Abstract
The outbreak of COVID-19 has become a worldwide pandemic. The
pathogenesis of this infectious disease and how it differs from other
drivers of pneumonia is unclear. Here we analyze urine samples from
COVID-19 infection cases, healthy donors and non-COVID-19 pneumonia
cases using quantitative proteomics. The molecular changes suggest that
immunosuppression and tight junction impairment occur in the early
stage of COVID-19 infection. Further subgrouping of COVID-19 patients
into moderate and severe types shows that an activated immune response
emerges in severely affected patients. We propose a two-stage mechanism
of pathogenesis for this unusual viral infection. Our data advance our
understanding of the clinical features of COVID-19 infections and
provide a resource for future mechanistic and therapeutics studies.
Subject terms: Proteomic analysis, Viral infection, Infectious diseases
__________________________________________________________________
How COVID-19 pathology differs from other drivers of pneumonia is
unclear. Here the authors analyze urine from patients with COVID-19 and
identify an immunosuppressive protein expression pattern that is
distinct from the pattern in healthy individuals or patients with
non-COVID-19 pneumonia.
Introduction
COVID-19, the disease caused by infection with the virus SARS-CoV-2,
has become a worldwide pandemic^[48]1–[49]5. By 24 April 2020, there
had been 2,649,859 confirmed cases, including 187,244 deaths, in 211
countries worldwide^[50]6. It is truly the first-time non-influenza
pandemic. As yet, we know very little about this new virus and its
pathogenesis^[51]7. At the beginning of the pandemic, research studies
focused on the management and treatment of severe and critical
patients^[52]8. Statistics showed that the elderly, especially those
with underlying conditions such as heart disease, lung disease,
obesity, and diabetes, have the most severe symptoms^[53]9,[54]10. As
the virus has now spread globally, more information is available about
younger infected people in their 30s to 50s. In a growing number of
cases, symptoms were expected to improve, but suddenly got worse.
Patients can develop acute respiratory distress syndrome (ARDS) or even
die suddenly in a short period of time. This sudden change implies a
“two-stage” pattern of disease progression, but the underlying
mechanisms are unknown.
Here, we applied a mass spectrometry-based, data-independent
acquisition (DIA) quantitative proteomic approach to analyze urine
samples from COVID-19 infection cases, healthy donors, and non-COVID-19
pneumonia cases (Fig. [55]1a). We found that immunosuppression and
tight junction (TJ) impairment specifically occur in COVID-19 patients.
Interestingly, we also found an activated immune response to some
extent in the late stage of infection compared with the early stage.
These data provide a map of molecular changes associated with the
COVID-19 disease and provide hints as to how two-stage pathogenesis
might occur.
Fig. 1. Quantitative urine proteomic studies.
[56]Fig. 1
[57]Open in a new tab
a An illustration of the experimental design. A total of 37 urine
samples were analyzed from three groups: healthy controls (10 samples);
COVID-19 patients (14 samples); and non-COVID-19 pneumonia patients (13
samples). The data-independent acquisition (DIA) technique was applied
for quantitative proteomics, which was performed by trapped ion
mobility spectrometry coupled to TOF mass spectrometry (TIMS-TOF MS)
with the parallel accumulation-serial fragmentation (PASEF) technique.
Integrated data analysis involved protein expression, clustering, and
functional correlational network strategies. b Venn diagram of
significantly changed proteins (cut-off value of fold change >2 and
fold change <0.5; p value (t test) <0.05) in the healthy control group
compared to the COVID-19 patient group (pink) and to the non-COVID-19
pneumonia patient group (blue). A total of 1841 proteins were
significantly changed specifically in COVID-19 patients compared to
healthy controls. Six hundred and ninety-one proteins were changed in
both disease groups. Among the 691 proteins, 145 proteins meet the
condition of fold change >2 and fold change <0.5, p value (t test)
<0.05. A total of 1986 proteins (the group of 1841 proteins and the
group of 145 proteins) were used for functional analysis. c Proteomic
changes in COVID-19 patients. Heatmap of 1986 proteins in COVID-19
patients in the comparison between healthy controls and non-COVID-19
patients. The color bar from red to blue represents the fold change
from increasing to decreasing of all proteins identified in each group.
Hierarchical clustering shows a clear group differentiation according
to similarity. Numbers of proteins, intensity profiles, and selected
enriched KEGG pathways are indicated for marked clusters. Color bar
represents Z-score change from −1 to 1.
