ARTICLE
Received 3 Jan 2016 | Accepted 13 Sep 2016 | Published 21 Oct 2016
Tuning thermal conductivity in molybdenum
disulfide by electrochemical intercalation
Gaohua Zhu
1,
*, Jun Liu
2,3,
*, Qiye Zheng
2
, Ruigang Zhang
1
, Dongyao Li
2
, Debasish Banerjee
1
& David G. Cahill
2
Thermal conductivity of two-dimensional (2D) materials is of interest for energy storage,
nanoelectronics and optoelectronics. Here, we report that the thermal conductivity of
molybdenum disulfide can be modified by electrochemical intercalation. We observe distinct
behaviour for thin films with vertically aligned basal planes and natural bulk crystals with
basal planes aligned parallel to the surface. The thermal conductivity is measured as a
function of the degree of lithiation, using time-domain thermoreflectance. The change of
thermal conductivity correlates with the lithiation-induced structural and compositional
disorder. We further show that the ratio of the in-plane to through-plane thermal conductivity
of bulk crystal is enhanced by the disorder. These results suggest that stacking disorder and
mixture of phases is an effective mechanism to modify the anisotropic thermal conductivity of
2D materials.
DOI: 10.1038/ncomms13211
OPEN
1
Materials Research Department, Toyota Research Institute of North America, Ann Arbor, Michigan 48105, USA.
2
Department of Materials Science and
Engineering, Frederick Seitz Materials Research Laboratory, University of Illinois at Urbana-Champaign, Urbana, Illinois 61801, USA.
3
Department of
Mechanical and Aerospace Engineering, North Carolina State University, Raleigh, North Carolina 27695, USA. * These authors contributed equally to this
work. Correspondence and requests for materials should be addressed to G.Z. (email: gaohua.zhu@toyota.com) to or J.L. (email: jun_liu@ncsu.edu) orto
D.G.C. (email: d-cahill@illinois.edu).
NATURE COMMUNICATIONS | 7:13211 | DOI: 10.1038/ncomms13211 | www.nature.com/naturecommunications 1
T
wo-dimensional (2D) layer-structured materials consist
of atomic layers with strong intra-layer covalent
bonding stacked together by weak van der Waals bonds.
Transition metal dichalcogenides, an important class of 2D
materials, have attracted extensive research interest recently
due to their unique electronic and chemical properties
1
.
In particular, molybdenum disulfide (MoS
2
) has been
extensively studied for potential applications in nanoelectronics,
optoelectronics and flexible electronic devices
2–5
. Although the
thermal conductivity of single, few-layer and bulk MoS
2
has been
reported recently, the effects of structural and compositional
disorder on the anisotropic thermal conductivity of layered
materials, which usually occurs during crystal growth, fabrication
and applications (for example, in energy storage, thermoelectrics
and nanoelectronics), have not yet been systematically
characterized
6–8
.
Guest ions can be intercalated into the van der Waals gaps in
MoS
2
. Intercalation causes changes in the electronic structure,
and optical and electrical properties
9
. Intercalation can also
induce structural and compositional disorder, including
variations in layer spacing, interaction strengths between
adjacent layers and phase transitions
10–12
. By monitoring the
potential during electrochemical intercalation, we can control the
amount of intercalated ions. Therefore, intercalation provides
an effective way to systematically vary the structural and
compositional disorder of many 2D materials, and enables
investigations of how disorder affects their thermal conductivity.
To understand how thermal conductivity of highly anisotropic
materials can be affected by disorder, here we study the thermal
conductivity of both pristine and lithium ion-intercalated bulk
and thin-film MoS
2
. In the bulk sample, the MoS
2
basal planes are
oriented parallel to the surface; whereas, in the thin-film sample,
the MoS
2
basal planes are vertically aligned. The thermal
conductivity of Li
x
MoS
2
samples with different degrees (x)of
electrochemical interaction of lithium ions were measured by
time-domain thermoreflectance (TDTR). We show that lithium
ion intercalation has drastically different effects on thermal
transport in these different forms of MoS
2
due to the differences
in crystalline orientation and initial structural disorder.
