ABSTRACT
Constrained RNA libraries of limited sequence complexity were constructed and used to select RNA molecules binding to the antigen
binding site of an anti-ferritin antibody. The sequences required as primer-binding sites for the selection cycle were designed to form a
predictable
secondary structure, which greatly facilitated the characterisation of the secondary structures of the selected RNAs. RNA-antibody interactions were studied by real-time interaction analysis to study the dynamic aspects of binding and by
circular dichroism spectroscopy to search for conformational changes upon
binding. The selected RNAs were analysed with a binding site sequestering assay
and were shown to compete with ferritin for binding to the antigen-binding site. The experiments described here indicate that the introduction of strong structural constraints does not have to interfere with the ability to select tightly and specifically binding RNA-molecules.
The number of building block types that can be used to generate molecules with a
desired feature by
in vitro
selection from combinatorial libraries is growing rapidly. The libraries can be
of very different structural classes: peptides displayed on the surface of
phages, synthetic peptide libraries, antibody libraries, RNA or single-stranded DNA containing natural or modified nucleotides and chemical libraries. From any of these libraries molecules binding specifically and with high affinity to a wide variety of ligands have
been isolated (for reviews see
1
-
3
).
Although nucleic acid libraries were developed only a few years ago (
4
,
5
), a large number of RNA and DNA molecules has been selected and characterised. These libraries have been used to define binding sites of RNA-binding proteins (
5
-
8
), to generate RNAs that inhibit polymerases (
9
) or proteases (
10
), to change properties of existing ribozymes (
11
,
12
) or to create novel ribozymes (
13
,
14
). It has been shown that specifically selected RNA, also called aptamer, is
capable of discriminating between molecules differing by as little as a single
methyl group (
15
).
The structural motif responsible for the activity of an aptamer is often quite
small, so it could appear that the structural elements employed by aptamers to
bind to ligands which normally do not bind to RNA are similar to the elements
found in naturally occurring RNA-protein interactions. These RNA elements include stem-loop structures with variable stem but fixed loop sequence,
symmetric and asymmetric internal loops, pseudo-knot and G-quartet structures (
16
). So it might be sufficient to limit the randomised library sequences to a
specific RNA motif in an otherwise constant background. This kind of strategy
is used in antibodies to generate diversity of binding specificity involving
only a small portion of the molecule. The variable part of the antibody is kept in loop structures that are structurally predisposed within the context of the whole antibody molecule to form binding sites.
To test the hypothesis that this concept could also be used for RNA libraries,
two structurally constrained RNA libraries of limited sequence complexity,
representing two different types of constraint, were designed. Both libraries
were used to isolate aptamers binding to the anti-ferritin antibody H107 (
17
). The antibody was selected because peptides have been isolated with H107 from
phage display libraries (
17
), allowing comparison of aptamer and peptide structures binding to the same
ligand.
Template was generated by amplifying 200 pmol of single-stranded DNA-library (sequence corresponding to the RNA sequences shown in Fig.
1
) for 5 cycles with 1 nmol of each primer (GQCL: primer T7P1, 5'-GGGAAGCTTAATACGACTCACTATAGGGTTGCACACCGTCTCG and primer 3'RSP1, CCCAAGCTTCCCGGGTTGCACACCGTCTCG; SSL
30
: primer T7P2, GGGAAGCTTAATACGACTCACTATAGGGATGCTTCGGCATCCC and primer 3'RSP2, CCCAAGCTTACGTACCGCCGAAGCGGTAC). For all PCR reactions buffer J (50
mM NaCl; 3.75 mM MgCl
2
; 10 mM Tris-HCl pH 8.0; 0.01% gelatine; 0.01% Triton X-100) was used. DNA was purified from 2.5% agarose gels and transcribed in a volume of 240 [mu]l for 16 h (including 30 [mu]Ci [[alpha]-
32
P]GTP) as described in reference
4
. RNA was purified by phenol/chloroform extraction and ethanol precipitation.
