GE Healthcare
Extended reports and posters from
the third international conference on
high-throughput process development
Siena, Italy, October 6–9, 2014
In these presentations from HTPD 2014
5 From the conference chairs
Extended reports of presentations
6 High-throughput process development: chromatography media volume definition
10 Minicolumns as a complement to 96-well filter plates in a process development
workflow
13 Rapid process development with UV absorption-based inverse modeling of protein
chromatography
16 Fourier transform assisted deconvolution of complex multidimensional
chromatograms
19 Catching up bioprocess development on the downstream side
23 Utilizing high-throughput techniques to evaluate next generation multimodal
chromatography media
26 High-throughput strategies for the development of viral vaccine processes
29 Perspectives of HTPD techniques for modeling and QbD implementation
31 BioLector Pro – Expansion of microbioreactor platform for strain screening under
full bioprocess control
33 High-throughput microscale platform to accelerate the development of particle
conditioning for biologics
38 Understanding the chromatography behavior of monoclonal antibodies using
quantitative structure-property relationship analysis
40 List of posters presented at HTPD 2014
42
Author index
HTPD 2014 | Extended reports 3
4 HTPD 2014 | Extended reports
From the conference chairs
The conference series devoted to high-throughput process development (HTPD) has already
established itself as a leading forum within its field. After the first two successful meetings in
Krakow, Poland, in 2010, and Avignon, France, in 2012, the third HTPD meeting was in October
2014 held in yet another UNESCO World Heritage site - Siena, Italy.
The goal of this scientific conference has not changed since the very first meeting in 2010; it is
to provide a leading forum for discussion and exchange of ideas surrounding the challenges
and benefits of employing high-throughput techniques in the development of production
processes for biological products.
The conference program for HTPD 2014 included: a workshop on the theme “Lessons learned in
building HTPD capabilities”; a plenary lecture from Arne Staby at Novo Nordisk on mechanistic
modeling; three case study sessions covering upstream, downstream, and formulations;
one session focused on modeling and data analysis; and one session presented an outlook
to HTPD 2020. The sixth session included reports from the very first HTPD Olympiad, which
intended to allow process developers to benchmark their internal HTPD workflows against
peers in the biopharmaceutical industry as well as academic peers. The HTPD Olympiad was
really an appreciated addition to the HTPD meetings program.
About 100 delegates from 13 different countries could enjoy in total 30 oral presentations and
20 scientific posters.
This extended abstract book captures some of the presentations from this very exciting
conference. We hope that this book will serve as a resource and summary of the first-rate
talks and discussions, as well as encourage you to participate in the next meeting in the
HTPD conference series.
Our thanks go the session chairs for their efforts in putting together excellent sessions,
the scientific board for the HTPD Olympiad, the presenters for their contributions, and the
participants for making this a truly valuable and enjoyable event. We also would like to thank
the conference sponsors.
Looking forward to seeing you at next HTPD meeting.
Philip Lester
Genentech
Mats Gruvegard
GE Healthcare
Karol Łącki
Novo Nordisk
HTPD 2014 | Extended reports 5
High-throughput process development:
chromatography media volume definition
T. Bergander and K. M. Łacki
GE Healthcare, Björkgatan 30, 751 84 Uppsala, Sweden
email: [email protected]
Full article to be published in Eng. Life Sci., DOI: 10.1002/elsc.201400240
Introduction
Use of microplates in high-throughput process development (HTPD) studies has become routine.
While this approach is typically used for investigating chromatography conditions such as salt
and pH (1–3), characterization of a multicomponent adsorption system (4), and estimation of
dynamic binding capacities (5), microplates can also be used for media screening.
In any screening experiment involving plates, results obtained will strongly depend on the actual
volume of chromatography media (resin) used. Errors in actual compared to assumed media
volume are probably of no importance when conditions screening is in focus (all conditions
are tested with the same medium, and results are reported as relative to each other); the
importance of correctly estimated volume of chromatography media is more significant for
chromatography media screening. Figure 1, shows a hypothetical effect of combined errors
originating from concentration measurement, ∆c, and in chromatography media volume
dispensed per well, ∆V, on the calculated value of binding capacity, for low- and high-equilibrium
protein concentrations. Further error can be introduced if one assumes an incorrect volume of
chromatography media used in a given experiment (Fig 2). If not accounted for, such an error
influences conclusions drawn on media performance if the media is later used in the packed
column format.
This study compares three criteria for defining media volume and investigates whether the
physicochemical properties of the chromatography medium—particle size, type of base matrix,
and surface chemistry—affect the medium actual volume used in screening experiments when
standard methods are employed for preparation of HTPD plates.
(B)
180
Adsorption isotherm
Assumed phase ratio
20% overestimation of media volume
20% underestimation of media volume
160
140
120
∆CEq
100
80
60
40
20
0
∆Vmedia
0
1
2
3
4
Equilibrium liquid concentration (mg/mL)
5
Equilibrium media capacity (mg/mL)
Equilibrium media capacity (mg/mL)
(A)
∆CEq
14
12
10
Adsorption isotherm
Assumed phase ratio
20% overestimation of VMedia
20% underestimation of VMedia
8
6
4
2
0
∆Vmedia
0
0.05
0.10
0.15
0.20
Equilibrium liquid concentration (mg/mL)
0.25
Fig 1. The effect of error in determination of equilibrium concentration and medium volume on binding capacity at
(A) saturated, and (B) linear binding conditions.
6 HTPD 2014 | Extended reports
Equilibrium media capacity (mg/mL)
80
Adsorption isotherm
Assumed phase ratio
∆CEq
Upper range, β
∆CEq
60
Lower range, β
Perceived adsorption isotherm
Actual phase ratio
40
20
0
β ± ∆β
0
β ± ∆β
1
2
3
4
Equilibrium liquid concentration (mg/mL)
5
Fig 2. Effect of incorrectly assumed value of phase ratio, (ratio between liquid and solid phases, β) on the shape of
adsorption isotherm compared with the true adsorption isotherm for the system studied.
Study outline
Five chromatography media from GE Healthcare’s Life Sciences business, differing in particle
size, base matrix, and surface chemistries were used in the study:
• Capto™ Q (agarose-based, average particle size ~ 90 μm)
• Capto Q ImpRes (agarose-based, ~ 40 μm)
• Capto S ImpAct (agarose-based, ~ 50 μm)
• Q Sepharose Big Beads (agarose-based, ~ 200 μm)
• SOURCE™ 30Q (divinyl benzene-based, 30 μm)
Chromatography media volume definition
Three methods for defining chromatography media volume were used in the study:
(i) Gravity settling method - slurries in 20% v/v ethanol were sedimented in calibrated
measuring cylinders. The measurements were used to calculate slurry concentrations
used for dry weight density (DWD) determination.
(ii) Medium plaque method - for preparation of resin plaques used for dry weight density
determination, Mediascout™ ResiQout (Atoll GmbH, Germany) was used. Three plaque
volumes were tested: 7.8, 20.8, and 50.9 μL. Media plaques were dried according to the
DWD method described below and an average for each plaque was determined.
(iii) Packed column method - chromatography columns were packed according to
recommended packing procedures for each media and column type (Tricorn™ or
HiScale™ columns). After packing, the media were carefully removed from the columns
and resuspended in 20% v/v ethanol to obtain 1% to 10% v/v slurry relative to the
column volume.
DWD determination
For DWD determination, media slurries (1% to10% v/v) from methods (i) and (iii) above were
pipetted into preweighed Petri dishes. The dishes were dried at 105°C for 12 h ± 2 h and after
equilibration at RT for 1 h, the dishes were reweighed. In the case of method (ii), media plaques
were directly transferred to Petri dishes, after visual examination and removal of irregular
plaques. Drying was performed under the same conditions as for methods (i) and (iii).
HTPD 2014 | Extended reports 7
Results
DWD of chromatography media used in the study determined by the packed column method
varied between 90 and 310 mg/mL (DWDColumn, data not shown). These results corresponded well to
known differences between media used such as intraparticle porosity, packability, and material of
construction. The RSD associated with DWDColumn measurement was estimated to 2.6%.
Some chromatography media required several hours to form a stable interface between the
sedimented bed and clear liquid phase (Fig 3). The relative DWD measured using the gravity
method for all media used in the study was lower than the DWD estimated using the packed
column method. This was an expected result considering the media compression during column
packing procedure.
DWDGravity Settled/DWDColumn (%)
120
100
80
60
40
20
0
SOURCE 30Q
Capto Q
Q Sepharose
Big Beads
Capto Q
ImpRes
Chromatography media
Fig 3. Dry weight density (DWD) data after 6 h of sedimentation (the dashed line denotes the DWD determined using
the column packed method, whereas error bars denote the 95% confidence interval for all mass and bed height
determination).
The media plaque method generated data that indicated a specific trend of the lower DWD
for the plaques of 7.8 μL volume. Since this result was observed for all media tested when a
standard plaque preparation protocol was followed, it was decided to perform a more extensive
study using one of the media to confirm the observed pattern. It should be emphasized the
plaque method did not provide DWD measured for each single plaque but instead provided an
average DWDPlaque calculated from the mass of ~ 96 or ~ 192 plaques prepared in four replicate
preparations. The RSD of 1.5% for the average result was compared with corresponding DWDColumn
(Fig 4). The detailed study confirm results observed earlier and showed that, for Capto Q, the
DWDPlaque determined using 7.8 μL plaques, was lower than the DWD based on packed column
method. In case of data generated using larger plaques, the differences in DWD were very small.
8 HTPD 2014 | Extended reports
Since the volume of medium is defined through geometrical dimensions, the differences in the
data obtained can be only explained by differences in the packing densities. These can be linked
to the specifics of plaque-generated procedures such as pressure, slurry concentration, length
of packing, etc., as well as specific properties of the media particles packed, which will also
influence the results. It should be emphasized here that the reported deviations between DWD
measured with the plaque and the column methods are not observed in preparation of equally
sized plaques (4, 6). The deviations reported here indicate that a volume calibration step might
be necessary if a certain size of plaques is used.
120
DWDPlaque/DWDColumn (%)
100
80
60
40
Capto Q DWDPlaque
20
0
Capto Q DWDColumn
0
10
20
30
40
Plaque volume (µL)
50
60
Fig 4. Comparison of DWD measurements between packed column generated medium slurries and the medium
plaques. Error bars in graphs denote the error associated with the DWD method.
Summary
When comparing results generated for different chromatography media with differing
physicochemical properties, care needs to be taken to assure that potential differences in the
volume of chromatography media used in those experiments are accounted for. The magnitude
of these differences is related to methods used for chromatography media volume definition.
With correctly defined and controlled phase ratios in HTPD screening experiments using
microplates, results will allow for comparison of various chromatography media and process
conditions, and also for predicting the media performance in packed columns.
References
1. Kelley, B. D. et al. High-throughput screening of chromatographic separations: IV. Ion-exchange. Biotechnol. Bioeng.
100(5), 950–963 (2008).
2. Kramarczyk, J. F. et al. High-throughput screening of chromatographic separations: II. Hydrophobic interaction.
Biotechnol. Bioeng. 100(4), 707–720 (2008).
3. Rege, K. et al. High-throughput process development for recombinant protein purification. Biotech. and Bioeng. 93(4),
618–630 (2006).
4. Linden, T. Untersuchungen zum inneren Transport bei der Proteinadsorption an poröse Medien mittelskonfokaler LaserRester-Mikroskopie., Heinrich-Heine University Dusseldorf. p. 227 (2001).
5. Bergander, T. et al. High-throughput process development: determination of dynamic binding capacity using microtiter
filter plates filled with chromatography resin. Biotechnol. Prog. 24(3), 632–639 (2008).
6. Herrmann T. et al. Generation of equally sized particle plaques using solid-liquid suspensions. Biotechnol. Prog. 22(3),
914–918 (2006).
HTPD 2014 | Extended reports 9
Minicolumns as a complement to
96-well filter plates in a process
development workflow
A. Grönberg, E. Brekkan, A. Edman Örlefors, and K. Nilsson Välimaa
GE Healthcare, Björkgatan 30, 751 84 Uppsala, Sweden
email: [email protected]
In order to increase speed and efficiency during process development, microscale formats
enabling parallel evaluation of process conditions can be used. 96-well filter plates and
minicolumns packed with chromatography media (resins) are useful tools for high-throughput
process development. There might be preferences for either of the formats. Here the benefits of
both formats are demonstrated, and an alternate use of plates and minicolumns in a process
development workflow is shown. The goal of the study was to optimize monoclonal antibody
(MAb) binding capacity and wash conditions for removal of host cell protein (HCP) when using the
multimodal cation exchange chromatography medium Capto™ MMC ImpRes for MAb polishing.
