ARTICLE
Received 7 Jun 2015
| Accepted 28 Jul 2016 | Published 10 Oct 2016
Infected erythrocyte-derived extracellular vesicles
alter vascular function via regulatory Ago2-miRNA
complexes in malaria
Pierre-Yves Mantel
1,2
, Daisy Hjelmqvist
1
, Michael Walch
2
, Solange Kharoubi-Hess
2
, Sandra Nilsson
1
,
Deepali Ravel
1
, Marina Ribeiro
1
, Christof Gru¨ring
1
, Siyuan Ma
3
, Prasad Padmanabhan
1
, Alexander Trachtenberg
4
,
Johan Ankarklev
1
, Nicolas M. Brancucci
1,5
, Curtis Huttenhower
3
, Manoj T. Duraisingh
1
, Ionita Ghiran
6
,
Winston P. Kuo
4,7
, Luis Filgueira
2
, Roberta Martinelli
8
& Matthias Marti
1,5
Malaria remains one of the greatest public health challenges worldwide, particularly in
sub-Saharan Africa. The clinical outcome of individuals infected with Plasmodium falciparum
parasites depends on many factors including host systemic inflammatory responses, parasite
sequestration in tissues and vascular dysfunction. Production of pro-inflammatory cytokines
and chemokines promotes endothelial activation as well as recruitment and infiltration of
inflammatory cells, which in turn triggers further endothelial cell activation and parasite
sequestration. Inflammatory responses are triggered in part by bioactive parasite products
such as hemozoin and infected red blood cell-derived extracellular vesicles (iRBC-derived
EVs). Here we demonstrate that such EVs contain functional miRNA-Argonaute 2 complexes
that are derived from the host RBC. Moreover, we show that EVs are efficiently internalized by
endothelial cells, where the miRNA-Argonaute 2 complexes modulate target gene expression
and barrier properties. Altogether, these findings provide a mechanistic link between EVs and
vascular dysfunction during malaria infection.
DOI: 10.1038/ncomms12727
OPEN
1
Department of Immunology and Infectious Diseases, Harvard T.H. Chan School of Public Health, Boston, Massachusetts 02115, USA.
2
Department of
Medicine, Unit of Anatomy, University of Fribourg, 1700 Fribourg, Switzerland.
3
Department of Biostatistics, Harvard T.H. Chan School of Public Health,
Boston, Massachusetts 02115, USA.
4
Harvard Catalyst Laboratory for Innovative Translational Technologies, Harvard Medical School, Boston, Massachusetts
02115, USA.
5
Wellcome Trust Center for Molecular Parasitology, University of Glasgow, Glasgow G12 8TA, UK.
6
Division of Allergy and Infection, Beth Israel
Deaconess Medical Center, Boston, Massachusetts 02115, USA.
7
Predicine, Inc., Hayward, California 94545, USA.
8
Center for Vascular Biology Research,
Beth Israel Deaconess Medical Center, Boston, Massachusetts 02115, USA. Correspondence and requests for materials should be addressed to P.-Y.M. (email:
Pierre-Yves.mantel@unifr.ch) or to M.M. (email: matthias.marti@glasgow.ac.uk).
NATURE COMMUNICATIONS | 7:12727 | DOI: 10.1038/ncomms12727 | www.nature.com/naturecommunications 1
D
espite continuing efforts, malaria remains a major
public health threat worldwide. Although the majority
of infections remain asymptomatic or uncomplicated,
a small proportion of infected people develops severe symptoms
and may die. The most virulent malaria parasite, P. falciparum
causes B650,000 deaths annually, most of them amongst
children below the age of 5 in sub-Saharan Africa
1
. The factors
that regulate the transition from an uncomplicated disease to
serious conditions including cerebral malaria (CM) and severe
malarial anaemia are poorly understood. Even when treated CM
has a poor prognosis with fatality rates of 30–50%, and survivors
often suffer from neurological complications.
