1
1
Developing an Appropriate Design Developing an Appropriate Design Space Strategy to Mitigate Variability in Space Strategy to Mitigate Variability in
Downstream Processing OperationsDownstream Processing Operations
Justin McCue
Biogen Idec Corporation
September, 2010
2
OverviewOverview
• Chromatography column scale up approach – Scale up considerations and challenges
• Chromatography Adsorbent Lot Variability– Two case studies
• HIC- Monomer/Aggregate Separation • Anion Exchange-Product-related Impurity Separation
• Design Space Approach – Ways to control adsorbent lot variability through QBD
• Conclusions and acknowledgements
3
3
General Chromatography Column Scale Up ApproachGeneral Chromatography Column Scale Up Approach
• Bed height constant during scale up• Scale up by increasing column diameter and volumetric flow rate
– Development scale to cGMP scale: ~10,000x Volumetric Scale up factor
100x
Scale Up
Development Scale Pilot Scale MFG Scale
1 cm column diameter
10 cm column diameter
100 cm column diameter
100x
Scale up
Bed Height held constant
4
4
Scale Up Challenges with Adsorbent Lot VariabilityScale Up Challenges with Adsorbent Lot Variability
• Different lots may need to be used across scales– Multiple lots/lot mixtures may be required to
pack a large scale column
• If adsorbent lot variability exists, risk that column performance/product quality could change during scale up
5
5
Case StudiesCase Studies
• Case Study #1– HIC (Phenyl Sepharose FF) adsorbent lot variability
• Differences in yield and aggregate removal using different adsorbent lots
• Case Study #2– Anion Exchange adsorbent lot variability
• Differences in yield and removal of process-related impurity species during column wash step
6
6
Case Study #1Case Study #1
HIC Adsorbent Lot variabilityHIC Adsorbent Lot variability
7
Motivation for StudyMotivation for Study– HIC commonly used to separate monomer and aggregate
species for protein therapeutics (monoclonal antibodies, fusion proteins)
– HIC adsorbent (Phenyl Sepharose FF) – must reduce aggregate levels from 10−20% to < 1%
• Difficult separation – Column yields limited to <70%– Superior adsorbent for this particular separation
• Separation sensitive to column residence time (operating velocity, bed height) and column loading
Column performance sensitive to adsorbent lot (ligand density)
8
• Similar elution profile for lots having lower ligand densities– Ligand density of 40 and 42 μmole/mL Similar elution peak profiles
• Smaller elution peaks when adsorbent ligand density was increased to 45−47 μmole/mL– Suggests product yield (and maybe purity) could be different when
using adsorbent lots containing higher ligand densities
Experimental Finding: Lot to lot Performance Experimental Finding: Lot to lot Performance VariabilityVariability
Development Data: Column Elution Profiles for Different adsorbent Lots
0
200
400
600
800
1000
0 2 4 6 8 10
Elution Column Volumes
UV
Abs
orba
nce
40
42
45
47
adsorbent Ligand Density (μmole/mL)
0
200
400
600
800
1000
0 2 4 6 8 10
Elution Column Volumes
UV
Abs
orba
nce
42
45
47
adsorbent Ligand Density (μmole/mL)
Isocratic Elution Condition #1 Isocratic Elution Condition #2
9
30
35
40
45
50
55
60
65
70
75
39 40 41 42 43 44 45 46 47 48
Ligand Density (umole/mL resin)
Col
umn
Yiel
d (%
)
Similar Yields
• Downward trend of yield with higher ligand densities– Lot to lot variability
Use modeling insights to evaluate impact of adsorbent lot variability prior to scale up
Potential Scale Up Challenge: Lot to lot Potential Scale Up Challenge: Lot to lot Performance VariabilityPerformance Variability
Lower Yields
Development Data: Column yields for Different adsorbent Lots (Isocratic elution conditions)
10
• Formulate mechanistic model to predict Monomer/Aggregate separation over range of column operating conditions– Determine governing separation performance
parameters– Predictive tool
• Understand mechanism responsible for lot-to-lot variability in performance – Apply model to evaluate acceptable operating conditions
prior to scale up– Developing a “use test” to aid in adsorbent screening
and potential design space prior to scale up
Modeling to Assist in Process ScaleModeling to Assist in Process Scale--Up Up
11
Adsorption of Monomer/Aggregate SpeciesAdsorption of Monomer/Aggregate Species
• Aggregate species binds irreversibly to the HIC adsorbent– Irreversible term included in the Adsorption Isotherm Model
Phenyl Sepharose Adsorbent Surface (Solid Phase)
k1,ad
Monomer Species (C1)
k-1,de
Aggregate Species (C2)
k2,ad k-2,de
k2,irrev
Irreversibly bound
aggregate
Q1 Q2 Q2, Irr
C1 C2
Competitive Langmuir Isotherm model with irreversible binding of aggregate species1
1J.T. McCue, et al., Bioprocess and Biosystems Engineering, 31 (2008) 261.