Initially, optimal binding conditions were identified using a 96-well filter plate filled with 6 μL
of Capto MMC ImpRes. Static binding capacity (SBC) was evaluated at binding conditions
of pH 4.5 to 7.0 and salt concentrations of 0 to 500 mM NaCl. The medium in the plate was
overloaded with MAb and SBC was calculated from loaded concentration and the analyzed
unbound sample concentration using Assist software, where also interpolation of data was
performed. The highest SBC was obtained in the pH range of 5.0 to 7.0 at 0 mM NaCl, with
an optimum at pH 6.0 where the SBC was 90 mg/mL. Dynamic binding capacity (DBC) was
thereafter evaluated using PreDictor™ RoboColumn™ Capto MMC ImpRes 600 μL minicolumns,
operated using a liquid handling workstation. Breakthrough curves were generated using
binding conditions within the area of interest identified in the SBC experiment. A flow rate of
1.67 μL/s, corresponding to a column residence time (RT) of 6 min was used. The highest DBC
was obtained at pH 6.0, without NaCl where 10% breakthrough of MAb occurred at a load of
85 g/L, which correlated well with the SBC data.
In a subsequent wash and elution study using a PreDictor Capto MMC ImpRes 6 μL plate, 60 mg
MAb/mL medium was loaded at a loading pH of 6.0. This corresponds to 70% of the dynamic
binding capacity at 10% breakthrough (QB10%). When applying different buffers (pH 5.0 to 7.0
and 0 to 500 mM NaCl) to the medium with bound MAb, the yield ranged from 0% to more than
90% (Fig 1). The dark blue area in the figure corresponds to the conditions where the yield is very
low (binding conditions) whereas the red area shows potential elution conditions where the yield
is high. Potential wash conditions were found on the border between binding conditions and
conditions where the MAb started to elute (light blue). It was of interest to evaluate whether pH or
salt was the most important parameter for removing HCP while maintaining a high MAb recovery.
Therefore, four conditions, one control and three potential wash conditions, were evaluated using
PreDictor RoboColumn Capto MMC ImpRes, 600 μL. The conditions were: (1) control (wash with
loading buffer); (2) low pH, high salt; (3) intermediate pH and salt; (4) high pH, no salt.
In the wash evaluation using PreDictor RoboColumn minicolumns, 200 µL fractions were
collected into UV-readable plates during sample load, wash, elution, strip, cleaning-in-place (CIP),
and the first part of the re-equilibration. The UV absorbance at 300 nm in the collected fractions
was plotted against the volume, creating chromatograms (Fig 2). Fractions from the wash
and elution peaks, respectively were pooled. MAb concentration and HCP concentration were
determined in these pools.
10 HTPD 2014 | Extended reports
Yield (%)
500
95
90
85
80
75
70
65
60
55
50
45
40
35
30
25
20
15
10
5
0
Elution conditions
450
400
NaCl concentration (mM)
350
300
2
250
200
150
100
3
50
1
0
5.0 5.2 5.4 5.6 5.8
4
6.0 6.2 6.4 6.6 6.8
7.0
pH
Fig 1. MAb yield for various conditions on a PreDictor Capto MMC ImpRes 96-well plate. The blue area corresponds
to the binding conditions and the red area corresponds to the elution conditions. The white circles represent the four
conditions that were chosen for further wash evaluation on PreDictor RoboColumn minicolumns.
MAb yield and HCP concentration in the elution pools are presented in Table 1. Wash 2 (pH 5.0,
250 mM NaCl) showed unexpectedly high loss of MAb in the wash fraction, which could be
explained by an immediate conductivity shift over the column whereas the pH shift lagged behind.
This was verified by introducing a wash with pH 5.0 without NaCl, for 2 column volumes (CV) prior to
a wash with pH 5.0, 250 mM NaCl, for 3 CV (wash 5). Wash 5 resulted in a significantly higher yield
in the eluate compared to wash 2. Judging from the five PreDictor RoboColumn runs, pH seemed
to be the most important parameter for HCP removal. The start material, a MabSelect SuRe™ LX
eluate, contained 364 ppm of HCP and a wash with pH 6.8 increased the purification factor over
Capto MMC ImpRes from 1.6 to 2.9 compared to a wash with loading buffer (pH 6.0).
(A)
(B)
7
Flow rate: 1.85 µL/s
Residence time: 5.4 min
6
Wash: 50 mM acetate
250 mM NaCl pH 5.0, 3 CV
3
1
0
5
A300 nm (AU)
A300 nm (AU)
4
Strip:
50 mM Tris
CIP:
1.0 M NaCl
pH 8.0 1.0 M NaOH
Sample load: 60 mg
MAb/mL pH 6.0
0
1
2
Elution
Elution pool:
Yield 97%
[HCP] 126 ppm
6
Elution pool:
Yield 79%
[HCP] 188 ppm
5
2
7
Elution: 50 mM phosphate
250 mM NaCl pH 7.0
3
4
5
6
7
8
Volume (mL)
9
10
11
4
3
2
Sample load
1
12
13
0
0
1
Wash: 50 mM phosohate
pH 6.8, 3 CV
2
3
4
5
Strip
8
6
7
Volume (mL)
9
CIP
10
11
12
13
Fig 2. Two chromatograms generated from wash evaluation on PreDictor RoboColumn: (A) wash 2; (B) wash 4.
Table 1. MAb yield and HCP concentration in eluates after wash with different conditions on PreDictor RoboColumn
minicolumns
No.
Wash buffer in PreDictor RoboColumn run
Yield eluate (%)
HCP eluate (ppm)
1
pH 6.0, 0 M NaCl (control, loading condition)
100
221
2
pH 5.0, 250 mM NaCl
79
188
3
pH 6.0, 75 mM NaCl
97
171
4
pH 6.8, 0 M NaCl
97
126
5
pH 5.0, 0 M NaCl (2 CV) followed by pH 5.0, 250 mM NaCl (3 CV)
93
212
HTPD 2014 | Extended reports 11
In order to explore if wash conditions could be further optimized, a design of experiments (DoE)
approach was employed. The factors and ranges were: pH 6.0 to 7.0; NaCl concentration
0 to 50 mM; wash volume 1 to 5 CV. A Box Behnken design, which supports quadratic models,
was selected. This resulted in 16 experiments including four center points and was conducted on
two runs on the robotic liquid handling system where eight PreDictor RoboColumn minicolumns
can be operated in parallel. The responses were HCP concentration and MAb yield in the elution
pool. The HCP model was good with R2 and Q2 values of 0.97 and 0.85, respectively. The HCP
concentration in the start material was 322 ppm and the highest HCP clearance observed was
obtained for wash with high pH and high NaCl concentration for 3 to 5 CV (Fig 3). The loss of
MAb during wash was low for the entire experimental space except for a wash with pH 7.0,
50 mM NaCl, 3 to 5 CV, where approximately 15% of the MAb was lost. No statistically significant
model was obtained for MAb yield. In order to ensure robustness in terms of MAb yield it was
decided not to include NaCl in the wash buffer and a wash with 50 mM phosphate, pH 6.8
was selected for verification on a HiScreen™ Capto MMC ImpRes column (4.7 mL). During the
verification run, the HCP concentration was reduced from 386 to 116 ppm at a MAb yield higher
than 90%, which correlated very well with the results from the PreDictor RoboColumn runs.
Wash volume 1 CV
45
3 CV
5 CV
[HCP] in eluates (ppm)
40
NaCl (mM)
35
30
25
20
15
Selected wash
10
5
0
6.0 6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8 6.9
pH
6.0 6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8 6.9
pH
6.0 6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8 6.9
pH
Fig 3. Response contour plots showing the HCP concentration in eluates after wash with different pH and NaCl
concentrations.
Conclusions
PreDictor 96-well filter plates are excellent tools for broad screening of buffer conditions.
Dynamic parameters, for example, DBC, effect of pH and conductivity shifts over a
chromatographic column, and wash volumes can be studied in PreDictor RoboColumn
minicolumns. Good correlation and reproducibility between PreDictor plates, PreDictor
RoboColumn, and other small-scale columns such as HiScreen, enabled the use of PreDictor
plates and PreDictor RoboColumn for optimization studies in a fast and efficient way.
12 HTPD 2014 | Extended reports
Rapid process development with UV
absorption-based inverse modeling of
protein chromatography
T. Hahn, P. Baumann, T. Huuk, and J. Hubbuch
Biomolecular Separation Engineering, Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany
email: [email protected]
Full article to be published in Eng. Life Sci., DOI: 10.1002/elsc.201400247
UV absorption measurements play an important role in bioprocessing. Preparative chromatography
steps are usually controlled using UV measurements and analytical chromatography runs
often determine peak areas of impurities as a measure of overall purity. While this is sufficient
for traditional process development based on a Design-of-Experiments approach, modelbased process development usually requires knowledge of the molar concentrations of all
components in the feed. Model-based process development is highly attractive as it makes
elaborate screening experiments redundant once the model has been calibrated to the specific
process step. With no a priori knowledge about the components’ behavior, the inverse method
is a suitable option which alters parameters in a systematic fashion to achieve a match of the
measured chromatogram and the model prediction.
The common models in liquid chromatography describe the transport in the column, transitions
into the pore system, and adsorption. The concentrations at the column outlet ci (L,t) are first
computed in molar or mass concentrations and then scaled to milliabsorbance units [mAU] with
the factors ai for comparison with the measurements m(tj) at a point in time tj:
2
𝑚𝑚𝑚𝑚𝑚𝑚 ∑𝑗𝑗 �𝑚𝑚(𝑡𝑡𝑗𝑗 ) − ∑ 𝑐𝑐𝑖𝑖 (𝐿𝐿, 𝑡𝑡𝑗𝑗 ; 𝑝𝑝) ⋅ 𝑎𝑎𝑖𝑖 � ,(1)
𝑝𝑝
𝑖𝑖≥1
𝜈𝜈𝑖𝑖
𝜈𝜈 +𝜎𝜎 ′ simulate
𝜕𝜕𝑞𝑞𝑖𝑖′
𝜈𝜈𝑖𝑖 in
′
′
parameter
set to
be
∑𝑘𝑘𝑗𝑗=1 𝑗𝑗 𝑗𝑗 𝑞𝑞To
= 𝑘𝑘𝑎𝑎𝑎𝑎𝑎𝑎,𝑖𝑖
− estimated.
𝑐𝑐𝑝𝑝,𝑖𝑖
− 𝑘𝑘directly
�𝛬𝛬
�
𝑑𝑑𝑑𝑑𝑑𝑑,𝑖𝑖 𝑐𝑐𝑠𝑠 𝑞𝑞𝑖𝑖 ,
𝑗𝑗
𝜕𝜕𝜕𝜕
𝑎𝑎
with ρ being the unknown
[mAU], we
𝑗𝑗
multiplied all molar concentration values in the model equations
with the unknown factors ai.
𝜈𝜈
We obtained a set of equations
that=uses
∑𝑘𝑘𝑗𝑗=1 𝑗𝑗 𝑞𝑞𝑗𝑗′ , boundary equations, which could easily
𝑞𝑞𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠
𝛬𝛬 − [mAU]-based
𝑎𝑎𝑗𝑗
be set by dividing the observed peak areas in [mAU∙mL] by the sample volume in [mL]. The
′
𝑞𝑞𝑗𝑗′ of ′the mobile phase,
′ in the
′ interstitial
equations for convection and 𝜕𝜕𝑞𝑞
diffusion
volume
as well as
𝑖𝑖
= 𝑘𝑘𝑎𝑎𝑎𝑎𝑎𝑎,𝑖𝑖
𝑞𝑞𝑚𝑚𝑚𝑚𝑚𝑚,𝑖𝑖
− ∑𝑘𝑘𝑗𝑗=1
�1
� 𝑐𝑐𝑝𝑝,𝑖𝑖 − 𝑘𝑘𝑑𝑑𝑑𝑑𝑑𝑑,𝑖𝑖 𝑞𝑞𝑖𝑖′ .
′
𝜕𝜕𝜕𝜕
𝑞𝑞𝑚𝑚𝑚𝑚𝑚𝑚,𝑗𝑗
the equations for the pore volume
are linear in the concentration
variables and do not change
structurally.
Depending on the isotherm equation, that describes adsorption and desorption, the unknown
scaling factors stay in the equation system or become hidden in other constants.
HTPD 2014 | Extended reports 13
For models including stoichiometric ion-exchange, such as the Steric Mass Action (SMA) model
2
(1) or the model for multimodal chromatography. (2), a second
isotherm equation for the
2
𝑚𝑚𝑚𝑚𝑚𝑚 ∑
𝑡𝑡𝑗𝑗 ; 𝑝𝑝) ⋅ 𝑎𝑎For
𝑗𝑗 ) − ∑ 𝑐𝑐a
𝑖𝑖 (𝐿𝐿,
𝑖𝑖 � ,the SMA model we obtain
𝑗𝑗 �𝑚𝑚(𝑡𝑡
counter-ion balance allows
for
determining
uniquely.