RNA was re-suspended in H
2
O and DNA digested in a volume of 300 [mu]l for 1 h with 50 U RQ DNase (Stratagene) in the recommended buffer. RNA was
purified by phenol/chloroform extraction, unincorporated nucleotides removed
(NAP
5
; Pharmacia), RNA precipitated with ethanol, suspended in 52 [mu]l NaK
150
, heated to 68oC for 10 min.
Dextran coated sample cells were used to couple H107 (50 [mu]g/ml) covalently to the cuvette in 10 mM NaAc pH 5.0. The base line was
obtained with 195 [mu]l NaK
150
, RNA to be analysed added in a volume of 5 [mu]l NaK
150
and the association monitored for 5-10 min. The sample cell was washed rapidly twice with 200 [mu]l NaK
150
and dissociation monitored for 5-10 min. The cell was regenerated by washing for 2 min with 200 [mu]l 1.5 M NaCl and three times with 200 [mu]l NaK
150
. The same H107 coated sample cell could be used for 20-30 measurements, stored in between measurements at 4oC (in PBS + 0.1% Tween-20). Curves were fitted using the fast fit software package
and association constants obtained by plotting
k
on
over aptamer concentration (error typically 1-3%).
Spectra were obtained with a JASCO J710 Spectropolarmeter at 4oC using a cuvette of 1 cm pathlength (V = 2.5 ml). Spectra were recorded
from 320-220 nm at 5 nm/min and five measurements sampled. Aptamer concentrations were 2.4 [mu]M, H107 concentration 132 [mu]g/ml (=825 nM bivalent antibody or 1.65 [mu]M binding sites), all samples in 100 mM KCl/10 mM Na-phosphate pH 7.5).
Due to the iterative nature of the
in vitro
selection process the 5'- and 3'-terminal sequences of all library molecules remain
invariant because they serve as the primer binding sites both for the reverse
transcriptase and for the polymerase chain reaction (PCR) step. If fixed
sequences could be used to create a structural constraint, it might be easier
to analyse the structure of RNA selected from libraries with short stretches of
variable sequences because of the formation of a predictable secondary
structure. To test the feasibility of this approach, two different libraries,
GQCL and SSL
30
, were constructed. The library GQCL should be compatible with the formation of
G-quartet-like structures and contained primer binding sites of complementary
sequence permitting the formation of a continuous 18 bp stem structure. The
stem sequences were interrupted by four groups of three G-residues separated by four blocks of variable sequence (1-4 random nucleotides for each block). These blocks should provide
sufficient structural variability for the formation of loops connecting a G-Quartet-like scaffold (Fig.
1
, GQCL). The library SSL
30
was not intended to favour a particular type of structure in the variable part
of the library. Rather, each primer sequence was designed to form a stable
structure independent from the rest of the molecule (Fig.
1
, SSL
30
). A single-stranded tail of 11 nt was added to both libraries to increase the
sensitivity of the reverse transcriptase-PCR reaction. Although this introduced
the risk that the 3'-tail could be used for base pairing interactions with bases from
the variable sequence of the library, it permitted the use of stringent
selection conditions. The stringency of the selection is likely to influence
the number of cycles required to select an aptamer, and also the affinity for
the selector molecule.
Since RNA binding to the antigen binding site of H107 was desired, aptamers
binding to the constant region of the antibody were depleted in every cycle by
using an unrelated antibody from the same species (anti-oncostatin M/[alpha]-OSM). Although the amount of RNA eluted from the initial
cycles of selection increased, after four to five cycles the amount of RNA
remaining decreased with each additional cycle (Table
1
). This observation was reproducible, because any given selection cycle could be
repeated and the previously observed amount of RNA remained after the washing
steps. This is in contrast to what has been reported previously in selection
experiments where the amount of selected material usually increased with each
additional cycle until the majority of the selector molecules had been
saturated. The decrease in the amount of RNA after five cycles of selection was
accompanied by the appearance of shorter PCR products. It is possible, that
once selection for binding affinity has reduced the RNA pool to a small number
of individual sequences, the selective pressure of the cycle might be limited
to the deletion of sequences outside the primer binding sites.
