Challenges and Opportunities in Biopharmaceutical ...focapo-cpc.org/pdf/Braatz.pdf · Challenges and Opportunities in Biopharmaceutical Manufacturing Control Moo Sun Hong, ... Pharmaprojects

Challenges and Opportunities in Biopharmaceutical Manufacturing Control

Moo Sun Hong, Kristen Severson, Mo Jiang, Amos E. Lu, J. Christopher Love, Richard D. Braatz

Outline

• Background, Current State, Technology Directions

• A Biomanufacturing-on-Demand Platform

• Comments on the State-of-the-Art in Control

• Closing

Background: Biopharmaceuticals Manufacturing

• Products derived from biological organisms for treating or preventing diseases

• Hundreds of approved products on the market

• Over 7000 medicines in development

• Usually any off-spec material requires rejection of entire lot

Deloitte (2016). 2016 Global Life Sciences Outlook: Moving Forward with Cautious Optimism. Deloitte LLP, Boston, MA.Informa (2016). Pharmaprojects Pharma R&D Annual Review 2016. Informa PLC, London, United Kingdom.

USD in billions

2010 2011 2012 2013 2014 2015 2016 2017 2018 2019

500

400

300

200

100

0

% non-bio drug candidates in pipeline

% bio drug candidates in pipeline95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16

100908070605040302010

0

Current State of Biopharmaceutical Manufacturing

• Sequence of batch unit operations

• Upstream: cell culture and harvest

• Downstream: purification using chromatography, filtration, diafiltration

• Bioreactor has T, dissolved oxygen, and maybe pH controls

• Minimal modeling/controls

Cell Culture Centrifugation

Protein A Chromatography

Low pHViral

Inactivation

Cation ExchangeChromatography

Depth and Membrane Filtration

Anion ExchangeChromatography

Viral Filtration

Ultrafiltration/Diafiltration

DOWNSTREAM

UPSTREAM

Kelley, B. (2009). Industrialization of mAb Production Technology: The Bioprocessing Industry at a Crossroad. MAbs, 1, 443-452.Shukla, A. A., Thömmes, J. (2010). Recent Advances in Large-Scale Production of Monoclonal Antibodies and Related Proteins. Trends in Biotechnology, 28, 253-261.

Motivation: Biopharmaceuticals Manufacturing

• Over 7000 medicines in development

• Industry driven to look for new technology to increase flexibility and reduce costs

continuous flow

novel process designs

Deloitte (2016). 2016 Global Life Sciences Outlook: Moving Forward with Cautious Optimism. Deloitte LLP, Boston, MA.Informa (2016). Pharmaprojects Pharma R&D Annual Review 2016. Informa PLC, London, United Kingdom.

USD in billions

2010 2011 2012 2013 2014 2015 2016 2017 2018 2019

500

400

300

200

100

0

% non-bio drug candidates in pipeline

% bio drug candidates in pipeline95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16

100908070605040302010

0

Moving Towards Continuous-flow Operations

http://www.drugdevelopment-technology.com/contractors/contract_research/biovian/biovian1.htmlhttp://www.genengnews.com/gen-articles/novasep-morphs-into-full-service-company/3048/

• Move manufacturing from batch to integrated continuous operations with minimal holdup in between

• Continuous flow enables higher process flexibility, consistency,

and volumetric capacity lower equipment cost and operational

complexity tighter specifications on product quality

→ need for research in process modeling/design/controls

Example Limitation of an Existing Process Design: Packed-Bed Chromatography

• Currently the dominant method for protein purification• High resolution but costly for high-dose pharmaceuticals• The resin in a single column can cost >$1 million• Operating cost scales

linearly with throughput• Strongly prefer

sublinear scaling→ see paper for design

alternatives

Inlet Outlet

sorption

convection &dispersion

Porous bead

pore diffusion

film mass transfer

→ need for research in process modeling/design/controls

Technology Directions

• Increased understanding and optimization of each unit operation and component

• Microscale technologies for high-speed continuous process development

• Availability of plug-and-play modular unit operations with integrated process control and monitoring systems to facilitate straightforward deployment in the laboratory

• Dynamic mathematical models for unit operations and entire plant to support process development and plant-wide control

• Easy-to-use model-based control technologies for optimizing startup, changeover, and shutdown

Design of Control Systems Based on “Virtual Plant”

• Constructed from first-principles models wherever possible, grey-box models where necessary

