Molecular Characterization of Arbuscular Mycorrhizal Fungi in an Agroforestry System Reveals the Predominance of Funneliformis spp. Associated with Colocasia esculenta and Pterocarpus officinalis Adult Trees and Seedlings

Pterocarpus officinalis (Jacq.) is a leguminous forestry tree species endemic to Caribbean swamp forests. In Guadeloupe, smallholder farmers traditionally cultivate flooded taro (Colocasia esculenta) cultures under the canopy of P. officinalis stands. The role of arbuscular mycorrhizal (AM) fungi in the sustainability of this traditional agroforestry system has been suggested but the composition and distribution of AM fungi colonizing the leguminous tree and/or taro are poorly characterized. An in-depth characterization of root-associated AM fungal communities from P. officinalis adult trees and seedlings and taro cultures, sampled in two localities of Guadeloupe, was performed by pyrosequencing (GS FLX+) of partial 18S rRNA gene. The AM fungal community was composed of 215 operational taxonomic units (OTUs), belonging to eight fungal families dominated by Glomeraceae, Acaulosporaceae, and Gigasporaceae. Results revealed a low AM fungal community membership between P. officinalis and C. esculenta. However, certain AM fungal community taxa (10% of total community) overlapped between P. officinalis and C. esculenta, notably predominant Funneliformis OTUs. These findings provide new perspectives in deciphering the significance of Funneliformis in nutrient exchange between P. officinalis and C. esculenta by forming a potential mycorrhizal network.


