Transcript Kinetic CHO Cell Modelling and Simulations

16 May 2014 –Hinxton
Kinetic CHO Cell Modelling and
Simulations
– The Modeling Cycle and Industrial Application
Sophia Bongard
1
Company Overview
Insilico Biotechnology AG
Insilico
Biotechnology
Biotech
Pharma
Process Analysis
& Optimisation
Testing of
New Drugs
Software
Founded in 2001, headquartered in
Stuttgart, Germany
Inter-disciplinary team comprised of
biologists, chemists, computer scientists,
physicists, and bioprocess engineers
Expertise in modelling and simulation of
biochemical networks
Solution provider for 40+ international
companies and academic research institutes
Services
Insilico quantifies and predicts
cellular processes for
the Life Science Industries.
2
Technology Platform
– Overview
Insilico Discovery™
Insilico Inspector™
Insilico
Software
> 8,000 reactions
> 2,000 compounds
> 20 strains
+ organs/body
Insilico Databases
Insilico Cells
Insilico Organs
> 100,000 cores
High-Performance
Computing
Insilico Technology Platform
Insilico delivers (i) quantitative insight into biotechnological
processes and (ii) predictive simulations.
3
Why do we need CHO Cells?
– Usability and industrial Application
What are CHO cells?
Immortalised cell strain from
chinese hamster ovaries
Picture Source: wikipedia
What do the cells produce?
Recombinant proteins
How are they produced?
Feed
cSubstrates
Sampling
cProduct
Fermentation processes
4
Model-supported
Cell Line Development
Cell line: host cell + recombinant expression construct
1. Transfection
2. Amplification
& Selection
3. Screening
4. Expansion
Many Clones (100‘s – 1000‘s)
µL - mL
5. Growth Evaluation
& Process Optimization
10 – 20 Clones
100 – 1.000 mL
Host
Cell
Cell Line Engineering
Clone Selection
Process Control
Computer-supported Processes
5
Case Scenario for this session
– IgG antibody Production
Most common antibody in blood needed for immune defense against bacteria and
viruses
Artificial generation via CHO cells to target proteins in the human body (used e.g. for
cancer therapies)
Case scenarios
Producer strain comparison with stationary CHO models
Bioprocess optimization of best clone using
dynamic CHO model
6
In silico Workflow
7 Hand over new Media
Design to customer
6 New Bioprocess
Design
5 Kinetic parameter
estimation
4 Determination of stationary
flux distribution
1 Model reconstruction
and network adaption
2 Experimental Design
3 Data integration
7
Model Reconstruction and
Network Adaption
Challenge
To gain a consistent model with all needed data information in a short time
and appropriate visualisation
Solution
KEGG: Pathway Research
UNIPROT: stoichiometry
PubChem: chemical compound structures
 multiple database information (reactions, compounds), which can be
used for model reconstruction, e.g. in COPASI and Cytoscape
Benefits
„All-in-one“ software solution which delivers model in universal SBML format
8
CHO Cell Metabolism
IgG
ER and Golgi
Glucose
Amino Acids
O- and N-Glycosylation
Glucose
glycoprotein
Pyruvate
BIOMASS
Purine and Pyrimidine Metabolism
Lipid Metabolism
Mitochondrium
TCA Cycle
Steroid Synthesis
Amino Acid Metabolism
Protein
DNA
RNA
Glycogen
Lipid
AKG
O2
CO2
Peroxisomes
Nucleus
Organic Acids
NH4 (e.g. Lactate)
9
CHO Model Network Adaption
Merge Host Cell
with Novel Reactions
»Super-Network«
+ Host cell
+ new strain-specific
reactions/pathway
+ known gene knock outs
Producer Strain Comparison
Target Identification,
Design New Media
Integration of Kinetics
+ experimental data
(extracellular and
intracellular)
Stationary Model
Identification
Dynamic Model
Identification
10
Definition of Model Kinetics
Ordinary Equation System:
Stationary model:
Dynamic model:
𝑑𝑐
𝑑𝑡
𝑑𝑐
𝑑𝑡
= 𝑟1 − 𝑟2 + 𝑟3 with r=const.
