https://orcid.org/0000-0002-4394-0866
ABSTRACT
Meloidogyne graminicola, also known as the rice root-knot nematode, is one of the most damaging plant-parasitic nematode, especially on rice. This obligate soilborne parasite induces the formation of galls that disturb the root morphology and physiology. Its impact on the root microbiome is still not well described. Here, we conducted a survey in Northern Vietnam where we collected infected (with galls) and non-infected root tips from the same plants in three naturally infested fields. Using a metabarcoding approach, we discovered that M. graminicola infection caused modifications of the root bacterial community composition and network structure. Interestingly, we observed in infected roots a higher diversity and species richness (+24% observed ESVs) as well as a denser and more complex co-occurrence network (+44% nodes and +136% links). We identified enriched taxa that include several hubs, which could serve as potential indicators or biocontrol agents of the nematode infection. Moreover, the community of infected roots is more specific suggesting changes in the functional capabilities to survive in the gall environment. We thus describe the signature of the gall microbiome (the ‘gallobiome’) with shifting abundances and enrichments that lead to a strong restructuration of the root microbiome.
INTRODUCTION
Plant-parasitic nematodes (PPNs) are known to cause significant crop losses (Nicol et al. 2011) and of these, Meloidogyne spp. are considered one of the most important PPNs in terms of economic importance (Jones et al. 2013). Meloidogyne spp. are telluric obligate PPNs that accomplish their life cycle in roots and have a short free-living stage in soil. Meloidogyne spp. are also known as root-knot nematodes (RKNs) because they distort the root vascular system by creating large deformations at the root tips, called galls, which are essential for their growth and reproduction. Indeed, the infectious juveniles settle in the root, where they form a feeding site by inducing giant plant cells near the endodermis, and accomplish several moults over their life cycle of 20−30 days (Cabasan, Kumar and De Waele 2012; Cabasan et al. 2014). Hyperplasia and hypertrophy of the surrounding cells of the feeding site lead to the formation of characteristic hook-shaped galls that appear two to four days after infection and that will limit root development. The giant feeding cells act as specialized sinks providing the nematodes with their nutrient requirements to reproduction (Jammes et al. 2005). As a result, the root system is atrophied, disrupting the transport of water and nutrients into the plant and compromising rice yield (Page 1982). Meloidogyne graminicola has a particularly detrimental impact in Asia where a large part of the world rice is produced and consumed. In flooded conditions, yield losses associated with M. graminicola infections of up to 80% have been reported (Plowright and Bridge 1990). Therefore, it is considered as a major threat to rice agriculture (see Mantelin, Bellafiore and Kyndt 2017 for a review).
Plants and their associated microorganisms (microbiome) form a holobiont and can be considered as co-evolved species assemblages consisting of bacterial, archaeal and diverse eukaryotic species according to Berg et al. 2017. Microbial communities indeed inhabit different plant compartments like the root endosphere (root interior), the rhizoplane (root surface) and the rhizosphere (soil influenced by the root) (Edwards et al. 2015; Ding et al. 2019). Key insights reveal close parasitic, mutualistic or pathogenic relationships between these microorganisms and the plant host (Newton et al. 2010). Through metabolic interplay and signaling, microorganisms can stimulate germination and plant growth, prevent diseases and promote stress resistance and general fitness (Berg et al. 2017). Due to the advance of -omic tools, access to the plant microbiome is possible (Sánchez-Cañizares et al. 2017) and microbe-based agronomical approaches such as the exploitation of the microbiome as a solution against PPNs are promising.
The physiological impact of M. graminicola on rice has been widely described (Jain, Khan and Kumar 2012; Cabasan et al. 2014; Patil and Gaur 2014). Contrastingly, its impact on the root microbiome is less known, although the impact of plant pathogens on the plant-associated microbiomes is suspected to have an importance on plant health and yield (Vannier, Agler and Hacquard 2019). At the plant level and particularly at the root site, Back, Haydock and Jenkinson (2002) have identified synergistic interactions between PPNs and soil-borne pathogens. In particular, the release of plant root exudates known as ‘rhizosphere effect’ is considered as an important factor in the attraction of microorganisms (Zhalnina et al. 2018). Due to the physiological impact of M. graminicola on the rice roots, the plant exudation pattern can be modified and so the nematode can indirectly affect the microbiome. Indeed, M. graminicola could affect the root-associated bacteria by modifying the plant hormonal balances (jasmonic, salicylic acids, and ethylene), inducing the production of secondary metabolites (terpenoids and flavonoids) or defense proteins (Pathogenesis-Related proteins, thaumatin and thionin) as described in the transcriptomic analysis of Petitot et al. (2017). Meloidogyne graminicola could also have an impact on root-associated microorganisms by carrying its own microbiome as it was shown for M. incognita, another root-knot nematode (Elhady et al. 2017). Finally, these direct and indirect effects could lead to modifications of the root microbiome that play an important role in plant health (Pieterse, de Jonge and Berendsen 2016).
The relationship between plant microbiomes and RKNs has been described in few studies. For instance, communities and functions of endophytes in tomato plants were compared before and after infection by M. incognita in a greenhouse assay (Tian, Cao and Zhang 2015). Some bacterial groups have been found specifically enriched in the root galls and carry genes that may be associated with the nematode pathogenesis. Another study focused on the characterization of rhizosphere microbiomes of eggplant and cucumber infected by M. incognita transplanted on tomato plants, in a greenhouse assay as well (Zhou et al. 2019). The authors highlighted some nematicidal effects and plant benefits that can be associated to taxa such as Pseudomonas sp. and Bacillus sp. with biocontrol activity. Bacillus strains were also antagonist toward one fungal pathogen of the Meloidogyne-based disease complex studied by Wolfgang et al. (2019). However, to our knowledge, no study has investigated the relationship between M. graminicola infection, rice and its root-associated microbiome in a natural environment.
In the present study, we characterized root-associated bacterial communities (comprising both endosphere and rhizoplane colonizing bacteria) of rice roots infected by M. graminicola (with apparent galls) and of non-infected roots (no apparent galls) from the same plants in three highly infested fields in Vietnam. Using a metabarcoding approach, we aimed to assess the impact of the infection by M. graminicola on the microbiome by investigating the differences in the following features: bacterial diversity and composition, community structure, enriched taxa and potential hub taxa in co-occurrence networks.