INTRODUCTION
Pterocarpus officinalis L. is one of the dominant wetland tree species of the seasonally flooded swamp forests in the Caribbean and the Guiana regions (Eusse and Aide, 1999;Bâ and Rivera-Ocasio, 2015). It covers large areas of the coastal floodplain as individual trees and small patches adjacent to mangroves, and along rivers and in mountains (Eusse and Aide, 1999). In the Caribbean, this unique P. officinalis swamp forest provides a habitat for many species of plants and animals and reduces soil erosion along the margins and riverbanks in coastal and mountain areas (Saur and Imbert, 2003;Bâ and Rivera-Ocasio, 2015). Despite its ecological interest, most of the populations of P. officinalis in the Caribbean islands are restricted to a small area due to wetland drainage and urban development (Bâ and Rivera-Ocasio, 2015). Furthermore, the low genetic diversity found within and between populations of P. officinalis is exacerbated by a strong inbreeding depression (Muller et al., 2009). As a consequence, management and conservation measures must be implemented to preserve the remaining P. officinalis populations.
In some Caribbean islands, a dominant management strategy for P. officinalis conservation is to plant agricultural crops under these stands. The Guadeloupean smallholder farmers notably conduct taro (Colocasia esculenta L. Schott) monocultures under the P. officinalis stands in freshwater flooding swamp forests because of higher crop yields compared to other agricultural practices (Saur and Imbert, 2003). No fertilizer or pesticide are used, and the P. officinalis trees are not impacted (no cutting), preserving the tree genetic diversity, and the ecosystem processes (e.g., nutrient cycling, local biodiversity) (Rivera-Ocasio et al., 2006;Bâ and Rivera-Ocasio, 2015). The ecological mechanisms sustaining the functioning of this traditional agroforestry system remains poorly investigated. Indeed heavy leaching of soils brought by seasonal flooding contributes to the shortage of available P and N and should be detrimental for taro monocultures (Bâ and Rivera-Ocasio, 2015). Pterocarpus stands might be beneficial to understory taro cultures by (i) maintaining humidity and temperature at certain levels to prevent water stress (Sanou et al., 2012), and (ii) improving the N input on soil and non-legume plants through biological nitrogen fixation process (Koponen et al., 2003;Saur and Imbert, 2003). The transfer of N from legume to the non-legume can occurred through root exudation, root and nodule decomposition and mineralization, as well as mediated by plant-associated arbuscular mycorrhizal (AM) fungi (Kaiser et al., 2015). Both P. officinalis and C. esculenta establish a symbiosis with AM fungi (Saint-Etienne et al., 2006;Wang and Qiu, 2006;Fougnies et al., 2007), notably improving plant phosphorus (P) nutrition (Smith and Read, 2008), but also potentially plant N nutrition (Hodge et al., 2001;Walder et al., 2012). In addition, P. officinalis requires P from AM fungi not only for their nutrition but also for efficient nodule formation and nitrogen fixation (Fougnies et al., 2007;Le Roux et al., 2014). Whereas, the diversity of nitrogen-fixing bacteria associated with P. officinalis in Caribbean swamp forests has been described (Le Roux et al., 2014), AM fungal diversity has been poorly investigated.
Arbuscular mycorrhizal fungi are the most common and widespread symbiosis involving 86% of land plants including many important crops (Davison et al., 2012). The extraradical phase of AM fungi acts as an extension of the root system for the uptake of nutrients in exchange for plant-synthesized carbon. AM fungi are assumed to exhibit non-specific symbiosis but a given AM taxa could have different effects depending on plant species (Thioye et al., 2017). The relatively low host specificity of AM fungi, increases the possibility that extraradical fungal hyphae links multiple plant species to form common mycorrhizal networks (CMNs) in a plant community (Cheng and Baumgartner, 2004;Ingleby et al., 2007). Mycorrhizal networks are known to drive nutrient transfers (mainly C and N) between adult trees and seedlings for one plant species, and among different plants species (e.g., legume and non-legume) (Cheng and Baumgartner, 2004;He et al., 2004;Selosse et al., 2006;Wahbi et al., 2016). However, mycorrhizal networks have been mainly assessed at the fungal strain level in controlled conditions (Walder et al., 2012) and rarely at the community level (Montesinos-Navarro et al., 2012). Thanks to the development of high-throughput sequencing approaches such as pyrosequencing, the complexity of AM fungal community among plant roots can be deeply assessed (Hart et al., 2015), and a wide range of ecological studies based on the diversity of AM fungal taxonomic markers such as the SSU rRNA gene has been performed (Öpik et al., 2006, 2014).
The current study aims the molecular characterization of AM fungal community composition and distribution between P. officinalis and C. esculenta crops in a traditional Guadeloupean agroforestry system (swamp forests) in order to evaluate if the sustainability of the system might be explained by a high similarity or a high dissimilarity of AM fungal community between Pterocarpus and taro. In addition, the comparison of AM fungal community of P. officinalis adult trees and seedlings with the ones of taro were investigated because of the potential importance of seedlings in the sustainability of the agroecosystems since they are conserved inside the taro cultures. Consequently, two main questions were assessed, (i) What are the predominant mycorrhizal taxa in the traditional agroforestry system, and (ii) What is the degree of similarity of AM fungal members among P. officinalis adult trees, seedlings and taro?

Study Sites and Sampling
We conducted the sampling during 2012 in two representative Pterocarpus swamp forests located in the Grande-Terre island of Guadeloupe: Grande Ravine (GR) (16 • 13 N, 61 • 28 W) and Belle Plaine (BP) (16 • 17 N, 61 • 31 W) (Supplementary Figure  S1). The P. officinalis stands, which comprise 45 and 52 adult trees for GR and BP sites, respectively, are fairly forested and contain dense populations of understory regenerating seedlings. The GR forest site (approximately 0.3 ha) is located along the GR river and taro plants (C. esculenta) were cultivated by smallholders farmers under adult trees and between naturally regenerating seedlings. Some individuals of understory plant species such as Ficus sp.1, Commelina sp.1, and Mimosa pudica were naturally associated with P. officinalis. The BP forest site (approximately 0.4-ha), is located around the bay of the Grand Cul-de-sac Marin, in the near mangrove area and taro plants were also cultivated between Pterocarpus trees and regenerating seedlings. Understory species like Musa sp. and Ficus sp. are represented by a few individuals and are widely spaced from one another. In each forest site, soil cores (200 g of fresh soil) were randomly collected near three adult trees (more than 25 m high), three seedlings (1 < height < 2 m) and three taro plants. Overall, 18 soil cores were stored at 4 • C before being processed. Roots were separated from soil, gently washed with tap water and dried with Silica-gel until molecular analyses. Soil physico-chemical parameters were measured at the Celesta-lab (Mauguio, France) (Supplementary Table S1).

