Introduction

Given the high diversity of soil microorganisms1,2 and their crucial roles in ecosystem functioning (e.g., nutrient cycling, soil stability and plant community dynamics)3,4, investigating soil microbial communities has become increasingly important, particularly in light of the global change. However, the belowground component of terrestrial ecosystems remains largely overlooked, especially in biodiversity hotspots.

New Caledonia, a subtropical archipelago located in the southwest Pacific (Fig. 1), is renowned for its remarkable biological diversity5,6,7 and recognized as one of the world’s highest priority areas for conservation8 and restoration9. The long-term isolation, along with the large variety of soils that have developed over geological time, are considered as major factors that have contributed to the high richness and endemism rates encountered in this biodiversity hotspot. Soils originated from ultramafic rocks, which cover about one-third ofthe main island (Grande Terre) (ca. 5500 km²), harbour a variety of soil and vegetation types. A previous study we conducted on ectomycorrhizal fungi (symbiotic organisms associated with the roots of plants) in the South of New Caledonia, via a barcoding approach (Sanger sequencing), suggested a high diversity and endemism rate, along with the dominance of the Cortinarius genus in monodominant ultramafic rainforests (i.e., where more than 50% of the trees in the canopy layer belong to a single species)5. In addition to this work, our team carried out environmental DNA (eDNA) metabarcoding investigations of soil fungal and bacterial communities using a high-throughput sequencing technology (Illumina MiSeq)4,10,11,12. Combination of eDNA and next-generation sequencing is a fast and efficient approach that has revolutionised the study of biodiversity across various ecosystems13, particularly in soil microorganisms, which are intrinsically cryptic and largely uncultivable. Our findings revealed changes in microbial phyla and fungal functional groups’ composition along vegetation successions. Subsequently, we proposed biological indicators of succession dynamics and land degradation, the latter referring to the negative trends in land condition resulting from direct or indirect human-induced processes14. We also observed a structure of microbial communities according to the vegetation type5,10,11. Additionally, a geographical effect was strongly suspected, suggesting that each ultramafic Massif could host a unique microbial community10.

Despite the advances made in recent years, these studies on microbial communities in ultramafic soils have been conducted on local scales, with no comparison between the predominant ultramafic soil types, particularly the two most widespread ones: iron crust and lateritic soils. Iron crust soils are composed of ironstones and gravels, and are empirically considered as harsh environments15. To date, no large-scale approach has been undertaken to ascertain whether local observations can be generalized, or not. This gap hinders our comprehension of the ecology of soil microorganisms in this well-renowned biodiversity hotspot. In this study, we thus conducted the first large-scale analysis of fungal and bacterial communities in ultramafic soils of New Caledonia. Both published4,10,11,12 and unpublished soil eDNA metabarcoding data (generated by Illumina MiSeq sequencing) from our own work were combined here. A range of ultramafic sites from the northern to the southern regions of Grande Terre, harbouring diverse vegetation types developing on iron crust and lateritic soils, as well as a non-ultramafic site on the island of Maré (used as an outgroup) (Fig. 1; Table 1, Fig. S1), were employed to address the following questions: (1) Is there differences in microbial diversity between iron crust and lateritic soils? (2) Are phyla and functional groups, previously identified at a local scale as markers of ecological succession and land degradation, relevant to ultramafic systems on a broader scale? (3) Is the high diversity and endemism rate in ectomycorrhizal fungi, along with the dominance of Cortinarius in monodominant rainforests previously observed, generalizable at a larger scale and on other vegetation types? And finally, (4) is there a geographical structure of soil microbial communities, with each investigated ultramafic massif exhibiting its own fungal and bacterial community?. This pioneering work provides a unique insight into the microbial dynamics of ultramafic soils and paves the way for adapted conservation strategies to this biodiversity hotspot.

Fig. 1
figure 1

(A) Location of New Caledonia in the southwest Pacific and (B) locations of the five studied ultramafic sites on the main island (Grande Terre). The non-ultramafic site (gibbsic Ferralsol) at Maré Island, used as an outgroup, is also presented. Ultramafic substrates are shown in grey. Numbers in superscript indicate the previously published research articles from which data are sourced; 1: Gourmelon et al. (2016); 2: Demenois et al. (2020); 3: Fernandez Nuñez et al. (2021); 4: Stenger et al. (2025); 5: this study).

Full size image
Table 1 Vegetation formations and soil types at the different study sites.
Full size table

Results

Soil microbial diversity

Among the 4281 fungal ASVs and the 1633 bacterial ASVs, only 117 (2.7%) and 32 (2.0%) were common between ultramafic and non-ultramafic soils (Fig. S2). Specific ASVs to ultramafic substrates were 3646 for fungi and 1476 for bacteria (Fig. S2).