12
0
10
20
30
40
50
60
70
80
90
100
39 40 41 42 43 44 45 46 47 48Ligand Density (μmole/mL resin)
Col
umn
Yiel
d (%
)
0
1
2
3
4
5
6
7
Bin
ding
Con
stan
t (K
)
• Able to distinguish lots containing different ligand densities using an alternative test (Binding Constant from Adsorption isotherms)
Similar Yields (“Acceptable Lots”)
Similar Binding Constants
Lower Yields (“Unacceptable Lots”)
Higher Binding Constants
Correlation between Binding constant and Column Performance
K ~ 2Acceptable (K < 2) (similar yields)
Not Acceptable (K > 2) (low yields)
13
• Can we use the model to determine acceptable operating conditions if the adsorbent lot or lot mixture is outside the “acceptable” range?
Further Model Applications: What happens if we are not able to “Cherry Pick” adsorbent lots?
14
Effect of Higher adsorbent Ligand Density: Experimental DataEffect of Higher adsorbent Ligand Density: Experimental Data
• Using adsorbent lots with lower ligand density levels will increase aggregates above acceptable levels (e.g. 0.35 M Ammonium Sulfate, < 2.0% Aggregate)
Feed: 20 % aggregate
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0.30 0.35 0.40 0.45 0.50Ammonium Sulfate Eluate Buffer Concentration (mM)
Agg
greg
ate
(%)
40
42
45
47
LIgand Density (μmole/mL)
Aggregates too high!
15
• Change of operating conditions will be required (increase Ammonium Sulfate concentration from 0.35 to 0.45 M)
Feed: 20 % aggregate
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0.30 0.35 0.40 0.45 0.50Ammonium Sulfate Eluate Buffer Concentration (mM)
Agg
greg
ate
(%)
40
42
45
47
LIgand Density (μmole/mL)
Effect of Higher adsorbent Ligand Density: Experimental DataEffect of Higher adsorbent Ligand Density: Experimental Data
16
0.25
0.30
0.35
0.40
0.45
0.50
35 37 39 41 43 45 47 49
ligand density [umol \ mL]
amm
omiu
m s
ulfa
te [M
]aggragte [%] yield [%] 1.6 46
2.0 50
2.5 58
3.0 62 4.0 70
• Model used to predict acceptable operating conditions• For a ligand density of 47 μmole/mL, aggregate levels will be 2.0% and
product yields 50% if Ammonium sulfate concentration in the elution buffer is 0.29 M
Modeling Predictions to Guide Scale UpModeling Predictions to Guide Scale Up
Feed: 20% Aggregate
17
0.25
0.30
0.35
0.40
0.45
0.50
35 37 39 41 43 45 47 49
ligand density [umol \ mL]
amm
omiu
m s
ulfa
te [M
]aggragte [%] yield [%] 1.6 46
2.0 50
2.5 58
3.0 62 4.0 70
• Similar product purity and yield achievable for lower ligand density lots (42 μmole/mL) if AS concentration increased from 0.29 M to 0.41 M AS
Modeling Predictions to Guide Scale UpModeling Predictions to Guide Scale Up
Feed: 20% Aggregate
18
• Mechanistic model provides additional insight on separation performance prior to scale up – Determine most sensitive (governing) input
parameters – Useful for predicting performance of the unit
operation outside the range explored during development
– Developed a protein-specific use test to assist in screening adsorbent lots
• Model applied to predict acceptable conditions for different adsorbent ligand density levels– Useful if selection of adsorbent lots with specific
ligand densities is not possible
Mechanistic Modeling to Guide Process Scale up for Adsorbent Variability
19
Model Case Study #2Model Case Study #2
Anion Exchange Adsorbent Lot variabilityAnion Exchange Adsorbent Lot variability
20
Background for Case Study #2Background for Case Study #2
• Anion exchange column performed in the “bind/elute” mode• Main function of column
– Removal of a product-related impurity species in wash step (prior to product elution)
• Desirable to have 5-10% product loss in wash to ensure the product-related impurity is effectively removed
• Difficult separation due to similarities between the product-related impurities and target product – Column performance sensitive to changes in column
loading and wash buffer composition (pH, Osmolality)• Significant adsorbent lot variability
21
Adsorbent Lot Variability During Wash StepAdsorbent Lot Variability During Wash Step
• Increase in protein loss during wash step for adsorbents lots containing higher Ionic Capacity – Some resin lots which did not follow the trend
Protein Loss During Wash Step for Different Adsorbent Lots1
1Cecchini, D.; Quality by Design for Biopharmaceuticals, 2009, P. 140
22
Adsorbent Lot Variability During Wash StepAdsorbent Lot Variability During Wash Step
• Increase in protein loss during wash step for adsorbents lots containing higher Ionic Capacity – Some resin lots which did not follow the trend
• May require an additional correlation beyond one provided on the COA or by the Vendor
