Bioprocess Biosyst Eng (2004) 27: 25–37

DOI 10.1007/s00449-004-0378-9

O RI GI N AL PAPER S

William J. Kelly Æ Kenneth R. Muske

Optimal operation of high-pressure homogenization for intracellular

product recovery

Received: 23 February 2004 / Accepted: 30 June 2004 / Published online: 7 October 2004

Springer-Verlag 2004

Abstract An optimal control methodology for the P Operating pressure (psig)

homogenization of bacterial cells to recover intracellular Pr Impact ring pressure (psig)
products is presented. A Fluent computational fluid Q Volumetric flow rate (ml/s)
dynamics (CFD) model is used to quantify the hydro- rp Fraction of free plasmid recovered from
dynamic forces present in the homogenizer, and homogenization
empirical models are used to relate these forces to b Proportionality constant
experimentally obtained cell disruption and product DE Post-channel turbulence energy dissipation
recovery data. The optimal homogenizer operation, in rate (m/s)
terms of either constant cell breakage or maximum DPc Pressure drop across the channel (psi)
intracellular product recovery, is determined using these DPi Channel inlet pressure gradient (Pa/m)
empirical models. We illustrate this methodology with an l Viscosity (cp)
Escherichia coli bacterial system used to produce DNA q Fluid density (g/ml)
plasmids. Homogenization is performed using an sm Maximum channel shear stress (psi)
industrial APV–Gaulin high-pressure homogenizer. The sw Fully developed channel wall shear stress (psi)
modeling and optimization results for this E. coli–DNA
plasmid system show good agreement with the experi-
mental data.

Introduction
List of symbols
d Impact distance (mm) This paper discusses the modeling and optimal control of
fc Total fraction of broken cells an industrial-scale bioprocess homogenizer. Many pro-
ff Fraction of cell breakage due to channel forces ducts produced by fermentation processes, such as med-
fs Fraction of cell breakage due to shear stress icinal proteins and antibiotics, are intracellular. These
fr Fraction of cell breakage due to impact ring products must first be liberated from the cell prior to the
impingement downstream isolation and purification processing steps.
gf Plasmid in the homogenizer feed pellet (g) A common method for liberating intracellular products is
gh Plasmid in the homogenate pellet (g) the homogenization of a solution of the cells using a high-
h Gap space (lm) pressure homogenizer. In this process, the cell walls are
L Channel length (m) broken open by the hydrodynamic forces present in the
homogenizer. Operation at the maximum attainable hy-
drodynamic forces in the homogenizer will produce the
maximum number of broken cells and, consequently,
maximize the release of the desired product. However, a
W. J. Kelly (&) Æ K. R. Muske number of negative effects can result from this practice
Department of Chemical Engineering, [1]. These effects include the possibility of thermal de-
Villanova University, Villanova, gradation of the product due to excessive heating in the
PA 19085-1681, USA
E-mail: [email protected]
homogenizer and the formation of cellular debris parti-
Tel.: +1-610-5194947 cles that are small enough to interfere with the down-
Fax: +1-610-5197354 stream isolation processes. For large intracellular
26