Comparisons among all ultramafic sites revealed that the diversity indices for fungi were significantly higher at Goro and Tiébaghi sites (Fig. 2; Table 2), where the studied vegetation occurs on iron crust soils. For bacteria, the diversity indices were overall higher at the Goro site than the other ultramafic sites (Fig. 2; Table 2).

Fig. 2
figure 2

Mean and standard deviation of soil fungal (A) and bacterial (B) diversity indices calculated for the different vegetation types and sites. Significative results from the Generalized Linear Mixed Model (GLMM) are highlighted in bold and p-values with codes: “***” 0.001, “**” 0.01, “*” 0.05. The model used the Formation as the fixed parameter and the Site as the random parameter. Reference was fixed as Arillastrum gummiferum-dominated rainforest (AgF) formation; see Table 1 for plant formation details.

Full size image
Table 2 Estimate values from the generalized linear mixed model (GLMM) for diversity indices.
Full size table

Soil microbial phyla composition

Taxonomic assignment led to the identification of 62% of soil fungal ASVs and 98% of soil bacteria ASVs (after discarding sequences identified as Archaea) at the phylum level. A substantial number of ASVs remain unidentified at lower taxonomic levels, underscoring the presence of potentially novel or understudied fungal and bacterial taxa within these environments. Most fungal ASVs (i.e., 89%) belonged to Ascomycota and Basidiomycota and represented the majority of the reads (i.e., 86%). Both phyla, in variable proportions, dominated in almost all vegetation types (Fig. 3A; Table S1). The taxonomic classification of fungal ASVs realized using a Bayesian model to compare the composition between sites and vegetations showed that among Basidiomycota, ASVs belonging to the Agaricomycetes class were the most represented, with the Agaricales one of the most present order (Table S2). Within this order, members of the Cortinariaceae and the Cortinarius genus appeared frequently. The Ascomycota ASVs were primarily distributed across the Eurotiomycetes, Leotiomycetes, Dothideomycetes, and Sordariomycetes classes. Strikingly, the Mucoromycota was also observed as dominant phylum at the Rivière Blanche and Kopéto sites in the first stages of the succession (Fig. 3A, Table S1). Within this phylum, well-represented ASVs were ascribed to Bifiguratus adelaidae species and the GS23 order. Indeed, 34 and 60% of the reads, as well as 9 and 98 ASVs out of the 138 Mucoromycota ASVs delineated, were assigned to B. adelaidae and the GS23 Clade, respectively (Table S3).

Regarding bacteria, Proteobacteria, Verrucomicrobia, Acidobacteria, Actinobacteria, Chloroflexi and Planctomycetes were found to be the most abundant phyla in ultramafic soils (Fig. 3B; Table S4). The relative abundance of these phyla varied among sites and vegetation types. Supporting these observations, the taxonomic classification of bacterial ASVs, based on a Bayesian model, which highlights specific taxa according to environmental conditions and sites, revealed the prominence of ASVs classified within these phyla (Table S5). Strikingly, a high proportion of Chloroflexi was observed in open vegetation, namely in the open low maquis and the revegetated sites at Goro and the sedge-dominated maquis at Rivière Blanche and Kopéto. In contrast, Proteobacteria seemed to be relatively more abundant in closed vegetation (i.e., within closed maquis and rainforest) (Fig. 3B; Table S4). Similarly, despite its low representation in all plant formations, Cyanobacteria were more represented within closed vegetation, except in the Nothofagus rainforest at Rivière Blanche (Table S4).

Interestingly, at the Tiébaghi site, Actinobacteria appeared more present than in the other sites (Fig. 3B; Table S4), with three actinobacterial genera (with no assignation at the species level) found in high proportions and displaying the highest number of ASVs: Acidothermus, Conexibacter and Mycobacterium (see Table S6). Regarding the non-ultramafic formations, the patterns obtained were characterized by the large occurrence of Firmicutes, a phylum observed in relatively very low abundances in ultramafic soils (Fig. 3B; Table S4).

Fig. 3
figure 3

Relative rarefied abundances of soil (A) fungal phyla and (B) bacterial phyla, within the different vegetation types studied at the five ultramafic sites and the non-ultramafic site used as an outgroup.

Full size image

Soil fungal and bacterial functional groups

In total, 1594 ASVs (ca. 37%), out of the 4281 fungal ASVs detected, were assigned to functional groups (i.e., groups of fungal species exploiting resources in the same manner12,13). Nineteen guilds were considered in our study (see Materials and Methods) (Fig. S3A; Table S7). Ectomycorrhizal and undefined saprotrophic fungi represented the most abundant functional groups (Fig. S3A). Ectomycorrhizal fungi were found in three clusters (Fig. S3A). These clusters correspond to vegetations where ectomycorrhizal trees or shrubs were dominant, i.e., in rainforests dominated by N. aequilateralis or A. gummiferum and maquis dominated by Tristaniopsis species, as well as in three plant formations at Goro plateau, i.e., the closed low maquis, the G. deplancheanum-dominated maquis and the preforest maquis (Fig. S3A and Table S7).When looking at the changes along the vegetation successions, or conversely, along the land degradation gradient, no clear tendency was observed (Fig. S3A).

