of 27
RNA origami design tools enable cotranscriptional folding of
kilobase-sized nanoscaffolds
Cody Geary
1,2
,
Guido Grossi
1
,
Ewan K. S. McRae
1
,
Paul W. K. Rothemund
2,*
,
Ebbe S.
Andersen
1,*
1
Interdisciplinary Nanoscience Center and Department of Molecular Biology and Genetics,
Aarhus University, Aarhus C, 8000, Denmark
2
Bioengineering, Computing + Mathematical Sciences, and Computation & Neural Systems,
California Institute of Technology, Pasadena, CA 91125, USA
Abstract
RNA origami is a framework for the modular design of nanoscaffolds that can be folded from a
single strand of RNA, and used to organize molecular components with nanoscale precision.
Design of genetically expressible RNA origami, which must cotranscriptionally fold, requires
modeling and design tools that simultaneously consider thermodynamics, folding pathway,
sequence constraints, and pseudoknot optimization. Here, we describe RNA Origami Automated
Design software (ROAD), which builds origami models from a library of structural modules,
identifies potential folding barriers, and designs optimized sequences. Using ROAD, we extend the
scale and functional diversity of RNA scaffolds, creating 32 designs of up to 2360 nucleotides, five
that scaffold two proteins, and seven that scaffold two small molecules at precise distances.
Micrographic and chromatographic comparison of optimized and nonoptimized structures
validates that our principles for strand routing and sequence design substantially improve yield. By
providing efficient design of RNA origami, ROAD may simplify construction of custom RNA
scaffolds for nanomedicine and synthetic biology.
The field of RNA nanotechnology began by extracting RNA structural modules from natural
RNA molecules and connecting them to create engineered constructs
1
,
2
,
3
. This approach
was enabled by the structural determination of biological RNA molecules, such as the
ribosomal subunits
4
,
5
, which provided a large library of RNA modules to build from. With
these modules, architectures ranging from multi-stranded tiles to single-stranded origami
have been explored. Of particular recent interest are RNA structures designed to fold
cotranscriptionally, during their synthesis by RNA polymerase, which have the benefit that
they can be genetically expressed and folded within cells. Previously, we introduced the
Users may view, print, copy, and download text and data-mine the content in such documents, for the purposes of academic research,
subject always to the full Conditions of use:
http://www.nature.com/authors/editorial_policies/license.html#terms
*
correspondence to: pwkr@dna.caltech.edu and esa@inano.au.dk.
Author Contributions
C.G., P.W.K.R and E.S.A conceived the project. C.G., G.G., and E.K.S.M. performed the research. P.W.K.R and E.S.A supervised the
project. C.G., P.W.K.R and E.S.A wrote the manuscript. All authors analyzed the data and commented on the manuscript.
Competing Interests
The authors declare no competing interests.
Europe PMC Funders Group
Author Manuscript
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Published in final edited form as:
Nat Chem
. 2021 June 01; 13(6): 549–558. doi:10.1038/s41557-021-00679-1.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
RNA origami method
6
—a highly regular architecture that arrange RNA helices into parallel
arrays held together by crossovers and kissing loops (KLs)—which is compatible with
cotranscriptional folding, but several bottlenecks in computational design methods limited
the size (450 nt) and folding yield. Later studies constructed somewhat larger (715 nt)
wireframe single-stranded cotranscriptional shapes by composing complex tertiary motifs
in
vitro
7
and
in vivo
8
. The largest currently achieved structures (6000 nts) require long (~18-
hour) thermal anneals
9
making them incompatible with cotranscriptional folding in cells.
RNA nanostructures can serve as functional scaffolds by directly incorporating RNA-protein
binding domains
10
,
11
, small molecule aptamers
12
,
13
, biosensors
14
, ribozymes
15
, siRNAs
16
,
or combinations of such modifications to create multifunctional nanoparticles
17
,
18
. RNA
nanostructures that fold cotranscriptionally
6
,
7
have been expressed in cells
8
,
12
where they
have the potential to be used as biosensors, scaffolds or regulators for synthetic biology
applications
19
—for example, to control product formation from colocalized enzymes
20
,
21
and perform gene regulation via recruitment of transcription factors
22
. To verify that two
proteins are located on the same scaffold, split fluorescent proteins
23
or Förster resonance
energy transfer (FRET) between fluorescent proteins
24
are often used. Similarly, fluorescent
RNA aptamers (split-Spinach
25
and apta-FRET
12
) have been used to verify scaffolding
effects. RNA origami structures may incorporate tertiary motifs such as the IRES
13
or
bKL
26
motifs to produce ~90° bends that allow out-of-plane functionalization. 2’-fluoro-
modified RNA origami scaffolds carrying the thrombin aptamer have been used to produce a
potent therapeutic anticoagulant
11
.
Computational methods have played a central role in developing RNA nanotechnology by
facilitating core tasks
27
,
28
. Dedicated software has been developed to ease the construction
of RNA nanostructures from 3D structural motifs
29
,
30
,
31
. However, no software exists for
the interactive 3D modeling of large and regular RNA scaffolds such as the RNA origami
architecture. Algorithms simulating RNA cotranscriptional folding have been developed for
predicting folding pathways,
32
,
33
which for small structures enables designers to verify that
their sequences will avoid kinetic traps; it has not been possible to do this for RNA origami.
RNA sequence design algorithms were originally developed based on secondary structure
thermodynamic folding algorithms
34
, but these lack the ability to efficiently predict
pseudoknots (such as KLs). RNA origami, which are stabilized by numerous KL interactions
along their strand path, necessarily contain numerous pseudoknots and are therefore not easy
to design. Another important element for RNA sequence design is the ability to incorporate
numerous sequence constraints to allow RNA sequence and structural motifs to be added,
but current design pipelines lack the ability to simultaneously incorporate the multiple
constraints necessary for the design of RNA origami structures
6
,
23
,
35
.
Here we introduce the RNA Origami Automated Design software (ROAD)—a computer-
aided design software to automate the 3D modeling of structures, analyze folding paths, and
design sequences and KLs that fold into the designated structures—allowing us to greatly
extend the scale and diversity of RNA origami scaffolds. ROAD allows us to rapidly
prototype multiple distinct scaffolds, and investigate the effects of different design
parameters with a short design cycle. We study the effect of curvature and cross-over
placement within RNA origami structures by atomic force microscopy (AFM) allowing us to
Geary et al.
Page 2
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
greatly increase the scale of the structures. To study the effect on yield, we then constructed
a set of non-optimal designs and analyzed yields by size-exclusion chromatography (SEC)
and negative stain transmission electron microscopy (TEM). Finally, we tested the ability of
ROAD to design RNA origami scaffolds embedded with aptamers for binding fluorescent
proteins and small molecule fluorophores, and used FRET between the fluorescent
molecules as a distance indicator to validate the precision of scaffolding.
Results
Design tools for creating RNA origami scaffolds
We have developed the ROAD software package (see code availability) to automate the main
design steps for RNA origami: model building, folding path analysis, and sequence design.
