Experimental design

Soil organo-physico-chemical analyses

All organic-physical-chemical analyses were carried out by an independent laboratory for analysis, study and advice on soil biology (Celesta lab, https://celesta-lab.fr); see methods S1 for more information.

Plant community inventory

Plots of 20 x 20m were inventoried for plant species with DBH > 5cm. In each plot, four 5 x 5m sub-plots (same as above) were established, where plant species over 1m in height were recorded and measured. Additionally, smaller plant species (less than 1m in height) were counted within these sub-plots.

Nematodes survey

On the same soil samples used for previous analyses, a survey of nematodes was realised by the independent engineering office Elisol environnement (https://www.elisol.fr). The taxonomic distinction was made up to the families. The abundance per site was also recorded (number of individuals per 100g of dry soil).

Molecular method

Bioinformatics

Statistical analyses

Soil eDNA pre-processing analysis

For the ITS2 marker (fungi) 2,594,514 raw sequences from 15 samples were obtained and then 270,160 sequences were kept after different cleaning steps (S1 and S2 Tables). Due to a calculated rarefaction of 12,582 reads, four plots were not kept for further analysis (namely, plots F2, LF2, LF5 and SF3) (S2 Fig). For the V4 marker (bacteria), 3,064,846 raw sequences from 15 samples were obtained and then 236,235 sequences were kept after the cleaning steps (S3 and S4 Tables). As a result of a calculated rarefaction of 4,483 reads, two samples were removed for subsequent analyses (i.e., F2 and LF2) (S3 Fig). Thus, 270,160 quality-filtered fungal sequences (ITS2) and 102,277 quality-filtered bacterial sequences (V4) from 11 and 13 soil samples respectively were finally generated and further analysed.

Soil fungal and bacterial diversity

In total, 383 and 94 fungal and bacterial ASVs, respectively, were delineated. For both fungi and bacteria, no significant differences were observed in diversity indices between the conditions (i.e., SF, LF, and F) (S4 and S5 Figs).

Soil fungal and bacterial composition and functional groups

Fig 2 presents the relative abundances of the fungal phyla (Fig 2A) and functional groups (i.e., guilds and trophic modes) (Fig 2B). Ascomycota was observed as the most abundant phylum in each condition (SF: 55.4% ±18.6%, LF: 63.9% ±6.9%, and F: 61.5% ±8.1%), followed by Basidiomycota (SF: 33.5% ±18.5, LF: 23.9% ±7.8%, and F: 28.9% ±11.1%). All other phyla (Rozellomycota, Chytridiomycota, Mucoromycota, Calcarisporiellomycota, Glomeromycota, and Mortierellomycota) showed a relative abundance inferior to 8%. No significant variations in the proportions of the relative phyla abundances between the three conditions were detected (Kruskal Wallis test).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2.

Bar plots of relative abundance (in per cent) of fungi phyla (A), their corresponding guilds (B), and of bacteria phyla (C) and their corresponding traits (D). Conditions legend: SF is for Short Fallow; LF is for Long Fallow and F is for Forest.

https://doi.org/10.1371/journal.pone.0311986.g002

Regarding fungal guilds, the undefined saprotroph guild was the most relatively abundant in short fallow (44.4% ±11.3%) and long fallow (40.4% ±9.4%), and the second most abundant in the forest (36.2% ±19.1%). In the forest, the animal pathogen guild was the most relatively abundant guild with a proportion a relative abundance of 41.0% ±20.2%, whereas it was the second most abundant guild in short fallow (25.4% ± 13.8%) and long fallow, as well (40.0% ± 21.5%). The plant-pathogen guild was the third most relatively abundant guild for all conditions (SF: 13.6% ±8.3%, LF: 9.0% ±8.9%, F: 9.4% ±6.3%). The guild of ectomycorrhizal fungi was the fourth most relatively abundant guild in all conditions (SF: 8.9% ± 5.3%, LF: 5.3% ± 3.1%, F: 5.9% ± 3.6%). All other guilds showed a relative abundance inferior to 8%. When comparing the different conditions, the Kruskal-Wallis test revealed no significant variation in the relative abundances of these guilds.

The relative abundance of each bacterial phyla (and the only archaeal phyla), and their corresponding functional traits (excluding Archaea) are presented respectively in Fig 2C and 2D. For the phylum composition, in the three conditions studied, two bacterial phyla dominated the soil communities, namely the Firmicutes (SF: 31.5% ±9.3%; LF: 27.3% ±1.5%; F: 38.9% ±5.4%) and the Verrucomicrobiota (SF: 17.4% ±9.3%; LF: 32.7% ±13.3%; F: 15.3% ±%5.4). The only detected archaeal phylum that was the Crenarchaeota, was also observed in relatively high proportions (SF: 25.9% ±12.5%; LF: 16.8% ±8.4%; F: 18.6% ±2.5%) (Fig 2C). All these phyla did not show any significative differences in their relative abundances between the three compared treatments. The only phylum that presented significant variations in its proportions (SF: 4.9% ±0.6%; LF: 1.2% ±1.4%; F: 9.4% ±2.3%) was the Proteobacteria (Kruskal Wallis test p-value = 0.004723).