MATERIALS AND METHODS
Field description
The survey was conducted in Vietnam on the 11th of March 2017 in Nam Sach district, Hải Dương province (21˚00’ 51.1’’ N and 106˚ 19’ 33.0’’ E). Prospected lowland fields were located within the Red river delta under a humid subtropical climate (Fig. 1). The three rice fields surveyed were inside a ten hectares area with three crop rotations per year: two rice cultures and one onion culture. Farmers have grown onions for a decade in winter before cultivating two cycles of rice in spring and summer. Chemical fertilizers consist of 800−850 kg NPK per ha for the rice crop and 1000−1500 kg P2O5 + 300 kg Urea + 200 kg KCl per ha for the onion crop. Some pesticides were applied whenever pests appeared in the field but the names of the chemicals could not be recovered. For the first rice cropping cycle in 2017, 15 days after tillering, rice variety Bac Thom No7 (Oryza sativa indica) was broadcasted. In spring of 2017, due to unusual water scarcity, the fields suffered a drought stress for up to 20 days. Nearly four weeks after seeding, almost all seedlings died, presenting leaf chlorosis and small root systems with formed swelling galls. The three fields were highly infested and devastated (Fig. 1E).
Sampling site and design. Climatic features (A, data from www.weather.gov) and soil features (B, data from the VNUA) of the site localised in Hải Dương (red point, C) near Hanoï (upper blue point) in Northern Vietnam. Map of the infested fields and blocks (D) and picture of one of the three-infested fields (E) used to constitute the two sample types for this article.
Plant sampling and nematode identification
Each of the three rice fields of 3000 m2 was subdivided in four plots of 100 m2, resulting in 12 plots in total (Fig. 1D). Fifty plants at the vegetative phase in each plot were randomly picked up four weeks after seeding and carefully scanned for the presence of hook-shaped galls characteristic of M. graminicola infection. As all rice plants were infected, each root system has been divided in two sample types: the infected roots with galls and the non-infected roots without any visible gall. Only non-necrotic roots were collected. An average of three root tips (about 2 cm) with or without galls according to the sample type were collected from the same plant. This part of the root corresponds to the growing zone including both the proliferation zone and the elongation zone, with galls (if any) since the nematodes usually settle in the root tip. We pooled per sample type the root tips of 50 plants per plot. So, we used in total 600 plants to constitute n = 24 samples. The samples were kept in separate labeled plastic bags at 4˚C until laboratory analysis within 24 h. The presence of M. graminicola was confirmed in galls collected at random by acid fuchsin staining and by molecular identification of PPNs extracted from roots. SCAR markers were used and a rDNA fragment including the ITS-1 and part of the 5.8S and 28S was sequenced (Bellafiore et al. 2015).
Soil sampling and analysis
Five soil samples were collected from each of the 12 plots at 0–5 cm depth and were mixed to create one composite sample per plot. Soil properties were analyzed at the Soil Science Department Faculty of Land Management at the Vietnam National University of Agriculture (VNUA in Hanoi, Vietnam) with methods described in Motsara and Roy (2008). Briefly, soil pH was determined using a 1:5 ratio of soil/distilled water-KCl 1 M mixture and measured with a pH meter D-51 (Horiba Ltd., Kyoto, Japan). Cation exchange capacity (CEC) was determined by the ammonium acetate method. Soil total organic carbon (OC) was determined by the Walkley and Black method and the quantification of total nitrogen (N) was determined by the method of Kjeldahl. Total phosphorus (P2O5) was determined by digestion with H2SO4 and HClO4/Colorimetric method. Total sulfur (S) was determined after di-acid (HNO3–HClO4) digestion and turbidimetric method, soluble or available sulfate (SO42−) by barium sulfate precipitation and turbidimetric method. Soil texture was determined by the pipette method (Robinson), and aggregate stability was determined using the wet sieving apparatus (Eijkelkamp instrument) (stroke = 1.3 cm, at about 34 times/min, 0.053 mm and 0.25 mm mesh sieve).
PCR amplification and metabarcoding sequencing
The root samples were washed with sterile water to remove the rhizospheric soil attached to the roots. The 50 root tips for each plot were pooled according to their sample type (gall/no galls) and grinded in liquid nitrogen in a sterile mortar. DNA was extracted from 15 mg of powder of root tissues using the PowerSoil® DNA Isolation Kit (Qiagen, Netherland) following the manufacturer's instructions. Samples were pooled contributing exactly the same amount (50 ng μL−1) of DNA in the final library. PCR amplification, library and MiSeq Illumina sequencing were performed by Macrogen (Seoul, South Korea) using primers 341F (16S_V3F, 5’-CCTACGGGNGGCWGCAG-3’) and 805R (16S_V4R, 5’-GACTACHVGGGTATCTAATCC-3’) to amplify the V3 and V4 regions of the 16S rRNA gene.
Sequences processing
Qiime2 bioinformatic platform (Bolyen et al. 2019) was used to obtain Exact Sequence Variants (ESVs) feature table and its taxonomy. More concretely, paired-end reads were primer and adaptor removed by cutadapt (Martin 2011). To extract the ESV feature table, forward and reverse read truncation at 277 and 242 bp, respectively based on quality plot inspection, default chimera removal, and denoising were conducted by Dada2 (Callahan, McMurdie and Holmes 2017). We initially had 187 8244 reads and we filtered out low frequency (less than ten) and singleton/doubleton features which represented 9.65% of the reads. Taxonomic affiliations were assigned by a Naive Bayes classifier, which was trained for V3+V4 regions from Greengenes 16S rRNA database v13.8. Finally, ESVs with no affiliation at the phylum level (10.78% of the initial reads) or assigned to mitochondria or chloroplast (3.15%) were removed. We finally kept 76.41% of the initial reads and that was enough to analyze the data with no need to rarefy it. After the removal of these reads, the sequencing depth was still very good, ranging from 39 679 reads (sample 2.2I) to 74622 reads (2.3 N) with homogeneous variances (standard deviation = 8484). We finally ended up with 2202 ESVs. R code written to generate the figures are being made available on GitLab (https://gitlab.com/anne-saw/gallobiome_haiduong_2017/). The data for this study have been deposited in the European Nucleotide Archive (ENA) at EMBL-EBI under accession number PRJEB37618 (http://www.ebi.ac.uk/ena/data/view/PRJEB37618).