ROAD is based on a library of compatible structural modules used to construct RNA
origami structures. Core modules such as helices, junctions, and 180KLs (KLs that interact
at an angle of 180°)
36
are used to build the central scaffold (Fig. 1a) and peripheral modules
such as tetraloops
37
, 120KL-connectors (KLs that interact at an angle of 120°)
38
, light-up
aptamers
39
,
40
and protein binding aptamers
41
,
42
are used to add functionality (Fig. 1b).
Schematic representations of the core modules can be used like Lego bricks to compose a
large diversity of different designs (Fig. 1c) that directly translate to atomic coordinates (Fig.
1d). Closely spaced crossovers between three helices result in ‘dovetail’ (DT) junctions
6
(Fig. 1e), which is an important design parameter for RNA origami, since the DT length (in
bps) changes the dihedral angle
Φ
between connected helices (Fig. 1f). DTs can only have
certain lengths to avoid helices to sterically clash, and are named
s
DT, where the spacing
s
can have values from -5 to +2 bps
43
(Supplementary Fig. 4).
The ROAD software package consists of three main algorithms: RNAbuild, RNApath and
Revolvr, that take a user-specified ‘RNA blueprint’ as input. RNA blueprints are text-based
diagrams that encode all Watson-Crick base pairs (bp), sequence constraints, pseudoknots,
base stacking at junctions, and 5’ to 3’ strand orientations (Fig. 1g). RNAbuild uses a
module library to build atomic models according to specifications in the blueprint (Fig. 1h).
The automated atomic modeling helps the user to design curvature and avoid steric clashes
within larger RNA structures that are otherwise unapparent in the RNA blueprint. RNApath
analyzes the folding path for potential topological barriers that may arise during the
cotranscriptional folding process (Fig. 1i). Topological barriers can arise if a KL interaction
(Fig.1g blue arrow) forms before the formation of helices in the loop region (Fig. 1g, pink
and orange arrows) since the formation of a double helix may be sterically hindered by the
closed loop region. RNApath determines topological barriers based on the relative rates of
KL and helix formation, as well as the speed of synthesis, and generates plots and 3D
folding animations (Supplementary Videos 1–6) to help the user avoid topology-based
misfolding. Revolvr is a sequence design algorithm that uses a multi-stage sequence
optimization procedure involving positive design by minimum free energy (MFE)
44
prediction, negative design by sequence symmetry minimization (SSM)
45
, and KL
orthogonalization, to develop a sequence that folds into the target structure (Fig. 1j). The
ROAD package and the analysis scripts are described in the Methods section, a tutorial is
Geary et al.
Page 3
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
provided as Supplementary Note 1, and a webserver has been established to make the
software easily accessible (see code availability).
Design of multivalent interfaces for RNA origami tiles
The ROAD software was used to design three-helix (3H) RNA origami tiles with edge
interactions to form fibers or rings, which make them easier to observe by AFM imaging. To
make the interactions stronger the 3H tiles were connected by two 120KL interactions
38
;
their relative in-plane positioning defines the tile-tile interaction angle
θ
(Fig. 2a) that can
deviate from 120° since the KL motif is flexible enough to accommodate a range of
angles
46
. Using RNAbuild, we designed three trapezoidal 3H tiles with different tile-tile
interaction angles (
θ
= 120°, 135°, 108°) that form closed polygonal objects (blue and white
models in Fig. 2c–e, Supplementary Fig. 6). The different
θ
angles were made possible by
changing the tile geometry by different DT spacings (named 3H
s
DT, where
s
= -2, -3 and -4
bps corresponding to
Φ
= 155°, 122°, and 89°, respectively). For characterization of the
designs, we introduce a new near native sample preparation protocol for AFM imaging to
capture structures formed in the transcription reaction on the mica surface (Fig. 2b). AFM
experiments (Fig. 2f–h, Supplementary Fig. 5) showed that of polygons observed for
3H-2DT (n = 27), 59% were hexagons, 30% were pentagons, and 11% were heptagons; for
3H-3DT only a few octagons were observed, but most tiles participated in open structures
that we interpret as helical fibers; for 3H-4DT (n = 72), 69% were pentagons, 26% were
hexagons, and 4% were heptagons or quadrilaterals (Supplementary Fig. 6). The data show
that 120KLs can be used to create multivalent binding interfaces with
θ
from 108° to 135°.
The folding yield of the individual RNA origami tiles was estimated to be 72-89% by
counting of well-formed versus broken structures in AFM images (Supplementary Fig. 12–
14, Supplementary Table 3).
Expanding the size of RNA origami structures
Motivated by a desire to make scaffolds large enough for organizing multiple proteins, we
explored geometric details and design approaches important for scaling up RNA origami.
The modular combination of smaller, already validated RNA motifs is a common and
successful approach to the design of larger structures
47
,
48
. Here, starting with domains from
tile 3H-2DT (Fig. 2c,f), we hierarchically applied duplication and fusion (Supplementary
Fig. 8) to design sets of taller and wider scaffolds (Fig. 3a). Extension of RNA origami in
the
x
-direction required no geometric innovation, but extension in the
y
-direction required
consideration of
Φ
-based curvature when adding multiple rows of helices. DTs that
alternates between 0 and -2 bp DTs result in minimum curvature of the RNA origami, but
unfortunately 0 bp DTs introduce a potential weakness into an RNA origami, since each 0
bp DT is effectively a 6-arm junction with at least three sterically plausible alternative
stacking conformations (
cf.
the two stacking isomers observed in 4-arm junctions
49
). To
better stabilize and specify desired folds, we used larger ‘offset DTs’—DTs displaced by a
helical turn of RNA, which maintain the same dihedral angle
Φ
as shorter DTs (e.g. -11 bp
and +11 bp DTs rather than 0 bp DTs).
To reach five helices tall, two copies of 3H-2DT were merged via +11 or -11 bp offset DTs
to create tiles ZigZag-A-1X and ZigZag-B-1X, respectively (Fig. 3b,c, Supplementary Fig.
Geary et al.
Page 4
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
8). Three 120KLs added to the edges of these tiles were programmed to join the tiles in a
trans
configuration, resulting in zigzag-shaped filaments (Fig. 3b,c, blue and grey models) in
which alternating tiles face up and down, a corrugated configuration which balances tile
curvature (cf. previous polygons in Fig. 2c–e in which all tiles face in the same direction).
Samples were imaged by AFM and analyzed to show a folding yield of 77-85% similar to
the 3H-2DT tiles (Fig. 3b,c, Supplementary Fig. 15–16). A few alternative 6H tiles that
contained isolated 0 bp DTs were shown to fold well (Supplementary Fig. 7 combining +11,
0, -2, -11 and -13 bp DTs). To create still taller tiles, two copies of ZigZag-A-1X were
merged, via -11 bp DTs, to create the core of a nine-helix tile (Supplementary Fig. 8).
Addition of 120KLs resulted in
trans
connections and filaments of alternating up-down
orientations for ZigZag-B-9H tiles (Fig. 3d); addition of 180KLs resulted in
cis
connections
and filaments of consistent orientation for Ribbon-9H tiles (Fig. 3e). The 9H tiles showed
more partial structures and had a reduced folding yield estimated to be 51-62%
(Supplementary Fig. 17–18), which could be caused by topological folding barriers (marked
in red and orange in Supplementary Video 1) as suggested by RNApath analysis.
