Transcript ceiling ss 2007

1
Second and third generation therapeutic and diagnostic proteins
Directed evolution of new proteins
Connect phenotype
(typically a binding)
to genotype
(typically DNA)
Membrane display (Rice paper)
Yeast surface display
E. coli surface display
Phage display
Ribosome display
Displayer
size
Number
High
Low
Low
High
Screening for natural antibodies using yeast
Aga2 becomes
associated with Aga1 in
the cell.
Aga1 becomes
covalently anchored to
the cell wall
Nature Biotechnology 21, 163 - 170 (2003)
Flow-cytometric isolation of human antibodies from
a nonimmune Saccharomyces cerevisiae surface
display library
Michael J. Feldhaus, et al. and K. Dane Wittrup
yeast
FACS
2
First purification to get numbers down to <
108
for FACS: magnetic (20 nm) microbeads
Avidin-coated
magnetic bead
μ
Av
Biotin-
B
Bead-labeled cells retained on
magnetic column
www.miltenyibiotec.com
3
4
Created a general scFv library from 58
people using PCR primers from various
classes of Igs.
Cloned in E. coli to maintain complexity of
109.
Cloned into yeast for surface expression
maintaining complexity
Screen for antigen binding by 2-steps:
Ag label
1) Too many cells for initial FACS, so:
Magnetic bead capture (avidin
micromagnetic beads [20 nm], bind via
biotinylated antigen) isolate on
column
2) FACS
Found Abs vs. many antigen tried.:
EGF, EGFR, p53 peptides, fluorescein, etc.
Kd ~ nM range, often.
5
Isolation of engineered, fulllength antibodies from libraries
expressed in Escherichia coli.
Mazor, Y., Van Blarcom, T.,
Mabry, R., Iverson, B.L., and
Georgiou, G. 2007. Nat
Biotechnol 25: 563-565.
Outer
memb.,
permeabilized
Fortunately,
100 pM avidity,
10x affinity.
Divalency?
Periplasm
Inner
memb.
Immunize mice  spleen cells
 mRNA  PCR VH and VL
cDNAs
Fluorescent
 reconstruct full length H&L
antigen
7
chains in one plasmid. Clone 10
ProteinA-E.coli
 E. coli with ProteinA/memb
anchor fusion protein (“ZZ”)  memb. anchor
fusion protein
displayed outside inner
membrane  lysozyme 
spheroplasts 
cytoplasm
add fluorescent antigen 
FACS  clone E. coli cells,
Yields regular Abs, no fusion protein.
rescue gene
6
Recovery of Ag-speific IgGs
On time Off time
PA = B. anthacis
protective antigen
7
Phage display for a human antibody
RT-PCR the rearranged VH and VL cDNAs
Reassemble full antibody molecules
Fully Synthetic Human Combinatorial Antibody Libraries (HuCAL) Based on
Modular Consensus Frameworks and CDRs Randomized with Trinucleotides
Knappik et al, and Andreas Pluckthun and Bernhard Virnekas
J. Mol. Biol., 296: 57-86 (2000)
VL
VH
CDR3 targeted here
CDRs bkgds colored
Framework gray
8
9
Starting material here = HuCAL:
Fully synthetic human combinatorial antibody library based on modular
consensus frameworks and CDRs randomized with trinucleotides.
These are single chain antibodies (scFv’s).
Framework consensus sequences of 7 of heavy chain V-regions
x 7 framework consensuses of light chain V-regions
= 49 combinations
Plus: randomized CDR3’s by inserting synthetic oligos beween restriction
sites. CDR3 = site of VJ (light chain) and VDJ (heavy chain) joinings.
CDR = complementarity determining region
Knappik et al, JMB, 296: 57-86 (2000)
10
Found 95% of human Abs are represented by 7 VH and 7 VL exons
Constructed: Human Combinatorial Antibody Library (HuCAL) as scFv’s
E. coli (2 x 109 recombinants [!])
Made all 49 (7x7) combinations of framework regions [but not represented equally]
Mutagenized with triplets to put in all AA substitutions into a run of 6 spots in CDR3
Simulated the AA frequencies found in natural Abs.
} cf.
Fully Synthetic Human Combinatorial Antibody Libraries (HuCAL) Based on
Modular Consensus Frameworks and CDRs Randomized with Trinucleotides
11
Knappik et al, and Andreas Pluckthun and Bernhard Virnekas
J. Mol. Biol., 296: 57-86 (2000)
VL
VH
CDR3 targeted here
CDRs bkgds colored
Framework gray
12
Different amino acid combinations achieved after insertion of synthetic oligonucleotides.