Microbiome structure analyzes
The rarefaction curves of the samples were checked (Fig. S1, Supporting Information) and there was no need to rarefy the data (McMurdie and Holmes 2014). Indeed, the sequencing depth was sufficient for all samples to reach a plateau around 20 000 reads. To visualize the infection and field effects on the microbiome structure, a non-metric multidimensional scaling (NMDS) with Bray−Curtis distance was drawn using the Vegan package (Oksanen et al. 2008) with R software (version 3.5.2). We performed a permutation test to check the multivariate homogeneity of variance with the function betadisper. After that, we performed an Adonis test (permutational multivariate analysis of variance using distance matrices) to look for an effect of the root infection by the nematode (infection effect) or the sampling localization (field effect) on the microbiome structure. An envfit test was used to look for correlations of the community structure with environmental variables. The ggplot2 package was used to build the figures.
Diversity analyzes
To assess the diversity of the bacterial communities within and between samples, we measured the observed ESVs richness and calculated the Shannon and Pielou's evenness indices using the Vegan package. After that, we performed statistical tests with a generalized linear model (GLM). The best fit for the three measurements was found with a gamma distribution and a log scale. Only significant effects (infection effect, field effect or dependency between the two) are written on the figures and indicated with asterisks. The tests were performed using the R packages Rmisc, MASS and car.
Phylogenetic and differential abundance testings
To better characterize the rice root microbiome composition at a phylogenetic level, we drew an unrooted phylogenetic tree with specifically enriched taxa in non-infected or in infected roots, using the Metacoder package (Foster, Sharpton and Grünwald 2017). Metacoder is a set of tools for parsing, manipulating and graphing data classified by a hierarchy such as taxonomic data. In a nutshell, it sums the read counts per taxon (i.e. calculates the total ESVs’ read count), converts them to proportions for every taxonomic level and represents them on a tree. We drew a detailed (with all enriched taxa) and a simplified (with total ESVs’ read count > 50) version of the same tree. The Metacoder trees allowed us to visualize the overall enrichment of bacterial clades along the phylogenetic tree, highlighting some signatures of enrichment, in order to focus our further analyzes. In parallel, we performed DESeq2 (Love, Huber and Anders 2014) with an adjusted P-value lower than 0.05. DESeq2 is used to calculate differential abundances of entities between two conditions, allowing us to compare the abundance of taxa in non-infected versus infected roots. One limit of DESeq2 in our analysis is that it ignores the compositionality of the community because it calculates the abundance for each individual ESV. Consequently, it can inform us about the differential abundance for one ESV, but it doesn't inform us about the enrichment at taxonomic levels that aggregate several ESVs. However, we could calculate the proportion of the enriched taxa in non-infected and in infected roots by the sum of ESVs’ read count, respectively in non-infected and in infected samples, divided by the total ESVs’ read count of this taxon. We focused on the enriched taxa at order and phylum levels according to the Metacoder trees. For the DESeq2 analysis, we focused on the enriched ESVs with extended affiliation at species level (affiliation extended at genus level if unassigned or uncultured at species level) to represent them on a graph, and on the enriched ESVs with full affiliation at species level. For those last ones, we calculated their relative abundance in the dataset.
Co-occurrence networks construction
Co-occurrence networks were generated with the packages SPIEC-EASI (Kurtz et al. 2015), igraph and ggnet. The ESVs present in non-infected or in infected roots were separated in two files in order to construct the two bacterial community networks with a threshold of 80% for taxa prevalence. We used the following parameters in SPIEC-EASI to compute the networks: method ‘mb’, lambda.min.ratio = 5e-4 and nlambda = 80. Network properties and taxa specificities were analyzed with the igraph package and detailed on the figures. To identify highly connected taxa in those networks, we considered the 5% most connected ESVs in term of betweenness centrality, closeness centrality and node degree as described and used by Agler et al. (2016). We identified these highly connected ESVs as ‘hubs’.
Functional predictions
We used the software PICRUSt2 (Douglas et al. 2019) in order to perform a predictive analysis of the functions carried by the bacterial communities present in each sample type. As this approach is more speculative, we present the detailed methodology in Sup. Method and results obtained in Figs S4 and S5 (Supporting Information).
RESULTS
The bacterial community structure is impacted by the root-knot nematode infection
First of all, the total number of reads is similar in both sample types: 715 488 in infected roots and 719 666 in non-infected roots, with a sequencing depth by sample allowing a full exploration of the bacterial communities as observed with rarefaction curves (Fig. S1, Supporting Information). The NMDS ordination drawn with the ESV table shows that the bacterial community structure of the infected roots is distinct from the one of the non-infected roots (Fig. 2A). Indeed, 17% of the variance is explained by the nematode infection (P<0.001) that we call the infection effect. The localization of the field also has a significant effect on the community structure (P < 0.001, 20%). This field effect on the community structure can be seen on the ordination along the NMDS2 axis (Fig. 3). We also looked into correlations between the bacterial community structure and environmental variables. We found that the pH H2O (P < 0.05, R² = 0.37), the available sulfate (P < 0.05, R² = 0.35), the total organic carbon (P<0.05, R² = 0.32), the clay texture (P<0.05, R² = 0.27) and the total phosphorus (P < 0.05, R² = 0.25) are the variables that are significantly correlated to the community structure. Field 3 for example has a higher total sulfur and clay texture and a lower pH H2O, total carbonates and the total phosphorus (Table S1, Supporting Information). There is no significant interaction between this field effect and the infection effect (Infection x Field effect: P = 0.25) meaning that the infection effect on the bacterial community structure is not dependent on the sampling location.
NMDS ordination (A), and diversity indices (B, C and D) of the bacterial community structure of infected rice roots by Meloidogyne graminicola (with galls) and non-infected roots (without galls). Observed ESVs Richness (B), Shannon Index (C) and Pielou's Evenness Index (D).
NMDS ordination of the bacterial community structure of infected rice roots by Meloidogyne graminicola (with galls) and non-infected roots (without galls). Environmental vectors significantly responsible for the field effect have been drawn.