product molecules, such as proteins, antibodies, and rate of 350 ml/min through a CARR centrifuge (P6
DNA plasmids, there is also a potential risk of the de- Powerfuge) operating at 15,000 rpm. A cell suspension
struction of these molecules after they are released, due to was formed from the resulting cell paste by the addition
the same hydrodynamic forces that disrupt the cell walls. of 25 ml of 100x TE buffer (Sigma) and 2.5 l of DI
For these reasons, operating practice might involve at- water. The cell suspension viscosity was approximately
tempting to operate the homogenizer at a constant per- 1 cp when the cell paste was formed from the contents of
centage of broken cells where these negative effects are two fermenters and approximately 5 cp when the cell
minimized. paste was formed from the contents of ten fermenters.
In this work, optimization of the homogenization
operation to maximize the recovery of intact intracellular
product is considered. We determine the optimal Homogenization
homogenizer operating pressure for an Escherichia coli
bacterial system used to produce an intracellular DNA The homogenizer under study is an industrial-scale
plasmid product. Interest in DNA plasmid production APV–Gaulin model CD-30 high-pressure homogenizer
stems from the fact that approximately 25% of the on- located in the Biotechnology Research Laboratory of
going clinical trials for gene therapy products use DNA Villanova University. For the cell breakage experiments,
plasmids to carry the corrective gene, and most DNA the cell suspension was pumped through the homo-
plasmids currently in clinical trials are produced from genizer at 3 l/min and 15C using a single pass. Plasmid
E. coli fermentation [2]. Recovery of intracellular DNA solutions were also used for the experiments conducted
plasmids using homogenization is a challenging problem to determine how homogenizer conditions affect the
because the released plasmids can be damaged in the breakage of plasmid that had already been released and
homogenizer. If the homogenizer operating pressure is isolated from the cells. To prepare the plasmid solution,
too high, excessive product loss can occur due to the the cell suspension was alkaline lysed [3]. Following
breakage of the product plasmid molecules. On the other the lysis, the pH 5.5 supernatant was separated from
hand, if the homogenizer is operated at a pressure that is the lysate precipitate using the CARR centrifuge. The
too low, a significant fraction of cells will not be broken supernatant solution was then used as the feed to the
and their intracellular plasmid product will be lost. For homogenizer at 3 l/min and 15C using a single pass.
this system, the optimal operation state is one in which
the cell wall breakage is maximized while the disruption
of the liberated DNA plasmids is minimized. Multiple Analytical
low-pressure passes through the homogenizer would
typically not be considered without a preceding plasmid Cell breakage assay
isolation step, due to the adverse effects on the previously
released plasmids. Pre- and post-homogenization samples (10 ml) were
centrifuged and the resulting pellet was washed with and
then resuspended in TE buffer. A standard DNA prep
Materials and methods (Qiagen spin mini-prep, cat #27106) was used to break
open the cells and release/purify the DNA plasmids. At
Organism, culture media, and fermentation the end of the prep, each Qiagen column eluate was run
for 7 h on a 0.75% agarose gel at 47 V. The gels were
The bacterial system in this study is an E. coli strain stained with ethidium bromide, destained, and then
(Life Technologies DH5a) that produces an intracellular imaged (Molecular Dynamics Inc., Fluor-Imager 595).
DNA plasmid product (pUC18). This plasmid contains The gel image was subsequently analyzed with the Image
a gene for the production of a b-lactamase enzyme Quant (Molecular Dynamics Inc.) software package. A
that provides ampicillin resistance. The media used to linear relationship between DNA concentration and
produce the cells for the homogenization experiments band intensity was established using a DNA mass ladder
was Lennox LB broth at a concentration of 20 g/l. After (Invitrogen high DNA mass ladder #10496-016). Cell
sterilization, 200 mg/l of ampicillin was added sterilely breakage was determined by comparing the quantified
to the media to ensure that the E. coli retained the intensity of the supercoiled form of the pUC18 (3 kB)
intracellular plasmid product. plasmid in each lane to that in the lane containing the
The cells containing the plasmid were stored at homogenizer feed sample. The percentage of broken

70C until the time of use. One ampoule (1 ml) was cells was determined from the relationship (1
gh/
used to inoculate 500 ml of broth contained in a 2-l flask gf)·100%, where gh is the amount (g) of plasmid in the
that was agitated in a shaker cabinet at 230 rpm and homogenate pellet and gf is the amount (g) of plasmid in
37C for 15 h. The 500-ml harvested flask was trans- the homogenizer feed pellet [1]. With a 95% confidence
ferred sterilely into a 20-l Bioflow IV fermenter (New interval, the average error in the cell breakage assay was
Brunswick Scientific). Fermentations were carried out 22%: 15% cell breakage ±2.5% (16% error), 40% cell
for 15 h after the transfer. The fermenter contents breakage ±10% (25% error), 65% cell breakage ±16%
(OD2) were then chilled to 10C and pumped at a flow (25% error).
 27