Concerning the bacterial functional traits, the organotroph-chemotroph functional group was dominant in all conditions (SF: 80.4% ±2.7%, LF: 72.0% ±11.2%, and F: 80.9% ±2.1%), and was followed by the heterotroph group (SF: 11.7% ±2.7%, LF: 21.5% ±12.2%, F: 10.4% ±2.5%). The organotrophs and the composite group represented systematically less than 8%. The heterotrophs were the only functional group that showed a significant variation between conditions (Kruskal Wallis test, p-value = 0.0333).

Microbial communities’ structure

The NMDS ordination, based on the Bray-Curtis dissimilarity index, suggests that soil fungal communities were distinct between the studied conditions, particularly between short fallow and forest (Fig 3A). The PERMANOVA analysis supports this observation (PERMANOVA: p-value = 0.018, R² = 0.277 for sites and 0.723 for residual, post-hoc test pairwise adonis F vs. SF p-value = 0.019; F vs. LF and LF vs. SF were non-significant). Conversely, for bacteria, no community structure was observed (Fig 3B; PERMANOVA non-significant). Thus, in contrast to fungi, bacterial communities did not exhibit any significant differences between land-use conditions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3.

Non-metric multidimensional scaling (NMDS) for fungi phyla (A) and bacteria phyla (B). SF is for Short Fallow (brown letters); LF is for Long Fallow (light green letters) and F is for Forest (forest green letters).

https://doi.org/10.1371/journal.pone.0311986.g003

Influence of physico-chemical parameters

The soil texture (i.e., the proportions of clays, silt, and sand) was homogeneous among plots and, hence, among the three related conditions investigated (S5 Table, S6 and S8 Figs) and was classified as silt loam. The organic matter, carbon (C), nitrogen (N), and pH showed significant differences (p-values < 0.05), with systematically higher values in the forest (S5 Table and S8 Fig). In linked organic matter analysis, C/N showed significant differences (p-value = 0.029), with a higher ratio in long fallow. Regarding the microbial biomass parameters, significant differences were observed (S5 Table and S8 Fig). Indeed, the carbon, the estimated total microbial biomass, as well as the estimated related parameters nitrogen, phosphorus, potassium, calcium, and magnesium stored in microbial biomass showed significant differences between conditions (p-values < 0.05). Looking at the pairwise comparisons, the soils in forests presented, for most parameters, higher significant values. In addition to these investigated parameters, significant differences were found in mineralized carbon (microbial activity), decreasing from forest to short fallow (p-value = 0.015, S5 Table; S8 Fig). Since a structure of communities according to the studied conditions was only observed for fungi (Fig 3), db-RDA analysis using soil physico-chemicals parameters as explanatory variables has only been proceeded on this microbial group. The db-RDA representation showed that fungal phyla were significantly related to the soil texture and not to other parameters (S9 Fig–with only the significant parameter, i.e., the soil texture). The PERMANOVA (nPerm = 9999; 20% of the variance explained, p-value = 0.007) and post-hoc tests (Clays, p-value = 0.031; and Silt, p-value = 0.009) supported this relationship. More precisely, the Basidiomycota were found to be related to the silt content, whereas the Ascomycota were inversely related to the clay content (S9 Fig). However, no relationships were detected regarding the soil fungal fallows and forest communities. So, the soil texture was homogeneous among plots, but the soil physico-chemical properties, including organic matter, carbon, nitrogen, and pH, showed significant differences between the forest and fallow conditions, with the forest having systematically higher values.

Influence of plants

Kruskal-Wallis tests, followed by Dunn post-hoc tests (S6 Table) revealed significant differences in plant species composition between the forest and fallow conditions, with several species showing significant presence or absence in specific conditions, such as Acacia spirorbis (p-value = 0.00621) being absent in the forest and Dodonaea viscosa (p-value = 0.00327) being more present in short fallow.

Influence of nematodes

One nematode family, Aphelenchoididae (Kruskal-Wallis test: p-value = 0.00918), showed significant changes between conditions, with differing abundance in long-term and short-term fallows (S7 Table).