The rice root-knot nematode infection increases species richness and diversity of the root bacterial community
We represented the observed exact sequence variants (ESVs) richness, Shannon index and Pielou's evenness index in Fig. 2. The number of observed ESVs is significantly impacted by the RKN infection (Fig. 2B, P < 0.01). Indeed, the ESVs richness is 24% higher in infected roots than in non-infected roots (Table S2, Supporting Information). The Shannon and Pielou's evenness indices also significantly increase with the infection (Fig. 2C and D, P < 0.001), with an increase of 9.96% and 6.46% in infected roots, respectively. It means that the infection by Meloidogyne graminicola is associated with an increase in species richness, diversity and evenness of the root bacterial community. We also observe a dependency between the infection and the field effect for both Shannon and Pielou's evenness indices (Fig. 2C and D, Infection x Field effects: P < 0.05) meaning that the soil properties shapes the bacterial diversity differently in response to the nematode infection. However, the infection effect always has a significantly higher impact on the bacterial diversity than the field effect: P < 0.001 for the infection effect on Pielou's evenness index versus P < 0.05 for the field effect (Fig. 2D) and P < 0.001 for the infection effect on Shannon index versus no significant impact for the field effect (Fig. 2C).
The root-knot nematode infection changes the relative abundance of several bacterial taxa
To explore the composition of the bacterial communities, we performed two independent analyzes to identify enriched taxa in non-infected or in infected roots: (i) one analysis with the Metacoder package that measures the enrichment of taxonomic groups at any level (grouped ESVs from phylum to species level) and (ii) the other analysis with the DESeq2 package that measures the enrichment for every single ESV.
First, the Metacoder analysis and additional calculations show that the overall distribution and the relative abundances of taxa on a phylogenetic tree highlight specific taxonomic signatures of bacterial enrichment for non-infected or for infected roots (Fig. 4). First, more taxa are enriched in infected rice roots, as shown by the higher prevalence of bright-red nodes and branches on the Metacoder representation, which is consistent with a higher observed richness and diversity in infected roots. Second, some branches are fully or partially colored meaning that the enrichment can be more or less restricted to some clades. Third, the enrichments are not only restricted to close taxa, they are spread among the branches, especially for infected roots. At the phylum level (Fig. 5A), Actinobacteria for example have a higher prevalence in infected roots: about 70% of these taxa in our samples are found in infected roots (Table S3 (Supporting Information), visualized on Fig. 4). In contrast, no phylum were found to be specifically enriched in non-infected roots. At order level (Fig. 4B), quantitatively more orders are enriched in infected roots (16 orders) than in non-infected roots (only 2 orders including Flavobacteriales). By rank of decreasing prevalence in infected roots, some enriched orders with high total abundance include Pedosphareales, Saprospirales, Actinomycetales, Sphingomonadales, Rhizobiales and Opitutales. For examples, about 65% of all the ESVs assigned to Rhizobiales (107 ESVs) and Actinomycetales (44 ESVs) are found in the infected roots (Table S3, Supporting Information). At species level, eight species are enriched in infected roots whereas four species are enriched in non-infected roots according to Metacoder enrichment trees (detailed tree in Fig. S2, Supporting Information). For instance, these species are affiliated to Rhizobiales (Agrobacterium sullae, Ensifer adhaerens and Pleomorphomonas oryzae) for the ones enriched in infected roots and Flavobacteriales (Flavobacterium succinicans), or Xanthomonadales (Silanimonas mangrovi and Stenotrophomonas maltophilia) for the ones enriched in non-infected roots. The Metacoder tree shows a high restructuration of the microbiome associated with the infection as shown with enrichments of diverse clades at different taxonomic level.
Phylogenetic tree of enriched taxa, simplified tree version. Each node represents a taxon. The color indicates the differential abundance (in binary logarithm scale) between the median counts of taxa in infected (bright-red) and non-infected (blue) rice roots. The size node and branch indicates the ESVs’ read count for each taxa. Only the taxa at order or higher levels are shown in this simplified tree version.
Enriched taxa according to Metacoder (A and B) and DESeq2 analyzes (C). Proportion and total ESVs’ read count of enriched taxa at phylum (A) and order (B) levels in non-infected or infected roots. Differential abundance of the enriched ESVs with extended affiliation at species level (affiliation extended at genus level if unassigned or uncultured at species level, C). The Proteobacteria phylum, which is not significantly different between the two sample types, is presented in the graph A to allow comparison with significant phyla and because it is the most dominant bacterial phylum associated with rice roots.
Secondly, the DESeq2 analysis enables the identification of single ESVs enriched or depleted between non-infected and infected roots with affiliation at different taxonomic levels. The results, summarized in Table 1, give 81 ESVs enriched in infected roots including 11 with a full affiliation at species level, and 17 ESVs enriched in non-infected roots including four with a full affiliation at species level (Rubrivivax gelatinosus, Pelomonas puraquae, Silanimonas mangrovia, Flavobacterium succinicans). In infected roots, Pleomorphomonas oryzae, Agrobacterium sullae, Ensifer adhaerens, Streptomyces lanatus, Leeia oryzae, Streptomyces reticuliscabiei, Asticcacaulis biprosthecium and Flavobacterium succinicans are enriched. Five additional species were identified with DESeq2 but not with Metacoder: Rubrivivax gelatinosus and Pelomonas puraquae in non-infected roots, Duganella nigrescens, Acidovorax delafieldii and Aeromonas caviae in infected roots. Enrichment of ESVs affiliated at species level (or extended at genus level if unassigned or uncultured at species level) are represented in Fig. 5C and given in the corresponding Table S4 (Supporting Information). In non-infected roots, 15 ESVs (over the 17 enriched) are affiliated at extended species level whereas 42 ESVs (over the 81 enriched) were found in infected roots. For instance, the most enriched in non-infected roots is Rubrivivax gelatinosus which is highly enriched (222.5 times, P < 0.001) but very rare in terms of relative abundance (0.02% with 226 counts in non-infected samples that are not significant with the Metacoder analysis). The least enriched but the most abundant in non-infected roots is Flavobacterium succinicans (21.11 times enriched, P < 0.01, 3.11% of relative abundance and 31,345 counts in non-infected samples). Moreover, there is another ESV assigned to Flavobacterium succinicans enriched in infected roots as well (23.28 times enriched, P < 0.01, 0.07% of relative abundance with 875 counts in infected samples). This species is actually the most abundant in both non-infected and infected roots but has a higher proportion in infected roots (Fig. S2, Supporting Information), hence a complementary aggregating tool such as Metacoder is required to visualize its type of enrichment. Only the ESVs with a full affiliation at species level that have been found enriched in the Metacoder trees and/or with the DESeq2 analysis are detailed in Table 2.
