-65 °C or below. This first cell bank is labeled pre-
master.
A frozen suspension from a pre-master vial was thawed
and used to prepare the master cell bank according to
the protocol described above. Finally, a frozen suspension
from a master vial was thawed and used to prepare the
working cell bank (vials containing ca. 4 mL) according
to the same protocol.
Preparation of Large-Scale Working Cell Banks.
Shake-Flask Cultivation. Two 250-mL baffled sterile
Erlenmeyer flasks were filled with 30 mL of DME-P4 and
inoculated by pipetting 30 µL from a vial of HIV-FL-gag
master cell bank. The flasks were incubated in the same
conditions described above until the optical density at
600 nm (OD600) of the flask reached a value between 5
and 9 (approximately 26 h post-inoculation). A volume
of 150 mL of the culture was used to inoculate a 30-L
bioreactor (B. Braun Biotech, Inc.; Allentown, PA) con-
taining 15 L of pDME-P4. The bioreactor was prepared
by sterilizing the pDME-P4 medium in situ for 30 min
at 121 °C. Post-sterilization, 125 mL of SSS and 15 mL
of TES were added to the bioreactor, and the pH of the
DME-P4 medium was adjusted to 7.1 ( 0.1 with a sterile
30% NaOH solution. The cultivation conditions were
temperature, 37 °C; back-pressure, 4.5 psig; airflow, 7.5
slpm. The dissolved oxygen level (DO %), carbon evolu-
tion rate (CER), oxygen uptake rate (OUR) (Prima V
mass spectrometer model 600, Thermo ONIX Corp.,
Houston, TX), cell density (OD probe, Monitek, Bedford,
MA), and pH were simultaneously measured and re-
corded on-line. The DO was maintained to a set-point
(greater or equal to 30% of air saturation) by automatic-
feedback cascade control of the agitation speed (between
250 and 750 rpm). The culture pH was maintained to
7.1 ( 0.1 by automatic addition of a sterile 30% NaOH
solution. Once mid-exponential growth was attained, as
1040 Biotechnol. Prog., 2005, Vol. 21, No. 4

Figure 1. Schematics of the three processes evaluated. (A) Three-stage baseline inoculum train. A frozen cell suspension (4 mL) is
used to inoculate a 2.8-L baffled shake flask containing 500 mL of medium. After overnight cultivation, an OD600 of 5-9 indicative
of mid-exponential growth is achieved and the culture is used to inoculate a 30-L inoculum reactor containing 15 L of medium (if a
larger inoculum reactor is used, the contents of several 2-L flasks can be pooled). Once the culture reaches mid-exponential growth
(about 12 h), the culture is used to inoculate the production bioreactor (in sizes ranging from 30 to 2000 L or larger). Once the
culture reaches mid-exponential growth, feeding of a nutrient solution [50% glycerol/25% MSG (w/v)] is initiated and lasts
approximately 40 h. (B) Two-stage inoculum train. A large frozen inoculum stock (300 mL or larger) is used to inoculate a 30-L
inoculum bioreactor, containing 15 L of medium. Once the culture reaches mid-exponential growth (about 10 h), the culture is used
to inoculate the production bioreactor. The remainder of the process is similar to that described above. (C) One-stage inoculum
train. The one-stage process consists of the direct inoculation of the production reactor (30 L or larger) with a frozen inoculum stock
(300 mL or larger). The remainder of the process is similar to that described above.

indicated by an on-line CER value of 35 mmol/L/h, 5 L
of the culture was aseptically transferred into a sterile
10-L Nalgene carboy containing 5 L of a cold 50% glycerol
solution. The carboy was placed on a magnetic stirring
plate in a biological safety cabinet and was continuously
mixed to provide even suspension of the cells throughout
the filling procedure (500-mL Nalgene bottles each
receiving 300 mL of the glycerol/culture solution). The
filled bottles were initially cooled by covering them with
dry ice and then were transferred to a -70 °C freezer
for storage. These bottles constituted the large-scale
working cell bank used throughout these studies.
Plasmid Production Procedure. Three-Stage Pro-
cess. (Figure 1A). A 2-L baffled shake flask containing
500 mL of DME-P4 medium was inoculated with 4 mL
of culture from a frozen vial (approximately 0.8% of
working volume). After overnight cultivation, an OD600
of 5-9, indicative of mid-exponential growth, was
achieved. The culture was used to inoculate a 30-L
inoculum reactor containing 15 L of DME-P4 medium.