Protein Loss During Wash Step for Different Adsorbent Lots1
1Cecchini, D.; Quality by Design for Biopharmaceuticals, 2009, P. 140
23
• Lot to lot variability measured during protein adsorption isotherm experiments using the wash buffer conditions
• Measure the binding constant (KL) for each of the adsorbent lots
Adsorption Isotherms for Different Adsorbent LotsAdsorption Isotherms for Different Adsorbent Lots
0
5
10
15
20
25
30
0.0 0.5 1.0 1.5 2.0CF (mg/mL)
q (m
g pr
otei
n/m
L re
sin)
99 μeq/mL
153 μeq/mL
150 μeq/mL
131 μeq/mL
Adsorbent Ionic Capacity
Liquid Phase: Column Wash Buffer
q = KL*C
24
• Correlation between binding constant and column performance– Process-specific use test can be used to detect differences in adsorbent
lot performance
ProcessProcess--Specific Adsorbent Use TestSpecific Adsorbent Use Test
y = -2.0007x + 39.525R2 = 0.9376
0
2
4
6
8
10
12
14
16
10 12 14 16 18 20
Binding Constant (KL)
Prot
ein
Loss
dur
ing
Was
h (%
)
Correlation between Protein Loss (in Wash step) and Binding Constant for different Adsorbent Lots
25
• How do we control differences in lot variability during GMP MFG?– Potential impacts on both process consistency and product quality
• Several Approaches Exist– Conventional Approach
• Select operating conditions/range in which all adsorbent lots will have acceptable performance
• May result in sub-optimal process performance – “Cherry pick” Approach
• Use only certain adsorbent lots in MFG processing • May or may not be possible• Not desirable
– Design Space/QBD Approach • Design space filing which includes ranges in column operating
conditions • Can control adsorbent lot variability through changes in the
column operating conditions
Manufacturing Process Control ConsiderationsManufacturing Process Control Considerations
26
• Change the wash buffer composition to achieve the appropriate level of product removal during the wash step
• Provides flexibility in managing different adsorbent lots– Potential impacts on both process consistency and product quality
Design Space Approach for Anion ExchangeDesign Space Approach for Anion ExchangeEffect of Wash Buffer pH and NaCl
Concentration on Product Loss in Wash
x
x
Adsorbent Lot “A”
27
0.25
0.30
0.35
0.40
0.45
0.50
35 37 39 41 43 45 47 49
ligand density [umol \ mL]
amm
omiu
m s
ulfa
te [M
]aggragte [%] yield [%] 1.6 46
2.0 50
2.5 58
3.0 62 4.0 70
• Change the Elution Buffer Ammonium Sulfate concentration to achieve consistent product quality and yield for different adsorbent lots
Feed: 20% Aggregate
Design Space Approach for HICDesign Space Approach for HIC
28
0.25
0.30
0.35
0.40
0.45
0.50
35 37 39 41 43 45 47 49
ligand density [umol \ mL]
amm
omiu
m s
ulfa
te [M
]aggragte [%] yield [%] 1.6 46
2.0 50
2.5 58
3.0 62 4.0 70
Feed: 20% Aggregate
• QBD filing can include ranges for elution buffer composition – Selection of the elution buffer composition can be based upon the
adsorbent lot binding constant• Provides additional process control to ensure consistent and optimal
process performance
Design Space Approach for HICDesign Space Approach for HIC
29
ConclusionsConclusions• Adsorbent lot variability should be evaluated, especially when
performing difficult separations – Illustrated with two case studies: HIC and Anion Exchange
columns– Formulated a mechanistic model useful as a predictive tool
• Adsorbent CoA information can be helpful, but a process/product specific use test may be required to correlate adsorbent lot variability with process performance– Measurement of adsorption isotherm binding constants-useful
to detect difference among adsorbent lots
• Flexibility in selecting elution or wash buffer composition to provide optimal process performance– Design Space/QBD Filing strategy could be used to implement
this approach in drug substance manufacturing
30
30
AcknowledgementsAcknowledgements
• Philip Engel• Austen Ng• Rich Macniven• Jason Dube
31
31
ReferencesReferences
• Phenyl Sepharose Modeling and Adsorbent Lot variability across scales– J.T. McCue, P. Engel, A. Ng, R. Macniven, J. Thömmes,
“Modeling of Protein Monomer/Aggregate Purification and Separation Using Hydrophobic Interaction Chromatography,”Bioprocess and Biosystems Engineering, 31 (2008) 261.
– J.T. McCue, P. Engel, J. Thömmes, “Effect of Phenyl Sepharose Ligand Density on Protein Monomer/Aggregate Purification and Separation Using Hydrophobic Interaction Chromatography,”Journal of Chromatography A, 1216 (2009) 902.
– J.T. McCue, P. Engel, J. Thömmes, “Modeling the Effects of Column Packing Quality and Residence Time Changes on Protein Monomer/Aggregate Separation,” Journal of Chromatography A, 1216 (2009) 4895.