A network of interacting proteins controls the activity of Cyclin-Dependent-Kinase
2 (Sherr, 1994; Morgan, 1995) and governs entry of higher eukaryotic cells
into S-phase. Analysis of this and other genetic regulatory networks would
be facilitated by intracellular reagents that recognized specific targets
and inhibited specific network connections. Here we expressed inside cells
a combinatorial library of constrained 20-mer peptides displayed in the
active site loop of E.coli thioredoxin, and used a two hybrid system to
select those that bound human Cdk2. These peptide aptamers were designed
to mimic the recognition function of the complementarity determining regions
of immunoglobulins. The aptamers recognized different epitopes on the Cdk2
surface with dissociation equilibrium constants in the nanomolar range;
those tested inhibited Cdk2 kinase activity. Our results show that peptide
aptamers are analogous to monoclonal antibodies, with the advantages that
they are isolated together with their coding genes, that their small size
should facilitate solution of their structures, and that they are designed
to function inside cells.
We based our approach on the fact that peptide loops that are anchored
at both their amino and carboxyl termini, such as those found in antibody
and T cell receptor complementary determining regions, are capable of specific
and high affinity molecular recognition. The active site loop of E. coli
thioredoxin (TrxA) can be used as a scaffold to display such conformationally-constrained
peptides (LaVallie, 1993), which, when expressed as fusions to the flagellin
gene product, can recognize distinct combinations of shape, charge, and
hydrophobicity (Lu, 1995). Moreover, TrxA is small, soluble, rigid, and
readily expressed and purified from E. coli.
We constructed a library that directed, in yeast, the synthesis of constrained
peptides of random sequence, displayed in the active site of E. coli TrxA
and fused to a modified set of protein moieties from the original interaction
trap (Gyuris, 1993), a yeast two-hybrid system (Fig. 1a). We chose 20-mers
for the displayed peptides, long enough so that they could fold into many
different patterns of shape and charge, but short enough so that many of
their encoding oligonucleotides lacked stop codons, and so that the structure
of the encoded proteins might be easily solved. The library contained 2.9
X 10 9 members, of which >109 directed the synthesis
of peptides. Because of the presence of fortuitous restriction sites in
some coding oligonucleotides and because some library members contained
double inserts, approximately one fifth of the constrained peptides were
longer or shorter than unit length. We introduced this interaction library
into a selection strain containing a LexA-Cdk2 bait. From 6.0 X 106
transformants, we isolated 14 plasmids that expressed peptide aptamers
that interacted with Cdk2 but not with control proteins (detailed in Fig
1b).
We used an interaction mating assay (Finley Jr., 1994) to examine the
strength and specificity with which the peptide aptamers bound Cdk2 (Fig.
2a). As judged by blue color on interaction matrix plates containing Xgal,
all 14 aptamers interacted with the LexA-Cdk2 bait but not with unrelated
proteins such as Max or Rb, or with certain Cdk family members such as
Cdk4, which shares 47% sequence identity with Cdk2. However, some aptamers
interacted with other Cdk family members. The fact that different peptide
aptamers showed distinct patterns of cross-reactivity with different Cdks
indicated that these aptamers recognized different epitopes conserved among
various Cdks. The sequence of the peptide loops is shown in figure 2b.
Non-unit-length peptides occured at the same frequency among the Cdk2 interacting
aptamers as in the library as a whole. No aptamer showed significant sequence
similarity to known proteins, as expected if the 20-mer peptides indeed
formed novel recognition structures. All were charged, suggesting that
some of their interactions with the Cdk2 target could be ionic. None of
the peptides showed more than random similarity to any other, suggesting
that we have not yet exhausted the peptide motifs capable of recognizing
Cdk2.
