The impact of the rice production system (irrigated vs lowland) on root-associated microbiome from farmer’s fields in western Burkina Faso

As a consequence of its potential applications for food safety, there is a growing interest in rice root-associated microbial communities, but some systems remain understudied. Here, we compare the assemblage of root-associated microbiota in rice sampled in 19 small farmer’s fields from irrigated and rainfed lowlands in western Burkina Faso, using an amplicon metabarcoding approach 16S (Prokaryotes, three plant sample per field) and ITS (fungi, one sample per field). In addition to the expected structure according to the root compartment (root vs. rhizosphere) and geographical zones, we show that the rice production system is a major driver of microbiome structure, both for prokaryotes and fungi. In irrigated systems, we found a higher diversity of prokaryotic communities from rhizosphere and more complex co-occurrence networks, compared to rainfed lowlands. Core taxa were different between the two systems, and indicator species were identified: mostly within Bacillaceae and Bradyrhizobiaceae families in rainfed lowlands, and within Burkholderiaceae and Moraxellaceae in irrigated areas. Finally, phylotypes assigned to putative phytobeneficial and pathogen species were found. Mycorrhizal fungi Glomeromycetes abundance was higher in rainfed lowlands. Our results highlight deep microbiome differences induced by contrasted rice production systems that should consequently be considered for potential microbial engineering applications.


41
Soil and rhizosphere host megadiverse and dynamic communities of microorganisms that are 42 crucial to the plants they associate with. Their role is particularly recognized for crops as the 43 below-ground microbiota supply plants with nutrients and provide protection against 44 pathogens Chialva et al. 2021). Recent research suggests that root-45 associated microbes can improve plant tolerance to environmental stressors (Chialva et al. 46 2020), and modify phenology (Lu et al. 2018) and morphological traits (Senthil Kumar et al. 47 a rapid (15s) rinse and then stored in another 50 mL sterile PBS-containing tube. We placed 165 the tubes in a cooler and then at 4°C when back to the laboratory, on the same day. 166 Soil physicochemical properties: data acquisition and analysis 167 The three geographical zones studied (Fig. 1a)  Soil sampling was performed on the same day as rice root sampling, and in three locations 171 nearby sampled plants, using a 10 cm depth auger. Back from the field, sampled soil was 172 dried in the shade at room temperature and stored until analysis. 173 INERA/GRN service performed the analyses of soil samples according to a standardized 174 methodology. Briefly, soil physical properties were assessed by soil particle size distribution 175 following Bouyoucos (1962), pH was estimated according to AFNOR (1981), total organic 176 carbon with the Walkley & Black (1934)'s method, total concentrations of nitrogen with 177 Kjeldhal method Hillebrand et al. (1953), and finally phosphorus and potassium content as 178 described respectively in Novozansky et al. (1983) and Walinga et al. (1989). Then, cation 179 exchange capacity (CEC), a measure of fertility, nutrient retention capacity, and the capacity 180 to protect groundwater from cation contamination, was estimated, as well as sorptive 181 bioaccessibility extraction (SBE), that relates to the environmental mobility, partitioning and 182 toxicity of soil pollutants, following Metson (1956). Soil data are publicly available on the 183 IRD Dataverse : https://doi.org/10.23708/ LZ8A5B. 184 Rice root conditioning, DNA extraction and sequencing 185 Less than 24h after sampling, root samples (rice roots including rhizosphere) stored at 4°C 186 were processed. In order to separate the different root compartments, the tubes were vortexed 187 vigorously one minute and then, roots were removed from the PBS solution using sterile 188 forceps. The remaining PBS solution was considered as the 'rhizosphere' compartment. Roots 189 were then surface-sterilized with 70% alcohol (30s), 1% bleach (30s) and finally rinsed three 190 times in sterile water. We considered these surface-sterilized roots as the 'root ' compartment, 191 which comprises both endosphere microorganisms as well as persistent DNA from the We specifically focused on the factors structuring prokaryotic communities (16S sequencing), 204 but were also interested to see whether similar tendencies hold for fungal communities (ITS 205 sequencing). We consequently chose the following approach: sequencing was performed for 206 each sampled plant (19 fields * 3 plants = 57 samples per compartment) for 16S sequencing 207 and for a composite sample (3 plant samples were pooled to result in one sample per field, so 208 that the total number of samples per compartment is 19) for the ITS marker. Negative controls 209 (three for 16S and one for ITS) were sequenced to remove potential contaminants. 