‘Targeted’ analysis for fungi and bacteria with the Anaconda R package

Eleven and 13 samples were used, respectively, for fungi and bacteria/archaea analyses (as a result of the deletion of samples due to the previous rarefaction step). An estimate of the dispersion by shrinkage can be visualised by plotting the dispersion estimates on the average ASVs presence strength (here ‘ASV abundance’ is used as a ‘count’) (S10 Fig) by adjusting only an intercept term. First, and following [31], the maximum likelihood estimate of the ASVs was obtained using only the respective ASVs data (black dots). Then, a curve (red) was fitted to the maximum likelihood estimate to capture the general trend of the dispersion-mean dependence. This fit was used as a prior mean for a second round of estimation, which resulted in the final estimates of the dispersion at the maximum a posteriori. This can be understood as a narrowing (blue circle) of the noisy estimates by ASVs towards the consensus represented by the red line. The black points circled in blue were detected as outliers of the dispersion and were not reconciled with the prior (the reconciliation would follow the dotted line). In our case, we see that few ASVs were not fitted in the (here, parametric) model (which is normal according to [31]) and that the results were very similar between the two kingdoms, although the bacteria showed few ASVs in comparison. The analysis of the inter-sample relationships after the previous transformation (Fig 4) showed us that the variability observed in the previous analyses (e.g., sections 3.2 to 3.4) was well preserved. For example, the similarity between the NMDS (Fig 3) and the PCA presented here was remarkable. Nevertheless, we can observe nuances in this variability.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4.

Analysis of the inter-sample relationships for fungi (A) and bacteria (B) with the Anaconda R package. For each, at the left, a hierarchical clustering on Euclidean distances on logarithm-transformed ASVs abundance with average clustering method, at the right a Principal component analysis (PCA), and at the bottom, a sample-to-sample heatmap based on rlog transformation with trim on too low represented ASVs. SF is for Short Fallow (brown letters); LF is for Long Fallow (light green letters) and F is for Forest (forest green letters).

https://doi.org/10.1371/journal.pone.0311986.g004

For fungi (Fig 4A), the hierarchical clustering on Euclidean distances on logarithm-transformed ASVs abundance with average clustering method showed higher differences in sample relationships than in the PCA. As an example, the samples from the Forest condition (‘F’) were tightly grouped in this PCA whereas they were fitted in three different sub-clusters in the hierarchical clustering. The 31% (18%+13%) total explained variation in the PCA showed that a small part of the data explained this convergence. The sample-to-sample heatmap based on rlog transformation with trim on too low represented ASVs showed a certain homogeneity of the samples, which could illustrate a variability homogeneously explained by some ASVs (with a variability not pulled by only some ASVs in a specific way, but also by several ASVs in the same direction).

For bacteria and archaea (Fig 4B), the hierarchical clustering on Euclidean distances on logarithm-transformed ASVs abundance with average clustering method showed similar differences in sample relationships than in the PCA. As an example, the samples F1, F3, F4, SF1, and SF2 were at the margin of the other samples in the PCA, which was well highlighted by a similar sub-cluster in the hierarchical clustering. The 34% (19%+15%) total explained variation in the PCA showed that a small part of the data explained the presented variation. The sample-to-sample heatmap based on rlog transformation with trims on too low represented ASVs, showed higher heterogeneity within some of the samples, as for LF5-SF5 for example, this could display a variability explained by few ASVs in a heterogeny way (with a statistical variation pulled by few ASVs, and not a lot of ASVs). Such similarities (e.g., between the previous and the Anaconda analyses) with nuances allowed us to ensure that the variability structure of the dataset (e.g., few ASVs highly over or under-represented in a condition versus a lot of similar ASVs slightly over or under-represented in a condition) was maintained while allowing us to explore the smallest variation so that we can answer our biological/ecological question with further analysis (see below).

The clustered heatmap of the 75 most abundant ASVs based on Euclidean distance with average clustering method for fungi (S11 Fig) and bacteria/archaea (S12 Fig) showed that there is no discernible pattern based on the most prevalent ASVs. This allows for condition-specific analyses to recover ASVs that are specifically over- or under-represented in the different conditions, which could therefore explain the observed variations.

The DeSeq2 algorithm allowed such comparison, and here with P-adjusted < 0.05 and LogFoldChange > |2|, F vs. LF, F vs. SF, and SF vs. LF comparison hooked, respectively, 43, 96, and 43 significantly under- or over-represented ASVs respectively for fungi (Fig 5A), and 33, 35, and 17 significantly under- or over-represented ASVs for bacteria (Fig 5B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5.

MA plots (A and B) from the mean of normalised count by their respective log fold change, and Venn diagrams from ASVs (C and D) realised for fungi (A and C) and bacteria (B and D) with the Anaconda R package. SF is for Short Fallow; LF is for Long Fallow and F is for Forest.