Total and free plasmid assay

The total plasmid concentration (intracellular and re-

leased undamaged pUC18 plasmids) in the eluate of the
Qiagen column was determined by taking pre- and post-
homogenization samples (2·250 ll each) through a
standard DNA prep to break any remaining intact cells
and release/purify the DNA plasmids. At the end of the
prep, each Qiagen column eluate was loaded onto
agarose gels as per the cell breakage assay and the total
concentration of the supercoiled pUC18 plasmid in the Fig. 1 Homogenizer valve diagram
homogenate eluate and the homogenizer feed eluate was
determined from the quantified band intensity. The
‘‘free’’ (released undamaged) plasmid concentration was stress [6, 7], post-channel turbulence, [6, 8], and impact
calculated by subtracting the plasmid concentration in ring impingement forces [9, 10]. Each of the hydro-
the intact cells (from the cell breakage assay) from the dynamic forces related to these effects can result in
total plasmid concentration in the homogenizer samples. increased cell breakage and also an increased possibility
The error in determining the total concentration of of thermal and/or mechanical damage to the released
plasmid, and, correspondingly, the concentration of intracellular product.
‘‘free’’ plasmids, is estimated to be higher than the error Common practice in the biotechnology industry is to
observed in the cell breakage assay because of the pre- specify a single operating pressure for the homogeniza-
sence of significant amounts of genomic DNA, com- tion of a particular cell type and strain. This practice is
parable in size to pUC18, in the gel lanes for these not optimal since batch-to-batch variation in the cell
samples. The genomic DNA was released from the concentration and solution viscosity will effect the per-
broken cells during homogenization and quickly broke formance of the homogenizer at a given operating
into smaller fragments that co-eluted with the plasmid pressure. Published operating pressures corresponding
DNA during the final Qiagen purification (chromato- to 50% E. coli breakage range from 6,000 psig [11] to
graphy) step. 8,000 psig [12]. These results indicate that variations in
the process, such as cell density and valve geometry, can
have a significant effect on homogenizer performance. A
Homogenization process study using high-pressure homogenization to release
DNA plasmids from E. coli showed that 30% of the
A high-pressure homogenizer consists of a positive dis- plasmid product could be recovered while operating
placement pump that delivers a constant volumetric flow either at 5,000 psig, where 30% of the cells were broken,
rate of a solution containing the cells through a homo- or at 7,000 psig, where 75% of the cells were broken [11].
genizer valve. A typical homogenizer valve geometry is This study clearly indicates that significant disruption of
shown in Fig. 1. The solution is fed axially into the valve the released DNA plasmid can occur in a homogenizer.
and is accelerated into the channel between the valve rod
and the seat, referred to as the gap space. The solution
emerges from this channel as a radial jet that stagnates Homogenizer model
on an impact ring before leaving the homogenizer,
usually at atmospheric pressure. The distance that the In this work, the hydrodynamic forces present in the
liquid travels after leaving the valve channel prior to homogenizer are quantified using a 2D computational
impinging on the impact ring is referred to as the impact fluid dynamics (CFD) simulation model of the homo-
distance. genizer valve developed under Fluent V5.3. Previous
The channel gap space between the valve rod and the work in [5, 6, 13] have also considered CFD modeling of
seat is often the only operating variable that can be the homogenizer valve. However, these modeling efforts
varied because the inlet flow rate is typically fixed by the were directed toward understanding the hydrodynamic
associated feed pump and the impact distance is fixed by forces present and not toward the optimization of
the impact ring installed in the valve. The gap space is homogenizer performance.
adjusted by a flywheel attached to the valve rod and is The steady-state flow field through the 2D valve
used to control the inlet or operating pressure. The op- model is determined assuming laminar flow upstream
erating pressure is typically indicated on the homo- and inside of the channel, and turbulent flow down-
genizer. stream of the channel. A realizable k–
turbulence model
Decreasing the gap space will increase the operating [14] is used for the downstream flow model. The model is
pressure and the hydrodynamic forces present. A num- comprised of an unstructured hexagonal mesh, dense in
ber of hydrodynamic mechanisms for cell breakage have the regions of high-velocity gradients, consisting of be-
been proposed for the high-pressure homogenizer, such tween 45,000 and 65,000 nodes, depending on the valve
as channel inlet pressure gradient [4, 5], channel shear geometry. Since the calculated pressure gradient, shear
28