Biotechnol. Prog., 2005, Vol. 21, No. 4 1041

The reactor was prepared and operated as described

previously. Once the CER of the culture reached a value
of 35 mmol/L/h (indicative of mid-exponential growth), a
volume of culture was transferred to a 30-L production
bioreactor containing 15 L of DME-P4 medium. During
the first part of the process, the production reactor was
prepared and operated under the same process conditions
as those described previously. Once the CER reached 35
mmol/L/h, feeding of a 50% glycerol/25% MSG (w/v)
solution at a rate of 3.2 g/L/h was initiated. At the same
time, a one-time automatic increase of the airflow and
back-pressure set-points from 7.5 to 12 L/min and from
4.5 to 15 psi, respectively, was performed. Feeding was
maintained at 3.2 g/L/h for approximately 40 h. The pH
was maintained at 7.1 ( 0.1 by feedback-controlled
automatic addition of sterile 15% (v/v) phosphoric acid
and 30% (v/v) sodium hydroxide throughout the fermen-
tation cycle.
Two-Stage Process. (Figure 1B). A large frozen in-
oculum stock (300 mL) prepared as described above was
thawed in a stationary 37 °C water bath (with periodic
vigorous manual shaking) and was used to inoculate a
30-L inoculum bioreactor, containing 15 L of DME-P4
medium. The remainder of the process was essentially
similar to that described above.
One-Stage Process. (Figure 1C). The one-stage pro-
cess consisted of the direct inoculation of the production
reactor with a frozen inoculum stock. The remainder of
the production process was similar to that previously
described.
Analytical Procedures. Sample Preparation and
Cell Lysis. Samples were prepared for lysis by adding a
culture volume required to reach an OD600 of 10 (in 1 mL
final volume) followed by centrifugation (Eppendorf
centrifuge 5415 C, Westbury, NY) for 5 min at 14 000
rpm. The supernatant was discarded, and the pellets
were stored in a -70 °C freezer. Prior to analysis, each
microfuge tube was thawed at room temperature and the
pellet was resuspended in 500 µL of STET buffer,
followed by the addition of 500 µL of 4 g/L lysozyme
solution. The tubes were incubated at 37 °C for 45 min
in an Eppendorf Thermomixer R (Westbury, NY), with
continuous shaking at 500 rpm. After incubation, the
tubes were inserted into a floating rack and set in boiling
water for 1 min. The cell debris was separated by
centrifugation in an Eppendorf centrifuge 5415 C for 15
min at 14 000 rpm. The supernatant of each tube was
transferred into an HPLC vial, and 10 µL of “Rnace-It!”
(Ribonuclease Cocktail, Stratagene, La Jolla, CA) was
added. The vials were capped, and the contents were
mixed prior to HPLC analysis. Figure 2. Plasmid yield and carbon evolution rate (CER)
HPLC Assay. Plasmid DNA was quantified using a profiles for three inoculum train strategies. The calculated
HPLC system (Gilson, Middleton, WI) equipped with a viable-cell concentrations achieved immediately post-inoculation
Waters Gen-Pak FAX column (4.6 mm × 100 mm) for each case were as follows. (A) Three-stage: shake-flask 7.8
(Milford, MA). Separation of the supercoiled plasmid × 106 vc/mL, inoculum bioreactor 2.5 × 107 vc/mL, production
bioreactor 5.0 × 107 vc/mL. (B) Two-stage: inoculum bioreactor
DNA was achieved with a gradient-based elution em- 9.8 × 104, production bioreactor 1.0 × 108 vc/mL. (C) One-
ploying a mobile phase consisting of Buffer A (containing stage: production bioreactor 9.8 × 106 vc/mL. The cultivation
per 973 mL of HPLC-grade H2O: 1 M Tris-HCl pH 8.0, conditions used are described in Materials and Methods.
25 mL; 0.5 M EDTA pH 8.0, 2 mL) and Buffer B
(containing per 773 mL of HPLC-grade H2O: 1 M Tris- (containing per L of HPLC-grade H2O: 85% H3PO4
HCl pH 8.0, 25 mL; 0.5 M EDTA pH 8.0, 2 mL; 5 M NaCl,
HPLC-grade, 4.61 mL] for 12 min. Detection was per-
200 mL) delivered at a rate of 0.75 mL/min. The concen-
tration of Buffer A decreased from 70% to 35% over the formed at 260 nm at 25 °C. Under these conditions,
first 2 min of the assay, while the concentration of Buffer supercoiled DNA eluted after approximately 5 min. The
B increased from 30% to 65%. At that point, the sample specific and volumetric plasmid yields were calculated
was injected into the column and the mobile phase on the basis of a standard curve obtained using pure DNA
continued at 35% Buffer A and 65% Buffer B for 7 min plasmid solutions of known concentrations and by the
while the sample components eluted. Following each automatic integration of the supercoiled DNA peak. The
injection, the column was cleaned by flowing Buffer C final calculations took into account the volume of culture
1042 Biotechnol. Prog., 2005, Vol. 21, No. 4

Table 1. Volumetric and Specific Plasmid Yields for Three Inoculum Train Configurations and Variable Inoculation
Levels
shake-flask inoculum bioreactor production bioreactor volumetric specific
inoculum inoculation concn inoculation concn inoculation concn plasmid yield plasmid yield
train type (vc/mL) (vc/mL) (vc/mL) (g/L) (µg/mg DCW)