𝑝𝑝
∑𝑗𝑗 �𝑚𝑚(𝑡𝑡
𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖≥1
i
𝑗𝑗 ) − ∑ 𝑐𝑐𝑖𝑖 (𝐿𝐿, 𝑡𝑡𝑗𝑗 ; 𝑝𝑝) ⋅ 𝑎𝑎𝑖𝑖 � ,
with q'i = ai · qi and
𝑝𝑝
𝑖𝑖≥1
𝜈𝜈𝑖𝑖
𝜈𝜈
𝑘𝑘 𝜈𝜈𝑗𝑗 +𝜎𝜎𝑗𝑗 ′
′
= 𝑘𝑘𝑎𝑎𝑎𝑎𝑎𝑎,𝑖𝑖 �𝛬𝛬 − ∑𝑗𝑗=1
𝑞𝑞𝑗𝑗 � 𝑘𝑘𝑐𝑐𝑝𝑝,𝑖𝑖
𝑘𝑘𝑗𝑗𝑑𝑑𝑑𝑑𝑑𝑑,𝑖𝑖
𝑐𝑐𝜈𝜈𝑖𝑖 𝑖𝑖 𝑞𝑞′𝑖𝑖′ , 𝜈𝜈𝑖𝑖 ′
𝜈𝜈−
(2)
𝑗𝑗 +𝜎𝜎
𝜕𝜕𝜕𝜕
= 𝑘𝑘𝑎𝑎𝑎𝑎𝑎𝑎,𝑖𝑖𝑎𝑎𝑗𝑗�𝛬𝛬 − ∑𝑗𝑗=1
𝑞𝑞𝑗𝑗′ � 𝑠𝑠 𝑐𝑐𝑝𝑝,𝑖𝑖
− 𝑘𝑘𝑑𝑑𝑑𝑑𝑑𝑑,𝑖𝑖 𝑐𝑐𝑠𝑠 𝑞𝑞𝑖𝑖 ,
𝜕𝜕𝜕𝜕
𝑎𝑎𝑗𝑗
𝜈𝜈𝑗𝑗
𝑞𝑞𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 = 𝛬𝛬 − ∑𝑘𝑘𝑗𝑗=1 𝑞𝑞𝑗𝑗′ ,
𝜈𝜈𝑗𝑗
𝑞𝑞𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑎𝑎𝑗𝑗 = 𝛬𝛬 − ∑𝑘𝑘𝑗𝑗=1 𝑞𝑞𝑗𝑗′ 2,(3)
𝑎𝑎
∑𝑗𝑗 �𝑚𝑚(𝑡𝑡𝑗𝑗 ) − ∑ 𝑐𝑐𝑖𝑖 (𝐿𝐿, 𝑡𝑡𝑗𝑗 ; 𝑝𝑝) ⋅′ 𝑗𝑗𝑎𝑎𝑖𝑖 � ,
𝑚𝑚𝑚𝑚𝑚𝑚
′
𝑞𝑞𝑗𝑗
𝜕𝜕𝑞𝑞𝑝𝑝
𝑖𝑖≥1
𝑘𝑘
′
′
𝑖𝑖
𝑞𝑞𝑗𝑗′ 𝑞𝑞𝑖𝑖 . ′
= 𝑘𝑘 ′ 𝑞𝑞 ′ 𝜕𝜕𝑞𝑞𝑖𝑖′ �1 −
𝑘𝑘− 𝑘𝑘𝑑𝑑𝑑𝑑𝑑𝑑,𝑖𝑖
′ ∑𝑗𝑗=1
′ 𝑞𝑞′ 𝜈𝜈 � 𝑐𝑐𝑝𝑝,𝑖𝑖
c'p,i = a
𝜕𝜕𝜕𝜕i ·′ cp,i. 𝑎𝑎𝑎𝑎𝑎𝑎,𝑖𝑖 𝑚𝑚𝑚𝑚𝑚𝑚,𝑖𝑖
𝑖𝑖 − ∑𝑗𝑗=1 ′
=
𝑘𝑘
𝑞𝑞
𝑐𝑐 − 𝑘𝑘𝑑𝑑𝑑𝑑𝑑𝑑,𝑖𝑖 𝑞𝑞𝑖𝑖′ .
�1
�
𝑚𝑚𝑚𝑚𝑚𝑚,𝑗𝑗
𝜈𝜈
+𝜎𝜎
𝑎𝑎𝑎𝑎𝑎𝑎,𝑖𝑖
𝑚𝑚𝑚𝑚𝑚𝑚,𝑖𝑖
𝜕𝜕𝑞𝑞𝑖𝑖
𝜈𝜈
𝑗𝑗
𝑗𝑗
′
𝑞𝑞𝑚𝑚𝑚𝑚𝑚𝑚,𝑗𝑗𝑖𝑖 ′ 𝑝𝑝,𝑖𝑖
= 𝑘𝑘𝑎𝑎𝑎𝑎𝑎𝑎,𝑖𝑖 �𝛬𝛬𝜕𝜕𝜕𝜕− ∑𝑘𝑘𝑗𝑗=1
𝑞𝑞𝑗𝑗′ � 𝑐𝑐𝑝𝑝,𝑖𝑖
− 𝑘𝑘𝑑𝑑𝑑𝑑𝑑𝑑,𝑖𝑖
𝑐𝑐𝑠𝑠 𝑞𝑞𝑖𝑖 ,
𝜕𝜕𝜕𝜕
𝑎𝑎𝑗𝑗
𝜕𝜕𝑞𝑞𝑖𝑖′
𝜕𝜕𝑞𝑞𝑖𝑖′
Binding models, such as the Langmuir isotherm do not include a second equation, such that the
𝜈𝜈𝑗𝑗 ′
ai cannot be determined𝑞𝑞uniquely:
= 𝛬𝛬 − ∑𝑘𝑘
𝑞𝑞 ,
𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠
𝜕𝜕𝑞𝑞𝑖𝑖′
𝜕𝜕𝜕𝜕
𝑗𝑗=1 𝑎𝑎
𝑗𝑗
𝑗𝑗
′
′
= 𝑘𝑘𝑎𝑎𝑎𝑎𝑎𝑎,𝑖𝑖
𝑞𝑞𝑚𝑚𝑚𝑚𝑚𝑚,𝑖𝑖
�1 − ∑𝑘𝑘𝑗𝑗=1
𝑞𝑞𝑗𝑗′
′
𝑞𝑞𝑚𝑚𝑚𝑚𝑚𝑚,𝑗𝑗
′
− 𝑘𝑘𝑑𝑑𝑑𝑑𝑑𝑑,𝑖𝑖 𝑞𝑞𝑖𝑖′ .(4)
� 𝑐𝑐𝑝𝑝,𝑖𝑖
The scaling factors will remain hidden in k'ads,i = kads,i/ai and q'max,i = qmax,i · ai. Still, [mAU]-based
modeling and inverse parameter estimation will lead to an applicable model.
An anion-exchange chromatography of an Escherichia coli SE 1 lysate, including Cherrytagged Glutathione-S-Transferase (GST) as a product, demonstrated practical applicability.
Five experiments in bind/elution mode were performed. Figure 1 shows the result obtained
with a 10 mL and 20 mL gradient and 0.5 mL sample volume, as well as a 10 mL gradient
with 12.4 mL sample volume at the column capacity limit. The peak areas of the target
component and 11 impurities needed for modeling were determined from the 20 mL gradient
run. The parameter estimation was performed first with a genetic algorithm and refined with a
deterministic method. Due to the visibility of the of Cherry-tagged GST in Vis 536 nm, a singlecomponent Vis trace is available for the product, allowing estimation of the scaling factor from
molar concentration to absorbance units (here 8.65 × 107 mAU/M at 280 nm). Analysis of the
feed material by capillary gel electrophoresis resulted in a molar concentration measurement
that translates to a scaling factor of 7.86 × 107 mAU/M at 280 nm, which shows only 9%
deviation to the estimated value. Considering the number of interacting species, this estimate
of a is very good. Additional reliability can be obtained by including samples with different
impurity proportions or fraction analyses that only need to provide peak percentages in one
of the observed wavelengths. In preparative chromatography process development, these
fraction analyses are performed on a regular basis, for example, by applying size-exclusion
chromatography or ion-exchange HPLC, such that no additional experiments are required.
14 HTPD 2014 | Extended reports
(A)
2.5
(B)
1.0
1.0
0.4
0.2
0
0
5
(C)
Volume (mL)
10
15
UV 280 nm (AU)
2.0
0
(E)
20
0.02
0.4
0.01
0.2
0
0
0.2
0
0
10
Volume (mL)
20
0
5
10
15
Volume (mL)
1.0
20
0
1.0
1.0
UV 536 nm (AU)
0.4
1.0
0
(F)
Salt (M)
UV 300 nm (AU)
0.6
0.5
0
0.6
0.8
1.5
15
0.03
1.0
2.0
10
0.8
3.0
2.5
Volume (mL)
0.04
0.2
10
15
Volume (mL)
5
0.8
0.4
5
0
(D) 0.05
0.6
0
0.2
1.0
1.0
0
0.4
0.02
0
1.5
0.5
0.03
0.01
UV 536 nm (AU)
0.5
0.6
Salt (M)
1.0
0.04
Salt (M)
0.6
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
Salt (M)
1.5
0.8
0.05
UV 536 nm (AU)
0.8
Salt (M)
2.0
Salt (M)
UV 280 nm (AU)
0.06
0
10
Volume (mL)
20
0
Fig 1. Comparison of measured and simulated chromatograms with UV signals (solid lines), conductivity
measurement (dashed lines), simulated Cherry-GST absorbance (dotted lines), impurity traces (light solid lines), and
sum of simulated proteins (dot-dashed lines). Plots A and B show the 10 mL gradient elution, C and D the 20 mL
gradient elution, and E and F the breakthrough experiment.
References
1. Brooks, C. A. and Cramer, S. M. Steric mass-action ion exchange: Displacement profiles and induced salt gradients.
AIChE J. 38, 1969–1978 (1992).
2. Nfor, B. K. et al., High-throughput isotherm determination and thermodynamic modeling of protein adsorption on mixed
mode adsorbents. J. Chromatogr. A 1217, 6829–6850 (2010).
HTPD 2014 | Extended reports 15
Fourier transform assisted deconvolution of
complex multidimensional chromatograms
A. T. Hanke1, P. D. E. M Verhaert1, L. A. M. van der Wielen1, E. J. A. X. van de Sandt2, M. H. M. Eppink3,
and M. Ottens1
1
Department of Biotechnology, TU Delft, Delft, The Netherlands
2
DSM Biotechnology Center, Delft, The Netherlands
3
Synthon Biopharmaceuticals B.V., Nijmegen, The Netherlands
email: [email protected]
This extended abstract is based on Hanke, A. T. et al., Fourier transform assisted deconvolution of skewed
peaks in complex multi-dimensional chromatograms, J. Chromatog. A, 1394, 54–61 (2015), © Elsevier B. V.
Reproduced with permission of the publisher.
Introduction
After the success of monoclonal antibodies, the biopharmaceutical drug candidate pipelines are
slowly increasing in protein-class diversity. Many of these classes, such as therapeutic enzymes,
do not share enough structural similarities to allow setting up new affinity-based platform
technologies. This leads to the need for more universally applicable process design approaches.
Hybrid process design approaches that combine mechanistic modeling with experimental
determination of the required physicochemical parameters have been recognized as a
promising strategy to arrive at economically feasible process designs, without having to rely
on affinity technology.
The parameters required for these models are typically regressed by fitting isotherm models
to batch-adsorption measurements (1), or by solving a mechanistic retention time formalism
for single-component peak positions in multiple chromatograms (2). While the causes and
propagation of errors have been thoroughly investigated and optimized for parameters
determined by batch adsorption studies (3), so far comparatively little has been done for
single-component peak parameters. In this study, we introduce a combined experimental
and algorithmic approach designed to increase the confidence in single-component peak
parameters determined directly from complex mixtures.
Peak detection and deconvolution
To accurately determine the position of a peak within a complex chromatogram requires
separating the contributions of co-eluting contaminants to the observed signal. This is usually
achieved by fitting an empirical peak model to the chromatogram with a least-squares
optimizer. As long as the detector used is operated within its linear range, the observed
chromatogram can be treated as the sum of the contribution of each of the components
present to the signal. Therefore, a superposition of multiple empirical single-peak models can
serve as the fitting function for an entire complex chromatogram. The success of such a fit
depends on the correct estimation of the number of peaks and the ability of the peak model to
describe the occurring peak shapes.