rate, and impact pressure did not change significantly empirical fit in Eq. 1. Figure 3 presents the gap space
with additional computational nodes, the model solu- over the operating range of the homogenizer shown in
tions are assumed to be grid independent. Convergence Table 1.
of the model solutions were assumed when the normal-
ized residuals were below 10
4 and the impact ring
pressure change was below 0.2%/1,000 iterations. This Hydrodynamic forces
convergence tolerance results in model errors that are
well within the normal error associated with the de- In this section, the magnitudes of each of the hydro-
termination of the operating pressure of the homo- dynamic forces proposed as mechanisms for cell break-
genizer and the viscosity of the cell solution. Further age in the homogenizer are related to the operating
information on the Fluent CFD model can be found in pressure and viscosity by empirical models. The form of
[1]. these empirical models is guided by theoretical re-
Due to the computational requirement for the CFD lationships, where possible, and then simplified to obtain
model of the homogenizer, it is not appropriate for a reasonably compact equation form that describes the
on-line optimization. For this reason, reduced complex- Fluent simulation results over the operating range in
ity empirical models relating the hydrodynamic forces Table 1.
considered important in cell and product disruption are
developed based on theoretical and CFD simulation re-
sults. Cell and product disruption models are determined Channel inlet pressure gradient
from experimental homogenization data obtained for the Kleinig and Middleburg [5] proposed the following re-
E. coli–DNA plasmid system considered in this work. lationship for the channel inlet pressure gradient:
These models consist of empirical relationships between
the hydrodynamic forces computed by the Fluent DPi ¼ qQ2 b
1 h
3 ð2Þ
simulation model and the fraction of broken cells and
DNA plasmids obtained experimentally. The operating where DPi is the pressure gradient, q is the fluid density,
condition range considered for the homogenizer is shown Q is the volumetric flow rate, b is a constant determined
in Table 1. by the valve, and h is the gap space. This relationship,
based on theoretical considerations, predicts that the
pressure drop is directly proportional to solution den-
Operating pressure sity, inversely proportional to the cube of the gap space,
and independent of the viscosity. However, Fluent si-
All Fluent simulations in this work are based on a spe- mulation results suggest that the pressure gradient does
cified gap space. Although the valve rod position, which have a slight viscosity dependence. Applying an em-
sets the gap space, is the manipulated variable for pirical correction for viscosity dependence to match the
the homogenizer, it is typically not possible to determine Fluent simulation data results in the following expres-
the actual rod position and corresponding gap space in sion for the inlet pressure gradient that is valid over the
the industrial unit. Therefore, the operating pressure is operating range in Table 1:
specified to describe the operating conditions. Following
DPi ¼ 6:67  106 P l2
9:00  107 P l þ 3:83  108 P
this convention, we use the operating pressure as the
manipulated variable for homogenizer operation and
1:73  1010 l2 þ 1:94  1011 l
5:77  1011
develop the following empirical relationship based on ð3Þ
Fluent simulations for the gap space as a function of
operating pressure and viscosity: Figure 4 presents this relationship where the points
represent Fluent simulation results and the lines
h ¼ ð252 þ 2:5lÞP ð
0:433þ1:7110 l
1:2710
3 l2 Þ

2
ð1Þ represent the prediction from Eq. 3. Because the solu-
tion density and volumetric flow rate remained
where h is the gap space in microns, P is the operating essentially constant for the experimental work in this
pressure in psig, and l is the viscosity in cp. Figure 2 study, no attempt was made to determine density and
presents this relationship in which the points represent flow rate effects in the Fluent simulation model.
Fluent simulation results and the lines represent the

Channel shear stress

Table 1 Operating variable ranges The shear stress in the fully developed section of the
channel occurs at the channel wall and can be
Variable Range determined from rearrangement of the Hagen–Poiseuille
equation, as discussed in [15]:
Gap space 1–11 lm
Viscosity 1–5 cp DPc h
Pressure 2–14 kpsig sw ¼ ð4Þ
4L
 29

Fig. 2 Gap space relation to

pressure and viscosity

where sw is the channel wall shear stress, DPc is the cells have limited exposure to the maximum shear
pressure drop across the channel, h is the gap space, and stresses in the channel. The following empirical re-
L is the channel length. This relationship predicts that lationship for the maximum channel shear stress mat-
the shear stress is directly proportional to the pressure ches the Fluent simulation data over the operating range
drop across the channel and the gap space. Fluent si- in Table 1:
mulation results, however, reveal that the maximum
shear stress in the channel, sm, occurs at the channel wall sm ¼ 11:6 log ðP Þ þ ð0:831 log ðP Þ
8:40Þl2
in the entrance region. This maximum value (sm) is three þ ð6:94 log ð P Þ
30:5Þl
89:1 ð5Þ
to five times the fully developed channel wall shear stress
(sw). Fluent simulations also reveal that approximately Figure 5 presents this relationship where the points
90% of the channel volume has shear stress values less represent Fluent simulation results and the lines re-
than 10% of the channel wall shear stress. Therefore, the present the prediction from Eq. 5. As shown in the fig-