We arbitrarily chose 6 aptamers for further study. Fig. 3a shows that
the peptide aptamers interacted with a GST-Cdk2 fusion protein in vitro,
demonstrating that these interactions were independent of any bridge proteins
native to yeast. Based on interpolation from interaction trap calibration
experiments (Estojak, 1995), the robust transcription that some aptamers
directed from the pSH18-34 reporter suggested that the equilibrium dissociation
constants (Kds) of the interactions was <10 -6M. In order
to precisely measure the binding affinity of the aptamers to Cdk2, we used
an evanescent wave instrument (BIAcore, Pharmacia). We coupled purified
His6-Cdk2 to CM-dextran chips, then flowed peptide aptamers in running
buffer (10mM Hepes pH 7.4 / 50mM NaCl) over the chip, allowed them to bind,
and rinsed the chip with running buffer without aptamer. Fig. 3b shows
a representative run. We determined association and dissociation rate constants
by fitting the association and dissociation phases of at least two runs
(typically four runs) for each aptamer to exponential functions using a
non-linear least squares algorithm as described (O'Shannessy, 1993). We
calculated Kds by dividing dissociation rate constants by association rate
constants. Table 1 shows that under these conditions, all aptamers exhibited
Kds between 30 and 120nM.
The above results raised the possibility that these peptide aptamers
might inhibit the activity of Cdk2, perhaps by binding to a face of the
molecule and by blocking its interaction with one of its partner proteins
or substrates. Accordingly, we tested the ability of the aptamers to inhibit
phosphorylation of Histone H1 by Cdk2/Cyclin E kinase (Fig.3c). All tested
aptamers did so; under standard conditions (pH7.5, 0mM NaCl) (Kato, 1993),
apparent half-inhibitory concentrations ranged from 1.5 to 100nM. To rule
out the possibility that a trace bacterial contaminant was responsible
for the inhibition, we removed the His6-peptide aptamer from the Pep2 preparation
with a rabbit polyclonal anti-thioredoxin antiserum; this immunodepleted
preparation no longer inhibited Cdk2 kinase activity (not shown). Half-inhibitory
concentrations of aptamers were lower than the Kds measured from evanescent
wave experiments, consistent with the idea that some of the energy of each
interaction is ionic and is reduced by the salt in the evanescent wave
instrument running buffer.
In co-precipitation experiments (Reymond, 1995), purified Pep2 did not
compete with in vitro-translated cyclin E for binding to in vitro-translated
Cdk2 (not shown). However, inhibition by Pep2 was reversed by addition
of a 10-fold excess of Histone H1 (not shown), suggesting that at least
Pep2 inhibits kinase activity by competing with its H1 substrate.
Previous studies have established that libraries of unconstrained peptides
contain sequences capable of recognizing targets in vitro (Devlin, 1990;
Cwirla, 1990; Lam, 1991; Songyang, 1994; Scott, 1994) and in yeast (Yang,
1995); such isolated peptide sequences often bear similarity to natural
interactors. By constrast, although constrained peptide libraries are less
conformationally diverse (McConnell, 1994), the lack of conformational
diversity should lower the entropic cost if binding causes the loop to
adopt a single conformation (Spolar, 1994); we imagine this reduction in
entropic cost accounts for the fact that our Cdk2 peptide aptamers recognize
their targets with higher affinity than is typically observed for unconstrained
peptides (Yang, 1995; Oldenburg, 1992; McLafferty, 1993). This high affinity
suggests that peptide aptamers could inhibit protein function in vivo,
in the simplest case by binding to specific faces of the target molecule
and disrupting its interaction with specific partners or effectors.
The ability to generate large numbers of aptamers from combinatorial
libraries, taken together with the interaction trap, which offers a powerful
selection for those that bind specific proteins, will facilitate the selection
of peptide aptamers against a variety of intracellular targets. The use
of such aptamers as inhibitors of protein contacts should aid the dissection
of the networks of protein interactions that govern division of higher
eukaryotic cells and significantly ease the genetic analysis of those metazoan
organisms for which isolation of specific missense alleles is now impractical.