Our most striking observation is that the thermal a nisotropy
ratio in bulk Li
x
MoS
2
crystals increases from 52 (x ¼ 0) to 110
(x ¼ 0.34) as a resu lt of lithiation-induced stacking disor der and
phase transitions. The thermal anisotropy ratio is the ratio
of in-plane to through-plane thermal conductivity, an important
material parameter in thermal management. The increase in
thermal anisotropy with increasing disorder is counter-intuitive:
previous studies show that structural disorder typically
decreases the thermal anisotropy ratio
13,14
.Ouranalysis
suggest that the enhanced thermal anisotropy ratio in
Li
x
MoS
2
bulk crystal is likely due to the combination of
phonon-focusing ef fects and pronounce d differences i n the in-
planeandthrough-planelengthscaleofthelithiation-induced
disorder.
Results
Sample preparation. The MoS
2
thin-film samples with vertically
aligned basal planes were grown by rapid sulfurization of a
Mo thin film
15
. Bulk samples of MoS
2
were obtained by
mechanical exfoliation of bulk MoS
2
crystals (SPI Supplies).
The cross-sectional transmission electron microscopy (TEM)
image in Fig. 1a reveals that the film thickness is E200 nm after
chemical-vapour deposition (CVD) growth and the MoS
2
atomic
layers are predominantly aligned perpendicular to the substrate
with the edges of the MoS
2
layers exposed to the surface. A typical
plan-view TEM image of the vertically aligned MoS
2
thin film is
shown in Fig. 1b. The thin film is polycrystalline with randomly
oriented strip-like or columnar grains. The cross-sectional area of
those columnar grains is B10 nm wide and several tens of
nanometres long. Bulk MoS
2
samples, with typical thickness
of 10–20 mm, were prepared by standard Scotch tape assisted
mechanical exfoliation. A plan-view TEM image of the bulk MoS
2
sample is shown in Fig. 1c, where the high-quality atomic plane of
MoS
2
can be seen.
Further structural analysis of bulk and thin-film MoS
2
samples
before lithiation were carried out by X-ray diffraction and Raman
spectroscopy. In X-ray diffraction, bulk samples show a strong
(002) peak located at 2y ¼ 14.4
o
, and (004), (006) and (008) peaks
with decaying magnitudes, indicating that the basal planes of the
bulk sample are parallel to the surface. The thin-film samples
have two peaks located at 32.8
o
and 58.5
o
, which are due to the
diffraction from (100) and (110) planes of individual columnar
grains, consistent with our conclusion from the TEM images that
the basal planes are vertically aligned (see details in Fig. 5). In
contrast to the bulk sample, (002) peak is not observed in the
X-ray diffraction spectrum of the thin-film sample because the
normal to the (002) planes is parallel to the substrate surface.
Figure 2a,b present the Raman spectra of the bulk and thin-
film MoS
2
samples. To probe the low-frequency interlayer E
2
2g
phonon mode, we used three reflective volume Bragg grating
filters (BragGrate notch filters) in combination with a single-pass
monochromator to access frequency shifts as small as
B10 cm
1
. The E
1g
mode (286 cm
1
) is observed only in the
thin-film samples while the E
2
2g
mode (32 cm
1
) is only present
in the bulk samples, as expected. According to the Raman
selection rules, the E
1g
mode is forbidden in backscattering
experiment on the basal plane of bulk MoS
2
(refs 16,17).
However, when the incident light scatters on the surface of
edge-terminated MoS
2
, the corresponding scattering Raman
tensor undergoes a rotation transformation, leading to a non-
zero differential scattering cross-section and hence the E
1g
mode
can be observed. The observation of the E
1g
mode in thin-film
samples therefore indicates that the basal planes of MoS
2
are
vertically aligned, consistent with the TEM and X-ray diffraction
data. The absence of the E
2
2g
mode (which is not forbidden by
selection rules) in thin-film samples is probably due to the
randomly oriented columnar grains and stacking disorder in
CVD-grown samples. In addition, although both A
1g
at 383 cm
1
ab c
Figure 1 | TEM images of MoS
2
bulk crystals and thin films. (a) Cross-sectional TEM image of the MoS
2
thin film with vertically aligned basal plane.