three-stage 7.8 × 106 2.5 × 107 5.0 × 107 0.635 33.5
9.8 × 105 2.5 × 107 5.0 × 107 0.626 26.3
9.8 × 105 2.5 × 107 5.0 × 107 0.554 26.6
two-stage 2.0 × 107 1.3 × 107 0.751 34.3
2.0 × 107 1.3 × 107 0.645 28.4
4.9 × 105 2.5 × 106 0.630 26.5
9.8 × 104 1.0 × 108 0.685 34.7
one-stage 9.8 × 106 0.846 30.7
9.8 × 106 0.830 34.2
9.8 × 104 0.634 28.5
4.9 × 105 0.681 31.5

Table 2. Relationship between Total Process Time and Plasmid Productivity

process time of stage (h)
inoculation concn(s) feeding total plasmid
in bioreactor shake inoculum production period process productivity
inoculum train type (vc/mL) flask bioreactor bioreactor (h) time (h) (µg/mg DCW/h)
standard three-stage 7.8 × 106/2.5 × 107/5.0 × 107 18 11 10 40 79 0.42
(0.8%/1%/2%)
two-stage (0.05%/0.1%) 4.9 × 105/2.5 × 107 15 14 40 69 0.44
two-stage (0.01%/4%) 9.8 × 104/1.0 × 108 22 7 40 69 0.46
one-stage (0.01%) 9.8 × 104 22 40 62 0.48
one-stage (4%) 3.9 × 107 9 40 49 0.57

needed to prepare the OD10 pellet and the dry cell weight
measurement (DCW).
Dry Cell Weight. Determinations were made by
filtering 5 mL of culture (a 1:10 dilution of the cell culture
was prepared for all samples with an OD600 > 20 to
decrease filtration time) through a 0.45 µm membrane
(Millipore, Bedford, MA). A 5-mL rinse with CM80 buffer
was completed prior to drying in a household microwave
oven on medium power for 5 min.
Cell Viability Assay. Serial dilutions of the fresh
culture were made in CM80 buffer to prepare 10-6, 10-7,
and 10-8 dilutions. Aliquots of 100 µL from each dilution
were spread onto LB agar plates (five plates per dilution).
The colonies were counted after 24 h of incubation at 37
°C.
Automated Scaled-Down Technology. The auto-
mated scaled-down cultivation platform technology uti-
lized a Labsystems Bioscreen C (Labsystems, Franklin,
MA) controlled by the Growth Curves software (Trans-
galactic Ltd, Helsinki, Finland). The system automati-
cally and simultaneously monitors the growth (via color
or turbidity change) of microorganisms in 200 wells, while
providing mixing. The resulting data (growth curves) are
automatically logged into Microsoft Excel 97.
Variable amounts of inoculum or compounds to test
were added to DME-P4 medium. An amount of 350 µL
of pre-inoculated medium was then aseptically dispensed
into the cultivation wells of the sterile plate. To increase
accuracy, each factor explored was allowed 10 wells and
the data from the 10 wells were averaged after the assay.
The growth conditions of the assay were temperature,
37 °C; detection wavelength, 600 nm; measurement
interval, 20 min; experiment length, 2 days; shaking
interval, continuous; shaking speed, medium.
Results and Discussion
Impact of Inoculum Train Configuration on Plas-
mid Productivity. The first set of investigations com-
pared the growth kinetics and plasmid production per-
formance of the baseline three-stage process against a
two-stage and a one-stage process. Figure 1 depicts these
three processes. The latter two strategies offer the
advantage of reducing the number of operations by
eliminating the usage of shake flasks and thereby lower
the potential for contamination. Consequently, we inves-
tigated these two latter alternate strategies in depth by
evaluating them over a wide range of inoculation viable-
cell concentrations (IVCC). The comparators utilized in
the evaluations were the time for the culture to reach
mid-exponential growth, measured on-line via carbon
evolution rate (CER), and plasmid production yield.
Preliminary experiments employing DME-P4 medium
had established that a CER value of 35 mmol/L/h
correlated with mid-exponential growth.
Three-Stage Process. Figure 2A presents the typical
process growth profiles achieved with the use of the
baseline inoculum train described in Materials and
Methods and outlined in Figure 1A. Figure 2A presents
the CER profile of the production stage. After reaching
a maximum of 75 mmol/L/h approximately 15 h post-
inoculation, the CER stabilized during the feeding portion
of the process and remained between 35 and 45 mmol/
L/h. The plasmid specific yield data presented in Figure
2A show that the yield rapidly increased during the first
15 h of feeding, and later slowly reached a plateau after
approximately 25 h of feeding. The data presented in
Table 1, obtained at the 30-L and 75-L scales, show that
these three batches achieved plasmid production yields
in the range of 26-33 µg plasmid DNA per mg of dry
cell weight (µg/mg DCW) and volumetric plasmid yields
ranging from 554 to 635 mg/L. These yields are fully
consistent with the historical range established at various
scales, from 30-L to 2000-L, over the course of our
investigations. When factoring the inoculum development
and production stages, the total cultivation cycle time of
the specific example presented in Figure 2A was 79 h
(Table 2). Based on this in-process cycle time, excluding
the equipment turnaround, this three-stage process
achieved an overall plasmid specific productivity of 0.42
µg/mg DCW/h.