The pathology of P. falciparum malaria is related to the
capability of parasite-infected red blood cells (iRBCs) to sequester
in deep tissues by adherence to the microvasculature. For this
purpose the parasite expresses a variant surface antigen,
P. falciparum erythrocyte membrane protein 1 (PfEMP1). Each
P. falciparum parasite genome encodes about 60 PfEMP1 variants
that enable the parasite to attach iRBCs to different host receptors
on the vascular lining of various tissues, including brain and
placenta
2
. Sequestration prevents splenic clearance of mature
iRBCs and contributes to pathology by reducing blood flow and
promoting inflammation in the capillaries. Inflammatory
cytokines can induce receptor expression on the surface of the
endothelium, thereby enhancing iRBC binding. In addition,
adhesion of iRBCs to the endothelium can increase vascular
permeability and apoptosis of the endothelial cells
3
. Indeed,
vascular dysfunction is a common feature of cerebral malaria
through destruction of the blood–brain barrier and the formation
of haemorrhages
4–6
. Interestingly, in vitro experiments have
demonstrated that increased vascular permeability does not only
require direct iRBC interactions but can also be caused by soluble
factors released by iRBCs
7
. Transfer of released malaria antigens
to endothelial cells has also been observed in autopsy studies in
the context of acute inflammatory lesions in cerebral tissue
5,8
.
Likewise tissue surveys in autopsy studies have detected parasite
markers in endothelial cells of multiple tissues
9
. Antigen transfer
from iRBCs to brain endothelial cells appears to occur via
membranous structures during transient iRBC—endothelial cell
interactions
7
, a process reminiscent of trogocytosis.
Release of membranous material in the form of extracellular
vesicles (EVs) from iRBCs during parasite development has
initially been described in the rodent malaria model and more
recently in P. falciparum
10–13
. EVs are small vesicles released by
almost all eukaryotic cell types. EVs derived from iRBCs are
increased in serum of malaria patients, in particular during severe
disease
14
. EVs can activate cells of the innate immune system in
the rodent malaria system and in human malaria
11,13
. We have
recently demonstrated that EVs can also facilitate cellular
communication between iRBCs by transferring signalling
molecules from a donor to a recipient iRBC. Moreover this
communication pathway is linked to the formation of malaria
transmission stages
11
and capable of transferring (episomal)
DNA
10
.
Compositional analysis of EVs derived from iRBCs identified a
specific set of RBC and parasite proteins as well as RNA species
including small RNA species
11
. In other systems, EVs have been
shown to contain nucleic acids, in particular messenger RNA
(mRNA) and microRNA (miRNA) that can be transferred to and
function in recipient cells. For example, Epstein-Barr virus
induces secretion of EVs that contain viral miRNAs from infected
cells, and these secreted viral miRNAs can downregulate cytokine
expression in uninfected monocyte-derived dendritic cells
15
.
Similarly, hepatitis C virus induces secretion of exosomes from
infected hepatocytes that are also able to transmit the virus to
naive cells
16
. Importantly, extracellular miRNAs in complex with
proteins and lipids are very stable and protected from
degradation by RNAses and are therefore particularly well
suited to transduce signals
17
. miRNAs are 19–24-nucleotide
noncoding RNAs that are derived from larger primary transcripts
encoded in the genome. Following initial processing by the
RNases Drosha and DGCR8 in the nucleus, the pre-miRNA is
exported to the cytoplasm where it is processed to a miRNA
duplex by another RNase termed Dicer
18
. One of the RNA
strands is loaded and incorporated into the effector protein
Argonaute 2 (Ago2) to form the RNA-induced silencing complex
(RISC)
19
. The sequence-specific miRNA guides the RISC complex
either to target sites in coding regions of target mRNAs for
endonucleolytic cleavage or to the 3
0
-untranslated regions (UTRs)
leading to translational repression
20
. The latter can occur through
a variety of reported mechanisms including decreased mRNA
stability due to deadenylation and uncapping, or via direct
inhibition of translation
21
.
In this study, we tested the hypothesis that iRBC-derived EVs
contain miRNAs that can modulate target genes in recipient cells.
We identified multiple miRNA species in EVs, and we
demonstrated that they are bound to Ago2 and form functional
complexes. Furthermore, we observed transfer of iRBC-derived
EVs into endothelial cells, repression of miRNA target genes and
alteration of endothelial barrier properties.
