Rapid mRNA-Display Selection of an IL-6 Inhibitor Using Continuous-Flow Magnetic Separation

Department of Molecular and Medical Pharmacology, University of California, Los Angeles (USA).
Angewandte Chemie International Edition (Impact Factor: 11.26). 08/2011; 50(36):8295-8. DOI: 10.1002/anie.201101149
Source: PubMed


Even flow: Microfluidics technology can rapidly generate affinity reagents that bind to protein targets using mRNA display. The continuous-flow magnetic separation (CFMS) method (see scheme; steps a-h) offers reproducible means to impose stringent selection conditions without the need for a specialized device. The method yields approximately a 30-fold greater partition efficiency over conventional methods when using interleukin-6 as a model target.


Available from: Ting-Ting Wu
Selection Methods
DOI: 10.1002/anie.201101149
Rapid mRNA-Display Selection of an IL-6 Inhibitor Using
Continuous-Flow Magnetic Separation**
C. Anders Olson, Jonathan D. Adams, Terry T. Takahashi, Hangfei Qi, Shannon M. Howell,
Ting-Ting Wu, Richard W. Roberts,* Ren Sun,* and H. Tom Soh*
Since the invention of hybridoma technology, methods for
generating affinity reagents that bind specific target mole-
cules have revolutionized biology and medicine.
In the
postgenomic era, there is a pressing need to accelerate the
pace of ligand discovery to elucidate the functions of a rapidly
growing number of newly characterized molecules and their
modified states.
Nonimmunoglobulin-based proteins such as
DARPins, affibodies, and monobodies represent attractive
alternatives to traditional antibodies as these are small,
soluble, disulfide-free, single-domain scaffolds that can be
selected from combinatorial libraries and expressed in
For example, monobodies—highly stable scaffolds
based on the immunoglobulin VH-like 10th fibronectin
type III (10Fn3) domain of human fibronectin
yielded antibody mimetics that bind to numerous targets for
applications including intracellular inhibition,
and biosensors.
These 10Fn3-based ligands can be
derived from highly diverse libraries using techniques such as
phage, ribosome, mRNA, bacterial, and yeast displays.
Among these techniques, mRNA display has the advant-
age of being an entirely in vitro method that uniquely pairs a
covalent, monovalent linkage between genotype and pheno-
type with a relatively high fusion yield.
Furthermore, it
becomes possible to use very large combinatorial libraries
because of the lack of an obligate in vivo step during iterative
rounds of enrichment, and mRNA-display libraries can
encompass 10
-fold more unique sequences than typical
phage- or cell-based display experiments.
Larger libraries
often produce higher-affinity binders,
and can also yield
diverse pools of target-specific ligands with unique proper-
Large libraries are especially beneficial for evolving
rare functionalities including enzymatic activity,
and the
capacity to target nonstructured biological targets.
ever, effective isolation of desired molecules from high-
complexity mRNA-display libraries typically requires many
rounds of selection, which is often resource intensive. Many
iterative selection cycles may also result in the enrichment of
suboptimal ligands as a result of compounding biases from
selective constraints other than binding efficiency (e.g. PCR
or translation efficiencies). Thus, novel technologies that can
accelerate and automate the selection process are urgently
We report herein a rapid, low-cost, highly efficient
method for generating high-affinity antibody mimetics using
small-scale, continuous-flow magnetic separation (CFMS).
Unlike previous microfluidic approaches, which have
required fabrication of specialized devices,
CFMS can be
performed within a small section of perfluoroalkoxy (PFA)
tubing to achieve highly stringent selection with minimal
background. This low background directly contributes to the
efficiency of selection, and continuous flow improves the
washing efficiency while promoting selection for low off-
rates; together, these factors contribute to the rapid con-
vergence of high-affinity ligands.
We also describe the implementation of an improved
10Fn3 library with enhanced expression (e10Fn3). Previously,
we utilized an in vivo expression screen based on a green
fluorescence protein (GFP) folding reporter
to improve the
expression of a phospho-specific IkBa-binding Fn3 variant.