Two-Stage Process. The two-stage process was per-
formed as described in Materials and Methods and
outlined in Figure 1B. Cells, obtained from a frozen
source, were used to directly inoculate the inoculum
Biotechnol. Prog., 2005, Vol. 21, No. 4
Figure 3. Representative metabolic profiles for various one-
stage inoculation viable-cell concentrations. Various inoculation
viable-cell concentrations (vc/mL post-inoculation) were inves-
tigated. Presented are data obtained with approximate percent-
age of final volume represented by thawed inoculum: (O, b)
inoculum of 3.9 × 107 vc/mL (4% of final volume); (0, 9)
inoculum of 9.8 × 106 vc/mL (1% of final volume); (4, 2)
inoculum of 4.9 × 105 (0.5% of final volume); (), () inoculum of
9.8 × 104 vc/mL (0.01% of final volume). The cultivation
conditions used are described in Materials and Methods.
bioreactor to a variable final IVCC. Figure 3 presents
composite CER profiles for this first cultivation stage and
shows that an increasing delay in reaching mid-expo-
nential growth was observed as the IVCC was reduced.
To further model this process behavior, the time for all
batches to reach mid-exponential growth was plotted,
using a semilog format versus the IVCC. The data
presented in Figure 4A clearly show correlation between
the IVCC and the time for the culture to reach mid-
exponential growth. Since the viable-cell concentration
of the inoculum can always be determined, this correla-
tion allows a logarithmic regression to be used as a tool
in predicting the time to mid-exponential growth for a
vessel of any size inoculated with a frozen inoculum bank.
Growth rates calculated from the on-line off-gas data
revealed that all growth rates (µ) were within in a close
range (0.32 < µ < 0.37 h-1). In addition, the data
presented in Figure 4A clearly show the existence of good
reproducibility between multiple batches when employing
separate inocula with a similar viable-cell concentration.
Similarly, sets of data were collected from production
fermenters inoculated at various IVCC from inocula
produced in an inoculum fermenter and harvested in mid-
exponential growth (Figure 1B). The logarithmic regres-
sion showed a trend similar to that of the previous
example (Figure 4B). However, when compared to the
data presented in Figure 4A, the higher correlation
coefficient is indicative of a higher level of reproducibility
achieved between repeat runs. We speculate that this
higher reproducibility may be attributed to the use of
actively growing versus frozen cells. The recovery and
acclimation period that the cells must endure upon
thawing likely adds an additional source of variability.
The decrease in time to mid-exponential growth observed
with the use of actively growing cells relative to the use
of an equal concentration of thawed cells can also be
similarly explained. As in the previous case, we found
that the concentration of viable cells upon inoculation did
not affect the growth rate.
Figure 2B presents the CER profile of the production
stage. After reaching a maximum of 80 mmol/L/h about
15 h post-inoculation, the CER stabilized during the
1043
Figure 4. Relationship between time to mid-exponential
growth and inoculation viable-cell concentration. Mid-exponen-
tial growth was assessed by the carbon evolution rate (CER)
reaching a value of 35 mmol/L/h (mid-exponential growth). The
inoculation viable-cell concentration (vc/mL post-inoculation) for
(A) one-stage inoculum train and (B) two-stage inoculum train
is the calculated immediate post-inoculation cell concentration
in the bioreactor. The regression line shown here can be used
to predict total future process time using the equations y )
-2.37 ln(x) + 50.5 (one-stage) and y ) -1.94 ln(x) + 62.2 (two-
stage).
feeding portion of the process and remained between 40
and 50 mmol/L/h. The plasmid specific yield data pre-
sented in Figure 2B show that it rapidly increased during
the first 15 h of feeding and later slowly reached a
plateau after about 25 h of feeding. The data presented
in Table 1, obtained from two-stage processes employing
variable IVCC, show that these representative batches
achieved plasmid production yields in the range of 26-
34 µg/mg DCW and volumetric plasmid yields ranging
from 630 to 751 mg/L. Overall productivity of a two-stage
process was found to range from 0.44 to 0.46 µg/mg
DCW/h (Table 2).
One-Stage Process. The one-stage process was per-
formed as described in Materials and Methods and
outlined in Figure 1C. Various IVCC, obtained from a
frozen source, were used to directly inoculate the produc-
tion bioreactor stage.