For bacteria, ASVs mostly belonged to the sulfate reducer, dehalogenation and ammonia oxidizer metabolism categories (Fig. S3B; Table S8). As for fungi, no significant changes were observed along the successions.

Ectomycorrhizal diversity

Regarding the diversity of ectomycorrhizal fungi, 317 ASVs were delineated over the entire dataset. From these, only 41 showed matches to described species, which correspond to 23 known fungal species (several ASVs can be ascribed to a same taxon) (Table S9). These fungi have been recorded from regions outside New Caledonia, notably Australia, New Zealand, Indonesia and Thailand; some of them are also encountered in the northern hemisphere. Thus, based on our data, 276 ASVs, equivalent to 87%, have so far only been recorded in New Caledonia. These 317 ASVs belonged to 33 distinct genera (Figs. 4, 5). Cortinarius was the most species-rich ectomycorrhizal genus, with 116 ASVs detected in total (36.6%) (Fig. 5). The over-representation of this taxonomic group was observed in several vegetation types, in terms of species richness and relative abundance (Fig. 4; Tables S10 & S11). In particular, Cortinarius members were well-represented in N. aequilateralis rainforests and Tristaniopsis maquis from sites located in the south (i.e., the Rivière Blanche site) and the north (i.e., the Kopéto site) of New Caledonia, This was also the case for the mixed rainforest at Kopéto and the three closed vegetation types at the Goro site (i.e., from the closed to the preforest maquis). The importance of this genus is also supported by the Bayesian analysis (Table S2). In terms of total richness, the G. deplancheanum-dominated maquis exhibited the highest value (with 76 ectomycorrhizal fungal ASVs), followed by the two monodominant rainforests (A. gummiferum and N. aequilateralis) and the Tristaniopsis-dominated maquis (Table S12).

Fig. 4
figure 4

Barplots of the (A) relative abundance and (B) relative number of ASVs of ectomycorrhizal fungi in the different locations and vegetations studied.

Full size image
Fig. 5
figure 5

Number of ectomycorrhizal fungal species (ASVs) for each of the 33 detected genus from the entire ITS2 dataset. The number of delineated ASVs within each genus is indicated at the top of the bar.

Full size image

Soil microbial community structure

The bipartite network on soil fungi (Fig. 6A) revealed a geographical clustering of communities, with Goro and Tiébaghi sites being particularly distinct from the other locations. In addition, fungal communities were also structured by the vegetation type (Fig. 6A). Plots in monodominant rainforests of N. aequilateralis from Rivière Blanche and Kopéto were closely related. The same observation was made for the maquis dominated by Tristaniopsis species at these two sites. These observations were corroborated by the results obtained from the partition of the fungal network into communities (Table S13). Similar tendencies were observed for bacteria, even if less evident (Fig. 6B; Table S14). NMDS visualisations of the three dissimilarities indexes calculated, i.e., Bray-Curtis for both fungi and bacteria (Fig. S4) and unweighted and weighted UniFrac indexes for bacteria only (Fig. S5), and PERMANOVA analyses on Bray-Curtis dissimilarity for the combined effect of sites and vegetations corroborate these observations (fungi: 35.77%, F = 1.9183, p-value < 0.001; bacteria: 34.57%, F = 1.9186, p-value < 0.001). Furthermore, the results from the betadisper analysis using the Bray-Curtis dissimilarity matrix, revealed significant differences in dispersion for both sites and vegetation types. For ITS, dispersion differed significantly between sites (F = 12.103, p-value < 0.001) and vegetation types (F = 5.129, p-value < 0.001). Similarly, for 16S, dispersion differed significantly between sites (F = 12.674, p-value < 0.001) and vegetation types (F = 3.142, p-value < 0.001). These findings suggest that variation in group dispersions may partly contribute to the observed differences in microbial community composition identified through the PERMANOVA.

Fig. 6
figure 6

Bipartite networks performed on soil (A) fungal and (B) bacterial communities. Coloured nodes correspond to the plots positioned at different sites and vegetation types in New Caledonia. Sizes of plot nodes are proportional to the number of edges they are connected to, so here, to their ASV richness. Black nodes represent the ASVs (i.e., molecular sequence variants). For ASV nodes, they are proportional to the number of plots they are connected to. Letters in nodes correspond to the site name (the same name can thus appear several times).

