ABSTRACT
The oligonucleotide r(GGACUUCGGUCC) has been observed to adopt a hairpin
conformation by solution NMR and a double helical conformation by X-ray diffraction. In order to understand this apparent conflict, we used time-resolved fluorescence depolarization and
19
fluorine NMR to follow the secondary structure of this dodecamer as the solution
composition was changed stepwise from the NMR experimental conditions to those used for crystallization. Calculation of the dodecamer concentration in the crystal (180 mM strands) and the cation
concentration needed for neutrality (>2 M) prompted investigation of a tethered
species, in which two dodecamers are connected by a string of 4 nt,
geometrically equivalent to
~100 mM strands, in 2.5 M NaCl. The RNA tetraloop and its DNA analog maintain a
single-strand hairpin conformation in solution, even under the conditions used to
grow the crystal. Under high salt conditions, the tethered RNA and DNA analogs
of this sequence yield secondary components which could be the double helical conformation. Crystal contacts in addition to solvent changes and high RNA
concentrations are needed to obtain the double helix as the predominant
species.
Our understanding of RNA function, stability and catalytic activity is hampered
by a paucity of three-dimensional structures. Nucleic acid structures, particularly of folded
RNA, are difficult to obtain using solution NMR and single crystal X-ray diffraction methods. The NMR study of RNA molecules larger than 30 nt requires isotope enrichment (
1
-
3
), while X-ray diffraction requires large, ordered RNA crystals, which have only been successfully produced
for tRNA and a few other small RNAs. Some notable RNA structures have recently
been solved, including: the P1 helix (20 nt) from Group I self-splicing introns, solved by heteronuclear multidimensional NMR (
4
); a 34 nt hammerhead ribozyme complexed to a DNA inhibitor, solved by X-ray diffraction (
5
); two complexes (38 and 40 nt) between RNA hairpins with complementary loops (
6
), termed `kissing hairpins' (
7
); a 30 nt RNA construct modeling a regulatory element of the human U1A pre-mRNA, both free and complexed with the RNA binding domain of the U1A
protein, solved by NMR (
8
); a 21 nt hairpin complexed to the U1A domain, studied by X-ray diffraction (
9
).
Although solution NMR and single crystal X-ray diffraction are the methods of choice for macromolecular structure determination, solvent composition for the two methods are very different and so too may
be the results. We ask why a solution NMR structure and an X-ray diffraction structure disagree in the case of an RNA dodecamer,
r(GGACUUCGGUCC) (Fig.
1
). This sequence is of interest because it is expected to adopt a particularly
stable hairpin structure (
10
). Approximately half of all hairpins in rRNAs are tetraloops, with over two
thirds of these sequences belonging to two motifs, UNCG and GNRA, where N is
any nucleotide and R is a purine (
11
). In fact, the most frequent tetranucleotide loop sequence in
Escherichia coli
16S rRNA is C(UUCG)G. When Tuerk
et al.
(
12
) searched a bacteriophage T4 sequence library for this hexanucleotide sequence,
they found that it occurred, flanked by inverted repeats, amazingly frequently
in intercistronic regions. Many of these predicted hairpins efficiently
terminated DNA synthesis by avian myeloblastosis virus (AMV) reverse
transcriptase, suggesting that the enzyme could not traverse the secondary
structure. Thermal denaturation of model hairpins (
12
,
13
) demonstrated that C(UUCG)G tetraloops have melting points ~20oC higher than homopolymeric tetraloops, indicating that the UUCG loop
sequence enhances the stability of the constructs. In fact, when an RNA
structure prediction algorithm assigns 2 kcal/mol extra stability to hairpins
with the eight most common rRNA tetraloop sequences (all belonging to the UNCG
and GNRA families and including UUCG), the results become more consistent with
phylogenetic analysis (
14
). Hairpins with UUCG are also abundant in many other RNAs, both eukaryotic and
prokaryotic, including precursor and mature tRNAs and the catalytic M1 RNA of
E.coli
RNase P (
12
).