= 𝑟1 − 𝑟2 + 𝑟3 with r=f(c, p), nonlinear kinetics
Kinetics
linlog kinetics: 𝑟 = 𝑣𝑚𝑎𝑥 ∙ (1 + 𝜀1 ∙ 𝑙𝑜𝑔
𝑐1
𝑐1 𝑠𝑠
+ 𝜀2 ∙ 𝑙𝑜𝑔
𝑐2
𝑐2 𝑠𝑠
+…),
with 𝜀(𝑃𝑟𝑜𝑑𝑢𝑐𝑡𝑠, 𝐼𝑛ℎ𝑖𝑏𝑖𝑡𝑜𝑟𝑠) ≤ 0 𝑎𝑛𝑑𝜀(𝑆𝑢𝑏𝑠𝑡𝑟𝑎𝑡𝑒𝑠) ≥ 0
Michaelis-Menten kinetics: 𝑟 = 𝑣𝑚𝑎𝑥 ∙
𝑐1
𝐾𝑀 ∙𝑐1
Convenience kinetics, Mass action kinetics, Hill kinetics …
11
In silico Workflow
7 Handover new Media
Design to customer
6 New Bioprocess
Design
5 Kinetic parameter
estimation
4 Determination of stationary flux
distribution
1 Model reconstruction
and network adaption
2 Experimental Design
3 Data integration
12
Experimental Design
Challenge
To find the minimum set of required measurements providing the maximum
of information for model identification
Solution
Optimal Experimental Design (CSIC, CWI, Joke Blom)
Benefits
Get maximum quality/quantity of information from a minimum of
experimental effort
Saves resources
13
Implementation of
General conditions
– Definitions of constraints, variables and parameters
Cell concentration in fermenter (500,000 cells/ml)
Fermenter Volume (6 Liters)
Fed-Batch/Batch/Continuous Process:
Feed Rates, Feed Concentrations, Bolus Shots
Process time: 300 h
Sampling
Biomass densitiy in cell
Biomass growth rate
Product protein composition (amino acids)
14
Identification of Optimal
Experimental Conditions
– Required inputs for CSIC Method
Mathematical Model Inputs:
Ordinary differential equations including external conditions (e.g. feeding,
temperature), kinetic parameters, state variables (e.g. fermenter volume)
Auxiliary functions describing the relation between model states and experimental
measurements (preliminary data)
Measurement Inputs:
Measurement quality
Limitations (e.g. glucose solubility)
Variables to be considered in experimental design (sampling times, feeds, to be
measured metabolites…) =Output of optimised experimental design
15
In silico Workflow
7 Handover new Media
Design to customer
6 New Bioprocess
Design
5 Kinetic parameter
estimation
4 Determination of stationary flux
distribution
1 Model reconstruction
and network adaption
2 Experimental Design
3 Data integration
16
In silico Workflow
7 Handover new Media
Design to customer
6 New Bioprocess
Design
5 Kinetic parameter
estimation
4 Determination of
stationary flux distribution
1 Model reconstruction
and network adaption
2 Experimental Design
3 Data integration
17
Determination of Stationary
Flux Distributions
Challenge
Determine phase-dependent flux distributions which best describe
measurements
Solution
COPASI: steady-state analysis and parameter estimation (UNIMAN)
Multiple Objective FBA (CSIC, Julio Banga)
Benefit
State-of-the-art parameter estimation and flux balance analysis making
integration of multiple objectives possible
Considerations of non-obvious criteria (Example??)
18
Case Study 1:
Producer Strain Comparison
19
Producer Strain Comparison
– Strategy
Input:
Time Series Data of extracellular Metabolites,
Offgas Data, Feeds, Samples
Procedure:
Calculation of according phase-dependent
uptake/consumption rates
Flux Balance Analysis for intracellular
distribution for multiple process phases
Analysis:
Either phase-wise comparison of performance
indicators or over whole process
20
Producer Strain Comparison
– Decision Criteria I
Product Titer
The final product
concentration in
fermenter
Product Yield
The ratio between
product produced to
glucose consumed
Biomass Yield
Cell Density
The ratio between
generated biomass to
consumed Glucose
Maximum viablecell
concentration in the
fermenter
Growth Rate
Productivity
Maintenance
Maximum or
average growth rate
(biomass formation
rate) over the
process
Maximum or
average specific
product generation
rate over the whole
process
Cellular rate of ATP
consumption for
maintaining the cell in
a viable state.