• Highest complexity models used for the invention and optimization of process designs and development

• Lower complexity plant-wide model runs in parallel with process operations, for process control and quality and equipment condition monitoring

Micromixer

pH

Outline




• Closing

Towards Biomanufacturing on Demand (BioMOD)

• Enable flexible methodologies for genetic engineering/modification of microbial strains to synthesize multiple and wide-ranging protein-based therapeutics

• Develop flexible & portable device platforms for manufacturing multiple biologics with high purity, efficacy, and potency, at the point-of-care, in short timeframes, when specific needs arise

• Include end-to-end manufacturing chain (including downstream processing) within a microfluidics-based platform

J.C. Love, Towards making biologic drugs on demand, 4th Int. Conf. on Accelerating Biopharmaceutical Development, January 25, 2015

Design Requirements Patient

Downstream

Concentration

Identity

Purity

pH, DO, T

Potency

Safety

Sterility

Analytics

Purified Product for Quality Testing

Multivariate Model

OK forReleaseNot OK

☐

Final Product

Sterile Media

Yeast Inoculum

Crude Product Hold Tanks

Upstream

On-line Reactor ControlPerfusion

UF/DF Membrane

Waste

Fill

Integrated and Scalable Cyto-Technology (InSCyT) Biomanufacturing Platform

Real-TimeProcess

Data

Affinity Chromatography

Waste

Column Tanks 1 2

EluteWaste

Polishing Membrane #1 Tanks

1 2

Polishing Membrane #1

Waste

1 2



J.C. Love, Towards making biologic drugs on demand, 4th Int. Conf. on Accelerating Biopharmaceutical Development, January 25, 2015

Plant-wide Control Approach

Characteristics of InSCyT• Multi-product manufacturing plant• Continuous & discrete operations• All of the process characteristics

articulated by Lee/Weekman 1976, Foss 1983, and more

• No SS & must align with regulatory requirements (no off-spec product)

Approach adapted from the chemical industry• Employing systematic & modular design of plantwide control strategies

for large-scale manufacturing facilities (Stephanopoulos/Ng, JPC 2000)• Using algorithms that can handle nonlinearities, uncertainties,

distributed states, time delays, unstable zero dynamics, constraints, and mixed continuous-discrete operations

A.E. Lu, J.A. Paulson, N.J. Mozdzierz, A. Stockdale, A.N. Ford Versypt, K.R. Love, J.C. Love, R.D. Braatz (2015). Control systems technology in the advanced manufacturing of biologic drugs. Proc. of the IEEE Conference on Control Applications, 1505-1515.

Downstream

Concentration

Identity

Purity

pH, DO, T

Potency

Safety

Sterility

Analytics


Multivariate Model

OK forReleaseNot OK

☐

Sterile Media

Yeast Inoculum


Upstream


UF/DF Membrane

Waste

Fill

Real-TimeProcess

Data


Waste

Column Tanks 1 2

EluteWaste


1 2


Waste

1 2



Plant-wide Control Approach

• Build first-principles dynamic modelsfor each unit operation (UO)

• Design control system for each UOto meet “local” material attributes

• Evaluate performance in simulationsand propose design modificationsas needed

• Implement and verify the control system for each UO

• Design and verify plantwide control system to ensure that the product quality specifications are met

Downstream

Concentration

Identity

Purity

pH, DO, T

Potency

Safety

Sterility

Analytics


Multivariate Model

OK forReleaseNot OK

☐

Sterile Media

Yeast Inoculum


Upstream


UF/DF Membrane

Waste

Fill

Real-TimeProcess

Data


Waste

Column Tanks 1 2

EluteWaste


1 2


Waste

1 2




Downstream

Concentration

Identity

Purity

pH, DO, T

Potency

Safety

Sterility

Analytics


Multivariate Model

OK forReleaseNot OK

☐

Final Product

Sterile Media

Yeast Inoculum


Upstream


UF/DF Membrane

Waste

Fill


Real-TimeProcess

Data


Waste

Column Tanks 1 2

EluteWaste


1 2


Waste

1 2



Robust Split-range Control of Bioreactor

• Designed hybrid robust controllers that switch during each manipulated variable change