Full size image

Discussion

A higher microbial diversity on iron crust soils

Surprisingly, fungal species richness was higher in iron crust soils (Goro and Tiébaghi sites). These soils are composed of ironstones and gravels and are, subsequently, empirically thought to harbour more stressful conditions than lateritic soils15. However, nickel content is known to gradually decrease from the bottom to the top along the weathering profiles of soils from ultramafic rocks16. Considering their lower nickel content and, subsequently, the potential lower nickel bioavailability, iron crust soils may thus represent less restrictive environments regarding this parameter than previously thought. One hypothesis would thus be that limitation in nickel constraints may promote fungal diversity. However, to the best of our knowledge, no studies have been performed concerning the constraints that iron crust soils have on living organisms, nor has a comparison been made with other soils originating from ultramafic rocks (e.g., in terms of nickel and the bioavailability of other heavy metals, water drainage and temperature). Another hypothesis would be an adaptation to iron crust soils, leading to fungal speciation and diversification. Further, we cannot also exclude a plant composition effect, since maquis on iron crust soils are distinct vegetation types compared to others17.

For bacteria, greater species richness was only observed in iron crust soils at the Goro site. A previous comparison18 of the molecular diversity of the symbiotic nitrogen-fixing bacteria genus Frankia from the root system of two Gymnostoma species (Casuarinaceae), endemic to New Caledonia, revealed that G. deplancheanum hosted a greater diversity of Frankia than G. chamaecyparis. G. deplancheanum occurs on iron crust soils, whereas G. chamaecyparis is found on hypermagnesian soils that develop at the base of ultramafic massifs. A greater selective effect of hypermagnesian soils on bacteria was given as a probable explanation18. This is consistent with our hypothesis of iron crust soils being less constraining in nickel content, but a vegetation composition effect must also be considered. Indeed, G. deplancheanum is a major component of the maquis in the South of the main island and could thus drive bacterial community diversity in this area. The highest values of fungal richness being observed in the G. deplancheanum-dominated vegetation suggest that the presence of this Casuarinaceae species may also locally contribute to increased soil fungal diversity, through direct (e.g., beneficial mutualistic interactions) or indirect (e.g., litter composition) effects.

Soil microbial phyla composition as markers of ecological succession, land degradation and geographic origin

Beyond variations in soil microbial diversity, differences in relative abundances of phyla among vegetation types were observed. Ascomycota and Basidiomycota, the two most abundant fungal phyla in the large-scale dataset, varied in their proportions, with Ascomycota being relatively well-represented in the first stages of the plant successions, especially at the Goro site (open low maquis), as also found by Fernandez Nuñez et al.11. At the Rivière Blanche and Kopéto sites, Ascomycota was also well present in the open vegetation types (i.e., sedge and Tristaniopsis formations); nevertheless, the pattern was not as strong as that detected in Gourmelon et al.10. Interestingly, we found that at these two sites, Mucoromycota were more prevalent in the first stages of the succession and decreased along the succession. Among Mucoromycota taxa, ASVs assigned to Bifiguratus adelaidae and GS23 clade were detected as the most common in terms of relative abundance and richness. B. adelaidae has been recovered mostly from soil and to a lesser extent as a plant endophyte19. Despite the broad geographic distribution and abundant site occurrence of this novel taxon, its functional roles are still unknown. The GS23 clade, to which most of Mucoromycota reads and ASVs were assigned, forms a new monophyletic lineage from tropical and subtropical acidic forest soils (2.5 to 4.0)20. Recently, fungal isolation and identification from plant roots in acidic and oligotrophic soil of northeast America led to the description of two new species, belonging to a new genus and a new family, Pygmaeomycetaceae, proposed to correspond to the Clade GS2321. The functions of Pygmaeomycetaceae members are not yet fully understood, however, Walsh et al.21 suggest that these fungi may be capable of degrading diverse substrates, allowing mobilization of nitrogen and subsequently contributing to the success of their associated host plants in acidic and nutrient-poor environments. Such a hypothesis suggests that ASVs belonging to the GS23 Clade are root symbionts that strongly influence ecosystem functioning at the two distant sites that are Rivière Blanche and Kopéto, especially in vegetation with poor plant coverage and soil nutrition. This reinforces the necessity of determining to which functional groups (in terms of guilds and trophic modes) these Mucoromycota members belong to. Their unknown functionality may explain why no clear trend in fungal functional groups was detected across the successions, or conversely, along the land degradation gradients, at least at Rivière Blanche and Kopéto sites. In addition to this, akin to Ascomycota, high relative abundances of Mucoromycota may be an indicator of land degradation10 as well as a sign of early stages of an ecological succession11. However, we do not know whether the prevalence of this GS23 Clade in ultramafic substrates is mainly restricted to lateritic soils in general or linked to particular vegetation types.