Molecular Analyses
For each root sample, the three replicates were pooled and subjected to liquid nitrogen grinding for homogenization. Total DNA was extracted from a sub-sample (100 mg of dried root) using a FastPrep-24 homogenizer (MP Biomedicals Europe, Illkirch, France) and the FastDNA R SPIN kit (MP Biomedicals Europe) according to manufacturer's instructions. The quality of DNA extracts was improved by adding 20-30 mg Polyvinylpolypyrrolidon (PVPP) during the first step of DNA extraction. Two DNA extractions were done per root sample.

Data Processing and Taxonomic Assignment of AM Fungal Sequences
Four hundred and fifty-four-sequencing data were analyzed using Mothur software according the standard operating procedure 2 proposed (Schloss et al., 2011), except for the quality cutoffs, for which it has been set up at Q30 (trim.seqs command). All sequence reads were then depleted of barcodes and primers (final length 230 bp), and sequences < 100 bp or with ambiguous base calls or with homopolymer runs exceeding 8 bp were also removed. A pre-clustering step (Huse et al., 2010) was also performed to remove sequences still likely due to pyrosequencing errors. Chimeric sequences were checked by using UCHIME (Edgar et al., 2011) and removed. Finally, sequences were identified using the Glomeromycota-based alignment database (Krüger et al., 2012) and sequence similarity ≥60% at the family level.
Clustering of sequences in operational taxonomic unit (OTUs) was performed using dist.seqs and cluster commands in Mothur.
Then, the number of sequences from each sample was normalized with sub.sample command. This sub-sampling step allows reducing the number of spurious OTUs and is crucial to obtain robust estimation of alpha and beta diversity (Gihring et al., 2012). Finally, OTUs were defined at 97% similarity level for taxonomic affiliation.

Statistical Analyses
Diversity (Shannon, inverse Simpson [1/D]), richness (number of OTUs, Chao1) and evenness (Pielou) indexes were estimated using R version 3.3.2 (R Core Team, 2016) and the R package vegan (Oksanen et al., 2016). The sequencing effort was evaluated using the coverage calculator and Boneh estimator (Boneh et al., 1998) implemented in Mothur. We assessed the effects of forest site, plant species, Pterocarpus age categories and their interactions on the AM fungal community composition by non-parametric permutational multivariate analysis of variance (PERMANOVA) (Anderson, 2001) implemented in the perm.anova() function from the R package RVAideMemoire (Hervé, 2016). The differences in AM fungal community structure among forest sites and plant species were assessed using PERMANOVA in adonis() function (McArdle and Anderson, 2001), both from the R package vegan. The AM fungal community structure was based on the Bray-Curtis dissimilarity index as defined in vegdist() function from the R package vegan. Multivariate dispersion was estimated using the betadisper() and permutest() functions (999 permutations; alpha = 0.05) from the R package vegan because it can affect PERMANOVA results. Differences in the relative abundances of AM fungal OTUs among forest sites or plant species were estimated using Kruskal-Wallis' test implemented in kruskal.test() function from the R package stats. The frequency of AM fungal OTUs was determined using the strassoc() function from the R package indicspecies (De Cáceres and Legendre, 2009).
The determination of AM fungal OTUs preferentially associated with a given forest site was performed using the corrected Pearson's phi coefficient of association ("r.g") implemented in the multipatt() function (De Cáceres et al., 2010) from the R package indicspecies. AM fungal OTUs preferentially associated with a given plant species was assessed using the corrected indicator value index ("IndVal.g"species) from the R package indicspecies. A procedure based on determination of species (i.e., OTU) and group (i.e., plant type) combinations was applied using successively combinespecies() and multipatt() functions. This procedure was demonstrated to bear more ecological informations and to determine more robust predictive indicator value than by considering species or group independently (De Cáceres et al., 2012). Two different probabilities were calculated, i.e., A (specificity), representing the probability of a sample to be defined by a group (i.e., plant type), given that the species or the species combinations have been detected, and B (sensitivity) representing the probability of finding the species or the species combinations in different samples characterized by a given group (i.e., plant type). Only AM fungal OTUs present in two samples among three groups defined (i.e., plant type) were subjected to analysis. We considered as valid indicators the OTUs showing both A (specificity) and B (sensitivity) superior to 0.8 and 0.6, respectively, as recommended in Suz et al. (2014).
The AM fungal community membership among C. esculenta and P. officinalis adult trees and seedlings was assessed using venn diagram analysis with the R package VennDiagram (Chen, 2016), and the relative abundance of AM fungal taxa shared among plants was characterized using bipartite network analysis with the plotweb() function from the R package bipartite (Dormann et al., 2008).