Results
iRBC-derived EVs contain a subset of human miRNAs.We
have previously developed a protocol for purification of iRBC-
derived EVs, and we have demonstrated the presence of specific
parasite and host proteins in purified EVs, including human
Ago2. Initial experiments also demonstrated that EVs from iRBCs
contain RNA, although the nature of these RNA species remained
unclear
11
. To further confirm these initial findings and investigate
the composition of nucleic acids in EVs, we prepared purified
EVs from iRBC cultures and analysed the size distribution of
RNA by the Agilent 2100 bioanalyzer (Fig. 1a). Starting with 1 mg
of purified EVs we extracted total RNA. For this purpose EVs
were pretreated with RNase A before lysis in order to remove
potential contaminating free RNA bound to EVs or co-purified
during the isolation process. We also excluded possible serum
contamination by pre-clearing the serum of host EVs with
ultracentrifugation before culturing the parasites. Analysis by
bioanalyser failed to detect RNA species above 150 nucleotides in
size, including the diagnostic ribosomal RNA bands
(Supplementary Fig. 1a). In contrast we detected small RNA
species in the range of 21–25 nucleotides suggesting the presence
of mature miRNAs (Fig. 1a). Since Plasmodium parasites lack
functional RNA interference machinery capable of producing
miRNAs
22
, we hypothesized that these miRNA were host-
derived. Indeed it has previously been shown that human RBCs
express a small set of miRNA species during maturation, with
various roles during RBC development
23
. To profile the
composition of human miRNAs in EVs from iRBCs, we used a
nanostring expression array comprising probes for all 800 human
miRNA species. Specifically, we compared the composition and
amount of human miRNAs from uninfected RBCs, infected RBCs
and EVs. As expected, we detected RBC-specific miRNAs in all
four preparations with a total of 21 miRNAs specifically present
in RBCs and a small subset of those in iRBCs and EVs from
iRBCs (Fig. 1b and Supplementary Data 1). However, except for
miR-451a and let-7b, all the miRNA species were depleted in
iRBC and EV samples as compared with uninfected RBCs
(Fig. 1c). Notably, miR-451a is the most abundant miRNA in
RBCs and plays an important role in erythroid homoeostasis
during RBC development
24
. To independently confirm presence
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12727
2 NATURE COMMUNICATIONS | 7:12727 | DOI: 10.1038/ncomms12727 | www.nature.com/naturecommunications
of mature miRNA species in RBCs, iRBCs and EVs, we performed
northern blots for miR-451a, let-7b, RNU6 and miR-106b (Fig. 1d).
AllfourmiRNAspeciesweredetectedinRBCsandEVsprepared
from iRBCs culture supernatants. Altogether these data demonstrate
the presence of a subset of RBC-specific miRNAs, notably the RBC-
specific miR-451a species, in EVs and iRBCs.
25 nt
21 nt
let-7b miR-106b
0
2
4
6
8
10
12
14
16
4
20
40
80
100
150
miRNAs
0
1,000
2,000
3,000
4,000
5,000
6,000
7,000
8,000
9,000
10,000
0 2,000 4,000 6,000 8,000 10,000
miR-451a
let-7b
let-7b
miR-451a
hsa−miR−16−5p
hsa−miR−374a−5p
hsa−miR−26b−5p
hsa−let−7d−5p
hsa−miR−185−5p
hsa−miR−363−3p
hsa−let−7c
hsa−miR−142−3p
hsa−miR−423−5p
hsa−miR−191−5p
hsa−miR−30d−5p
hsa−miR−20a−5p+hsa−miR−20b−5p
hsa−miR−222−3p
hsa−miR−144−3p
hsa−miR−15a−5p
hsa−miR−106a−5p+hsa−miR−17−5p
hsa−miR−15b−5p
hsa−miR−106b−5p
hsa−let−7a−5p
hsa−let−7b−5p
hsa−miR−451a
iRBC EV from iRBCs RBC
18
(FU)
(nt)
miRNA ladder
EVs from iRBCs
RBCs
EVs from iRBCs
RBCs
miRNA ladder
Mature
miRNA
Mature
miRNA
Pre-miRNA
25 nt
21 nt
EVs from iRBCs/RBCs
iRBCs/RBCs
RBC (transcript counts)
EVs from iRBC and iRBC
(transcript counts)
EVs from iRBCs
RBCs
miRNA ladder
miR-451 RNU6
EVs from iRBCs
RBCs
a
b
c
d
Figure 1 | Detection of human miRNA species in EVs. (a) EVs derived from iRBCs contain small RNAs. RNA samples were prepared from EVs and
analysed by Agilent Bioanalyzer Small RNA Chip. Small RNA species of up to 150 nucleotides are detected, including species between 20 to 25 nucleotides
(range marked by green lines). FU, fluorescence units. (b) miRNA profiling of iRBCs, EVs from iRBCs and RBCs by nanostring. miRNA preparations from
four experiments were analysed using a Nanostring miRNA array containing 800 human miRNA species. Normalized data from the entire array are shown
in the left panel. Twenty-one miRNAs are expressed in RBCs, of which a subset is also present in iRBCs and EVs from iRBCs (right). These are (in order of
expression levels) miR-451a, let-7b, let-7a, miR-106b and miR-15b. Those that are further characterized in this study are marked with black arrows.