Similarly, the relative abundances of soil bacteria varied between vegetation types. These discrepancies were overall due to the higher relative abundances of Chloroflexi in the open vegetation and Proteobacteria in the closed vegetation. Despite their low proportions, Cyanobacteria were also mostly encountered in closed maquis and rainforests. These results allow us to generalize the findings made at a local scale by Fernandez Nuñez et al.11, regardless of the type of ultramafic soil (i.e., iron crust soils versus lateritic soils). Indeed, the higher proportion of Chloroflexi that characterized the open low maquis at Goro11 can be extended to the sedge-dominated maquis, two ecosystems characterized by a more or less herbaceous layer with sparse shrubs, which results in lower plant coverage compared to other vegetation types (approximately 10% for Goro’s open low maquis and 30–80% for the sedge maquis). The large relative occurrence of Proteobacteria, as well as the increased presence of Cyanobacteria, previously highlighted from Goro’s closed vegetations11, were also found in most of the closed maquis and rainforests. The prevalence of Chloroflexi and Proteobacteria, in three out of the four successions investigated in our analysis (i.e., at Goro, Rivière Blanche and Kopéto sites, but not at Tiébaghi), may be explained by the availability of nutrients in soil along ecological successions: Chloroflexi suspected as oligotrophs11,22, survive in nutrient-poor environments, while Proteobacteria, usually copiotrophs23, prefer nutrient richer substrates.

Another striking finding was the high relative abundance of bacteria belonging to the Actinobacteria phyla at Tiébaghi site independent of the vegetation type. Acidothermus, Conexibacter and Mycobacterium were by far the most dominant taxa in terms of relative abundance and ASVs richness. The Acidothermus genus contains to date only one described species, A. cellulolyticus, that has been isolated and characterized from acidic hot springs24, and amplicons assigned to this genus have since been recovered from acidic soils from China25,26. The complete genome sequencing of A. cellulolyticus revealed the presence of several genes encoding for enzymes involved in the breakdown of plant and fungal cell wall components27. We, therefore, hypothesise that Acidothermus species possess the ability to use a range of carbon sources and could play a significant role in the degradation of biomass and carbon cycling. For the second most abundant actinobacterial genus, Ma et al.28. recently found that Conexibacter may be involved in carbon and nitrogen biogeochemical cycles in natural forests of eastern China. Both genera may therefore play a central role in nutrient cycles at the Tiébaghi site. In addition, the slow-growing and high genomic GC content observed in Conexibacter strains are suggested as potential traits that could favour the survival of this group in stressful environments29,30. It is not worthy that besides the Tiébaghi site, members of this taxonomic group were mostly recovered from Goro’s iron crust soils. The third most abundant genus was Mycobacterium which is usually positively related to the prevalence of certain iron minerals in soil31,32. Overall, the over-representation of these three genera in iron crust environments supports the interest of further characterizing and comparing environmental constraints of distinct ultramafic soils, as well as identifying the key microbial taxa involved in soil biogeochemical cycle regulation.

Focus on ectomycorrhizal fungi

A remarkable finding was the considerable proportion of ectomycorrhizal fungal ASVs that did not find matches at the species level to the existing databases. Out of the 317 delineated ASVs only 41 were assigned to known species (more exactly to 23 described species; several ASVs can be assigned to one species). Based on this dataset, this would indicate a hypothetical endemism rate of 87% in New Caledonia. This result, obtained from a soil eDNA and high-throughput sequencing approach, is consistent with the 95% calculation made by Carriconde et al.5 on ectomycorrhizas and fruit bodies typed by Sanger sequencing. New Caledonia clearly hosts a high and unique ectomycorrhizal fungal diversity. This raises questions about the underlying evolutionary and ecological mechanisms that have led to this exceptional and exclusive diversity in this remote territory. The New Caledonian biodiversity is also facing various severe threats (e.g., open cast mining, pollution, frequent and extended wildfires and invasive species introduction) and fungi, as all other living forms, are prone to extinction risks. This high endemism rate points out the effort that should be made to identify fungal species under threat and establish on-the-ground species-oriented conservation plans, an aspect that has never been considered and addressed in this biodiversity hotspot.

Concerning the fungal ectomycorrhizal taxonomic composition, the Cortinarius genus was over-represented in numerous ultramafic vegetation types, corroborating the observations made by Carriconde et al.5. on ectomycorrhizas and fruit bodies collected in N. aequilateralis, A. gummiferum and mixed rainforest stands in the south of New Caledonia. With respect to the generalization of this pattern on a larger geographical scale and to other vegetation types than rainforests, it is likely that ultramafic rainforests and shrublands dominated by ectomycorrhizal plant species are characterized by the prevalence of Cortinarius. Nevertheless, despite the limited number of samples, Cortinarius was not the main group in ectomycorrhizas collected in ultramafic substrates from Acacia spirorbis roots33, a shrub legume that can dominate the ecosystems where it occurs. Future in-depth examinations must be undertaken to confirm or not the pre-eminence of Cortinarius in ecosystems dominated by a single plant species in New Caledonia. Such dominance by this fungal ectomycorrhizal group has direct implications in terms of ecosystems functioning: it appears that Cortinarius has retained from their saprophyte ancestors the capacity to degrade organic matter for potentially mobilizing nitrogen34,35, a limited element in ultramafic soils5.