The solution structure of the RNA dodecamer r(GGACUUCGGUCC) was determined by
Tinoco and co-workers using NMR NOE distance constraints and scalar coupling constants,
followed by distance geometry and constrained energy minimization (
4
,
15
,
16
). The dodecamer did indeed adopt a hairpin conformation. Structural features in the loop convincingly explained the unusual stability, including extended base stacking, a U-G base pair closing the loop and a contact from the loop cytidine to a phosphate oxygen. The results agreed with chemical modification experiments of CUUCGG hairpins in 16S rRNA (
17
) in that the more reactive, and variable, second residue of the loop lacked
specific contacts and stuck out into the solvent in the NMR structure. The
solution structure, together with the biophysical and biochemical data, painted
a satisfying, convincing picture of this stable UUCG hairpin. A year later,
however, Holbrook
et al.
(
18
) published the single crystal X-ray diffraction structure of the same oligomer. The data were collected to
2.0 Å resolution and the structure was solved by molecular replacement with a
4 bp probe and refined to a crystallographic
R
factor of 18%. Surprisingly, the RNA adopted a double helical conformation with a remarkably regular structure, despite four consecutive non-Watson-Crick base pairs in the center of this helix. This tract consisted
of two U-G wobble base pairs and two U-C base pairs, with the latter composed of one base-base hydrogen bond and one water-mediated hydrogen bond. In the crystal, the duplexes stacked end-to-end forming pseudo-infinite helices. These interacted through water-mediated hydrogen bonds and four direct
hydrogen bonds per dodecamer, which all involved ribose 2'-hydroxyls.
Oligoribonucleotides were made on a Cruachem Model PS250 Synthesizer using 2'-
O
-fluorophenyl methoxypiperidinyl-protected phosphoramidites (Cruachem) in 1 [mu]M batches. Labeling is achieved by selectively replacing
uridines on the RNA (or thymidines on DNA analogs) with 5-fluoro-2'-deoxyuridine (FdU). FdU-protected phosphoramidite monomers were purchased from American Bionetics, while all of the
canonical nucleotide monomers were obtained from Cruachem. The DNA oligonucleotides were made in automated solid-phase syntheses on a Milligen Biosearch Cyclone DNA Synthesizer. Fifteen micromole (for NMR
samples) or 1 [mu]mol syntheses were performed using standard phosphoramidite methodologies. All oligomers were separated
from failure sequences and shorter products by trityl-on reverse phase HPLC over a Beckman C18 column, with the RNA purified prior to 2'-OH deprotection. The oligonucleotide sequences utilized in this research are listed in Figure
2
.
L buffer (
NMR samples initially contained 3-4 mM DNA in L buffer containing 10% D
2
O in a volume of 350-400 [mu]l. The pH of each sample was measured (without regard to D
2
O content) and adjusted by adding submicroliter amounts of 1 M NaOH or H
3
PO
4
in sample buffer. For studies at higher salt concentrations, the samples were lyophilized to dryness and redissolved in 1* D buffer plus the additional salt. Samples for fluorescence studies typically
contained 2.5 OD oligonucleotide in 250 [mu]l 1* buffer (~100 [mu]M). Ethidium bromide (EtBr) in the same buffer was added to
a final concentration of 5 [mu]M. All samples were heated in sample buffer to 90oC and allowed to reach a thermodynamically stable conformation by slow
cooling to room temperature over 12 h.
The hydrodynamics of the oligonucleotides were probed by bound EtBr, with time-resolved fluorescence decays collected on a Single Photon Counting
apparatus in the Regional Laser and Biology Laboratories at the University of
Pennsylvania.
Exciting a population of chromophores with a pulse of vertically polarized light
selectively excites those fluorophores whose absorption dipoles are nearly aligned with the
z
-axis. Subsequent fluorescence emission is polarized with anisotropy
defined as
r
(
t
) =
I
||
(
t
) -
I
^
(
t
)/
I
0
(
t
)
1
where
r
(
t
) is the time-dependent anisotropy,
I
||
(
t
) is the time-dependent decay of the vertical component of the fluorescence,
I
^
(
t
) is the time-dependent decay of the horizontal component of the fluorescence and
I
0
(
t
) is the time-dependent total fluorescence intensity. If the chromophore exhibits significant rotational Brownian diffusion before fluorescence emission occurs, the rate of anisotropy decay will be proportional to the rotational mobility of the particle. For spherical particles,
r
(
t
) =
r
0
e
-
t
/[tau]rot
2
where
r
0
is the initial anisotropy and [tau]
rot
is the rotational correlation time.