The most intuitive approach of counting the number of observed local maxima tends to
underestimate the number of underlying peaks, as highly fused peaks may only appear as
shoulders without their own local maxima. Analysis of local derivatives on the other hand
can reveal such hidden peaks, but is sensitive to noise and can lead to false positives and an
overestimation of the number of peaks. To strike a balance between these extremes, we propose
16 HTPD 2014 | Extended reports
a fast and robust local maxima detection, but sharpen the chromatogram before peak
detection to reveal local maxima also for strongly fused peaks. In the past, Fourier transforms
have been successfully demonstrated to increase the observed resolution of chromatograms.
This is achieved by capturing the systems' peak broadening and skewing contributions in a
transfer function and removing them through division in the frequency domain. By calibrating
the transfer function to not compensate for more than these effects, the chance of introducing
artefacts is kept at a minimum. The peaks in the resulting chromatogram not only appear more
pronounced, but their position can also slightly shift, which is why the number and relative
positions of the detected peaks are only used as starting estimates and the model is still fit to
the original chromatogram. The residuals of the fit are used to determine the uncertainty of
the fitted peak parameters. The residuals not only serve as error margins for later simulations,
but allow judging the quality of the fit, without having to look at each fit, making it a suitable
diagnostic tool for an automated high-throughput environment.
To serve as a peak model, we adapted an extended exponentially modified Gaussian (EMG)
formalism that allows accurate numerical calculation over a broad parameter range both for
fitting and for creating a transfer function for a resolution enhancement operation with Fourier
transforms. The model was chosen based on its ability to create a variety of peak shapes with a
comparatively small number of parameters. The model also corresponds well to a stirred tank
and plug flow reactor in series and so can easily capture the extra column and dispersion effects
to be reduced by the resolution enhancement operation.
Even when the fit yields a small number of residuals, the certainty of the fitted parameters
greatly increases when the peaks in the original chromatogram are well-resolved (Fig 1).
By introducing a second offline separation dimension, the system’s peak capacity is greatly
increased and the properties of peaks that are highly fused in the first dimension can still be
accurately determined.
(B)
1200
1200
1000
1000
800
800
Abs UV280 (mAU)
Abs UV280 (mAU)
(A)
600
400
200
600
400
200
0
0
10
12
14
16
18
20
tR,D1 (min)
22
24
26
10
12
14
16
18
20
tR,D1 (min)
22
24
26
Fig 1. (A) Untreated partial chromatogram (black line) of an IgG1-containing CHO cell culture supernatant separated
by 15 column volumes (CV) linear pH-gradient on a Mono S 4.6/100 column at 1.5 mL/min, together with the Fourier
transform sharpened version of itself (red line) used for peak detection. (B) Fit of multiple EMG peaks to the original
1-D chromatogram. Even though the high degree of overlap between original (solid line) and fitted chromatogram
(dotted line) indicates a good fit, the uncertainties on the calculated moments of the individual peaks (colors) mostly
exceed 100%, deeming the fitted data not useful.
HTPD 2014 | Extended reports 17
(B)
50
50
25
25
Abs280 nm (mAU)
Abs280 nm (mAU)
(A)
0
-25
0
-25
-50
10
-50
10
15
tR,D1 (min)
20
25
6.0 6.5
5.0 5.5 t (min)
4.5
R,D2
4.0
7.0 7.5
15
tR,D1 (min)
20
25
6.0 6.5
5.0 5.5 t (min)
4.5
R,D2
4.0
7.0 7.5
Fig 2. (A) Comprehensive 2-D chromatogram of the fractions collected during the separation shown in Figure 1. The
second separation dimension was size exclusion UHPLC on a BEH 200 4.6/150 column at 0.3 mL/min. (B) 2-D fitted
chromatogram with individual deconvoluted peaks shown as flat contour line projections. As in the 1-D case, the fit
corresponds well to the original chromatogram, but the additional resolution allows the moments to be determined
with 10-fold higher certainty. Reprinted from J. Chromatog. A, Publication title, Vol. 1394, Hanke, A. T. et al., Fourier
transform assisted deconvolution of skewed peaks in complex multi-dimensional chromatograms, pages 54–61,
© 2015, with permission from Elsevier B. V.
Conclusions
Fourier transform assisted peak detection and deconvolution was demonstrated to be a robust
and yet powerful tool for the deconvolution of complex chromatograms that works well for both
single and multidimensional datasets. The integration of automatic uncertainty estimation into
the procedure revealed both the benefits of multidimensionality and serves as an objective and
an automation-friendly measure of fit quality.
References
1. Nfor, B. K. et al. Model-based high-throughput process development for chromatographic whey proteins separation.
Biotechnol J. 7(10), 1221–1232 (2012).
2. Nfor, B. K. et al. Multi-dimensional fractionation and characterization of crude protein mixtures: toward establishment of
a database of protein purification process development parameters. Biotechnol Bioeng. 109(12), 3070– 3083 (2012).
3. Osberghaus, A. et al. Detection, quantification, and propagation of uncertainty in high-throughput experimentation by
Monte Carlo methods. Chem. Eng. Technol. 35(8), 1456–1464 (2012).
18 HTPD 2014 | Extended reports
Catching up bioprocess development on
the downstream side
C. Walther1, M. Berkemeyer2, and A. Dürauer1
University of Natural Resources and Life Sciences, Vienna
2
Boehringer-Ingelheim RCV
1
email: [email protected]
Diverse formats from microliter to liter scale are available for process development and
optimization of bioprocesses. This leads to an increasing number of screening conditions in
the upstream development, putting high pressure on the process development of downstream
unit operations. High-throughput techniques in downstream process development are
mainly focused on chromatographic steps employing standard batch adsorption and
miniaturized column formats. Less experience is available in high-throughput screening
of nonchromatographic unit operations such as centrifugation, filtration, or precipitation,
and protein recovery from inclusion bodies (IB). The formation of IB could be considered
as advantageous for biopharmaceutical production since they usually contain the target
compound highly enriched and pure. Due to the individual needs of each refolding procedure,
the process development presents a bottleneck during downstream processing.
Not only the refolding conditions are of importance for a successful functionality of a product,
but also the solubilization conditions of the IB have been discussed as influential on yield
and productivity of such processes. Including these factors increases screening complexity
exponentially. We present an automated platform for parallel screening of IB solubilization and
protein refolding conditions compatible with the concept of design of experiments (Fig 1).
Solubilization screening
Refolding screening
Turbidity by absorbance @ 600 nm
Turbidity indicates aggregation /
precipitation = unfavored conditions
Discard turbid samples
Intrinsic fluorescence emission scan
Determine wavelength shift
and scattering
Selective capture of refolded protein
cGE analysis of eluted protein
Unfolded and misfolded
protein found in flow through
Binding and elution of
successfully refolded
protein
Fig 1. Experimental design of IB renaturation screening process.
HTPD 2014 | Extended reports 19
The solubilization of the IB is monitored over time via turbidity measurement enabling the
screening of 24 solubilization conditions in quadruplicate per 96-well plate. The efficiency
of the refolding is tested using a hierarchical analysis consisting of turbidity measurement,
fluorescence measurement, and capillary gel electrophoresis (cGE) analysis after selective
capture of the correctly folded protein by affinity chromatography media (resins). The
incompatibility of numerous compounds in refolding buffers with the subsequent highthroughput analysis by cGE is overcome by elution of the captured protein in a suitable buffer.
Figure 2 shows the application of the high-throughput system for determining solubilization
kinetics and thereby optimal solubilization kinetics. Increasing the pH and higher amount
of chaotrope enhances IB solubility. For this product, addition of GuHCl to urea buffers
further improves the yield of solubilized protein. The established method enables not only to
determination of the equilibrium yield of solubilized protein but also kinetics of the solubilization
process. Such solubilization kinetics provides information about the minimal time required
for complete solubilization but also about the maximum time not to be exceeded during
solubilization. This can be valuable information as it has been shown before that longer holding
times of solubilized IBs can have a negative impact on the subsequent refolding.
(A)
Solubilized protein (mg/mL)
8
4 M Urea, pH 9.0
4 M GuHCl, pH 7.5
8 M Urea, pH 10.5
8 M Urea, 0.5 M GuHCl,
pH 10.5
6
4
2
0
0
20
40
60
Time (min)
80
100
120
(B)
Solubilized protein (%)
100
80
60
40
20
6
4
M
Ur
M ea,
U p
8 rea H 7
M , .5
U p
4 rea H 7
M , .5
Ur pH
6 ea 7
M , .5
U p
8 rea H 9
M
.0
4 Ure , pH
M a 9
Ur , p .0
6 ea H
M , 9
U p .0
8 rea H 1
M , 0
Ur pH .5
e
3
M a, p 10.
5
4 GuH H 1
M C 0.
5
5 GuH l, pH
M C 7
G l
3 uH , pH .5
M C 7
l
4 GuH , pH .5
M C 7
5 GuH l, pH .5
M C 9
3 Gu l, p .0
M HC H
9
4 GuH l, pH .0
8
M C
M
9
l
G
,
U 5 u p .0
8
8 rea M HC H 1
M
M , Gu l, p 0
Ur
0
U
8
8 ea M re .5 HC H .5
M , 0 U a M l, 10
Ur .5 re , 0 G pH .5
ea M a .5 uH 1
, 0 G , 0. M Cl 0.5
.5 uH 5 M Gu , p
M Cl
H H
Gu , 1 Gu Cl, 7.5
HC 0 m HC pH
l, 2 M l, p 9.0
0 DT H 1
m T, 0
M p .5
DT H 1
T, 0.
pH 5
10
.5
0
Fig 2. Application of high-throughput system for (A) determination of solubilization kinetics for an affinity scaffold
and (B) for screening of optimal solubilization conditions for an Fc-fusion protein.
20 HTPD 2014 | Extended reports
When using high-throughput methods, there is usually debate about the scalability of the
results. Therefore, solubilization was investigated at micro- and laboratory scale (Fig 3). At
microscale, 96-well microplates with 180 µL working volume were used on an integrated shaker
with a shaking diameter of 3 mm. The microplates were shaken at 650 rpm. At laboratory scale,
a bioreactor (Mettler Toledo) with a working volume of 100 mL was used. An upward pitchedblade element with four impellers was used as stirrer and the IB was kept homogeneously in
suspension at 200 rpm. The Phase number, Reynolds number, energy input, or parameters
limiting solubilization were investigated. Afterwards, solubilization kinetics were determined for
various proteins under various conditions and results from both scales were comparable (1).
96-well microplate,
turbidity measured
at OD 600 nm
96-well microplate
Height, H: 10.9 mm
Fill level (180 µL), lF: 5 mm
Diameter top, d: 6.96 mm
Diameter bottom, d´: 6.58 mm
Integrated shaker:
Bioreactor: Tecan™ Te-Shake™
Mixing orbit: 3 mm
Mixing frequency: 650 rpm
Temperature: 21°C
Lab-scale reactor,
turbidity measured at
880 nm (backscattering)
d
Turbidity
measurement
in microplate
wells
lF
H
h
LF
d‘
h
d
D
Laboratory-scale reactor
Height, H: 90 mm
Fill level (100 µL), lF: 52 mm
Inner diameter, i.d. : 52 mm
Impeller stirring:
Pitched blade element, upward
Four impellers, 45°
Height, H: 9 mm
Diameter, d: 25 mm
Stirrer frequency: 200 rpm
Temperature: 21°C
Fig 3. Scale-up of the high-throughput system for determination of solubilization conditions from microplates
(180 µL working volume, shaken at 650 rpm) to a laboratory-scale reactor (100 mL working volume, stirred at
200 rpm, overhead stirrer).
Figure 4 shows the results of the hierarchical analysis after the refolding screening for an
Fc-fusion protein. Turbidity (Fig 4A) was determined directly after refolding by absorbance
measurement at 600 nm. Samples showing turbidity higher than 0.2 were discarded as high
turbidity indicates aggregation rather than refolding. Figure 4B shows fluorescence and
scattering measurements for a selection of conditions. For fluorescence measurements,
excitation was carried out at 290 nm and an emission wavelength scan was recorded. The
maxima of a native standard and the solubilized protein were different–a wavelength shift was
observed. Samples showing a comparable maximum to the native standard were regarded
as strong evidence for successful refolding conditions. Additionally, scattering of the samples
was determined. High-scattering intensity is an indicator of aggregation even if the wavelength
maxima are comparable to the maximum of the native standard. Finally, the refolded protein
was captured using MabSelect Xtra™ and analyzed via cGE (Fig 4C). Based on a calibration,
the refolding yields can be determined. Using these three data sets for a DoE analysis, optimal
refolding conditions can be determined.