Scale bar, 20 nm. (b) Plan-view TEM image of the MoS
2
thin film. Scale bar, 10 nm. (c) Plan-view TEM image of the bulk MoS
2
crystal. Scale bar, 2 nm.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms13211
2 NATURE COMMUNICATIONS | 7:13211 | DOI: 10.1038/ncomms13211 | www.nature.com/naturecommunications
and E
1
2g
at 408 cm
1
modes are present in bulk and thin-film
MoS
2
, the peak intensity of the out-of-plane A
1
g
mode is similar to
that of the in-plane E
1
2g
mode in the bulk sample and B3 times
that of the E
1
2g
mode in the thin-film sample under the same
measurement condition. Such preferred excitation of an out-of-
plane mode is also consistent with the vertical-aligned crystal
texture of the thin-film sample considering the polarization
dependence of the Raman scattering cross-section
15
.
We carried out electrochemical intercalation of lithium ions in
both bulk and thin-film MoS
2
to study how lithiation affects the
thermal conductivity differently in MoS
2
samples with different
orientations. Lithium ion intercalation of thin-film MoS
2
samples
was performed through a galvanostatic discharge process in a
glass vial inside a glovebox; electrochemical intercalation of
bulk MoS
2
samples as performed using a coin cell battery setup
18
.
In both cases, MoS
2
samples were used as the working electrode,
and the lithium foil was used as the counter and reference
electrode.
The discharge curves for thin-film and bulk MoS
2
samples are
shown in Fig. 3. On lithium ion intercalation, a well-defined
plateau is observed at potentials between 1.1 and 1.2 V, as the
host lattice of MoS
2
undergoes a phase transition from 2H
to 1T phase
19
. The voltage d ip at the initial stage of the
discharge curve for bulk MoS
2
sample i s caused by the mass
transport limitation of lithium ions. The voltage gradually
recovers when the lithium ion transport is facilitated by defects
formed duri ng intercalation. The voltage dip was not observed
in the discharge process of thin-film MoS
2
, which can be
attributed to the high density of the edge sites in edge-
terminated thin-film samples.
We describe the lithium ion intercalation process in the range
of 1.1–3.0 V as reaction (1):
MoS
2
þ x Li
þ
þ x e
! Li
x
MoS
2
1:1V versus Li=Li
þ
; 0 x 1ðÞ
ð1Þ
We calculate the average lithium composition from the electronic
charge transferred to MoS
2
, based on the theoretical specific
100
a
b
c
d
80
Bulk
A
1g
E
1g
Bulk
Thin film
Thin film
60
40
20
0
0
30
60
90
120
150
180
x=0
x=0.34
x=0.86
x=0
x=0.34
x=0.86
0
30
60
90
120
150
180
100
80
60
40
20
0
250 300 350 400 450 500
250
300 350 400 450
500
02040
Raman shift (cm
–1
)
Raman shift (cm
–1
)
Raman shift (cm
–1
)
Raman shift (cm
–1
)
60 80 100
0 20406080100
S/P (cps mW
–1
)
S/P (cps mW
–1
)
S/P (cps mW
–1
)
S/P (cps mW
–1
)
E
2
2g
E
2
2g
E
1
2g
A
1g
E
1
2g
Figure 2 | Raman spectra for bulk and thin-film MoS
2
samples before lithiation. Spectra obtained at (a) low frequencies and (b) high frequencies.
The y-axis is the signal intensity normalized by laser power (S/P) in the unit of counts per second per milliwatt (cps mW
1
). The bulk spectrum is shifted
up by 50 cps mW
1
. Raman spectra for bulk Li
x
MoS
2
samples at different degrees of lithiation (x ¼ 0, 0.34 and 0.68) at (c) low frequencies, (d) high
frequencies. The x ¼ 0.34 and x ¼ 0 spectra are shifted up by 65 and 130 cps mW
1
, respectively.
1.6
1.4
Wang et al.
1.2
Thin film
Bulk
1.0
0.8
0.0 0.2 0.4
x in Li
x
MoS
2
0.6 0.8 1.0
Voltage versus Li/Li
+
(V)
Figure 3 | Initial discharge curves for MoS
2
samples. Thin-film (black)
and bulk (red) MoS
2
samples, compared with data by Wang et al.
55
(dashed line) for thin-film Li
x
MoS
2
.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms13211 ARTICLE
NATURE COMMUNICATIONS | 7:13211 | DOI: 10.1038/ncomms13211 | www.nature.com/naturecommunications 3
charge capacity for full intercalation of 167 mA h g
1
(ref. 19).