(c) Differential abundance of miRNA species in RBCs, EVs from iRBCs and RBCs. The normalized Nanostring data from b are represented as correlations
between the mean expression across four experiments of RBC and the mean expression of either iRBCs (black boxes) or EVs from iRBCs (grey diamonds).
miR-451a and let-7b show similar abundance across all three populations, while the other miRNA species are greatly reduced in iRBCs and EVs from iRBCs.
(d) Detection of miRNAs by northern blot. RNA samples from RBCs and EVs from iRBCs are hybridized with probes specific for miR-451a, let-7b and
miR-106b. The northern blot shows specific bands at the expected size of 22 nucleotides for miR-451a and let-7b, and of 21 nucleotides for miR-106b. In
addition unprocessed Pre-miRNA species at the expected size of 70 nucleotides are detectable for let-7b and miR-106b. Shown is also a blot for U6 small
nuclear RNA (RNU6) as a control for RBC-derived non-miRNA RNA species
23
. Samples were normalized using equal RNA input.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12727 ARTICLE
NATURE COMMUNICATIONS | 7:12727 | DOI: 10.1038/ncomms12727 | www.nature.com/naturecommunications 3
Mature RBCs and iRBCs contain human Ago2 protein. There
is increasing evidence for the EV-mediated transfer of miRNAs
between cells. Although generally subject to degradation by
nucleases, mature miRNAs are protected and functional when
bound to Ago2 (ref. 25). We therefore hypothesized that the
miRNA detected in iRBCs and EVs may be bound to Ago2.
Functionality of RNAi has been described in erythroid cells
26
, but
there is no evidence so far that the machinery remains functional
once erythroid cells mature to normocytes (that is, mature RBCs).
To investigate whether functional miRNAs may be present in
mature RBCs, we first monitored expression of components of
the RNAi machinery during in vitro RBC maturation from
hematopoietic stem cells to normocytes
27
. We detected Ago2 in
mature RBCs, while Drosha and Dicer were only detectable in
erythroid cells, before their enucleation and maturation into
reticulocytes (Fig. 2a). To determine the localization of Ago2 in
mature RBCs, we performed immunofluorescence assays (IFA)
and flow cytometry using specific antibodies to human Ago2. In
normocytes, Ago2 was primarily localized to the cell periphery,
close to the RBC membrane (Fig. 2b). Analysis of infected RBCs
demonstrated a similar localization pattern in RBCs containing
young ring stage parasites (up to B20 h post invasion). Later
during development, however, the signal was progressively lost
(Fig. 2b,c). We have previously observed a similar dynamic in
subcellular localization for the RBC lipid raft protein stomatin,
with stomatin disappearance coinciding with the peak of EV
release from iRBCs during late schizont maturation
11
. We further
characterized the subcellular localization of Ago2 during
iRBC development by sequential permeabilization using the
pore-forming agents tetanolysin and saponin. Tetanolysin forms
pores in the RBC membrane but leaves the internal membranes
intact, while saponin also permeabilizes the parasitophorous
vacuole membrane while leaving the parasite intact
28
(Fig. 2d).