HTPD 2014 | Extended reports 21
(A)
(B)
Wavelength (mm)
350
0.7
0.6
0.4
S39
S14
S24
340
335
STD
S32
S91
S66
S21
330
0.3
0.2
0.1
0
12 11
10
9 8
C
7 6
5 4
B
3 2
A
1 0
D
E
F
H
G
Scattered light intensity
Turbidity
0.5
Den 1:25
345
S66
80 000
60 000
40 000
S21
Den 1:25
20 000
0
STD
S1
S14
S24
S32
S39
S71
S91
(C)
Fig 4. Hierarchical analysis of refolding screening of an Fc-fusion protein. (A) Turbidity measurement at 600 nm.
(B) Intrinsic fluorescence and scattering measurement. (C) cGE analysis after capture of refolded protein.
Conclusion
This platform enables the screening for optimized conditions of the entire protein recovery from
inclusion bodies creating a holistic view on all crucial impact factors in an early stage of process
development.
Reference
1. Walther, C. et al. Prediction of inclusion body solubilization from shaken to stirred reactors. Biotechnol. Bioeng. 111,
84–94 (2014).
22 HTPD 2014 | Extended reports
Utilizing high-throughput techniques to
evaluate next generation multimodal
chromatography media
J. Welsh1, H. Bao1, K. Barlow1, E. Brekkan2, K. M. Łacki2, T. Linden1, and D. Roush1
Merck & Co., Inc., Kenilworth, NJ, USA
2
GE Healthcare, Björkgatan 30, 751 84 Uppsala, Sweden
1
email: [email protected]
Full article published in Eng. Life Sci., DOI: 10.1002/elsc.201400251
Introduction
Automated, scale-down screening has become a critical piece of chromatography process
development in protein purification. High-throughput chromatography techniques allow for
screening of more chromatography media (resins) and wider operating spaces while reducing
material, time, and labor requirements. In addition, the increased data points generated from
high-throughput screens facilitate improved process models.
In this study, prototype multimodal media of varying ligand densities were supplied by
GE Healthcare’s Life Sciences business for evaluation in monoclonal antibody (MAb) purifications.
The ligands, adhere and MMC, were evaluated in flowthrough and bind-and-elute modes,
respectively, for three different MAbs. Both ligands were evaluated with commercial CaptoTM
media as well as with prototype Capto ImpRes media (smaller particle size) at three different
ligand densities referred to as Prototype Low, Mid, and High. Prototype High ligand densities
correlated with commercial Capto media. Screens were implemented to quantitate performance
in terms of both capacity and separation of key product-related impurities. These impurities
included aggregates, product degradates, and hydrophobic fragments that had proven difficult
to remove using traditional chromatography media.
The screening strategy implemented for these studies utilized multiple techniques in series (1).
Batch slurry plates were initially screened to quickly evaluate capacity and partitioning of both
MAb and impurity species at a wide set of conditions. Results of batch screens were used to set
a more narrow range of conditions to explore using miniature columns (0.6 mL bed volumes and
3 cm bed height). These screens were implemented to quantitate precise impurity resolution in
a packed bed, unidirectional flow format. All high-throughput screens were fully automated on
TecanTM liquid handling systems. Finally, select conditions were run on 1 mL HiTrapTM columns
with ÄKTATM systems to confirm high throughput findings. Additional material presented from
this study can be found in reference 2.
Results – batch slurry plates
In the example provided here, media with adhere ligands were screened in the batch slurry plate
format in flowthrough mode. A total of 96 conditions were explored at pH between 3.5 and 9.0
and NaCl addition from 0 to 800 mM (Fig 1). The outputs measured in this example were yield
and clearance of aggregates. The responses for both of these outputs were modeled using a
design of experiments (DoE) approach. A minimum success criterion was set at 80% yield (less
than 20 mg/mL binding to the medium at a 100 mg/mL load) with less than 0.5% aggregates
present in the flowthrough from a feed stream with 2.2% aggregates.
HTPD 2014 | Extended reports 23
Figure 1 displays the results for the slurry DoE models for all four media tested. The general trend
for yields was stronger binding at low salt and high pH. Higher ligand densities also allowed for
stronger media binding. In terms of aggregate clearance, optimal conditions were observed at
moderate pH and salt concentrations. The circular ranges identified in these models represent
a balance of two factors—weaker aggregate clearance at low pH and salt conditions and
formation of aggregates in the flowthrough product at high pH and salt due to protein instability
at these conditions. Higher ligand densities provided improved aggregate clearances.
For the criteria set for this screen, the Capto adhere and Prototype High media provided the
optimal balance of acceptable yields and adequate aggregate clearance. Yellow regions represent
conditions where both criteria were met. The Prototype Mid and High media gave higher yields
but inadequate aggregate clearance at all conditions. For follow-up screenings with miniature
columns, the Capto adhere and Prototype High media were screened at a range of salt conditions
at pH 6.5, the approximate midpoint for optimal pH.
Capto adhere
Prototype Low
800
Yield
boundary
Aggregates
boundary
600
NaCl (mM)
NaCl (mM)
600
800
400
200
0
400
200
0
3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 8.5 9.0
pH
Prototype High
800
800
600
600
Yield
boundary
400
200
0
NaCl (mM)
NaCl (mM)
Prototype Mid
3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 8.5 9.0
pH
400
Yield
boundary
Aggregates
boundary
200
3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 8.5 9.0
pH
0
3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 8.5 9.0
pH
Fig 1. Media slurry plate screen for Capto adhere and Capto adhere ImpRes prototype media. Yellow regions
represent areas where both yield and aggregate clearance criteria were met. Higher aggregate levels seen at pH
extremes were due to protein instability at these conditions. Red highlighted boxes indicate regions selected for
further screening. No areas were identified for Prototype Low and Mid.
24 HTPD 2014 | Extended reports
Results – miniature column screens
Yield (%) or aggregate removal (%)
Miniature column screening conditions were selected from slurry plate model results, namely
pH 6.5 at 25, 200, 400, and 600 mM NaCl. Consistent with slurry plate predictions, aggregate
clearances were poor at low salt whereas salt had only a moderate effect on yields (Fig 2). The
miniature columns showed improved aggregate clearance at high salt than was observed in
slurry plates. However, this is likely an artifact of the reduced time between sample measurements
for the miniature columns (< 1 d) rather than a true performance discrepancy. Also as observed
in slurry plates, the Capto adhere and Prototype High media provided similar performance, with
Capto adhere giving only slightly improved clearance of aggregate.
100
80
60
Capto – Yield (%)
Prototype High – Yield (%)
40
Capto – Aggregate removal (%)
20
Prototype High – Aggregate removal (%)
0
0
200
400
600
NaCl (mM)
Fig 2. Comparison of miniature column performance at pH 6.5 across different salt concentrations for
Capto adhere and Prototype High media.
Conclusion
This study demonstrates that differences in particle size and ligand density, often factors
providing only moderate changes in resolution, can be identified with appropriate highthroughput chromatography screening techniques. In the example shown here, improved
performances were observed with higher ligand densities, and marginal performance
improvements were observed with a larger particle size in flowthrough mode. In bind-and-elute
studies with MMC ligands, however, the smaller particle size gave clearly improved aggregate
separation with optimal performances seen at both high and low ligand densities depending
on the impurity challenge (2). These examples highlight the necessity of using high-throughput
methods for mapping out performance space on a case-by-case basis.
References
1. Welsh, J. P. et al. A practical strategy for using miniature chromatography columns in a standardized high-throughput
workflow for purification development of monoclonal antibodies. Biotechnol. Prog. 30 (3), 626–35 (2014).
2. Welsh, J. P. et al. Utilizing high throughput techniques to evaluate next generation multimodal cation exchange resins.
Eng. Life Sci. In press (2015).
HTPD 2014 | Extended reports 25
High-throughput strategies for the
development of viral vaccine processes
M. Wenger, S. Christanti, C. Daniels, K. Huff, E. Lim, J. Olson, J. Rodriguez, and S.-c. Wang
Vaccine Bioprocess R&D, Merck Research Laboratories, Merck & Co., Inc., Kenilworth, NJ USA
email: [email protected]
Introduction
Process development of live virus vaccines presents many novel challenges owing to their size,
molecular complexity, and structural heterogeneity. Further, because of the diversity of vaccine
candidates, vaccine bioprocesses must often be developed de novo, in which the cell line varies
with virus type and the purification requires alternative strategies from those of conventional
protein chromatography. For virus production, this may include the need to use novel cell lines
that have not been suspension-adapted and therefore require static (e.g., roller bottle or cell
stacks) or microcarrier processes. For downstream processes, unit operations that are amenable
to the purification of large particles are required, including large-pore tangential-flow filtration
(TFF), membrane chromatography, precipitation, and aqueous two-phase separation (ATPS).
Adding to these challenges, fully aseptic processing may be required for those viruses too large
to pass through a 0.22-µm filter, thereby further limiting processing options.
High-throughput (HT) strategies provide a means to cope with these diverse challenges
through a flexible suite of automation and scale-down tools. However, prior high-throughput
process development (HTPD) efforts, particularly for upstream cell culture processes, have been
hampered by low-throughput analytics and automation challenges surrounding biosafetylevel-2 (BSL-2) containment and aseptic operation. Recently, we have begun executing on a
roadmap of HT solutions to overcome these obstacles and better support live virus process
development. Platforms that have been implemented to date include robotic systems for plateand tube-based screening, automated minibioreactors for bioreactor development, the TAP
Biosystems (Sartorius Stedim) SelecT™ for automated cell culture of adherent cell lines, and a
96-well microplate format for membrane chromatography screening.
Upstream HTPD
Specifically, as part of our upstream HTPD roadmap (Fig 1), we recently implemented a
HighRes Biosolutions integrated robotic platform with full BSL-2 containment and aseptic
pipetting capabilities for plate-based culture screening and advanced HT analytical capabilities.
This system consists of an ACell track robotic arm (HighRes Biosolutions), a Hamilton
Microlab™ STAR™Plus liquid handler with clean-air canopy, and multiple peripheral devices
including incubators, an imager, flow cytometer, plate reader, plate washer, dispenser, and
labware carousel. This system has been subsequently applied to the screening of cell culture
media and viral production conditions in support of multiple vaccine programs, with the goals
of increased virus productivity and serum reduction. Prior to introduction of this system, media
screening was an important but under-utilized lever for live-virus vaccine development because
of the vast design space involved. Since its inception, the automated robotic platform has
enabled thousands of media conditions to be interrogated and parameter interactions to be
better understood for rapid media development and more robust processes.
26 HTPD 2014 | Extended reports
Highest throughput
(screening models)
Integrated
robotic systems
(plate-based, microliter scale)
Spin-tube systems
(tube-based, shake-flask
surrogate)
Microbioreactors
Throughput
(< 15 mL scale; e.g., ambr™ 15,
BioLector™, Micro-24)
SelecT
(automated T-flask operation
for adherent cell lines)
ambr 250
(250-mL automated
minibioreactors)
Most representative
(scale-down models)
Representativeness
Fig 1. Roadmap of HTPD technologies for upstream vaccine development.
For bioreactor development, a number of minibioreactor options have emerged on the market,
including the ambr 15 (Sartorius Stedim/TAP Biosystems), Micro-24 (Pall Corporation), and
BioLector (m2p-labs). In addition, TAP Biosystems recently introduced the ambr 250, a 250-mL
bioreactor with independent temperature, pH, dissolved oxygen, feeding, agitation, and gassing
control of 12 to 24 single-use bioreactors. While this system has been successfully demonstrated
for CHO cell culture, P. pastoris, and E. coli platforms (1), there has been little-to-no experience
to-date with microcarrier systems. We therefore evaluated the application of the ambr 250 to
microcarrier cultures with two different cell lines for vaccine production. These experiments
suggest that the current bioreactor design, with its baffled sides and pitch-blade impeller, is
insufficient to suspend the microcarrier beads at the slower impeller speeds required to maintain
cell viability and growth. Conversely, cell detachment and poor growth were observed at the
increased impeller speeds necessary for complete microcarrier suspension. As such, alternate
impeller designs and non-baffled bioreactors are currently being evaluated in collaboration with
TAP Biosystems for improved microcarrier operation.
Downstream HTPD
In the downstream space, we have expanded our HT chromatography platform to include
membrane chromatography, using a 96-well format on a Tecan workstation to examine the
effect of feed matrix, mobile-phase composition, and membrane chemistry on relative capacity,
yield, and impurity clearance. While the 96-well format has proven to be useful for relative
assessments, it is less effective for predicting absolute dynamic binding capacity because
of the very thin membrane thickness and poor flow control (by centrifugation or vacuum
filtration). In addition to chromatography, we are broadening our downstream HTPD platforms
to precipitation and ATPS. While ATPS has been minimally employed in vaccine manufacturing
processes because of the broad design space that must be evaluated for its development, HTPD
makes this unit operation much more tractable and something worthy of examination in the
development of future vaccines.