Pulsed excitation was obtained from a cavity dumped dye laser, synchronously
pumped by a Nd:YAG laser (Coherent Antares 76-s). Two time-dependent fluorescence decays were collected through a polarization
filter mechanically rotated in the horizontal and vertical orientations. The
data were stored into two memory addresses in an IBM PC. A third decay was
collected at the magic angle (54.7o to the polarization orientation of the exciting beam).
In the data analysis, the fluorescence decay taken at the magic angle was fitted
to two exponentials. This data represents the fluorescence decay in the absence
of polarization effects and yields the fluorescence lifetime of ~22 ns, in good agreement with previous measurements for EtBr bound to
oligonucleotides (
22
). This represented ~80% of the total decay. The complete fit required a 20% contribution of a 5-15 ns component. The parameters computed from this fit of the magic
angle decay (lifetimes and pre-exponential factors) were then fixed in the simultaneous analysis of the
vertically and horizontally polarized fluorescence decays, where the anisotropy
function is expressed as a single exponential.
I
||
(
t
) = 1/3
I
0
(
t
)[1 + 2
r
(
t
)]
3
I
^
(
t
) = 1/3
I
0
(
t
)[1 -
r
(
t
)]
4
A simultaneous analysis of the polarized fluorescence decays insured an improved
accuracy on the recovered parameters. The [chi]
2
parameter, employed for judging the quality of the fit, was optimized by our analysis software using the Marquardt-Levenberg algorithm. A value of <1.4 is reasonable, while a value >2.0 indicates a poor fit.
Care was taken in these experiments to account for any residual free EtBr and to
ensure negligible depolarization due to energy transfer by minimizing the
number of molecules having more than one bound EtBr (Duhamel,J., Kanyo,J.,
Dinter-Gottlieb,G. and Lu,P., manuscript submitted).
The viscosities ([eta]) of the solutions were measured on a Brookfield Rheometer with a CP-40 cone. The temperature was regulated by a thermostated circulating
water bath. (L buffer: [eta](4oC) = 1.67 cP, [eta](22oC) = 0.93 cP, [eta](40oC) = 0.65 cP; D buffer: [eta](4oC) = 1.92 cP, [eta](22oC) = 1.06 cP, [eta](40oC) = 0.74 cP).
NMR data were obtained on a Bruker AMX500 spectrometer equipped with a 5 mm
1
H/
19
F probe. The
19
F base frequency was 470.27 MHz, with D
2
O serving as an internal lock. Typically 8192 complex data points were collected
over a frequency window of 4000 Hz. A relaxation delay of 1.0 s was utilized
and sufficient (16-256) transients were signal averaged to obtain an acceptable signal-to-noise ratio. The free induction decay was multiplied by an exponential window function having a time
constant of 5.00 Hz prior to Fourier transformation. All
19
F spectra were referenced to free FdU (20 mM) in L buffer, pH 6.7, as 0.0 p.p.m.
We have investigated the effect of solution conditions on the hairpin/duplex
equilibrium (Scheme
1
) for the r(CUUCGG) tetraloop and several analogs.
The oligonucleotides used in these studies are shown in Figure
2
. The original RNA dodecamer is TinRNA (r-I). Thermal melting data indicate that, although the RNA C(UUCG)G
tetraloop is unusually stable, its DNA analog, C(TTCG)G, is not (relative to
homopolymeric tetraloops) (
3
). Also, the loop region of the DNA hairpin is unstructured in solution, as
examined by a complete solution NMR analysis (
23
). There is no evidence for hydrogen bonding in the DNA loop, nor for stacking
of T6 on the C7 sugar, as in the RNA. Varani points out that the U-G base pair
in the RNA dodecamer involves not only a base-base hydrogen bond, but also a base-sugar hydrogen bond (to the 2'-OH of the first loop uridine residue), thus
explaining to some degree the loss of stability in analogs of the hairpin containing deoxyribose sugars (
10
). We therefore performed parallel experiments with the DNA analog (d-I), thinking that, by removing the stabilization energy of the loop, we
would be able to manipulate the transition to the double helical conformation
in solution.
Since we are exploring large size and shape changes, we examined the hydrodynamic behavior of the oligonucleotides when the solvent was changed
from that used in the solution structure determination to that of the crystal
growth. Rotational correlation times ([tau]
rot
values) for the RNA tetraloop and various analogs were compared with those of control molecules of defined size and shape.
