Fractal assembly
of micrometre-scale DNA origami arrays
with arbitrary patterns
Supplementary information
Grigory Tikhomirov
1
?
, Philip Petersen
2
?
and Lulu Qian
1
,
3
WWW.NATURE.COM/NATURE | 1
SU
PPLEMENTAR
Y INFORMATION
doi:1
0.1038
/nature
2
46
55
Contents
1 Materials and methods
3
1.1 Sample preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
1.2 Echo protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
1.3 AFM imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
2 Challenges for uniquely-addressable DNA origami arrays
6
3 Key definitions and abstractions
9
4 Melting temperature measurement
10
5 Yield estimation
11
5.1 Assessing the accuracy of AFM-based yield estimation . . . . . . . . . . . . . . . .
11
5.2 Plain arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12
5.3 Arrays with an example pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13
6 Effect of design and experimental conditions
14
6.1 Rotation rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14
6.2 Annealing temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
6.3 Non-interactive locations in the edge code . . . . . . . . . . . . . . . . . . . . . . .
16
6.4 Inert edges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18
6.5 Bridge and interior staples near the seams . . . . . . . . . . . . . . . . . . . . . . .
19
6.6 Interactive locations in the edge code . . . . . . . . . . . . . . . . . . . . . . . . . .
20
6.7 Automatic liquid handler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
22
6.8 Plate sealing method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
23
6.9 Annealing time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
7 Analysis on 8 by 8 arrays with patterns
26
7.1 The Mona Lisa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
7.2 A rooster . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
7.3 A bacterium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
7.4 A circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
30
8 Spin-filter purification
31
9 Strengthening the origami arrays after fractal assembly
32
10 Cadnano diagram
33
11 DNA sequences
34
References
48
WWW.NATURE.COM/NATURE | 2
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.103
8/nature
2
4655
1 Materials and methods
1.1 Sample preparation
Single-stranded M13mp18 DNA (scaffold strand) was purchased from Bayou Biolabs (catalog #
P-107) at 1 g/L in 1
×
TE buffer (10 mM Tris-HCl, 1 mM EDTA, pH 8.0). The concentration
of the scaffold strand was calculated on the basis of DNA ultraviolet absorbance measurements at
260 nm using NanoDrop2000 (Thermo Scientific). Staple strands (sequences listed in Supplementary
Tables 1 to 5) were purchased unpurified from Integrated DNA Technologies in 1
×
TE buffer (pH
8.0) at 100
μ
M each. Negation strands (sequences listed in Supplementary Table 6) that are
complementary to the edge staples were purchased at 200
μ
M each. The strands were diluted to
15
μ
M in 1
×
TE buffer and loaded into Echo qualified 384-well source microplate (Labcyte).
Individual DNA origami tiles were mixed from the source plate by Echo 525 liquid handler
(Labcyte) to yield 10
μ
L total volume with 10 nM scaffold strand and 75 nM staples in 1
×
TE
buffer with 12.5 mM Mg
2+
, after the addition of the negation strands. (Before the negation strands
were added, during annealing of the individual DNA origami tiles, the volume was slightly lower
than 10
μ
L and the concentrations of the scaffold and staples were slightly higher than 10 nM
and 75 nM, respectively.) The scaffold and staples were kept at 90
◦
C for 2 min and annealed
from 90
◦
C to 20
◦
C at 6 sec per 0.1
◦
C in a twin.tec 96-well skirted PCR plate (Eppendorf,
catalog # 951020401) sealed with domed cap strips (Eppendorf, catalog # 0030124839) on a Nexus
Mastercycler (Eppendorf). After the anneal, a five-fold excess (relative to the concentration of the
staple strands) of a full set of 44 negation strands were added to each type of DNA origami tile and
quickly cooled down from 50
◦
C to 20
◦
C at 2 sec per 0.1
◦
C.
2 by 2 arrays were prepared by mixing equal volumes of four individual tiles and annealing from
55
◦
C to 45
◦
C at 2 min per 0.1
◦
C and from 45
◦
C to 20
◦
C at 6 sec per 0.1
◦
C. The total annealing
time is roughly 3.5 hours.
