PLANT MICROBIOTA
Pathogen-induced activation of disease-suppressive
functions in the endophytic root microbiome
Víctor J. Carrión
1,2
, Juan Perez-Jaramillo
1,3
*, Viviane Cordovez
1,2
*, Vittorio Tracanna
4
*,
Mattias de Hollander
1
, Daniel Ruiz-Buck
1
, Lucas W. Mendes
5
, Wilfred F.J. van Ijcken
6
,
Ruth Gomez-Exposito
1,7
, Somayah S. Elsayed
2
, Prarthana Mohanraju
7
, Adini Arifah
7
,
John van der Oost
7
, Joseph N. Paulson
8
, Rodrigo Mendes
9
, Gilles P. van Wezel
1,2
,
Marnix H. Medema
4
†, Jos M. Raaijmakers
1,2
†
Microorganisms living inside plants can promote plant growth and health, but their genomic and
functional diversity remain largely elusive. Here, metagenomics and network inference show that fungal
infection of plant roots enriched for Chitinophagaceae and Flavobacteriaceae in the root endosphere
and for chitinase genes and various unknown biosynthetic gene clusters encoding the production of
nonribosomal peptide synthetases (NRPSs) and polyketide synthases (PKSs). After strain-level genome
reconstruction, a consortium of Chitinophaga and Flavobacterium was designed that consistently
suppressed fungal root disease. Site-directed mutagenesis then revealed that a previously unidentified
NRPS-PKS gene cluster from Flavobacterium was essential for disease suppression by the endophytic
consortium. Our results highlight that endophytic root microbiomes harbor a wealth of as yet unknown
functional traits that, in concert, can protect the plant inside out.
P
ast and present plant microbiome studies
have generated a large amount of se-
quence data and a wealth of (mostly)
descriptive information on the diversity
and relative abundance of different tax-
onomic groups in the rhizosphere, phyllo-
sphere, spermosphere, and endosphere of a
multitude of plant species (1, 2). To date, how-
ever , relatively few studies have demonstrated
the functional importance of microbiomes
for specific plant phenotypes, tha t is, plant
growth, development, and health (3–9). Fur-
thermore, the molecular and chemical basis of
the causal relationships between these plant
phenotypes and microbiome structure and
functions are, in most cases, still unknown.
The aim of this study was to investigate the
genomic diversity and functional potential
of the endophyt ic roo t micro biom e in the pro-
tection of plants against fungal infections. To
this end, we integrated multiple approaches,
including network inference and metagenom-
ics, to identify root endophytic bacterial consor-
tia and functional gene clusters associated with
a soil that is suppressive to disease caused by
Rhizoctonia solani, a fungal root pathogen of
several plant species, including rice, wheat, and
sugar beet.
Disease-suppressive soils are exceptional eco-
systems in which plants are protected from
root pathogens as a result of antagonistic
activities of the root-associated microbiome.
Suppressive soils have been described for var-
ious soil-borne pathogens, including fungi,
bacteria, oomycetes, and nematodes (3, 5, 10–15).
Disease suppression can be eliminated by se-
lective heat treatment and can be transplanted
to nonsuppressive (conducive) soils, analogous
to fecal transplants in humans (5, 16). Specific
suppression of soils to fungal root pathogens,
such as R. solani, is induced in field soils by a
disease outbreak during continuous cultivation
of a susceptible host plant (17). Once estab-
lished, the suppression can dissipate if nonhost
plants are grown but is regained in the pres-
ence of the host plant and the specific fungal
pathogen. Therefore, the three-way interactions
between the fungal pathogen, the host plant,
and its root microbiome are key elements of
the onset and persistence of specific disease
suppression. We previously showed that in a
soil suppressive to the fungal root pathogen
R. solani, several bacterial genera inhabit-
ing the rhizosphere of sugar beet, in particu-
lar Pa raburkholder ia, Pse udomonas,and
Streptomyces (5, 18, 19), act as a first line of
defense. To understan d what role micro-
organisms that live within plant root tissues
(endophytes) play in disease suppression,
we conducted a metagenomic analysis of the
endosphere of sugar beet seedlings grown in
field soil suppressive to R. solani and identi-
fied the microorganisms associated with disease
suppres sion, distinguished which biosynthetic
gen
e clusters (BGCs) were up-regulated during
infection, reconstructed synthetic endosphere
consortia, and finally made site-directed muta-
tions to test the role of specific BGCs in disease
suppression.