Hence, by controlling the duration of the galvanostatic discharge
process as shown in Fig. 3, vertically aligned Li
x
MoS
2
thin-film
(x ¼ 33 at%, 46 at%, 68 at% and 100 at%) and lithiated bulk
samples (20 at%, 40 at%, 60at%, 80 at% and 100 at%) with various
lithiation content (x) were prepared.
Since the charge transfer during electrochemical intercalation
can have a contribution from complicated side reactions, the
actual degree of lithiation (x) for each sample might be different
from the calculated values. To determine the actual x in bulk
MoS
2
samples, inductively coupled plasma mass spectroscopy
measurements were performed on one fully lithiated bulk sample.
The mass ratio of Li, Mo and S atoms is calculated by measuring
the element masses in the bulk Li
x
MoS
2
sample with metal
coating (total massE1 mg). The measured x is 86 at% suggesting
reasonable agreement between calculated and measured lithium
contents for the lithiation process. The actual x in other bulk
Li
x
MoS
2
samples is then scaled by the same factor of 0.86.
Thermal conductivity. MoS
2
samples (schematics shown in
Fig. 4a) were characterized by TDTR to determine the change in
thermal conductivity caused by lithium ion intercalation. The
TDTR data of both thin-film and bulk Li
x
MoS
2
samples are
presented as a function of x in Fig. 4b. The through-plane thermal
conductivity of thin-film Li
x
MoS
2
decreases monotonically from
E3.4 W m
1
K
1
(x ¼ 0) to E1.7 W m
1
K
1
(x ¼ 1) with
increasing lithium content. The through-plane thermal con-
ductivity of bulk Li
x
MoS
2
decreases first from E2.0 W m
1
K
1
(x ¼ 0) to E0.4 W m
1
K
1
(x ¼ 0.34) and then increases to
E1.6 W m
1
K
1
(x ¼ 0.86). The in-plane thermal conductivity
follows a similar trend, which decreases from E105 W m
1
K
1
(x ¼ 0) to E45 W m
1
K
1
(x ¼ 0.34) and then increases to
E80 W m
1
K
1
(x ¼ 0.86).
To understand such a drastic decrease of thermal conductivity,
we calculated the minimum thermal conductivity L
min
of
Li
x
MoS
2
using a simplified model by Cahill et al.
20
at the
high-temperature limit. The minimum thermal conductivity is
L
min
¼
p
48
1
3
k
B
n
2=3
ðv
L
þ 2v
t
Þ, where k
B
is the Boltzmann
constant, n is the atomic density (atoms cm
3
), v
L
is the
longitudinal speed of sound, and v
t
is the transverse speed of
sound
21
. Atomic densities of Mo and S atoms in Li
x
MoS
2
thin
films was measured by Rutherford backscattering spectrometry;
v
L
is measured by picosecond acoustics; v
t
is measured by
detecting surface acoustic waves using a phase-shift mask
22
.We
accounted for the change of atomic density and longitudinal
speed of sound due to lithiation. More details can be found in the
‘Methods’ section and the Supplementary Methods.
In thin-film Li
x
MoS
2
samples, the polycrystalline structure with
randomly oriented columnar grains is transverse isotropic so that
this conventional minimum thermal conductivity model can be
applied. In the bulk Li
x
MoS
2
samples, however, the strongly
anisotropic structure introduces a significant phonon-focusing
effect
23
, which suppresses the through-plane average group
velocity due to the relatively high in-plane group velocity. We
adopted the modified minimum thermal conductivity model
recently proposed by Zhen et al.
41
(equation S(7) in their
Supplementary Material) and followed their procedure to
calculate the minimum thermal conductivity of bulk Li
x
MoS
2
(see details in Supplementary Note 1 with parameters shown in
Supplementary Table 1).
If the measured lowest through-plane thermal conductivity of
bulk or thin-film Li
x
MoS
2
agrees with the predicted minimum
thermal conductivity, the phonons are glass-like lattice vibrations
in a disordered crystal at this composition. The calculated
minimum thermal conductivity of bulk and thin-film Li
x
MoS
2
is
plotted in Fig. 4b as dashed lines. Both measured through-plane
thermal conductivity of bulk and thin-film Li
x
MoS
2
is higher than
the predicted minimum thermal conductivity, which suggests that
a significant fraction of the phonons in these samples are
propagating modes.