The growth kinetics of a one-stage process, until
feeding is initiated at mid-exponential growth, are es-
sentially similar to those of the first stage of a two-stage
process (Figure 1C). Therefore, the same relationship
1044
between the IVCC and the time to reach mid-exponential
can be used in process modeling.
Figure 2C presents the CER profile of the production
stage. After reaching a maximum of 70 mmol/L/h ap-
proximately 8 h post-inoculation, the CER stabilized
during the feeding portion of the process to between 35
and 40 mmol/L/h. The plasmid specific yield data pre-
sented in Figure 2C exhibited a behavior similar to those
observed in the production stages of the three- and two-
stage processes. The data presented in Table 1, obtained
from one-stage processes using variable IVCC, show that
these representative batches achieved specific plasmid
production yields in the range of 28-34 of µg/mg DCW
and volumetric plasmid yields ranging from 634 to 846
mg/L. Overall, productivity of a one-stage process was
found to range from 0.48 to 0.57 µg/mg DCW/h (Table
2).
In general, the first part of these evaluations estab-
lished that, when using an inoculum with a known
viable-cell concentration, the time to mid-exponential
growth can be accurately derived. This is a very useful
piece of information in term of process control and
scheduling when scaling up. The data presented here also
showed that all of the inoculum train strategies evaluated
are viable options as they all supported similar growth
kinetics and similar final plasmid yields.
As previously discussed, although the three-stage
process supported acceptable plasmid production, we
believe that the presence of two inoculum stages and the
use of shake-flasks allows room for potential process
improvements. The data presented here fully support the
removal of the shake-flask stage from the baseline
process. In addition, although inoculum train configura-
tion does not appear to impact final plasmid yield, due
to changes in process cycle time, it does affects plasmid
specific productivity. Data compiled in Table 2 outline a
clearly visible relationship between decreasing cycle time
and increasing plasmid specific productivity, and suggest
that a one-stage inoculum train with an IVCC of 3.9 ×
107 vc/mL (or higher) would be the best choice here, since
it supports the highest plasmid specific productivity.
Sensitivity Analyses. Because of increased process
efficiency, a single-stage process appears highly desirable.
However, to be implemented in a manufacturing envi-
ronment, it is important that such an inoculum train be
robust and reliable. As previously noticed, the one-stage
process exhibits a slightly higher level of process vari-
ability in terms of time required to reach mid-exponential
growth, relative to the two-stage inoculum train (Figure
4). This variability is likely to be exacerbated with a lower
IVCC. Evaluating the robustness of the one-stage process
by means of a sensitivity analysis will either support or
rule out its use under stringent conditions (Patnaik,
1996).
We opted to evaluate the effect of changes in external
factors that may occur as a result of discrepancies in
medium and vessel preparation or component quality on
the growth kinetics of our E. coli construct. Variation in
the concentration of trace elements (TES) in the medium
was chosen as a marker to model inconsistencies in the
preparation of the medium. The addition of detergent to
the medium was selected to reflect potential residuals
that may remain in the bioreactor after cleaning. In
addition, the quality of monosodium glutamate (MSG),
a process specific factor previously identified, represented
the potential influence of ingredient lot-to-lot variation.
A scaled-down high-throughput automated cultivation
technology was implemented in order to rapidly evaluate
the robustness of the one-stage inoculum train, employ-
Biotechnol. Prog., 2005, Vol. 21, No. 4
ing a Bioscreen C apparatus. Preliminary experiments
established that the time for the OD600 to increase by 20%
of the initial value (early-exponential growth) was a good
reflection of the duration of the apparent lag phase.
Effect of Variations in Trace Elements. The effect
of variations in the concentration of TES is summarized
in Figure 5A. The time-values plotted versus IVCC for
the four concentrations of TES evaluated in the assay
show a dual effect caused by TES concentrations and
inoculum concentration. The data indicate that the time
to reach early-exponential growth, at a constant IVCC,
increases with increasing TES concentration, although
no other significantly adverse effects on growth were
observed with the higher concentrations. The data also
show that the time to reach early-exponential growth
increased as IVCC decreased. The delaying effect ob-
served was found to be slightly magnified at inoculation
cell concentrations below 4.8 × 106 vc/mL. However, this
could be due to the fact that at higher IVCC, more
accurate optical density readings were achieved. Even
though increases in the time to mid-exponential growth
are apparent at 2× and 4× TES in the medium, this level
of elevated deviation is unlikely to occur with any
regularity.
Effect of Residual Detergent. Next, two concentra-
tions of detergent were evaluated, a 1:10 and a 1:100
dilution of the detergent normally added to the bioreactor
during cleaning (Contrad 70, Decon Laboratories, Bryn
Mawr, PA). These concentrations were chosen to be
significantly larger than the realistic detergent concen-
tration that may remain in the bioreactor after rinsing,
since draining and refilling with water to prepare the
medium would create more than a 1:100 dilution of the
detergent. Figure 5B shows that a relatively large
concentration of residual Contrad 70 detergent in the
medium has little effect on culture growth. Again, and
consistent with expectations, the time to reach early
exponential grew shorter as IVCC increased.