https://doi.org/10.1371/journal.pone.0311986.g005

Venn diagram representations (Fig 5C and 5D) allowed us to recover species (ASVs) that were i) specific to a comparison, and most importantly that were ii) specific to a condition (the latter correspond to those with a common denominator, e.g., F vs. SF compare to F vs. LF will show ASVs specific to F). For fungi (Fig 5C), of the 43 ASVs significantly over- or under-represented in the F vs. LF pairwise comparison, nine were specific to this comparison. Out of the 96 ASVs significantly over- or under-represented in the F vs. SF comparison, 40 were here specific to this comparison. For the SF vs. LF comparison, of the 43 ASVs significantly over- or under-represented, 13 were specific to this pairwise comparison. Thirty ASVs, 4 ASVs, and 26 ASVs significantly over- or under-represented were specific to the forest, the long fallow and the short fallow, respectively (condition-specific ASVs). For the bacteria/archaea (Fig 5D), of the 33 ASVs significantly over- or under-represented in the F vs. LF comparison, two were only recovered from this comparison. Out of the 35 ASVs significantly over- or under-represented in the F vs. SF comparison, only one was specific to this pairwise comparison. Of the 17 ASVs significantly over- or under-represented in the SF vs. LF comparison, none were specific. Twenty-four, 7, and 10 ASVs that were significantly over- or under-represented were, respectively, restricted to the forest, the long fallow and the short fallow (condition-specific ASVs).

Looking at the fungal ASVs (p-value < 0.05; LogFoldChange > |2|) that were condition-specific (Fig 6A), 15 ASVs were over-represented in short fallows (present only in short fallows), particularly Sarocladium kiliense, Acrocalymma fici and Exophiala aquamarina, and 11 were under-represented in short fallows (present in forests and long fallows, but not in short fallows), notably Acrocalymma walkeri, Angustimassarina acerina, and Mortierella minutissima. In long fallows condition-specific, three ASVs were found to be over-represented, such as Trechispora invisitata, and one was under-represented, namely Agaricales sp. 01. In forests condition-specific, 20 ASVs were over-represented, like Mortierella bisporalis, Lycogalopsis solmsii and Hygrocybe sp.), and 10 were under-represented, like Spizellomyces punctatus, Botryosphaeria sp. and Mortierella alpina.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6.

‘Targeted’ analysis results graph for fungi (A) and bacteria (‘b’) and Archaea (‘a’) (B) with the Anaconda R package. Results are in Log2FC for the significant ASVs according to the precision of the taxonomic rank: species (‘s’), genus (‘g’), family (‘f’), order (‘o’), class (‘c’), phyla (‘p’), or kingdom (‘k’). SF is for Short Fallow (brown); LF is for Long Fallow (lightgreen) and F is for Forest (forestgreen).

https://doi.org/10.1371/journal.pone.0311986.g006

For ASVs (p-value < 0.05; LogFoldChange > |2|) that were condition-specific for bacteria and archaea (Fig 6B), nine ASVs were over-represented, such as Burkholderiales, Gemmataceae, or Gaiella, and one was under-represented in short fallows, an ASV assigned at the Vicinamibacteraceae family. In long fallows, four ASVs were over-represented, for instance, bacteriap25 sp., Entotheonellaceae sp. and Vicinamibacteraceae sp., and three were under-represented, i.e., Hyphomicrobium sp., Candidatus Udaeobacter and Nitrososphaeria sp. (archaea). Finally, nine ASVs were over-represented, like Candidatus xiphinematobacter, Bacillus sp. and Acidibacter sp., and 15 were under-represented in F, such as three archaeal ASVs belonging to the Nitrososphaeria genus.

So here, hierarchical clustering analysis revealed greater differences in fungal sample relationships, with distinct sub-clusters in Forest samples, while PCA showed a more compact grouping. Bacterial and archaeal samples exhibited similar patterns, with some distinct sub-clusters and others showing higher heterogeneity. Condition-specific ASVs were identified in fungi and bacteria/archaea, with distinct ASVs over- or under-represented in each condition, including Sarocladium kiliense, Acrocalymma fici, and Exophiala aquamarina in short fallows, and Mortierella bisporalis, Lycogalopsis solmsii, and Hygrocybe sp. in forests.

‘Global’ analysis of fungal and bacterial communities with the Anaconda R package

Concerning the ‘global’ analysis for fungi (Fig 7), 1174 ASVs matched to 653 NCBITaxon different terms, and 639 NCBITaxon were remaining. After the secondary clustering, the MWU test output 23, 31, and 21 NCBITaxon terms at 10% FDR for the pairwise comparisons F vs. SF, F vs. LF, and SF vs. LF, respectively. Here, compared to the ‘targeted’ analysis, when an affiliation is made at a higher rank than species (e.g., family, order, or genus), this corresponds to several ASVs that share the same taxonomic rank. When a group of ASVs are ascribed at the species level, it means that several ASVs share this taxonomic affiliation and can correspond to different sub-species, or strains.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. ‘Global’ analysis results graph for fungi with the Anaconda R package for the three comparisons.

