MANCHF$TER 1824

Enabling rapid liquid and freeze-dried formulation design for the manufacture and

delivery of novel biopharmaceuticals

Paul Dalby Dept. Biochemical Engineering, UCL

Robin Curtis & Jim Warwicker University of Manchester

~cu ·--~:o ~..c ·-U cc

MANCHF$TER 1824 Introduction

~~ • Biological medicines account for >25o/o of all new drug approvals: 8 of the top 10 selling drugs .c.

0

• Biopharmaceuticals market is rapidly growing with reported sales of £197 billion in 2016 (compared with total drug market of £816 billion)

• Next generation therapies are increasingly complex and engineered for biological activity at the expense of physical and chemical stability (eg protein fusions, fragments, conjugates with small drug molecules)

• Formulation development of biopharmaceuticals:

1. Dosage formulations fixed quickly - in time for clinical trials. Not much material is available. Shelf life over 2 years not known until mid-way through clinical trials.

2. Formulations require stability, potency, and ease of delivery to patient

3. Many therapeutics require high concentrations which leads to increased physical degradation, poor rheological properties, and phase separation

MANCHF$TER 1824 Protein aggregation

• Predicting and controlling aggregation is an outstanding challenge: 1. Key intermediates are transient, have low populations, and are difficult to isolate I study 2. Multiple mechanisms for aggregation, depend on protein and environment (solvent properties, temperature)

N u f(pH IS T)

• Rapid experimental screens are too indirect: 1. Unfolding temperature or free energy, colloidal stability (eg aggregation temperatures and protein-protein

interaction measurements) 2. Accelerated (eg. high T) aggregation assumes Arhenius-type extrapolations

Aim1: Understand factors affecting aggregation in formulation

?p - ~?

)? ? E:>

Local Dynamics I Partial

unfolding

Global Conformational

Stability

3° I 4° conformation

Surface properties: charge/

hydrophobicity

Sequence features eg. APRs

Excipient crowding

Aggregation Excipient binding

Fragmentation

Solubility Viscosity & diffusion

MANCHF$TER 1824

Aim2: How can we predict better formulations?

Do conformational and colloidal stabilities correlate to aggregation rates?

Does forced degradation at high temperature predict shelf-life?

Can alternative methods be developed for predicting aggregation rates?

Aim3: How can we engineer based on predictions?

Can we engineer lower aggregation rates?

Can we develop novel (GRAS-based) excipients?

MANCHF$TER 1824 Objectives

01. Use high-throughput automation to generate a large experimental formulation dataset for protein:excipient combinations, that will include aggregation kinetics, conformational stability, colloidal stability, phase behaviour, and rheology measurements.

02. Molecular informatics and modelling will improve predictability of formulation attributes and excipient effects

03. Analytical advances will enable earlier, more sensitive, and lower-volume assessments of formulated protein degradation kinetics.

Heat maps of Fab aggregation kinetics at 4-65 °C

400

- 300 ::2: .s (/) - 200

100

0 4 6

pH pH 500 500 S' 1500

~ ~ ~

v(%.day E ~

400 400 c ~· 0.0 0 ~ N i: N 1000

0.1 - 300 1000

(IJ (..)

::2: c .s Cll .,

..Cl 0.3 2000 0

.. ,. (/) - 200 CJ)

..Cl 500 <( 0.4 3000

100 100

0.5 4000

0 4 5 6 8 9 4 5 6 7 8 9 3.0 3.5

Kinetics at low Tine often dont correlate with melting temperature

Where Tine<< Tm• fraction unfolded is <<0.0001:

Global unfolding (and hence Tm) is not relevant

Native ensemble dynamics & colloidal stability control aggregation kinetics.

u

u·

15

10

........ 'l""'I

I

> 5 ta

"'C

~ -........ 0 ::iii -c

• Fab 4C pH3.5-9

• Fab 23C pH3.5-9

• Fab 45C pH3.5-9

• Fab 65C pH3.5-9

>99.99o/o folded

o GCSF variants pH4/8, 37C

• • • • • • • • • •

: ! ~ : ,-~ = ~ G> @ : ,f'

! ~ .- , • Fab variants pH 4, 65C

• TK variants, pH 7, 60C

•\e . ! .. ' • '" :