Finally, it is noteworthy that, in addition to the high relative abundances of ectomycorrhizal fungi in ultramafic vegetation dominated by ectomycorrhizal trees (N. aequilateralis or A. gummiferum) in forests and shrubs (Tristaniopsis spp.) in maquis, this functional group was also prevalent in Goro’s closed vegetation, where only sparse individuals of known ectomycorrhizal plants have been recorded. Isolated ectomycorrhizal plants are unlikely to account for such high ectomycorrhizal fungal abundance and richness (G. deplancheanum-dominated vegetation harbouring the highest ectomycorrhizal fungal richness). As stated by Fernandez Nuñez et al.11, these observations clearly highlight the need to determine the mycorrhizal status of the New Caledonian flora, with only 14 plant species inventoried as ectomycorrhizal to date. This will improve our understanding of the extent to which this mutualistic association influences the ecosystems functioning in this extraordinary hotspot of biodiversity.

A geographical structure of soil microbial communities

Finally, in our analysis, structure investigations (through PERMANOVA analyses, bipartite networks and communities’ partitioning) revealed a geographical clustering of soil microbial communities, a result in accordance with the initial work of Gourmelon et al.10 undertaken at two distant massifs (Rivière Blanche and Kopéto). For fungi, the two studied iron crust sites, Tiébaghi and Goro, were particularly apart from the other locations and distinct from each other as well. For bacteria, only Goro appeared clearly different. Overall, such geographical structure seems to indicate that each site exhibits its own microbial community. Ultramafic outcrops are patchily distributed and represent edaphic islands. This spatial discontinuity could have led to geographical (allopatric) speciation, contributing to this species diversity36. Soil microorganisms, like plants37, may thus display microendemic patterns.

In terms of soil microbial conservation, this geographical structure has a direct implication: it indicates that each ultramafic massif might be considered as a conservation unit by itself, especially for the isolated massifs such as Tiébaghi and Kopéto. For the “Massif du Grand Sud”, with a surface area of 3015 km², making it the largest continuous ultramafic outcrop in the world38, different hotspots of fungal and bacterial diversity likely exist and may result in many priority areas for soil conservation. Distinct and unique microbial communities may also be hosted by the other types of soils encountered in New Caledonia. Samples from Maré island collected on aluminic-rich soils (gibbsic Ferralsols) were, indeed, completely dissimilar from ultramafic soils. The interest in focusing on soil biota for establishing priority areas has been very recently spotlighted by Guerra et al.39 at the Earth scale. However, despite the input of this work, the approach remains imprecise, with many regions being not considered, including well-known biodiversity hotspots like New Caledonia. A high-resolution investigation of microorganisms, and other soil-living organisms (e.g., protists, nematodes, earthworms, and arthropods), using high-throughput sequencing technologies, on a large array of soil types could constitute a next future research challenge. At the scale of this territory and more widely, it will undeniably help to put on the map this still largely neglected hidden biodiversity and will contribute to a more holistic perception of nature conservation.

Conclusion

Our eDNA metabarcoding comprehensive analysis allowed us to depict several global patterns in microbial communities from ultramafic soils in New Caledonia:

  1. 1)

    Iron crust soils harbour a significantly higher fungal diversity than lateritic soils.

  2. 2)

    The composition of microbial phyla shifts along ecological successions. Fungal Ascomycota and Mucoromycota phyla, which were prevalent in the first stages of the succession, may be markers of ultramafic land degradation. Among bacteria, the high relative abundance of Chloroflexi seems to characterize vegetation with limited plant cover, and conversely, both Proteobacteria and Cyanobacteria characterize closed maquis and rainforests.

  3. 3)

    The ectomycorrhizal fungal functional group exhibits a high and unique species richness, with a hypothetical endemism rate of 87%. In addition, the Cortinarius genus was prevalent in ultramafic rainforests and maquis (shrublands) dominated by ectomycorrhizal plants.

  4. 4)

    Finally, each ultramafic Massif seems to display its own soil microbial community.

All these results highlight the relevance of further identifying the evolutionary and ecological mechanisms that have shaped the microbial diversity of ultramafic ecosystems in New Caledonia. Future investigations are also needed to better understand how the dominant groups are involved in the functioning of ultramafic systems, notably in nutrient cycling. In addition, our observations underscore the necessity of describing microbial species and assessing their threat status to establish species-conservation plans. They also support the emerging point of view of considering soil biota for identifying hotspots of diversity and defining conservation priority areas for a more holistic perception of nature conservation. This analysis represents a first step. In the coming years, expanded high-resolution investigations into soil biota, covering a wider range of vegetation and soil types (including volcano-sedimentary and calcareous soils) will provide a deeper insight into how environmental factors, including soil physicochemical parameters, climate and land use, influence soil communities. This would help to define conservation strategies and policies for this biodiversity hotspot. This will notably be achieved through the sensibilization of local populations (of all origins) to soil life and its significance, alongside the consideration of their perception of soil and nature conservation, as well as their involvement in defining priority areas.