Effect of MSG Lot to Lot Variation. Finally, we
evaluated the effect of three different lots of MSG.
Included were a lot that exhibited no inhibitory effects
on cell growth in the past (lot A), and two lots that
exhibited severe inhibitory effect on growth (lots B and
C). Although analytical investigations did not highlight
any significant differences in the composition and impu-
rities profiles of these lots, we speculate that the presence
of a low-level inhibitory agent may be the culprit for the
inconsistencies observed. Data presented in Figure 5C
explicitly show, when employing lots B and C, that the
apparent lag phase increased dramatically as the inocu-
lum concentration decreased. A similar, but slight, effect
was also observed when using a historically “good” lot
(A). Interestingly, with IVCC greater than 4.8 × 106 vc/
mL, little discrepancy in growth was observed between
the three lots of MSG. One explanation, consistent with
several theories of lag phase (Monod, 1949; Pirt, 1975)
is that the chemical additive in the MSG has a toxic effect
on a finite number of cells, which represent a small
fraction of the cells when the IVCC is large, but repre-
sents a majority of the cells at low concentrations. Once
the toxic chemical is completely adsorbed by the cells,
the cultivation medium is effectively detoxified, and the
remaining unaffected viable cells may proliferate nor-
mally.
Taken together, these results support the argument
that the one-stage inoculum train configuration is suf-
ficiently robust for implementation in a manufacturing
environment, provided that a minimum IVCC (here 4.8
× 106 vc/mL) is employed as a safe guard against
Biotechnol. Prog., 2005, Vol. 21, No. 4
Figure 5. One-stage inoculum train sensitivity analysis pat-
terns to several external factors. The results show the relation-
ship between the time required to reach early-exponential
growth, inoculation density (viable-cells/mL post-inoculation),
and three external factors at variable concentrations: (A) trace
elements solution (TES), (B) detergent concentration (Contrad
70, Decon Laboratories), (C) monosodium glutamate lot. All data
were obtained using an automated high throughput automation
microbial cultivation device. The cultivation conditions employed
are described in Materials and Methods.
1045
potential variations in the quality of the ingredients or
the medium/equipment preparation.
Inoculum Optimization. The next step of these
studies was to address remaining practical issues facing
the implementation of a one-stage process. Specifically,
the size of the frozen inoculum can have a major impact
on storage facilities and handling operations. This was
addressed by attempting to increase the ratio of cells to
total volume by harvesting the culture at a higher OD600
and by using a concentrated glycerol solution that
reduced the dilution effect.
A cell bank was prepared by cultivating cells in a
nutritionally enriched medium containing 25 g/L glycerol
instead of the normal 10 g/L, while the cells were
harvested later in their exponential growth phase. This
strategy allowed a 5-fold increase in the harvested
biomass. In addition, the inoculum was diluted only by
10% using a 90% solution of glycerol instead of our
routinely used 40% solution. These three modifications
yielded a frozen working cell bank with an approximate
10-fold increase in viable-cell concentration when com-
pared to three previous batches of working inoculum (9.5
× 109 vs 7.0 × 108 vc/mL, respectively). This strategy
effectively supports the implementation of a one-stage
process, even with the use of large production bioreactors.
Analyses and Recommendations. Since the results
have established the limits for the robustness of the one-
stage process and have supported elimination of the
shake-flask stage from the current three-stage inoculum
train, both one- and two-stage configurations are viable
options. For all inoculum train configurations, the use
of a durable inoculum bag is recommended (Junker et
al., 2002), because the bags can be filled and emptied
from and to the bioreactor via a sterile tube-welder,
thereby reducing the overall risk of contamination. Based
on our sensitivity studies, we estimate that the choice of
one versus multiple stages for the process is directly
dictated by the size of the production vessel and by the
size of the frozen inoculum bag a specific facility is
comfortable working with. For this particular process,
IVCC greater than 4.8 × 106 vc/mL are recommended in
order to ensure robustness. Therefore, for relatively small
bioreactors with volumes ranging from 1000 to 5000 L,
a one-stage process is likely to be the best option. With
bioreactors with working volumes greater than 5000 L,
the addition of an inoculum bioreactor may be necessary,
since the large volume of frozen working inoculum
required to inoculate such a vessel (>10 L) may not be
practical for storing, handling, or thawing (Figure 6).
Additional investigations that aim at optimizing inocu-
lum preparation procedures in order to maximize viable-
cell concentration could push these limits further and
allow the application of a reliable and robust one-stage
process to larger vessels.