Table 1
Course of selection
Aptamers derived from the 4th cycle of the GQCL selection were analysed directly
for binding activity by immunoprecipitation or real time interaction analysis.
Specific binding to H107 was observed for 7/32 RNAs. No binding of the
unselected libraries of GQCL and SSL
30
to H107 could be detected, neither by immunoprecipitation, nor by real time
interaction analysis. This indicates that the scaffold structures selected for
the libraries are not responsible for determining binding specificity.
Sequencing showed that 6/7 were identical (clone GQCL-4.7), while the seventh clone had a single base substitution. In all
clones selected the 4th group of G-residues (G
4
) which was invariant in the starting library, had been mutated from GGG to AAG.
Instead, new G-residues had been created from the variable regions N
3
and N
4
(Fig.
2
). It is possible that the reason for the lack of recovery of the initial
constrained structure is that a G-quartet like structure and a stem structure might have to be separated by
loop sequences on both sides of the G-quartet. In the GQCL-design used for this selection a loop was provided only at the 5'-end of the G-quartet scaffold. The sequence outside of the
primer binding sites is extremely purine-rich (
20
/
22
), and the selection of additional G-residues from the variable blocks made an unambiguous allocation of a G-quartet scaffold in GQCL-4.7 impossible (Fig.
2
; 1+2). Immunoprecipitation of GQCL-4.7 with various antibodies demonstrated that it did bind specifically to
H107 (Fig.
3
a). Therefore, to characterise the GQCL-4.7 structure, various types of mutagenesis were employed. The 3'-tail was removed by cutting the template with a restriction
enzyme (
Sma
I) and RNA transcribed from this template was analysed for binding to H107. No
difference in binding was detected. Next, GQCL-4.7 derivatives with the ability to form stems of 16, 14, 12, 8 or 4 bp
were constructed and tested for binding to H107 and [alpha]-OSM antibodies. Shortening the stem to 8 bp or less resulted in
complete loss of binding to H107 (Fig.
3
b and data not shown). This indicates that a stem is formed by the primer
binding sites as anticipated with the constraint. Furthermore, the formation of
a stem of a length of >8 bp is required for the formation of the active
structure.
The selection strategy used should have resulted in the isolation of aptamers
binding to the antigen binding site of H107, because the library was depleted
specifically in each selection cycle. To test whether this indeed was the case,
a binding site sequestering assay was performed (Fig.
6
a). A limiting amount of H107 (7 pmol) was coupled to magnetic beads and incubated for 30 min with 400 pmol of unlabelled RNA (specific and non-specific RNA) or with ferritin (28 [mu]g = 1200 pmol monomeric ferritin = 50 pmol ferritin 24mer, the
solution form of ferritin). Subsequently, radioactively labelled aptamer (10
pmol) was added and the immunoprecipitation was continued for 15 min.
Pre-incubation of H107 with tRNA or GQCL-4.7G
4
G3
did not interfere with binding of SSL-2.5 to H107. Pre-incubation with ferritin or cold SSL-2.5 RNA abolished binding almost completely. Pre-incubation of cold GQCL-4.7 reduced the amount of SSL-2.5 immunoprecipitated significantly but not
completely, reflecting the lower affinity of GQCL-4.7 (Fig.
6
b, SSL-2.5). This interpretation is supported by the observation that in the
reverse experiment, pre-incubation of cold SSL-2.5 RNA excluded the immunoprecipitation of GQCL-4.7 completely (Fig.
6
b, GQCL-4.7). The results of the competition experiments indicated that both
aptamers bind to the antigen binding site of H107 (or to two sites very close
in space), and confirmed that also binding of aptamer and ferritin is mutually
exclusive.