Fig. 3 Gap space over

homogenizer operating range
30

Fig. 4 Channel inlet pressure

gradient model

ure, this relationship begins to approach the linear form dissipation to the operating pressure and viscosity was
in Eq. 4 at operating pressures above 6,000 psig. determined from the Fluent simulation results:


DE ¼
2; 600 þ 4; 060l
208l2 P ð2:10
0:305lþ0:0300l Þ
2

Post-channel turbulence
ð6Þ
Although an equation for the energy dissipated from
post-channel turbulence can be developed from theory, where DE is the maximum post-channel energy dissipa-
it would not be useful as an on-line predictive model due tion rate. Figure 6 presents this relationship where the
to its complexity. For this reason, the following points represent Fluent simulation results and the lines
empirical relationship relating the maximum energy represent the prediction from Eq. 6.

Fig. 5 Channel shear stress

model
 31

Impact ring pressure Fluent simulation results (no more than a 6 psi change
over the operating pressure range). Based on these
Keshavarez-Moore [10] derived the following relation- results, it does not appear that the relationship in Eq. 7
ship for the fluid pressure at the impact ring where the is valid for large impact distances. We also note that
jet emanating from the channel strikes the ring: linear terms in l were sufficient to achieve a good fit for
the smallest impact distance (CD-6). Figure 7 presents
Pr ¼ ad
2 h
2 ð7Þ
the relationship for the CD-0 impact ring where the
where Pr is the impact ring pressure, d is the impact points represent Fluent simulation results and the lines
distance, h is the gap space, and a is a proportionality represent the prediction from Eq. 8. Similar fits were
constant. The impact distance is a function of the impact obtained for the other impact rings, but are not shown.
ring installed on the homogenizer valve. In this work, we
consider the three impact rings in Table 2.
Fluent simulation results suggest that the ring Cell breakage
pressure is strongly influenced by the solution viscosity.
According to the Fluent simulation results, the impact A series of homogenizer experiments that isolated each
pressure is the only fluid dynamic parameter that of these hydrodynamic forces as the major mechanism
depends on impact distance as well as the viscosity and for cell disruption were carried out [1]. Based on the
operating pressure [1]. The following empirical expres- results of this experimental study, empirical relation-
sions are used to relate the impact ring pressure to the ships between cell breakage and the hydrodynamic
operating pressure and viscosity for each of the three forces, as computed by Fluent simulation, are devel-
impact rings in Table 2: oped. Because of the quantity of data available, linear
expressions are developed in this work. These expres-
CD
0 : sions are based only on the magnitude of the forces. The
Pr ¼ 133
17:8l þ 1:5l2 ð8Þ time that the cells are exposed to these forces is not


þ 0:0185
0:0093l þ 0:00108l2 P considered in the cell breakage models.

CD
5 : Pr ¼ 80
3l ð9Þ Table 2 Impact rings and distances

CD
6 : Pr ¼ 105
5l þ ð0:028
0:004lÞP ð10Þ Impact ring Impact distance (mm)

We note that the impact ring pressure for the largest CD-0 0.25
CD-5 2.5
impact distance (CD-5) is not a function of the operating CD-6 0.15
pressure, due to the very weak dependence noted in the