Planned Found
Figure 7. Comparison between design and
experimental composition of CDR3 libraries used.
For each position of the CDR3 region (numbering
according to Kabat et al., 1991; for HCDR3 the
position before H101 is numbered 100z, the length
variable region is numbered from H95 to H100s),
the amino acid composition in the planned libraries
(P, left columns) is compared with the composition
found from sequencing 257 clones of the initial
libraries (F, right columns). The TRIM mixture
indicates the mixtures of trinucleotides used in the AAs
oligonucleotide synthesis (see Table 3 of the
Supplementary Material). Occupied indicates the
number of amino acids encoded by the respective
mixture and found in the sequenced clones,
respectively.
There’s a similar table for VL
Knappik et al, JMB, 296: 57-86 (2000)
13
From Biacore:
Kd (nM)
on time (t1/2) off time (t1/2)
*
* From subsequent mutagenesis and ribosome diplay
14
Nature Biotechnology (2000) 18:1287
1
2
6
3
5
4
imer
ust
rry T7
omoter
15
Start here:
Library in DNA form
via an E. coli S30 extract
Insulin (target)
16
Ribosome display
Absence of a stop codon is necessary to prevent dissociation of the newly synthesized
protein from the ribosome.
biotin
antigen
RT-PCR
product of
affinity
captured
material
Affinity selection in rabbit reticulocyte ribosome display is antigen-dependent
and requires the absence of a stop codon to keep the protein-ribosome-mRNA
complex together.
In the presence of a stop codon, the protein synthetic complex dissociates.
From Hanes et al., FEBS Letter, 450, 105-110, 1999)
17
Prokaryotic
5’
stem- ribosome
T7 promoter
loop binding site
scFv
3’
stem3’ AA spacer loop
DNA
mRNA
DNA construct assembled by ligation and PCR: no cloning in E. coli
Here: 2 x 109 molecules.
Stem-loops protect frm degradation
Later conservative estimate of no. of molecules that can be screened:
2.6 × 1011 per ml of reaction (Nature Methods 4:269 (2007))
scFv: single chain Fragment of variable chain (VH + VL
18
VL
VH
CDR3 targeted here
CDRs bkgds colored
Framework gray
19
Back to:
Nature Biotechnology (2000) 18:1287
Primer
must
carry T7
promoter
1
2
via an E. coli S30 extract
6
3
5
4
Insulin (target)
Anti-insulin?
B= biotin
B
Avidin bead
20
120
counts per minute
Specific insulin binding activity (gray-white
bars) was evident over a smaller amount of
non-specific (or different-specific) binding, as
assayed by binding of 35S-labeled translated
protein in the translational complex.
100
80
60
40
Specific
Non-specific
50-50 mix
20
0
0
50
100
150
200
nM cold insulin
35S-labeled anti-insulin
“Cold” (unlabeled) anti-insulin
Insulin
B
B
B
B
B
B
B
dish
Radioimmune assay (RIA)
B
21
Plenty of framework mutations introduced by the PCR (i.e., not in the HuCAL library)
Varied in
the
HuCAL
library
(CD3)
Length
variations
22
7x7 grid
Only a 8 of the 49 possible
framework combinations
were used. Three were used
most often. So there may be
a preference for particular
frameworks for this antigen.
23
“Consensus WT” HuCAL
sequence
(without the PCR
mutations)
All selected clones exhibit
tighter binding. That is, the
addiitonal mutations
introduced by PCR were
selected.
Entirely cell-free: no bottlenecks due to relatively inefficient DNA transfer steps.
One is always dealing here with populations of molecules.
Eventually clone in a plasmid in E. coli for expression and testing.
24
DNA shuffling (Stemmer - Maxygen)
(by PCR)
Fragment DNA
with DNase
Reassemble
fragments
“recombination”
via PCR priming
Iterate
PCR
(no
oligo
primers)
Base changes can be purposely introduced prior to shuffling
• by chemical mutagenesis)
• by error-prone PCR during the process
• by “faithful” PCR during the process (lower level of mutation)
• by using different members of a gene family (paralogs or homologs)
Analogous functions
in the same organism;
e.g. families of proteases
Stemmer, W., Proc Natl Acad Sci U S A. 1994 Oct 25;91(22):10747-51.