210 Sequence data are retrievable from NCBI (National Center for Biotechnology Information) 211 under the Bioproject ID: PRJNA763095. 212 Bioinformatic analyses of obtained sequences 213 All bioinformatics and statistical analyses were performed in R software v 3.6.3 (R core 214 Team, 2018) and the package ggplot2 was used for the visualization. 215 Raw sequences were processed using a custom script from the dada2 pipeline, which is 216 designed to resolve exact biological sequences (ASVs for Amplicon Sequence Variants) from 217 Illumina sequence data without sequence clustering (Callahan et al. 2016). Raw sequences 218 were first demultiplexed by comparing index reads with a key, and paired sequences were 219 trimmed. Sequences were dereplicated, and the unique sequence pairs were denoised using 220 the dada function. Primers and adapters were screened and removed using a custom script 221 with cutadapt (Martin, 2011 Members of the core microbiota were identified for 16S and ITS communities (including both 237 rhizosphere and roots compartments) in each rice grown system using the prevalence 238 threshold of 60%. 239 For 16S dataset only, we inferred co-occurrence networks using the SpiecEasi pipeline (Kurtz 240 et al. 2015), independently for each rice growing system: rainfed lowland and irrigated areas. 241 Networks were calculated for ASVs present in more than 15% of the samples using the 242 method 'mb' and setting the lambda.min.ratio to 1e-3 and nlambda to 50. We identified hub 243 taxa, i.e. the potential keystone of the microbial network belonging to the most connected 244 ASVs, based on their node parameters (method developed by Berry & Widder, 2014): a low 245 betweenness centrality (lower quantile, < 0.9), and a high closeness centrality (higher 246 quantile, > 0.75), transitivity (higher quantile, > 0.25) and degree (higher quantile, > 0.75). 247 The node and network parameters were determined using the R package igraph (Csardi & 248 Nepusz, 2006) and qgraph (Epskamp et al. 2012). Complete networks were further described 249 by calculating the number of nodes and hubs, the network mean degree, mean closeness and 250 betweenness centralities, the total number of edges, and the positive to negative edges ratio. 251 Core taxa (prevalent in more than 60% of the samples) were also identified for each system. 252 The taxonomic affiliation of all ASVs was refined using nucleotide basic local alignment 253 search tool (BLASTn) analyses on NCBI nr database. We screened the table of blast best hits of all 16S and ITS ASVs in order to search the genus or species names of a number of 255 pathogen species listed a priori (Table S6), based on the reference book Compendium of Rice 256 Diseases (Cartwright et al. 2018). We also searched within blast best hits for ITS ASVs 257 assigned to the Glomeromycetes class, as well as 16S and ITS ASV assigned to a species 258 name's including "oryz", as this may correspond to a specific interaction (pathogen or 259 beneficial) with rice. 260 Statistical analyses 261 We first analyzed soil physico-chemical parameters. To this purpose, PERMANOVA were 262 performed on soil physical properties (texture, i.e. relative amount of sand, silt, and clay) on 263 the one hand, and soil chemical properties (7 variables: pH, total carbon, total nitrogen, total 264 phosphorus, total potassium, SBE, CEC) on the other hand. For both models, we included as 265 explanatory factors the 'geographical zone' and 'rice-growing system', as well as their 266 interaction, using adonis2 function from the vegan R package (Oksanen et al. 2007 The effect of edaphic variables (i.e., pH, total organic carbon, total phosphorus, total nitrogen, 279 total potassium, CEC and SBE) in structuring β -diversity was tested using the envfit function 280 (9999 permutations) in R package vegan. Structuring soil properties were thus fitted onto the 281 ordination space. 282 We tested for an effect of the rice growing system on obtained indices of alpha diversity 283 (Shannon diversity index and observed richness). To this purpose, we performed non-284 parametric statistical tests (kruskal_test function from the library rstatix) independently for each kingdom (prokaryotes and fungi respectively) and each compartment (rhizosphere and 286 root associated). In addition, for 16S data only (because of insufficient sample size for ITS), 287 we also tested for an effect of the specific site, using Kruskal-Wallis test, and then performed 288 posthoc tests using dunn_test function. 289 Then, we identified particular soil taxa that were associated with lowlands or irrigated