Fig. 6 Post-channel turbulence

model
32

Fig. 7 CD-0 impact ring

pressure model

Channel inlet and outlet forces channel shear stress. In this case, channel shear stress
can be assumed responsible for cell breakage. Analysis
The impact distance for the CD-5 impact ring is large of high viscosity operating data (5 cp) using the CD-0
enough such that impingement against the ring is not a impact ring shows a linear response between the fraction
significant source of cell wall disruption. Cell breakage of cells broken and the maximum channel shear stress
data with this impact ring can then be used to isolate the computed by Fluent simulation. The relationship be-
effect of the forces associated with the valve channel. At tween the fraction of broken cells and the channel shear
low viscosity, channel shear stress is also not likely to stress, determined from high viscosity CD-0 impact ring
contribute significantly to cell disruption. The remaining experiments, is presented in Eq. 12, where fs is the
forces, channel inlet pressure gradient and post-channel fraction of broken cells due to channel shear stress and
turbulence, can then be assumed responsible for cell sm is the maximum channel shear stress in psi:
breakage under these conditions. Since it is not possible 
to isolate the individual effect of these forces experi- 0 sm \95
fs ¼ ð12Þ
mentally, we quantify cell breakage based on the inlet 0:011sm
1:04 95 6 sm 6 170
pressure gradient in this work. The relationship between
the fraction of broken cells and the pressure gradient, The channel shear stress range in Eq. 12 represents
determined from low viscosity (1 cp) CD-5 impact ring the threshold value for cell breakage and the maximum
experiments, is presented in Eq. 11, where ff is the value for which there is experimental data.
fraction of broken cells due to the inlet and outlet
channel forces and DPi is the channel inlet pressure Impact ring pressure
gradient in Pa/m:
 Operation at low pressure and low viscosity will pro-
0 DPi \1:1  1012
ff ¼ DPi DPi ð11Þ duce channel inlet pressure gradients, post-channel
0:35 1012
0:385 1:1 6 10 12 6 2:6
turbulences, and channel shear stresses below the
threshold required to disrupt the cell wall. For low
The channel inlet pressure gradient range in Eq. 11
operating pressure experiments using the CD-6 impact
represents the threshold value for cell breakage and the
ring, impingement with the ring can be assumed
maximum value for which there is experimental data.
responsible for cell breakage where the impact ring
pressure is used as a measure of the impingement for-
Channel shear stress ces. The relationship between the fraction of broken
cells and the impact ring pressure, determined from low
Operation at high viscosity will reduce the channel inlet viscosity (1 cp) CD-6 impact ring experiments, is pre-
pressure gradient, post-channel turbulence, and ring sented in Eq. 13, where fr is the fraction of broken cells
impingement forces below the threshold required to due to impingement with the ring and Pr is the impact
disrupt the cell wall while significantly increasing the ring pressure in psig;
 33
8
<0 Pr\160 operation. Even with the limited amount of experi-
fs ¼ 0:01Pr
1:6 1606Pr 6240 ð13Þ mental data and the error associated with the analytical
: technique used to determine cell breakage, the model
0:8 Pr > 240
appears quite adequate for predictive purposes.
The impact ring pressure range in Eq. 13 represents the
threshold and limiting values for cell breakage due to Plasmid recovery
impingement forces based on experimental data.
Similar experimental studies to those carried out on the
Total fraction of broken cells cells can be undertaken with isolated DNA plasmid to
determine empirical expressions for plasmid recovery as
An expression for the total fraction of broken cells can a function of the hydrodynamic forces computed by
be determined from the three previous mechanisms as Fluent simulation. Because of the difficulty in obtaining
follows: and isolating the quantity of DNA plasmid necessary to
replicate the experimental study carried out for cell
fc ¼ ff þ ð1
ff Þfs þ ð1
ff
ð1
ff Þfs Þfr breakage, experimental data is presented for the CD-0
¼ ff þ fs þ fr þ ff fs fr
ðff fs þ ff fr þ fs fr Þ ð14Þ ring only. The expression for the total fraction of plas-
where fc is the total fraction of broken cells. This model mid recovered for low viscosity (1 cp) operation of the
accounts for the fact that the cells can only be broken CD-0 ring as a function of operating pressure is:
once either by the upstream channel forces or by impact rp ¼ 1:07
3:47  10
5 P 2; 000 6 P 6 12; 000 ð15Þ
ring impingement.
As shown in Fig. 8, this model describes the CD-5 where rp is the fraction of plasmid recovered and the
and CD-6 impact ring experimental data very well. The operating pressure is in psig. Figure 9 presents this
discontinuity of the slope in the model-predicted cell model along with the experimental data. It is anticipated
breakage is due to the use of the linear breakage models that the slope of the recovery model is a function of the
with thresholds. To verify the predictive capability of the plasmid size such that lower recovery would be expected
model, we compare the model predictions to cell for a larger plasmid.
breakage data for the CD-0 impact ring where both
pressure gradient and impact ring impingement effects
are present. The model appears to describe the high Optimal operation
viscosity (5 cp) CD-0 impact ring data quite well, except
for the two intermediate pressure data points. The model Constant cell breakage
describes the low viscosity (1 cp) CD-0 impact ring data
well at low operating pressure, but underpredicts the cell Industrial practice might involve attempting to operate
breakage by approximately 5% for the higher pressure the homogenizer at a constant fraction or percentage of