HTPD 2014 | Extended reports 27
HPTD 2020
Looking forward, we plan to onboard additional HT-enabling technologies, including
an automated subcloning workflow to isolate high-producing clones and an improved
infrastructure for rapid expression screening of different cell substrates, both for adherent and
suspension cell lines. Improved informatics, feedback loops, and modeling tools will also be
an area of emphasis in order to deal with and make sense of the large volumes of data being
generated. Figure 2 provides a summary of our current progress and where we are headed.
In executing on this roadmap, it is our expectation that these HT platforms will have a direct
impact on our vaccine pipeline, enabling a next-generation of viral vaccine processes that are
potentially more productive, robust, and significantly better understood.
Exression
systems
Today
(in progress)
Clonal isolation
ATPS/precipitation
Minibioreactor
Medium development/toxicity
Adherent cell culture (SelecT)
Pipeline impact
Advanced
informatics
Membrane chromatography
Conjugation chemistry screening
Chromatography
Implementation
Fig 2. HTPD 2020 for vaccines: pathway to pipeline impact.
References
1. Bareither R.et al. Automated disposable small scale reactor for high throughput bioprocess development: a proof of
concept study. Biotechnol. Bioeng. 110, 3126–3138 (2013).
28 HTPD 2014 | Extended reports
Perspectives of HTPD techniques for
modeling and QbD implementation
L. Sejergaard, M. H. Hansen, A. A. Olsen, P. Østergaard, T. Strøm-Hansen, P. Østfeldt,
S. Valentin, T. B. Hansen, S. Kidal, and A. Staby
Novo Nordisk, Novo Alle, 2880 Bagsværd, Denmark
email: [email protected]
Introduction
Speed in manufacturing development is synonymous with experimental high-throughput
process development (HTPD) techniques, however, HTPD techniques have also resulted in
a much broader application area, for example, for Quality by Design (QbD) implementation
according to ICH guidelines. Implementation of QbD covers employment of a variety of activities
and concepts such as product, process, and facility understanding under a risk-based approach
(Fig 1), and improved process understanding is directly linked to the methodology applied, from
theoretical knowledge to mechanistic modeling (Fig 2).
HTPD techniques have proven very efficient for numerous applications in the past, for example,
chromatographic media (resin) screening (1), protein crystallization (2), filtration membrane
screening, preformulation, formulation, solubility and stability studies, and identification of
critical quality attributes (CQAs) due to the inherent low material consumption rates and/or
the high number of experiments that may be conducted with a limited amount of material.
These capabilities have also made HTPD techniques very interesting as a supporting tool for
development of mechanistic models of various unit operations.
Product
understanding
• Quality target product profile
• CQAs
Risk
assessment
Requirements
Process
understanding
Risk
assessment
Requirements
• Process parameters, e.g. DoE
• Process models, e.g. simulation
• Design space
Facility
understanding
• Control strategy
• PAT etc.
Fig 1. Factors governing QbD implementation.
HTPD 2014 | Extended reports 29
Discussion
A mechanistic model is the ultimate level of process understanding achievable, and it has been
implemented successfully for a number of unit operations, for example, chromatographic
purification (IEC, SEC, AC), chemical/enzymatic reactions, and various analytical chromatography
techniques. Other unit operations such as mixing, chromatographic purification based on
hydrophobic interaction, and fermentation are still difficult to implement over a broader utility
range. In this context, HTPD techniques are very useful for generation of model parameters
from otherwise material- and time-consuming measurements of chromatographic binding
capacities in static mode and screening of reaction conditions including time. A potential issue
of model parameters established through HTPD techniques is a higher experimental uncertainty
due to scaling issues. However, process development times and experimental handling errors
often decrease using HTPD pushing bottlenecks from traditional manufacturing development to
analytical capacity concerns, while speed of experimentation increases dramatically.
Once reliable mechanistic (or statistical/empirical) models are established, they may be used for
a long list of applications including process development/optimization (3), troubleshooting, and
deviation handling (4), and mechanistic models are a very efficient tool in process understanding
discussions with regulatory authorities.
Mechanistic models
Extent of knowledge
Advanced statistical models/PCA etc.
Simple statistical models/DoE
OPAT
(One parameter at a time)
Preliminary experimentation
Experience from other projects
Theoretical knowledge
Fig 2. Process understanding elements and level of process knowledge obtained.
References
1. Kramarczyk, J., Masters Thesis. HTS of Chromatography Resins and Excipients for Optimizing Selectivity (2003).
2. Faber, C., PhD thesis. Measurement and Prediction of Protein Phase Behaviour and Protein-Protein-Interactions (2006).
3. Model Based Design of Experiments – Case Studies. M. Degerman, L. Sejergaard, E. Hansen, A.-M. Ludvig, E.B. Riis,
K.G. Jensby, and A. Staby. Oral presentation at: 241st ACS National Meeting Anaheim, CA, March 27 – 31, (2011).
4. Quality by Design: Regulatory Aspects. I. Mollerup, E. Hansen, J. Krarup, T.B. Hansen, S. Kidal, L. Sejergaard, T. StrømHansen, and A. Staby. Oral plenary lecture at: 238th ACS National Meeting, Washington DC, August 16–20 (2009).
30 HTPD 2014 | Extended reports
BioLector™ Pro – Expansion of
microbioreactor platform for strain
screening under full bioprocess control
A. Grimm, N. Frische, T. Olfers, and F. Kensy
m2p-labs GmbH, Baesweiler, Germany
email: [email protected]
Early phases in bioprocess development are based on large numbers of small-scale
experiments including clone and cell-culture medium screening. These are often conducted
in shaken bioreactors in uncontrolled processes, which are characterized by pH shifts and
constantly changing medium concentrations. In contrast, final production processes usually
take place in scales several magnitudes larger with comprehensive bioprocess control.
This disparity in operating conditions can cause significant problems during scale-up. For
this reason, liter-scale stirred-tank bioreactors (STR) are still largely used for bioprocess
development, allowing a higher level of control over the process. Unfortunately, the large scale
significantly limits throughput at a stage where rapid mass screening is desired.
While it is possible to simulate fed-batch fermentations with controlled-release systems, these
have certain drawbacks. Current controlled-release systems are often limited to a fixed feeding
rate and the nature of the applicable substrates. To rapidly test feeding strategies and feed
medium compositions, cultivation systems have to be flexible. Significantly more flexibility can
be warranted by miniaturized bioreactors (1–3).
m2p-labs here reports the extension of the application scope of their established BioLector
technology to yield a microbioreactor capable of full bioprocess control at microscale, BioLector
Pro. The application of the microbioreactor for screening of E. coli cultivations under different pH
conditions and feed rates is also described.
The BioLector Pro is a user-friendly microbioreactor system that can incorporate and substitute
the early stages of process development, and permits the direct transition to industrial large-scale
bioprocesses. As a high-throughput fermentation system, the main characteristics of the BioLector
Pro include high parallelization, a small working volume (typically 800 to 2400 μL), and the use of
standardized microplates (MTP) as bioreactors featuring 48 parallel reactors. The standardized
MTP format allows for universal automation. As a ready-to-use, disposable component to the
bioreactor system, the use of MTPs minimizes setup time, lowers costs, and increases efficiency.
m2p-labs’ FlowerPlate™ well geometry is designed to increase oxygen mass transfer (OTR). As a
reactor, it has the same performance as an industrial stirred-tank bioreactor (4).
Most importantly, the BioLector Pro system is based on online, noninvasive measurement.
Its fiber-optic measuring system is capable of monitoring the most relevant fermentation
parameters such as biomass concentration, pH, dissolved oxygen, and fluorescent proteins
and metabolites.
HTPD 2014 | Extended reports 31
Based on the same 48-well MTP format as the BioLector, microfluidic process control in the
BioLector Pro enables pH-controlled cultivations and fed-batch fermentations in up to 32 parallel,
individually controlled microbioreactors. Microfluidic structures are integrated in the bottom of a
FlowerPlate. Microchannels with microvalves connect reservoir wells containing feeding solutions
and pH adjusting agents with cultivation wells. The actuators of the microvalves are placed in the
fixation of the microplate. Therefore, with the separation of actuators of the valves and the valves
themselves, a user-friendly plug-and-play operation was constructed. By including microfluidic
structures, the fully instrumented microfluidic FlowerPlate is delivered as a ready-to-use sterile
disposable.
In conjunction with the novel features of the BioLector Pro, m2p-labs’ intuitive BioLection software
was also redesigned. For each specific culture well, several process conditions can be defined
individually. These include pH profiles, feed profiles, and trigger conditions for activation and
deactivation of feeding and pH control. Evaluation and visualization of growth and product
building kinetics such as the calculation of the specific growth rate, µ, are implemented.
Therefore, the same software package can be used for the cultivation setup and subsequent
data analysis. Evaluated data can also be exported into common spreadsheets for further
data analysis.
Due to the internal microfluidic fluid control on the microplate, no additional liquid handling
system is required, providing a more accurate scaled down model of industrial processes. The
capability to control bioprocesses unlocks a whole new array of applications. With the BioLector
Pro, clone screenings can be performed under different process conditions. Fermentation
parameters, fed-batch feed profiles, and media can be optimized at different pH values.
Bioprocesses themselves can be better and faster characterized. Coupled with this new potential,
the full scalability of the BioLector Pro to lab-scale bioreactors, pilot and production scales makes
it an excellent tool in bioprocess development. Cultivation process optimization can be shifted
entirely into microscale, omitting developmental runs in stirred tanks. STR fermentations then only
have to be conducted to verify results at scale.
Microscale results of fermentations with the BioLector Pro had previously been shown to be
scalable to fully controlled 2 L stirred tank bioreactors, whereby scale-up ranged over three
orders of magnitude from 500 µL to 1 L. Both pH and feeding control were reproduced, yielding
superimposable time courses regarding biomass concentration, dissolved oxygen, and pH
during cultivation of E. coli (3).
References
1. Betts J. I. and Baganz F. Miniature bioreactors: current practices and future opportunities. Microb. Cell Fact. 5:21. [Online.]
10.1186/1475-2859-5-21 (25 May 2006).
2. Scheidle M. et al. High-throughput screening of Hansenula polymorpha clones in the batch compared with the
controlled-release fed-batch mode on a small scale. FEMS Yeast Res. 10(1), 83–92 (2010).
3. Funke M. et al. Bioprocess control in microscale: Scalable fermentations in disposable and user-friendly microfluidic
systems. Microb. Cell Fact. 2010, 9:86 [Online.] DOI: 10.1186/1475-2859-9-86 (13 November 2010).
4. Funke M. et al. The baffled microtiter plate: Increased oxygen transfer and improved online monitoring in small scale
fermentations. Biotech. Bioeng. 30(6), 1118–1128 (2009).
32 HTPD 2014 | Extended reports
High-throughput microscale platform to
accelerate the development of particle
conditioning for biologics
A. Noyes1,2, B. Huffman3, A. Berrill3, N. Merchant4, R. Godavarti1, N. J. Titchener-Hooker2, K. Sunasara3,
and T. Mukhopadhyay2
1
Pfizer Bioprocess R&D, 1 Burtt Road, Andover, MA 01810, USA
2
The Advanced Centre for Biochemical Engineering, Department of Biochemical Engineering,
University College London, Bernard Katz Building, Gordon Street, London, WC1E 7JE, UK
3
Pfizer Bioprocess R&D, 700 Chesterfield Parkway, Chesterfield, MO 63017, USA
4
Pfizer Vaccine Research Unit, 401 N. Middletown Road, Pearl River, NY 10965, USA
email: [email protected]
This extended abstract is based on content previously published in references 1 and 2 and is
reproduced with kind permission from the publishers, Wiley. © 2015 Wiley Periodicals, Inc.
Introduction
Particle conditioning comprises a series of operations wherein insoluble particulates are formed
via flocculation and/or precipitation, dynamically conditioned to improve particle characteristics
for subsequent processing, and then separated from the clarified liquid. Particle conditioning is
effective for the removal of cells, cellular debris, proteins, DNA, and polysaccharides and has been
often employed for primary recovery in biopharmaceutical processes. The efficient development
of unit operations for particle conditioning, specifically flocculation and precipitation, has
been constrained by lab-scale models that require large volumes and considerable time to
evaluate. These problems are compounded for products that have several components requiring
purification, such as multivalent, capsular polysaccharide (CPS) vaccines. Development of
HTPD systems in this area has been hindered by a lack of (i) suitable high-throughput analytics,
(ii) qualified systems and hardware to control and measure the performance of the associated
unit operations, and (iii) established scaling methodologies for particle conditioning.
