Extended reports and posters from the ... - HTPD Meetings

GE Healthcare

Extended reports and posters from the third international conference on high-throughput process developmentSiena, Italy, October 6–9, 2014

HTPD 2014 | Extended reports 3

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

4 HTPD 2014 | Extended reports


From the conference chairs

Philip Lester Genentech

Mats Gruvegard GE Healthcare

Karol Łącki Novo Nordisk

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.


Equilibrium liquid concentration (mg/mL)0 1 2 3 4 5

Equi

libriu

m m

edia

cap

acity

(mg/

mL)

0

20

40

60

80

100

120

140

160

180 Adsorption isothermAssumed phase ratio20% overestimation of media volume20% underestimation of media volume

∆CEq

∆CEq

∆Vmedia ∆Vmedia

Equilibrium liquid concentration (mg/mL)0 0.05 0.10 0.15 0.20 0.25

Equi

libriu

m m

edia

cap

acity

(mg/

mL)

0

2

4

6

8

10

12

14

Adsorption isothermAssumed phase ratio20% overestimation of VMedia

20% underestimation of VMedia

High-throughput process development: chromatography media volume definitionT. Bergander and K. M. ŁackiGE Healthcare, Björkgatan 30, 751 84 Uppsala, Sweden

email: [email protected]

Full article to be published in Eng. Life Sci., DOI: 10.1002/elsc.201400240

IntroductionUse 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.

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.

(A) (B)


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 outlineFive 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 definitionThree 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 determinationFor 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).

Equilibrium liquid concentration (mg/mL)

0 1 2 3 4 5

Equi

libriu

m m

edia

cap

acity

(mg/

mL)

0

20

40

60

80 Adsorption isothermAssumed phase ratio

Upper range, β

Lower range, βPerceived adsorption isothermActual phase ratio

∆CEq

∆CEq

β ± ∆β β ± ∆β


DW

DG

ravi

ty S

ettle

d/D

WD

Col

umn (

%)

Chromatography media

0

20

40

60

80

100

120

SOURCE 30Q Capto Q Q SepharoseBig Beads

Capto QImpRes

ResultsDWD of chromatography media used in the study determined by the packed column method varied between 90 and 310 mg/mL (DWD

Column, 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 DWD

Column 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.

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 DWD

Plaque 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 DWD

Plaque 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.


Plaque volume (µL)0 10 20 30 40 50 60

DW

DPl

aque

/DW

DC

olum

n (%

)

0

20

40

60

80

100

120

Capto Q DWDPlaque

Capto Q DWDColumn

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.

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.

SummaryWhen 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.

References1. 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 Laser-Rester-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).


Minicolumns as a complement to 96-well filter plates in a process development workflowA. Grönberg, E. Brekkan, A. Edman Örlefors, and K. Nilsson VälimaaGE Healthcare, Björkgatan 30, 751 84 Uppsala, Sweden


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 (Q

B10%). 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.


pH

NaC

l con

cent

ratio

n (m

M)

5.0 5.2 5.4 5.6 5.8 6.0 6.2 6.4 6.6 6.8 7.0

0

50

100

150

200

250

300

350

400

450

500 95908580757065605550454035302520151050

2

3

1 4

Elution conditions

Yield (%)

7

6

5

4

3

2

1

00 1 2 3 54 6 7 9 108 12 1311

Volume (mL)

A30

0 nm

(AU

)

7

6

5

4

3

2

1

00 1 2 3 54 6 7 9 108 12 1311

Volume (mL)

A30

0 nm

(AU

)

Sample load: 60 mgMAb/mL pH 6.0

Wash: 50 mM acetate250 mM NaCl pH 5.0, 3 CV

Elution: 50 mM phosphate250 mM NaCl pH 7.0

CIP: 1.0 M NaOH

Strip:50 mM Tris1.0 M NaCl

pH 8.0

Flow rate: 1.85 µL/sResidence time: 5.4 min

Elution pool: Yield 79%[HCP] 188 ppm

Elution pool: Yield 97%[HCP] 126 ppm

CIPStripSample load

Wash: 50 mM phosohatepH 6.8, 3 CV

Elution

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).