Summary on numbers of the enriched ESVs according to the DESeq2 analysis.
| Number of ESVs enriched in… | …Infected roots | …Non-infected roots |
|---|---|---|
| Total | 81 | 17 |
| With extended affiliation at species level (affiliation extended at genus level if unassigned or uncultured at species level, see Fig. 5C) | 42 | 15 |
| With full affiliation at species level… (see Table 2) | 11 | 4 |
| …found with the Metacoder analysis too | 8 | 2 |
| …not found with the Metacoder analysis | 3 | 2 |
| Number of ESVs enriched in… | …Infected roots | …Non-infected roots |
|---|---|---|
| Total | 81 | 17 |
| With extended affiliation at species level (affiliation extended at genus level if unassigned or uncultured at species level, see Fig. 5C) | 42 | 15 |
| With full affiliation at species level… (see Table 2) | 11 | 4 |
| …found with the Metacoder analysis too | 8 | 2 |
| …not found with the Metacoder analysis | 3 | 2 |
Summary on numbers of the enriched ESVs according to the DESeq2 analysis.
| Number of ESVs enriched in… | …Infected roots | …Non-infected roots |
|---|---|---|
| Total | 81 | 17 |
| With extended affiliation at species level (affiliation extended at genus level if unassigned or uncultured at species level, see Fig. 5C) | 42 | 15 |
| With full affiliation at species level… (see Table 2) | 11 | 4 |
| …found with the Metacoder analysis too | 8 | 2 |
| …not found with the Metacoder analysis | 3 | 2 |
| Number of ESVs enriched in… | …Infected roots | …Non-infected roots |
|---|---|---|
| Total | 81 | 17 |
| With extended affiliation at species level (affiliation extended at genus level if unassigned or uncultured at species level, see Fig. 5C) | 42 | 15 |
| With full affiliation at species level… (see Table 2) | 11 | 4 |
| …found with the Metacoder analysis too | 8 | 2 |
| …not found with the Metacoder analysis | 3 | 2 |
Compilation of information for ESVs with full affiliation at species level that are enriched (according to the DESeq2 and Metacoder analyzes) and identified as hub taxa (according to the SPIEC-EASI analysis).
| Assigned species | Enrichment according to DESeq2 | Enrichment according to Metacoder | Connectivity according to SPIEC-EASI | Proportion in enriched sample type | Relative abundance over all samples |
|---|---|---|---|---|---|
| Flavobacterium succinicans | −1.1 (ESV 1) and 3.3 (ESV 2) | −1.6 | Specific taxa hubs in both networks | 70% (ESV 1) and 88% (ESV 2) | 3.11% (ESV 1) and 0.07% (ESV 2) |
| Acidovorax delafieldii | 2.1 | Not enriched | Not hub | 76% | 1.12% |
| Asticcacaulis biprosthecium | 2.1 | 1.9 | Shared taxa hub in both networks | 78% | 0.79% |
| Aeromonas caviae | 1.8 | Not enriched | Not hub | 79% | 0.71% |
| Streptomyces reticuliscabiei | 2.8 | 2.6 | Not hub | 83% | 0.64% |
| Duganella nigrescens | 2.3 | Not enriched | Not hub | 55% | 0.25% |
| Chitinimonas taiwanensis | Not enriched | −1.6 | Not hub | 80% | 0.20% |
| Agrobacterium sullae | 4.2 | INF | Not hub | 93% | 0.18% |
| Pelomonas puraquae | −2.6 | Not enriched | Not hub | 87% | 0.17% |
| Streptomyces lanatus | 3.0 | 2.6 | Not hub | 86% | 0.16% |
| Silanimonas mangrovi | −2 | −2.3 | Shared taxa hub in infected network | 79% | 0.13% |
| Leeia oryzae | 2.9 | 1.8 | Not hub | 87% | 0.12% |
| Pleomorphomonas oryzae | 5.0 | INF | Not hub | 96% | 0.09% |
| Ensifer adhaerens | 3.2 | INF | Not hub | 88% | 0.08% |
| Desulfovibrio putealis | Not enriched | INF | Not hub | 95% | 0.04% |
| Stenotrophomonas maltophilia | Not enriched | −4.3 | Not hub | 73% | 0.03% |
| Rubrivivax gelatinosus | −22.5 | Not enriched | Not hub | 88% | 0.02% |
| Assigned species | Enrichment according to DESeq2 | Enrichment according to Metacoder | Connectivity according to SPIEC-EASI | Proportion in enriched sample type | Relative abundance over all samples |
|---|---|---|---|---|---|
| Flavobacterium succinicans | −1.1 (ESV 1) and 3.3 (ESV 2) | −1.6 | Specific taxa hubs in both networks | 70% (ESV 1) and 88% (ESV 2) | 3.11% (ESV 1) and 0.07% (ESV 2) |
| Acidovorax delafieldii | 2.1 | Not enriched | Not hub | 76% | 1.12% |
| Asticcacaulis biprosthecium | 2.1 | 1.9 | Shared taxa hub in both networks | 78% | 0.79% |
| Aeromonas caviae | 1.8 | Not enriched | Not hub | 79% | 0.71% |
| Streptomyces reticuliscabiei | 2.8 | 2.6 | Not hub | 83% | 0.64% |
| Duganella nigrescens | 2.3 | Not enriched | Not hub | 55% | 0.25% |
| Chitinimonas taiwanensis | Not enriched | −1.6 | Not hub | 80% | 0.20% |
| Agrobacterium sullae | 4.2 | INF | Not hub | 93% | 0.18% |
| Pelomonas puraquae | −2.6 | Not enriched | Not hub | 87% | 0.17% |
| Streptomyces lanatus | 3.0 | 2.6 | Not hub | 86% | 0.16% |
| Silanimonas mangrovi | −2 | −2.3 | Shared taxa hub in infected network | 79% | 0.13% |
| Leeia oryzae | 2.9 | 1.8 | Not hub | 87% | 0.12% |
| Pleomorphomonas oryzae | 5.0 | INF | Not hub | 96% | 0.09% |
| Ensifer adhaerens | 3.2 | INF | Not hub | 88% | 0.08% |
| Desulfovibrio putealis | Not enriched | INF | Not hub | 95% | 0.04% |
| Stenotrophomonas maltophilia | Not enriched | −4.3 | Not hub | 73% | 0.03% |
| Rubrivivax gelatinosus | −22.5 | Not enriched | Not hub | 88% | 0.02% |
Compilation of information for ESVs with full affiliation at species level that are enriched (according to the DESeq2 and Metacoder analyzes) and identified as hub taxa (according to the SPIEC-EASI analysis).