305
Structure of rice soil properties and rice root associated microbial communities 306 PERMANOVA analysis performed on the Bray-Curtis distance matrix of soil characteristics 307 to describe the overall soil properties, highlighted no significant influence of the rice growing 308 system, but a differentiation according to the geographical zone, for both physical (F = 6.420, 309 r 2 = 0.448, p = 0.006, Fig. 1b) and chemical (F = 4.121, r 2 = 0.346, p = 0.026) soil parameters. 310 More precisely, we found no effect of the rice growing system but a significant effect of the 311 geographical zone on the clay and sand contents, as well as total Phosphorus, total Potassium, SBE and CEC (Table S1). Posthoc tests revealed that the geographic zones that differed 313 statistically were the same for the six above-mentioned variables, namely Banzon and 314 Karfiguela zones (Fig. 1c). 315 To determine whether the geographical zones (Karfiguela, Bama or Banzon), the rice-growing 316 systems (irrigated vs. rainfed lowlands), or their interactions, structured root-associated or 317 rhizosphere microbial communities, we performed PERMANOVA analysis on the Bray-318 Curtis distance matrix of 16S and ITS ASVs, respectively (Table S2, Fig. S2). As different 319 structures were revealed between root and rhizosphere communities of both prokaryotes (F= 320 6.863, r 2 = 0.052, p < 0.001) and fungi (F= 3.753, r 2 = 0.080, p < 0.001), we further subsetted 321 both datasets to observe the relative influence of the rice-growing systems and the 322 geographical zones in shaping root and rhizosphere communities separately (Table 1, Fig. 2). 323 In the rhizosphere, prokaryotic communities were mainly structured by the rice growing 324 system (F=5.096, r 2 = 0.079, p < 0.001, Table 1), the geographical zone (F=2.118, r 2 = 0.069, 325 p < 0.001) and the interaction between rice growing system and zone (F=1.877, r 2 = 0.058, p 326 < 0.001). Posthoc tests revealed that most of the pairs of sites (i.e., irrigated and lowland from 327 the same geographical zone) were significantly different, but interestingly revealed no 328 significant difference between communities originating from irrigated systems, whereas all 329 communities from rainfed lowland sites exhibited distinct structures (Table S3). Fungal 330 communities of the rhizosphere were also mostly structured by the rice growing system 331 (F=2.452, r 2 = 0.115, p < 0.001, Table 1). The geographical zone (F=1.555, r 2 =0.146, p = 332 0.007) and the interaction between rice growing system and the geographical zone (F=1.335, 333 r 2 = 0.126, p = 0.036) were also driving the rhizosphere fungal microbiome. The low number 334 of samples did not allow to detect, if any, statistically significant differences between sites in 335 communities' structures for ITS (Table S3). 336 As observed for rhizosphere communities, the root-associated prokaryotic communities were 337 mainly shaped by the rice growing system (F=5.155, r 2 = 0.079, p < 0.001, Table 1), the 338 interaction between rice growing system and the geographical zone (F=2.451, r 2 = 0.075, p = 339 0.002) and the geographical zone (F=2.247, r 2 = 0.069, p < 0.001). Most of the pairs of sites 340 (from the same geographical zone) were significantly different. No significant difference was 341 detected between communities originating from irrigated systems, whereas all communities 342 from rainfed lowland sites exhibited distinct structures (Table S3). Root-associated fungal 343 communities were also mostly influenced by the rice growing system (F=2.289, r 2 = 0.111, p = 0.002, Table 1), and by the interaction between rice growing system and the geographical 345 zone (F=1.457, r 2 =0.141, p = 0.013). The effect of the geographical zone on root-associated 346 fungal communities was not evidenced (F=1.206, r 2 = 0.117, p =0.123). As for rhizosphere, 347 posthoc tests on root-associated fungal communities were all non-significant (Table S3). 