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

Fig 2. Two chromatograms generated from wash evaluation on PreDictor RoboColumn: (A) wash 2; (B) wash 4.

(A) (B)


NaC

l (m

M)

45

40

35

30

25

20

15

10

5

0

pH6.0 6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8 6.9

pH6.0 6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8 6.9

pH6.0 6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8 6.9

Wash volume 1 CV 3 CV 5 CV

[HCP] in eluates (ppm)

Selected wash

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.

Fig 3. Response contour plots showing the HCP concentration in eluates after wash with different pH and NaCl concentrations.

ConclusionsPreDictor 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.


Rapid process development with UV absorption-based inverse modeling of protein chromatographyT. Hahn, P. Baumann, T. Huuk, and J. HubbuchBiomolecular Separation Engineering, Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany


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, model-based 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:

𝑚𝑚𝑚𝑚𝑚𝑚𝑝𝑝

∑ �𝑚𝑚(𝑡𝑡𝑗𝑗) − ∑𝑖𝑖≥1𝑐𝑐𝑖𝑖(𝐿𝐿, 𝑡𝑡𝑗𝑗;𝑝𝑝) ⋅ 𝑎𝑎𝑖𝑖�

2

𝑗𝑗 ,

𝜕𝜕𝑞𝑞𝑖𝑖′

𝜕𝜕𝜕𝜕= 𝑘𝑘𝑎𝑎𝑎𝑎𝑎𝑎,𝑖𝑖 �𝛬𝛬 − ∑ 𝜈𝜈𝑗𝑗+𝜎𝜎𝑗𝑗

𝑎𝑎𝑗𝑗𝑞𝑞𝑗𝑗′𝑘𝑘

𝑗𝑗=1 �𝜈𝜈𝑖𝑖𝑐𝑐𝑝𝑝,𝑖𝑖′ − 𝑘𝑘𝑎𝑎𝑑𝑑𝑎𝑎,𝑖𝑖𝑐𝑐𝑎𝑎

𝜈𝜈𝑖𝑖𝑞𝑞𝑖𝑖′,

𝑞𝑞𝑎𝑎𝑎𝑎𝑠𝑠𝜕𝜕 = 𝛬𝛬 − ∑ 𝜈𝜈𝑗𝑗𝑎𝑎𝑗𝑗

𝑘𝑘𝑗𝑗=1 𝑞𝑞𝑗𝑗′ ,


𝜕𝜕𝜕𝜕= 𝑘𝑘𝑎𝑎𝑎𝑎𝑎𝑎,𝑖𝑖

′ 𝑞𝑞𝑚𝑚𝑎𝑎𝑚𝑚,𝑖𝑖′ �1 −∑

𝑞𝑞𝑗𝑗′

𝑞𝑞𝑚𝑚𝑚𝑚𝑚𝑚,𝑗𝑗′

𝑘𝑘𝑗𝑗=1 � 𝑐𝑐𝑝𝑝,𝑖𝑖

′ − 𝑘𝑘𝑎𝑎𝑑𝑑𝑎𝑎,𝑖𝑖𝑞𝑞𝑖𝑖′.

(1)

with ρ being the unknown parameter set to be estimated. To simulate directly in [mAU], we multiplied all molar concentration values in the model equations with the unknown factors ai. We obtained a set of equations that uses [mAU]-based boundary equations, which could easily be set by dividing the observed peak areas in [mAU∙mL] by the sample volume in [mL]. The equations for convection and diffusion in the interstitial volume of the mobile phase, as well as 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.


For models including stoichiometric ion-exchange, such as the Steric Mass Action (SMA) model (1) or the model for multimodal chromatography. (2), a second isotherm equation for the counter-ion balance allows for determining a

i uniquely. For the SMA model we obtain



2

𝑗𝑗 ,











𝑞𝑞𝑗𝑗′




(2)



2

𝑗𝑗 ,











𝑞𝑞𝑗𝑗′




(3)

with q'i = ai · qi and c'p,i = ai · cp,i.