SF is for Short Fallow; LF is for Long Fallow, and F is for Forest.

https://doi.org/10.1371/journal.pone.0311986.g007

In the forest, compared to the short fallow, numerous ASVs assigned to the fungal entomopathogen Metarhizium robertsii were over-represented (p-value < 0.05) (Fig 7). In contrast, other ASVs were under-represented, hence, over-represented in the short fallow, such as the ones ascribed to the Curvalaria plant pathogen genus (p-value < 0.05), the potential plant pathogens that are Spizellomyces punctatus (p-value < 0.01) and Fusarium oxysporum (p-value < 0.05), and the potential human pathogen that is Exophiala equina (p-value < 0.05) (Fig 7).

In comparison to the long fallow, Metarhizium robertsii was again found in higher abundance than in the forest. Only Spizellomyces punctatus was significantly observed as over-represented (p-value < 0.05) in the long fallow, but to a lesser extent than in the short fallow (Fig 7).

The comparison between the short and the long fallow periods (Fig 7) showed an over-representation in the former for numerous taxa: Sarocladium, Fusarium oxysporum species complex, Pyrenochaetopsis leptospora, Curvularia, Pleosporaceae, Alternaria, Chaetomella, Chytridiomycota, and Spizellomyces punctatus (p-value < 0.01). Leotiomycetes and Talaromyces sect. Talaromyces were also over-represented in the short fallow (p-value < 0.05). In the long fallow, Hymenochaete acerosa, Glutinoglossum, Lycogalopsis solmsii, Trechispora invisitata, Saksenaea trapezispora, Metarhizium robertsii, and Exophiala equina were found in higher abundances than in the short fallow (p-value < 0.01).

For bacteria and archaea, 486 ASVs matched to 108 NCBITaxon different terms. One hundred NCBITaxon were remaining. After the secondary clustering, the MWU test output zero NCBITaxon terms at 10% FDR for all comparisons. This result mirrors the ones presented in Fig 4, which displays a variability explained by a few ASVs in a heterogeny way (with a variability pulled by only some ASVs in a specific way, and not by several ASVs in the same direction). Here it means that some ASVs are very strongly over- or underrepresented in a condition (which is why the ‘targeted’ analyses worked) but that there are not enough similar ASVs that are slightly over- or underrepresented in a condition in the same direction (which is why the ‘global’ analyses cannot be realised because it is not significant).

To summarize, the analysis of fungal ASVs in the forest, short fallow, and long fallow conditions revealed significant differences in taxonomic affiliations, with Metarhizium robertsii being over-represented in the forest and short fallow, and Curvalaria, Spizellomyces punctatus, and Fusarium oxysporum being under-represented in the forest and over-represented in the short fallow, whereas Leotiomycetes and Talaromyces sect. Talaromyces were over-represented in the short fallow. The analysis of bacterial and archaeal ASVs revealed that only 108 taxonomic groups were significantly represented, with the majority of ASVs remaining unclassified, and no significant differences were found between the forest, short fallow, and long fallow conditions, indicating that a few ASVs strongly drive the variation in this community composition.

Anaconda package on ecological data

To determine the usefulness of this tool for other types of data, the Anaconda package was used on plants’ and nematodes’ ecological data. Using the ‘targeted’ analysis, six plant species were observed as forest-specific (p-value < 0.05; LogFoldChange > |2|), notably Aglaia elaeagnoidea, Diospyros fasciculosa, and Schefflera gabriellae (Fig 8). Acacia spirorbis, Dodonaea viscosa, and Psidium guajava were only encountered in the two fallowing periods (Fig 8). Acalypha grandis and Pitpturus argenteus were only observed in the short fallow (p-value < 0.05; LogFoldChange > |2|), whereas Podonephelium homei and Polyscias bracteata were only found in the two other conditions. Schinus terebinthifolius was found significant in the short fallows compared to the forests (p-value < 0.01), which means that this species was found in larger quantities in short fallows compared to forests. In forests vs. long fallows, Acronychia laevis (p-value = < 0.01) and Glochidion billardieri (p-value = < 0.05) were encountered in greater quantities in long fallows compared to forests, meaning that these species were mostly found in long fallows compared to forests. Finally, Diospyros samoensis (p-value = < 0.05) was observed in higher abundance in forests compared to long fallows, indicating that it was mostly present in the forest and slightly in the long fallow land.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Venn diagram from the ‘targeted’ analysis results graph for plantae for the three comparisons.

Values are corresponding fold changes. Conditional blue values are under-represented ASVs (negative fold change) and conditional red values are over-represented ASVs (positive fold change). SF is for Short Fallow; LF is for Long Fallow, and F is for Forest.

https://doi.org/10.1371/journal.pone.0311986.g008

For nematodes, the comparison between the forest and the short fallow showed significant variation in the abundance of one family, the Tylenchidae, which was more present in the forest (p-value = 0.01665, LogFoldChange = 2.28). A significant variation in another family, the Aphelenchoididae, was also observed when the short and low fallow were compared, with a higher representation in the former (p-value = 0.00974, LogFoldChange = -2.23).