Fig. 8 Cell breakage model

34

broken cells. Constant pressure operation, however, will solution viscosity. The optimization objective function
not produce constant cell breakage when batch-to-batch in Eq. 16 represents the normalized concentration of
viscosity variation arising from the variation in the undamaged intracellular product recovered from the
fermenter operation supplying the cells occurs. This homogenization process. It represents the trade-off
variation can be accounted for by making adjustments between cell wall disruption to liberate the intracellular
to the operating pressure based on the model in Eq. 14. product and product disruption, which destroys the
Figure 10 presents the operating pressure as a function liberated product. This optimization approach assumes
of solution viscosity for constant cell breakage with the that all cells behave in a uniform manner, all cells
CD-0 impact ring. The other two impact rings behave experience the same hydrodynamic forces in the
similarly and are not shown. As the viscosity increases, homogenizer, all cells contain the same amount of
the channel inlet and outlet forces and the impact ring intracellular product, all of the intracellular product is
impingement forces decline faster than the increase in released when a cell is broken open, the fraction of the
the channel shear, requiring an increase in operating product damaged does not depend on the concentration
pressure. As the viscosity continues to increase, the of the free product released from the broken cells, and
channel shear stress effects dominate and the operating the cell concentration does not vary during the proces-
pressure must be reduced. sing of a single batch. Although these assumptions are
not strictly accurate, they are not unreasonable for the
operating conditions considered in this work.
Optimal plasmid recovery Assuming that the expression presented for plasmid
recovery with the CD-0 impact ring is valid for liberated
Instead of attempting to operate at a constant cell plasmids during the homogenization of the cells,
breakage, an alternative approach is to attempt to the objective function in Eq. 16 can be compared to
maximize the recovery of released, intact intracellular experimental free plasmid recovery data. Figure 11
product. A batch-to-batch, model-based, optimal con- presents this comparison for low viscosity (1 cp)
troller can be obtained from the nonlinear optimization operation with the CD-5 impact ring and Figure 12
problem: presents this comparison for low viscosity (1 cp) op-
eration with the CD-6 impact ring. The model does
maxff ðP ; lÞrp ðP ; lÞ ð16Þ predict increasing recovery with operating pressure, as
P
shown experimentally. The consistent underprediction
where P is the homogenizer operating pressure, l is the of the recovery is expected in both cases because the
solution viscosity, ff(P, l) is the fraction of broken cells, plasmid recovery rp is based on CD-0 impact ring data
rp(P, l) is the fraction of recoverable, undamaged where ring impingement would be expected to con-
intracellular product, and both of these fractions are tribute to plasmid breakage. In the case of the CD-5
functions of the homogenizer operating pressure and the impact ring, ring impingement is not significant due to

Fig. 9 CD-0 impact ring

plasmid recovery model
 35

Fig. 10 Constant cell breakage

for CD-0 impact ring

the large impact distance. For the CD-6 impact ring, the used to generate the 1 cp model in Fig. 13) is not valid at
majority of cell breakage is expected to occur at the high viscosity. Modification of the low viscosity plasmid
impact ring where there may not be sufficient exposure recovery expression to fit the high viscosity plasmid
time for plasmid breakage to occur also. recovery data from this experiment results in the
Figure 13 presents a comparison of the objective following:
function in Eq. 16 with experimental free plasmid rp ¼ 1:34
8:62  10
5 P 4; 0006P 612; 000 ð17Þ
recovery data for high viscosity (5 cp) operation with the
CD-0 impact ring. Although the first two data points are where plasmid breakage starts at a higher operating
questionable, it is clear from this figure that the low pressure (4,000 psig) but the slope is over twice as large
viscosity plasmid recovery expression in Eq. 15 (which is as in the low viscosity expression. Based on this analysis,

Fig. 11 Plasmid recovery for

CD-5 impact ring
36

Fig. 12 Plasmid recovery for

CD-6 impact ring

it appears that the channel shear stress (the dominant plasmid recovery model predictions, which are con-
breakage mechanism at high viscosity) is more destruc- sistent with the experimental data, the CD-6 impact ring
tive to released plasmids than the channel inlet and is recommended for this system.
impact ring impingement forces (the dominant breakage
mechanisms at low viscosity) for operating pressures
above 4,000 psig. Conclusions
Figure 14 presents a comparison of the model-pre-
dicted plasmid recovery for each impact ring at low This paper presents a methodology for the optimal
viscosity (1 cp) operation. This figure clearly shows that control of a high-pressure homogenizer based on a
the CD-6 impact ring provides the highest plasmid CFD model prediction of the hydrodynamic forces
recovery at lower operating pressure. Based on these present in the valve. The forces that are believed to be