Binding models, such as the Langmuir isotherm do not include a second equation, such that the ai cannot be determined uniquely:



2

𝑗𝑗 ,











𝑞𝑞𝑗𝑗′



′ − 𝑘𝑘𝑎𝑎𝑑𝑑𝑎𝑎,𝑖𝑖𝑞𝑞𝑖𝑖′. (4)

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 Cherry-tagged 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 single-component 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.


2.5(A) (B)

(E) (F)

(C) (D)

2.0

1.5

UV

280

nm (A

U)

Salt

(M)

Volume (mL)0 5 10 15

1.0

0.5

0

1.0

0.8

0.6

0.4

0.2

0

2.0

1.5

UV

280

nm (A

U)

Salt

(M)

Volume (mL) Volume (mL)0 5 10 15 20

1.0

0.5

0

1.0

0.8

0.6

0.4

0.2

0

0.05

0.04

UV

536

nm (A

U)

Salt

(M)

0 5 10 15 20

0.03

0.02

0.01

0

1.0

0.8

0.6

0.4

0.2

0

2.0

2.5

3.0

1.5

UV

300

nm (A

U)

Salt

(M)

Volume (mL)0 10 20

1.0

0.5

0

1.0

0.8

0.6

0.4

0.2

0

0.8

1.0

0.6

UV

536

nm (A

U)

Salt

(M)

Volume (mL)0 10 20

0.4

0.2

0

1.0

0.8

0.6

0.4

0.2

0

UV

536

nm (A

U)

Salt

(M)

Volume (mL)0 5 10 15

0.02

0.03

0.04

0.05

0.06

0.01

0

1.0

0.8

0.6

0.4

0.2

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.

References1. 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).


Fourier transform assisted deconvolution of complex multidimensional chromatogramsA. 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 Netherlands2 DSM Biotechnology Center, Delft, The Netherlands3 Synthon Biopharmaceuticals B.V., Nijmegen, The Netherlands


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.

IntroductionAfter 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 deconvolutionTo 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


1200

0

10 2624222018161412

200

400

600

800

Abs

UV 2

80 (m

AU)

tR,D1 (min)

1000

1200

0

10 2624222018161412

200

400

600

800

Abs

UV 2

80 (m

AU)

tR,D1 (min)

1000

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.

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.

(A) (B)


0

25

50

-25

-5010

15

20

25 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5

Abs

280

nm (m

AU)

tR,D1 (min)tR,D2 (min)

0

25

50

-25

-5010

15

20

25 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5

Abs

280

nm (m

AU)

tR,D1 (min)tR,D2 (min)

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.

ConclusionsFourier 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.

References1. 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).

(A) (B)


Binding and elution of successfully refolded

protein

Unfolded and misfolded protein found in �ow through

Selective capture of refolded protein

Turbidity by absorbance @ 600 nm

Refolding screening

Turbidity indicates aggregation / precipitation = unfavored conditions

Discard turbid samples

Determine wavelength shift and scattering

cGE analysis of eluted protein

Intrinsic �uorescence emission scan

Solubilization screening

Catching up bioprocess development on the downstream sideC. Walther1, M. Berkemeyer2, and A. Dürauer1

1University of Natural Resources and Life Sciences, Vienna2Boehringer-Ingelheim RCV


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).

Fig 1. Experimental design of IB renaturation screening process.


00

2

4

Solu

biliz

ed p

rote

in (m

g/m

L)

6

8

1201008060Time (min)

4 M Urea, pH 9.0

4 M U

rea, p

H 7.5

8 M U

rea, p

H 9.0

6 M U

rea, p

H 9.0

4 M U

rea, p

H 9.0

8 M U

rea, p

H 10.5

3 M G

uHCl, pH 7.5

4 M G

uHCl, pH 7.5

5 M G

uHCl, pH 7.5

3 M G

uHCl, pH 9.0

4 M G

uHCl, pH 9.0

5 M G

uHCl, pH 9.0

3 M G

uHCl, pH 10.5

4 M G

uHCl, pH 10.5

5 M G

uHCl, pH 10.5

8 M U

rea, 0

.5 M G

uHCl, pH 7.5

8 M U

rea, 0

.5 M G

uHCl, pH 9.0

8 M U

rea, 0

.5 M G

uHCl, pH 10.5

8 M U

rea, 0

.5 M G

uHCl, 10 m

M DTT

, pH 10.5

8 M U

rea, 0

.5 M G

uHCl, 20 m

M DTT

, pH 10.5

6 M U

rea, p

H 10.5

4 M U

rea, p

H 10.5

8 M U

rea, p

H 7.5

6 M U

rea, p

H 7.5

4 M GuHCl, pH 7.5

8 M Urea, pH 10.5

8 M Urea, 0.5 M GuHCl, pH 10.5

4020

0

20

40

Solu

biliz

ed p

rote

in (%

)

60

100

80

(A)

(B)

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 high-throughput 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.