We used lateral duplication and fusion (Supplementary Fig. 8) of ZigZag-B-1X to create
tiles with two repeats (ZigZag-B-2X in Fig. 3f and Supplementary Fig. 10; Supplementary
Fig. 11 shows unexpected edge interactions) or four repeats (ZigZag-B-4X in Fig. 3g and
Supplementary Fig. 10). The 2X duplication did not seem to affect yield (estimated to be
78%), whereas the 4X duplication had a reduced yield of 58% (Supplementary Fig. 19–20).
The reduction in yield of the large 12 nm x 48 nm ZigZag-B-4X could not be explained by
RNApath analysis (Supplementary Video 2), but is more likely caused by the misfolding and
aggregation of its long transient 5’ single stranded end. Tiles with alternating -2 bp and
+/-11 bp DTs will be flat but have steeply sloped sides. To obtain a more rectangular tile we
replaced each -2 bp DTs with +9 bp DTs (-2 bp offset by +11 bp), so that every repeat unit
has a counterbalanced set of +9 bp and -11 bp DTs. As an example of this architecture we
designed the three-repeat Ribbon-5H-3X with 180KLs connectors resulting in straight linear
chains as observed by AFM with a yield of 42% (Fig. 3h, Supplementary Fig. 21). As a
second example the tile was extended to 9 helices tall and designed without intermolecular
connections, as a standalone scaffold, reaching a length of 2360 nts and a size of 20 nm x 36
nm (Fig. 3a, Rectangle-9H-3X), however, the expansion resulted in only few examples of
rectangular shapes that all had folding defects (Fig.3i). Finally, we designed RNA origamis
with shorter or longer double crossover spacing: ZigZag-B-2X-Mini with two-turns between
crossovers (Supplementary Figs. 7) and Ribbon-5H-3X-bumps with four-turns between
crossovers (Supplementary Figs. 7 and 10). The latter was designed with six out-of-plane
dumbbells placed in the middle of the 4-turn stem regions, however, the complexity of the
design resulted in a low observed yield of 30% (Supplementary Fig. 22), and the three-
dimensionality of the design resulted in poor imaging by AFM.
The manual evaluation of folding yields from the AFM images is summarized in
Supplementary Table 3. The folding yields negatively correlates with increasing length of
the RNA origami structures tested, and with RNApath-predicted topologically blocked
positions. This observation is supported by an apparent correlation between the number of
observed misfolded structures and RNApath-predicted topologically blocked positions. The
data indicates that folding topology is important and that increasing the height of the tile
Geary et al.
Page 5
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
results in the increased occurrence of predicted topological barriers, that arise because of the
longer delay between the synthesis of KL partners. The large Rectangle-9H-3X was
predicted to have several topological barriers, this correlated with the larger folding defects
observed (Fig.3i, Supplementary Fig. 23). Another example is a merged version of ZigZag-
A-1X and ZigZag-B-1X that is 10 helices tall, where we observe partly-formed tiles that
again have large defects that seem to correspond to the regions with predicted topological
barriers (Supplementary Fig. 23).
Effects of design parameters on folding yield
To support our AFM yield analysis, we performed negative stain TEM imaging of SEC-
purified monomer RNA origami structures. A monomeric 5-helix scaffold (5HS, Fig. 4a),
based on one of our best performing RNA tiles ZigZag-B-1X (85% yield by AFM) resulted
in 86% yield of monomer as determined by SEC analysis (Supplementary Fig. 24) and TEM
images of the monomer sample revealed homogeneous and monodisperse particles with
class averages displaying highly resolved details of tight helix packing (Fig. 4c and
Supplementary Fig. 24). The TEM analysis revealed a clear preference for observing either
front or back face views of the 5HS structure (Fig. 4c, Supplementary Fig. 27 for plot of
orientation distribution) even though a few edge views were observed as well
(Supplementary Fig. 24). Although we were not able to obtain an
ab initio
model, a 3D
reconstruction could be made by using the theoretical model as input search volume (Fig. 4b
and Supplementary Fig. 24).
To investigate the robustness of the RNA origami method in relation to core design
parameters we generated a challenging monomeric design with 5-helices and 2-KL columns,
with an unconventional meandering strand path and generated two different versions with
different 5’ start sites (Path1 shown in Fig. 4d and Path2 shown in Fig. 4e). The two strand
paths are equivalent in 3D structure, but the different positioning of the 5’ start sites (Fig.
4d,e, blue circles) has a large effect on folding topology as predicted by RNApath. During
transcription, Path1 has a long transient 5’ single strand but no predicted topological barriers
(Fig. 4f, Supplementary Video 3), whereas Path2 has no transient 5’ single strand but has
substantial topological barriers predicted (Fig. 4g, orange and red regions, Supplementary
Video 4). Previously, we have avoided designs with a long 5’ transient single stranded region
because transient single strands are expected to increase aggregation during cotranscriptional
folding
7
. To investigate the effect of sequence design optimization, a third design was
created based on Path1 satisfying the MFE structure (stage 1-4 of Revolvr, see Methods and
Supplementary Fig. 1–2) but lacking the final KL optimization (stage 5 of Revolvr). As
expected these designs display a substantial amount of aggregation that resulted in relatively
low yield of monomers of 26-44% as determined by SEC analysis (Fig. 4k,l), which can be
compared to 5HS which displays 86% monomer yield by SEC analysis (Fig. 4l and
Supplementary Fig. 24). The monomers were observed to be stable post SEC purification
(Fig. 4k) indicating that aggregation is happening during the cotranscriptional folding
process and is not the result of a subsequent equilibration process.
TEM imaging was performed on the purified monomer and aggregate peaks and the
monomers were observed to be monodisperse (Supplementary Fig. 25). To be able to
Geary et al.
Page 6
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
address the folding yield, the TEM grids were prepared from the same concentration of
purified RNA samples and quantified from the same number of acquired images. Unbiased
blob picking was used to identify particles and 2D class averages showed that the number of
face views of the RNA origami structure were very different between the three samples and
that several alternative particle views and shapes were observed (Fig. 4h–j and
Supplementary Fig. 26). Since each design has the same predicted 3D structure, they should
have the same angular distribution on the foils. We used the number of, easily recognizable,
face views observed in the 2D class averages as an estimate of the cotranscriptional folding
yield of the three samples (Fig. 4m, see Methods for description of folding yield
calculation). Path1 with an optimized sequence had a 30% folding yield. Path2 with
optimized sequence had 25% folding yield, but of these only 3/4 adopted the designed
structure, whereas 1/4 displayed a “purse-handle” phenotype (Fig. 4i, blue arrow) that we
suggest to correspond to distortions in the long topologically blocked helix (Fig. 4e) due to
partial inhibition of Watson-Crick base pairing (Fig. 4g, blue arrow). Path1 with non-
optimized KL sequences had a reduced folding yield of 6% (Fig. 4m) and a large fraction of
alternative shapes (Fig. 4j right, Supplementary Fig. 26).