• Vastly reduces variability by optimized switching


Downstream

Concentration

Identity

Purity

pH, DO, T

Potency

Safety

Sterility

Analytics


Multivariate Model

OK forReleaseNot OK

☐

Final Product

Sterile Media

Yeast Inoculum


Upstream


UF/DF Membrane

Waste

Fill


Real-TimeProcess

Data


Waste

Column Tanks 1 2

EluteWaste


1 2


Waste

1 2



Chromatography Modeling

Nonlinear PDEs• Species mass balance

in liquid phase• Adsorbed phase balance

θ∂

−=∂

*( )ii ik qq q

Inlet Outlet

SorptionConvection &

Dispersion

Interstitial volume Porous bead

Pore diffusion

Film mass transfer


Chromatography Modeling

Nonlinear PDEs• No steady-state, always in transient mode• No measurements in the column, large time delay to exit

measurement• Switch between multiple sets of nonlinear PDEs• Practice is to run open loop or switch on exit concentrations

Inlet Outlet

SorptionConvection &

Dispersion

Interstitial volume Porous bead

Pore diffusion

Film mass transfer


Hybrid Simulation of Chromatography Operation

• Four-stage bind-and-elute operation mode

# Stage Time Protein Conc. Salt Conc.1 Load tl cp,i cs,l

2 Wash tw 0 cs,w

3 Purge tp 0 cs,e

4 Elute te 0 cs,e

# Specified by upstream units

# #

→ Sensors at entrance and exit only, i.e., feedback limited→ Open-loop optimal control can be used to determine

“optimal” operation based on first-principles modelA.E. Lu, J.A. Paulson, N.J. Mozdzierz, A. Stockdale, A.N. Ford Versypt, K.R. Love, J.C. Love, R.D. Braatz (2015). Control systems technology in the advanced manufacturing of biologic drugs. Proc. of the IEEE Conference on Control Applications, 1505-1515.

Multiple Performance Objectives

Performance Objective Description

Purity

Recovery

Exit Concentration

Productivity

0

,0 0

,

,

e

e e

pdt elution

pdt elution cont elution

t

t t

c dt

c dt c dt+

∫∫ ∫

,

,

0

et

pdt elution

pdt loal d

c dt

t c∫

,0

e

pdt elution

t

e

c dt

t∫

,0

e

pdt elution

l w p e d

tQ c dt

t tt t t+ ++ +∫


Region oflow quality

localextrema

Global optimum

Multiobjective Optimization

• Purity, recovery, exit concentration, and productivity are competing objectives

• Look to maximize exit concentration & productivity while constraining purity to be >99% and recovery to be >90%

• Multiobjective problem solved using stochasticoptimization to avoid being trapped in low-quality local extrema


Pareto Optimality (Tradeoff Curve)

Dilution Factor

Curves are optimal tradeoff for various feed concentrations

Select “optimal” value based on measured feed concentration using a lookup table


Nonlinear feedforward control of a hybrid PDE system, with minimal online computational cost

Column Tanks 1 2

Downstream


Sterile Media

Yeast Inoculum


Upstream

1 2 1 2




Waste Waste Waste


UF/DF Membrane

Waste

Concentration

Identity

Purity

pH, DO, T

Potency

Safety

Sterility

Analytics


Multivariate Model

OK forRelease

Not OK

☐

Final Product

On-line Reactor Control

Perfusion

Fill

Elute

Real-TimeProcess

Data

Need buffer makeup unit to avoid carrying all possible buffers on the platform


Batch Buffer Makeup “UO” Modeling

pH Model (Waller, Kravaris, ...)Method of reaction invariants

Mass balances are linear equationspH is a static nonlinear function of the invariants

Multicomponent Mixing RuleRequires estimation of effective ionic radius from data

Single Component Conductivity First-principles Model

Flowrates Fin are the manipulated variables

Semi-batch Tank Reaction Invariant Species Conservation

;

Many Systems and Control Problems Appear in the Buffer “UO”

• Dynamics are highly nonlinear• Slow mixing and uncertain parameters• Model-based control can overcome these issues by predicting the effect of inputs on the process

– Need good parameter values based on process data– Need to develop robust control strategies that operate well

under uncertainty

Design of experiments for conductivitypH control in micromixers

examples


Example of pH Control in Buffer Mixing “UO”

• Use of in-line micromixers allows for process intensification• pH modeled using the reaction-invariant method• Material balances modeled using tanks-in-series

