“Development application and single use probes for building Design Space of live cell concentration by dielectric spectroscopy”

BY: JOHN CARVELL

ABER INSTRUMENTS, Wales, UK

ABER Instruments Ltd – originally a spin out company from the University of ABERystwyth, Wales.

Wales also has the longest station name in the world!

Fortunately it is shortened to Llanfair PG

Scope of Talk

• Principle of Dielectric Spectroscopy (DS) –emphasis on scanning

• Correlation with offline readings

• Case study Gedeon Richter

• Case study GSK

• Case study Fujifilm Diosynth

• Single use probe applications

• Conclusions

Industry Initiative

“Systems that promote greater product and process understanding can provide a high assurance of quality on every batch and provide alternative, effective mechanisms to demonstrate validation” FDA Guidance for Industry: PAT—A Framework for Innovative Pharmaceutical Manufacturing and Quality Assurance, 2004.

Four types of tools for generation and application of process understanding:

1. Multivariate tools for design, data acquisition and analysis

2. Process Analyzers

3. Process Control tools

4. Continuous improvements and knowledge management tools

PAT perspective

Principle of Dielectric Spectroscopy (DS)

•When an alternating electrical field is applied to an ionic solution of cells, a charge separation or polarization effect develops at the cell membrane

•Cells behave like a capacitor under an electric field

•The magnitude of this charge separation can then be measured by the capacitance of the solution at varying radio-frequencies in the range 0.1-20MHz

•The resulting capacitance is generally proportional to the total enclosed bio-volume, which itself is directly correlated with cell density or cell weight

Dielectric Spectroscopy (DS) in bioreactors

Dielectric Spectroscopy Key Outputs: • Capacitance Data Over Time • Correlates with Viable Cell Density

++ +

++

++

+ +

+

+ ++

++

+

- - -

--

-

---

-

--

- --

-

++ +

+

++ +

+

-- - -

---

-

DS Probe

DS Probe

Cell Populations

Single Cell

Bioreactor

Aber DS Probe

Population of Cells

Single Cell

0

5

10

15

20

25

0 50 100 150 200 250 300

Pe

rmit

tivit

y (

or

Ca

pa

cit

an

ce

)

Time (hrs)

Total Permittivity (or Capacitance) vs. Time

+

+

+ +

+ +

- - - -

- - - - -

+ + +

0.E+00

1.E-09

2.E-09

3.E-09

4.E-09

1.E+05 1.E+06 1.E+07

Pe

rmit

tivi

ty (

F/m

) αC

apac

itan

ce

Frequency (Hz)

Capacitace/Permittivity vs Frequency

Day 3-66 hr

Day 6-137 hr

Day 8- 194 hr

Day 11-252 hr

Dielectric Spectroscopy Key Outputs: • Frequency Scanning Data Over Time • Tells us about Cell Populations

Graphics courtesy of Bend Research,Oregon

Correlation of online capacitance data with offline readings

• Plot of capacitance at specific single frequency versus VCD should produce straight line correlation with an R2 value close to 1

• Correlation during log phase of cell growth is well maintained

• Divergence observed as cells enter stationary/death phase at which point cell viability decreases and LDH levels increase proportionally

• Published data on models for correcting this divergence available eg Bend, UMASS, University of Manitoba

Correlation of Viable Cell Density v Capacitance

Viable Cell Density

0 2 4 6 8 10 12 14 16

Cap

acita

nce

0

5

10

15

20

25

R2 = 0.9787

CHO Cell Density Monitoring in Bioreactor Culture – Comparison of four methods (data courtesy) University of Manitoba )

10

Fed-Batch HEK cells (data courtesy of CNRC, Canada)

0.6 MHz 2.2 MHz 10 MHz

Change in shape of Spectrum- fundamentals of frequency scanning

101

102

103

104

105

-0.2

0

0.2

0.4

0.6

0.8

1

Frequency (kHz)

Norm

aliz

ed C

apacitance

Timeseries of Normalized Capacitance vs. Frequency Timeseries of Normalized Capacitance vs. Frequency Timeseries of Normalized Capacitance vs. Frequency

No

rma

lize

d C

ap

acita

nce

No

rma

lize

d C

ap

acita

nce

Frequency (kHz) Frequency (kHz)

Curves move from

Blue – 0hrs

Red – 72 hrs

Control CHO cell undergoing induced apoptosis

Case Study 1

Gedeon Richter, Hungary

Contact Dénes Zalai zalaid@richter.hu

Details of study at Gedeon Richter

• 6 runs with 1L bioreactors and ATF2 technology (Refine,USA) with CHO expressing Mab

• High cell densities up to 75 x10E6 cells/ml

• Utilized Aber Biomass Monitor 220 with AberScan software

• Scans at 25 pre-defined frequencies every 8 minutes

• Spectral data exported for linear regression or multivariate data analysis

• Cells/ml derived from Aber using 3 different methods:

- Simple measurement at 580KHz (linear modelling)