Fig. 13 Plasmid recovery for

CD-0 impact ring
 37

Fig. 14 Predicted optimal

plasmid recovery for each
impact ring

responsible for cell wall and intracellular product dis- 2. Kelly W (2003) Perspectives on plasmid-based gene therapy:
ruption are quantified using empirical models, and ex- challenges for the product and the process. Biotechnol Appl
Biochem 37:219–233
perimental data is used to relate cell and product 3. Clemson M, Kelly W (2003) Optimizing alkaline lysis for DNA
disruption to these forces. The resulting predictive plasmid recovery. Biotechnol Appl Biochem 37:235–244
model can then be used to determine the operating 4. Brookman J (1974) Mechanism of cell disruption in a high-
conditions for constant cell breakage and optimal in- pressure homogenizer. Biotechnol Bioeng 16:371–383
5. Kleinig A, Middleberg A (1996) The correlation of cell dis-
tracellular product recovery. ruption with homogenizer valve pressure gradient determined
We illustrate the methodology in this work by using an by computational fluid dynamics. Chem Eng Sci 51:5103–5110
Escherichia coli bacterial system for the production of a 6. Middleburg A, O’Neill B (1994) A simplified model for the
DNA plasmid product. The models presented in this disruption of Escherichia coli: the effect of cell septation. Bio-
work, however, are specific to the homogenizer, cell, and technol Prog 10:109–113
7. Shamlou PA, Titchener-Hooker NJ (1995) A physical model of
bioproduct system under study. For the general applica- high-pressure disruption of Baker’s yeast cells. Chem Eng Sci
tion of this methodology, a CFD simulation model of the 50:1383–1391
homogenization system of interest is necessary to de- 8. Doulah M, Brookman J (1975) A hydrodynamic mechanism
termine the magnitude of the hydrodynamic forces pre- for the disintegration of Saccharomyces cerevisiae in an
industrial homogenizer. Biotechnol Bioeng 17:845–858
sent and experimental data is required to quantify the 9. Engler C, Robinson C (1981) Disruption of Candida utilis cells
effect of these forces on cell disruption and product re- in high-pressure flow devices. Biotechnol Bioeng 23:765–780
covery. The model complexity and modeling effort re- 10. Keshavarez-Moore E (1991) New findings on fundamentals of
quired will depend on the system and the desired model cell disruption in high pressure homogenizers. In: Biological
Recombinant—Microorg. Amin Cells. pp 27–92
prediction accuracy. The linear cell breakage and plasmid 11. Carlson A, Jem K (1995) Mechanical disruption of Escherichia
recovery models in this work were easily obtained from a coli for plasmid recovery. Biotechnol Bioeng 48:303–315
limited amount of experimental data and provided very 12. Agerkvist I, Enfors S (1990) Characterization of E. coli cell
good agreement with the experimental product recovery disintegrates from a bead mill and a high pressure homo-
data. genizers. Biotechnol Bioeng 36:1083–1089
13. Steveson M, Chen X (1997) Visualization of the flow patterns
in a high-pressure homogenizing valve using a CFD package.
Acknowledgements We would like to thank APV–Gaulin for pro- J Food Eng 33:151–165
viding technical and financial support for this work. We would also 14. Shih T, Zhu J (1995) A new k–
eddy viscosity model for high
like to acknowledge the contributions of Justin Miller, Mark Reynolds number turbulent flows—model development and
Rogowski, and Mike Clemson in obtaining the Fluent simulation validation. Comp Fluids 24:227–238
and experimental homogenizer results in this paper. 15. McCabe W, Smith J, Harriott P (1993) Unit operations of
chemical engineering, 5th edn. McGraw-Hill, New York

References

1. Miller J, Rogowski M, Kelly W (2002) Using a CFD model to

understand the fluid dynamics promoting E. coli breakage in a
high-pressure homogenizer. Biotechnol Prog 15(5):1060–1067