Sequential treatment of iRBCs with tetanolysin and saponin
results in a RBC cytosol, a parasitophorous vacuole and a parasite
and host membrane fraction (pellet). Western blot analysis
demonstrated Ago2 presence in the soluble fractions representing
RBC cytosol and parasitophorous vacuole. During the first 24 h of
parasite development, the protein was also present in the pellet
fraction, confirming the observed IFA localization in the parasite
R1
R2
R3 R4
R1
R2
R3
R4
Ring
10
3
10
3
10
2
10
2
10
1
10
1
10
10
10
–1
10
–1
0
20
40
60
80
100
0.017
8.2435.8
50.8 5.90
0.56
99.4
44.1
3.54
45.9 6.43
2.22 3.1 2.67
12.6
Ago2
Ago2
10
3
10
2
10
1
10
10
–1
Ago2
10
3
10
2
10
1
10
10
–1
Ago2
Ago2
Exp-1
Stomatin
Histone H3
12 h 24 h 36 h 48 h
Tetanolysin
(TTL)
Saponin
(SAP)
Untreated
(total iRBC)
Pellet
(P)
Time pi:
Ago2
Drosha
Dicer1
Histone H3
Stomatin
Day 11
Day 18
Day 21
Mature
RBC
SchizontTrophozoiteRingUninfected
Merged DAPI Anti-AGO2
Pellet
SAP
TTL
Total
Pellet
SAP
TTL
Total
Pellet
SAP
TTL
Total
Pellet
SAP
TTL
Total
Sybr green (DNA)
10
3
10
2
10
1
10
10
–1
Sybr green (DNA)
10
3
10
2
10
1
10
10
–1
S
y
br
g
reen (DNA)
10
3
10
2
10
1
10
10
–1
S
y
br
g
reen (DNA)
Isotype control
Trophozoite
97
97
17
31
15
159
218
15
31
2 µm2 µm2 µm2 µm
ab
cd
Figure 2 | Detection of human Ago2 in RBCs and iRBCs. (a) Expression of RNAi machinery during RBC development. Cell lysates from different
developmental stages of erythropoiesis were prepared (day 11: basophilic and polychromatic erythroblasts, day 18: orthochromatic erythroblasts and
reticulocytes, day 21: reticulocytes and normocytes; representative images are shown). Ago2 is detectable in mature RBCs while dicer, drosha (and human
histone H3) are not present. Stomatin is present in all preparations. (b) IFA of Ago2 in RBCs and iRBCs. Ago2 is localized to the RBC periphery in uninfected
and infected RBCs, however, labelling is reduced in later parasite stages. Notably, ring stage parasites also show Ago2 accumulation in the parasite. Scale
bar; 1 mm. (c) Flow cytometry analysis of Ago2 labelling in RBCs and iRBCs. Uninfected RBCs and iRBCs were gated based on SYBR staining (nuclear
content, n) and Ago2 labelling. Young iRBCs (rings, and trophozoites with n ¼ 1) show the highest Ago2 labelling, confirming IFA data. (d) Sequential
fractionation of purified infected parasites analysed by western blot. iRBC samples were collected at four time points post invasion, separated from
uninfected RBCs and fractionated. The host cytosol was released with 1 HU of tetanolysin (TTL) and subsequently the PV contents were released with
0.035% of saponin (SAP). The pellet contains parasite material and host membranes while the TTL supernatant contains RBC cytosol and the SAP
supernatant contains soluble parasitophorous vacuole material. The parasitophorous vacuole membrane marker PfExp-1 and parasite nuclear histoneH3
are detectable in the pellet fraction across the asexual parasite cycle. Ago2 is present in cytosol (SAP, TTL) across the cycle but only present in the pellet
fraction in early parasite stages, supporting the dynamic distribution observed by IFA. Host stomatin is present in cytosol (SAp, TTL) and membranes
(pellet) throughout the cycle. Data are representative of three independent experiments. pi: post invasion.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12727
4 NATURE COMMUNICATIONS | 7:12727 | DOI: 10.1038/ncomms12727 | www.nature.com/naturecommunications
and at peripheral RBC membranes and/or cytoskeleton. At later
time points, the protein could not be detected in the pellet
anymore, suggesting relocalization or secretion, possibly via EVs.
Ago2 and miRNA are found inside EVs. Indeed, we detected
Ago2 (and stomatin) in EV preparations from iRBCs by western
blot analysis (Fig. 3a), corroborating our previous proteomics
data
11
. We independently confirmed that Ago2 is present within
EVs by immune EM with specific anti-Ago2 antibodies using
purified EVs (Fig. 3b). Previous work has demonstrated that most
of the miRNAs found in the plasma of healthy donors are mainly
associated with proteolipid complexes and not with EVs
29
.To
exclude the possibility that Ago2 and miRNA molecules originate
from protein–lipid complexes present in the serum, we used two
complementary approaches, as recently described
29
. First, we
performed size exclusion chromatography and demonstrated that
human serum indeed contains a subpopulation of miRNAs
(and proteins) that elute in the same fractions as those from EV
preparations. Importantly, EV preparations do not elute miRNA
or protein in the major serum fractions, demonstrating that our
EV samples are devoid of serum contaminants (Fig. 3c,d).
Second, Proteinase K protection assays with EV preparations
demonstrated that Ago2 and miRNA molecules (and stomatin
control) were protease resistant while the surface exposed
Glycophorin C was sensitive (Fig. 3e,f), as we have shown
previously
11
. Altogether these experiments demonstrate that
Ago2 and miRNA molecules are present within EVs.
miRNAs form a functional RISC complex with Ago2 in EVs.
miRNAs form a stable silencing complex with Ago2 (that is, the
RISC complex) in order to be functional. RISC binds to the tar-
geted mRNA through the seeding sequence of the miRNA.
Having established presence of both specific miRNA species and
Ago2 in EVs we next wanted to determine whether they formed a
functional silencing complex. To test sequence-specific endonu-
cleolytic activity of RISC we incubated purified EV lysates from
infected RBCs with a radio-labelled RNA sensor probe containing
a complementary sequence to either miR-451a or let-7b. This
Relative expression
(fold induction)
6 8 10 12 14 16 18 20 22
0
50
100
150
miR-451a
let-7a
Fraction number
6 8 10 12 14 16 18 20 22
0
50
100
150
miR-451a
let-7a
Fraction number
Relative expression
(fold induction)
Ago2
GlyC
Stomatin
060453015
Min
Proteinase K
(5 mg ml
–1
)
– + + + +
miR-451a
0 15304560
0.0
0.5
1.0
Min
let-7a
0 15304560
0.0
0.5
1.0
Proteinase K
Control
Proteinase K
Control
Min
Relative expression
(fold induction)
Relative expression
(fold induction)
Ago2
Stomatin
SerumEVs
RBC
EV1
EV2
EV3
A280 (protein, percentage)
6 8 10 12 14 16 18 20 22 24
0
20
40
60
80
100
BSA
EVs
Serum
Tyr
Fraction number
97
37
31
ab c
d
ef
Figure 3 | Ago2 and miRNA are present within purified EVs. (a) Expression of Ago2 and stomatin in multiple EV preparations was probed by western
blot. (b) Immuno electron microscopy. EVs were prepared and processed as described in materials and methods for labelling with human Ago2 antibody.
The field of view shows multiple EVs with internal immunogold labelling, demonstrating that Ago2 is present in EVs rather than associated by surface
attachment. Inserts: three representative EVs in higher magnification (scale bar, 100 nm). (c) Elution profile of purified EVs, human serum, BSA and tyrosine
(Tyr) after size-exclusion separation. Protein abundance was determined by absorbance at 280 nm. Points represent the mean
±
s.d. of three experiments
performed in triplicate. (d) Fractions from purified EVs (left) and serum (right) were assayed for miR451a (black) and let-7a (red) using absolute
quantification by TaqMan qPCR. The mean
±
s.d. of one representative experiment is shown (n ¼ 2). (e,f) The Ago2-miRNA complexes are protected from
protease K digestion. (e) EVs were treated with proteinase K (5 mg ml
1
, or untreated control) at 55 °C. At times indicated, aliquots were removed and
assessed for Ago2, stomatin and Glycophorin C expression by western Blot. In contrast to Glycophorin C, Ago2 and stomatin were protected from digestion
by proteinase K. (f) Untreated (open symbols) or treated samples (close symbols) were also analysed for miR-451a and let-7a expression by qPCR. Points
represent miRNA copies detected at each time relative to the control sample at time point 0. The mean
±
s.d. of one representative experiment is shown
(n ¼ 3). qPCR, quantitative PCR.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12727 ARTICLE
NATURE COMMUNICATIONS | 7:12727 | DOI: 10.1038/ncomms12727 | www.nature.com/naturecommunications 5