Taxonomic diversity and network inference of
the endophytic microbiome
Sugar beet plants were grown in disease-
conducive (C) and disease-suppressive (S)
soils inoculated (or not) with the root path-
ogen R. solani (fig. S1). Disease incidence in
the pathogen-inoculated suppres sive soil (S+R)
was 15 to 30%, whereas disease incidence in
the pathogen-inoculated conducive soil (C+R)
exceeded 80% (fig. S1A), typical of our pre-
vious studies (5, 16). Given the high disease
incidence in C+R, there was not enough root
material left for in-depth microbiome analysis
of this condition. The taxonomic diversity and
functional potential of the root endophytic
microbiome of plants grown in the remaining
three soil conditions (C, S, and S+R) was in-
vestigated after 4 weeks of plant growth. After
metagenome sequencing and bioinformatic
analyses (fig. S2 and tables S1 and S2), tax-
onomic assignment of the microbial cell frac-
tion from the sugar beet endosphere showed
that 76.1, 10.5, and 0.0065% of the sequence
reads corresponded to the domains Bacteria,
Eukarya, and Archaea, respectively (fig. S3, A
and B). For the eukaryotic reads, constrained
analysis of principal coordinates showed sig-
nificant differences [permutational multivariate
analysis of variance (PERMANOVA), P <0.05]
between the endophytic fungal community
composition in C, S, and S+R (fig. S4A). This
was largely due to a statistically significant
increase in Rhizoctonia-related sequence reads
in the suppressive soil inoculated with R. solani
(S+R) (fig. S4, B and C). Most of the other se-
quence reads could not be reliably assigned
to specific fungal taxa. Collectively, these re-
sults indicate that after inoculation into the
disease-suppressive soil, R. solani colonized
and penetrated the plant roots but caused
little disease.
16S ribosomal RNA (rRNA) data from the
metagenome sequences (fig. S2) showed that
Proteobact eria and Bacteroide tes dominated
the endophytic bacterial community, with 10
operational taxonomic units spanning Pseud o-
monadaceae (two), Xanthomonadaceae (fo ur ),
Chitinophagaceae (one), Flavobac teriaceae
(two), and Veillonellaceae (one) (fig. S5), all
of which became enriched in the S+R condi-
tion compared with the S condition (Fig. 1A).
Co-occurrence network analysis revealed in-
creased complexity in the S+R condition (fig. S6,
A to C, and table S3) compared with C and
S conditions (table S3). Highly connected
RESEARCH
Carrión et al., Science 366, 606–612 (2019) 1 November 2019 1of7
1
Department of Microbial Ecology, Netherlands Institute of
Ecology (NIOO-KNAW), Droevendaalsesteeg 10, 6708 PB
Wageningen, Netherlands.
2
Institute of Biology, Leiden
University, Sylvius weg 72, 2333 BE Leiden, Netherlands.
3
PECET, University of Antioquia, Medellín, Antioquia 050010,
Colombia.
4
Bioinformatics Group, Wageningen University,
Droevendaalsesteeg 1, 6708 PB Wageningen, Netherlands.
5
Cell and Molecular Biology Laboratory, Center for Nuclear
Energy in Agriculture (CENA), University of Sao Paulo (USP),
Piracicaba, Brazil.
6
Erasmus MC, University Medical Center
Rotterdam, Department of Cell Biology, Center for Biomics,
3025 CN Rotterdam, Netherlands.
7
Laboratory of
Microbiology, Wageningen University and Research,
Stippeneng 4, 6708 WE Wageningen, Netherlands.
8
Department of Biostatistics, Product Development,
Genentech Inc., South San Francisco, CA 94080, USA.
9
Laboratory of Environmental Microbiology, Brazilian
Agricultural Research Corporation, Embrapa Environment,
Rodovia SP 340, Km 127.5, 13820-000 Jaguariúna, Brazil.
*These authors contributed equally to this work.