Materials and methods

Experimental sites description and soil sampling

This study encompasses, five sites located on ultramafic soils across the main island of New Caledonia (Fig. 1), namely from the south to the north: Goro (22°16’S; 166°58’E), Bois du Sud (22°10’S; 166°46’E), Rivière Blanche (22˚9’S; 166˚41’E), Kopéto (21˚10’S; 165˚0’E) and Tiébaghi (20°27’S; 164°12’E).

The three first sites are present on the worldwide largest continuous ultramafic outcrop38, called the “Massif du Grand Sud”, and the two others on isolated ultramafic Massifs located on the west coast, namely Kopéto and Tiébaghi Massifs. For these sites, 69 plots of 20 × 20 m were set up within 12 ultramafic vegetation types representing different plant successions (Fig. S1; Table 1), with the addition of seven revegetated plots from Goro plateau11 which were grouped under the label revegetated sites in this analysis. Based on Poorter et al.40 definition of succession, a “plant succession” or “vegetation succession” represents, here, changes in plant communities over time following a disturbance (most likely, in our case, after wildfires). Because ecosystem recovery after disturbance requires observations over long time periods that can be beyond human scale (over centuries in ultramafic systems41), chronosequences were used (Fig. S1). A chronosequence is defined as a “space-for-time substitutions using multiple sites with similar starting conditions but of different ages”42). The studied ultramafic vegetations developed on to two types of soils: (i) iron crust soils at Goro and Thiébaghi sites (Fig. 1; Fig. S1), classified as Petroplinthic Plinthosols (Ferritic)43, and (ii) lateritic soils at Bois du Sud, Rivière Blanche, and Kopéto, classified as Posic Ferralsols (Ferritic)43 (Fig. 1; Fig. S1; Table 1).

Forests and yam plantations developing on non-ultramafic soils from the Maré Island (Fig. 1) were used as outgroups and preliminary comparisons were here made to the investigated ultramafic systems. Studied soils from Maré Island developed from exogenous matter, notably volcanic ash and pumice (with gibbsite has major component and very high organic matter content (> 20%))44. These soils are classified as Gibbsic Ferralsols (Humic)43, named Gibbsite and humic soils in Table 1, and were collected at 15 distinct plots.

At each 20 × 20 m plot, 25 soil samples of 10 cm depth and 5 cm diameters were collected and grouped to constitute one sample per plot. All samples were placed at 4 °C and stored at −20 °C within the 4 following hours. In total, 91 plots, i.e. 91 soil samples, were considered in this work.

DNA extraction and next generation sequencing protocols

The next generation sequencing (NGS) data presented here were generated by our research team and come from our four published studies, i.e., Gourmelon et al.10, Demenois et al.4, Fernandez Nuñez et al.11, and Stenger et al.12, supplemented by new unpublished data from Tiébaghi Massif. All these samples were treated in a systematic manner. First, eDNA from all soil samples was extracted using the PowerSoil® DNA Isolation Kit (Mo Bio Laboratories, Carlsbad, CA, USA), according to the manufacturer’s instructions. Fungal communities were targeted by amplifying the nuclear rDNA internal transcribed spacer 2 (ITS2) using the fwd-ITS7 (5’-GTGARTCATCGAATCTTTG-3’)45 and rev-ITS4 (5’-TCCTCCGCTTATTGATATGC-3’)46 primers. For bacteria, the V4 region of the 16S RNA gene was amplified using the primers fwd-515 (5’-GTGCCAGCMGCCGCGGTAA-3’) and rev-806 (5’-GGACTACHVGGGTWTCTAAT-3’)47. Both regions are commonly targeted for fungal and bacterial communities’ investigation and are standard for the Earth Microbiome Project48. Each region was amplified at least twice to reduce PCR biases4,10,11,12. An equimolar pool of PCR products by fragment has been created and used for ITS2 and V4 libraries sequencing. Paired end 2 × 250 nt length fragments were sequenced on an Illumina MiSeq with a V2 reagent kit, following Illumina recommendations. In total, three runs were performed with the Illumina MiSeq technology (Kopéto and Rivière Blanche sites were of the first run; Tiebaghi and Bois du Sud sites were of the second; and, Goro plateau and Maré were of the third) (Table 1). As these were independent studies over time, no samples were split between the three series, with each sample belonging to a single series. As all the generated reads were sequenced with the same kit and on the same Illumina sequencer, we performed a learnErrors test from the DADA2 R package49 to confirm the absence of technical bias (Fig. S6 & S7). This was completed by a clustering analysis with pheatmap R package50 (Fig. S8) that revealed the absence of clustering by sequencing runs. Out of the 91 plots used for this analysis, 26,960,796 ITS2 reads and 29,024,538 V4 reads were obtained.

