continuous improvement in biologics ...defining pat process analytical technology is: a system for...
TRANSCRIPT
CONTINUOUS IMPROVEMENT IN BIOLOGICS
MANUFACTURINGAn Opportunity for PAT
Charles L. CooneyDepartment of Chemical
EngineeringMIT
Cambridge, MAFebruary 20, 2007
Points for Discussion
• Drivers of Change in Pharmaceutical Manufacturing
• Managing Uncertainty and Risk• PAT and QbD• Process modeling• Miniaturization and process understanding• Genomic arrays and process design
WHERE IS MANUFACTURING GOING?
• Insight from the Clinic– Increasing molecular complexity– Larger pipelines, e.g. no. of molecules– Higher late stage rejection of candidate drugs– Personalized medicine defining smaller markets– Products with multiple API’s
• Insight from the Regulators– Increased focus on risk– Increased interest in understanding process science– Reduced tolerance to uncertainty– Focus on consistent quality– Increased regulatory burden from new products and
amended applications
WHERE IS MANUFACTURING GOING? (continued)
• Insight from the Market– Increasing number of dosage forms– Increasing diversity of package inserts (global
Markets)– Novel delivery formats
• Insight from Discovery– Increased potency and decreased dose– Use of molecular diagnostics in prescribing therapy– Large number of monoclonal antibody and vaccine
products
Process Design and Operation
We need to know where we want to go (critical metrics for quality), have the analytics to measure where we are (relative to our product specifications) and understand how the process affects the important properties of the product.
“There is uncertainty and thus risk”
UNCERTAINTY
• Uncertainty is the foundation of risk• Risk is the probability of an event and its
impact if it occurs?• We need to assess and reduce uncertainty
to an acceptable level to manage risk• Need to know where we want to go and
have a measure of where we are
WHAT ARE THE PROBLEMS?• We look at what we can measure and not
necessarily what is important• Our understanding of the underlying science is
often inadequate, creating uncertainty • While the biology of the drug is critical, we often
do not know how this relates to molecular or process complexity
• Formulation and storage add an additional layer of complexity to complex APIs
• There are organizational and intellectual barriers between therapeutic evaluation and process evaluation
DEFINING PATProcess Analytical Technology is:a system for designing, analyzing, and controlling manufacturing through timely measurements (i.e., during processing) of critical quality and performance attributes of raw and in-process materials and processes with the goal of ensuring final product quality.
http://www.fda.gov/cder/OPS/PAT.htm.
Need to know what to measure
Need the analytical tools
Do we know where we want to go?
Process Modeling and Simulation with Mapping of
Uncertainty
A PAT Tool
Process Flow for a MABInoculum Prep
P-1 / TFR-101
T-Flask (225 mL)
P-2 / RBR-101
Roller Bottle (2.2 L)
P-5 / SBR1
First Seed Bioreactor (1000 L)
P-6 / V-102
Media Prep
P-7 / DE-101
Sterile Filtration
P-11 / V-104
Media Prep
P-12 / DE-102
Sterile Filtration
P-10 / SBR2
Second Seed Bioreactor (5000 L)
P-20 / PBR1
Production Bioreactor (20000 L)
P-21 / V-106
Media Prep
P-22 / DE-103
Sterile Filtration
S-101
S-102
S-103
S-111
S-112 S-113
S-115
S-114
S-119
S-130S-121
S-122S-123
S-125
S-124
S-129
S-149
S-140
S-141 S-142
S-144
S-143
S-120
P-30 / V-101
Surge Tank
S-148
P-33 / V-103
Storage
P-34 / UF-101
Concentration
S-155
S-156
S-157
P-35 / V-105
Virus InactivationP-36 / DE-104
Polishing FIlter
P-37 / V-107
Storage
S-158
S-159
P-40 / C-101
Prot-A Chromatography
P-42 / V-108
Prot-A Pool Tank
P-38 / DF-102
Diafiltration
S-160
S-162
PrA-Equil
PrA-Wash
PrA-Eluat
PrA-Reg
S-181
S-164
S-163
S-161
S-165
S-169
P-41 / DE-105
Polishing FIlter
S-180
S-183
S-182
P-50 / C-102
IEX Chromatography
S-184
P-52 / V-109
IEX Pool Tank
IEX-Equil
IEX-Wash
IEX-WFI
IEX-Eluat
IEX-Strip
S-190
S-194
P-60 / C-103
HIC Chromatography
P-62 / DE-106
Dead-End Filtration
P-70 / V-110
HIC Pool Tank P-71 / DF-103
Diafiltration
P-72 / DE-107
Final Polishing Filtration
P-73 / DCS-101
Freeze in Plastic Bags
S-195
HIC-Equil
HIC-Wash
HIC-Eluat
HIC-Reg
S-200
S-205
S-204
S-210
S-211S-212
S-213
S-215
S-217
S-216
Final Product
IEX-Rinse
P-31 / DS-101
Centrifugation
P-32 / DE-108
Polishing Fitler
S-150
S-152
S-151
S-154
S-153
P-3 / BBS-101
Bag Bioreactor (20 L)
S-104
S-105
S-107
Bioreaction
Primary Recovery
Protein-A
IEX Chrom HIC Chrom
Final Filtration
P-4 / BBS-102
Bag Bioreactor (100 L)
S-106
S-108 S-110
S-109
S-214
P-39 / MX-101
Mixing
S-166
S-167
S-168
P-51 / MX-102
MixingP-61 / MX-103
Mixing
S-191
S-192
S-193
S-201
S-202
S-203
P-9 / AF-101
Air Filtration
P-8 / G-101
Gas Compression
S-116
S-117
S-118
P-13 / G-102
Gas Compression
P-14 / AF-102
Air Filtration
S-126
S-127
S-128
P-24 / AF-103
Air Filtration
P-23 / G-103
Gas Compression
S-145
S-146
S-147
Allocation of operating costs for monoclonal antibody production across different process sections
Operating Cost
0 2 4 6 8 10 12 14 16 18
Inoculum preparation
Bioreaction
Primary recovery
Protein A chrom
Ion exchange chrom.
Hydrophobic interaction chrom.
Final filtration
$ million/year
UPC for the MAb process model at different MAb concentrations
0
50
100
150
200
250
0 0.5 1 1.5 2 2.5 3 3.5Final Product Concentration (g/l)
UPC
($/g
)
What is the effect of “normal” variation in process performance on the desired outcome?
Allocation of Cost to UPC at Varying MAb Concentrations
0
25
50
75
100
125
150
0 0.5 1 1.5 2 2.5 3 3.5
Final Product Concentration (g/l)
($/g
)
Raw MaterialsFacilityLaborConsumables
Uncertainty Analysis
Process Model (SuperPro Designer)
Monte Carlo SimulationsScenarios Sensitivity
Analyses
Uncertainty Analysis
Process Data, Literature, Estimates
Economic Assessment
Monte Carlo Simulation
Return on Investment
Environmental Indices
Unit Production Cost
Market parameters e.g. product selling price
Technical parameterse.g. product
concentration
Supply chain parameters
e.g. media price
Uncertain variables: Objective functions:
P-1 / V-1Blending / Storage Me P-4 / ST-1
Heat Sterilizat
P-2 / V-1Blending / Storage Glu
P-3 / MX-1Mixing
P-5 / G-1Gas Compres
P-6 / AF-1Air Filtratio
P-7 / V-1Fermentati
P-8 / AF-1Air Filtratio
P-20 / RVF-Removal Biom
P-21 / HX-1Coolin
P-22 / MX-1Acidificatio
P-23 / CX-1Centrifugal Extrac
P-25 / V-1Re-ectraction + Crystall
P-26 / BCF-Basket Centrifuga
P-29 / MX-1Adding Fresh Butyl Ac
P-31 / FBDR-Fluid Bed Dry
P-32 / V-1Storage Penicillin Sodium
S-10
S-10
S-10
S-10
S-10
S-10
S-10
S-10
S-11 S-11
S-11
S-11
S-11
S-15 S-15
S-15
S-15
S-15
S-15
S-15
S-16
S-16
S-16
S-16
S-16
S-16S-16
S-17
S-17
S-17
S-17
S-17
S-17
S-10
S-11
S-11 S-11
P-27 / CSP-Component Split
S-16
S-17
S-15P-9 / V-1Storag
S-11S-11
P-24 / MX-1Neutralizati
S-15
S-15
S-16
P-28 / MX-1Neutralizati
S-16
S-17
S-17
Monte Carlo simulations
Probability Distribution: Input Variables
0.00000.00500.01000.01500.02000.0250
2.1 7.6 13.0 18.4 23.8
0.0000
0.0100
0.0200
0.0300
44.6 52.3 60.0 67.6 75.3
0.00000.00500.01000.01500.0200
1.5 1.9 2.3 2.7 3.1
Final product concentration:Normal distribution, Std.-Dev.: 10%
Agitator power:Normal distribution, Std.-Dev.: 20%, min: 1.5 kW/m3, max: 3.5 kW/m3
Price glucose:Beta distribution, α = 3.49; β = 1.2,Distribution type fits best actual data
Probability distribution of the UPC(10,000 trials)
0.00
0.01
0.02
0.03
0.04
0.05
1 21 41 61 81UPC ($/g MAb)
Prob
abili
ty
Probability distribution of the unit production costs
(10,000 trials, 100 groups in each graph, The peak of the MCS-
SCMP is at p = 0.4 )
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
50 100 150 200 250 300 350
UPC ($/g)
Prob
abili
ty
MCS-AP
MCS-TP
MCS-SCMP
MCS-FP
MCS-DSP
-75% -50% -25% 0% 25% 50% 75%
Final MAbConcentration
Product Yield HIC
Fermentation Time
Product Yield IXC
Product Yield Prot AChr.
Rplc. FrequencyProtein A
Technical Parameters
Contribution to the unit production cost variance
-75% -50% -25% 0% 25% 50% 75%
Media Powder Price
Unit Cost Protein AResin
Unit Cost IXC Resin
Unit Cost HIC Resin
SC parameters
-75% -50% -25% 0% 25% 50% 75%
Final MAb Concentration
Product Yield HIC
Fermentation Time
Product Yield Prot AChr.
Product Yield IXC
Rplc. Frequency ProteinA
All Parameters
P-11 / V-104
Media Prep
P-12 / DE-102
Sterile Filtration
P-10 / SBR2
Second Seed Bioreactor (5000 L)
P-20 / PBR1
Production Bioreactor (20000 L)
P-21 / V-106
Media Prep
P-22 / DE-103
Sterile Filtration
S-122S-123
S-125
S-124
S-129
S-149
S-140
S-141 S-142
S-144
S-143
Bioreaction
P-13 / G-102
Gas Compression
P-14 / AF-102
Air Filtration
S-126
S-127
S-128
P-24 / AF-103
Air Filtration
P-23 / G-103
Gas Compression
S-145
S-146
S-147
Measuring Where You AreSynthesisPost translational modificationProteolytic clipping
STRATEGIES FOR PROCESS DESIGN AND OPTIMIZATION
Drill Down in Understanding& Building Robustness
Drill Out for Optimization& Maximizing Performance
Microscale Analysis
Macroscale Analysis
22© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
Overcoming Limitations of Common Cell Culture Development Platforms
ThroughputExperimental capacity per
researcher
QualityAbility to predict manufacturing bioreactor performance
Benchtop bioreactor
Well plates
Shake flask
1’s
10’s
100’s
1000’s
Low High
High qualityHigh quantity
Add automationand software
Improve model data quality
Micro-Reactor Technology Enabling Large Scale Design of Experiments
Multiple Bioreactors
Working volume~ 300µl
Microfluidic channels for inoculation. feeds, pH adjustment & sampling
• Mammalian Cell Cultures with 5x107 Viable Cells/ml
• Simulates standard production modes: Batch, Fed Batch and Perfusion
• Enables full factorial DoE
Gas-permeable materials
Culture monitoring via optical interrogation
Source:BioProcessors, Inc. Woburn, MA USA
24© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
SimCell™ On-Line Measurements: Cell Density• Measured using Optical Density at 633 nm on Sensor
Station– Forward light scatter method
•OD calibration based on hand-counted cells (> 95% viable) and serial dilution of stock cell suspension (8x106
cells/ml)
•Each data point is average of 9 measurements (3 chambers x 3 OD readings/chamber)
•Error bars are +/- 1 std. dev. (+/- 16% variance)
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
0
1x106
2x106
3x106
4x106
5x106
6x106
7x106
8x106
9x106
1x107
biom
ass
OD633nm
OD calibration curve
MicroBioreactor Performance CHO-SEAP Batch Comparison
Cell Density Comparison
100,000
1,000,000
10,000,000
0 50 100 150
Hours
Log
cells
/ml
MBABRFL
Doubling Time Variability
0
5
10
15
20
MBA Flask Bioreactor
Vessel
Dou
blin
g Ti
me
(hou
rs)
N=22 N=10 N=3
• Growth in MBAs is similar to flasks and bench-top bioreactors.
• MBAs have low variability.
-125 -100 -75 -50 -25 0 25 50 75 100 125
123456789
101112131415161718192021222324252627282930
Spot
Num
ber
Average IntensityBR MBA
-125 -100 -75 -50 -25 0 25 50 75 100 125
123456789
101112131415161718192021222324252627282930
Spot
Num
ber
Average IntensityBR Flask
BioreactorMBA Flask
MBAs are more similar to bench-scale bioreactors
MBA vs Bioreactor
MicroBioreactor Performance CHO-SEAP : Gene Expression Profiles
Source:BioProcessors, Inc. Woburn, MA USAFLASK vs. Bioreactor
27© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
Observations• SimCell™ Microbioreactor Arrays are a highly representative
bioreactor scale-down model.– Flexible design and operation facilitates control of micro
bioreactor conditions and enables matching culture conditions in bioreactors.
– Automated online measurement of OD, pH, Oxygen and CO2
– Comparable results for:• Growth rate• Metabolism• Production
– Protein expression profiles show good similarities to bench-scale bioreactors
28© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
Automation Enables Very High Throughput
Incubation Modules
Loading Cell
Sampling Module
Optical Sensing Modu
Fluidic Module
Central Robot
Flexible, modular cluster design around central robot.
Layout used in semiconductor manufacture for >10 years
Platform can support 252 micro-bioreactor arrays in 6 incubators (1512 cultures).
Design of micro-bioreactor array and operation of system can be tailored to simulate common bioreactor modes
•Batch operation•Fed-batch operation•Perfusion
• Control of Shear Magnitude– Size of mobile secondary phase
– Chamber geometry
– Rotation rate
THE OXYGEN DILEMMA
•Required for efficient growth and recombinant protein expression
•Potential in vivo or in vitro protein oxidation e.g. Met, Cys
•Oxygen induced stress response
O2 Gradients in Large-Scale Fermentors
10,000 L
DO
10%
40%
100 mL
Homogeneous
10 L
Homogeneous
• How do O2 gradients affect cells ?
• How do cells respond?• Effects on recombinant
protein production?
100μl
Homogeneous
Heterogeneous
Model System: α1-Antitrypsin• Elastase inhibitor (44 kDa) 10 met and 1 unpaired cys• Activity lost with oxidation of active site MET358
Oxidation of met358 --> partial loss of neutrophil elastase activity & complete loss of porcine pancreatic elastase
• Use in protein replacement therapy• Cytoplasmic expression in E. coli BL21
methioninemethioninesulfoxide
H2N-C-HCOOH
CH2
CH2
CH3
S
H2N-C-HCOOH
CH2
CH2
CH3
S Ooxidation
M358
M351
0
0.2
0.4
0.6
0.8
1
1.2
0 10 20 30 40 50 60
N2AIRO2
Nor
mal
ized
a1A
T
Time (min)
0
0.2
0.4
0.6
0.8
1
1.2
0 10 20 30 40 50 60
N2AIRO2N
orm
aliz
ed a
1AT
Time (min)
Wild Type SG1146A (ClpP-)
Some background degradation (~18%) remains• Other proteases responsible (?)
O2-Enhanced Degradation is Eliminatedin ClpP- Strain
Do we have the analytical techniques to probe a cell’s
global response to its physical environment?
Can we locate where the problems are and then focus on solving the
right problem
DNA Microarray Experiments
• 3,812 Genes representing 89% of the E. coli genome
• Multi-Gene Groups– 167 protein complexes– 190 pathways– 333 transcription units
Selecting Differential Expression• List of Up-Regulated Genes Includes
– dnaJ– dnaK– ibpA– ibpB– htpG– topA
-10
-8
-6
-4
-2
0
2
4
6
-10 -8 -6 -4 -2 0 2 4 6Expression Values Pre-Induction
Expression Values60 min
FollowingInduction
GenesNotSelectedGenesSelected
DifferentialExpression
*Richmond et al. (1999) Nucleic Acids Res 27(19):3821. Lesley et al. (2002) Protein Eng 15(2):153. Rohlin et al. (2002) J Chin Inst Chem Eng 33(1):103.