4 by 4 arrays were prepared by mixing equal volumes of four 2 by 2 arrays obtained in the
previous step and annealing from 45
◦
C to 35
◦
C at 8 min per 0.1
◦
C and from 35
◦
C to 20
◦
C at 6
sec per 0.1
◦
C. The total annealing time is roughly 13.5 hours.
8 by 8 arrays were prepared by mixing equal volumes of four 4 by 4 arrays obtained in the
previous step and annealing from 35
◦
C to 25
◦
C at 32 min per 0.1
◦
C and from 25
◦
C to 20
◦
C at
6 sec per 0.1
◦
C. The total annealing time is roughly 53.5 hours.
For arrays with patterns, a 10-fold excess of poly-A strand was added to the arrays before AFM
imaging, allowing at least ten minutes for the poly-A strand to hybridize to the poly-T staple
extensions at room temperature.
Important note:
it is absolutely essential that all tiles or arrays are mixed at equal concentra-
tion in all stages, otherwise the yield will decrease significantly. To achieve the best yield, a tight
seal of the plate during annealing is necessary. We explored several sealing options including films,
foils and caps, and the cap strips specified above produced the most reliable results. However, even
with a tight seal, it is still possible that some wells in the plate will evaporate more than other
wells. Taking this evaporation into consideration, it is also necessary to transfer all solution from
the wells for mixing the tiles or arrays.
WWW.NATURE.COM/NATURE | 3
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.103
8/nature
2
4655
1.2 Echo protocol
The transfer volume in a protocol for an Echo 525 liquid handler must be multiples of 25 nL.
Additionally, the volume of sample in each well of the Echo qualified 384-well source plate must be
15 to 65
μ
L, resulting in 15
μ
L of unusable sample. Because of both constraints, we diluted the
edge staples to 15
μ
M before storing them in a source plate. This resulted in a transfer volume of
50 nL for each edge staple to have the 75 nM target concentration in a 10
μ
L total volume. If all
staples were at 15
μ
M, the total volume would have exceeded 10
μ
L. Thus we diluted the interior
staples (including those with extensions) to 30
μ
M, resulting in a transfer volume of 25 nL for each
interior staple.
Because the bridge staples are the same in all tiles, we mixed them together and divided the
mixture into five wells in a source plate. The concentration of the bridge staple mixture was at
100
/
38 = 2
.
63
μ
M, for 38 distinct bridge staples. Based on the target concentration and volume,
the desired transfer volume of the bridge staple mixture should be 57 nL per well. Rounding it to
the multiples of 25 nL yielded the actual transfer volume of 50 nL. The smaller volume is ok because
the staples are in large excess relative to the M13 scaffold. In the FracTile Compiler, we wanted
to make it convenient for the users to organize their strands, by fitting all strands and buffer in a
single 384-well source plate. To do that, we divided the bridge staple mixture into two wells instead
of five, resulting in a transfer volume of 125 nL per well.
The concentration of the M13 scaffold varied from batch to batch, but typically the difference
is no more than 10%. We used 0.343
μ
M of M13 divided into twelve wells in the source plate. The
desired transfer volume is 24.3 nL per well, for 10 nM target concentration in 10
μ
L. We rounded
it to 25 nL per well, which is the nearest multiples of 25 nL. Again, for the purpose of fitting all
strands in one 384-well source plate, we used only three wells in the FracTile Compiler, resulting in
a transfer volume of 100 nL per well.
We used eight wells of 1
×
TE/10
×
Mg
2+
, resulting in a transfer volume of 125 nL per well for the
target volume of 10
μ
L per tile. We used sixteen wells of 1
×
TE, generally resulting in a transfer
volume of 175 to 200 nL per well, distributed as evenly as possible. The difference in the volume
of 1
×
TE is due to the fact that the total number of edge staples varies in different tiles. In the
FracTile Compiler, we kept the same number of wells for 1
×
TE/10
×
Mg
2+
, but reduced it to fifteen
wells for 1
×
TE, also for the purpose of fitting all strands in one source plate.