From the limited, but equivalent, datasets acquired for each design only the particles picked
from the Path1 data produced a reasonable
ab initio
reconstruction (Fig. 4n). As observed
previously in the TEM analysis of the 5HS, the Path1 structure had preferential face
adsorption to the carbon foil, but in this case one face was strongly preferred (see
Supplementary Fig. 27 for plot of orientation distribution), which indicates that the larger
monomer structure has an asymmetric shape in solution that affects the adsorption to the
carbon. Although the tested designs can all fold into the correct 3D structure, the choice of
strand path and sequence optimization have large effects on both the yield and structural
homogeneity of the origami particles.
Scaffolding of proteins and small molecules
To test the ability of RNA origami to scaffold proteins, we used the high-yield 5HS scaffold
(Fig. 4a) containing 10 hairpin sites that can be used for functionalization (Fig. 5a).
RNAbuild was used to design a series of five scaffolds that positioned two different protein-
binding aptamers at increasing distances of approximately 2.5, 5, 7.5, 10, and 22 nm using
scaffolds named M
x
P
y
, where
x
refers to the position of the MS2 aptamer
41
and
y
to the
position of the PP7 aptamer
42
(Fig. 5b,c). All scaffolds were designed by Revolvr to have
unrelated sequences, except for the fixed sequence of the aptamers. Similar to a previous
scaffolding study
24
, we fused mTurquoise2
50
(a Cyan Fluorescent Protein, CFP) and YPet
51
(a Yellow Fluorescent Protein, YFP) with viral coat proteins MS2 Coat Protein (MCP)
41
and
PP7 Coat Protein (PCP)
42
, respectively (Fig. 5b,c, sequences in Supplementary Table 8).
When the M5P3 scaffold was transcribed in the presense of excess fluorescent proteins it
resulted in a FRET signal that reached saturation after 20 min (Supplementary Fig. 28),
showing that the scaffold cotranscriptionally folds and brings the two proteins together
within FRET distance. To compare several RNA scaffolds, we normalized concentrations of
co-transcriptionally folded RNA products and incubated them with excess amounts of
fluorescent proteins. The FRET signal was observed to generally decrease with increasing
distance between aptamers (Fig. 5d, full spectra in Supplementary Fig. 30), however, some
Geary et al.
Page 7
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
constructs with a spacing differing by ~2.5 nm were not significantly different in FRET
signal (M5P4
M5P3; M5P2
M5P1, p ≥ 0.05 in Student’s t-test), and the control
constructs, M5P10 with a nominal distance beyond the Förster radius and 5HS with no
aptamers, showed measurable levels of FRET. These non-ideal effects may be explained by
the large size of the fusion proteins with long linkers used as well as the documented
tendency of the fluorescent proteins to form dimers in a colocalized context
52
. In general,
the results may also be affected by scaffold flexibility and sequence-specific conformations
of particular constructs.
RNAbuild was used to design two series of scaffolds that positioned the fluorescent
aptamers Spinach
53
,
54
and Mango
40
in various structural contexts (Fig. 5e–h,
Supplementary Fig. 29). The first series was based on a two-helix scaffold S2T (short, 2
turns) with short stems to position Spinach and Mango aptamers and 2 helical turns between
crossovers (Fig. 5e), which was previously shown to produce a strong FRET signal between
the fluorophores DFHBI-1T and YO3-biotin
12
. Two variations of the S2T scaffold were
produced: S3T (short, 3 turns) with wider crossover spacing (Fig. 5f), and L3T (long, 3
turns) with longer stems for positioning fluorescent aptamers and wider crossover spacing
(Fig. 5g). The S2T scaffold transcribed in the presence of fluorophores shows slowly
increasing fluorescence and FRET signals over at least 90 min (Supplementary Fig. 28),
which is likely caused by the slow folding of the fluorescent aptamers. To compare several
RNA scaffolds, we normalized the RNA concentrations before incubation with excess
amount of fluorophores. Fluorescence measurements show ~35% FRET for S2T, ~30%
FRET for S3T and ~5% FRET for L3T scaffolds (Fig. 5i, full spectra in Supplementary Fig.
30). While RNAbuild models predict that all three scaffolds have the same distance and
orientation between donor and acceptor fluorophores (Fig. 5e–g), the large decrease in
FRET signal with increasing construct size suggests that scaffold flexibility (due to longer
stems and to a lesser extent larger crossover spacing) strongly influences the FRET signal.
The second series was based on the three-helix scaffold from Fig. 2 with fluorescent
aptamers placed on top and bottom helices and 2 turns between crossovers (L2T
s
DT in Fig.
5h, Supplementary Fig. 29). This scaffold is able to tune fluorophore spacing (from 1.3 to
3.2 nm in increments of 0.6 nm) by changing DT length (from
s
= -5 bp to -2 bp in
increments of 32.7°), respectively. Fluorescence measurements for the L2T
s
DT scaffolds
show a decrease in FRET signal as the predicted distance between the fluorophores is
increased (Fig. 5i) (statistically significant p<0.05 in Student’s t-test except for L2T-3DT).
Within this series, care was taken to maintain the relative orientation of the Spinach and
Mango aptamers to avoid the possible effects of oriented dipoles on FRET
55
(Supplementary
Fig. 29). Comparing between series, we attribute the low FRET signal of the sterically
overlapped construct L2T-5DT relative to construct S2T which shares a similar crossover
spacing, primarily to the flexibility contributed from a longer aptamer bearing arm.
Discussion
The design and synthesis of cotranscriptional RNA structures in high yield is very
challenging, and in our previous work
6
we were only able to achieve cotransctiptional folds
of 440 nts in length with yields so low that only a few correctly formed objects could be
identified. In the current work we have improved the RNA origami method to greatly expand
Geary et al.
Page 8
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
both the size and functional complexity of RNA nanostructure designs, as well as
dramatically improving the yields of correct products that are able to be produced by
cotranscriptional folding. We have rapidly prototyped 32 different RNA origami designs in
this work, which allowed us to explore the effect of multiple RNA origami design
parameters: DT geometry, multivalent interfaces, taller and wider structures, different strand
routing strategies, as well as designs incorporating aptamers for scaffolding proteins and
small molecules. The achievements were enabled by the development of the ROAD software
package, comprising the programs RNAbuild, RNApath and Revolvr that work together to
facilitate the design of large and complex RNA structures and were all found to be crucial
for obtaining high-yield RNA scaffolds.
RNAbuild automates the rough 3D modeling of RNA origami structures, which were
previously constructed by hand, allowing us to design much larger and more sophisticated
designs than before. In this work we demonstrated that the DT seam can be used to adjust
the curvature of RNA origami structures to tune tile-tile interaction angle to form rings of
defined size (Fig. 2) and to tune the distance between attached fluorescent aptamers (Fig.
5h,i). RNAbuild further allowed us to expand the RNA origami architecture by domain
duplication and fusion (Supplementary Fig. 8) reaching sizes of approximately 2000 nts
albeit with decreasing yields as estimated from AFM images (Fig. 3; Supplementary Table
3). Interestingly, TEM analysis revealed preferred landing of larger RNA origami structures
on the carbon film (Fig. 4h; Supplementary Fig. 27), which indicates that larger RNA
origami structures may have a curved structure in solution. Even though
ab initio
reconstruction from TEM images showed that the RNA origami structures are flat this may
be an artifact of the deposition on the carbon film. RNAbuild can in the future be improved
by extending the library of functional motifs and by supporting for alternative architectures
such as parallel crossover RNA origami
9
and wireframe RNA origami
8
as well as allowing
physical simulation of the structures to address strain-induced distortions using e.g.
oxRNA
46
. The recently developed program RNAmake—which specializes in grafting and
stabilizing tertiary motifs onto an input model
31
—could complement and extend RNAbuild.