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.


96-well microplateHeight, H: 10.9 mmFill level (180 µL), lF: 5 mmDiameter top, d: 6.96 mmDiameter bottom, d´: 6.58 mm Integrated shaker:Bioreactor: Tecan™ Te-Shake™Mixing orbit: 3 mmMixing frequency: 650 rpmTemperature: 21°C

Laboratory-scale reactorHeight, H: 90 mmFill level (100 µL), lF: 52 mmInner diameter, i.d. : 52 mm Impeller stirring:Pitched blade element, upwardFour impellers, 45°Height, H: 9 mmDiameter, d: 25 mmStirrer frequency: 200 rpmTemperature: 21°C

lF

h

d‘

d

96-well microplate, turbidity measured

at OD 600 nm

Lab-scale reactor, turbidity measured at

880 nm (backscattering)

D

HLF

d

h

Turbidity measurement in microplate wells

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.

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 pitched-blade 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).


Turb

idity

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0

0123456789101112

AB

CD

EF

GH

350

Den 1:25

Den 1:25

S14

S14S1

S24

S24

S39

S39S71

S32

S32

STD

STD

S91

S91

S21

S21

S66

S66

330

335Wav

elen

gth

(mm

)

340

345

80 000

0

20 000

Scat

tere

d lig

ht in

tens

ity

40 000

60 000

(A)

(C)

(B)

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.

Reference1. Walther, C. et al. Prediction of inclusion body solubilization from shaken to stirred reactors. Biotechnol. Bioeng. 111,

84–94 (2014).


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

1Merck & Co., Inc., Kenilworth, NJ, USA2GE Healthcare, Björkgatan 30, 751 84 Uppsala, Sweden


Full article published in Eng. Life Sci., DOI: 10.1002/elsc.201400251

IntroductionAutomated, 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 platesIn 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.


Capto adhere

Prototype HighPrototype Mid

Prototype Low

pH pH

pH pH

NaC

l (m

M)

800

600

400

200

03.5 9.08.58.07.57.06.56.05.55.04.54.0

NaC

l (m

M)

NaC

l (m

M)

NaC

l (m

M)

Yield boundary

Yield boundary

Yield boundary

Aggregates boundary

800

600

400

200

03.5 9.08.58.07.57.06.56.05.55.04.54.0

800

600

400

200

03.5 9.08.58.07.57.06.56.05.55.04.54.0

800

600

400

200

03.5 9.08.58.07.57.06.56.05.55.04.54.0

Aggregates boundary

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.

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.


0

20

40

60

80

100

0 200 400 600

Yiel

d (%

) or

aggr

egat

e re

mov

al (%

)

Capto – Yield (%)

Prototype High – Yield (%)

Capto – Aggregate removal (%)

Prototype High – Aggregate removal (%)

NaCl (mM)

Results – miniature column screensMiniature 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.

Fig 2. Comparison of miniature column performance at pH 6.5 across different salt concentrations for Capto adhere and Prototype High media.

ConclusionThis study demonstrates that differences in particle size and ligand density, often factors providing only moderate changes in resolution, can be identified with appropriate high-throughput 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.

References1. 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).


High-throughput strategies for the development of viral vaccine processesM. Wenger, S. Christanti, C. Daniels, K. Huff, E. Lim, J. Olson, J. Rodriguez, and S.-c. WangVaccine Bioprocess R&D, Merck Research Laboratories, Merck & Co., Inc., Kenilworth, NJ USA


IntroductionProcess 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 biosafety-level-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 plate- and 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 HTPDSpecifically, 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.