-5 ' ' ' ~. i:. '. " '4 • : . •t ...... -c ~ • • : · • ·r· · · ,. · · ·;· · · t · · · · · · · · · · · · · · · · · · 12% -~~~~ · · · · -10

I : 2yrs

Zhang et al. (2018) Molecular Pharmaceutics. 15, 3079-3092 1 E l6 Robinson et al. (2018) Molecular Pharmaceutics. 15, 256-267. · - 1.E-13 1.E-10 1.E-07 1.E-04 1.E-01 Chakroun et al. (2016) Molecular Pharmaceutics. 13, 307-319. See also Roberts (2013) - review on non-Arrhenius protein aggregation. f 1

Small-angle x-ray scattering of Fab under native conditions

-e ' -i:n B.:

2.7

2.6

pH-dependent conformation

3 s s 1 a 9 pH

Aggregation kinetics correlate

T --1

--"O

+ -:>

- .5

2.60 2.65 2.70

- g (nm)

• Conformational change with pH correlates with aggregation kinetics, at 23 °C

Godina at al. 2019 JMB

Molecular dynamics simulation for Fab

Equilibrium RMSF {300K) pH?, 25°C, 50ns, 50mM IS pH3.5, 25°C, 50ns, 50 mM IS OPLS-AA/L force field & SPC/E water Triplicated

pH 7 pH 3.5

CL domain displacement

Godina at al. 2019 JMB

Which molecular dynamics simulation structures explain SAXS?

A Light Chain

pH 7.0 pH 5.5

• PH 3.5

Back view

*' ----

I

B Heavy Chain

Side view

CL domain displacement

Godina at al. 2019 JMB

~~-

Single-molecule fluorescence

smFRET analysis of pH-dependent Fab conformations

A smFRET

Dis 1

00

00 05 10 pp fi nl FR T

00 p 35

300 !.! 097 ~ 200

100

0 00 05 10

pp t n FE

F

0 H 70

00

00

300 0.78 c 200

8 00

0 10

Dist 1: LC-K126pAzF + LC-S156C: CL to CL domain Dist 2: HC-S117pAzF + LC-S156C: CL to HC linker

SAXS

70

0

0 -2.5 .s 3,5

3,0

35 2.5

2.5 :t 0.1

smFRET, SAXS and MD reveal the same dynamics and conformational shift with pH Godina at al. 2019 JMB

Best-fit SAXS structures reveal APR exposure at low pH

( )

(a} (b} pH 7.0

(b)

on

(c) pH 3.5

Godina at al. 2019 JMB

Protein engineering and formulation Activity '·00 --,,-,--,-.,-~

Activity

100

0

3.75

{ "" e.... <( 2.50

(/)

I

di

1.25

OOO

0.00 0.13 0.25 0.38 0.50

A: Tween (%v/v)

MD-guided protein engineering to slow aggregation

180"

D H35

Designed 12 stabilising and 5 destabilising mutations using ROSETTA at flexible sites.

Tm did not increase, but ~Svh increased. Aggregation slowed where ASvh increased. ~~~~n~~

LC-154

\

HC-227/228

\

/ HC-134 to 136 I

HC-194 Zhang et al. (2018) Computational-design to reduce conformational flexibility and aggregation rates of an antibody Fab fragment. Molecular Pharmaceutics. 15, 3079-3092

´·¹¸

¬ ­½

¿¬¬

»®·

²¹ ·

²¬»

²­·

¤

Iv1ANCH E51 ER 182..t STPP for tuning protein phase behaviour

Increasing STPP concentration

•

I

I I I I• I I

• I I •

sodium tripolyphosphate (STPP)

Region I

s t

•

•

Region Ill •

• Zeta Potential

Solubility

•

Region I soluble protein

Region II protein precipitate

I • Region Ill D 5 10 1S

protein re-solubilized SlPP conc _n _r _ 'tlon (m1M)1

Lysozyme, pH 8

• Reentrant condensation phenomena observed with positively charged proteins and STPP (or ATP)

• Initial [STPP] causes protein precipitation through forming ion bridges across proteins

• Resolubilization at higher [STPP] due to overcharging protein

• STPP can direct formation of reversible colloidal gels with glassy dynamics for formulating proteins

Na

5

1VlANCH F.s1 El{. IH24 0 ''''"'""'""'''"'''"''''"''''"'''''"'''"''''"''''"''''''''''"'''''''''''''' '

RED: NaCl , GREEN: ATP

ovalbumin, pH 7 > E -5 -PURPLE: STPP

• 0.0012 "' ~ -10

- i · QJ .... 0 c.. & -15 N

llO ........