RNApath makes a simple folding path analysis based on the RNA blueprint (while not
taking account of the designed sequence) to predict possible topological barriers for the
cotranscriptional folding process. Comparing the number of predicted topological barriers to
the folding yield estimation from AFM images revealed a strong correlation, where the most
severe cases did not result in any correctly folded objects (Supplementary Table 3).
However, the effect of size and number of topological barriers could not easily be separated
in this evaluation, since topological barriers arise when designs get larger (and especially
taller). The effect of folding path choice was investigated further by designing an RNA
origami structure with two alternative folding paths. TEM analysis revealed that there was
an approximately 30% decrease in folding yield for the path with topological barriers (Fig.
4m) and that misfolds could be observed with severe distortions of the topologically trapped
helix (forming a “purse handle”) (Fig. 4i). The observation that only the structure without
topological barriers resulted in a reasonable 3D reconstruction further underscores the
importance of taking this design parameter into account (Fig. 4n). The kinetic folding
analysis of RNApath may be improved by using thermodynamic kinetic folding algorithms
like Kinefold
32
or by using coarse-grained molecular simulations like oxRNA
46
.
Geary et al.
Page 9
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
Revolvr designs sequences for RNA blueprints with a high content of pseudoknots - a task
that has not been approached by any other RNA design program. Revolvr solves this task by
using a multi-stage sequence optimization procedure involving MFE-based positive design,
SSM-based negative design, and KL orthogonalization, that makes it very efficient in the use
of computational time (Supplementary Fig. 31). The sequence design by Revolvr has a high
success rate, since most of the structures presented in this study worked on first try. The high
success rate prompted us to design new sequences for each new RNA origami for
scaffolding of proteins and small molecules with the assumption that the geometry of the
RNA origami and not the precise sequence was important, which was verified to some extent
by the overall ability to control distances on the scaffolds. The effect of sequence design was
investigated by designing a non-optimal sequence, where only the last stage of the 5-step
design procedure (KL orthogonalization) was omitted. Here the folding yield was observed
to decrease from 30% to 6%, showing that this sequence design step has a substantial effect.
Future improvements to Revolvr sequence design could be to include an RNA secondary
structure partition function in the design optimization, like NUPACK
35
. The partition
function optimization may become especially important when designing more challenging
RNA structures with smaller stem regions. A great future challenge would also be to include
pseudoknot-prediction and kinetic folding simulation directly in the sequence design
algorithm.
While RNA nanostructures have previously been used to scaffold protein binding
aptamers
10
,
11
and small molecule aptamers
12
,
13
, here we demonstrate distance control by
changing the position of aptamers on parallel helix ends and by tuning the DT length to
gradually change the distance between helices. Our FRET studies highlight a potential size-
flexibility tradeoff in RNA scaffold design: based on our current architecture, larger
structures enable complex spatial arrangements of proteins to be constructed, but smaller,
more rigid structures are required if more precise distances are desired. The elaboration of
RNA origami to multilayer 3D structures may obviate this tradeoff by achieving
simultaneously large and rigid structures, as has been achieved for DNA origami design
56
.
Rigidity and precision of arrangement will also be improved by exchanging large, flexible,
dimeric linkers such as MS2 and PP7 aptamer-protein constructs with smaller, monomeric
RNA-binding proteins or peptides such as L7Ae
10
or BIV-Tat
57
. With improved protein
scaffolding methods, the RNA origami scaffolds may be used to control product formation
from colocalized enzymes
20
,
21
and perform gene regulation via recruitment of transcription
factors
22
.
In this study we have improved the RNA origami method to allow the design of
cotranscriptionally folding RNA nanostructures approaching the size of ribosomal RNAs.
However, the structural complexity and strategies for cotranscriptional folding of RNA
origami and the ribosome are very different. The ribosome is constructed from a high
percentage of tertiary structural motifs, with almost 50% non-Watson-Crick base pairs. In
contrast, the RNA origami architecture is mainly constructed from Watson-Crick base pairs
formed by secondary structure elements and pseudoknots. The cotranscriptional folding of
the ribosome involves transiently stable helices, protein chaperones and structural switches
to guide the strand into a final native state that doesn’t correspond to the MFE. By contrast,
RNA origami takes advantage of a very different, very unnatural design construction, in
Geary et al.
Page 10
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
which every helix of the design is able to rapidly find its MFE structure during the kinetic
folding process. Thus, we are engineering very smoothed-out folding landscapes, with
strand paths designed to minimize the possibility for the strand to misfold during the
process. A recent computational study
58
suggests a general method for choosing strand-
paths that minimizes the risk of topological barriers, and finds that KLs arranged into
columns connected by a single common helix will result in the fewest topological barriers.
While many of our designs have this property (minimized topological barriers), this work
deserves to be further explored quantitatively; and determining whether or not RNA origami
contain minor misfolded elements may require the adaptation of SHAPE-seq
59
or other
techniques to very large RNA structures, or perhaps high-resolution cryo-EM. Future
challenges for the cotranscriptional RNA origami method will be to increase the structural
complexity by 3D architectures and tertiary motifs. Likewise, the ability to program RNA
origami scaffolds with functional, dynamic features and molecular computing elements (like
strand displacement logic gates), as have been achieved with DNA origami structures, to
create biosensor devices and nanorobots is still to be explored.
Methods
RNA Origami Automated Design (ROAD) package
To automate key processes of RNA origami design, a set of algorithms were made that
together constitutes a design pipeline: RNAbuild – Builds a PDB structure from an RNA
blueprint. RNApath – Analyses the folding path of a given RNA blueprint and highlights
topological barriers. Animations can be generated in the form of a series of keyframes and a
Chimera command file for automatic generation of a movie. Revolvr – Designs RNA
sequences that fold into target structures (requires installation of the Vienna RNA package
https://www.tbi.univie.ac.at/RNA/
). Trace – Takes an RNA blueprint without sequence
assigned to it, and a candidate sequence, and creates a new blueprint with the candidate
sequence threaded onto the blueprint. Trace_pattern – Converts RNA blueprint diagrams
into dot-paren notation and candidate sequence for Revolvr. Trace_analysis – Analyzes an
input blueprint and annotates it with features such as duplicated sequences, unintended
complementary sequences, GC content, restriction sites, etc. Flip_trace – Flips a blueprint
horizontally, vertically, and in both directions, which aids in the design of more complex
patterns using domain duplication and fusion. The analysis package is available for
download at GitHub (
https://github.com/esa-lab/ROAD/
) and has been made available as a
webserver with accompanying tutorials (
https://bion.au.dk/software/rnao-design/
).