A full set of 44 negation strands were mixed together at 200
μ
M each, resulting in 200
/
44 =
4
.
545
μ
M of the mixture. As the target concentration is 75
×
5 = 375 nM in 10
μ
L, the desired
transfer volume is 825 nL. The negation strand mixture was added to each tile either by Echo
or by manual pipetting after the tiles were individually annealed. We had sixteen wells of the
negation strand mixture in a source plate, each was used to transfer 825 nL four times into a 96-
well destination plate, for adding the mixture to a total of 64 tiles in an 8 by 8 array. In the FracTile
Compiler, we left out the negation strand mixture from the source plate to make room for other
strands.
The manual protocol generated by the FracTile Compiler uses the same concentrations and
transfer volumes as in the Echo protocol.
WWW.NATURE.COM/NATURE | 4
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.103
8/nature
2
4655
1.3 AFM imaging
Samples for AFM imaging were prepared by diluting the annealed samples ten fold, resulting in
1 nM of the scaffold strand (i.e. the sum of all monomers) in 1
×
TE/Mg
2+
buffer. After dilution,
40
μ
L of the sample was deposited onto freshly cleaved mica (SPI Supplies, 9.5 mm diameter,
catalog # 01873-CA). After 30 sec the solution was removed by sucking up all the liquid that comes
off in a single thumb-up movement while keeping the pipette attached to and almost perpendicular
to the mica surface. To minimize the background during imaging, the excess of staple and negation
strands was removed as follows. The mica surface was washed three times with 40
μ
L TE buffer
containing 10 mM MgCl
2
and 100 mM NaCl, by performing 10 down-and-up thumb movements
for each wash. After that, 80
μ
L of 1
×
TE/Mg
2+
buffer was added onto the mica and the sample
was imaged. AFM images were taken using tapping mode in fluid on a Dimension FastScan Bio
(Bruker) with FastScan-D tips (Bruker). Typical scanning parameters were: scan rate = 5 Hz, lines
= 1024, amplitude set point = 30–50 mV, drive amplitude = 180–240 mV, drive frequency = 110
Hz, integral gain = 1, proportional gain = 2.
WWW.NATURE.COM/NATURE | 5
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.103
8/nature
2
4655
2 Challenges for uniquely-addressable DNA origami arrays
Multi
-
stage
One
-
pot
Two
-
stage
Figure 1:
Overview of uniquely-addressable two-dimensional DNA nanostructures.
Each point in the
log
−
log plot corresponds to the size of a DNA nanostructure and the number of nucleotides in unique strands
self-assembled in the nanostructure. Although Wang et al explored both one-pot and two-stage experiments, we
show the work as one-pot because no interior or edge strands were reused. Similarly, Rajendran et al explored both
two-stage and three-stage experiments, and we show the work as two-stage because no edge strands were reused.
There are three types of approaches for creating uniquely-addressable two-dimensional DNA nanos-
tructures (Supplementary Fig. 1). Distinct DNA strands can be annealed together in one pot to
create DNA origami
10,14
and arrays of single-stranded tiles
32
and double-crossover tiles.
33
The one-
pot approaches are easy to implement, but the number of nucleotides in unique strands increases
linearly with the size of the structure. The increasing number of nucleotides makes it difficult to
scale up, due to the cost of the strands, the design challenges for controlling the spurious interac-
tions among distinct strands, and the resulting decrease in yield. Using hierarchical approaches,
DNA molecules can be annealed in two stages, first self-assembling into smaller structures such as
cross-shaped DNA tiles
34
or DNA origami tiles,
11–13
and then the individual tiles coming together to
form larger structures. In the two-stage approaches, the interior strands can be reused for different
tiles, but an increasing number of unique edge strands is still required. Multi-stage hierarchical
approaches make it possible to reuse both interior and edge strands, as shown in a 4 by 4 array
of small DNA tiles.
17
Because of the potential for creating arbitrarily complex structures from a
simple set of tiles, the strategies for multi-stage self-assembly have been explored extensively in
theory.