It was interesting to analyse the dynamic aspects of the interaction of the
aptamers with H107 precisely, to gain information about on- and off-rates of aptamers. The antibody was linked covalently to a dextran-coated sample cell of a Affinity Sensors IAsys real time
interaction analysis system. Unlabelled RNA was added at various concentrations
and binding as well as dissociation were monitored in real time. Figure
7
shows the superimposition of binding curves obtained with various
concentrations of aptamer. These measurements confirmed the observed
differences in the relative affinity of GQCL-4.7 and SSL-2.5. The calculated dissociation constants were
K
D
= 6.4 * 10
-8
M for SSL-2.5 and
K
D
= 1.6 * 10
-6
M for GQCL-4.7. The higher affinity of SSL-2.5 compared to GQCL-4.7 seems to be due to a ~15-fold higher on-rate and a 2-fold lower off-rate of SSL-2.5 (Fig.
7
). For comparison, the equilibrium signal was plotted against the concentration
of aptamer to obtain a conventional binding curve. Defining
K
D
as the concentration at which the half-maximum response was obtained, produced numbers similar to those above [
K
D
-values obtained in this way were usually lower ( <= 20%) than those obtained by integrating the binding curves with the
fast fit program, data not shown].
CD spectroscopy can be used to obtain qualitative information about a
conformational change of an aptamer when binding to a ligand (
23
). For this reason, CD-spectra of the H107 aptamer with the highest affinity for H107 (SSL-2.5) in the presence or absence of H107 were recorded. As a control,
the mutant SSL-2.5CB, which did not show detectable binding to H107, was used for
comparison. The spectra of SSL-2.5 and SSL-2.5CB RNA were almost identical in the absence of H107 (Fig.
8
a). Furthermore, upon addition of H107 the spectrum of SSL-2.5CB did not change in the region around 270 nm (Fig.
8
c). In contrast, in the presence of H107 the spectrum of the aptamer SSL-2.5 indicated a clear reduction of the signal around 270 nm (Fig.
8
b and d). Although this is a mixed spectrum of the RNA and the antibody, at the
concentrations used for these measurements the antibody is not likely to have
contributed to the changes of the spectrum around 270 nm. To visualise
conformational changes of aromatic side chains in proteins much higher
concentrations would have been required. The observed differences of the CD-spectra of the RNA upon binding to H107 are similar to those reported for
the binding sites of Rev (
24
), Tat (
25
) and arginine binding DNA aptamers (
19
). Although the conformational change observed upon binding does not prove that
it is required for binding, it indicates that the structure of the SSL-2.5 RNA might be different in the bound and the unbound state.
Consequently, detailed studies of the RNA structure in the absence of the
ligand are unlikely to be sufficient to identify the reactive groups involved
in ligand-RNA interaction. A dramatic change of the backbone conformation similar
to that described for the Tat-binding site could not be deduced from the RNA-structure alone.
Commonly used RNA-libraries are of high sequence complexity and no attempts are made to
limit diversity by introducing structural constraints. This is motivated by the
reasonable assumption that the higher the sequence complexity of the library
is, the higher will be the structural complexity of the library and the higher
the resulting chance of obtaining tightly binding RNA. However, a problem could
arise from the fact that the number of individual RNA-sequences that can be handled in a single experiment is much lower than
the theoretical complexity of the library. The result is partial pool coverage.