RNAbuild
The RNAbuild algorithm automates structural modeling of RNA origami structures. As
input the algorithm takes an RNA blueprint and parses the RNA structure from 5’ to 3’ end
to identify a set of predefined 2D motifs. The 3D atomic model is constructed from 5’ to 3’
end by serial addition of 3D structural modules from a library that matches the 2D motifs
and 3D structural modules. When each structural module is added to a structure, it is rotated
and translated to the correct position using a single reference base or base pair having A-
form parameters (Supplementary Table 1 and Supplementary Fig. 3); these reference bases
are added to modules when they are put in the library (Supplementary Fig. 1). For building
Geary et al.
Page 11
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
RNA helices, there are four different trivial nucleotide modules, simply the PDB coordinates
for each RNA nucleotide. Modules for KLs, terminal tetraloops, fluorogenic aptamers, and
the RNA-protein binding domains are all based on known crystal structures. The helix axes
of the crossover module are modeled as parallel, rather than at the 60°-70° angle found in
crystal structures (see PDB-ID: 1HP6)
60
, under the assumption that the coupling of adjacent
crossovers forces them flat. The spacing of crossovers between three or more parallel helices
defines the geometry of the DT junction and RNAbuild helps model these.
RNApath
The RNApath algorithm analyses the folding path of RNA origami designs to identify
possible topological barriers during the cotranscriptional folding process. The algorithm
takes an RNA blueprint as input and analyses the RNA structure as it is being extended from
5’ to 3’ end. For each subsequence of length k, RNApath takes the fold computed for the k-1
subsequence, and decides what new base pairs can be added to the fold. RNApath adds a
secondary structure (e.g. a particular hairpin) to the fold for the subsequence having the
smallest k possible, which models the situation that secondary structure folds immediately,
as soon as the necessary sequence is transcribed. By default, a particular KL is added to the
fold of a subsequence k only when k is at least 150 nt longer than the smallest subsequence
that contains both halves of the KL. This feature roughly captures the KL formation time,
modeling the situation that KL formation is delayed by ~0.7 s (assuming a transcription
speed of 4.3 ms/nt)
61
after the KL sequence has been transcribed. Where the folding of KLs
might topologically clash with secondary structure formation, RNApath labels barrier loops
by ‘~’ and topologically blocked nts by ‘X’ in an analysis blueprint output. It additionally
outputs a list of substructures in dot-paren where transient single strands are shown as ‘,’ and
crossovers are shown as ‘^’. The delay is adjustable from 0 nt (for which almost all KLs
cause clashes), to N nt (for which no clashes will occur). In addition, RNApath can output a
series of PDB models which can be rendered to create a movie in UCSF Chimera v1.10
where pseudoknot loops are colored orange and topologically blocked nts are colored red
(Supplementary Videos 1–6 and Note 1). Alternatively, the program trace_analysis provides
a fast summary of any patterns in the sequence as well as positions of wobbles within the
design, and lastly the strand path analysis.
Revolvr
The Revolvr algorithm designs sequences for target structures by using a five-stage variant
of stochastic gradient descent where each stage has an increasingly restrictive cost function
(see algorithm flowchart in Supplementary Fig. 1 and 2). The cost function is a score that
combines the minimum free energy (MFE) folding prediction and different measures of
sequence symmetry, and for each successive round of design becomes stricter. The input file
defines the target secondary structure, pseudoknots and sequence constraints, and is initially
seeded with a random sequence that satisfies the constraints (or with user inputted
sequence). The first stage optimizes the MFE structure over five rounds of positive design to
stabilize helix ends and multi-junctions at a cost of raising the GC content. The current
sequence’s MFE structure, as computed by the ViennaRNA package
44
is used to calculate
the Hamming distance of the current structure to the target structure, which is used as the
cost function score for this round of design. The second stage applies alternating positive
Geary et al.
Page 12
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
and neutral design for a variable number of rounds, until the target structure is achieved:
mutations are either targeted to regions that misfolded in the previous round (two out of
every three rounds) or spread randomly throughout the sequence to enable neutral drift
(every third round). To increase the speed of design, we scale the rate of mutation per design
based on the success/failure of each iteration. A design round is considered successful when
the cost function remains the same or decreases. The third stage uses negative design to
decrease the probability of misfolding, prevent the inclusion of particular sequences (e.g.
restriction sites) where undesired, and make the DNA template from which RNA origami are
transcribed easier to synthesize and PCR amplify. Sequence symmetry minimization
62
(SSM) is applied to remove all repeated sequences or regions of undesired complementarity
above a threshold length (default setting = 10 nt). Similarly, removal of long homopolymer
stretches, and a known transcriptional pause site
63
, may help to reduce the frequency of
unwanted transcriptional termination. The fraction of GC bp is reduced to below 55%, to
encourage correct folding at 37°C. Finally, GU wobble pairs are introduced to
simultaneously preserve the helix within the desired RNA structure, and weaken it within
the corresponding DNA templates. All of the above constraints are applied through
successive rounds of targeted mutation, until they and the MFE fold are simultaneously
satisfied. The fourth stage eliminates repetition from the set of KLs. Repeated (and thus also
palindromic) KL sequences are targeted for mutation in successive rounds until all KL
interactions are unique. The fifth stage optimizes the sequences of the KLs to have uniform
binding energy and greater specificity. Energies for all possible KL interactions are
estimated with the Duplex function of ViennaRNA
44
and KLs are targeted for mutation until
all desired KLs have energies between -10.7 and -7.2 kcal/mol, and all undesired KL
interactions have energies greater than -6.0 kcal/mol. Revolvr enables potentially conflicting
requirements for positive vs negative design, sequence vs secondary structure constraints,
and pseudoknotted vs non-pseudoknotted structure to be balanced and satisfied. User-
specified sequence constraints supersedes user-specified secondary structure, which
supersedes all other constraints. Sequences explicitly specified in a blueprint (such as
aptamers) are left unmutated, even if the secondary structure specified for them cannot be
achieved in an MFE structure, or if they violate a sequence symmetry constraint. Upon
termination, Revolvr outputs an analysis of the designed sequence which includes a
blueprint populated with the sequence, KL energies, and the position of potential topological
clashes, violations of sequence symmetry constraints, and GU wobbles.
Synthesis of RNA origami structures
DNA templates were commercially synthesized (Integrated DNA Technologies) as double-
stranded gBlocks. DNA gBlocks were PCR-amplified using 19-20 nt primers (T
m
56°C)
complementary to the ends of the gBlock, using standard Taq DNA polymerase, and purified
using a Qiagen PCR purification kit. RNAs were transcribed and cotranscriptionally folded
in a one-pot reaction containing: template DNA (~4 ng/μl final of PCR amplicon), 6 mM
Mg(OAc)
2
, 40 mM Na OAc, 40 mM KCl, 50 mM Tris-OAc (pH 7.8), rNTPs (0.5 mM each)
and 1 mM DTT. Reactions were initiated by adding T7 RNA polymerase (~0.2 U/50 μl).
Transcription reactions were carried out in 50 μl volumes at 37.0°C for 45 minutes to 2 hrs,
depending on the sequence length. Larger designs required longer synthesis time (1-2
Geary et al.
Page 13
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
hours), whereas smaller designs (e.g. 2AE) required just a few minutes to reveal multimeric
products by AFM.