We demonstrate here that these framework mutations plus an
additional rational mutation enhance expression of four
unrelated 10Fn3 variants both in vivo and in vitro (see
Figure S1 in the Supporting Information). The enhancement
in expression may be due to the replacement of three solvent
exposed hydrophobic residues, which are localized on the
three-dimensional structure, with polar residues, as well as
replacing a buried Leu for Ile, which may enhance stability
because of a significantly higher b-sheet-forming propen-
The substantial 2–4-fold expression increase for
individual e10Fn3 clones in rabbit reticulocyte lysate (Fig-
ure S1c) is also seen for the nave e10Fn3 library as a whole as
expected (data not shown). One additional change includes
limiting variation of the final BC loop random position to the
hydrophobic residues Leu, Ile, and Val, as this position is a
[*] Prof. H. T. Soh
Materials Department, Department of Mechanical Engineering,
University of California, Santa Barbara, CA 93106 (USA)
Dr. C. A. Olson, H. Qi, Prof. T. Wu, Prof. R. Sun
Department of Molecular and Medical Pharmacology, University of
California, Los Angeles (USA)
Dr. J. D. Adams
Department of Physics, University of California, Santa Barbara
Prof. T. T. Takahashi, S. M. Howell, Prof. R. W. Roberts
Department of Chemistry, University of Southern California (USA)
[**] We thank Prof. Vaithilingaraja Arumugaswami and Prof. Christo-
pher T. Denny for helpful comments and discussions as well as
Jenny L. Anderson for technical assistance. We are grateful for the
financial support from the National Institutes of Health, ARO
Institute for Collaborative Biotechnologies (ICB), Armed Forces
Institute of Regenerative Medicine (AFIRM), and California Institute
for Regenerative Medicine.
Supporting information for this article is available on the WWW
1Angew. Chem. Int. Ed. 2011, 50, 1 5 2011 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
These are not the final page numbers!
Page 1
buried core residue in the wild-type 10Fn3 structure and may
interact in the transition-state folding nucleus.
We chose to target interleukin 6 (IL-6) as a model to
generate high-affinity e10Fn3-based ligands using CFMS
mRNA display. This cytokine contributes to the regulation
of the immune response and hematopoiesis,
and aberrant
IL-6 serum levels are implicated in various inflammatory
diseases and cancers.
We show that CFMS selection offers
significantly improved (ca. 30-fold) partition efficiencies
compared to conventional methods, and report the generation
of a high-affinity IL-6 ligand (K
= 21 nm) with an excellent
off-rate (8.8 10
). This high-affinity IL-6 ligand is
capable of inhibiting signaling through gp130, thus indicating
the molecules potential value and demonstrating the effec-
tiveness of CFMS for rapidly identifying clinically relevant
A key advantage of our method is the ability to perform
highly stringent molecular selections on very small scales
using readily available materials while maintaining high
recovery of target binders. The selection process is shown in
Figure 1 A, including standard mRNA-protein fusion prepa-
Briefly, after incubating the target-coated beads
with the library, the beads are efficiently captured by the
NdFeB rare-earth magnets in the PFA tubing, and stringently
washed under continuous flow conditions. Almost all of the
beads are recovered after selection by cutting the section of
the tubing containing the beads and eluting directly into the
PCR buffer. We performed numerical simulations to deter-
mine optimal flow conditions for washing and dissociation of
weakly bound ligands, and found that a flow rate of 30 mLh
(ca. 16 cmmin
) through tubing having a diameter of
approximately 1 mm would deplete a 500 mL suspension
containing 1 10
beads by more than four orders of
magnitude after magnetic capture (Figure 1B). We modeled
capture using a convection-diffusion equation, with convec-
tive particle velocity accounting for the magnetic force on the
beads (calculation details provided in the Supporting Infor-
mation). We also calculated the magnetic field gradient
horizontally across the magnet surface, and confirmed that
captured beads would remain trapped at high flow rates
(Figure 1 C). Even at 30 mL above the tube wall, regions exist
where the magnetic field gradient is suitable for bead capture.
Using CFMS, trapping, washing, and bead recovery can be
performed in less than 10 minutes, and highly efficient
recovery is possible within a minimal volume (less than
20 mL). We demonstrated high recovery using FACS by
passing 1 10
biotin-phycoerythrin-labeled beads through
the system; only 0.01% of the beads were detected in the
flow-through (Figure 1 D). Some residual
beads remained in the microcentrifuge tube,
which is in part due to the concentration of the
beads at the descending buffer surface, but we
could minimize loss with a single rinse step.