348 The influence of soil chemical parameters on microbial community structure is reported in 349  Table S4. We noticed that the prokaryotic communities of both 350 rhizosphere and roots were affected by the same three parameters: SBE (r 2 =0,482, p < 0,001 351 for rhizosphere, and r 2 = 0,175, p = 0,006 for roots), CEC (r 2 =0,314, p < 0,001 for 352 rhizosphere, and r 2 = 0,204, p = 0,003 for roots) and total phosphorus (r 2 = 0,132, p = 0,023 353 for rhizosphere, and r 2 = 0,179, p = 0,004 for roots). For fungi, although various parameters 354 were marginally significant in each compartment (Table S4), we only detected a significant 355 effect of total nitrogen on rhizosphere communities (r 2 =0,320, p = 0,043). 356 Composition of rice root microbiomes and comparison of alpha-diversity 357 While 16S data were assigned at the genus level for 64% of ASVs, only 34% of ITS ASVs 358 could be assigned (see assignations at the phyla level in Fig. S4). Assignations at the phylum 359 level were obtained for all (100%) 16S ASVs, but only for 62% for ITS ASVs (see Fig. S4). 360 Assigned prokaryotic taxa represent 17 phyla, most abundant ones being Proteobateria, 361 Firmicutes, Mixoccoccota and Acidobacteriota. For ITS, seven phyla were found, with the 362 most abundant ones being Ascomycota followed by Basicomycota. 363 We tested the effect of the rice growing system on the diversity indices (alpha-diversity). No 364 effect could be evidenced on the fungal (Shannon: H = 0.026, p =0.87 for the rhizosphere, and 365 H = 0.107, p =0.74 for root-associated communities), or root-associated prokaryote diversities 366 (Shannon: H = 1.05 p = 0.306). On the other hand, the rice growing system had a significant 367 effect on the prokaryote diversity of the rhizosphere (Shannon: H = 11.6, p <0.001), with a 368 higher Shannon diversity index in irrigated areas (5.03 ± 0.13), compared to rainfed lowlands 369 (4.39 ± 0.12) and a higher observed richness (275.6 ± 26.4) in irrigated areas compared to 370 rainfed lowlands (143.7 ± 17.4) (Fig. 3). We noticed however an opposite pattern for fungal 371 communities of the rhizosphere with higher observed richness in rainfed lowlands, compared 372 to irrigated areas (Fig. 3). 373 This effect of the rice growing system on 16S rhizosphere data was also clearly observed 374 when plotting the diversity indices by site (Fig. S5). In addition, we found that the specific 375 site had an effect on the prokaryotic communities of the rhizosphere, and also, but to a lesser 376 extent, in roots ( Fig. S5 and Table S5). In the rhizosphere, the highest diversity was found in 377 the irrigated perimeter of Karfiguela, and to a lesser extent in the irrigated area of Bama. A 378 particularly low diversity was found in the rainfed lowland of Karfiguela zone, and to a lesser 379 extent in the rainfed lowland of Banzon. Conversely, we noticed a slightly higher diversity in 380 fungal root associated communities in the rainfed lowland of Karfiguela (Fig. S5). 381 Core microbiome and co-occurrence networks in the two rice-growing systems 382 ASVs belonging to the core microbiome of lowland vs. irrigated rice were respectively 383 identified with a prevalence threshold set to 60%. For 16S, we identified 26 core ASVs 384 associated with the irrigated systems, and two core ASVs in lowlands (Fig. 4). Among the 385 core taxa in irrigated areas, the vast majority of phylotypes (25/26) belonged to the 386 Burkholderiaceae family, with 24 assigned to Ralstonia pickettii and one to Paraburkholderia 387 kururiensis. One of the core ASVs is common to both irrigated area and rainfed lowlands 388 systems. Its best blast hit corresponds to Bradyrhizobium tropiciagri (Bradyrhizobiaceae) 389 with a 99.5% sequence similarity. Another core phylotype in rainfed lowlands is assigned to 390 the same species with 99.3% sequence similarity. 391 For ITS, we identified 5 core ASVs in the irrigated systems, compared to 11 core ASVs 392 associated with the lowlands, 4 of them being common to both rice growing systems (Fig. 4). 393 Then, we compared the prokaryotic co-occurrence networks in each rice growing system 394 respectively (Table 2). We identified 15 hub ASVs in the irrigated systems and 20 in rainfed 395 lowlands. We found a higher edge number in irrigated compared to rainfed lowlands: 1720 396 positive and 269 negative resulting in 2029 total edges in irrigated areas, while only 1163 397 positive and 85 negative resulting in 1248 total edges were found in rainfed lowlands. Finally, 398 the network computed from irrigated areas had higher connectivity compared to the one from 399 rainfed lowlands (9.8 vs 7.9 node mean degrees, respectively). 