Interlayer distance characterization. Large expansion in layer
spacing along the c-axis is often observed after intercalation of
large-diameter molecules or ions
11,24
. We collected X-ray
diffraction spectra for bulk Li
x
MoS
2
samples with different
amount of lithium ion intercalation (0rxr0.86) to determine
the change in layer spacing due to lithium ion intercalation.
Samples were placed inside an air-tight sample holder with a
beryllium (Be) window in an argon-filled glovebox before they
were transferred out for X-ray diffraction characterization.
Figure 5b shows that the (002) peak of the pristine MoS
2
is
located at 14.40
o
, and the peak position downshifts to 14.28
o
for
Li
0.86
MoS
2
. The corresponding layer spacing is 6.16 and 6.19 Å
b
100
10
Bulk (in-plane)
Bulk
(through-plane)
Thin film
(through-plane)
1
Thermal conductivity (W m
–1
K
–1
)
0.1
0.0 0.2 0.4
x in Li
x
MoS
2
0.6 0.8 1.0
Pump and probe beam
a
Pump and probe beam
In-plane
Through-plane
Al
Al
Thin film Li
x
MoS
2
Bulk Li
x
MoS
2
Sapphire
Figure 4 | Thermal conductivity measurement of Li
x
MoS
2
samples.
(a) Schematics of bulk and thin-film Li
x
MoS
2
samples for TDTR
measurements. (b) Thermal conductivity of Li
x
MoS
2
samples with different
degrees of lithiation x. Blue squares: through-plane thermal conductivity of
thin-film MoS
2
; black squares: through-plane thermal conductivity of bulk
MoS
2
; black open squares: in-plane thermal conductivity of bulk MoS
2
.The
minimum thermal conductivity for bulk and thin-film samples are plotted as
black and blue dashed lines, respectively. The total uncertainties of the
measured thermal conductivity are calculated by taking into account the
systematic errors that propagate from uncertainties in the film thickness,
laser spot size, and thermal properties of the transducer film and substrate.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms13211
4 NATURE COMMUNICATIONS | 7:13211 | DOI: 10.1038/ncomms13211 | www.nature.com/naturecommunications
for pristine MoS
2
and Li
0.86
MoS
2
, respectively, a 0.5% change in
lattice constant. We estimate the uncertainty of the lattice spacing
measurement as E0.6%. The weak dependence of lattice constant
on lithium content is consistent with the fact the effective ionic
radius of lithium ion
25
(76 pm) is slightly smaller than the
octahedral site in the van der Waals gap of in MoS
2
. If we assume
an effective ionic radius of S
2
as 1.84 Å, and close packing of
S
2
atoms, the radius of the octahedral site is 76 pm (ref. 25).
Our measurement on the layer spacing of Li
0.86
MoS
2
agrees with
recent observations reporting on minimal change in interlayer
distance of LiMoS
2
(refs 26,27).
Elastic constants. We further measured elastic properties using
pump-probe techniques to help understand the thermal con-
ductivity change in our samples. Polycrystalline MoS
2
thin films
with vertically aligned basal planes are transverse isotropic, which
has five effective independent averaged elastic constants: C
0
11
, C
0
12
,
C
0
13
, C
0
33
and C
0
44
. Figure 6 plots the effective C
0
33
and C
0
44
elastic
constants of thin-film Li
x
MoS
2
. The elastic constant C
0
33
of
Li
x
MoS
2
thin film decreases from 147 GPa (x ¼ 0) to 121 GPa
(x ¼ 1). Even though the density of Li
x
MoS
2
thin films increases
by 11% from x ¼ 0tox ¼ 1, the decrease of longitudinal speed of
sound from 5,720 ms
1
to 4,930 ms
1
(measured by picosecond
acoustics) dominates the decrease of C
0
33
. The elastic constant C
0
44
increases from 22 GPa (x ¼ 0) to 32 GPa (x ¼ 0.34) and then
decreases to 18 GPa (x ¼ 1). We do not yet understand the trend
of the elastic constants of thin-film Li
x
MoS
2
changing with x. One
possible reason is the combined effect of increasing binding
energy due to the intercalation of lithium ions and increasing
structural and compositional disorder (for example, point defects
and mixture of phases). More measurement details are described
in the ‘Methods’ section and Supplementary Methods.