16,35–37
However, none of these theoretical strategies has yielded a successful experimental
demonstration.
WWW.NATURE.COM/NATURE | 6
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.103
8/nature
2
4655
In prior work, we developed a framework for creating random tilings with both unbounded and
finite DNA origami arrays.
18
The finite arrays self-assembled in two stages had a four-fold rotational
symmetry, which only required
n
2
/
4 distinct types of tiles and
n
(
n
−
1)
/
2 distinct types of edges
for arrays of size
n
by
n
. For example, 4 by 4 arrays were constructed with 4 distinct tiles and
6 distinct pairs of edges (Supplementary Fig. 2, top row). The small number of distinct tiles and
edges allowed us to successfully construct arrays of sizes 3 by 3 to 5 by 5. However, the rotational
symmetry of these arrays does not allow unique addressability.
To create uniquely-addressable DNA origami arrays, we had to use four times the number of
distinct types of tiles and edges compared to the prior work. For example, 4 by 4 arrays now require
16 distinct tiles and 24 distinct pairs of edges (Supplementary Fig. 2, middle and bottom rows).
In prior work, we used two types of edge codes with a stronger and a weaker binding energy: the
stronger one was composed of 11 edge staples, each has a stacking bond and a one-nucleotide sticky
end; the weaker one was composed of 4 edge staples, each has a stacking bond and a two-nucleotide
sticky end (Supplementary Fig. S54 of ref.
18
). The stronger edge code was used for interactions
between tiles composing each of the four 2 by 2 arrays and the weaker edge code was used for
interactions between the 2 by 2 arrays. Taking advantage of the M13 sequences being different on
the four sides of the square origami tile, each edge code can provide a maximum of four pairs of
distinct edges. Therefore, we cannot use the two types of edge codes in the arrays with rotational
symmetry to create 24 distinct pairs of edges for the uniquely-addressable arrays (16 for interactions
between tiles composing each of the four 2 by 2 arrays and 8 for interactions between the 2 by 2
arrays).
Keeping the edge codes palindromic as in prior work, there are a total of
(
4
2
)
= 6 types of edge
codes, using 4 out of the 8 edge staples (positions 2 through 5 and 7 through 10) that each has a
stacking bond and a two-nucleotide sticky end. Using all six edge codes, we now have 6
×
4 = 24
distinct edges to create the uniquely-addressable 4 by 4 arrays. However, the yield of these arrays
was substantially lower than the 4 by 4 arrays with rotational symmetry (Supplementary Fig. 2,
middle row), presumably due to the increased spurious interactions among the increased number
of distinct tiles. More importantly, the 24 distinct pairs of edges used in the 4 by 4 arrays already
reached our limit for designing orthogonal edges, and thus it is not possible to create larger uniquely-
addressable arrays using the same method.
Dividing the self-assembly process into more stages could provide a natural solution for (i)
reducing spurious interactions (by reducing the total number of possible reactions at any given time
during self-assembly), and (ii) using fewer distinct edges (by reusing the same edge interactions for
tiles that are in different test tubes at the same stage). However, multi-stage self-assembly cannot
work unless the edge design is compatible with a multi-stage annealing protocol: there must exist
an annealing temperature for each stage that is both low enough to keep the structures from a
previous stage stable and high enough to melt the spurious interactions at the current stage. When
we attempted to create the 4 by 4 arrays in three stages, we first annealed the sixteen individual
tiles from 90 to 20
◦
C, then annealed the four 2 by 2 arrays in separate test tubes from 50 to
20
◦
C, and finally annealed the four 2 by 2 arrays together from 30 to 20
◦
C. Because the melting
temperature of the 2 by 2 arrays is about 35
◦
C (Supplementary Fig. S23 of ref.
18
), we had to keep
the annealing temperature of the third stage below that, which was not high enough to melt the
spurious interactions and resulted in an extremely poor yield (Supplementary Fig. 2, bottom row).