Highest throughput(screening models)

Integrated robotic systems

(plate-based, microliter scale)

Microbioreactors(< 15 mL scale; e.g., ambr™ 15,

BioLector™, Micro-24)

ambr 250(250-mL automated

minibioreactors)

SelecT(automated T-�ask operation

for adherent cell lines)

Spin-tube systems(tube-based, shake-�ask

surrogate)

Most representative(scale-down models)

Thro

ughp

ut

Representativeness

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.

Fig 1. Roadmap of HTPD technologies for upstream vaccine development.


Today(in progress)

Chromatography

Membrane chromatography

Adherent cell culture (SelecT)

Medium development/toxicity

Minibioreactor

ATPS/precipitation

Clonal isolation

Advancedinformatics

Exressionsystems

Conjugation chemistry screening

Pip

elin

e im

pac

tImplementation

Fig 2. HTPD 2020 for vaccines: pathway to pipeline impact.

HPTD 2020Looking 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.

References1. 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).


Product understanding

Processunderstanding

Facilityunderstanding

• Quality target product pro�le• CQAs

• Process parameters, e.g. DoE• Process models, e.g. simulation• Design space

• Control strategy• PAT etc.

Risk assessment

Risk assessment

Requirements

Requirements

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. StabyNovo Nordisk, Novo Alle, 2880 Bagsværd, Denmark


IntroductionSpeed 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.

Fig 1. Factors governing QbD implementation.


Theoretical knowledge

Simple statistical models/DoE

Advanced statistical models/PCA etc.

Mechanistic models

OPAT(One parameter at a time)

Preliminary experimentation

Experience from other projects

Exte

nt o

f kno

wle

dge

DiscussionA 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.

Fig 2. Process understanding elements and level of process knowledge obtained.

References1. 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. Strøm-Hansen, and A. Staby. Oral plenary lecture at: 238th ACS National Meeting, Washington DC, August 16–20 (2009).


BioLector™ Pro – Expansion of microbioreactor platform for strain screening under full bioprocess controlA. Grimm, N. Frische, T. Olfers, and F. Kensym2p-labs GmbH, Baesweiler, Germany


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.


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).

References1. 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).


High-throughput microscale platform to accelerate the development of particle conditioning for biologicsA. Noyes1,2, B. Huffman3, A. Berrill3, N. Merchant4, R. Godavarti1, N. J. Titchener-Hooker2, K. Sunasara3, and T. Mukhopadhyay2

1Pfizer 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

3Pfizer Bioprocess R&D, 700 Chesterfield Parkway, Chesterfield, MO 63017, USA4Pfizer Vaccine Research Unit, 401 N. Middletown Road, Pearl River, NY 10965, USA


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.

IntroductionParticle 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 systemA 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.


Mounting platform

Heater block

Copper shot

Multichannel liquid dispensing

Magnetictumblestirrer

Stir bar

Squaremicrowell

Supernatant:≤ 300 µL/well

CPS titerHb Ppt/UPLC-MALS

130 µL/well

Particle characterizationMFI

10 µL/well

DNAPICOGREEN™

10 µL/well

Turbidity OD 405/600300 µL/well

Centrifugation/vacuum

Particle conditioning platformTimeFlocculation: 1 dAnalytics: 1 dAnalysis: 1 d

Load material:≤ 600 µL/well

ProteinCoomassie™ 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.

(A)

(B)


6.0

5.8

5.6

pH

Floc agent concentration (%)

SA1 DNA

≤ 1%

≤ 5%

≤ 9%

≤ 13%

≤ 17%

≤ 21%

> 21%

1.0 1.5 2.0 2.5 3.0 3.5 4.0

5.4

5.2

5.0

ApplicationsFour 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.

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.

(A) (B)

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 (θ

0.95), the lowest rpm required to maintain particle suspension (N

js), 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 (v

tip) and/or power per volume in the impeller zone (P/V

imp)

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/V

imp and v

tip 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/V

imp t, v

tip t , or a Camp number (Ca) permutation substituting P/V

imp for P/V

avg represent the

most appropriate scaling terms.