AFM sample preparation and imaging
1-5 μl of transcription product was mixed with 40 μl AFM dilution buffer (12.5 mM
Mg(OAc)
2
, 40mM KCl, 40mM NaCl, Tris-Borate pH 7.8) directly on the surface of a
freshly-cleaved mica puck. Mixing is performed by vigorously pumping a 200 μl pipette tip
ten times, before removing and discarding the fluid. The mica was washed with a solution of
60 mM NiCl
2
. Most AFM images were collected using a Multimode AFM (Digital
Instruments) with a Nanoscope IIIA controller and a J-scanner. Olympus TR400PSA silicon
nitride probes with a spring constant of ~.08 N/m were used for imaging, with a drive
frequency of ~6-9 kHz. AFM in Supplementary Figs. 10 and 11 were collected with a
Bruker Fastscan Bio AFM (Bruker) under buffer using FastScan-D probes (Bruker).
Purification of RNA Origami
RNA origamis were transcribed from linearized pUC19 plasmid for large-scale synthesis
and purification. Briefly, 25 μg of linearized plasmid was used as template in a 0.5 mL
reaction containing 40 mM Tris-Cl pH 8.1, 1 mM spermidine, 0.001% Triton X-100, 100
mM DTT, 12 mM MgCl
2
, 0.1 mM CaCl
2
, 0.5X ribolock (Thermo Fisher Scientific) and in
house prepared T7 polymerase. Transcription was carried out at 37°C for 3 hours before the
addition of 40 μL of DNase I (NEB). After 30 minutes of DNA digestion the reaction was
centrifuged at 17,000 RCF (x g) for 10 minutes to pellet precipitated pyrophosphate. The
supernatant was loaded onto a Superose 6 column (GE) equilibrated with 25 mM Hepes pH
7.5, 50 mM KCl and 5 mM MgCl
2
.
Negative Stain TEM
CF400 Au grids (Electron Microscopy Sciences) were glow discharged for 45 seconds at 25
mA before application of 3 μL of sample and then thrice blotted with 3 μL of 0.5% uranyl
formate. Peak 2 from the 5HS purification was diluted to 25ng/μL prior to blotting and Peak
2 from the Path1-optimized, Path2-optimized and Path1-nonoptimized purifications were
diluted to 50ng/μL. TEMImages were obtained on a 120 kV Tecnai Spirit TEM equipped
with a 4K TVIPS CMOS camera at 67k magnification. The images were contrast inverted
and converted from tif to mrc using Eman2
64
before being imported into CryoSparc V2.0
65
.
CTF correction was applied with CTFFIND4
66
. For 3D reconstructions, ~300 particles were
manually picked in Cryosparc and used to generate templates for the first round of templated
particle picking. These particles were sorted into 50 2D classes, the best of which were used
for 3D reconstructions.
Ab initio
3D reconstruction of the Path1-optimized design produced
a volume with real space slices similar to what would be expected from our design and was
further refined by a non-uniform refinement.
Ab initio
3D reconstruction repeatedly failed
for the 5HS design and so a homogeneous refinement of the 5HS structure was performed
using a 40 Å masked volume of our predicted RNA origami structure as an initial volume.
For the comparison of the Path1-optimized, Path2-optimized and Path1-nonoptimized
datasets blob picking was performed on 88 images of each design with the default settings in
CryoSparc V2.0, followed by a single round of 2D class averaging into 50 classes. Structural
deformities observed were measured with Eman2.
Geary et al.
Page 14
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
Folding yield calculation
The SEC yield was calculated as the monomer to aggregate fraction based on the average
peak heights of the SEC chromatogram (UV 255 nm) for 2-3 transcription reactions. The
TEM folding yield was estimated by counting the number of face views of the RNA origami
particles based on the assumption that correctly folded particles would have a similar
preference of adsorbing to the carbon film in this orientation. The number of face views
were adjusted to the amount of ng of RNA loaded on the grids, the number of images
obtained, and the molar mass. The folding yield of the 5HS was assumed to be 95% based
on analysis of TEM and AFM images (data not shown) and was used to calculate the relative
folding yield of the other samples. The transcription folding yield was calculated by
multiplying the SEC monomer yield with the TEM folding yield. An alternative fold, named
the “purse-handle”, was identified by measurement of a helix gap of minimum 1.5 nm in the
TEM class average images of the Path2-optimized sample accounting for 25.3% of the face
views.
Protein design, expression, and purification
MCP-ΔFG/V29I
67
and PCP-ΔFG
68
were codon-optimized for expression in
E. coli
.
Expression plasmids pJ431 encoding for His(6)-tagged mTurquoise2-MS2 Coat Protein
(mTq-MCP) or His(6)-tagged YPet-PP7 Coat Protein (YPet-PCP) under the control of an
IPTG-inducible T7 RNA polymerase promoter were ordered from ATUM (USA). Proteins
were expressed in E. coli BL21 Star (DE3). Cells were inoculated from a single colony in
Luria Bertani (LB) medium with Kanamycin and grown at 37°C overnight, under shaking.
The next day media was refreshed and cells were grown at 37°C under shaking to a density
of 0.3 OD/ml. Cells were then induced with 1 mM IPTG and incubated at 29°C under
shaking for 5 hours. Induced cells were harvested and sonicated with a Q125 microtip
sonicator (Qsonica Sonicators, USA). Both His(6)-tagged proteins were purified by gravity-
flow chromatography with TALON® Metal Affinity Resin (Takara Bio Inc, Japan). After
purification, the proteins were dialyzed overnight at 4°C using a Spectra-Por® Float-A-
Lyzer® G2 dialysis device (8-10 kDa MWCO; Spectrum Labs Inc, USA) in Protein Storage
Buffer (25 mM Tris/HCl (pH 8.0) + 0.3 M NaCl) and stored at 4°C.
Fluorescent proteins experiments
For protein scaffolding experiments fluorescence measurements were performed on a
VarioskanFlash 4 (Thermofisher). Excitation of mTurquoise2-MCP and YPet-PCP were
performed at 434 nm and 505 nm, respectively. Emissions of mTurquoise2-MCP and YPet-
PCP were recorded at 474 nm and 525 nm, respectively. Excitation bandwidths were set to 5
nm, and measurement time was 0.1 s. Measurement during transcription was done for a 58
μl transcription reaction containing transcription mix, 100 ng of DNA template, and protein
concentrations of 500 nM each. Transcription was started by adding 2 μl NTPs (25 mM
each) and measured every 5 min for 50 min (Supplementary Fig. 28). To compare several
constructs, RNA was transcribed (NEB T7 RNApol protocol, incubated overnight at 37°C)
and the reaction was stopped by adding DNase I (NEB) at 10 U/100 μl and incubated for 45
min at 37°C. Produced RNA was quantified on denaturing PAGE using a Typhoon Laser
Scanner (Amersham). Florescence measurement was performed on 60 μl samples with 330
Geary et al.
Page 15
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
nM RNA and protein concentration of 500 nM, each incubated for 5 min to allow proteins to
bind the scaffolds.