To demonstrate the utility of CFMS selec-
tion, we immobilized biotinylated IL-6 onto
nonporous superparamagnetic beads (epoxy
M-270) through neutravidin, which displays
particularly low background library binding
(Figure S2). We performed the first round of
selection with a library containing greater than
10 copies of 1 10
unique sequences. Round
one was performed with 5 10
beads in 5 mL
binding buffer, and all subsequent rounds used
beads in 500 mL binding buffer. After
binding the library against IL-6-labeled beads
for one hour at room temperature, we applied
the sample to the device and washed for
5 minutes at 30 mLhr
. We then transferred
the minimal segment of tubing containing the
trapped, library-bound beads into a microcen-
trifuge tube containing PCR buffer and eluted
the beads for direct amplification.
We performed CFMS selection under two
slightly different conditions. After round three
under the initial selection conditions (selec-
tion A), we noted that the number of PCR
cycles required to amplify affinity-enriched
fusions had decreased by three, indicating that
pool 2A was becoming enriched for binding
over background. We sequenced a random
sampling of ten clones from pool 3A and found
that approximately 70% of these represented a
single sequence, eFn-3A02 (Figure 2A). How-
Figure 1. CFMS mRNA display selection. A) In vitro selection of mRNA-display libraries
using CFMS. 10Fn3 scaffold: blue; random regions: red, gold, and purple. The 10Fn3
scaffold is represented as greek key motif with random BC and FG loops. Selections
begin with PCR amplification (a) to produce copies of approximately one trillion unique
sequences, in vitro run-off transcription (b), splint-mediated ligation of a 3 puromycin-
containing oligonucleotide (c), in vitro translation and fusion formation (d), and
subsequent reverse transcription (e), pool binding (f), and affinity enrichment. Target-
coated beads are captured within the PFA tubing by three small NdFeB magnets (g)
and washed under continuous flow. Bead recovery is achieved by cutting a minimal
length of tubing containing the beads and eluting directly into the PCR mix (h).
B) Simulation of the depletion in bead concentration resulting from vertical magnetic
capture. C) Simulation showing the regions in the tube where the magnetic field is
sufficient to retain beads attracted to the tube surface against the fluid flow. In the
regions shown in red, the magnetic force in the negative horizontal direction exceeds
the maximum Stokes drag force experienced by a stationary bead in the flow field within
the tube. We identified four regions that extend to a distance of 10–30 microns into the
tubing. D) Bead recovery analysis using FACS. All but 0.01% of beads applied to the
device were successfully trapped.
2 2011 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim Angew. Chem. Int. Ed. 2011, 50,15
These are not the final page numbers!
Page 2
ever, this sequence, as well as two nondominant sequences
from this pool, contained cysteines, and we were concerned
that disulfide bond formation may mediate binding to IL-6,
which contains two solvent-exposed disulfide bonds, one of
which is important for function.
While eFn-3A02 has clear
target specificity, we felt that a noncysteine-containing ligand
would eliminate the possibility of nonspecific disulfide bond
formation and preserve an orthogonal chemical handle, a
useful feature of this scaffold.
We thus repeated the selection against pool 1 while
blocking cysteines in our library with iodoacetamide (selec-
tion B) to convert free thiols into thioethers (Figure S3). By
round 4B, the number of PCR cycles required to amplify
enriched binders had decreased by four, thus indicating pool
convergence. eFn-4B02 was the dominant sequence, repre-
senting approximately 80% of the 20 clones we sequenced.
Neither this clone nor Fn-4B01 (composing ca. 15 % of pool
4B) contained cysteines (Figure 2A).
Our results show that CFMS can generate highly con-
vergent pools in considerably fewer rounds (3–4) compared to
previous mRNA display experiments (6–10 rounds).
CFMS, we observed much lower background binding, based
on semiquantitative PCR during initial rounds of selection
(ca. 27 PCR cycles for detection after CFMS compared to ca.