0 .0008 I I 0

E ...J

E -N N

CQ

1 ! I

0 .0004

i

!

•

-20+-------------....------.._..-0 25 50 75 100

T

•

Ionic Strength (mM} 125

T

• 0.0000 .... --.-..... --.... --...... --......... --.... --...... ~ .........

0 20 40 60 80 100

Ionic Strength (m M)

• ATP and TPP prevent thermal-induced aggregation of negatively charged proteins.

• Stabilization is due to reduction in aggregate growth rates through electrostatic stabilization

ATP and STPP prevent protein aggregate growth

• Multivalent anions ATP and TPP supercharge negatively charged proteins through ion binding

• Supercharging protein with ATP or TPP increases protein-protein repulsion and colloidal stability as reflected by increase in 8 22 values at fixed IS

BLACK: No Salt, RED: NaCl , GREEN: ATP PURPLE: STPP

E 500 ovalbumin, pH 7, • E 200 c , HSA, pH 7, - 10 mM salts • ..s 10 mM salts Ill 400 • • :J Ill

:J 150 "'C

"' . , "'CJ er: 300 "' ' a:: .!:! .!:! E . , E 100

"' 200 • • c "' > II c "'C > 0 100 "'CJ 50 .... • 0

"'C .... > "'CJ :c >

0 :c 0 60 65 70 75 80 50 60 70

Temperature (0 C) Temperature (0 C)

• •

••

1V1ANCH Esi EI~ 182..t

Polyvalent anion binding and solubility A model (not a web tool), following on from Bye & Curtis:

Controlling Phase Separation of Lysozyme with Polyvalent Anions ] rdan " d in A urti *

60,----------------,

+10 -10 40

w c '§ 0 e c.

" "' ~ -20

'° u

-40

-60+---~~-~~~-------<

- 60 - 40 - 20 0 20 40 60 Exper imental zeta-potential/ mV

model for anion binding fits measured zeta potl

Increased polyanion concentration

Protein net-charge reversal

model applied to human proteins suggests that over-charging and potential solubilisation is general

Bye and Curtis (2019) J Phys Chem B 123:593; Kalayan et al (2020) Mai Pharm 17:595.

1V1ANCH E51 ER 182..t Solubility prediction server

Sequence Prediction The protein-sol software will take a single amino acid sequence and return the result of a set of solubility prediction calculations, compared to a solubility database.

Please enter a single sequence of single letter amino acid codes in the FASTA format.

For example > P00547

MVKVYAPASSANMSVGFDVLGAAVTPVDGALLGDVVTVEAAETFSLNNLGRFADKLPSEPRENI VYQCWERFCQELGKQI PVAMTLEKNMP I GSGLGSSACSVVAALMAMNEHCGKPLNDTRLLALMGELEGRISGSIHYDNVAPCFLGGMQLMIEENDI T c:nnuor.:vni:cwT WUT.n VP!.: T KUC:'T'n v n Rn TT .Pnnvoonnr Tn J:.lf.:Rl.n .nr.:vT wnrvc:RnPVT. n n KT .MKnUT n TCPVRPRT.T.P

Originally (2017): Sequence-based solublity prediction, based E. colidata Currently (2020), also: Patches, Heatmap, Abpred, pKa In development (2020): Excipient predictions (for next meeting)

(2017) Bioinformatics (2017) 33:3098; ; PNAS (2009) 106:4201

1V1ANCH E51 ER 182..t

Patches: with Fab fragment non-polar

2.3

1.45

0.6

polar

polarity

A positive 7SmV

OmV

-7SmV

negative

charge

Hebditch and Warwicker (2019) Sci Rep 9: 1969

B

1V1ANCH E51 ER 182..t

0.3

0.25

0.2

0.15

0.1

0.05

0.0

B 2.0

0.3

0.25

0.2

0.15

0.1

0.05

0.0

D 2.0

Heatmap: pH and ionic strength ~100

1~0 Predicted pH and ionic 100 strength dependence of folded ~ state stability {upper panel), 0 J/aa and net charge {lower panel). -~

- 100 Upperpanelshows'phase - 1~ diagram' fit to experimental

data. 2.5 3.0 3.5 4 .0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 """"*

pH ~20

Method can be used to identify 0

regions that lower stability in 0\0

conditions such as a low pH of o~

bioprocessing. uo e/a These regions can then be -00) engineered out, e.g. switching

Asp/ Glu for Asn/Gln.