| Assigned species | Enrichment according to DESeq2 | Enrichment according to Metacoder | Connectivity according to SPIEC-EASI | Proportion in enriched sample type | Relative abundance over all samples |
|---|---|---|---|---|---|
| Flavobacterium succinicans | −1.1 (ESV 1) and 3.3 (ESV 2) | −1.6 | Specific taxa hubs in both networks | 70% (ESV 1) and 88% (ESV 2) | 3.11% (ESV 1) and 0.07% (ESV 2) |
| Acidovorax delafieldii | 2.1 | Not enriched | Not hub | 76% | 1.12% |
| Asticcacaulis biprosthecium | 2.1 | 1.9 | Shared taxa hub in both networks | 78% | 0.79% |
| Aeromonas caviae | 1.8 | Not enriched | Not hub | 79% | 0.71% |
| Streptomyces reticuliscabiei | 2.8 | 2.6 | Not hub | 83% | 0.64% |
| Duganella nigrescens | 2.3 | Not enriched | Not hub | 55% | 0.25% |
| Chitinimonas taiwanensis | Not enriched | −1.6 | Not hub | 80% | 0.20% |
| Agrobacterium sullae | 4.2 | INF | Not hub | 93% | 0.18% |
| Pelomonas puraquae | −2.6 | Not enriched | Not hub | 87% | 0.17% |
| Streptomyces lanatus | 3.0 | 2.6 | Not hub | 86% | 0.16% |
| Silanimonas mangrovi | −2 | −2.3 | Shared taxa hub in infected network | 79% | 0.13% |
| Leeia oryzae | 2.9 | 1.8 | Not hub | 87% | 0.12% |
| Pleomorphomonas oryzae | 5.0 | INF | Not hub | 96% | 0.09% |
| Ensifer adhaerens | 3.2 | INF | Not hub | 88% | 0.08% |
| Desulfovibrio putealis | Not enriched | INF | Not hub | 95% | 0.04% |
| Stenotrophomonas maltophilia | Not enriched | −4.3 | Not hub | 73% | 0.03% |
| Rubrivivax gelatinosus | −22.5 | Not enriched | Not hub | 88% | 0.02% |
| Assigned species | Enrichment according to DESeq2 | Enrichment according to Metacoder | Connectivity according to SPIEC-EASI | Proportion in enriched sample type | Relative abundance over all samples |
|---|---|---|---|---|---|
| Flavobacterium succinicans | −1.1 (ESV 1) and 3.3 (ESV 2) | −1.6 | Specific taxa hubs in both networks | 70% (ESV 1) and 88% (ESV 2) | 3.11% (ESV 1) and 0.07% (ESV 2) |
| Acidovorax delafieldii | 2.1 | Not enriched | Not hub | 76% | 1.12% |
| Asticcacaulis biprosthecium | 2.1 | 1.9 | Shared taxa hub in both networks | 78% | 0.79% |
| Aeromonas caviae | 1.8 | Not enriched | Not hub | 79% | 0.71% |
| Streptomyces reticuliscabiei | 2.8 | 2.6 | Not hub | 83% | 0.64% |
| Duganella nigrescens | 2.3 | Not enriched | Not hub | 55% | 0.25% |
| Chitinimonas taiwanensis | Not enriched | −1.6 | Not hub | 80% | 0.20% |
| Agrobacterium sullae | 4.2 | INF | Not hub | 93% | 0.18% |
| Pelomonas puraquae | −2.6 | Not enriched | Not hub | 87% | 0.17% |
| Streptomyces lanatus | 3.0 | 2.6 | Not hub | 86% | 0.16% |
| Silanimonas mangrovi | −2 | −2.3 | Shared taxa hub in infected network | 79% | 0.13% |
| Leeia oryzae | 2.9 | 1.8 | Not hub | 87% | 0.12% |
| Pleomorphomonas oryzae | 5.0 | INF | Not hub | 96% | 0.09% |
| Ensifer adhaerens | 3.2 | INF | Not hub | 88% | 0.08% |
| Desulfovibrio putealis | Not enriched | INF | Not hub | 95% | 0.04% |
| Stenotrophomonas maltophilia | Not enriched | −4.3 | Not hub | 73% | 0.03% |
| Rubrivivax gelatinosus | −22.5 | Not enriched | Not hub | 88% | 0.02% |
The co-occurrence network of taxa is more complex and specific in roots infected by M. graminicola
Co-occurrence networks were constructed in order to visualize and analyze the impact of the infection on the microbiome structure and taxa prevalence, and to identify hub taxa. The networks computed with the SPIEC-EASI package are different between non-infected and infected roots, in terms of node number, connectivity and specificity (Fig. 6). The community network of non-infected roots (Fig. 6A) is composed of 180 ESVs with 260 links in total (174 positive and 86 negative) whereas the network of infected roots (Fig. 6B) is composed of 260 taxa with 616 links (424 positive and 192 negative). The bacterial network of infected roots is denser (260 taxa versus 180 taxa), which is consistent with the higher species richness described earlier, and it is also more connected (degree distribution of 3.14 in non-infected versus 5.02 in infected network). This suggests potentially more interactions and a higher stability in the infected community. In both networks, positive links ratio is similar (174/260 = 0.67 in non-infected network whereas 424/616 = 0.69 in infected network) meaning that the types of predicted interactions seem unaffected in the overall community. Furthermore, in the non-infected network, the majority of ESVs are shared with the infected network, with only 27% specific taxa (48/180), while 49% of the taxa (128/260) are specific of the infected network. This suggests a higher specificity of the bacterial network of roots infected by M. graminicola.
Co-occurrence networks of taxa in non-infected (A), and in infected roots (B). Each taxon is represented by a node. The green lines between nodes are positive links (positive co-occurrence) and the red lines are negative links (negative co-occurrence). Identified ‘hub’ taxa are represented with triangles. Non-infected-specific taxa are in blue in A and infected-specific taxa are in bright-red in B. The other shared taxa are in grey. Only the ESVs present in 80% of the samples are included in these networks.