AM Fungal Community Composition among Forest Sites
The global 454-pyrosequencing data were composed of 210,676 reads, and 155,544 reads (74%) passed the quality control steps. The average read length was 230 bp. After trimming, pre-clustering and chimera detection steps, 70,949 sequences were classified using a Glomeromycota-based alignment database. A total of 31,803 non-Glomeromycota sequences were removed from the dataset (70,949), as well as singletons (1515 sequences) that are mostly considered as artifacts and can lead to overestimations of AM fungal diversity. The AM fungal sequences (37,631 reads) were assigned to a total of 215 OTUs based on a sequence similarity threshold ≥ 97%. The sequence number between samples was rarefied to 440 sequences per sample (threshold based on the sample with the lower number of sequences) to improve statistical robustness. A high diversity coverage (94-98%) was reached for all samples, with less than eight potential OTUs that were not retrieved (Boneh estimation, Supplementary Table S2). The Boneh estimation showed that the sequencing depth was sufficient to estimate and compare the microbial diversity of the samples (Supplementary Table S2).
Richness, diversity and evenness were calculated for the different AM fungal communities. No significant difference was observed at both forest sites (Supplementary Table S2). AM fungal community structure analysis, based on Bray-Curtis index, also showed no significant difference between both forest sites ( Table 1). However, four OTUs affiliated to Acaulospora, Gigaspora, and Incertae sedis Glomus showed significantly higher abundance in BP, and three OTUs affiliated to Rhizophagus and Funneliformis in GR (Figure 1, for all comparisons see Supplementary Table S4). Among the 215 AM fungal OTUs, only four were determined as preferentially associated at a given forest site (Supplementary Table S5), among which two of them, i.e., Gigaspora (OTU28) and Rhizophagus (OTU16) were exclusively found in BP and GR sites, respectively.

AM Fungal Community Composition among Plant Types
Arbuscular mycorrhizal fungal community richness (for the two indices Chao1 and OTUs number) was significantly different between taro and Pterocarpus (P = 0.033; Supplementary Table  S2) whereas only a locality-dependent plant type effect was observed on AM fungal community structures (P = 0.011; Table 1). The analysis of AM fungal community structure   Venn diagram analysis of OTU-based AM fungal community revealed a low membership characterized by a high number of OTUs specific to one plant and few overlapping OTUs among taro and the two age categories of Pterocarpus, i.e., 10 and 8% of common OTUs in BP and GR sites, respectively (Figure 2A). Taro showed the highest number of specific OTUs (>35%) compared to Pterocarpus (<25%) in both forest sites. Bipartite network analysis showed that the overlapping AM fungi between plants species was mainly composed of Funneliformis OTUs, with 53 and 61% of sequences in BP and GR forest sites, respectively ( Figure 2B). Rare OTUs mainly constituted the plant-specific OTUs, which fit with the low number of indicator taxa associated with the different types of plants.