Micromixer

pH

Fw, wwFb, wbFp, wpFs, ws

w1 w2 wi wn

11

1 1

1 tin

Fddt V V

−= Fw W w

T

w b p sF F F F = F

t w b p sF FF F F+ + += 1 )(ti i

i

id Fdt V −= −w ww

[ ]in w b p s=W w w w w pH ( )nf= w


Direct PID vs Reaction-invariant PID Control

• Direct PID

• Reaction-invariant PID

Slow response

Marginally stable

Fast response

Slow response

PID based on the reaction invariants improves performance across pH range


Reaction-invariant Controller:Performance Under pH Model Uncertainty

• No model uncertainty

• 20% error in pKas within pH model

InstabilitySlow response

Caveat: Reaction-invariant control requires accurate models to achieve good

performance


Model Adaptation of Reaction-invariant Control via Dual Micromixer Configuration

Micromixer

pH

Micromixer

pH

First micromixer used to excite system to enable accurate parameter estimation of pH modelSecond micromixer used to perform corrections to the output of the first micromixer and effect setpoint tracking

Rhinehart & Riggs proposed multi-tankLinear control of n tanks studied by SkogestadOur approach:


Dual Micromixer Configuration

Input signal to generate data for model adaptation

First micromixer used to excite system to enable accurate parameter estimation in pH model

Second micromixer used to perform corrections to the output of the first micromixer and effect setpoint tracking


Closed-loop Performance Comparison

• Single micromixer without model adaptation

• Dual micromixer with model adaptation

Model adaptation allows for rapid response in spite of high uncertainty


Outline




• Closing

Comments on the State-of-the-Art in Model-based Control Technology

• The best commercial plant simulation software handles nonlinearities, time delays, unstable zero dynamics, constraints, mixed continuous-discrete operations, and some uncertainty analysis methods (e.g., Si, MC)

• Distributed states facilitated by various math tricks, e.g., moment analysis, transform methods, characteristics

• More sophisticated polynomial series and uncertainty analysis tools can be wrapped around of integrated into such software

• More control research is needed to prove stability, simplify tuning, reduce on-line computational cost, and improve closed-loop performance (e.g., MPC)

Outline




• Closing

Summary

• Described some research challenges and opportunities in the manufacturing of biopharmaceuticals

• Discussed some control technology within the context of a specific biopharmaceuticals-on-demand platform

• Model-based control solutions are a mix of nonlinear feedback & feedforward strategies

• Described hybrid controls and a new type of adaptive control strategy motivated by a specific application need

• Described state-of-the-art in control technology and some research needs

Closing Comments

• Small-molecule pharma PSE went from very little in the late 1990s to becoming well established

• We believe that the rise of PSE in biologic manufacturing will mirror its rise in small-molecule pharmaceuticals that occurred over the last 20 years, but will be faster

D2

D1

R2

R1LC LC

M1 M2 S1

C1 C2

W1

LC LC

C3 C4

LC

LC LC

S3S4 W2

M3 M4

M5

S5 S6 E1 CS

CAT

A

B

S2

S1

S3

S1

PU1

PU2

PU4

PU3

S1

S1

S1

C

D

E

S1

EX1

EX2FP

LC

S1

FCsp

FT

FCsp

spsp

sp

FCsp

CC FT

CTFC

FCsp

TC

sp

spsp

FCsp

DCsp

sp

S1RC CC sp

DC

LC

sp

1CATA B I+ ←→

1 2 BPI C I P+ → +

2I E API+ →

Bequette, Comp Chem Eng, 20:S1583–S1588, 1996

Togkalidou+, AIChE J, 47:160-168, 2001

Lakerveld+, Org Process R&D, 19:1088-1100, 2015

Simon+, Org Process R&D, 19:3-62, 2015

AcknowledgmentsHassan Ait HaddouMonica AminPaul BaroneCatie BartlettLisa BradburyDanielle CampJicong CaoDivya ChandraMichael CostaSteve CramerAleksander CvetkovicAmanda Del RosarioSusan DexterJuyai-Ivy DongChaz GoodwineJongyoon HanPeter HanWilliam HancockWilliam HerringtonShan JiangPankaj KarandeSojin KimSung Hee KoRalf KuriyelTaehong KwonRachel LeesonJames LeungLihong LiuKerry LoveAmos LuTimothy Lu