†Corresponding author. Email: j.raaijmakers@nioo.knaw.nl
(J.M.R.); marnix.medema@wur.nl (M.H.M.)
on May 19, 2021 http://science.sciencemag.org/Downloaded from
Carrión et al., Science 366, 606–612 (2019) 1 November 2019 2of7
Fig. 1. Pathogen-induced changes in endophytic microbiome diversity and
functions. Differential abundance of endophytic bacterial communities from
plants grown in S or S+R soils. (A) Taxonomic differences are based on 16S rRNA
sequences extracted from the metagenome. The largest circles represent phylum
level, and the inner circles represent class, family, and genus. (B) Functional
differences are based on the metagenome sequence data and assigned to
taxonomic groups. The smallest circles represent the COG categories groups.
The circle sizes represent the mean read relative abundance of the differentially
abundant taxa and functions. Bacterial taxa or functions that are significantly
enriched (FDR < 0.1) in the comparison between S and S+R are indicated in
green for S and in blue for S+R; nonsignificant taxa and functions are indicated in
yellow. (C) Strip plot depicting the average abundance ratios of all genes from
Bacteroidetes belonging to core COG functional categories that contain
significantly enriched genes in S+R (Sr) compared with S and in S compared
with C. Categories are sorted from top to bottom by S+R/S ratio. Each COG type
is abbreviated as follows: C, energy production and conversion; D, cell cycle
control, cell division, and chromosome partitioning; E, amino acid transport
and metabolism; F, nucleotide transport and metabolism; G, carbohydrate
transport and metabolism; H, coenzyme transport and metabolism; I, lipid
transport and metabolism; J, translation, ribosomal structure, and biogenesis;
K, transcription; L, replication, recombination, and repair; M, cell wall, cell
membrane, and cell envelope biogenesis; O, posttranslational modification,
protein turnover, and chaperones; P, inorganic ion transport and metabolism;
Q, secondary metabolites biosynthesis, transport, and catabolism; T, signal
transduction mechanisms; U, intracellular trafficking, secretion, and vesicular
transport; and V, defense mechanisms.
RESEARCH | RESEARCH ARTICLE
on May 19, 2021 http://science.sciencemag.org/Downloaded from
networks, like those in the S+R samples, can
occur when microbiota face environmental
perturbation, such as pathogen invasion (20).
Interestingly, 80% of the interacting nodes in
the S+R network belonged to Chitinophaga,
Flavobacterium,andPseudomonas species
(table S4). When sequence reads from the
Bacteroidetes were removed from the datasets,
the endophytic signals from the C and S soils
were indistinguishable (fig. S7, A and B), once
again indicating an association of the Bacteroi-
detes genera Chitinophaga and Flavobacterium
with the disease-suppressive phenotype.
Functional diversity of the endophytic
microbiome
Of the genes retrieved from the metagenome
data, 50 to 70% were assigned to a known
function (fig. S3, C to E). For the other genes,
grouping annotations indicated 56,175 taxa-
associated functions, of which 402 functions
were significantly enriched in the endophytic
bacterial community of plants grown in the
S soil compared with that of plants grown in
the C soil [false discovery rate (FDR) < 0.1;
fig. S8, B and C]. In the S+R condition, this
proportion of functional enrichment increased
more than 10-fold (4443) (FDR < 0.1; Fig. 1B).
These genes belonged mainly to pathways clas-
sified as “carbohydrate transport and metabo-
lism” and “signal transduction mechanisms.”
Several endophytic bacterial families—including
Chitinophagaceae and Flavoba cteriaceae
(Bacteroidetes); Pseudomonadaceae and
Xanthomonadaceae (Gammaproteo bacteri a);
Hyphomicrobiaceae and Rhizobiaceae (Alpha-
proteobacteria); and Burkholderiaceae
(Betaproteobacteria)—were specifically asso-
ciated with the functional enrichment we
observed (Fig. 1, B and C, and fig. S9A). The
majority of the overrepresented genes in S+R
(3138 genes of 4443) were associated with
Chitinophagace ae and Flavobacteriaceae (Fig.
1B and fig. S9A). When we used a more strin-
gent significance level of P <0.05,2063of
56,175 taxa-associated functions were over-
represent ed, with 461 functions associated
mainly with Chitinophagaceae and Flavo-
bacteriaceae. Cumulative differential abun-
dance analyses of all Bacteroidetes ’ genes
between samples highlighted that genes from
cluster of orthologous groups (COG) category
Q (secondary metabolit es biosynthesis, trans-
port, and catabolism) were among the most
differentially abundant between S+R and S,
whereas genes from category G (carbohy-
drate transport and metabolism) were among
the most differentially abundant between S
and C (Fig. 1C).