DISCUSSION
Arbuscular mycorrhizal fungi might play a major role in the functioning of the traditional Guadeloupean agroforestry system associating P. officinalis trees, their naturally regenerating seedlings, and an understory taro monoculture, and the degree AM fungal community similarity among plants has been hypothesized as one of the main factors. Significant differences in AM fungal community richness and structure were observed among the different plant types, but not in terms of diversity. Some AM fungal OTUs were preferentially associated with a given plant type, but a highly predominant AM fungal OTU affiliated to Funneliformis was detected among all plants.

Characteristics of AM Fungal Community
A relatively high AM fungal richness (>120 OTUs) was observed compared to other AM fungal surveys using high-throughput methods from a wide range of in tropical forest ecosystems (22-207 OTUs;De Beenhouwer et al., 2015;Holste et al., 2016;Rodríguez-Echeverría et al., 2016). As shown by Bainard et al. (2011), tree-based cropping systems, combining different tree species (white ash, hybrid poplar and Norway spruce) with annual crops (corn, soybean, and winter wheat), can present a highly diverse AM fungal community. However, the robustness of comparisons between different tropical agro-ecosystems remains questionable due to the scarcity of studies in tropical regions. Furthermore, OTU-based fungal richness is highly dependent on the bioinformatics treatment applied (mainly sequence quality filtering and clustering methods) (Bálint et al., 2016); the biological material analyzed (roots, spores, and extraradical mycelium) (Varela-Cervero et al., 2015) or the methodology used (spore identification, PCR-cloning, pyrosequencing) . The predominance of Glomeraceae, which is the most widespread family in natural and managed ecosystems (Oehl et al., 2010;Brearley et al., 2016) and Acaulosporaceae was in agreement with several surveys carried out in tropical environments (Leal et al., 2013;De Beenhouwer et al., 2015;Holste et al., 2016). Only 4% of all detected AM fungal OTUs were poorly affiliated to a reference taxa in databases, which contrasted with previous data in dry afromontane forests (Wubet et al., 2003) and dry tropical regions (Rodríguez-Echeverría et al., 2016), where up to 18 and 15% of OTUs were considered as new taxa, respectively.
Characteristics of AM fungal community (diversity, richness, and structure) were relatively comparable in this cropping system between the two forest sites (separated of 15 km). However, some AM fungal were associated to a given forest site, notably for BP site. The soil characteristics and soil hypoxia are known as major drivers of AM fungal community (Helgason and Fitter, 2009;Oehl et al., 2010;Alguacil et al., 2016) and differences observed between the both forest sites in soil nutrient contents (N, Ca, and Na; Supplementary Table S1) and flooding duration might have favored specific taxa. The low number of indicator species observed in our work was consistent with the study by Moora et al. (2014), which showed that forest plantations or cultivated lands have very few AM indicator species compared to primary forests or permanent grasslands.