- Cole- Cole

- PLS using Umetrics SIMCA software

• at-line measured VCD values throughout Run 1-4

were plotted against predicted VCD values and

depicted as triangles

• linear equation for prediction:

VCD = 414492 × C580 + 1.187 × 106

• linear model over-estimated VCD at the later phases

of the cultivation

Linear Modelling at 580KHz

Deriving VCD using Cole-Cole using frequency scanning between 100 and 20,000KHz

•

Off-line cell counts (VCD) for 6 runs with REFINE system

Score plots for 6 runs

• Multivariate analysis (PCA) of capacitance spectra resulted in detection of measurement disturbances

• Principle component t(1) captured 92% of variance in spectra and showed same trend for all 6 runs

• DIELECTRIC SPECTRA NOT INFLUENCED BY BATCH TO BATCH DIFFERENCES OF CRITICAL CULTIVATION PARAMETERS INDICATING ROBUSTNESS OF probe

• VCD prediction with multivariate PLS model developed

CHO with continuous feed-Comparison of 3 methods to predict VCD

0

200

400

600

800

1000

1200

0.0E+00

1.0E+07

2.0E+07

3.0E+07

4.0E+07

5.0E+07

6.0E+07

7.0E+07

0 2 4 6 8 10 12 14

fc [

kH

z]

VC

D [

ce

lls

/ml]

Process time [day]

VCD prediction Run2 VCD at-line

linear (C580)

Cole-cole

PLS

fc

CHO with continuous feed-Comparison of 3 methods to predict VCD

0

200

400

600

800

1000

1200

0.0E+00

1.0E+07

2.0E+07

3.0E+07

4.0E+07

5.0E+07

6.0E+07

7.0E+07

8.0E+07

0 2 4 6 8 10 12 14

fc [

kH

z]

VC

D [

ce

lls

/ml]

Process time [day]

VCD prediction Run1 VCD at-line

linear (C580)

Cole-cole

PLS

fc

18 February 2014 Dielectric Spectroscopy and its

application in process development

Dielectric Spectroscopy and its application in process development Andrew Heinrich Biopharm R&D, GSK Stevenage

PAT & QbD Forum, Feb 18th

Goettingen, Germany

Case study 2

GSK, Stevenage, UK

Use of capacitance in scale-up and technology transfer

• Background

– Cytokine produced in modified CHO cells

– Initial cell line selection carried out in shake flask studies

– Process scoping in 2L bioreactor system

• In-situ Aber FUTURA Probe used in 2L glass bioreactors with multiple frequency scanning

• What information can we glean from this data?

– Relatively similar process profiles

– Perturbations in capacitance profiles appear to correspond with feed points in process

– Can this data be used to help build a ‘process model’ for scale up (and scale-down)?

Single frequency measurements for various bioreactor runs

Run Time Hrs

0 50 100 150 200 250 300 350

Me

asu

red

Ca

pa

cita

nce

0

10

20

30

40

50

Scanning frequency analysis Curve fitting the data

• GSK developed an XL-fit macro for rapid and reproducible automatic analysis of scanning data sets

Use of capacitance in scale-up and tech transfer

• Analysis of various scanning data parameters highlights clearer differences between some of the processes

• Using this information one can eliminate runs that appeared previously to be consistent and therefore make more informed conclusions of the ‘predictability’ of the process conditions

• Further processing of data is still required to fully analyse these results and gain a greater confidence in the ‘process model’

Biovolume profiles for Bioreactor runs

Run Time Hrs

0 100 200 300 400

Bio

vo

lum

e P

0

10000

20000

30000

40000

50000

Multivariate analysis

• PCA modelling can then be used to further interpret the key components and specific characteristics of individual batches

• Pre-processing of the raw data is critical in this sort of analysis in order to ensure that inferences from the statistical tools are ‘genuine’

• This analysis then offers a method for investigating design space and process ‘limits’ suitable for development and scale up

Understanding the data

18 February 2014 Dielectric Spectroscopy and its

application in process development

Using DS in scale up/scale down experiments

Using conventional probes volumes

down to 100ml can be monitored

Samples down to 1ml can be taken from shake flasks for DS using an inverted probe

Scale up data from GSK

• Capacitance as a scale parameter from shake flask to 50L single use bioreactor

• Inverted probe technology from Aber Instruments provides real-time offline capacitance readings that correlate well with online data

Technology comparison

Correlation of Off-line v In-situ Capacitance probes at scale

Run Time Hrs

0 100 200 300 400

Measure

d C

apacita

nce

0

5

10

15

20

25

Via

ble

Ce

ll D

ensity

0

2

4

6

8

10

12

2L Bioreactor

50L SUB

Offline Probe (1L S/Flask)

Viable Cell Density (Flask)

Case study 3

Fujifilm Diosynth, UK

y = 0.663x + 0.4493R² = 0.9949

0

2

4

6

8

10

12

0 2 4 6 8 10 12 14 16 18

Via

ble

ce

ll C

on

cen

trat

ion

(x 1

06 )

ce

ll/m

L

Capacitance (pF)