400 None of the identified hub taxa were also core in any of the two rice growing systems. Only 401 one ASVs was identified as a hub in both irrigated and rainfed lowland systems, assigned to 402 Enterobacter mori (Enterobacteriaceae). Hub taxa in irrigated areas (15 ASVs) were assigned to 8 different species from 5 families, while hub taxa in rainfed lowland (20 ASVs) only 404 corresponded to 4 species from 2 families ( Table 2). 405 Indicator taxa of the two rice growing systems 406 For 16S data, we found 128 indicator taxa in irrigated areas, including ASVs from eight 407 bacterial families, most of them assigned to Acinetobacter, Ralstonia, Aeromonas, 408 Comamonas, Clostridium and Enterobacter (Table 3). On the other hand, only 63 were 409 identified in rainfed lowlands, most of them within the Bacillaceae family, including ASVs 410 assigned to Exiguobacterium and Priestia, and Bradyrhizobiaceae family, genus 411 Bradyrhizobium (Table 3). The ASV assigned to Paraburkholderia kururiensis 412 (Burkholderiaceae) revealed as indicator in irrigated areas (Table 3) was also a core taxa in 413 irrigated areas. Also, among the 24 indicator ASVs in irrigated areas assigned to Ralstonia 414 pickettii (Burkholderiaceae), 16 were also core in irrigated areas. In addition, four indicator 415 ASVs in irrigated areas were also hubs in this system: two assigned to Aeromonas hydrophilai 416 (Aeromonadaceae), another assigned to Enterobacter cloacae (Enterobacteriaceae), and 417 finally one corresponding to Acinetobacter soli (Moraxellaceae). Three ASVs assigned to 418 Priestia flexa (Bacillaceae) were hubs in rainfed lowlands. 419 For ITS data, we found 16 indicator taxa in irrigated areas, and 27 in rainfed lowlands (Table  420 3). Indicator taxa in irrigated areas were assigned to seven classes: Agaricomycetes, 421 Chytridiomycetes, Dothideomycetes, Geoglossomycetes, Leotiomycetes, Sordariomycetes, and 422 Ustilaginomycetes. Indicator taxa in rainfed lowlands were assigned to nine classes: 423

Agaricomycetes, Chytridiomycetes, Dothideomycetes, Glomeromycetes, Saccharomycetes 424
Schizosaccharomycetes, Sordariomycetes, Tremellomycetes, Ustilaginomycetes. One ITS 425 ASV identified as indicator taxa in irrigated, with best hit Pulveroboletus sinensis 426 (Agaricomycetes), was also core in this rice growing system, and two indicator taxa in rainfed 427 lowlands were also core in this system: one assigned to Pseudobaeospora wipapatiae 428 Next, we made a subset of the ITS dataset for ASVs assigned to the Glomeromycetes class 437 (total of 14 ASVs). AMF summed abundance was affected by the compartment (χ 2 = 101.22, 438 p<0.001) and by the rice growing system (χ 2 = 951.12, p<0.001), with higher abundances in 439 rhizosphere compartment and in rainfed lowlands (Fig. 5). In addition, differential abundance 440 testing between rice growing systems detected an ASV assigned to Racocetra crispa as 441 preferentially found in rainfed lowlands (l2FC = 24.59; p<0.001). We also noticed that 442 another Glomeromycetes (Dentiscutata savannicola) was identified as indicator taxa in 443 rainfed lowland environments (Table 3). 444 We then screened the list of all assigned ASVs for a set of pathogen species defined a priori 445 (see the list in the Table S6). For Prokaryotes (16S data), a number of ASVs corresponded to 446 the genera of pathogens, but only Burkholderia glumae (two ASVs), Acidovorax avenae (four 447 ASVs) and Dickeya chrysanthemi (six ASVs) were identified at the species level. These 12 448 ASVs identified at the species level were however only found in one sample. The 449 Xanthomonas genus was found, but with no assignation to X. oryzae (instead, assignation to 450 X. theicola which is phylogenetically closed to the rice associated X. sontii (Bansal et al. 451 2020). A similar situation was observed for the genera Pseudomonas, Pantoea, and 452 Sphingomonas. The same analysis of putative pathogens for ITS revealed the presence of the 453 following ten genera: Alternaria, Bipolaris, Ceratobasidium, Curvularia, Fusarium, 454 Helminthosporium, Microdochium, Rhizoctonia, Sarocladium. We notice that one ASV 455 whose best blast hit was Curvularia chonburiensis was core in both irrigated and rainfed 456 lowlands (Fig. 4). 457

458