Degree of AM Fungal Community Similarity between Tree and Culture
The analysis of AM fungal community composition and structure demonstrated a site-dependent host plant effect due to low abundant OTUs notably belonging to Gigaspora for taro, Geosiphon and Archaeospora for Pterocarpus adult trees. In addition, the significant differences observed between Pterocarpus adult trees and seedlings confirmed the modification of AM fungal communities according to the plant age (Wubet et al., 2009). Our data corroborate previous molecular studies conducted in tropical forests where divergent AM fungal communities of co-occurring plant species were reported (Uhlmann et al., 2004;Wubet et al., 2009;Mangan et al., 2010). However, three highly abundant AM fungal taxa (80% of sequences) were associated to the three plant types, notably Funneliformis (OTU1) that is considered as a generalist AM fungus (Öpik et al., 2006). Funneliformis was shown to form CMNs for transport of N, particularly in tropical environments where N is poorly available (Walder et al., 2012;Munroe and Isaac, 2014). CMNs might play a major role in the sustainability of Pterocarpus-taro agroforestry systems and high crop yields. First, the CMNs maintained by Pterocarpus could provide an AM fungal mycelium reservoir enabling a faster colonization of shortlived crops under swamp forests (Kuyper et al., 2004) compared to an AM spore reservoir (Brundrett and Abbott, 1994). Secondly, the CMNs could be involved in the N transfers from Pterocarpus trees or seedlings to taro in N-deficient soils of swamp forests, as observed between other legume and non-legume associations (Thilakarathna et al., 2016).
Funneliformis has been, however, described as a ruderal and stress tolerator taxa mainly involved in plant protection against biotic and abiotic stress rather than in plant nutrition (competitor) (Chagnon et al., 2013). Indeed, Funneliformis was shown to protect tropical plants against certain pathogens (Cardoso and Kuyper, 2006). Several strategies could be set up to experimentally test the significance of Funneliformis in nutrient and/or plant protection in systems associating Pterocarpus-taro. Compartmented microcosms with inoculated Funneliformis strains and the use of a 15 N-labeled growth substrate as designed in Walder et al. (2012) could be used to demonstrate the N transfers through Funneliformis CMNs between the two studied plants (Figure 3A). Moreover the benefits of Pterocarpus-taro associations compared to taro monoculture could be estimated experimentally in pot . This type of screen is pervious for fungal hyphae but not for roots and allows the separation between two plants. Three months after planting, an isotopic labeling experiment is conducted utilizing 15 N. Plants are harvested 20 days after labeling. Percentage of root colonized by the inoculated Funneliformis strain and 15 N abundance of taro plants are determined. (B) Pot culture experiments, consisting either to a model of monoculture (taro/taro) or a culture association (P. officinalis/taro). Plants are harvested after 12 weeks of growth and taro growth parameters are measured. The experiment aims the determination of the "competitor" status of Funneliformis. (C) Pot culture experiments, consisting in taro monoculture, to study the potential role of Funneliformis in plant pathogen defense. Plants are grown for 3 months to allow the establishment of Funneliformis and then are inoculated with a pathogen. Plants are harvested after 4 weeks of growth and growth parameters of both plants are measured. The experiment aims the determination the "ruderal" status of Funneliformis. cultures (Figure 3B), allowing the evaluation of Funneliformis contribution as competitor taxa (Chagnon et al., 2013). Finally, the introduction of a pathogen in inoculated or not inoculated pot cultures could be used to evaluate the significance of Funneliformis regarding plant protection (Figure 3C), but also the contribution of Funneliformis as ruderal taxa (Chagnon et al., 2013).

CONCLUSION
Our study highlights the high AM fungal diversity and richness associated with roots of Pterocarpus (adult trees and seedlings) and taro. Although the AM fungal community is significantly different in terms of membership and structure between the types of plants, Pterocarpus and taro had few but predominant overlapping AM fungi, notably Funneliformis spp. (OTU1). From an agricultural point of view, in addition to the good tolerance of taro to waterlogging and shade under Pterocarpus swamp forests (Saur and Imbert, 2003), the preservation of Pterocarpus adult trees and their seedlings could be one of the main factors leading to high taro crop yields by maintaining N input on soil and a source of AM fungal inoculums that might form potential CMNs crucial for the establishment of taro.

AUTHOR CONTRIBUTIONS
AB and AnG designed the research and collected the samples. AlG and HS developed the methodology and performed statistical analyses. AlG and HS generated data. AB, AlG, and HS wrote the initial manuscript. AB, AlG, AnG, and HS contributed to the final manuscript. All the authors shared, edited and approved the final manuscript.