2.5 3.0 3.5 4 .0 4 .5 5.0 5:-5 6.0 6.5 7.0 7.5 8.0 pH

Hebditch and Warwicker (2019) Sci Rep 9:1969; Hebditch et al (2020) Comp Struct Biol J 18:897

1V1ANCH E51 ER 182..t

AbPred: ML models from mAb137

l !O

s

120

-0 I -5 Q)

t :0 t 0 ~

0...

90

A machine learning model for each of the 12 methods, this example is prediction of HIC {orange), against the background {green). Models predict hydrophobicity {HIC etc, below) a problem for this mAb.

97b95a7 Olef46419fl5 Ar;iproved

Phase 3 Phase 2

mAb137 is a dataset of 12 biophysical measurements {HIC etc) for 137 mAbs {Jain et al 2017)

Experimental value (min)

problem no problem {predicted) ETA lllC S ·IAC SGAC CIC CSI AC HE PSR ELISA BVP DSF AS

Hebditch and Warwicker (2019) PeerJ 7:e8199; Jain et al (2017) PNAS: 114:944.

Take-home messages

Aim1: Understand factors affecting aggregation in formulation ~t

- local dynamics/unfolding - exposure of aggregation hotspots (APRs) - colloidal stability (net charge) - Optimize formulations to increase Tm, decrease dynamics and increase net charge

Aim2: How can we predict better formulations?

- Local dynamics and aggregation hotspots can be predicted computationally - Excipient interactions can be predicted by molecular docking or LCMS - Net charge and effect of charged excipient binding can be predicted computationally - see webserver at protein-sol.manchester.ac.uk

Aim3: How can we engineer based on predictions?

- Mutations that suppress dynamics sometimes decrease aggregation kinetics - Novel (GRAS-based) excipients based on dipeptides, and supercharging with STPP

§111 ~.c -u cc: ':J Q.J~

.r: t-o

l\ilANCHFs1 f:R 1~2.;

UG/PhD/PDRA: Yuki Abe Lucy Rayner Nesrine Chakroun Teresa Barata Cheng Zhang David Hilton Mathew Robinson Victoria Wood Akash Pandya Nuria Codina Maariyah Samad

Collaborators: SAXS &AUC GCSF Mass Spee Fabs lgGs Molecular docking

Acknowledgements

where are they now? - Biopharm Services - Lonza - Merus (NL) - IPSEN - PORA UCL -GSK - Lonza -GSK - PhD UCL - NeoLeukin (Seattle, USA) - 04 Research

Steve Perkins (UCL) Paul Matejtschuk - NIBSC Milena Quaglia - LGC UCB Pharma & Darren Nesbeth (UCL) Lonza, UCB, Medlmmune, FujiDSB Mire Zloh (Univ Herts I UCL)

EPSRC EP/1033270/1 : CIM EP/L015218/1 : CDT EP/K005030/1 : Equipment EP/M028100/1 : Equipment EP/N025105/1 : Formulation EP/P006485/1 : FTHM Hub

MANCHE'.'II ER ll"24

•

•

•

•

•

•

BBSRC/BRIC/Royal Academy of Engineering (funding) James Austerberry, Mariam Nuhu, Rose Keeling, Rana Dajani, Dorota Roberts, Luke Holloway Jim Warwicker, Jeremy Derrick, Alexander Golovanov Chris van der Walle, Shahid Uddin and the formulation development group at Medlmmune Chris Roberts, Jai Pathak, Prasad Sarangapani Wyatt Technology, Unchained Labs (Avacta)

PDRAs on EPSRC Formulation project • John Hales - UCL

• Cheng Zhang- UCL • • •

Samir Aoudjane - UCL Jordan Bye - University of Manchester Max Hebditch - University of Manchester

EPSRC CDT PhD students aligned to Formulation project • Nuria Codina - UCL • Akash Pandya - UCL • •

Nikita Vekaria - University of Manchester Jas Kalayan - University of Manchester

BRIC PhD: BB/J003824/1 BRIC: BB/1017119/1 BRIC: BB/K011162/1 BRIC: BB/E005942/1