Results
Urine proteome analysis of COVID-19 disease
We collected 14 patients who tested positive for nasopharyngeal swab
real-time PCR for SARS-CoV-2 infections, among which were six males and
eight females with ages ranging from 30 to 77. We subdivided the 14
COVID-19 patients into nine with moderate disease type and five with
severe disease type according to the information provided by the
attending physicians at the hospitals where the samples were obtained
(Supplementary Tables [58]1 and [59]2). The two control groups were 13
non-COVID-19 pneumonia patients and 10 healthy people. The raw data
were first processed using a double boundary Bayes imputation method
and were standardized for further analysis (Supplementary Figs. [60]1
and [61]2 and Supplementary Data [62]1–[63]3). Our data showed clear
stratification among different cohorts according to the principal
components analysis (PCA) (Supplementary Fig. [64]3 and Supplementary
Data [65]3). For the data analysis, we firstly compared the three
groups of COVID-19 infection cases, healthy donors, and non-COVID-19
pneumonia cases. A total of 5991 proteins were identified in all 37
samples. The levels of 1986 proteins were significantly changed in the
COVID-19 group compared to the healthy donors and the non-COVID-19
pneumonia controls (Fig. [66]1b, Supplementary Fig. [67]4, and
Supplementary Data [68]4–[69]6). Surprisingly, we identified nearly ten
times more down-regulated proteins than up-regulated ones in the
COVID-19 group (Fig. [70]1c and Supplementary Data [71]7). KEGG (Kyoto
Encyclopedia of Genes and Genomes) enrichment analysis revealed the
molecular landscape associated with COVID-19 infections. More than ten
pathways were significantly changed. In particular, COVID-19 had a
strong impact on immune-related pathways, TJ pathways, and metabolic
pathways (Fig. [72]1c and Supplementary Data [73]7).
Immune system is suppressed in the early stage of COVID-19 disease
We found that a large number of proteins associated with the immune
response were down-regulated (Supplementary Fig. [74]5 and
Supplementary Data [75]7). This suggests that the immune responses are
suppressed in COVID-19 patients. We observed dramatically decreased
protein levels of protein tyrosine phosphatase receptor type C, leptin,
and tartrate-resistant acid phosphatase type 5, which are involved in
lymphopenia, and platelet basic protein in COVID-19 patients
(Fig. [76]2a, Supplementary Fig. [77]6, and Supplementary Data [78]7).
These results are consistent with the lower lymphocyte and platelet
counts in the blood tests of COVID-19 patients^[79]11. Complement C3,
complement C1q subcomponent subunit C, complement C1r subcomponent, and
PZP-like alpha-2-macroglobulin domain-containing protein 8 were
down-regulated, which suggests the significant impairment of the
complement system (Supplementary Fig. [80]7 and Supplementary
Data [81]7). The level of spleen tyrosine-protein kinase, which is
involved in FcγR-mediated phagocytosis, was dramatically decreased in
patients, indicating that the phagocytosis of microphages, neutrophils,
natural killer cells, and monocytes was suppressed (Fig. [82]2a,
Supplementary Fig. [83]8, and Supplementary Data [84]7). A decreased
level of apolipoprotein A-I in the serum has been reported during the
transition of COVID-19 patients from mild to severe illness^[85]12. We
also observed that COVID-19 patients showed the down-regulation of
apolipoprotein A-IV (APOA4) and apolipoprotein E in COVID-19 patients,
possibly associated with macrophage function.
Fig. 2. Functional analysis of 1986 specific proteins in COVID-19 patients.
[86]Fig. 2
[87]Open in a new tab
a The interaction diagram of the FcγR-mediated phagocytosis, chemokine
signaling pathway, natural killer cell-mediated cytotoxicity and
antigen processing and presentation. Network nodes and edges represent
proteins and protein–protein associations. Green solid lines represent
inhibition; the red solid lines represent activation; the blue dotted
lines represent the KEGG pathways. Color bar from red to green
represents the fold change of protein level from increasing to
decreasing. The significance of the pathways represented by −log(p
value) (Fisher’s exact test) was shown by color scales with dark blue
as most significant. b The interaction diagram of proteins involved in
tight junctions. Green solid lines represent inhibition; blue dotted
lines represent the KEGG pathways. Color bar from red to green
represents the fold change of protein level from increasing to
decreasing. The significance of the pathways represented by −log(p
value) (Fisher’s exact test) was shown by color scales with dark blue
as most significant. c The scatter plot graphs show six proteins, which
are potential diagnostic markers for COVID-19. One-way ANOVA was used
to analyze the data. For Health group, n = 10; for nCoV group, n = 14;
for non-CoV group, n = 13. Data are presented as mean ± SEM. Source
data are provided as a Source Data file.