So here, the application of the ‘targeted’ analysis to plant and nematode ecological data revealed significant differences in species abundance between the forest and fallow conditions, with specific plant species showing preferences for either forest or fallow environments, and nematode families exhibiting altered abundance patterns in response to different land-use regimes. This clearly demonstrates the usefulness of this package for ‘classic’ ecological data, and its use can therefore be extended beyond metabarcoding data.

‘Classical’ community analysis: No effect of agricultural practice changes in the first instance

The so-called ‘classical’ community analysis (which refers to the diversity, composition, and structure investigations that are commonly made in community analyses), revealed no effects of cultural practice changes on soil microbial communities. Indeed, no differences in microbial diversity were found between short- and long-term fallowing, and forest, for both fungi and bacteria/archaea. Variations in phylum composition were only observed for Proteobacteria, with a higher proportion in the forest, but not between fallow periods. Based on [25], the relative abundance of Proteobacteria may indicate soil and land degradation, suggesting that both short- and long-fallow periods (the latter lasting over a decade) could be considered as degraded systems. As for the diversity and the phyla composition, the functional groups did not reveal a clear tendency, especially in terms of pathogens and beneficial microbe relative abundances. Looking at the soil microbial community structure, a significant partitioning was only observed for fungi, which resulted from differences with the forest, not from any fallowing period effect.

It’s noteworthy that despite our inability to detect soil microbial changes due to the agricultural practice, some ‘global’ tendencies seem to emerge from those ‘classical’ approaches. Indeed, for fungi, in all conditions Ascomycota was observed as the most abundant phylum, followed by Basidiomycota. In the literature, the dominance of Ascomycota over Basidiomycota has been recently suggested as an indicator of ecosystem degradation [25]. This may thus suggest that even the forests used as reference ecosystems are in some extent degraded. Regarding bacteria, the Firmicutes and Verrucomicrobiota phyla dominated the soil communities. The Firmicutes have been classified as copiotrophic [83] and Verrucomicrobiota as oligotrophs [84]. However, a recent study has conversely shown a positive correlation between Verrucomicrobiota and soil carbon content [85]. The high soil organic carbon content of Maré’s Gibbsic Ferrasols, even in cultivated soils [49], could thus be a probable explanation for the over-representation of these two bacterial groups. In addition to these bacterial phyla, the Crenarchaeota was also well-represented in all conditions and was the only archaeal representative. The dominance of archaeal communities by Crenarchaeota on Maré island is in accordance with the observations made by [86] on divers’ soils at a worldwide scale. This group may play central roles in biochemical cycles in soils [86, 87]. However, deeper investigations are needed to better understand the roles of microorganisms in Gibbsic Ferralsols on Maré island. Indeed, except soil texture, environmental variables were not found to influence soil microorganisms.

At this stage, based on “classical” analyses, we cannot conclude that changing agricultural practices at Maré Island have any effect on soil microbial communities. We cannot rule out a lack of effect, but we can also acknowledge the need for more in-depth approaches to highlight potential changes, particularly in the soil health and One Health context.

Revolutionising soil health and One Health through advanced detection of soil pathogens with the Anaconda package

The two newly developed statistical analyses implemented in the Anaconda package, namely the ‘targeted’ and ‘global’ analyses, highlighted the over-representation of microbial ASVs, particularly for fungi, ascribed to plant and animal pathogens, including humans, in the short fallow. Indeed, fungal pathogens such as Acrocalymma fici, known as a pathogen of cultivable trees [88], Chaetomella raphigera, recognised as a fruit rot pathogen [89], and Gibellulopsis chrysanthemi, identified as a root rot pathogen [90] were detected in significantly higher proportions through the ‘target’ approach in the short-term fallow (Fig 6). Additionally, an undetermined species belonging to the Botryosphaeria genus, a taxon known to be associated with grapevine decline [91], was absent in the forest and present in both fallows, with higher abundance in the short-duration fallow. In congruence with all these results, an increase of plant fungal pathogens in the short fallow was observed using the ‘global’ statistical investigation. For instance, taxa such as Fusarium oxysporum, Alternaria and Curvularia, known to be pathogenic to many plant species [92–94], were particularly present in the short fallow compared to both the long fallow and the forest ecosystem.