For more detailed resolution of the specific
functions associated with COGs G and Q,
we searched for carbohydrate-active enzymes
(CAZymes) and secondary metabolite bio-
synthetic gene clusters within the me ta-
genome sequences using dbCAN (21, 22)and
antiSMASH (23), respectively. Using dbCAN,
we were able to annotate 1822 genes in the
endophytic metagenome with glycoside hy-
drolase, glycosyltransferase, polysaccharide
lyase, and carbohydrate esterase domains, as
well as noncatalytic carbohydrate-binding
modules. Because many of these domains are
evolutionary related and have related func-
tionalities, we mapped the domain diversity
in a protein family similarity network constructed
using the hhsearch algorithm (24). Glycoside
hydrolases and glycosyltransferases were more
abundant in the S+R endophytic microbiome
Carrión et al., Science 366, 606–612 (2019) 1 November 2019 3of7
Fig. 2. Diversity and distribution of carbohydrate-active enzymes in the
endophytic microbiome. (A) Similarity network of known and putative protein
domains of enzymes involved in carbohydrate metabolism (CAZymes). From
the endophytic metagenome of plants grown in suppressive soil (S) or in
suppressive soil inoculated with the fungal root pathogen R. solani (S+R), a
total of 1822 genes were annotated as CAZymes. Domain-domain distances and
their relatedness are shown in the network. Nodes were grouped into five
functional classes: glycoside hydrolases (GH, blue), glycosyltransferases (GT,
ora nge), polysaccharide lyases (PL, purple), carbohydrate esterases
(CE, green ), an d the noncataly tic carbohydr ate-binding modules (C BM, red).
Unknown domains or domains for which the function has not been
experim entally validated are shown in yello w. Squared nodes represent
enzymes that are significan tly overrepresented (F DR < 0.1) in S+R compared
with S and taxonomically assigned t o the Chitinophagaceae. Enzymes
significantly overrepresented in S+R and taxonomically class ified as Burkhol-
der iace ae and Xanthomon adaceae a re shown in fig. S9, B and C, respectively.
(B) Venn diagram with different CAZymes annotated for three endophytic
bac terial families enriched in S+R, t hat is, Burkholderiaceae (yellow),
Chitinophag aceae (blue), and Xanthomonadaceae (green). For each of the
CAZymes, the Pfam number is shown in parentheses. The Venn diagram shows
the number of domains detected exclusively for each bacterial family and the
domains shared by these endophytic bacterial families.
RESEARCH | RESEARCH ARTICLE
on May 19, 2021 http://science.sciencemag.org/Downloaded from
Carrión et al., Science 366, 606–612 (2019) 1 November 2019 4of7
Fig. 3. Diversity and distribution of biosynthetic gene clusters in the
endophytic microbiome. (A) Sequence similarity network [constructed with
BiG-SCAPE (32), threshold: 0.8] of the different classes of BGCs detected
in the endophytic microbiome. Taxonomic assignment and BGC class annotation
of the nodes are shown. Nodes with fewer than three connections were removed;
the original network with all nodes can be found in fig. S10. Node colors
represent statistical significance based on a Welch’s t test (FDR < 0.1): Yellow
nodes are nonsignificant, and blue nodes are significantly overrepresented in the
S+R condition. (B) Number of overrepresented BGCs (two-tailed Welch’s t test,
P < 0.1) detected by the antiSMASH and Clusterfinder algorithms for the
different bacterial phyla in the endophytic root microbiome of plants grown in
C, S, and S+R soils. (C to E) Number and type of BGCs assigned to
Proteobacteria (C), Bacteroidetes (D), and unclassified (E) bacterial phyla that
were significantly (two-tailed Welch’s t test, P < 0.1) more enriched in S+R
BGCs that could not be classified are not included in the (C) to (E) barplots.
(F) Clustered heat map of relative abundances [cumulative sum scaling (CSS)–
normalized RPKM (reads per kilobase per million reads) values] of the 33 NRPS
gene clusters that were significantly overrepresented in the different replicate
samples of S or S+R versus C. The NRPS cluster number and the corresponding
taxonomic assignment are shown on the right side of the panel.