Description of system
A system for high-throughput particle conditioning (HTPC) integrating ultra scale-down (USD)
technology was developed. HTPC comprises USD flocculation, microplate centrifugation, and a
suite of high-throughput analytics as illustrated in Figure 1. The USD hardware for HTPC consists
of off-the-shelf parts and is designed to enable automated, temperature-controlled flocculation
of 96 different reactors in parallel (Fig 1). A robotic liquid handling system introduces the feed
material, flocculants, and titrant. A centrally positioned magnetic stirrer (0.64 cm diameter
stir bar) provides the agitation for each well in a standard 2 mL, deep, square-well microplate.
Following dynamic aging of particulates, clarified supernatant is produced with centrifugation
in a microplate. High-throughput analytics enable comprehensive analysis of product titer,
product quality, impurity clearance, clarification efficiency, and particle-size characterization.
The appropriate analytical methods can be selected from a library of high-throughput assays
qualified for CPS-containing microbial feedstreams. A typical 80-point screen requires 50 mL
total volume and can be performed in two to three workdays, including setup and multiplexed
analyses. The remaining wells on a HTPC plate are necessary for standard curves, blanks, and
controls during analysis.
HTPD 2014 | Extended reports 33
(A)
Multichannel liquid dispensing
Square
microwell
Stir bar
Copper shot
Heater block
Magnetic
tumble
stirrer
Mounting platform
(B)
Time
Flocculation: 1 d
Analytics: 1 d
Analysis: 1 d
Particle characterization
MFI
10 µL/well
Particle conditioning platform
Load material:
≤ 600 µL/well
Centrifugation/vacuum
Supernatant:
≤ 300 µL/well
CPS titer
Hb Ppt/UPLC-MALS
130 µL/well
Turbidity
OD 405/600
300 µL/well
DNA
PICOGREEN™
10 µL/well
Protein
Coomassie™ Plus
10 µL/well
Fig 1. (A) Schematic of the USD flocculation device and (B) flow diagrams for HTPC. Details for a typical screen
are provided. Adapted from a figure first published in reference 1. © 2015 Wiley Periodicals, Inc., reproduced
with permission from the publisher.
Although not geometrically similar, the USD flocculation system was designed to match or
minimize differences between key engineering elements of larger scale reactors as closely
as possible. Hydrodynamics in the microwell were examined with a paper-based engineering
characterization and empirical studies. Blend time studies and published correlations suggested
that turbulent conditions were present in the microwell, although the Reynolds numbers were
below typical ranges for turbulence. The power number for the magnetic stir bar was also
determined to be 0.7. When combined, these findings supported studies into the effects of
maintaining different hydrodynamic parameters constant during scaling.
34 HTPD 2014 | Extended reports
Applications
Four different capsular polysaccharides (CPS) from three bacterial species were employed to
assess scale-down from pilot-scale (2, 3, 13 L reactors) to USD-scale (≤ 1 mL reactor). The aim
was to align responses hypothesized to be dominated by thermodynamic equilibria, such as
impurity removal, product yield, and quality with more hydrodynamically sensitive responses
such as particle distributions and clarification. For staphylococcal broths containing anionic CPS,
full factorial experiments confirmed consistent scale-down of impurity removal, product yield,
and product utility across two and three-factor experimental spaces, with minor dependency on
scaling approach. DNA clearance from a CPS-containing broth from S. aureus is shown at two
different scales in Figure 2. Subsequently, protein and DNA clearance were demonstrated to be
scalable between the 1 mL and 700 L scales.
(A)
(B)
6.0
Floc agent concentration
2.0%
2.5%
3.0%
pH 5.8
16.8%
pH 5.5
1.2%
1.3%
0.9%
1.2 mL pH 5.8
7.1%
1.0%
pH 5.2
5.6
≤ 13%
≤ 9%
≤ 17%
5.4
≤ 21%
> 21%
5.2
1.1%
1.7%
≤ 5%
1.2%
pH 5.2
pH 5.5
≤ 1%
5.8
pH
3.5 L
SA1 DNA
0.8%
5.0
1.0
1.5
2.0
2.5
3.0
3.5
Floc agent concentration (%)
4.0
Fig 2. DNA clearance (percentage of load control) in the supernatant as a function of the floc agent concentration and
pH during the flocculation process for a staphylococcal CPS. Results for identical conditions at the 1.2 mL (average of
duplicates) and 3.5 L scales are presented in the matrix on the left, with gray boxes denoting data withheld to maintain
an identical comparison pattern (A). In (A), conditional formatting was performed as a single group to facilitate interscale
comparison. Full factorial results for the 1.2 mL system (in duplicate) are provided in the contour plot (B) where floc agent
percentage = 0, 1.0, 2.0, 2.3, 2.5, 2.7, 3.0, 4.0 and pH = 5.0, 5.3, 5.5, 5.7, 6.0, with each condition represented by small black
squares. The large black squares represent conditions evaluated at both the 1.2 mL and 3.5 L scales. First published in
reference 1. © 2015 Wiley Periodicals, Inc., reproduced with permission from the publisher.
Particle characterization was the response most sensitive to scaling between USD and
pilot-scale (Fig 3). With a meningococcal broth containing an anionic CPS, scaling down by
maintaining blend time (θ0.95), the lowest rpm required to maintain particle suspension (Njs), or
power per volume in the reactor tank (P/Vavg) (confounded with Njs) produced fewer particles
and larger average diameters in HTPC as compared to the pilot-scale benchmark. In contrast,
scaling by maintaining tip speed (vtip) and/or power per volume in the impeller zone (P/Vimp)
enabled closer reproduction of particle populations across flocculation scales. This phenomenon
is consistent with enhanced particle breakup mediated by increased levels of input energy and
shear. P/Vimp and vtip characterize the most vigorous local environments that a given particle will
be exposed to during aging. Since all flocs will pass through the impeller zone multiple times
during the aging process, it is reasonable that the local energy dissipation rate (and therefore
characteristic scale of turbulent eddies) and maximum shear rates will be decisive factors in
determining the extent of particle break-up. The number of passes a particle makes through
the impeller region will be a function of aging time. Although time was not a variable in this
trial, in a separate study, protein removal and clarification were dependent on the aging time,
necessitating linear time-integration of scaling factors. Therefore, the time-integrated factors,
P/Vimp t, vtip t, or a Camp number (Ca) permutation substituting P/Vimp for P/Vavg represent the
most appropriate scaling terms.
HTPD 2014 | Extended reports 35
(A)
Pilot-scale reference
2.5 × 106
USD; Scale by P/Vimp, vtip 1 mL
1 mL
USD, Scaling by Njs, θ0.95
9
7
6
5
4
3
2
Scale
2350 mL
USD; Scale by P/Vimp, vtip 1 mL
USD, Scaling by Njs, θ0.95
8
Particle count (particles/mL)
Number-average particle diameter (µm)
10
(B)
Scale
2350 mL
Pilot-scale reference
1 mL
2.0 × 106
1.5 × 106
1.0 × 106
5.0 × 105
1
0
1%
2%
Floc agent concentration
0
3%
1%
2%
Floc agent concentration
3%
Fig 3. Scaling by θ0.95, Njs, P/Vavg, Ca, or νtip, P/Vimp.Effect on particle characterization of scaling meningococcal
flocculation using different scaling rules. Number-average particle diameters (A) and concentrations of the raw floc
suspension diluted 100-fold diluted (B). The pH did not influence results and for a given floc agent concentration,
the results for all pH set points were averaged. For HTPC, a full factorial screen was performed with floc agent
percentage = 1.0, 2.0, 3.0 and pH = 4.5, 5.0, 5.5. For pilot-scale, the floc agent percentage/pH conditions were
1.0/pH 4.5, 1.0/pH 5.5; 2.0/pH 5.0, and 3.0/pH 4.5; 3.0/pH 5.5. The centerpoint was run in duplicate. First published in
reference 2. © 2015 Wiley Periodicals, Inc., reproduced with permission from the publisher.
With a refined HTPC system, a multifactor full factorial design of experiments (DoE) study was
performed with pneumococcal broth containing a neutral CPS. Standard DoE analysis led to
quadratic statistical models relating responses to input factors. These models were combined
with prior knowledge to create operating windows, as shown in Figure 4. Previously, untangling
these relationships would have been impractical. Figure 4 illustrates the utility of HTPC in
understanding many of the multivariate interactions that underpin flocculation, ultimately
leading to more robust and productive processes.
4.0
3.8
3.6
pH
ivb
iva
3.4
iii
Design space
3.2
i
ii
1.0
1.5
2.0
Floc agent (%)
2.5
3.0
Fig 4. Window of operation for a particle conditioning operation performed with a neutral pneumococcal CPS. The
unshaded area delineates the design space, the operating region where acceptable performance was achieved.
Statistical models for acceptable downstream processing as defined by: filtration performance (i), ≤ 10% of load
DNA (ii), ≤ 50% of load A600 (iii) and ≤ 3% of load protein (iva) were combined to create the operating window when
Ca = 6.5 × 105. If Ca were increased to 1.3 × 106 (ivb), the operating window would expand, owing to improved
protein clearance. First published in reference 2. © 2015 Wiley Periodicals, Inc., reproduced with permission from the
publisher.
36 HTPD 2014 | Extended reports
Conclusions
The innovative system and methods of HTPC enable the assessment of flocculation and
centrifugation performance with up to 96 conditions in parallel. A comparison of potential scaling
rules suggests that the most consistent scale-up is achieved by maintaining constant P/Vimp and/
or tip speed. Time-integration of either parameter (i.e., P/Vimp t or vtip t) or P/Vimp permutation of
Camp number is supported by time-sensitive clarification and protein clearance results. It was
found that evaluating scalability in experimental areas where a response changes sharply as
a function of input parameter(s) represents the most rigorous test for the suitability of a given
scaling rule. The described system forms the basis for platform workflows, enabling one scientist
to develop a robust particle conditioning operation in less than one week with 50 mL of feed.
References
1. Noyes, A. et al. High throughput screening of particle conditioning Operations: I. System design and method
development. Biotechnol. Bioeng. 112, 1554–1567 (2015).
2. Noyes, A. et al. High throughput screening of particle conditioning operations: II. Evaluation of scale-up heuristics with
prokaryotically expressed polysaccharide vaccines. Biotechnol. Bioeng. 112, 1568–1582 (2015).
HTPD 2014 | Extended reports 37
Understanding the chromatography
behavior of monoclonal antibodies
using quantitative structure-property
relationship analysis
B. Tran, B. Connolly, T. Patapoff, and P. McDonald
Genentech
email: [email protected]
High-throughput partition coefficient (Kp) determination enables the rapid mapping of antibody
binding behavior on different chromatography media (resins). In our group, these experimental
screens are used to guide chromatography development by assessing the fit of new therapeutic
antibodies to different purification process options. Through this work, we have generated a rich
library of Kp data reported as a log Kp for antibodies as a function of pH and ionic strength on
several chromatography media. Correlating the results from this library of experimental data
to molecular structure is an avenue to allow us to predict the behavior of antibodies on our
chromatography media solely from their sequence. An in silico tool for predicting binding behavior
on chromatography media would enable the mapping of antibody behavior on our media earlier
in the drug discovery process where material availability can preclude experimentation.
Herein, we evaluate quantitative structure-property relationship (QSPR) analysis as a tool for the
in silico prediction of binding behavior of our antibodies on anion-exchange chromatography
media. QSPR analysis is an established technique used to relate molecular descriptors that
quantitatively describe the chemical structure of small molecules to experimentally determined
properties. We determined chemical structure information for eight antibodies using homology
and molecular dynamics modeling using antibody primary amino acid sequences. Molecular
descriptors were then calculated using specific generated outputs including, but not limited to,
residue position, charge, solvent-accessible surface area, and trajectories.
Using single-variable ordinary least squares regression analysis to probe individual molecular
descriptor correlations, we found that molecular descriptors encoding for the geometric
spatial distribution of only the negatively charged and solvent-accessible residues are
effective determinants of antibody behavior on anion-exchange chromatography media. We
subsequently performed a multivariate partial least squares regression analysis using two
latent variables. Figure 1 compares the predicted vs experimentally determined log Kp values for
eight monoclonal antibodies (MAbs) on an anion-exchange medium using this approach. The
cross-plot indicates good agreement between the experimentally determined and predicted
log Kp values. The residuals did, however, increase with increasing log Kp. A visual comparison
of the contour plots nevertheless indicates similar log Kp contour profiles for each of the MAbs
between the experimentally determined and predicted log Kp values (Fig 2, data shown for four
representative MAbs).
We are continuing to refine this approach for predicting the log Kp of our MAbs in order to
improve the correlation between the experimentally determined results and the results
obtained with QSPR analysis.