The elastic constants of bulk Li
x
MoS
2
samples C
33
are also
plotted in Fig. 6. C
33
gradually changes from 52 GPa (x ¼ 0) to
58 GPa (x ¼ 0.86) with a transition point at x E0.34, which
suggests a phase transition in bulk Li
x
MoS
2.
Raman spectroscopy. We used Raman spectroscopy to further
characterize lithiation-induced structural and compositional dis-
order of bulk Li
x
MoS
2
to gain more insights about other phonon
scattering mechanisms that could lead to the thermal conductivity
change. All bulk Li
x
MoS
2
samples except pristine MoS
2
were
loaded inside the glovebox into a home-made air-free sample
holder sealed by O-ring and screw-on connectors and measured
through the glass window of the holder. As shown in Fig. 2c,d, the
intensity of the low-frequency peak at B32 cm
1
corresponding
to the E
2
2g
shear mode in 2H-MoS
2
decreases as the degree of
lithiation (x) increases. The decreasing peak intensity of E
2
2g
mode
is attributed to the increasing stacking disorder resulting from the
lithium ion intercalation. In the 2H to 1T phase transition, the
stacking of Mo atom planes changes from ABA (two molecular
layers per unit cell) to AA (one molecular plane per unit cell)
28,29
.
As the atomic structure of the MoS
2
layer changes from prismatic
aBa (2H) structure to octahedral aBc (1T) (upper letters
correspond to Mo planes, lowercase letters correspond to the
S planes)
30
, the high-frequency E
1
2g
and A
1g
modes redshift from
383 to 377 cm
1
and 408 to 402 cm
1
, respectively, as the
200
100
50
20
10
0.0 0.2 0.4
x in Li
x
MoS
2
Elastic constants (GPa)
0.6
Thin film C'
33
Bulk C
33
Thin film C'
44
0.8 1.0
Figure 6 | Effective elastic constant of thin-film Li
x
MoS
2
with different
degrees of lithiation. C
0
33
(square) decreases from 147 GPa (x ¼ 0) to
121 GPa ( x ¼ 1). C
0
44
(diamond) increases from 22 GPa (x ¼ 0) to 32 GPa
(x ¼ 0.34) and then decreases to 18 GPa (x ¼ 1). As comparison, the C
33
of
bulk Li
x
MoS
2
samples (circle) are also plotted. C
33
gradually changes from
52 GPa (x ¼ 0) to 58 GPa (x ¼ 0.86) with a transition point at xE0.34. The
error bars are calculated by taking into account the experimental errors and
the systematic errors that propagate from uncertainties in the Al film
thickness, Li
x
MoS
2
film density and the input elastic constants.
1,000
a
b
100
Intensity (cps)
Intensity (cps)
Lattice spacing d (Å)
10
10
6.3
6.2
6.1
10
6
10
5
10
4
10
3
10
2
10
1
6.0
5.9
0.0 0.2 0.4
13.5 14.0 14.5 15.0
x in Li
x
MoS
2
0.6 0.8 1.0
20 30 40
2 (°)
2 (°)
x=0
x=0.86
x=0.17
x=0.34
x=0.52
x=0.69
Sapphire
MoS
2
(110)
MoS
2
(100)
50 60 70
Figure 5 | X-ray diffraction characterization of Li
x
MoS
2
samples.
(a) X-ray diffraction spectra for MoS
2
thin-film samples on sapphire
substrate. The two diffraction peaks coincide with the standard X-ray
diffraction powder patterns of MoS
2
(100) and (110). Therefore, the
dominant lattice orientation in the MoS
2
thin-film samples are (100) and
(110). The diffraction peak at 32.8° and 58.5° corresponds to a lattice
constant of 2.73 and 1.58 Å, respectively. (b) Lattice spacing d between
MoS
2
layers in bulk MoS
2
samples. The error bars are calculated by taking
into account the experimental errors and the systematic errors that
propagate from uncertainties in the fitting of X-ray diffraction spectra. The
inset plot is the change of MoS
2
(002) peak position with x in Li
x
MoS
2
.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms13211 ARTICLE
NATURE COMMUNICATIONS | 7:13211 | DOI: 10.1038/ncomms13211 | www.nature.com/naturecommunications 5