Conventional Calibration showing online cell diameter from ABER SCADA software

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

0

2

4

6

8

10

12

14

16

0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00 18.00 20.00

Ce

ll D

iam

ete

r (A

rbit

rary

un

its)

Via

ble

Ce

ll C

on

cen

trat

ion

(x

10

6/

mL

Elapsed Time (Days)

Online VCC

Offline VCC

Cell size

Offline vs on line Viable cell concentrationusing Aber calibration method

• PCA analysis of Raw B-dispersion Spectra from CHO Mammalian cell culture

• P1 – Strong Biomass concentration signal

• PC2 and 3 Remaining Orthogonal variance in signal

Principal Component Analysis Raw scores plot of B-Dispersion Spectra

-10

-8

-6

-4

-2

0

2

4

6

8

-40

-30

-20

-10

0

10

20

30

40

0 2 4 6 8 10 12 14 16 18 20

PC

2 a

nd

PC

3 S

core

s

PC

1 S

core

Elapsed time (days)

P1

P2

P3

VCD- Variable importance plots vs Frequency

32

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

1.9

2.0

2.1

2.2

2.3

Capaci5

0C

apaci6

4C

apaci8

2C

apaci1

06

Capaci1

36

Capaci1

74

Capaci2

24

Capaci2

87

Capaci3

68

Capaci4

73

Capaci6

07

Capaci7

79

Capaci1

000

Capaci1

284

Capaci1

648

Capaci2

115

Capaci2

714

Capaci3

484

Capaci4

472

Capaci5

740

Capaci7

368

Capaci9

457

Capaci1

2139

Capaci1

5582

Capaci2

0000

Conduc50

Conduc64

Conduc82

Conduc106

Conduc136

Conduc174

Conduc224

Conduc287

Conduc368

Conduc473

Conduc607

Conduc779

Conduc1000

Conduc1284

Conduc1648

Conduc2115

Conduc2714

Conduc3484

Conduc4472

Conduc5740

Conduc7368

Conduc9457

Conduc12139

Conduc15582

Conduc20000

VIP

[3]

Var ID (Primary)

131216 PLS predictionset.M2 (PLS), Viable cells only

VIP[Comp. 3]

SIMCA-P+ 11.5 - 09/01/2014 20:28:52

Capacitance 50-20K Hz

For VCD the lower

frequencies correlate more

strongly than the higher

frequencies as expected

Viability-Variable importance plots vs Frequency

33

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

1.9

2.0

2.1

Capaci5

0C

apaci6

4C

apaci8

2C

apaci1

06

Capaci1

36

Capaci1

74

Capaci2

24

Capaci2

87

Capaci3

68

Capaci4

73

Capaci6

07

Capaci7

79

Capaci1

000

Capaci1

284

Capaci1

648

Capaci2

115

Capaci2

714

Capaci3

484

Capaci4

472

Capaci5

740

Capaci7

368

Capaci9

457

Capaci1

213

Capaci1

558

Capaci2

000

Conduc50

Conduc64

Conduc82

Conduc106

Conduc136

Conduc174

Conduc224

Conduc287

Conduc368

Conduc473

Conduc607

Conduc779

Conduc1000

Conduc1284

Conduc1648

Conduc2115

Conduc2714

Conduc3484

Conduc4472

Conduc5740

Conduc7368

Conduc9457

Conduc1213

Conduc1558

Conduc2000

VIP

[6]

Var ID (Primary)

131216 PLS predictionset.M3 (PLS), % Viability Only

VIP[Comp. 6]

SIMCA-P+ 11.5 - 09/01/2014 20:29:14

-Two peaks in the VIP plot vs

frequency with a peak in the

mid range.

-change in the shape of the

peak in the mid range over

time probably due to a shift in

the cell size distribution.

Capacitance 50-20K Hz

PLS Calibration

0

20

40

60

80

100

120

0

2

4

6

8

10

12

14

16

0 2 4 6 8 10 12 14 16 18 20

% C

ulture

Via

bili

ty

Cell

Concentr

ation

(x 1

0^6

cells

/mL)

Elapsed Time (Days)

Viable cell Conc prediction Total Cell Conc Prediction "offline VCC"

% Viability Prediction "offline % Viability"

Single use applications

Results from Aber single use probe in a Sartorius rocking bag

0 48 96 144 192 240 288

0

10

20

30

40

50

60

70

0

50

100

150

200

250

300

0

1

2

3

4

5

6

7

8

5

10

15

20

Ca

pa

cita

nce

[p

F/c

m]

Time [h]

Ce

ll D

ensity [

10^5/m

l]

Via

ble

Ce

ll V

olu

me

[%

]

Ce

ll D

iam

ete

r [µ

m]

Flow past cells for use on external loops with silicone tubing

First version of Flow past cell will be tested at Shire this month

Conclusions

• Spectral scanning of capacitance becoming increasingly used in PAT

• Software based on Cole-Cole or PLS can be used to correct VCD during death phase ,derive viability and in scale up studies

• Promising technology for studies on apoptosis on certain cell lines

• Development of single use probes has enabled technology to be applied to single use bioreactors