20 in previous studies). As long as target binding is not
significantly less efficient, this improved signal-to-noise ratio
will result in higher selection efficiency. To quantify this, we
utilized quantitative PCR to directly compare the binding of
pool 4B relative to a nave pool in binding to beads with or
without IL-6 using either CFMS or a conventional agarose
bead format (Figure 2 B). Pool 4B showed approximately a
100-fold enhanced binding to the target over background in
the agarose bead format, which is in agreement with previous
However, the background binding in the
CFMS format is significantly lower, whereas for pool 4B
target binding remains nearly as efficient (ca. 30% lower),
thus leading to a greater than 3000-fold target-to-background
binding ratio. This difference in selectivity explains the
improved rate of enrichment observed in CFMS and asso-
ciated reduction in the number of selection rounds needed.
To determine if our clones can efficiently and specifically
bind IL-6, we used bacterially-expressed e10Fn3 variants
immobilized onto nickel affinity resin to pull down IL-6 from
solution (Figure 2C) in the presence of a 500-fold excess of
bovine serum albumin (BSA). It is unlikely that the cysteine-
containing eFn-3A02 binds as a result of nonspecific disulfide
bonding in this assay since BSA contains 35 cysteines (17
disulfides and one free cysteine), and should therefore
prevent such association. eFn-4B02 was the most efficient
binder and was chosen for subsequent analysis by surface
plasmon resonance (SPR; Figure 2D). eFn-4B02 exhibited an
excellent off-rate, k
= 8.8 10
, which gives a half-life of
approximately 13 minutes; this illustrates the advantages of
continuous flow washing. The on-rate (k
) was 4.2
, thus giving a calculated equilibrium binding
constant of K
= 21 nm.
Finally, we explored whether our selected 10Fn3 variants
could inhibit IL-6 signaling in human hepatocytes (HuH
7.5.1), which express both the IL-6-specific co-receptor IL-
6Ra and the signaling receptor gp130. We performed a
western blot to analyze signaling inhibition by measuring
phosphorylation of STAT3, a downstream transducer of
gp130 activation (Figure 3). As expected, eFn-4B02 was a
more effective inhibitor than eFn-3A02, and a nonselected
Figure 2. Characterization of selected IL-6-binding sequences. A) The sequence of the e10Fn3 library (x= random, z= L, I, or V). After three
rounds, pool 3A converged to clone eFn-3A02, which represented 70 % of selected clones. Selection B was performed to avoid cysteine-containing
clones, and converged after four rounds to eFn-4B02 (16 out of 20 clones) and eFn-4B01 (3 out of 20 clones). B) Quantitative PCR-based binding
assay to demonstrate increased selectivity of CFMS for pool 4B, which bound to IL-6 target beads more than 3000-fold more efficiently than a
non-enriched pool (background binding); the selectivity of pool 4B using cross-linked agarose beads was approximately 100-fold. C) Pull-down
results of IL-6 using immobilized e10Fn3 samples, as visualized by western blot (input= 5 % of total). D) SPR measurements to determine affinity
of eFn-4B02 for IL-6 (k
= 8.810
; k
= 4.2 10
; K
= 21 nm).
3Angew. Chem. Int. Ed. 2011, 50, 1 5 2011 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
These are not the final page numbers!
Page 3
control e10Fn3 variant had little effect (Figure 3A). eFn-
4B02 inhibits IL-6 signaling in a concentration-dependent
manner (based on three independent experiments) with an
= 419 nm (Figures 3 B and C).
In summary, we report a simple and cost-effective micro-
fluidic method for the rapid generation of functional non-
immunoglobulin affinity reagents. Increased partitioning
efficiencies in CFMS result in a greater than 30-fold increase
in enrichment over conventional, agarose-based selections,
thus enabling faster selection. Notably, CFMS requires no
special materials or devices, involves minimal cost, and the
miniaturized format decreases the amount of target material
needed for the selection. Continuous washing in the CFMS
format should allow selection of ligands with better binding
kinetics, as demonstrated by the excellent off-rate of the IL-6-
binding eFn-4B02 ligand. Importantly, CFMS allows many
affinity enrichment experiments to be carried out in parallel
and therefore represents a significant step toward the goal of
massively parallel discovery of affinity reagents for a wide
range of molecular targets.
Experimental Section
For detailed methods see the Supporting Information.
Received: February 15, 2011
Published online: && &&, &&&&
Keywords: antibodies · directed evolution · mRNA ·
ligand design · selection methods
[1] D. H. Margulies, J. Immunol. 2005, 174, 2451.
[2] O. Stoevesandt, M. J. Taussig, Proteomics 2007, 7, 2738.