The
K
d
values for EtBr dissociation from the DNA analogs of the dodecamer hairpin (d-I) and the fully complementary duplex (d-V) suggest that EtBr binds equally tightly to the hairpin and duplex conformations. Therefore, addition of the dye to the samples does not perturb the hairpin-dimer equilibrium.
The [tau]
rot
values were calculated by fitting decays of fluorescence anisotropy to a single exponential (equation
2
), i.e. the molecules were modeled as spheres. The rotational correlation time ([tau]
rot
) for a spherical molecule is directly proportional to the volume of the
rotating unit (
V
)
[tau]
rot
= [eta]
V
/
RT
5
where [eta] is the viscosity of the solution and
T
is the absolute temperature. The term
V
is both size and shape dependent.
The spherical model is a useful first approximation for following nucleic acid
conformation in the range of solutions of interest using fluorescence
depolarization, as demonstrated by studies of double helical DNA control molecules. Calibration curves of [tau]
rot
versus size (
BP
, number of base pairs) were constructed for eight DNA duplexes ranging in
length from 6 to 32 bp (axial ratios of 1:1-5:1). These lengths span the sizes of interest for the tetraloop analogs.
The experiments were carried out in both L and D buffers and at three different
temperatures (4, 22 and 40oC). For each set of buffer/temperature conditions, rotational times were
found to be linearly proportional to the number of duplex base pairs over the
range of helix lengths studied (Fig.
3
A and B). When temperature and viscosity of the solutions are taken into account
according to equation
5
, all of the experimental rotational times fall on a single line, as illustrated
in Figure
3
C. The average equivalent spherical volume occupied by 1 bp, calculated from the slope of this plot, is found to be ~1700 Å
3
, consistent with the volume of hydrated DNA helices (and equivalent to the
volume of a cylinder with a height of 3.4 Å and a diameter of 25 Å). Thus, the decay of fluorescence anisotropy of bound EtBr with a
spherical approximation is an internally consistent approach for following
conformation in the solvents of interest for these short oligonucleotides.
We turned to
19
F NMR to confirm and extend the fluorescence anisotropy decay data. This allowed
observations in the actual solvent environment of the crystal mother liquor.
Furthermore, because we have incorporated only one fluorouracil per strand,
19
F NMR spectra would show the presence of multiple conformations in slow exchange, which would have been hidden by the exponential fitting routines used for the fluorescence data analysis.
There are three sites (uracils or thymines) on each dodecamer which can be
followed using
19
F NMR. We would expect the greatest change upon duplex formation for the
19
F nucleus at uridine 6 (see Fig.
1
), which is unpaired in the solution structure and stacked as a U-C base pair in
the crystal structure. In the hairpin conformation, that base is stacked on a
sugar, while in the duplex conformation, it is stacked between two other bases.
Finally, since the ribose of U6 is in a C2'-
endo
conformation in the RNA tetraloop, substitution with a deoxyribose residue
should not be deleterious to the structure. We found that the fluorine label at
that position has little effect on either the RNA (r-I.F6) or the DNA (d-I.F6) structure by comparison of NOESY cross-peaks in the aromatic H6/H8 to sugar H2'/H2'' region for the labeled versus unlabeled
oligonucleotides (data not shown).
We looked for evidence of a conformational change in the oligonucleotide when
the solvent was changed in steps from that used in the solution NMR structure
determination to that of the crystal growth. Under the various conditions,
spectra were acquired as a function of pH (6-10 range) for the fluorinated RNA and DNA dodecamers (r-I and d-I), for the tethered species (r-II and d-II) and for the control samples FdU and the fully
complementary duplex (d-V). In this pH range each species gives one peak per fluorine probe.
Therefore, we found no evidence for a double-stranded duplex coexisting with a hairpin in the 6-10 pH range in any of the buffers for either the RNA or DNA
dodecamer of the Tinoco RNA sequence or for the tethered analogs. For all of
the tetraloop analogs, the apparent p
K
a
of the imino proton of the FdU base, the midpoint of the titration curve, was
shifted lower by salt addition (100 mM sodium citrate and 50 mM Tris). Addition
of 2.5 mM MgCl
2
and 20% PEG400 brought about no further change to the titration curves.