The analogy with antibodies suggests that peptide aptamers may also be
useful in other applications in which immunological reagents are now employed,
such as ELISAs, immunofluorescence experiments, and sensors, and suggests
routes by which their affinity might be increased: increasing their valency,
and using existing interaction technology to select mutants that bind more
tightly.
Finally, the fact that the variable region is displayed by TrxA, a platform
of known structure, should facilitate solution of the loop structure by
NMR and X-Ray difference methods and help guide searches for peptide-mimetic
compounds more useful as drugs. This first generation of peptide aptamers
may one day lead to recognition modules for intracellular nanotechnologies
aimed at destroying, modifying, and assembling macromolecules inside cells.
FIG.2. a, Specificity of peptide aptamer recognition. b, Sequence of
Cdk2 interacting peptide loops.
METHODS.a. We transformed EGY48 with plasmids expressing the different
anti-Cdk2 aptamers and one that contained a control 20-mer peptide loop,
and then mated these transformants to different bait strains as described
(Finley Jr., 1994). b. We sequenced the DNA encoding the peptide loops
using a dideoxy kit (USB).
FIG 3. a, In vitro interaction between peptide aptamers and Gst-Cdk2.
b, Representative affinity measurement. Figure shows portions of the curve
used to calculate rate binding constants. c, Inhibition of Cdk2 kinase
activity in vitro by peptide aptamers.
METHODS. a, We purified Gst and Gst-Cdk2 as described (Lee, 1995).
We constructed pALHISTRX by annealing the oligonucleotides 5'TAATGAGCGATAAACACCACCACCACCACCACGACGACGACGACAAAGG3'
and 5'TACCTTTGTCGCTGTCGTCGTGGTGGTGGTGGTGGTGTTTATCGCTCATTA3', and ligating
into NdeI-cut pALTRX-781 (LaVallie, 1993). We cloned AvaII fragments encoding
peptide loops from the library plasmids into RsrII-cut pALHISTRX. We expressed
His6-TrxA and His6-aptamers in GI724 as described (Ausubel, 1987-1994),
purified them on a Ni2+-NTA-Agarose according to manufacturer's directions
(Qiagen), and dialysed them against 10mM Hepes pH 7.4 / NaCl 50mM. We precipitated
1µg of His6-TrxA or His6-aptamers with Gst or Gst-Cdk2 sepharose
beads as described (Lee, 1995) , and detected the aptamers by Western blot
with an anti-TrxA rabbit antiserum and ECL reagents (Amersham). b, We cross-linked
His6-Cdk2 in 10mM MES pH6.1/ 50mM NaCl to CM5 chips with an amine-coupling
kit (Pharmacia). We flowed purified aptamers in running buffer (Hepes 10mM
pH7.4; NaCl 50mM) onto the chips at 5µl/min. We recorded association
and dissociation of the His6-Cdk2-aptamer complexes as variations in resonance
angle with time. Association phase starts upon aptamer injection and dissociation
phase upon running buffer injection. We fitted portions of association
and dissociation curves that excluded the sudden variations in resonance
angle caused by transitions between running buffer and aptamer-containing
running buffer, which differed slightly in refractive index ("buffer fluxes").
c, We co-infected 2x10 7 Sf9 cells with recombinant bacculoviruses
expressing hemagluttinin-tagged Cdk2 and His6-Cyclin E as described (Kato,
1993; Desai, 1992). We lysed cells 40 hrs after infection in 500 microL
of 1X Kinase Buffer (Kato, 1993). We used 5 microliter of 100-fold diluted
extract in 30 microliter reactions. We performed 20 min reactions at 250C
by adding 2.5 microCi of [g -32P]ATP (3000 Ci/mmol), 25 microM ATP, 100ng
of Histone H1 (Sigma), and indicated amounts of His6-TrxA or His6-aptamers.
We ran samples on 15% SDS-PAGE gels and exposed by autoradiography.