For this discussion, it is obvious that the decision-
making process does not rest solely on scientific feasibility
but also on economics and ease of implementation. A
multi-stage inoculum train has a higher degree of process
robustness and requires significantly lower frozen inocu-
lum volumes, which reduces inoculum storage, handling,
and thawing issues. The one-stage process, on the other
hand, has a shorter process cycle time leading to in-
creased plasmid productivity and a reduction in the cost
of labor and supplies and in validation efforts due to the
elimination of the shake-flask step. The information
presented here permits many inoculum train configura-
tions to be considered for implementation at large-scale
and allows factors such as scheduling requirements,
equipment availability, and the costs of labor and sup-
1046 Biotechnol. Prog., 2005, Vol. 21, No. 4

Figure 6. Process recommendations schematics. Presented here are two potential inoculum development strategies. The selection
of one versus multiple stages for the process is dependent on the sizes of the production vessel and of the frozen inoculum bag.
Bioreactors with working volumes greater than 5000 L may require the use of an inoculum bioreactor, since the large volume of
frozen working inoculum required to inoculate such a vessel (>10 L) may not be practical for storing, handling, or thawing (Scheme
A), whereas smaller bioreactors can be directly inoculated from a frozen bag (Scheme B).

Table 3. Feasibility/Decision Grid

partial total
ratinga (importance × rating)
one-stage two-stage one-stage two-stage
process characteristic importancea process process process process
minimum cycle time 3 9 4 27 12
high cell-growth reproduciblility 9 5 7 45 63
minimum number of seed train stages 4 10 5 40 20
high specific plasmid yield 10 10 10 100 100
minimum process sensitivity to external factors 8 3 7 24 56
minimum freezer space required for inoculum storage 5 2 6 10 30
minimum overall cost 7 10 6 70 42
minimum risk for contamination or mechanical failure 7 9 6 63 42
total pracitcality score 379 365
a 1 ) least; 10 ) most.

plies to ultimately influence the choice of process config-
uration that will be transferred to manufacturing. A
weighted feasibility grid, presented in Table 3, may aid
future decision-making when choosing an inoculum train
to transfer to an existing facility or when designing a new
manufacturing facility (Design, 2001; Meredith and
Mantel, 2003). This simple deterministic analysis assigns
numerical weighted values to the various positive and
negative characteristics of each inoculum train configu-
ration, such as process cycle time, cost of equipment and
supplies, or inoculum storage space.
Conclusions
The work presented here highlights the importance of
proper inoculum development. Establishing a large da-
tabase supports process characterization and offers the
tools needed in decision-making when implementing the
process at large scale.
The data presented here established that a wide range
of IVCC and inoculum development strategies were
viable approaches in supporting the large-scale produc-
tion of plasmid DNA by cultivation of E. coli. This
information also provides a resource that, based on the
number of viable cells inoculated, allows prediction of the
timing to initiate feeding in the production stage at mid-
exponential growth. These studies also demonstrated
that the IVCC and configuration of the inoculum train,
single versus multiple stages, did not influence final
plasmid titers. Only overall productivity was affected; it
increased with both the reduction in the number of stages
and the increase in IVCC. Establishing such a collection
of data can be extremely useful when planning plant
configuration, capacity and utilization.
These studies also demonstrated that because it de-
livered a higher productivity, implementing a single-
stage process was an attractive option, although the
process could exhibit higher sensitivity to physicochem-
ical variations such as the quality of a key ingredient.
The database established suggests that the process
configuration retained will likely be different, with the
decision influenced by the facility configuration and by
the scheduling constraints. To help in deciding the
approach to implement, a decision grid was developed
in order to estimate the influence of multiple parameters.
We believe that this type of process information is of
Biotechnol. Prog., 2005, Vol. 21, No. 4
critical importance in order to develop an efficient and
cost-effective microbial cultivation process.
One clear advantage of the use of frozen cell banks for
the direct inoculation of bioreactors is the elimination of
shake flasks. In an industrial vaccine production envi-
ronment, eliminating operations where containers, such
as shake flasks, are opened is always advantageous
because it eliminates the need for environmental moni-
toring and can reduce the probability of contamination.
The data presented here clearly support the implementa-
tion of such simplification.
In addition, on the basis of our experience with about
10 DNA vaccine high producer E. coli clones that each
harbored plasmids with different inserts ranging over a
wide molecular weight range, we believe that this type
of approach is completely applicable to any construct.
References and Notes
(1) Babuik, L.; Babiuk, S.; Loehr, B.; van Drunnen Littel-van
den Hurk, S. Nucleic acid vaccines: Research tool or com-
mercial reality. Vet. Immunol. Immunopathol. 2000, 76,
1-23.
(2) Barouch, D. et al. Augmentation of immune responses to
HIV-1 and simian immunodeficiency virus DNA vacines by
IL-2/Ig plasmid administration in rhesus monkeys. Proc. Natl.
Acad. Sci. 2000, 97, 4192-4197.
(3) Barouch, D. et al. Control of verimia and prevention of
clinical AIDS in rhesus monkeys by cytokine-augmented DNA
vaccination. Science 2001, 290, 486-492.