In addition to these plant-detrimental microbes, a fungal taxon of primary interest for Human health has also been detected in the short fallow soils, namely Sarocladium kiliense (formerly Acremonium kiliense). S. kiliense is a soil saprophytic fungus that can cause opportunistic infections in immunocompetent and immunocompromised individuals, with diverse manifestations, such as dermatophytosis, onychomycosis, mycetoma, pneumonia and fungemia [95, 96]. Outbreaks of S. kiliense in immunodepressed patients have been reported in the literature [97–99]. These clusters were likely linked to infections in clinical settings [98, 99], but a probable environmental source has also been suggested by [97]. Recently, a fatal disseminated infection in a diabetic patient with coronavirus disease 2019 (COVID-19) has been reported by [100] in Iran. The severity of the diseases that can result from S. kiliense underlines the necessity of a high level of clinical attention in this area. In New Caledonia, at the public hospital, hitherto two cases involving undefined Sarocladium species have been reported (data on geographical origin and patient health non-available) (Arnaud Cannet, pers. com.). In light of the aforementioned fatal case in Iran [100], Sarocladium risk infections have to be in regards to the substantial diabetic population in New Caledonia (ASSNC, 2022), as well as the high prevalence of COVID-19 in the archipelago (WHO Coronavirus (COVID-19) Dashboard, https://covid19.who.int/). The ‘global’ analysis of this study shows an over-representation of the genus in the comparison of short versus long fallows, which also confirms the results of the ‘targeted’ analysis. The over-representation of this harmful fungus in a traditional agricultural system could result in a higher probability of infection and, therefore, support the need to raise awareness about this pathogen among healthcare workers and the local populations.

From the ‘targeted’ analysis (Fig 6), Exophiala aquamarina, an opportunistic fungal pathogen causing cutaneous and disseminated infections in cold-blooded vertebrates (so far restricted to fishes) [101], was also found to be significantly over-represented in the short fallow. Based on the ‘global’ approach (Fig 7), another Exophiala species in the same phylogenetic clade, E. equina, was significantly present in soil samples from both short-term and long-term fallows, with greater representation in the latter. This suggests that agricultural establishment, regardless of the fallowing period, increased this pathogen in Maré’s soils. Similar to S. kiliense, this underscores the need to monitor potential human infections by E. equina, which, although rare, can cause cutaneous and subcutaneous infections [101, 102]. Supporting the necessity of paying attention to this genus, two cases of Exophialum infections have been to date reported for the public hospital in New Caledonia (data not available on the geographical origins of the patients) (Arnaud Cannet, pers. com.).

Regarding bacteria, the Anaconda results were less clear than for fungi. Despite no findings from the ‘global’ analysis, likely due to high intra-sample variability, the ‘targeted’ analysis identified ASVs in the short fallow attributed to taxonomic groups containing or suspected of containing pathogens, such as the Gemmataceae (Planctomycetes) [103] and Burkholderiales [104] (Fig 6). Indeed, molecular-based detection has revealed the presence of Planctomycetes in the blood of leukemic of two aplastic patients with neutropenia, rash, diarrhoea and micronodular pneumonia [105]. The phylogenetic analysis revealed for one of the clinical cases a close relationship to Gemmata obscuriglobus, a species that belongs to the Gemmataceae. For the second case, according to [106], when sequences of the 16S rRNA gene were compared, the second hit with a described taxon was with another Gemmata species, G. massiliana. This bacterium was originally recovered and characterised from a hospital water distribution system in France [107], thus in proximity to patients, as pointed out by [103]. Gemmata-related sequences have also been found in human stool specimens, including individuals with infective endocarditis [108]. From these constatations and other cellular and molecular findings, Gemmataceae representatives which clinical microbiologists have overlooked [106], have been suggested to potentially behave as opportunistic pathogens [103, 106]. Concerning the Burkholderiales (given as a second example), it encompasses a large variety of organisms, in particular plant and animal pathogens, including for humans [109]. Certain Burkholderiales bacteria are considered particularly dangerous for individuals suffering from chronic lung diseases [104].

Uncovering beneficial soil microbes and ecological links with Anaconda’s statistical approaches

As just seen above, the approach implemented in the Anaconda package revealed an increase of soil microbial pathogens with a reduction in the fallowing period. In complement to this compelling constatation, other soil microorganism types that displayed differences in their occurrence and deserve great attention were also recovered from Anaconda analyses. Indeed, several fungal saprophytes, i.e., Glutinoglossum sp, Hymenochaete acerosa, Lycogalopsis solmsii, Trechispora invisitata and Sakseneae trapezispora, were detected in lower prevalence in the short fallow’s soils (Figs 6 and 7). It has been shown that saprophytic fungi can be involved in the regulation of pathogens [92, 110, 111]. Competition for resources [92] and antagonist interactions, via saprophyte fungi promoting soil antifungal bacteria [110, 111], are underlying mechanisms leading to soil pathogen suppression. The lowest value of mineralised carbon in the short fallow, which reflects a lower microbial activity, argues in favour of a diminution of saprophyte activity. We could, thereby, hypothesise that the specific decrease of these saprophytic fungi has favoured the increase of the detrimental microorganisms observed in the short fallow plots. Alongside saprophytes, the fungal animal pathogen Metarhizium robertsii was in a decreasing order well-represented in the forest, then in the long and the short fallow (Fig 7). This fungus is an entomopathogen infecting a wide range of arthropods, and can consequently be involved in insect pests’ regulation [112]. It can also establish itself as a root endophyte and favour plant growth and defence against plant pathogens [113]. The specific lower abundance of this entomopathogenic and plant-endophyte fungus in the short fallow may similarly favour an increase of detrimental organisms. Thus, in the context of soil suppressiveness (i.e., the capacity of any given soil to reduce pathogens and disease incidence), specific suppression mechanisms, through individual species or selected groups of antagonist microorganisms [17, 114, 115], seem to regulate soil borne-pathogens in our system, rather than microbial diversity [7, 115].