We identified hub taxa according to the betweenness centrality, closeness centrality and node degree value of the ESVs included in the networks (Supplementary Fig. S3). We assumed that hubs are highly connected taxa based on these three features, so we selected the 5% most connected to have a few numbers of taxa. Hence, we identified 18 hub taxa in the non-infected network, and 23 hub taxa in the infected network (Table S5, Supporting Information). In the non-infected network (Fig. 6A), only four hub taxa (22%) are specific to the non-infected condition whereas in the infected network (Fig. 6B), 11 hubs (48%) are specific to the infected condition. Moreover, 12 hubs (66%) in the non-infected network are shared taxa between both networks and 10 hubs (43%) in the infected network are shared taxa. All these above-mentioned characteristics about potentially important taxa for the structure and the composition of the network suggest a higher specificity in the overall structure of the infected bacterial community.
DISCUSSION
Our characterization of the root-associated bacterial communities (combined endosphere and rhizoplane) of non-infected and infected rice root tips in three highly infested paddy fields in Northern Vietnam shows that the RKN infection leads to a large restructuration of the microbiome. There are indeed two distinct structures of the bacterial community based on ESVs taxonomy and abundances, with differences in diversity including a higher richness and evenness in infected roots, and different phylogenetic composition with a strong signature of some taxa enriched or depleted in either non-infected or infected roots. This shift in the gall's microbiome, for which we propose the name ‘gallobiome’, suggests different interactions in these infected root tissues. Computational analysis performed in this study indeed showed a denser, more connected and more specific community network in the case of the infection.
A signature in the microbial diversity that characterizes the gallobiome
We observed an increase in bacterial diversity in the context of nematode infection. Not only the species richness increases but also the evenness, meaning that the taxa abundances are more equal and that there are fewer rare taxa in infected roots. It is indeed clearly visible on the Metacoder tree (Fig. 4) where there are diverse and numerous enriched taxa in infected roots, whereas in non-infected roots only two branches are enriched. Differences in the structure and diversity of the bacterial communities are explained by both the infection effect and the field effect independently.
Concerning the infection effect, such a shift in the microbiome leading to a microbiome restructuration is called dysbiosis and this phenomenon is under current broad investigations in human and other animals (DeGruttola et al. 2016). Imbalanced human microbiomes, compared to healthy microbiomes, can be associated to diseases according to many studies (Xuan et al. 2014; Casén et al. 2015). In plants, few studies describe a shift in the microbiome as a function of plant health or disease. Koskella et al. (2017) examined interactions between the bacterial pathogen Pseudomonas syringae pv aesculi (Pae), the leaf miner moth pest Cameraria ohridella, and the bark‐associated bacterial microbiome of the horse chestnut tree. They found a clear loss of diversity and associated shift in the microbiome composition of trees as a signature of disease. In our study, we found on the contrary an increase of species richness in infected roots. It could mean that the infection is not only associated to a change of the microbiome as a dysbiosis, but it is also associated to a change of the ecological niche because of root morphological and physiological modifications induced by the nematode. As sedentary biotrophic parasites, RKNs modify the plant's metabolism in order to complete their life cycle (Trudgill and Blok 2001). Therefore, the gall is a nutrient rich environment from which a large diversity of bacteria seems also to benefit. If so, the microbial restructuration would be an indirect consequence of the infection. However, the microbial restructuration that we observe could be a temporal state of the plant health status. To know if the shift in the microbiome happens before (if so it would be a cause of the infection) or after the infection, a dynamic sampling would be required. This is what has been done by Lebreton et al. (2019) with the parasite Plasmodiophora brassicae on the cabbage Brassica rapa. They analyzed the bacterial and fungal communities during the infection in both roots and rhizosphere. They observed a drastic shift of the fungal community from healthy plants between the last two sampling dates, especially in plant roots. The shift of the bacterial community in our study would fit with these observations, knowing that the plantlets were highly infected in the fields in Vietnam. Thus, the microbiome state described here would characterize a relatively late stage of the infection, a snapshot of well-formed galls on highly infected plants.
Concerning the field effect, as the physicochemical properties of the fields are slightly different, we expected different ecological niches and consequently different microbiomes in the three fields. We indeed observed different microbiome structures and diversity based on ESVs and KOs between the fields. According to a multifactorial analysis (Shakya et al. 2013), soil properties can be responsible for 9.1% of the variances in beta-diversity (pairwise UniFrac distances). These authors confirmed after Lauber et al. (2009) that among all soil factors, pH has the largest effect on the bacterial rhizosphere communities. We indeed found in our study that pH is the measured environmental factor that affects the most the structure of the bacterial community (Fig. 3).
As just mentioned above, we found that the alpha diversity of ESVs increases in the presence of M. graminicola. That is consistent with the simple explanation that the nematode carries its own microbiome inside the gall and that the taxa enriched in infected roots are part of the nematode's microbiome. M. graminicola’s microbiome has not been studied, but other nematode's microbiomes have been published. The closest to date is the one of the other RKN M. hapla collected from different soils in Germany (Topalović et al. 2019). Although the microbiome depends on host as well as many other factors (Shakya et al. 2013; Hacquard et al. 2015), this study confirmed that only a few microorganisms (14 strains) are able to attach to the nematode's cuticle.
Little is known about the internal microbiome of RKNs, but these obligatory plant parasites ingest plant solutes through a stylet whose diameter of 340–510 nm limits the entry of microorganisms (Hussey and Mims 1991). Other plant parasitic nematodes may nevertheless carry endosymbionts (Haegeman et al. 2009) but to date none have been isolated from the RKNs (Brown 2018). In other words, even if our data include M. graminicola’s microbiome in the infected roots, it could explain only a small part of the increased richness (+157 ESVs i.e. +24%). Besides, specific taxa are also revealed in non-infected roots suggesting indirect reasons for the modifications of the root microbiome. We propose that the shift in the bacterial composition is mainly due to changes in the plant physiology and morphology caused by infection that benefited to opportunistic bacteria from the surrounding soil, and slightly due to the colonization of the nematode microbiome. The rhizosphere effect, or so-called gall effect in our case, could be responsible for the shift in the gallobiome via root exudation (Sasse, Martinoia and Northen 2018) especially at the root tip where flux of primary metabolites are mostly located (Canarini et al. 2019).