Huaiya MengSylvia MessierNicholas MozdzierzJustin NelsonRussell NewtonWei OuyangSantosh PandeJoel PaulsonPablo Perez-PineraOliver PurcellGK RajuRajeev RamNigel ReuelDaniel SalemKartik ShahDivya ShastryGajendra SinghAnthony SinskeyStacy SpringsAlan StockdaleMichael StranoKyOnese TaylorSteve TimmickJP TrasattiNicholas VecchiarelloAshley Ford VersyptAnnie WangJames WooDi WuAnna Zhang

This material is based upon work supported by the Defense Advanced Research Projects Agency (DARPA) and SPAWAR Systems Center Pacific (SSC Pacific) under Contract No. N66001-13-C-4025. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Defense Advanced Research Projects Agency (DARPA) and SPAWAR Systems Center Pacific (SSC Pacific).

High-Dimensional and Heterogeneous Data

Heterogeneity in data type and time scale in bioreactor has motivated the use of• Kernel-based methods• Dynamic time warping (DTW)

Very high dimensions in hyperspectral images have motivated the use of• Hierarchical clustering• Greedy algorithms

Charaniya, S., Hu, W. Karypis, G. (2008). Mining Bioprocess Data: Opportunities and Challenges. Trends in Biotechnology, 26, 690-699.

Bioreactor example

Supplementing First-principles with Black-box Models

First-principles model• Based on detailed process

understanding• Conservation equations, reaction

mechanisms, constitutive relations

Black-box model• Based on statistical correlations

observed in experimental data• Principal component analysis

(PCA), partial least squares (PLS)

“Grey-box” model• Combine first-principles and black-box models• Generate more complete and accurate predictions• Parallel approach• Embedded approach• Surrogate output approach

Packed-Bed Chromatography

• Currently the dominant method for protein purification• High resolution but costly for high-dose pharmaceuticals• Operating cost scales linearly with throughput• Increased product titers achieved with modern bioreactor

Non-chromatography separation methods• Aqueous two-phase extraction• Precipitation• Crystallization

Crystallization

• Effective and inexpensive industrial operation to achieve adequate purity and production in large quantities

• Cost scales sub-linearly with throughput• Used for small-molecule drugs and some proteins

• Insulin: treatment of diabetes (room T, pH 8.2, ¼~72 h)• Insulin lispro: fast-acting insulin analogue

5°C, pH 9, NaCl 37.5 mM,

0.75 M acetic acid, 24 h

Insulin lispro 20 g/L

Fermentation Cell Separation

Proteolytic Excision Chromatography Crystallization

Jackson, R. (1973). Process for the Crystallization of the Ammonium and Alkali Metal Salts in Insulin. U.S. Patent 3,719,655.Baker, J. C., Roberts, B. M. (1997). Preparation of Stable Insulin Analog Crystals. U.S. Patent 5,597,893.

Challenges for Therapeutic Proteins

• Hard to find conditions for reproducible crystallization using pharmaceutical-grade buffers and precipitants

• Protein can be denaturated by pH variation, temperature change, precipitant addition, and agitation

→ Phase equilibrium model to characterize multidimensional solubility surfaces

• Slower nucleation & growth rates than small-molecule drugs→ Continuous-flow crystallizers must be designed to operate

at high crystal surface area

Control of Biopharmaceutical Crystallization

First-principles approach• Optimizing objective function

related to the crystal size distribution by using a first-principles model

• Optimal experimental design and data collection

–Parameter estimation–Mechanism selection

• Robust optimization–Worst-case–Weighted

Fujiwara, M., Nagy, Z. K., Chew J. W., Braatz, R. D. (2005). First-Principles and Direct Design Approaches for the Control of Pharmaceutical Crystallization. Journal of Process Control, 15, 493-504.

Control of Biopharmaceutical Crystallization

Direct design approach• Feedback control to follow a

setpoint supersaturation curve in the experimentally determined metastable zone

• Concentration control (C-control)

–Manipulating the temperature (cooling crystallization)

–Solvent addition (antisolventcrystallization)

Fujiwara, M., Nagy, Z. K., Chew J. W., Braatz, R. D. (2005). First-Principles and Direct Design Approaches for the Control of Pharmaceutical Crystallization. Journal of Process Control, 15, 493-504.

Documents

Challenges and Opportunities in Biopharmaceutical ...focapo-cpc.org/pdf/Braatz.pdf · Challenges and Opportunities in Biopharmaceutical Manufacturing Control Moo Sun Hong, ... Pharmaprojects