Floc agent concentration2.0% 2.5% 3.0%

3.5 L pH 5.8 16.8% 1.2%

pH 5.5 1.2%

pH 5.2 1.3% 0.9%

1.2 mL pH 5.8 7.1% 1.0%

pH 5.5 1.1%

pH 5.2 1.7% 0.8%


ii

iii

iva

Design space

Floc agent (%)1.0

3.2

3.4

3.6

3.8

4.0

2.52.01.5 3.0

pH

ivb

i

Fig 3. Scaling by θ0.95

, Njs, P/V

avg, Ca, or ν

tip, P/V

imp.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.

(A) (B)

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.

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 A

600 (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.

0

1

2

3

4

5

6

7

8

9

10

1% 2% 3%

Num

ber-

aver

age

part

icle

dia

met

er (µ

m)

Floc agent concentration

Pilot-scale reference

USD; Scale by P/Vimp, vtip

USD, Scaling by Njs, θ0.95

Scale2350 mL

1 mL

1 mL

Pilot-scale reference

USD; Scale by P/Vimp, vtip

USD, Scaling by Njs, θ0.95

Scale2350 mL

1 mL

1 mL

0

5.0 × 105

1.0 × 106

1.5 × 106

2.0 × 106

2.5 × 106

1% 2% 3%

Part

icle

cou

nt (p

artic

les/

mL)

Floc agent concentration


ConclusionsThe 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/V

imp 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.

References1. 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).


Understanding the chromatography behavior of monoclonal antibodies using quantitative structure-property relationship analysisB. Tran, B. Connolly, T. Patapoff, and P. McDonaldGenentech


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 K

p data reported as a log K

p 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 K

p 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 K

p values. The residuals did, however, increase with increasing log K

p. 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.


Residuals vs predicted log Kp

-1

0

1

Experimental log Kp vs predicted log Kp

Expe

rimen

tal l

og K

p

Resi

dual

s

-2

-1

0

1

2

3

4

-2 -1 0 1 2 3 4Predicted log Kp

-2 -1 0 1 2 3 4Predicted log Kp

MAb 1MAb 2MAb 3MAb 4

MAb 5MAb 6MAb 7MAb 8

Log Kp

Low binding

Intermediate binding

High binding

7.0

7.5

8.0

8.5

pH

0 40 80 120Salt conc. (mM)

7.0

7.5

8.0

8.5

pH

0 40 80 120Salt conc. (mM)

7.0

7.5

8.0

8.5

pH

0 40 80 120Salt conc. (mM)

7.0

7.5

8.0

8.5

pH

0 40 80 120Salt conc. (mM)

7.0

7.5

8.0

8.5

pH

0 40 80 120Salt conc. (mM)

7.0

7.5

8.0

8.5

pH

0 40 80 120Salt conc. (mM)

7.0

7.5

8.0

8.5

pH

0 40 80 120Salt conc. (mM)

7.0

7.5

8.0

8.5

pH

0 40 80 120Salt conc. (mM)

Experimental Predicted

Fig 1. Comparison of experimentally determined log Kp values vs predicted log K

p values for eight MAbs. Predicted

values were determined from a multivariate partial least squares regression analysis.

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.


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 fragmentV. 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 characterizationC. Williams†, J. Winderl, T. Hahn, and T. HuukGenentech Inc., Karlsruhe Institute of Technology, Germany

3. Mini-columns as a complement to 96-well filter plates in a process development workflowA. Grönberg†, E. Brekkan, A. Edman Örlefors, and K. Nilsson-VälimaaGE Healthcare Life Sciences, Björkgatan 30, 751 82 Uppsala, Sweden

4. Buffer plate preparation using AKTA™ avant and BufferProS. Westin, G. Rodrigo†, and E. CarredanoGE 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 ImpResL. Kärf, E. Brekkan†, K. Nilsson-Välimaa, and M. AhnfeltGE 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 modelingN. Patel†1,2, E. Sorensen2, and D. G. Bracewell11 Advanced Centre for Biochemical Engineering, Department of Biochemical Engineering, UCL, London, UK2 Centre for Process Systems Engineering, Department of Chemical Engineering, UCL, London, UK