(4) Bergmann-Leitner, E. S.; Leitner, W. W. Danger, death and
DNA vaccines. Microbes Infect. 2004, 6 (3), 319-327.
(5) Berzofsky, J.; Terabe, M.; Oh, S.; Belyakov, I.; Ahlers, J.;
Janik, J.; Morris, J. Progress on new vaccine strategies for
the immunotherapy and prevention of cancer. J. Clin. Invest.
2004, 113, 1515-1525.
(6) Board on Manufacturing and Engineering Design. Theoreti-
cal Foundations for Decision Making in Engineering Design;
National Academies Press: Washington, DC, 2001.
(7) Donnelly, J.; Berry, K.; Ulmer, J. Technical and regulatory
hurdles for DNA vaccines. Int. J. Parasitol. 2003, 33, 457-
467.
(8) Falk, L.; Ball, L. Current status and future trends in vaccine
regulation-USA. Vaccine 2001, 19, 1567-1572.
(9) FDA. Points to consider in the production and testing of new
drugs and biologics produced by recombinant DNA technol-
ogy. 1985.
(10) Ferber, D. Gene therapy: Safer and virus free. Science
2001, 294, 1638-1642.
(11) F.I.P. (International Federation of the Pharmaceutial
Industries) Committee on Microbial Purity. Validation and
environmental monitoring of aseptic processing. J. Parenter.
Sci. Technol. 1990, 44, 272-277.
(12) Gurunathan, S.; Wu, C.; Friedag, B.; Seder, R. DNA
vaccines: a key for inducing long-term cellular immunity.
Curr. Opin. Immunol. 2000, 12, 442-447.
1047
(13) Gurunathan, S.; Klinman, D.; Seder, R. DNA vaccines:
immunology, application, and optimization. Annu. Rev. Im-
munol. 2000, 18, 927-974.
(14) Ignova, M. M.; G. A.; Ward, A. C.; Glassey, J. Fermentation
seed quality analysis with self-organising neural networks.
Biotechnol. Bioeng. 1999, 64 (1), 82-91.
(15) Junker, B.; Seeley, A.; Lester, M.; Kovatch, M.; Schmitt,
J.; Oryscwicz, S.; Lynch, J.; Zhang, J.; Greasham, R. Use of
frozen aged seed inoculum for secondary metabolite and
bioconversion processes at the pilot scale. Biotechnol. Bioeng.
2002, 79, 628-640.
(16) Leitner, W.; Thalhamer, J. DNA vaccines for non infectious
diseases: new treatments for tumour and allergy. Expert
Opin. Biol. Ther. 2003, 3, 627-638.
(17) Leitner, W.; Ying, H.; Restifo, N. DNA and RNA-based
vaccines: principles, progress and prospects. Vaccine 2000,
18, 765-777.
(18) Martinkova, L. M., F.; Ujcova, E.; Kolin, F.; Zajicek, J.
Effect of Age, Amount of Inoculum and Inoculation Medium
Composition on Lactic Acid Production from Glucose by
Lactobacillus casei subsp. rhamnosus. Folia Microbiol. 1991,
36 (3), 246-248.
(19) Meredith, J.; Mantel, S., Jr. Project Management: A
Managerial Approach. John Wiley & Sons: New York, 2003.
(20) Monod, J. The growth of bacterial cultures. Annu. Rev.
Microbiol. 1949, 3, 371-394.
(21) Mountain, A. Gene therapy: the first decade. TIBTECH
2000, 18, 119-128.
(22) Patnaik, P. R. Start-up sensitivity to the initial state of a
batch bioreactor for a recombinant Escherichia coli in a
complex medium. Bioprocess Eng. 1996, 15, 215-219.
(23) PDA. Fundamentals of an environmental monitoring pro-
gram. PDA J. Pharm. Sci. Technol. 2001, 55, 1-35.
(24) Pirt, J. S. Growth lag. Principles of Microbe and Cell
Cultivation; Blackwell Scientific: Oxford, U.K., 1975; pp 194-
198.
(25) Powell, K. DNA vaccines-back in the saddle again? Nat.
Biotechnol. 2004, 22, 799-801.
(26) Prather-Jones, K.; Sagar, S.; Murphy, J.; Chartrain, M.
Industrial scale production of plasmid DNA for vaccine and
gene therapy: plasmid design, production, and purification.
Enzyme Microb. Technol. 2003, 33 (7), 865-883.
(27) Shroff, K.; Smith, L.; Baine, Y.; Higgins, T. Potential for
plasmid DNAs as vaccines for the new millenium. Pharm.
Sci. Technol. Today 1999, 2, 205-212.
(28) Warr, S.; Gershater, C.; Box, S. Seed stage development
for improved fermentation performance: increased milbemy-
cin production by Streptomyces hygroscopicus. J. Ind. Micro-
biol. 1996, 16, 295-300.
Accepted for publication May 18, 2005.
BP040041P