21
Producer Strain Comparison
– Decision Criteria II
Product Yield
Maximum Yield
Producer
Strain 1
Producer
Strain 2
Producer
Strain 3
Biomass Yield
Producer Strain 1 has best performance
indicators, but high ammonia release
 Strain 1 for next generation strain
development
22
In silico Workflow
7 Handover new Media
Design to customer
6 New Bioprocess
Design
5 Kinetic parameter
estimation
4 Determination of stationary flux
distribution
1 Model reconstruction
and network adaption
2 Experimental Design
3 Data integration
23
Kinetic Parameter Estimation
Challenge
Complexity of large dynamic models (large number of parameters, stability issues,
model too robust or fragile)
High resource demand of calculations
Solution
AMIGO: ScatterSearch optimisation method in combination with ensemble modelling
(CSIC, Julio Banga)
COPASI: integrated optimisation algorithms
High performance computing (Super Computer)
Benefit
Improved assessment of predictive value due to quantified uncertainty
Saving time by reducing number of required restarts during parameter estimation
24
Dynamic Model Identification
+ dynamic model rate equations
(use stationary flux distribution as
reference flux in dynamic model)
+ initial parameter guess
Identified stationary
model
Preliminary dynamic
model
Target Identification, New
Media Design
+ experimental data (extracellular
and intracellular)
Parameter Estimation
Identified dynamic
model
25
In silico Workflow
7 Handover new Media
Design to customer
6 New Bioprocess
Design
5 Kinetic parameter
estimation
4 Determination of stationary flux
distribution
1 Model reconstruction
and network adaption
2 Experimental Design
3 Data integration
26
Case Study 2:
NH4 Reduction in CHO Strain 1
27
NH4 Reduction in a CHO Process
– Case Study 2: Summary
Challenge
NH4 accumulation in a CHO fed-batch processes for monoclonal antibody production
impairs process performance
Solution
Identification of sources of NH4 formation in different process phases
Identification of intracellular and extracellular substrate limitations/bottlenecks
New media design for better performance through feed optimisation
Benefits
Reduce NH4 levels, improve viability late in the process and product
28
NH4 Reduction in a CHO Process
Amino Acid Synthesis View in Insilico Inspector™
>80% Degradation
Phase 2
<40% Degradation
Fluxes in µmol Carbon/(gDW*h)
Asparagine and glutamine exhibit the highest degradation rates of the amino
acids taken up in Phase 2
29
NH4 Reduction in a CHO Process
– Case Study 2: Phase-specific identification
of NH4 sources
Nitrogen Metabolism View in Insilico Inspector™
Phase 2
Intracellular degradation of
asparagine and glutamine
is responsible for the
majority of NH4 released
until Phase 3
Fluxes in µmol Nitrogen/(gDW*h)
Process Phase
Amino acids with significant
degradation (>30% of uptake)
Phase 1
Phase 2
Phase 3
Asn > Gln > Leu
> Val > Tyr
Asn > Gln > Leu
> Val > Tyr
Asn > Gln >
Ser
30
Identification of New Media Design
for NH4 Reduction
+ definition of objectives
 reduce NH4, increase product titer, …
Identified dynamic
model
Definition of elements
to be optimized
Prediction of altered cell dynamics
and new performance indicator
values
+ constraint definitions
Optimization
New Media
Composition
31
In silico Workflow
7 Handover new Media
Design to customer
6 New Bioprocess
Design
5 Kinetic parameter
estimation
4 Determination of stationary flux
distribution
1 Model reconstruction
and network adaption
2 Experimental Design
3 Data integration
32
NH4 Reduction in a CHO Process
– Case Study 2: Implementation
Customer reduced asparagine in the feed by 38%
Optimized
Result: reduced ammonium levels, improved viability, and product
titer increased by >30% relative to the reference run
33
Benefits of kinetic CHO models
Gain quantitative insight
Save experimental studies and reduce development time
Improve yield, productivity, and quality of biotech products
Generate new know-how and intellectual property
Taking decisions based on quantified processes
34
Many thanks to
Joachim Schmid, Dirk Müller, Klaus Mauch (Insilico Biotechnology AG)
And to the BioPreDyn consortium!
35
Contact
Sophia Bongard
Insilico Biotechnology AG
Meitnerstr. 8
70563 Stuttgart | Germany
T +49 711 460 594-25
[email protected]
36