19
F NMR chemical shift versus pH curves of the tethered DNA and RNA are superimposable on those of the non-tethered species. Thus, the tethering did not bring about a transition to
the extended duplex conformation for either the RNA or DNA, but each half of
the molecule adopted an independent hairpin structure, as the fluorescence data
showed.
Since over 2 M monovalent cations must accompany the RNA in the crystal to
ensure charge neutrality, we subjected tethered RNA and DNA species,
geometrically equivalent to ~100 mM strands, to 2.5 M NaCl in D duffer. Selected RNA spectra are shown
in Figure
4
. Upon tethering two RNA analogs and adding 2.5 M NaCl, the RNA shows a
significant second component at -1.5 p.p.m. (Fig.
4
D) not seen under identical conditions for the RNA dodecamer (Fig.
4
C), indicating that the tethering to increase the apparent concentration of the
dodecamer was necessary to obtain the second spectral component. The DNA analog
of the tethered sequence shows a broad second component upfield, although the
effect is less pronounced than for the tethered RNA. The relative proportions
of the two components is not altered in 20% PEG (data not shown). To summarize,
evidence for a second solution conformation, presumably the extended hairpin
which mimics the double helix in the crystal, was obtained only after tethering
two dodecamers together in order to increase the local oligonucleotide
concentration and after the addition of 2.5 M NaCl to mimic the cation
concentration in the crystal.
We demonstrate that an RNA CUUCGG tetraloop and its DNA analog maintain hairpin
conformations in a continuum of solvent conditions, including conditions
closely approximating those used to grow crystals. This contrasts with the
double helices observed by single crystal X-ray diffraction.
In order to mimic oligonucleotide concentrations in the crystal, we constructed
tethered RNA and DNA species, geometrically equivalent to ~100 mM strands. Although the tethered RNA remains as two independent
hairpins under both the solution NMR and crystallization conditions, a significant second component, presumably the extended hairpin which mimics the double helical conformation in
the crystal, appears in our NMR spectra if additional NaCl (2.5 M) is added to
the crystallization buffer. The same holds true, although to a lesser extent,
for the tethered DNA analog. Although counterion condensation theory predicts
that the cation concentration at the surface of the DNA helix is constant,
regardless of bulk ion concentration, Honig and Nicholls (
24
) remind us that the distribution near the molecular surface (which is a
function of bulk ion concentration) does play a role in polynucleotide
conformation.
Careful attention must be paid to the conditions under which nucleic acid
structures are solved. There are at least two other cases where related RNA
tetraloops show double helices in the diffraction-derived structure. Fujii
et al.
(
20
) claimed that r(UGAGCUUCGGCUC) is a double helix with 5' U residues unpaired in their crystal structure and a hairpin with an
unpaired 5' U in solution when examined by NMR spectroscopy. Although solvent
conditions were not given for either experiment, they note that the `duplex is
favored by high RNA and salt concentrations', an identical conclusion to that
drawn by Holbrook
et al.
(
18
). Similarly, the sequence r(GCUUCGGC) is a hairpin in solution and a double
helix in the crystal form, even though it is short and thus presumably has less
base pairing stabilization in the stem (
21
). A similar dimorphism occurs in the case of DNA. X-ray diffraction (using 20 mM potassium cacodylate, pH 7.0, 10 mM MgCl
2
, 6 mM spermine, 40 mM KCl and 5-40% 2-methly- 2,4-pentane diol) and solution NMR (with 50 mM NaCl, pH
7.0) yielded structures for the G quartet with very different pathways for the
phosphodiester backbone (
25
,
26
).
Holbrook and
co-workers (
18
) state that high RNA and salt concentrations favor an intermolecular double
helix over an intramolecular hairpin loop conformation, without suggesting how
high the concentrations should be. Our experiments demonstrate that although
high RNA and salt concentrations do shift the solution equilibrium toward the
bimolecular double helix, they are not sufficient to make the duplex the major
conformation. The duplex conformation adopted by the RNA dodecamer in the
crystal must also be a consequence of crystal contacts. Since the predominant
species under all solution conditions is the hairpin, we believe that the
duplex conformation seen in the crystal results from crystallization of a
secondary component of the crystallization mixture. Perhaps the ability of the
duplexes to stack end-to-end in an orderly array selectively precipitates that conformation.
+
Present address: Department of Chemistry, University of Waterloo, Waterloo,
Ontario, Canada
REFERENCES
Return