Conversely, to the reduction of saprophyte and entomopathogen-plant endophyte fungi, an over-representation of the chemoorganotroph Gaiella bacterial genus [116] was observed in the short-term fallow via the ‘targeted’ analysis (Fig 6). In tomato cropping soils, after organic amendment, a strong relationship was observed between this genus and the inhibition of the soil pathogen responsible for Fusarium wilt [117]. Therefore, in our short fallow system, certain beneficial taxon acting against detrimental soil microorganisms may also be present. The intrinsic balance of soil between its relative abundance of beneficial and detrimental microbes is a crucial factor in determining its capacity to express or suppress diseases. One of the major questions consequently arising is when this threshold leading to one situation or the other would be met (Fig 9).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 9. Synthesis of soil environment components potentially involved in the accumulation of plant and human pathogens with the reduction of the fallowing period in traditional yams culture in Maré Island.

Detected increases and decreases of each component, via the approaches implemented in Anaconda, are represented by arrows within boxes on the left. Related probable positive and negative effects on soil augmentation of detrimental organisms are indicated, i.e., that may favour (+) or suppress (-) pathogens. Potential interactions between other parameters than pathogens are also shown (between A. spirorbis and ectomycorrhizal fungi). The absence of apparent abiotic effects is represented. An important aspect also represented is the potentiality of soil pathogen populations reaching a threshold that could lead to substantial plant and human infections.

https://doi.org/10.1371/journal.pone.0311986.g009

Other biotic components of the soil environment than fungi and bacteria can contribute to soil suppressiveness [114]. As earlier seen, the fungal pathogenic Botryosphaeria genus (Botryosphaeriaceae) was present in both fallows, but particularly in the short one. A Botryosphaeria species has been recovered from Acacia plant species (Fabaceae) in Australia [118]. According to Anaconda results on plant communities, Acacia spirorbis was significantly present in both fallows, with higher relative abundances in the short fallow. The larger abundance of this fungus could thus be related to Acacia’s abundance. Interestingly, another microorganism type, a nematode of the Aphelenchoididae family has been experimentally demonstrated to feed on a Botryosphaeriaceae member [119, 120]. Using again the Anaconda package, significant variations in the abundance of this nematode family were detected, with higher abundances in the short fallow. Preferential grazing of an Aphelenchoididae species on ectomycorrhizal fungi has also been revealed in the literature [119]. A. spirorbis is recognised as an ectomycorrhizal shrub [121], which by the way would explain the over-representation of ectomycorrhizal fungi in the short-term fallowing (i.e., Thelephoraceae and Cortinarius species) (Fig 6). Multiple biotic interactions may thus intervene in the regulation of Botryosphaeria in soil. This fungus could, as previously indicated, benefit from the larger abundance of Acacia, but, at the same time, may be regulated by the predation of nematodes, which are also stimulated by the presence of ectomycorrhizal fungi. A. spirorbis is also able to form another type of mycorrhiza, i.e., endomycorrhiza. This characteristic could explain the over-representation in the short fallow, underlined by the ‘global’ analysis (Fig 7), of Spizellomyces punctatus, a chytrid species that has been suggested to attack and colonises dead endomycorrhizal spores [122]. S. punctatus could also be an indicator of perturbation. Lozupone and Klein (2002) [123] showed that Spizellomyces populations increased in response to disturbance (i.e., after experiencing agricultural cultivation); an observation supporting the aforementioned facts that short fallow constitutes a degraded system. The statistical approaches implemented in our Anaconda package may, thus, help to disentangle and better understand the multiple biological interactions occurring in a given ecosystem, particularly those leading to an over-representation of certain harmful microbes in soil (Fig 9). It can, additionally, participate in defining ‘targeted’ agricultural management practices to control pathogen populations, for instance, here, by regulating A. spirorbis occurrence.

Besides biotic factors, abiotic soil properties can also, directly and indirectly (via influencing other soil organisms), be involved in regulating plant and human pathogens populations in soil [7, 124]. Soil attributes, such as pH, soil moisture, organic matter content, and nutrient availability, can have a role in soil pathogen’s establishment, survival and growth [7, 124]. However, in our study, when significant differences occurred (e.g., pH, organic matter content, carbon content, and C/N ratio), they were mostly between the short fallow and the ecosystem of reference (not with the long fallow). It seems likely that biotic rather than abiotic factors regulate plant and human pathogens in our traditional agricultural system.