Identification of enriched and highly connected bacterial taxa in the roots
Back, Haydock and Jenkinson (2002) described the mechanisms by which PPNs and soil-borne pathogens can work together to infect a plant. These mechanisms imply nematode-induced wounds, nematodes-induced physiological changes to the host plant (giant cell for example), reduction of host resistance and, as shown earlier, modifications within the rhizosphere (microbial substrate preference or rhizosphere effect). Interactions including competition (e.g. antagonism) and mutualism (e.g. syntrophy) play an important role in these modifications. In our study, the complex structure and the higher number of bacterial taxa (+44%) and co-occurrence evidences (+136%) in the infected network suggests more interactions than in the non-infected network. Another study conducted by Carrión et al. (2019) also showed an increased complexity in the co-occurrence network when facing a pathogen invasion of Rhizoctonia solani inoculated on sugar beet plants. Many enriched species in infected roots by M. graminicola are known to evolve mechanisms that allow them to grow and survive in highly competitive environments like soil and rhizosphere. For instance, among the enriched species in infected roots (Table 2), Ensifer adhaerens is a predator of Gram negative bacteria (Casida 1982). It is able to attach to other bacteria and to cause their lysis. It has already been described as an endophyte of rice roots (Xiaoxia et al. 2010) and, interestingly, as an occupant of Fabaceae nodules (Rogel et al. 2001) which is also a nutrient-rich environment. About Duganella nigrescens, another enriched species in infected roots, little is known but it is closely related to Duganella violacienigra, a rice endophyte (Sun et al. 2008). This later is known to produce violacein, a blue-purple secondary metabolite that has numerous biological activities involved in competitive interactions, including antibacterial, antiviral, antitrypanocydal, antiprotozoan, and antitumor effects (Ballestriero et al. 2014). Thus, some bacterial taxa are enriched at the infected root tips potentially for different reasons that may be involved in plant defense. For example, a study conducted by Berendsen et al. (2018) showed that the infection of Arabidopsis plants with a biotrophic pathogen can promote growth of a specific microbiota in the rhizosphere to aid in their defense.
Flavobacterium succinicans is the most abundant species in all roots and is slightly enriched in non-infected roots according to Metacoder analysis. It has been described as a freshwater commensal and may possess opportunistic pathogenic responses according to Bernardet and Bowman (2006) that is consistent with the state of the plants exposed to water in the field during the sampling and to the nematode infection. In Tian, Cao and Zhang (2015), Flavobacteriales have been found enriched in tomato roots that are infected by the RKN M. incognita. To know about their functional role in the community, they identified a vast range of CAZymes mainly involved in oligosaccharide degradation or simple sugar utilization, suggesting that these bacteria might be involved in carbohydrate metabolism. In our study, metabolic pathways related to carbohydrate degradation are found enriched in non-infected roots (e.g. sucrose degradation). Moreover, both enriched ESVs affiliated to F. succinicans are identified as hubs in their respective networks, meaning that this bacterium may play a key role in the community structure regardless of the infection. It would be interesting to check experimentally in a synthetic community (SynCom) experiment if it is indeed a keystone taxon, i.e. a highly connected taxa that individually or in a guild exerts a considerable influence on microbiome structure and functioning irrespective of its abundance across space and time (Banerjee, Schlaeppi and Van Der Heijden 2018).
Stenotrophomonas maltophilia, the second most enriched species in non-infected roots, has been identified as a rice root endophyte in a field in China (Zhu et al. 2012). Many strains of this species can produce antibiotics that protect plants and compounds that can promote plant growth. The strain S. maltophilia R3089 for example, can produce an antifungal compound named maltophilin. Although it has been found inactive against bacteria (Jakobi et al. 1996), it can play an important role within the overall microbiome. More importantly for this study, another strain, S. maltophilia G2, isolated from soil in China was found to have a high nematotoxic activity against a free-living nematode (Panagrellus redivivus) and a plant-parasitic nematode (Bursaphelenchus xylophilus) (Huang et al. 2009). It acts via a serine protease degrading the nematode cuticle and kills nematodes. In our study, S. maltophilia is present only in non-infected roots and it would be interesting to test under controlled conditions its potential role in preventing the establishment of M. graminicola in rice roots. If such bacteria would exhibit beneficial effects on the phytobiome, it could be an interesting candidate for biocontrol solutions.
CONCLUSION
In this study, we aimed at improving our understanding on the impact of the rice RKN infection on the root microbiome by describing the two bacterial communities of non-infected roots and infected roots (with galls of M. graminicola). We clearly observed a specific signature of the gallobiome with shifting taxa abundances and specific ESVs found in the two community networks that was also associated with a restructuration of the microbiome. It would be interesting to know if these deep taxonomic and structural modifications of the microbiome are also associated with a change in their functional capabilities. PICRUSt2 is a prediction tool commonly used for functional abundances and metagenome content and is only based on marker gene sequences such as the 16S rRNA gene, which was used in this study. It is important to note that this analysis concerning the environmental context of gall with this tool (see Sup. Discussion) remains potentially speculative, as to date our knowledge of environmental genomes is fragmentary and limits predictions (Sun, Jones and Fodor 2020). Moreover, efforts to link important taxa to putative functions can be pointless without experimental validations, and such validations can be difficult to obtain from a complex community influenced by many environmental factors. However, in the perspective to limit M. graminicola infection by an integrated pest management, our study could help to select candidate bacteria for biocontrol. For example, Actinobacteria are enriched in infected roots and are known for antibiotic production, degradation of complex polysaccharides or as plant pathogens. Some strains are able to release lytic enzymes and secondary metabolites, and some other have nematicidal activity (Xu et al. 2011). Because of these properties and their ability to colonize and survive in the gall environment, they could be used as biocontrol agents, particularly against PPNs. An ideal candidate would carry nematicidal activity and would be able to survive in the same environment than the nematode. Thus, the enrichment of predicted functions specialized for bacterial survival in the gall can be an important criterion for selection. One perspective would be to match our results with a cultivable approach in order to select such bacteria and to validate them as diagnostic tools of the plant health status or as biocontrol agents.
ACKNOWLEDGEMENTS
The authors acknowledge the farmers in Hải Dương where the samples have been collected and the employees at the Vietnam National University of Agriculture (VNUA) where the soil analysis has been performed. They also acknowledge the IRD iTrop HPC (South Green Platform) at IRD Montpellier for providing HPC resources that have contributed to the research results reported within this paper. URL: https://bioinfo.ird.fr/- http://www.southgreen.fr.
FUNDING
This work was supported by the Consultative Group for International Agricultural Research (CGIAR) Program on Rice Agri-food Systems (RICE) and the ‘Mission Longue Durée’ (MLD) fellowship program of Research Institute for Development (IRD). Anne-Sophie Masson was supported by a fellowship from the French Ministry of Higher Education, Research and Innovation.
Conflicts of interest
None declared.