WWW.NATURE.COM/NATURE | 7
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.103
8/nature
2
4655
Design
diagram
Annealing
protocol
AFM image
Yield
4 by 4
array with a
four
-
fold rotational
symmetry
Tikomirov
et. al.
Nat.
Nanotechnol
.
2016
two
-
stage
15%
4 by 4 array with
unique
addressability
two
-
stage
4.7%
three
-
stage
0
%
Figure 2:
Prior work and failed attempts for creating uniquely-addressable DNA origami
arrays.
All AFM images are 10 by 10
μ
m. The yield of each sample was estimated using AFM
images of 30 by 30
μ
m.
WWW.NATURE.COM/NATURE | 8
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.103
8/nature
2
4655
3 Key definitions and abstractions
DNA origami tile
inert edges
receiving edges
giving edges
abstract tile
N
S
E
W
N
S
E
W
N
S
E
W
edge code
nucleotide
truncation
nucleotide
extension
two
hairpins
0
1
1
1
1
0
1
1
1
1
0
0
1
1
1
1
0
1
1
1
1
0
0
1
1
1
1
0
1
1
1
1
0
0
x
1
x
2
x
3
x
4
0
x
4
x
3
x
2
x
1
0
0
1
1
1
1
0
1
1
1
1
0
0
x
1
x
2
x
3
x
4
0
x
4
x
3
x
2
x
1
0
Figure 3:
Key definitions and abstractions.
A square DNA origami tile has four edges: north (N), east (E),
south (S) and west (W). Each edge has a maximum of eleven staples. Because the four triangles composing the
square tile are folded from different parts of the M13 scaffold, the staples on the four edges naturally have different
sequences. “Inert edges” are created using five edge staples that are each capped with two hairpins to inhibit their
interactions with other edge staples. Inert edge staples are colored black. Six edge staples are left out and the
remaining scaffold loops are colored gray. There are two types of “active edges”: “receiving edges” are created using
eight or less edge staples that each has a two-nucleotide truncation on the 3’ end; “giving edges” are created using
eight or less edge staples that each has a two-nucleotide extension on the 5’ end. An “edge code” is associated with
each receiving or giving edge: the code consists of eleven 0s and 1s. Each 0 corresponds to a scaffold loop and each
1 corresponds to a staple. Receiving staples on the north, east, south and west edges are colored blue, green, orange
and yellow, respectively. Giving staples are colored based on the sequence identity of the extension: extensions that
are complementary to the truncations on the north, east, south and west edges are colored blue, green, orange and
yellow, respectively. Although a giving edge can be complementary to any receiving edge, we only use north giving to
west, east giving to north, south giving to east, and west giving to south in the design of fractal assembly. Abstract
tiles are used to simplify the illustration. The edge colors in an abstract tile correspond to the staple colors in an
origami tile. Each indentation corresponds to a 1 in an edge code for a receiving edge. Each bump corresponds to a
1 in an edge code for a giving edge. In fractal assembly, we use palindromic edge codes that have three 0s at fixed
locations. Therefore, only the second to fifth digits are needed to infer any edge code, and only the 1s in these four
digits are shown as indentations or bumps in an abstract tile.
WWW.NATURE.COM/NATURE | 9
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.103
8/nature
2
4655
4 Melting temperature measurement
a
b
25
30
35
40
45
50
55
Relative fluorescence intensity
Temperature,
ºC
Decreasing
Temperature
Increasing
Temperature
Figure 4:
Fluorescence experiments for melting temperature measurement. a,
Tile ab-
straction and edge design of a 2 by 2 array. A ROX fluorophore (shown as a red dot) is attached to
the 5’ end of an edge staple in one DNA origami tile, and a Iowa Black RQ quencher (shown as a
black dot) is attached to the 3’ end of an edge staple in another tile. When the tiles self-assemble
into 2 by 2 arrays, the fluorophore and quencher will come into proximity and result in low fluo-
rescence intensity. When the arrays melt, the fluorophore and quencher will become separated and
result in increased fluorescence intensity.
b,
Melting graph showing relative fluorescence intensity
during heating and cooling of the 2 by 2 arrays. The fluorescence intensity was measured in a
Mx3005P QPCR system (Agilent Technologies). The sample was heated up from 25 to 45
◦
C at
5 sec/0
.