Moreover, we observed decreased levels of proteins related to the
chemokine signaling pathway (Supplementary Fig. [88]9 and Supplementary
Data [89]7). C-C motif chemokine 14, C-C motif chemokine 18, and C-X-C
motif chemokine ligand 12 were dramatically down-regulated in COVID-19
patients, indicating decreased monocyte activation, B cell migration,
and T cell-mediated immune response (Fig. [90]2a and Supplementary
Data [91]7). Neutrophil chemotactic factor C-X-C chemokine receptor
type 2 and signal transducer and activator of transcription 3, 5B, and
6 were also down-regulated, which is suggestive of impaired cytokine
production and degranulation of neutrophils, macrophages, and T
lymphocytes (Supplementary Fig. [92]9 and Supplementary Data [93]7).
Acute hypoxia and ARDS are two of the major causes of the high case
fatality rate in COVID-19 patients^[94]13. Previous research shows that
a significant increase in the permeability of the alveolar epithelial
barrier results in alveolar edema and exudate formation, and represents
one of the major factors that contribute to the hypoxemia in
ARDS^[95]14. The barrier function of the lung epithelium depends on a
set of TJ heteromeric complexes, which seal the interface between
adjacent epithelial cells^[96]15. TJs also exist between epithelial
cells in other organs, such as intestine, kidney, and
brain^[97]16,[98]17. The disruption of TJ complexes is the major cause
of epithelial barrier breakdown during virus infection^[99]18. We found
that a number of proteins involved in TJ formation and cell–cell
adhesion junctions were drastically down-regulated in COVID-19
patients, including TJ protein ZO-1, TJ protein ZO-2, claudin-2,
claudin-3, claudin-11, claudin-19, Afadin, cingulin, protein crumbs
homolog 3, cAMP-dependent protein kinase catalytic subunit alpha
(PRKACA), and Rho GTPase-activating protein 17 were drastically
down-regulated in COVID-19 patients (Fig. [100]2b, Supplementary
Fig. [101]10, and Supplementary Data [102]7). This indicates that the
virus may alter intercellular TJ formation and epithelial morphogenesis
during viral invasion. This in turn may damage the physical barrier
that protects the underlying tissues.
Among the significantly up- or down-regulated proteins in COVID-19
patient urine, metallothionein-1G, lipoprotein lipase, β2M, PRKACA,
FOLR2, and APOA4, showed significant changes compared to both healthy
controls and non-COVID-19 pneumonia cases (Fig. [103]2c). These
proteins are potential biomarkers for differential diagnosis of
COVID-19 to make it more precise. It is worth noting that four out of
the six proteins are associated with the immune response and TJs, which
are two featured pathways identified in COVID-19 patients.
Immune response is activated in severe COVID-19 patients
It is reported that a cytokine storm happens at the late stage of
COVID-19 patients^[104]19–[105]21. To understand the pathogenesis of
COVID-19, it will be essential to find out how the transitions take
place during the progression of the disease. To further investigate
this, we subdivided the COVID-19 patients into nine moderate cases and
five severe cases. PCA revealed a good separation of the moderate and
severe COVID-19 patient samples (Supplementary Fig. [106]1c, d,
Figs. [107]2c, d and [108]3c, d, and Supplementary Data [109]8 and
[110]9). Gene Ontology enrichment analysis showed that most of the
up-regulated proteins are involved in the complement and coagulation
cascades, natural killer cell-mediated cytotoxicity, and platelet
activation (Fig. [111]3a and Supplementary Data [112]10). Looking in
detail the moderate and severe subgroups, we found, interestingly, that
an activated immune response emerged to a certain extent in the late
stage of the disease, while the immunosuppression effect remained in
the early stage (Fig. [113]3b, c, Supplementary Figs. [114]11–[115]13,
and Supplementary Data [116]10 and [117]11). These results indicate
that in the late stage of the disease the immune response was
activated, which is consistent with an excessive immune response and
cytokine storm in patients in severe and critical stages of COVID-19
patients^[118]19–[119]21. Our study also identified two characteristic
proteins, immunoglobulin lambda variable 3–25 and elongation factor
1-alpha 1, that may indicate the progression of the two stages of the
COVID-19 disease. (Fig. [120]3d).
Fig. 3. Immune system response in moderate and severe COVID-19 patients.
[121]Fig. 3
[122]Open in a new tab
a Heatmap depicting the levels of differentially identified proteins in
patients with moderate and severe COVID-19. The graphs show the
relative intensity of differentially expressed proteins. Proteins
included in the heatmap meet the requirement that fold change >2 or
<0.5 and p value (t test) of <0.05 comparing severe to moderate patient
samples. Color bar represents the relative intensity of identified
proteins from −2 to 2. b The interaction diagram of proteins involved
in the innate immune response, response to the virus, antigen
processing and presentation, and T cell activation. Network nodes and
edges represent proteins and protein–protein associations. Green solid
lines represent inhibition; gray dotted lines represent GO pathways.
Color bar from red to green represents the fold change of protein level
from increasing to decreasing. The significance of the pathways
represented by −log(p value) (Fisher’s exact test) was shown by color
scales with dark blue as most significant. c The interaction diagram of
proteins involving in antigen processing and presentation, complement
activation, cellular response to chemokine, regulation of immune
response, T cell activation, and T cell receptor signaling pathway.
Green solid lines represent inhibition; red solid lines represent
activation; gray dotted lines represent GO pathways. Color bar from red
to green represents the fold change of protein level from increasing to
decreasing. The significance of the pathways represented by −log(p
value) (Fisher’s exact test) was shown by color scales with dark blue
as most significant. d The scatter plot graphs showing two proteins
that are potential diagnostic markers for severe COVID-19. One-way
ANOVA was used to analyze the data. For Health group, n = 10; for nCoV
Moderate group, n = 9; for nCoV Severe group, n = 5. Data are presented
as mean ± SEM. Source data are provided as a Source Data file.
Discussion
We applied the most advanced mass spectrometry technology to perform
quantitative proteomics analysis of urine samples from COVID-19
patients and healthy controls and non-COVID-19 pneumonia patients.
Several studies have shown that the compositions of proteins detected
in urine samples can genuinely reflect the changes of the body health
condition. The majority of urinary proteins originate from plasma
components that pass through the glomerular filtration barrier, as well
as liberated proteins from the kidney and urinary tract. Thus, in the
absence of primary urological disease, the protein composition of urine
is an appropriate mirror of general health status^[123]22,[124]23. We
uncovered changes in the protein landscape that revealed
immunosuppression and impaired tight junctions in COVID-19 patients in
the early stage of infection. Intriguingly, we also detected immune
activation to a certain extent in the late stage of infection
(Fig. [125]4). These results will provide an important molecular basis
for understanding the clinical symptoms of patients, and will shed
light on how the two stage of the infection proceeds. Our data suggest
that more attention should be paid to the dysregulation that occurs in
the early onset of the infection. The limitation of this study is that
the dataset is correlative but not longitudinal. Nevertheless, these
unusual features of COVID-19 during the course of human infections will
guide us to further understanding of COVID-19 pathogenesis, mechanistic
study, and clinical treatments.
Fig. 4. A “two-stage” model of COVID-19 pathogenesis.
[126]Fig. 4
[127]Open in a new tab
The first stage of COVID-19 might involve suppression of the immune
system and damage to tight junctions. The second stage might involve
activated immune responses, contributing to cytokine storm and organ
damage.
Methods
Sample collection
Urine samples were collected from Beijing Youan Hospital, Capital
Medical University, and Chinese Center for Disease Control and
Prevention between March 25 and April 10, 2020. Detailed information is
shown in Supplementary Tables [128]1 and [129]2. COVID-19 patients were
diagnosed according to the Chinese Government Diagnosis and Treatment
guideline (Trial 5th Version) (Medicine 2020).
Ethics statement
Ethics approval was exempted by the institutional review board of the
hospital as we collected and analyzed all data from the patients
according to the policy issued by the National Health Commission of the
People’s Republic of China. Informed consent was obtained from all
participants.
Sample preparation for DIA analysis
A measure of 200 µl urine was centrifuged at 6000 × g for 10 min at
4 °C. The supernatant was precipitated by trichloroacetic acid solution
at 4 °C for 4 h, and then centrifuged at 16,000 × g for 30 min at 4 °C.
After washing three times with acetone, the precipitate was dried with
vacuum concentrator (Labconco, USA). The dried precipitate was
resuspended in 40 µl 8 M urea in 500 mM Tris-HCl buffer (pH 8.5),
incubated with 20 mM (2-carboxyethyl) phosphine hydrochloride (TCEP)
(500 mM in 100 mM Tris/HCl pH 8.5) at room temperature for 20 min, and
then alkylated with 40 mM iodoacetamide in the dark for 30 min. The
mixture was diluted with 200 µl 100 mM Tris-HCl buffer (pH 8.5) to a
final concentration of 1.3 M urea, followed by digestion with 3 µg
trypsin protease to a final concentration of 0.0125 µg/µl at 37 °C for
16 h. Digestion was quenched by the addition of the formic acid (FA) at
a final concentration of 5%. The sample was desalted using Monospin C18
column (GL Science, Tokyo, Japan).
The 1/3 volume of eluent was taken out and dried in vacuum. The
peptides were re-dissolved with Milli-Q water and the concentration was
measured using a BCA Peptide Assay Kit following the manufacturer’s
instructions.
The 2/3 remaining purified peptides were vacuum-centrifuged to dryness
and reconstituted in Milli-Q water with 0.1 vol% FA for liquid
chromatography-mass spectrometry analysis. For DIA experiments, iRT
(indexed retention time) calibration peptides were spiked into the
sample.
Sample preparation for spectral library
Eighty-eight micrograms of mixed protein from 10 healthy and 11
COVID-19 patients were processed as above and the added enzyme to
protein ratio is 1:50. Purified peptides were reconstituted in 80 µl
fraction buffer A (98% H[2]O, 2% acetonitrile).
High-pH reversed-phase fraction
Approximately 88 µg mixed peptides were fractioned on a Chromatographic
column (BEH C18, 300 Å, 1.7 µm, 1 mm × 150 mm) coupled to a Waters
Xevo^TM ACQUITY UPLC (Waters, USA) within 80 min gradient and
concatenated into 62 fractions. The first fraction is mixed with the
last fraction, and the rest is mixed with two fractions every 30
fractions sequentially. Finally, 31 fractions were obtained. All
fractions were vacuum-centrifuged to dryness and reconstituted in 10 μl
Milli-Q water with 0.1 vol% FA. iRT peptides were spiked before
data-dependent acquisition (DDA) analysis.
Liquid chromatography
We employed a nanoElute liquid chromatography system (Bruker
Daltonics). Peptides (200 ng of digest) were separated within 90 min at
a flow rate of 300 nL/min on a 25 cm × 75 μm column with a laser-pulled
electrospray emitter packed with 1.5 μm ReproSil-Pur 120 C18-AQ
particles (Dr. Maisch). Mobile phases A and B were water and
acetonitrile with 0.1 vol% FA, respectively. The %B was linearly
increased from 2 to 22% within 70 min, followed by an increase to 37%
within 8 min and a further increase to 95% within 5 min before the last
7 min 95% process.
Mass spectrometry
All 31 fraction samples were analyzed on a hybrid trapped ion mobility
spectrometry (TIMS) quadrupole time-of-flight mass spectrometer (MS)
(TIMS-TOF Pro, Bruker Daltonics) via a CaptiveSpray nano-electrospray
ion source. The MS was operated in data-dependent mode for the ion
mobility-enhanced spectral library generation. We set the accumulation
and ramp time was 100 ms each and recorded mass spectra in the range
from m/z 100–1700 in positive electrospray mode. The ion mobility was
scanned from 0.6 to 1.6 Vs/cm^2. The overall acquisition cycle of
1.16 s comprised one full TIMS-MS scan and 10 parallel
accumulation-serial fragmentation (PASEF) MS/MS scans. When performing
DIA, we define quadrupole isolation windows as a function of the TIMS
scan time to achieve seamless and synchronous ramps for all applied
voltages. We defined up to eight windows for single 100 ms TIMS scans
according to the m/z-ion mobility plane. During PASEF MSMS scanning,
the collision energy was ramped linearly as a function of the mobility
from 59 eV at 1/K0 = 1.6 Vs/cm^2 to 20 eV at 1/K0 = 0.6 Vs/cm^2.
Generation of spectral library and DIA-PASEF processing
Raw files were processed using a developmental version of Spectronaut
(v14.0.200409.43655, Biognosys). The ion mobility-enhanced library was
generated from DDA-PASEF raw data using Spectronaut’s Pulsar database
search engine with 1% false discovery rate control at peptide-spectrum
match, peptide, and protein level. Carbamidomethyl (C) was set as fixed
modifications, and oxidation (M) and acetyl (protein N-term) were set
as variable modifications. For the subsequent targeted analysis of
DIA-PASEF data, DIA files were processed using Spectronaut with default
settings, but the correction factor of XIC IM extraction window was set
to 0.8 instead of 1.0. Q values at the precursor and protein level were
set to <1%. Each patient sample is treated as a biological duplicate. A
quality control sample of mixed aliquots from each sample was applied
for every four sample run. The median coefficient of variation for
quantification was 18.6% on the protein level after median
normalization.
Statistical analysis
To impute the proteomic data, we first used locally weighted polynomial
regression^[130]24 (lowess in R v.3.6.3^[131]25) to compute the local
polynomial fit for protein number and protein-detecting rate in each
stage(time point). Two boundary thresholds, 0.15 and 0.5, were used to
separate the data into three parts. When a protein-detecting rate is
<0.15, it is probably because the detected value is due to a technical
error. For these proteins, no imputation was applied. When a
protein-detecting rate is >0.5, the missing value was probably due to
the detection accuracy limitation of the LC/MS. In this case, the
missing value was replaced with a median value. When a
protein-detecting rate is between 0.15 and 0.5, it is probably because
the protein expression is unstable for detection.
In this case, we first calculated the missing probability of a protein
using Bayes theory,
[MATH: missp=PA(PBA/(
(PBA×PA)+(0.05×(1−PA)))), :MATH]
where PBA is the group missing rate and PA the total missing rate of
each protein.
Then, we determined the predicted imputation number (IN) of each
protein in each group,
[MATH: IN=(1<
mo>−missp)Mi, :MATH]
where Mi is the number of undetected sample number of a protein in
group i.
Finally, the random method was used to determine the samples to be
imputed. The imputation value^[132]15 was then defined by,
[MATH: IV=min(Mi/2,IN). :MATH]
Imputed data were then normalized using LogNorm algorithm. PCA (muma
v1.4 package, [133]https://www.rdocumentation.org/packages/muma) and
fastcluster v.1.1
([134]https://www.rdocumentation.org/packages/fastcluster/versions/1.1.
25/) using euclidean distance was used to perform the clustering
analysis of samples.
R package Genefilter
([135]https://www.rdocumentation.org/packages/genefilter/versions/1.54.
2) was used in the calculation of the fold-change values of proteins.
Fold change of 2, fold change <0.5, and p value (t test) of 0.05 were
used to filter differential expression proteins.
Mfuzz v.2.46.0
([136]https://www.bioconductor.org/packages/release/bioc/html/Mfuzz.htm
l) was used to detect different sub-clustering models of gene
expression among groups. R v.3.6.3^[137]26 was used to implement
Fisher’s exact test. String version 11^[138]27 was used for
protein–protein interaction network analysis.
The KEGG ligand database was used to obtain the compound and enzyme
relationship. Venn diagram, heatmap, and network visualization were
performed using the ggplot2^[139]28 packages and Cytoscape
v.3.5.1^[140]29 implemented in the omicsbean workbench. Ingenuity
pathway analysis was performed to explore the downstream effect in
significant regulated proteins dataset. The z-score algorithm was used
to predict the activation state (either activated or inhibited)^[141]30
of the biological process. If the z-score ≤ −2, the process is
predicted to be statistically significantly inhibited.
Reporting summary
Further information on research design is available in the [142]Nature
Research Reporting Summary linked to this article.
Supplementary information
[143]Supplementary Information^ (1.4MB, pdf)
[144]Peer Review File^ (738.3KB, pdf)
[145]41467_2020_19706_MOESM3_ESM.docx^ (11.8KB, docx)
Descriptions of Additional Supplementary Files
[146]Supplementary Data 1^ (1.9MB, xlsx)
[147]Supplementary Data 2^ (1.9MB, xlsx)
[148]Supplementary Data 3^ (3.8MB, xls)
[149]Supplementary Data 4^ (272KB, xls)
[150]Supplementary Data 5^ (131.5KB, xls)
[151]Supplementary Data 6^ (18.8KB, xlsx)
[152]Supplementary Data 7^ (730.7KB, xlsx)
[153]Supplementary Data 8^ (1.8MB, xlsx)
[154]Supplementary Data 9^ (3.8MB, xls)
[155]Supplementary Data 10^ (29KB, xlsx)
[156]Supplementary Data 11^ (274.5KB, xls)
[157]Reporting Summary^ (379.9KB, pdf)
Acknowledgements