RESEARCH | RESEARCH ARTICLE
on May 19, 2021 http://science.sciencemag.org/Downloaded from
andcorrelatedwithdiseasesuppression(Fig.2A
and fig. S9, B and C). Three endophyte fam-
ilies (Chitinophagaceae, Burkholderiaceae, and
Xanthomonadaceae) showed statistically sig-
nificant differences in CAZyme composition
between S+R and S (FDR < 0.1; Fig. 2A and
fig.S9,AandB).Furthermore,wefoundthat
Chitinophagaceae harbored several enzymes
with domains associated with fungal cell-wall
degradation, such as chitinases, b-glucanases,
and endoglucanases (Fig. 2A), and also possessed
debranching enzymes, including a -1,6-mannanase
and a-
L-rhamnosidase. Burkholderiaceae and
Xanthomonadaceae families (fig. S9, B and C)
also contributed two chitinase domains and
three other enzymes involved in chitin degra-
dation, including chitin deacetylase and
chitosanase. Only five domains were shared
between Chitinophagaceae, Burkholderiaceae,
and Xanthomonadaceae (Fig. 2B), indicating
limited functional redundancy among these
endophytes for this trait. The enrichment of gene s
encoding chitin-degradingenzymespointstoarole
in disease suppression for these endophytes (25).
Bacterial genomes contain a large diversity
of BGCs, the vast majority of which have not
yet been linked to specific molecules or func-
tions (5, 26–28). Our antiSMASH analysis for
secondary metabolites revealed a total of 730
BGCs associated with the biosynthesis of non-
ribosomal peptides, polyketides, terpenes, aryl
polyenes, ribosomally synthesized and post-
translationally modified peptides (RiPPs),
Carrión et al., Science 366, 606–612 (2019) 1 November 2019 5of7
Fig. 4. Transcriptional and functional analyses of disease-suppressive
consortia. (A) Genetic organization of BGC298, BGC396, BGC471, and BGC592
identified in both the Flavobacterium MAG nbed44b64 and in the genome
sequences of the four endophytic Flavobacterium isolates. Shown below the NRPS
and PKS genes are the module and domain organizations of the encoded proteins.
The domains are labeled as follows: C, condensation; A, adenylation; KS, ketosynthase;
AT, acyltransferase; PCP, peptide carrier protein; and TE, thioesterase. Predicted
substrates of the NRPS and PKS modules in BGC298 are glycine, malonyl-CoA,
and, again, glycine. (B and C) Quantitative polymerase chain reaction (qPCR)–
based analysis of the expression of BGC298, BGC396, BGC471, BGC592, and
chitinase genes (GH18) in the rhizosphere and endosphere of sugar beet seedlings
treated with the synthetic endophytic consortium of Chitinophaga and Flavo-
bacterium isolates (syncom). LogRQ represents the gene expression levels by
relative quantification scores: Values below 0 indicate lower expression of the BGC
relative to that of the housekeeping gene (glyA) used for data normalization.
Bars represent the average of three to five biological replicates per treatment, and
error bars indicate the standard error of the mean. Different letters indicate
statistically significant differences between treatments as determined by one-way
ANOVA with post hoc Tukey honestly significant different (HSD) test (P < 0.05).
Rs, R. solani.(D to F) Results of thr ee independent bioassays showing
Rhizoctonia damping-off disease incidence of sugar beet seedlings treated with
single Chitinophaga (Ch93, Ch94, and Ch95) and Flavobacterium (Fl96, Fl97, Fl98,
and Fl5B) isolates and with a consortium of all seven endophytic isolates (synthetic
community, syncom 7) [(D) and (E)] or treated with single Chitinophaga (Ch94)
and Flavobacterium (Fl98) isolates, two independent Fl98 mutants (Fl98-1 and
Fl98-2) with a deletion in BGC298, the consortium of Ch94 and Fl98 (syncom 2),
and syncom 7 (F). For (D) to (F), single isolates and the two syncoms were applied
at an initial density of 10
7
colony forming units/g of Rhizoctonia-conducive field
soil. Bars represent the average of four to eight biological replicates per treatment,
and error bars represent the standard error of the mean. Disease incidence
was scored 21 to 28 days after R. solani inoculation. Different letters indicate
statistically significant differences between treatments as determined by one-way
ANOVA with post hoc Tukey HSD test (P < 0.05). For (B) to (F), box plots with the
individual data of each replicate are provided in figs. S20 and S21.
RESEARCH | RESEARCH ARTICLE
on May 19, 2021 http://science.sciencemag.org/Downloaded from