[3] H. K. Binz, P. Amstutz, A. Pluckthun, Nat. Biotechnol. 2005, 23,
[4] A. Koide, C. W. Bailey, X. Huang, S. Koide, J. Mol. Biol. 1998,
284, 1141.
[5] H. I. Liao, C. A. Olson, S. Hwang, H. Deng, E. Wong, R. S. Baric,
R. W. Roberts, R. Sun, J. Biol. Chem. 2009, 284, 17512.
[6] C. A. Olson, H. I. Liao, R. Sun, R. W. Roberts, ACS Chem. Biol.
2008, 3, 480.
[7] L. Bloom, V. Calabro, Drug Discovery Today 2009, 14, 949.
[8] F. N. Ishikawa, H. K. Chang, M. Curreli, H. I. Liao, C. A. Olson,
P. C. Chen, R. Zhang, R. W. Roberts, R. Sun, R. J. Cote, M. E.
Thompson, C. Zhou, ACS Nano 2009, 3, 1219.
[9] A. M. Levin, G. A. Weiss, Mol. Biosyst. 2006, 2, 49.
[10] R. W. Roberts, J. W. Szostak, Proc. Natl. Acad. Sci. USA 1997,
94, 12297.
[11] T. T. Takahashi, R. W. Roberts, Methods Mol. Biol. 2009, 535,
[12] D. S. Wilson, A. D. Keefe, J. W. Szostak, Proc. Natl. Acad. Sci.
USA 2001, 98, 3750.
[13] B. Seelig, J. W. Szostak, Nature 2007, 448, 828.
[14] Y. Liu, J. D. Adams, K. Turner, F. V. Cochran, S. S. Gambhir,
H. T. Soh, Lab Chip 2009, 9, 1033.
[15] G. S. Waldo, B. M. Standish, J. Berendzen, T. C. Terwilliger, Nat.
Biotechnol. 1999, 17, 691.
[16] D. L. Minor, Jr., P. S. Kim, Nature 1994
, 367, 660.
[17] C. K. Smith, J. M. Withka, L. Regan, Biochemistry 1994, 33,
[18] E. Cota, A. Steward, S. B. Fowler, J. Clarke, J. Mol. Biol. 2001,
305, 1185.
[19] T. Taga, T. Kishimoto, Annu. Rev. Immunol. 1997, 15, 797.
[20] S. Rose-John, J. Scheller, G. Elson, S. A. Jones, J. Leukocyte Biol.
2006, 80, 227.
[21] F. L. Rock, X. Li, P. Chong, N. Ida, M. Klein, Biochemistry 1994,
33, 5146.
Figure 3. Inhibition of IL-6 function. IL-6 signaling inhibition was
assayed by western blot for STAT3 phosphorylation at Y705. A) IL-6
(50 ng mL
) was added to HuH-7 cells pre-incubated with 500 nm of
various e10Fn3s for 12 min. B) Quantitative western blot to determine
concentration-dependent inhibition; imaged using an infrared dye-
labeled secondary antibody. Shown is one of three independent experi-
ments. eFn-4B02 (10 nm–10 mm) and IL-6 (10 ng mL
) were pre-
incubated for 30 min before applying to HuH-7 cells for 12 min.
C) Four-parameter nonlinear regression curve fit of quantitative west-
ern blot analysis (IC
= 422 nm).
4 2011 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim Angew. Chem. Int. Ed. 2011, 50,15
These are not the final page numbers!
Page 4
Selection Met hods
C. A. Olson, J. D. Adams, T. T. Takahashi,
H. Qi, S. M. Howell, T. Wu,
R. W. Roberts,* R. Sun,*
H. T. Soh*
&&&& &&&&
Rapid mRNA-Display Selection of an IL-6
Inhibitor Using Continuous-Flow
Magnetic Separation
Even flow: Microfluidics technology can
rapidly generate affinity reagents that
bind to protein targets using mRNA
display. The continuous-flow magnetic
separation (CFMS) method (steps a–h)
offers reproducible means to impose
stringent selection conditions without the
need for a specialized device. The method
yields approximately a 30-fold greater
partition efficiency over conventional
methods when using interleukin-6 as a
model target.
5Angew. Chem. Int. Ed. 2011, 50, 1 5 2011 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
These are not the final page numbers!
Page 5
Show more