Fluorescent aptamer experiments
RNA was prepared as described in the previous section with the addition of 100 mM KCl to
the transcription reactions to facilitate folding of G quadruplex aptamers. DFHBI-1T was
purchased from Lucerna Technologies (USA) and YO3-biotin was purchased as custom
synthesis from Apigenex (Czech Republic). Fluorescence measurements were performed on
a FluoroMax 4 (Horiba, Jobin Yvon) by excitation of DFHBI-1T and YO3-biotin at 450 nm
and 580 nm, respectively. Emissions of DFHBI-1T and YO3-biotin were recorded at 503 nm
and 620 nm, respectively. Monochromator slits were set to 5 nm, and integration time was
0.2 s. Measurement during transcription was done for a 58 μl sample of transcription mix
containing 100 mM KCl, 30 ng of DNA template, 2 μM DFHBI-1T and 10 μM YO3-biotin.
Transcription was started by adding 2 μl NTPs (25 mM each) and measurements were
performed every 5 min for 90 min (Supplementary Fig. 28). To compare several constructs,
the produced RNA was quantified on denaturing PAGE as described above. Florescence
measurement was performed on 60 μl samples with 150 nM RNA incubated at RT for 20
min with 2 μM DFHBI-1T and 10 μM YO3-biotin.
Ensemble FRET calculations
The emission intensity arising from the donor tail at acceptor wavelength was calculated to
obtain the leak of the donor emission using the equation:
D
leak
=I
D
(ex
D
,em
A
)/I
D
(ex
D
,em
D
)
,
where
I
D
(ex
D
,em
A
)
is the emission at acceptor wavelength after donor excitation with only
donor present, and
I
D
(ex
D
,em
D
)
is the emission at donor wavelength after donor excitation
with only donor present. Relative FRET values were calculated using the equation:
FRETOutput
=
I
DA
ex
D
,
em
A
D
leak
I
DA
ex
D
,
em
D
I
DA
ex
D
,
em
A
D
leak
I
DA
ex
D
,
em
D
+
I
DA
ex
D
,
em
D
where
I
DA
(ex
D
,em
A
)
is the emission at acceptor wavelength after donor excitation, and
I
DA
(
ex
D
,
em
D
) is the emission at donor wavelength after donor excitation with both donor
and acceptor present.
Supplementary Material
Refer to Web version on PubMed Central for supplementary material.
Acknowledgements
We thank Lulu Qian and Erik Winfree for the use of their atomic force microscopes. Grigory Tikhomirov for help
with AFM and Mette Jepsen for help with FRET. We acknowledge the EteRNA community for conducting an
experiment which suggested that kissing loop sequences are less constrained than previously assumed; this inspired
us to add
de novo
design of KLs to Revolvr. C.G. acknowledges a fellowship from the Carlsberg Research
Foundation. E.K.S.M. acknowledges the Natural Sciences and Engineering Research Council of Canada for his post
doctoral fellowship. P.W.K.R. acknowledges funding by NSF grants (CCF-1317694 and CMMI-1636364) and ONR
grants (N00014-16-1-2159, N00014-17-1-2610, and N00014-18-1-2649). E.S.A. acknowledges funding by the
ERC Consolidator Grant (RNA ORIGAMI - RNA-protein nanostructures for synthetic biology, 683305) that
supported the work of C.G., G.G. and E.K.S.M. and the Independent Research Fund Denmark (9040-00425B) that
supported the work of E.K.S.M.
Geary et al.
Page 16
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts
Data availability
The data supporting the findings of this study are further documented in the associated
Supplementary Information. All raw data and analysis files used in the study are available
upon request from the authors.
Code availability
The code used to generate RNA origami designs in this study is included in the associated
Supplementary Information. Future updates to the code will be made available on Github
(
https://github.com/esa-lab/ROAD
) and on a dedicated web server with accompanying
tutorials (
https://bion.au.dk/software/rnao-design/
). The code is licensed under the MIT
License.
References
1. Grabow WW, Jaeger L. RNA self-assembly and RNA nanotechnology. Acc Chem Res. 2014;
47:1871–1880. [PubMed: 24856178]
2. Jasinski D, Haque F, Binzel DW, Guo P. Advancement of the Emerging Field of RNA
Nanotechnology. ACS Nano. 2017; 11:1142–1164. [PubMed: 28045501]
3. Weizmann Y, Andersen ES. RNA nanotechnology—The knots and folds of RNA nanoparticle
engineering. MRS Bull. 2017; 42:930–935.
4. Ban N, Nissen P, Hansen J, Moore PB, Steitz TA. The complete atomic structure of the large
ribosomal subunit at 2.4 A resolution. Science. 2000; 289:905–920. [PubMed: 10937989]
5. Wimberly BT, et al. Structure of the 30S ribosomal subunit. Nature. 2000; 407:327–339. [PubMed:
11014182]
6. Geary C, Rothemund PWK, Andersen ES. A single-stranded architecture for cotranscriptional
folding of RNA nanostructures. Science. 2014; 345:799–804. [PubMed: 25124436]
7. Geary C, Chworos A, Verzemnieks E, Voss NR, Jaeger L. Composing RNA Nanostructures from a
Syntax of RNA Structural Modules. Nano Lett. 2017; 17:7095–7101. [PubMed: 29039189]
8. Li M, et al. In vivo production of RNA nanostructures via programmed folding of single-stranded
RNAs. Nat Commun. 2018; 9:2196. [PubMed: 29875441]
9. Han D, et al. Single-stranded DNA and RNA origami. Science. 2017; 358
10. Shibata T, et al. Protein-driven RNA nanostructured devices that function in vitro and control
mammalian cell fate. Nat Commun. 2017; 8:540. [PubMed: 28912471]
11. Krissanaprasit A, et al. Genetically Encoded, Functional Single-Strand RNA Origami:
Anticoagulant. Adv Mater. 2019; 31 e1808262 [PubMed: 30972819]
12. Jepsen MDE, et al. Development of a genetically encodable FRET system using fluorescent RNA
aptamers. Nat Commun. 2018; 9:18–18. [PubMed: 29295996]
13. Chopra A, Sagredo S, Grossi G, Andersen ES, Simmel FC. Out-of-Plane Aptamer
Functionalization of RNA Three-Helix Tiles. Nanomaterials (Basel). 2019; 9
14. Porter EB, Polaski JT, Morck MM, Batey RT. Recurrent RNA motifs as scaffolds for genetically
encodable small-molecule biosensors. Nat Chem Biol. 2017; 13:295–301. [PubMed: 28092358]
15. Oi H, et al. Programmable formation of catalytic RNA triangles and squares by assembling
modular RNA enzymes. J Biochem. 2017; 161:451–462. [PubMed: 28096453]
16. Afonin KA, et al. Design and self-assembly of siRNA-functionalized RNA nanoparticles for use in
automated nanomedicine. Nat Protoc. 2011; 6:2022–2034. [PubMed: 22134126]
17. Cui D, et al. Regression of Gastric Cancer by Systemic Injection of RNA Nanoparticles Carrying
both Ligand and siRNA. Sci Rep. 2015; 5 10726 [PubMed: 26137913]
18. Afonin KA, et al. Co-transcriptional assembly of chemically modified RNA nanoparticles
functionalized with siRNAs. Nano Lett. 2012; 12:5192–5195. [PubMed: 23016824]
Geary et al.
Page 17
Nat Chem
. Author manuscript; available in PMC 2021 November 10.
Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts