1
A high throughput cell culture platform for bioprocess optimization
Seth Rodgers, CTO Bioprocessors
CPAC Rome
March 20, 2007
2© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
Outline
• The role of model systems in process understanding • A scale-down model bioreactor and the data sets we get now• Challenges remaining – especially the data sets we’d like to get
3© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
After discovery comes development, lots and lots of it!
Expression System
DevelopmentFlasks
Clone Evaluation
Media Development*
Process Optimization**
Note : ** Represent iterative processesSource : Nature Biotechnology Vol. 22 (11) 2004
• Screen and select the highest producing and most stable clone
• Develop optimal growth and production media for each cell line
• Optimize conditions for biomanufacturing process in a “scale-down” version
Scale Up
• Scale up process for use in large bioreactors for production of therapeutic
• Identify target, isolate gene, and develop expression system
• Knowing gene for the protein you want is great, but what cell line to use? What clone form that cell line is best. 100s of possibilities!
• 60 or more nutritional components in culture media, how many combinations? When to feed them? Inducers, promoters?
• What temperature? What oxygen level? CO2? pH any shifts? When to harvest?• A strategy of multi-factorial design is the natural way to attack this type of
problem, but is difficult to execute in cell culture because the parameters interact strongly-requiring a lot of experiments. This means models!
4© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
The role of model systems
• Here, the data is the product, faithful representation of process equipment is the goal
• Experiments with the systems that provide the best data, and the most understanding, i.e. production bioreactors themselves, are very time consuming and expensive.
• Model systems are universally used, but represent a compromise: reduced time and expense in exchange for imperfect data, which leads to imperfect understanding
• The same cost vs. data quality trade off that restricts experimentation in plant scale equipment often dictates the choice of model system
Process understanding
Model systems: Data is the product Real system:The API is the product
5© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
Can current models provide deeper process understanding?
• I can compromise on quality to a point, but only so far. The data still has to tell us what we want to know
– Are we at optimum?– Where to go next?
• But we need to know NOW!
• And we are going to need to know more soon!
– QbD ideas of variance identification and reduction.
– Statistical process control– Follow-on biologics
• Particular challenge with animal cells. (long experiment times, sensitive to culture conditions)
• A new model system could be very helpful
ThroughputExperimental capacity per
researcher
QualityAbility to predict manufacturing bioreactor
performance
Well plate
1’s
10’s
100’s
1000’s
Flask
Low High
High Quality, High Quantity
Bioreactor
6© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
Some High-Throughput Cell Culture System Requirements
• Deliver meaningful scalable data• Sustain cells, control temperature, O2, CO2, pH, agitation• Maintain sterility• Monitor cell density, pH, DO, metabolites, product titer• Operate with accuracy and precision and provide control of process
parameters comparable to bench top bioreactor systems• Automatic operation with minimal operator supervision• Integration with tools for designing experiments and handling data
7© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
SimCell MicroBioreactor Array
• 6 micro-bioreactors per array.• Working volume: ~700 µL.• Fluidic ports and channels for
inoculation, feeds, pH adjustment and sampling.
• Culture monitoring of biomass (OD), pH (immobilized sensors) and DO (immobilized sensors) by optical interrogation of micro-bioreactors.
• Proprietary gas permeable materials result in kLa ~ 10 hr-1 for oxygen and ~ 25 hr-1 for CO2.
• Experimental factors such as media composition, inoculation density, pH and feeds can be adjusted at the micro-bioreactor level.
• But sensing through thin plastic windows can be a challenge!
8© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
SimCell Automated Management System
• Incubators– One to five per system.– T, CO2, O2 and agitation control.
• Sensing module– Total biomass by OD.– pH by immobilized sensors.– DO by immobilized sensors.
• Sampling module– Sample removal to well plate.– Capable of dilution with single diluent
(PBS).– Capable of volumetric dilution or
dilution to specific cell density in well plate or MBR.
• Dispensing module– One to eight pumps.– Real-time mixing at point of delivery.– Fluid sources may be swapped in
between cycles for increased capacity.
9© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
SimCell Automated Management System (SAMS)
10© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
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
SimCell™ On-Line Measurements: Cell Density
• Measured using Optical Density at 633 nm on Sensor Station– OD is linear to 2.2– Working OD range is extended by dynamic neutral density filtering– accurate measurements up to 50M cells/ml have been demonstrated
• OD vs. cells/ml curve is specific to cell type• OD yields total intact cells: live + dead
• Error bars are +/- 1 std. dev. (+/- 16% variance)
• Inteferences matter how to compensate for cell size? aggregation?
• A better measurement might also tell us how many live and how many dead – dielectric spectroscopy?
11© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
Measurements: Immobilized pH Sensors
0 2 4 6 8 105.8
6.0
6.2
6.4
6.6
6.8
7.0
7.2
7.4
7.6
7.8
8.0
8.2
CD-CHO media 1 media 2 media 3 media 4, w/phenol red curve fit: CD-CHO data
pH
fluorescent ratio
• Response is independent of media
• Precision is < 0.06 (±3 std. dev.) over
the pH range 6.0 – 8.0
manual pH I2/I1 (R) precision+/- 3 std. dev.
5.98 0.76615 0.026.40 1.58483 0.016.76 2.76599 0.026.91 3.44210 0.027.26 5.41638 0.037.91 8.96870 0.06
media #1
• pH Sensor Composition – Hydrogel (Water-swellable polymer)– Covalently bound dye fluorescent pyrene derivative.
• Sensor manufactured by screen-printing, followed by UV polymerization
12© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
SimCell On-Line Measurements: pH Measurement TechnologyImmobilized pH Sensor
• Covalently bond fluorescent pH dye to hydrogel
• Hydrogel polymerized to bottom surface of MBA
• Retains ratiometric pH response
Four sensors/chamber
13© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
Automated pH Control
)()(
3
1010][
2 pKapHpKapH
addH
CO
totaladdif
HCOk
PVV
• Vadd: volume of solution of base to add
• Vtotal: total volume of the sample in the microbioreactor before addition
• PCO2: pressure of CO2
• kH: Henry’s Law constant for CO2
• [HCO3-]add: concentration of bicarbonate in the adjustment solution
• pHinitial and pHfinal: starting pH value and pH setpoint, respectively
• Similar equations are derived for use of sodium carbonate, sodium hydroxide, and monoprotic acids for pH adjustment.
14© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
Measurement and Control: Maintaining pH Setpoints
•3 pH setpoints•18 subprotocols•9 μBR/subprotocol•pH adjusted 2x/day
•Chart shows average pH for each subprotocol over the course of the experiment.
15© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
SimCell™ On-Line Measurements: Dissolved Oxygen (DO) Measurement • Oxygen-sensitive dye (platinum porphyrin
derivative)– Excitation of dye yields emissive triplet state.
– As [O2] increases, dye emission is quenched and τF decreases
– τF is correlated to phase shift (φ) between modulated excitation and emission signals
0 5 10 15-2.0
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
Excitation Emission
Sig
nal (
arbi
trary
uni
ts)
time
φ
10 15 20 25 30 35 40 45-10
0
10
20
30
40
50
60
70
80
90
100
110
120
pO
2 (%
air
sa
tura
tion
)
phase
Correlation of φ to DO: error bars are +/- 1SD (+/- 10% variance).
Phase shift between excitation (blue) and emission (red) signals.
16© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
Novo’s Comparison with Current Technologies
Significant improvement n process yield at lower cost and shorter time
28293031323334353637
0
5000
10000
15000
20000
25000
Yield
pHTemperature
Peak Production
Summary• 84% increase in yield• Scalable to 1,000 liter production vessels
17© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
Application across platforms and processes
• The central question is “To what extent is the MBA performance a predictor of the bioreactor result?”
• The R^2 statistic is a well-established way to quantify the answer to this question, computed by constructing ordered pairs of MBA and reference model system results
• Advantages of R^2 are that is can be constructed independent of platform, process cell line, etc.
– Compare relative predictive power across model systems: flask, MBA, bioreactor, well plate, etc.
– Metric of continuous improvement as technology evolves
• This graphic shows results over many cell lines, processes and vessels, predictive power might be even better for data with these factors kept constant
R2=0.96
R2 for 2006 client evaluation projects
y = 1.0926x + 36.042
R2 = 0.9628
0
1000
2000
3000
4000
0 1000 2000 3000 4000
MBA Titer
Refe
renc
e Ve
ssel
Tite
r
18© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
What’s missing
• Protein titer of course!– Enzymatic like ELISA is the most common, but it takes work, even with automation– Something else?
• Viability• Some understanding of the protein quality (glycosylation, aggregation)• All those media components in the culture broth
– Nutrients: glucose, glutamine, amino acids, vitamins– Metabolic products: lactate, etc.
• Can spectroscopy (NIR, MIR, Raman work here?)• Anything else useful in characterizing and ‘fingerprinting’ the process, that is, a
useful predictor of process outcomes.
• Ideal measurement (for us at least) is – Non invasive – it it needs a sample, best case is
• Small sample• Works with crude broth, no pre- treatment
– Matched throughput– Calibrated less frequently than once per MBA– Compatible with flexible! plastic cell culture device (challenge for some spectroscopy)– Cost competitive pulling samples and using well plates
19© 2005 BioProcessors • http://www.bioprocessors.com • [email protected] • (781) 935-1400
Conclusions
• Model systems are indispensable tools, and increasing demands for data will be difficult to meet with current platforms.
• A high-throughput cell culture system presents a possible solution if the data is of sufficient quality to predict process outcomes.
• BioProcessors SimCell system represents one possible solution that combines high throughput with highly representative data.