The selection of aptamers from libraries of high sequence complexity requires
many rounds of selection (10-12 cycles) to enrich individual sequences sufficiently to allow their
identification. This incorporates the risk of selecting against RNAs with
stable secondary structures because their lower replication efficiencies might
favour enrichment of less stable structures. High enrichment factors for
individual sequences were observed with the motif-libraries used for this study already after four to five cycles of
selection. The characterisation of the secondary structure of the H107 aptamers
was greatly facilitated by the predictable structure formed by the primer
binding sites. For
in vitro
selection studies aiming at a general understanding of RNA structure, the
relatively easy identification of the approximate overall secondary structure
of the aptamer selected from constrained libraries might not seem to be of
great importance. However, if aptamers are selected with the intention of
generating molecules interfering with the function of proteins, either
in vitro
or
in vivo
, a high resolution structure might not be required. In addition, the stability
of the aptamer will be an important factor for applications. The secondary
structure constraints used here were partly inspired by those of UsnRNAs. For
example, the binding site for the common UsnRNP-proteins (Sm-binding site) is a single-stranded region flanked by stable stem-loop structures (
26
). So the aptamers selected from these constrained libraries might be used
directly for experiments, without the need to engineer a stable structure.
Pilot experiments performed with the H107 aptamers showed that both aptamers
were stable for >= 24 h after injection into Xenopus oocytes (unpublished data).
One of the constrained libraries used for this study, GQCL, was designed to
favour a three-layer G-quartet structure. GQCL-4.7, the aptamer selected from this library carried a 2 nt
mutation compared to the original constraint, indicating that the chosen
backbone was not compatible with the formation of the binding site for the H107
antibody. However, a constraint is likely to limit the structural complexity of
a library by limiting its flexibility. The type and the right amount of
constraint, sufficient for displaying the binding regions without creating a
structure that is too rigid to allow binding to occur, will have to be
determined experimentally by designing a constraint and analysing the outcome
of the selection. Constrained libraries have been used already to select RNA (
7
,
27
). In the former selection a library with a randomised internal loop was used to
obtain optimised Rev-binding elements, but it was already known that the binding site of Rev
(an RNA binding protein) was an internal loop. The selection of RNA binding an
antibody raised against a peptide derived from the g10 fusion protein with a
library based on the stem-loop II of U1snRNA (
27
) is more relevant to the work presented here. However, there, the RNA structure
was not characterised in detail, and the affinity of the RNA for the antibody
was not determined.
The dynamic aspects of aptamer binding were analysed with a real time interaction analyser. This method permits the determination of kinetic constants under quasi-solution conditions in real time (
28
). So far the affinities of aptamers have been determined only by indirect
methods and no on- or off-rates were reported. The relative values for the two aptamers
characterised are in good agreement with the relative affinities observed in
competition assays and immunoprecipitation experiments. The affinity of GQCL-4.7 is comparable to the average affinity of aptamers isolated from
unconstrained libraries (1-500 [mu]M), the aptamer SSL-2.5 is part of the group of high affinity aptamers. This
indicates that the introduction of a constraint does not necessarily interfere
with the ability to select tightly and specifically binding aptamers.
CD-spectroscopy has been used to predict conformational changes in the RNA-elements binding to Rev, Tat and arginine binding DNA aptamers, and
at least in the case of Tat were confirmed later by the determination of the
Tat-binding RNA structure in the presence and absence of ligand by NMR (
29
). Since the determination of RNA structure by NMR is far from trivial, CD-measurements might be a fast way to look qualitatively for conformational changes occurring in the RNA upon binding to a ligand.
Although it is not clear whether it will be possible to attribute certain
shapes of spectra to specific secondary structure elements of RNA, as is
possible for proteins, it seems likely that the signal around 270 nm
corresponds to the stacking interactions of the bases (
23
). If bases would move out of helical structures to interact with the ligand (as
in the case of Tat) a reduction of the signal might be expected (as observed
for Tat and Rev). Of course the observation of a conformational change of RNA
or DNA upon binding to a ligand on its own is not sufficient to prove that the
conformational change is required for binding. However, since CD measurements
can be performed rapidly it could be informative to employ this type of
analysis as a standard characterisation for aptamers. It might turn out that conformational changes are the rule rather than the exception.