7. Microfluidics on liquid handling stations (μF-on-LHS): closing the gap between high-throughput experimentation and microfluidicsC. P. Radtke†1, J. Kittelmann1, A. Waldbaur2, B. E. Rapp2, and J. Hubbuch1

1 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

8. A high-throughput based approach to assess resin aging – an automated life time study on protein A column materialD. Boeth†, C. Atzkern, H. Geier, and H. RoglBoehringer Ingelheim Pharma GmbH & Co. K, Germany

9. High-throughput process development: Quality by Design studies across multiple formatsJ. Feliciano1, M. Ahnfelt2, A. Berrill3, E. Brekkan2, B. Evans3, Z. Fung1, R. Godavarti4, K. Nilsson-Välimaa2, J. Salm4, M. Switzer4, and K. M. Łacki21 GE Healthcare Life Sciences Fast Trak Services, 800 Centennial Ave., Piscataway, NJ 08854, USA2 GE Healthcare Life Sciences, Björkgatan 30, 751 82 Uppsala, Sweden3 Pfizer R&D Global Biologics, 700 Chesterfield Parkway, Chesterfield, MO 63017, USA4 Pfizer R&D Global Biologics, 1 Burtt Road, Andover, MA 01810, USA

10. High-throughput development of non-protein A monoclonal antibody purification process using minicolumns and bio-layer interferometryV. Ravault† and V. BrochierPall, 48 avenue des Genottes, 95800 Cergy Saint Christophe, France


11. Automated small-scale protein purification for accelerated development of protein therapeuticsN. Verlinden†1, L. Jordan1, R. Beighley1, K. McGuire1, B. Gerwe1, J. Lambert1, M. Costioli2, C. Kaltenbach2, and X. LeSaout2

1 PerkinElmer, Inc., 940 Winter Street, Waltham, MA USA2 Merck KGaA, Darmstadt, Germany

12. High-throughput solution screening documented and evaluated with an electronic systemR. Nachtigall†, G. Pollinger, A. Kleinjans, and F. ZettlRoche Diagnostics GmbH, Penzberg, Germany

13. Influence of chromatography media sampling in plate-based benchmarking studiesT. Bergander† and K. M. ŁackiGE Healthcare Life Sciences, Björkgatan 30, 751 72 Uppsala, Sweden

14. Tuning evolutionary multiobjective optimization to estimate chromatographic operating conditionsR. Allmendinger†, S. Gerontas, N. J. Titchener-Hooker, and S. S. FaridDepartment of Biochemical Engineering, University College London, London, Torrington Place, London WC1E 7JE, UK

15. RoboChrom: a highly flexible chromatography screening toolD. Harde† and M. BerkemeyerBoehringer-Ingelheim RCV Vienna, Austria

16. Purification of interferon α-2a – a process development studyS. Grönlund†, K. Eriksson, J. Shanagar, C. Brink, E. Pool, A. Moberg, M. Winkvist, and A. GrönbergGE Healthcare Life Sciences, Björkgatan 30, 751 82 Uppsala, Sweden

17. Utilization of high-throughput screening in the development of HyCell™ CHO production mediumE. Garner†, A. Elwood, J. Manwaring, and M. WightGE Healthcare Life Sciences, HyClone™ Cell Culture, Logan, Utah, USA

18. BioLector™ Pro - Expansion of microbioreactor platform for strain screening under full bioprocess controlA. Grimm†, N. Frische, T. Olfers, and F. Kensym2p-labs GmbH, Baesweiler, Germany

19. Self-optimization of robotic liquid classesF. Nossek†, K. Ehrhard, S. Werz, S. Hepbildikler, and C. BellRocwhe Diagnostics GmbH Dovnstream Development, Germany

20. SHARC – software integration for high-throughput systemsK. Doninger†, A. Kurtenbach, J. Griesbach, S. Markert, C. Musmann, S. Fan, H. Nibbrig, and A. JuxRoche Diagnostics GmbH, Penzberg, Germany


Author index

Ahnfelt, M. 40 Allmendinger, R. 41 Atzkern, C. 40

Bao, H. 23 Barlow, K. 23 Baumann, P. 13 Beighley, R. 41 Bell, C. 41 Bergander, T. 6, 41 Berkemeyer, M. 19, 41 Berrill, A. 33, 40 Boeth, D. 40 Bracewell, D. G. 40 Brekkan, E. 10, 23, 40 Brink, C. 41 Brochier, V. 40