This study aimed at describing the rice root-associated microbiome by comparing contrasted 459 rice growing systems in farmer's fields in Burkina Faso. We found that the rice growing 460 system was a structuring factor for rice root-associated microbiomes, and that the diversity of 461 prokaryotic community from the rhizosphere was higher in irrigated areas compared to 462 rainfed lowland. In addition, we identified a number of phylotypes with potential key roles 463 (hub, core, indicators) in the two contrasted systems, as well as putative phytobeneficial and 464 pathogen species. Although the results on fungi (ITS region) must be taken with caution due 465 to a smaller sample size and the poor representation of obtained sequences in available 466 databases, this study shed light on some drivers of assemblage of rice root associated 467 microbial communities in a sparsely documented African system. 468 The structuring of microbial diversity is affected by the rice growing system We noticed however that organic fertilization remained rare, and its frequency was not 534 drastically affected by the rice growing system. Finally, transplantation was always performed 535 in irrigated areas, while direct sowing was the most common practice in rainfed lowlands. 536 On the other hand, paddy soils studied in western Burkina Faso (all over the six sites) are 537 particularly poor if compared for example to a study of more than 8 000 soils in Hunan 538 Province (Duan et al. 2020), where average organic carbon was 1.972%, compared to 0.922% 539 in our study, total nitrogen was 0.191%, higher than 0.072%, and total phosphorus was 540 0.71g.kg -1 , compared to 0.24g.kg -1 . The studies previously cited evidencing fertilization 541 effects, were performed in soils with higher carbon and nitrogen contents (see for example 542 Ullah et al. 2020, where minimum average organic carbon was 2% and total nitrogen 0.1%). 543 The effect of fertilization on microbial diversity may actually depend on various aspects, 544 including the soil type. Notably, a positive relationship was found between rice fertilization 545 and soil bacterial richness and diversity in a 19-years inorganic fertilization assay in a reddish 546 paddy soil in southern China (Huang et al. 2019); while Wang & Huang (2021) showed an 547 effect of the fertilization on paddy soils microbial community composition but no effect on 548 the diversity. In poor soil systems such as in this study, fertilization input may actually 549 increase microbial diversity. 550 Finally, a complementary hypothesis could be the higher fragmentation of rainfed lowlands 551 compared to irrigated areas. Indeed, irrigated sites correspond to larger areas cultivated in 552 rice, possibly with two rice seasons per year, so that rice fields are likely to be more 553 connected to each other than in rainfed lowlands. Higher connectivity generally leads to 554 higher biodiversity (Fletcher et al. 2016). The principles of metacommunity theory could also 555 be applied to micro-organisms, with reduction in host habitats and fragmentation potentially increasing extinction rates (Mony et al. 2020), but as our study misses an explicit 557 characterization of the rice landscape structure, this hypothesis could not be formally tested. 558 Distant rainfed lowlands differ more than distant irrigated perimeters 559 Our results showed that the prokaryotic communities in the rice rhizosphere and roots from 560 the three irrigated sites do not differ significantly from each other. On the other hand, the 561 same analysis revealed significant differentiation between the three rainfed lowland study 562 sites (in all three cases for rhizophere and two out of three comparisons in roots). Also, we 563 found very few core phylotypes in rainfed lowland, with only two core ASVs for 16S, what 564 reinforces the above-mentioned observation. These results are likely driven by a higher 565 heterogeneity between rainfed sites, in terms of water control, agricultural practices or rice 566 genotypes. 567 Indeed, in irrigated rice, the farmer has the potential to control irrigation water during the 568 whole growing season. On the other hand, irrigation in rainfed lowland is dependent on 569 precipitations that differ between the three geographical zone sampled within the rice growing 570 season. In addition, we showed a high heterogeneity of agricultural practices in rainfed 571 lowlands: for example, legume rotation was common in the rainfed lowland of Bama zone, 572 but rare or absent in the two other rainfed lowland sites, and organic fertilization was more 573 frequent in the rainfed lowland of Karfiguela zone, than in the other sites (Barro et al. 2021). 