Same function
in a different organism; e.g., interferons
From:
Advances in directed protein evolution by recursive genetic
recombination: applications to therapeutic proteins
Aaron L Kurtzman et al. Current Opinion in Biotechnology 2001, 12:361–370
Why alter a natural protein?
For therapeutic proteins:
• Tighter target binding (enabling lower doses, perhaps)
• Enable desirable side reactions
• Disable undesirable side reactions
•Improve half-life in bloodstream
• Improve production characteristics (e.g., secretion,
stability)
For industrial enzymes:
• Altered (even new) substrate specificity
• Greater stabilty
• More robust (e.g., function at extreme pH’s)
• Faster
25
26
Example 1:
Human interferon alpha (IFNa )
20 IFNa genes in a gene family
Each human protein is poor at inducing virus resistance in mouse cells
Shuffle the family, Only 2 rounds. Mixing the non-conserved a.a. differences
Select for induction of viral resistance in mouse cells
Achieve 185X more activity in shuffled protein.
Best was even 3.5 X more active than the best mouse IFN.
And still equally effective vs. human cells.
So an example of a dramatic change in biological specificity (binding?)
27
Shuffling an engineered human gene with a mouse gene
Boxed nts are not conserved. So lots of differences here.
Low temp PCR was required due to the poor homology (Klenow instead of Taq)
Stemmer, W., Proc Natl Acad Sci U S A. 1994 Oct 25;91(22):10747-51.
28
Example 2: Antibodies: breaking the natural limit on affinity selection
Natural affinity ceiling of 10-10 M (100 pM):
Kd = off time / on time
Endocytosis rate ~~ 10 min to several hours
So no selection for off –rate longer than ~3 h (104 sec)
Diffusion-limited on-rate ~ 10-6 M/s
Selection limit of affinity of natural antibody evolution =
off rate / on rate = (1/10-6)/(1/10+4) = 10-10 = Kd
---------------------------------------------------------------------------------------------------
J Foote and HN Eisen
Kinetic and Affinity Limits on Antibodies Produced During Immune Responses
PNAS 1995; 92: 1254-1256
Off time (t1/2)
29
2x10-6 per
second =
5 days
half-life
Iterations
30
Directed evolution of antibody fragments with monovalent femtomolar
antigen-binding affinity. Eric T. Boder, Katarina S. Midelfort, and K. Dane
Wittrup. Proc Natl Acad Sci U S A. 2000; 97(20): 10701–10705.
Used the FACS to selected single chain anti-fluorescein antibodies
displayed on the surface of yeast cells. Competed with free fluorescein.
DNA shuffled. 4 cycles. Selected for slow off times.
Achieved 50 fM affinities. That’s femtomolar, 50 x 10-15 M = 5 x 10-14 M
= 0.05 pM (compare 100 pM above)
Slower off-rate than even biotin-avidin (>5d).
31
Other selections than Ag binding:
scMAb + DNA shuffling + ribosome display + selection in DTT (reducing agent)
Ab folding without need for disulfide bond (-SH oxidation to –S-S- not needed)
as well as high ligand affinity
Lutz Jermutus, Annemarie Honegger, Falk Schwesinger, Jozef Hanes, and
Andreas Pluckthun, PNAS 98:75-80 (2001)
32
Example 3: Enzyme stabilization:
PAI-1, a protease inhibitor (TPA inhibitor)
Error prone DNA shuffling  245X increase in stability (days)
Assay = tPA binding
Found 11 aa changes, presumably affecting protein folding
Back-cross to remove non-contributory mutations:
DNA shuffle best clone with original WT DNA.
Maintain selective pressure.
Analyze “progeny”: see 2 of the 11 aa changes lost,
not needed, replaced by WT sequence.
J Mol Biol. 2001 Jan 26;305(4):773-83.
Different structural requirements for plasminogen activator inhibitor 1 (PAI-1) during
latency transition and proteinase inhibition as evidenced by phage-displayed
hypermutated PAI-1 libraries.
Stoop AA, Eldering E, Dafforn TR, Read RJ, Pannekoek H.
33
Example 4: Viral tropism: murine leukemia virus, a retrovirus
MLV, 6 strains, all  poor infection of CHO cells.
DNA shuffled envelope gene of 6 strains
chimeric virus that can infect CHO cells
And selected incidentally for non-inactivation under conditions of
laboratory manipulation (100X centrifugation resistant)
Nat Genet. 2000 Aug;25(4):436-9.