Interestingly, the differences in affinity of SSL-2.5 and GQCL-4.7 are mainly due to different on-rates. If SSL-2.5 would simply provide more/better contact points for
H107, on- and off-rates might have been affected similarly. Therefore the lower
affinity of GQCL-4.7 might be the consequence of a more rigid structure, which would
require more time/energy to undergo a conformational change (if it should be
required for binding).
Although the GQCL-4.7 and the SSL-2.5 aptamers seem to have a rather different secondary structure,
they have some common features. Both of them are purine rich, contain purine
residues opposed to each other at the border of helical regions, and both
include the motif UGGAAG. However, no common secondary structure is
recognisable, and the differences in affinity for H107 suggest the formation of
distinct structures.
Purine-rich sequences have been reported to be important for the activity of
other aptamers, what might be the special feature of purine clusters? An
attractive hypothesis would be that structural flexibility could be
incorporated into an RNA structure by non Watson-Crick interactions. Hoogsteen base pairs have been found in the binding
sites of ribosomal proteins, Rev, the E loop of 5SRNA, and splice sites of pre
mRNA . A family of motifs involving G-A mismatches in ribosomal RNAs was studied and provided evidence for the
formation of a sheared tandem structure. Since the A-residues could always maintain a similar exposure in the minor groove it
was suggested that they might represent recognition or anchoring units (
30
). The formation of purine-purine interactions might be the RNA analogue of a hydrophobic pocket in
proteins, which upon conformational change becomes available for hydrophobic
interactions with a ligand. Furthermore, the functional groups of the bases
used for Hoogsteen or reversed Hoogsteen interactions are different from those
of the standard base pairs. This leaves for example the NH
2
-group of guanine available for interactions with the ligand.
Finally, do the aptamers share some similarity to the natural H107 ligand,
ferritin or the linear peptide selected from the phage display library? The
exact structure of the H107 epitope is not known, however, the selected peptide
has been modelled to the surface of ferritin. These simulations showed that the
peptide-consensus sequence YDAxxxW could be super-imposed onto the ferritin surface residues Y
39
D
42
W
93
. These residues are distant in the linear sequence but close in space in the
three-dimensional structure (
31
). The ferritin residues identified to be important for binding to H107 are
compatible with the recognition or the failure to do so of ferritin mutants by
H107 (
32
). It would not be difficult to imagine that the purine cores identified in the
H107 aptamers could make a spatial arrangement of contacts in a way equivalent
to that of the aromatic side chains likely to represent the H107 epitope of
ferritin (see fig. 4 in
31
). In this case the selected aptamers might be structurally equivalent to the
protein structure and not simply binding to the same site of the ligand. A
similar proposal has been made for aptamers selected with an anti-human insulin receptor antibody (
33
). In the absence of structural data, further evidence for this hypothesis would
be if an antibody could be generated against one of these aptamers and if the
antibody raised against the aptamer would recognise the peptide antigen.
I would like to thank Philippe Neuner for the 10
th
year of high-speed/tech oligonucleotide synthesis, John Aherne of Affinity Sensors for excellent customer service and for his introduction to real
time interaction analysis, Manuela Emili for artworks, Sven Behrens for helpful
discussions, Elisabetta Bianchi and Raffaele DeFrancesco for advice on CD-measurements, Joe Jiricny and Iain Mattaj for valuable comments on the
manuscript, and Alessandra Luzzago and Alfredo Nicosia for helpful suggestions,
antibodies and encouragement when approaching local minima during the course of
these experiments.
Cycle
SSL
30
[pmol RNA]
GQCL [pmol RNA]
1st (wash without NP-40)
1.66
1.19
2nd (wash with NP-40)
0.15
0.15
3rd (wash with NP-40)
0.26
0.35
4th (wash with NP-40)
1.69
1.20
5th (wash with NP-40)
1.69
4.21
6th (wash with NP-40)
2.68
1.30
7th (wash with NP-40)
2.66
8th (wash with NP-40)
1.64
REFERENCES