574 Finally, in terms of rice genetics, a high rice genetic differentiation was found between the 575 rainfed lowland site of Karfiguela zone and the five sites: a distinct genetic group O. sativa 576 Aus, and other distinct landraces were found in this peculiar site, compared to the five others 577 where only O. sativa indica was grown (Barro et al. 2021). These specificities of the rainfed 578 lowland from Karfiguela zone, in terms of rice grown and agricultural practices, may also 579 drive its specific patterns of alpha diversity, with a particularly low prokaryote diversity (in 580 rhizosphere and also, in a lesser extent, in roots), and a tendency for higher fungal diversity in 581 roots. 582 Our sampling size was much lower for ITS and this likely explains the absence of such a 583 pattern, with no significant differences obtained between pairs of sites. Alternatively, the 584 pattern may be different for fungal diversity, as suggested by the higher number of core taxa 585 in rainfed lowlands than in irrigated areas. 586 587 Identification of core microbiota and hub phylotypes 588 The prevalent taxa, indicator taxa and hubs may be considered as having an important 589 ecological role in microbiome assembly and ecosystem functions (Banerjee et al. 2018). In 590 this study, we identified the core prokaryote and fungal microbiota in both irrigated and 591 rainfed lowland environments. While four fungal taxa were found to be cores in both systems, 592 only one bacterial core taxa was shared between the two rice growing systems: assigned to 593 Bradyrhizobium tropiciagri, a nitrogen-fixing symbiont isolated from tropical forage legumes 594 We identified a few potentially beneficial taxa that could be investigated further. In particular, 629 AMFs of the class Glomeromycetes were found preferentially in rainfed lowlands, with one 630 ASV, Racocetra crispa, enriched in rainfed lowland system compared to irrigated areas, and 631 one ASV, Dentiscutata savannicola, identified as indicator in rainfed system. This was 632 expected considering the lower frequency of mineral fertilization in rainfed lowlands 633 compared to irrigated areas. Indeed, AMF colonization was shown to be affected by farming found ASVs assigned to ten genera comprising rice pathogens, including Bipolaris, causing foliar diseases, namely Pyricularia oryzae and Xanthomonas oryzae, known from 648 symptom observations to be present in the study sites (Barro et al. 2021), were not detected in 649 this root-associated metabarcoding data. Some of these putative fungal pathogens are 650 frequent, particularly one, whose best blast hit is Curvularia chonburiensis, identified as core 651 taxa in both irrigated perimeters and rainfed lowlands. Various Curvularia species are known 652 to be pathogenic in rice, potentially causing contrasted symptoms (Gao et  land (in red).
for each rice s obtained for ot associated Figure 4 Venn diagram representing the core sequence variants for each rice growing system : irrigated vs rainfed lowlands. A. For prokaryotes B. For fungi Figure 5 Spatial repartition of summed abundance of 14 ITS ASVs assigned to the class Glomeromycetes in the two rice root microbiome compartments (rhizosphere and roots), and in each rice growing system (irrigated perimeters vs rainfed lowlands).
The shape of the point corresponds to each geographic zone.   a. Analysis based on 16S rRNA gene reflecting Prokaryote communities. One point corresponds to one plant. b. Analysis based on ITS reflecting fungal communities. One point corresponds to one field. Figure S4: Prokaryote (16S) and fungi (ITS) taxonomic diversity obtained for each study site and each compartment Figure S5: Comparison of root-associated microbiota α -diversity in the six study sites: for Prokaryotes (16S data) on the left and for data) on the right.
Data obtained for the rhizosphere compartment are presented on top and roots data are on the bottom of the figure. The study sites fro areas are represented in blue, while the ones from rainfed lowlands appears in red.
for fungi (ITS from irrigated

Table S1
Non-parametric tests (Wilcoxon tests) on the soil physico-chemical parameters Rice      Results of the statistical analyses testing for the effect of soil chemical parameters on microbiome communities (each compartment analyzed separately).