Molecular breeding of viruses.
Soong NW, Nomura L, Pekrun K, Reed M, Sheppard L, Dawes G, Stemmer WP.
A more global version of DNA -- genome shuffling:
A different (unnatural) kind of genetic mixing using whole genomes
34
Zhang et al., Nature 415, 644 (2002)
Can mix Streptomyces (bacteria) genomes by protoplast fusion.  effective diploid bact
The fused cells will generate recombinant haploid spores.
Target – tylosin production (an antibiotic)
Mutagenize a culture, collect 22,000 survivors.
Screen them all for tylosin synthesis, pick the top 11.
Protoplast fuse all 11 with each other. Collect 1000 progeny.
Screen 100 for tylosin, collect the best 7.
Protoplast fuse again. Collect 1000 again. Screen for tylosin again.
Characterize the best 2: Tylosin production is up 9-fold.
So productivity is up 9-fold, without a tremendous amount of work (22000 screen the mo
35
A more supervised version of DNA shuffling
Multivalent avimer proteins evolved by “exon” shuffling of a family of
human receptor domains
Nature Biotechnology 23: 1556 (2005)
A misnomer; really domain shuffling)
Joshua Silverman, et al & Willem Pim C Stemmer
Avidia, Inc
Strategy:
Create therapeutic proteins by combining hundreds of known binding
domains from receptor proteins in new random combinations and selecting
for binding to a specific target by phage display
36
Organization of binding domains in typical mammalian receptors
A-domains:(~35-40 AA’s/domain):
determine binding speficicity of many
receptors
Typical receptor structures
of 217 A-domains
(~metaphorically?)
as a spacer
between domains
2 domains
cooperating
Bipartite
domain
Dual
specificity
domain
Degenerate oligos synthesized to coding for 35-40 AAs of the A domains
Only AA’s naturally found at each position were coded for.
Conserved structural AAs were kept constant.
Complexity = 1023 . Actually realized = 1010 as phage display particles
Select one domain at a time, serially, by panning:
LRP = LDL receptor related protein
37
Isolation of a high affinity binding protein to IL6 ( interleukin 6 ) by iterative
selection (IL6 is a target for cancer and inflammation)
Phage display (M13)
IL6 immobilized on plates.
Recovered proteins from first cycle, cloned and tested for IL6 binding; 20 top
binders pursued.
Add the domain library to each of the 20 first round winning domains. Again
pick best 20 overall.
Monomer protein
Monomer displayed Screened for
on phage coat
binding
Build 20 dimers pools
from 20 best monomers
After a third cycle pick the very best binder: = “C326”
M13 phage
One
domain
IL6 = interleukin 6
Two
domains
Three
domains
38
Add an IgG-binding domain at the end to prevent rapid clearance
(measured half-life of 89 hours in monkeys)
Binding measured by a competition assay (“AlphaScreen”)
Laser
Luminescence
Reactive Oxygen
IL6
IL6
receptor
39
Reactive oxygen
species can react
only over a short
distance with and
“acceptor” bead
Avidin bead: biotinylated IL6 :
gp130-Fc : Protein A bead
Competition: IL6 (nonbiotinylated) or C326 avimer
Activity of the anti-IL-6 tetramer C326. (a) AlphaScreen competition
analysis comparing ability of C326 relative to IL-6 itself in inhibiting the
interaction of IL-6 with its receptors. An avimer that does not bind to
IL-6 is included as a negative control.
40
More AlphaScreens: effect of combining the 3 domains
41
Physical assay: Biacore surface plasmon resonance to measure binding kinetics
42
Biological assay:
Stimulation of proliferation of TF-1 cells (erythroleukemia line)
16 h of 3H-TdR incorporation to measure promotion of DNA synthesis
Commercial anti-IL6 antibodies
43
Table 1 Selected avimer affinities and activitiesa
Avimer
Target
No. of
domains
Affinity (nM)
IC50
(Biochemical)
(nM)
IC50 (Biological) (nM)
C426
cMet
2
<0.1b
0.170c
Actived
C65
CD40L
2
<0.1e
0.06c
0.1f
C326
IL-6
3
<0.2e
0.05c
0.0008g
C2810
CD28
3
0.1e
n.d.h
0.6f
C2
BAFF
3
n.d.h
0.1c
0.4f
44
Acute phase inflammatory response induced by IL6 is reversed by avimer C326
(in mice)
Specific for IL6-induced inflammation