38 HTPD 2014 | Extended reports
Experimental log Kp vs predicted log Kp
4
MAb 1
MAb 2
MAb 3
MAb 4
2
Residuals vs predicted log Kp
1
1
0
Residuals
Experimental log Kp
3
MAb 5
MAb 6
MAb 7
MAb 8
-1
0
-1
-2
-2
-1
0
1
2
Predicted log Kp
3
4
-2
-1
0
1
2
Predicted log Kp
3
4
Fig 1. Comparison of experimentally determined log Kp values vs predicted log Kp values for eight MAbs. Predicted
values were determined from a multivariate partial least squares regression analysis.
Experimental
Predicted
8.5
8.0
8.0
pH
pH
8.5
7.5
7.5
7.0
7.0
0
0
40
80
120
Salt conc. (mM)
8.0
8.0
Log Kp
Low binding
pH
8.5
pH
8.5
40
80
120
Salt conc. (mM)
7.5
7.5
7.0
7.0
0
40
80
120
Salt conc. (mM)
0
8.0
8.0
Intermediate binding
pH
8.5
pH
8.5
40
80
120
Salt conc. (mM)
7.5
7.5
7.0
7.0
0
40
80
120
Salt conc. (mM)
8.0
8.0
40
80
120
Salt conc. (mM)
0
40
80
120
Salt conc. (mM)
High binding
pH
8.5
pH
8.5
0
7.5
7.5
7.0
7.0
0
40
80
120
Salt conc. (mM)
Fig 2. Comparison of log Kp contour profiles for four MAbs using high-throughput screening and prediction using
multivariate partial least squares regression analysis of selected molecular descriptors.
HTPD 2014 | Extended reports 39
List of posters presented at HTPD 2014*
* As listed in the HTPD 2014 conference program. Some differences between the list and the final posters presented might exist.
†
Denotes author for correspondence.
1.
High-throughput sorbent screening using 96-well plates and prepacked columns for the
purification of a Fab antibody fragment
V. Ravault and V. Brochier†
Pall, 48 avenue des Genottes, 95800 Cergy Saint Christophe, France
2.
The challenges and relationships in using high-throughput experiments for mechanistic
modeling of an industrial chromatography characterization
C. Williams†, J. Winderl, T. Hahn, and T. Huuk
Genentech Inc., Karlsruhe Institute of Technology, Germany
3.
Mini-columns as a complement to 96-well filter plates in a process development workflow
A. Grönberg†, E. Brekkan, A. Edman Örlefors, and K. Nilsson-Välimaa
GE Healthcare Life Sciences, Björkgatan 30, 751 82 Uppsala, Sweden
4.
Buffer plate preparation using AKTA™ avant and BufferPro
S. Westin, G. Rodrigo†, and E. Carredano
GE Healthcare Life Sciences, Björkgatan 30, 751 82 Uppsala, Sweden
5.
Quality by Design studies — effect on dynamic binding capacity and selectivity from ligand
density variations on Capto™ adhere ImpRes
L. Kärf, E. Brekkan†, K. Nilsson-Välimaa, and M. Ahnfelt
GE Healthcare Life Sciences, Björkgatan 30, 751 82 Uppsala, Sweden
6.
Process development of aqueous two-phase systems combining integration of automated
liquid handling with process modeling
N. Patel†1,2, E. Sorensen2, and D. G. Bracewell1
1
2
7.
Advanced Centre for Biochemical Engineering, Department of Biochemical Engineering, UCL, London, UK
Centre for Process Systems Engineering, Department of Chemical Engineering, UCL, London, UK
Microfluidics on liquid handling stations (μF-on-LHS): closing the gap between highthroughput experimentation and microfluidics
C. P. Radtke†1, J. Kittelmann1, A. Waldbaur2, B. E. Rapp2, and J. Hubbuch1
Institute of Process Engineering in Life Sciences, Section IV: Biomolecular Separation Engineering (MAB), Karlsruhe Institute of
Technology (KIT), Germany
2
Institute of Microstructure Technology (IMT), Karlsruhe Institute of Technology (KIT), Germany
1
8.
A high-throughput based approach to assess resin aging – an automated life time study on
protein A column material
D. Boeth†, C. Atzkern, H. Geier, and H. Rogl
Boehringer Ingelheim Pharma GmbH & Co. K, Germany
9.
High-throughput process development: Quality by Design studies across multiple formats
J. Feliciano1, M. Ahnfelt2, A. Berrill3, E. Brekkan2, B. Evans3, Z. Fung1, R. Godavarti4,
K. Nilsson-Välimaa2, J. Salm4, M. Switzer4, and K. M. Łacki2
GE Healthcare Life Sciences Fast Trak Services, 800 Centennial Ave., Piscataway, NJ 08854, USA
GE Healthcare Life Sciences, Björkgatan 30, 751 82 Uppsala, Sweden
Pfizer R&D Global Biologics, 700 Chesterfield Parkway, Chesterfield, MO 63017, USA
4
Pfizer R&D Global Biologics, 1 Burtt Road, Andover, MA 01810, USA
1
2
3
10. High-throughput development of non-protein A monoclonal antibody purification process
using minicolumns and bio-layer interferometry
V. Ravault† and V. Brochier
Pall, 48 avenue des Genottes, 95800 Cergy Saint Christophe, France
40 HTPD 2014 | Extended reports
11. Automated small-scale protein purification for accelerated development of protein
therapeutics
N. Verlinden†1, L. Jordan1, R. Beighley1, K. McGuire1, B. Gerwe1, J. Lambert1, M. Costioli2,
C. Kaltenbach2, and X. LeSaout2
1
2
PerkinElmer, Inc., 940 Winter Street, Waltham, MA USA
Merck KGaA, Darmstadt, Germany
12. High-throughput solution screening documented and evaluated with an electronic system
R. Nachtigall†, G. Pollinger, A. Kleinjans, and F. Zettl
Roche Diagnostics GmbH, Penzberg, Germany
13. Influence of chromatography media sampling in plate-based benchmarking studies
T. Bergander† and K. M. Łacki
GE Healthcare Life Sciences, Björkgatan 30, 751 72 Uppsala, Sweden
14. Tuning evolutionary multiobjective optimization to estimate chromatographic operating
conditions
R. Allmendinger†, S. Gerontas, N. J. Titchener-Hooker, and S. S. Farid
Department of Biochemical Engineering, University College London, London, Torrington Place, London WC1E 7JE, UK
15. RoboChrom: a highly flexible chromatography screening tool
D. Harde† and M. Berkemeyer
Boehringer-Ingelheim RCV Vienna, Austria
16. Purification of interferon α-2a – a process development study
S. Grönlund†, K. Eriksson, J. Shanagar, C. Brink, E. Pool, A. Moberg, M. Winkvist, and A. Grönberg
GE Healthcare Life Sciences, Björkgatan 30, 751 82 Uppsala, Sweden
17. Utilization of high-throughput screening in the development of HyCell™ CHO production
medium
E. Garner†, A. Elwood, J. Manwaring, and M. Wight
GE Healthcare Life Sciences, HyClone™ Cell Culture, Logan, Utah, USA
18. BioLector™ Pro - Expansion of microbioreactor platform for strain screening under full
bioprocess control
A. Grimm†, N. Frische, T. Olfers, and F. Kensy
m2p-labs GmbH, Baesweiler, Germany
19. Self-optimization of robotic liquid classes
F. Nossek†, K. Ehrhard, S. Werz, S. Hepbildikler, and C. Bell
Rocwhe Diagnostics GmbH Dovnstream Development, Germany
20. SHARC – software integration for high-throughput systems
K. Doninger†, A. Kurtenbach, J. Griesbach, S. Markert, C. Musmann, S. Fan, H. Nibbrig, and A. Jux
Roche Diagnostics GmbH, Penzberg, Germany
HTPD 2014 | Extended reports 41
Author index
Ahnfelt, M. Allmendinger, R. Atzkern, C. 40
41
40
Bao, H. Barlow, K. Baumann, P. Beighley, R. Bell, C. Bergander, T. Berkemeyer, M. Berrill, A. Boeth, D. Bracewell, D. G. Brekkan, E. Brink, C. Brochier, V. 23
23
13
41
41
6, 41
19, 41
33, 40
40
40
10, 23, 40
41
40
Carredano, E. Christanti, S. Connolly, B. Costioli, M. 40
26
38
41
Daniels, C. Doninger, K. Dürauer, A. 26
41
19
Edman Örlefors, A. Ehrhard, K. Elwood, A. Eppink, M. H. M. Eriksson, K. Evans, B. 10, 40
41
41
16
41
40
Fan, S. Farid, S. S. Feliciano, J. Frische, N. Fung, Z. 41
41
40
31, 41
40
Garner, E. Geier, H. Gerontas, S. Gerwe, B. Godavarti, R. Griesbach, J. Grimm, A. Grönberg, A. Grönlund, S. 41
40
41
41
33, 40
41
31, 41
10, 40, 41
41
42 HTPD 2014 | Extended reports
Hahn, T. Hanke, A. T. Hansen, M. H. Hansen, T. B. Harde, D. Hepbildikler, S. Hubbuch, J. Huff, K. Huffman, B. Huuk, T. 13, 40
16
29
29
41
41
13, 40
26
33
13, 40
Jordan, L. Jux, A. 41
41
Kaltenbach, C. Kensy, F. Kidal, S. Kittelmann, J. Kleinjans, A. Kurtenbach, A. Kärf, L. Łacki, K. M. Lambert, J. LeSaout, X. Lim, E. Linden, T. 41
31, 41
29
40
41
41
40
6, 23, 40, 41
41
41
26
23
Manwaring, J. Markert, S. McDonald, P. McGuire, K. Merchant, N. Moberg, A. Mukhopadhyay, T. Musmann, C. 41
41
38
41
33
41
33
41
Nachtigall, R. Nibbrig, H. Nilsson Välimaa, K. Nossek, F. Noyes, A. 41
41
10, 40
41
33
Olfers, T. Olsen, A. A. Olson, J. Ottens, M. 31, 41
29
26
16
Patapoff, T. Patel, N. Pollinger, G. Pool, E. 38
40
41
41
Radtke, C. P. Rapp, B. E. Ravault, V. Rodrigo, G. Rodriguez, J. Rogl, H. Roush, D. 40
40
40
40
26
40
23
Salm, J. Sejergaard, L. Shanagar, J. Sorensen, E. Staby, A. Strøm-Hansen, T. Sunasara, K. Switzer, M. 40
29
41
40
29
29
33
40
Titchener-Hooker, N. J. 33, 41
Tran, B. 38
Valentin, S. van de Sandt, E. J. A. X. van der Wielen, L. A. M. Verhaert, P. D. E. M. Verlinden, N. 29
16
16
16
41
Waldbaur, A. Walther, C. Wang, S.-c. Welsh, J. Wenger, M. Werz, S. Westi, S. Wight, M. Williams, C. Winder, J. Winkvist, M. 40
19
26
23
26
41
40
41
40
40
41
Zettl, F. 41
Østergaard, P. Østfeldt, P. 29
29
GE Healthcare Bio-Sciences AB
Björkgatan 30
751 84 Uppsala
Sweden
www.gelifesciences.com
GE, GE monogram, ÄKTA, Capto, HiScale, HiScreen, HiTrap, HyCell, HyClone, MabSelect SuRe, MabSelect Xtra, SOURCE, and Tricorn are
trademarks of General Electric Company.
Ambr, and SelecT are trademarks of Sartorious Stedim Biotech. BioLector and FlowerPlate are trademarks of m2p-labs. Coomassie
is a trademark of Thermo Fisher Scientific LLC. MediaScout and RoboColumn are trademarks of Atoll GmbH. Microlab and STAR are
trademarks of the Hamilton Company. PICOGREEN is a trademark of Life Technologies Corporation. Tecan and Te-Shake are trademarks
of Tecan Group, Ltd. All other third-party trademarks are the property of their respective owners.
© 2016 General Electric Company. First published Feb. 2016
All goods and services are sold subject to the terms and conditions of sale of the company within GE Healthcare which supplies them.
A copy of these terms and conditions is available on request. Contact your local GE Healthcare representative for the most current information.
GE Healthcare UK Ltd., Amersham Place, Little Chalfont, Buckinghamshire, HP7 9NA, UK
GE Healthcare Europe GmbH, Munzinger Strasse 5, D-79111 Freiburg, Germany
GE Healthcare Bio-Sciences Corp., 100 Results Way, Marlborough, MA 01752, USA
GE Healthcare Dharmacon Inc., 2650 Crescent Dr, Lafayette, CO 80026, USA
HyClone Laboratories Inc., 925 W 1800 S, Logan, UT 84321, USA
GE Healthcare Japan Corp., Sanken Bldg., 3-25-1, Hyakunincho Shinjuku-ku, Tokyo 169-0073, Japan
For local office contact information, visit www.gelifesciences.com/contact
29179765 AA 02/2016