• Heat-Shock (σ32) Genes Activated*
Differentially Expressed Genes
• Many Genes Affected by N2
• Focus on O2-Air Differences
050
100150200250300350400450500
0 30 60 90Post-Induction Time (min)
Numberof Genes
DifferentiallyExpressed
N2AIRO2
Hyperoxic Stress Responses
• Increasing N2 → Air → O2
• Sustained Response from SoxRS Regulon
• Increasing Air → O2
• Short-Term Response from OxyR Regulon
OxyR Regulon - ahpC /ahpF /dps /grxA /katG
-1.5-1
-0.50
0.51
1.52
0 30 60 90Post-Induction Time (min)
RelativeExpression(Average
Log Ratio)
O2AIR
SoxRS Regulon - fur /nfo /sodA /soxS
-5-4-3-2-101234
0 30 60 90Post-Induction Time (min)
RelativeExpression(Average
Log Ratio)
O2AIRN2
O2 Dependent Genes
• SoxRS Response– soxS, fur, sodA, nfo
• Iron Uptake– fur, sodA, fepB
• Fe-S Proteins– bioB, ilvD, leuB, mutY, fdx,
yfhI• Fe-S Cluster Assembly
– b2530 (iscS), b2531, hscA, fdx
Iron Supplementation• Improves Aerobic Growth of SOD Mutants in E.
coli* and Yeast‡
Effects of Iron – Alleviates α1AT
degradation
*Benov & Fridovich (1998) J Biol Chem 273:10313.‡De Frietas et al. (2000) J Biol Chem 275:11645.
Pulse-Chase Data - Wild-Type BL21 in O2
0
0.2
0.4
0.6
0.8
1
1.2
0 10 20 30 40 50 60Time After Chase (min)
NormalizedAntitrypsin
Level
FeCl3FeCl2Water
MediumSupplement
Medium Supplements in O2(Water or 500 μM Autoclaved Iron)
500 μM Autoclaved FeCl2
0
0.2
0.4
0.6
0.8
1
1.2
0 10 20 30 40 50 60Time After Chase (min)
NormalizedAntitrypsin
Level
N2O2AIR
– Eliminates oxygen dependence of α1AT degradation
Proteomics vs. Genomics• Transcriptional response provides insight into
what a cell wants to do• Protein synthesis tells us what the cell is doing• Protein isoforms may be important, e.g. narrow
glycosylation pattern for Enbrel.• DNA arrays are comprehensive but not precise• Broader dynamic range, increased resolution
and shorter duty cycle makes mass spec of complex protein mixtures feasible.
MANUFACTURING SCIENCEIn
tens
ity o
f Pr
oces
s U
nder
stan
ding
LEVEL 1
LEVEL 2
LEVEL 4
LEVEL 5
LEVEL 3
CORRELATIVE KNOWLEDGEWhat Is Correlated to What?
“CAUSAL" KNOWLEDGEWhat “Causes” What?
MECHANISTICKNOWLEDGE
How?
FirstPrinciples Why?
DESCRIPTIVE KNOWLEDGE: What?
IMPLEMENTATION OF PAT
• Develop and deploy the sensors to monitor process behavior and understand the process
• Identify critical control points that link process operation to product quality and if maintained assure product quality
• Replace the original quality assurance and control measures with those that link process understanding to product quality
This will require multiple sensors, new sampling techniques and strategies to interrogate the process.
Need to put in place an IT infrastructure that can handle the date and relationships.
This is an evolutionary process to accommodate new knowledge into new processes.
There needs to be a risk and learning focused organization in place.
STEPS INTO THE FUTURE• Understand what metrics (quality attributes) are
important• Build a roadmap for manufacturing science to focus on
what’s important • Expand high-throughput processes to high-insight
processes• Link advances in process & manufacturing science with
biomedical science – systems approach• Establish PAT and QbD early in process and product
development, aligning critical quality attributes along the value chain