1
◦
C, from 45 to 55
◦
C at 30 sec/0
.
1
◦
C, held at 55
◦
C for 30 sec, cooled down from 55
to 45
◦
C at 30 sec/0
.
1
◦
C, and then cooled down from 45 to 25
◦
C at 5 sec/0
.
1
◦
C. Fluorescence
intensity was measured with 585 nm excitation wavelength and 610 nm emission wavelength. Each
data point shown in the graph was an average of all data points within the same degree. Note that
the fluorophore and quencher labeled staples were added to the edge design shown in Fig. 2b. They
introduced two additional stacking bonds to the interaction between tiles, with which we expect a
slightly increased melting temperature.
WWW.NATURE.COM/NATURE | 10
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.103
8/nature
2
4655
5 Yield estimation
5.1 Assessing the accuracy of AFM-based yield estimation
How accurate AFM-based yield estimation is depends on if there exists a bias for the binding of
structures to mica surface, specifically based on their sizes. To explore that, we mixed 4 by 4 arrays
with monomers at an equal concentration of individual origami tiles (Supplementary Fig. 5a). If the
monomers have the same extent of binding as the 4 by 4 arrays, we expect the estimated yield of 4
by 4 arrays will decrease to half compared to the estimated yield without monomers, while greater
or smaller than half indicates less or more binding of the monomers, respectively. The AFM images
showed a bias of more monomers than 4 by 4 arrays on the mica surface (Supplementary Fig. 5b and
c). Our hypothesis is that monomers diffuse faster in solution and thus land faster on mica surfaces,
and once landed, the binding is strong enough for them to stay on mica after washing the surface
before imaging. This result suggests that the estimated yield could be lower (or higher) than the
actual yield if the off-target structures are mostly incomplete structures (or aggregations). Since
the off-target structures in the fractal arrays contain both incomplete and aggregated structures,
the AFM-based yield estimation should be reasonably accurate, in the sense that it is not obviously
lower or higher than the actual yield.
a
b
c
Binding
of
monomers
compared
to that of 4 by 4
arrays
to
mica surface
Yield
4 by 4
array
푥
4 by 4 array mixed with
monomers
at a
1
:
1
ratio
less
>
Τ
푥
2
same
Τ
=
푥
2
more
<
Τ
푥
2
4 by 4 array
4 by 4
array + monomers
Yield: 44.70%
Yield: 13.88%
Figure 5:
Assessing the accuracy of AFM-based yield estimation. a,
Three cases of scenarios
for estimating the yield of DNA origami arrays using AFM images.
b,
The yield of plain 4 by 4
arrays was estimated to be
x
= 44
.
70%.
c,
The yield of plain 4 by 4 arrays mixed with monomers
was estimated to be 13.88%, which is smaller than
x/
2.
11
WWW.NATURE.COM/NATURE | 11
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.103
8/nature
2
4655
5.2 Plain arrays
a
b
c
Figure 6:
AFM images of the plain arrays. a,
A 10 by 10
μ
m image of 2 by 2 arrays. The
yield was estimated to be 92
.
81
±
1
.
74%.
b,
A 30 by 30
μ
m image of 4 by 4 arrays. The yield was
estimated to be 47
.
91
±
1
.
76%.
c,
A 30 by 30
μ
m image of 8 by 8 arrays. The yield was estimated to
be 1
.
81
±
1
.
27%. The yield was determined as the total pixels in complete arrays of the designed size
(yellow pixels) divided by the total pixels above the threshold of background (blue pixels + yellow
pixels). The standard for identifying complete arrays is explained in ref.
18
Supplementary Fig. S59.
The calculation was aided by a custom software tool.
38
The error was calculated as
p
√
1
−
p/
√
n
,
where
p
is the estimated yield and
n
is the number of complete arrays in each image, treating the
yield as a Bernoulli probability.
39
12
WWW.NATURE.COM/NATURE | 12
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.103
8/nature
2
4655