Carredano, E. 40 Christanti, S. 26 Connolly, B. 38 Costioli, M. 41

Daniels, C. 26 Doninger, K. 41 Dürauer, A. 19

Edman Örlefors, A. 10, 40 Ehrhard, K. 41 Elwood, A. 41 Eppink, M. H. M. 16 Eriksson, K. 41 Evans, B. 40

Fan, S. 41 Farid, S. S. 41 Feliciano, J. 40 Frische, N. 31, 41 Fung, Z. 40

Garner, E. 41 Geier, H. 40 Gerontas, S. 41 Gerwe, B. 41 Godavarti, R. 33, 40 Griesbach, J. 41 Grimm, A. 31, 41 Grönberg, A. 10, 40, 41 Grönlund, S. 41

Hahn, T. 13, 40 Hanke, A. T. 16 Hansen, M. H. 29 Hansen, T. B. 29 Harde, D. 41 Hepbildikler, S. 41 Hubbuch, J. 13, 40 Huff, K. 26 Huffman, B. 33 Huuk, T. 13, 40

Jordan, L. 41 Jux, A. 41

Kaltenbach, C. 41 Kensy, F. 31, 41 Kidal, S. 29 Kittelmann, J. 40 Kleinjans, A. 41 Kurtenbach, A. 41 Kärf, L. 40

Łacki, K. M. 6, 23, 40, 41 Lambert, J. 41 LeSaout, X. 41 Lim, E. 26 Linden, T. 23

Manwaring, J. 41 Markert, S. 41 McDonald, P. 38 McGuire, K. 41 Merchant, N. 33 Moberg, A. 41 Mukhopadhyay, T. 33 Musmann, C. 41

Nachtigall, R. 41 Nibbrig, H. 41 Nilsson Välimaa, K. 10, 40 Nossek, F. 41 Noyes, A. 33

Olfers, T. 31, 41 Olsen, A. A. 29 Olson, J. 26 Ottens, M. 16

Patapoff, T. 38 Patel, N. 40 Pollinger, G. 41 Pool, E. 41

Radtke, C. P. 40 Rapp, B. E. 40 Ravault, V. 40 Rodrigo, G. 40 Rodriguez, J. 26 Rogl, H. 40 Roush, D. 23

Salm, J. 40 Sejergaard, L. 29 Shanagar, J. 41 Sorensen, E. 40 Staby, A. 29 Strøm-Hansen, T. 29 Sunasara, K. 33 Switzer, M. 40

Titchener-Hooker, N. J. 33, 41 Tran, B. 38

Valentin, S. 29 van de Sandt, E. J. A. X. 16 van der Wielen, L. A. M. 16 Verhaert, P. D. E. M. 16 Verlinden, N. 41

Waldbaur, A. 40 Walther, C. 19 Wang, S.-c. 26 Welsh, J. 23 Wenger, M. 26 Werz, S. 41 Westi, S. 40 Wight, M. 41 Williams, C. 40 Winder, J. 40 Winkvist, M. 41

Zettl, F. 41

Østergaard, P. 29 Østfeldt, P. 29

GE Healthcare Bio-Sciences AB Björkgatan 30 751 84 Uppsala Sweden

www.gelifesciences.comGE, 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. 2016All 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, UKGE Healthcare Europe GmbH, Munzinger Strasse 5, D-79111 Freiburg, GermanyGE Healthcare Bio-Sciences Corp., 100 Results Way, Marlborough, MA 01752, USAGE Healthcare Dharmacon Inc., 2650 Crescent Dr, Lafayette, CO 80026, USAHyClone Laboratories Inc., 925 W 1800 S, Logan, UT 84321, USAGE Healthcare Japan Corp., Sanken Bldg., 3-25-1, Hyakunincho Shinjuku-ku, Tokyo 169-0073, JapanFor local office contact information, visit www.gelifesciences.com/contact

29179765 AA 02/2016

Documents

Extended reports and posters from the ... - HTPD Meetings