Bioinformatic pipeline

The Bioindic pipeline (freely available at https://gitlab.com/IAC_SolVeg/CNRT_BIOINDIC), based on QIIME251,52 and DADA249 tools, was used to generate a table of amplicon sequence variants (ASVs). This pipeline allows the quality of raw reads to be filtered, by removing PCR primers and sequencing adapters prior to the use of QIIME2 plugins. After the quality filtering steps, which are performed on a per-sample basis, the samples were checked with the DADA2 learnErrors model to explore potential sequencing errors specific to each sequencing run. The learnErrors model uses a subset of the sequences present in the samples to estimate the error rate. This process was repeated 20 times to ascertain the deviation due to sampling. Then, models were performed by sequencing runs and with all runs together (Fig. S6 and S7). For all models, the estimated error rate and the observed error are well-covered (Fig. S6 and Fig. S7), and no deviation was observed. As the samples were sequenced using the same technology, the model worked well with all the samples. The reliability of the estimated model was confirmed, and it was subsequently utilised for the remainder of the pipeline, where the NGS data was treated as a single dataset, thereby reducing statistical bias for subsequent steps. Furthermore, to avoid potential noise due to sequencing contamination, we used water samples as negative controls to correct all datasets. Dereplication, sample inference and chimeras’ removal were performed with the DADA2 algorithm and ASVs were taxonomically assigned with the RDP’s naive Bayesian classifier. The resulting ASV samples were rarefied using the beta- and alpha-rarefactions of QIIME2 with 99 iterations, to compare beta diversity between samples, resulting in 1,231 sequences for the ITS2 region and 1,292 sequences for the V4 region, giving a total of 100,942 and 114,988 quality-filtered sequences, respectively, used for further analyses. Due to the very low number of fungal sequences from the sedge-dominated vegetation at Rivière Blanche site, the corresponding plots were excluded from the rest of the analyses of fungal communities (Table 1). Finally, out of the 91 plots, 81 plots were further retained for fungi and 89 plots for bacteria (Table 1).

The reference database used for the taxonomic assignment for bacteria was Silva v132 for 16S (SILVA-V132-2018.04.10–99 QIIME pre-formatted database with 99% homology clustering), including 412,168 sequences53. Reference database used for the fungal taxonomic assignation (ITS2 ASVs) was UNITE (UNITE-V7.2-2017.10.10-dynamic QIIME pre-formatted database with dynamic homology clustering), including 8,756 reference sequences and 21,793 representative sequences (RepS)54, combined to an in-house database of 311 New Caledonian ectomycorrhizal reference sequences5. Species level assignments were performed based on exact matching with sequenced reference strains according to 99% of homology clustering.

Network visualisation

Bipartite networks were generated with Gephi v0.9.249 using ForceAtlas2 network layout55. ASVs abundance matrices were used, with plots and ASVs as nodes and ASVs abundances as weighted edges. Partition of networks into communities (a non a priori approach) was carried out using Blondel et al.56 algorithm and Lambiotte et al.57 resolution.

Functional assignments

Fungal functional assignments were undertaken using the FUNGuild database58. Regarding the identification of the fungal functional groups, as in Fernandez Nuñez et al.11, “Possible”, “Probable” and “Highly Probable” confidence assignments were kept. Fungal functional groups refer to groups of species, related or not, that exploit the same class of environmental resources in a similar way58,59. ASV assigned to a unique trophic mode (i.e., Pathotroph, Symbiotroph and Saprotroph) and to the well-defined Ericoid Mycorrhizal guild (identified as Pathotroph-Symbiotroph in FUNGuild) were considered for further analyses. Bacterial functional assignments and statistical analysis were realized using the Metagenassist60 web server, and metabolism categories were retained. Both fungal and bacterial functional assignments were clustered and plotted using the pheatmap R package50, on ASVs abundances.

Statistical analysis

Diversity of soil microbial communities were assessed by calculating species richness (i.e., the number of observed ASVs), Hill-Shannon-entropy and Hill-inverse-Simpson, using vegan61 and hillR62 R packages. Diversity indices were analyzed with a Generalized Linear Mixed Model (GLMM) using the glmmTMB R package63. The model used the Formation as the fixed parameter and the Site as the random parameter. Reference was fixed arbitrarily as Arillastrum gummiferum-dominated rainforest (AgF) formation (see Table 1 for plant formation details). The distribution of each indices was checked and fixed as negative binomial in the GLMM model.

Similarity between communities was assessed with the Bray-Curtis dissimilarity index followed by a PERMANOVA analysis (vegan61 and pairwiseAdonis64 R packages) with 9999 permutations on the combined effect of plant formation and site. The increased number of permutations helps to improve the accuracy of p-value estimation. For routine analyses, 9999 permutations are sufficient; beyond that the gains in precision are often negligible in relation to the computation time required. To assess whether the assumption of homogenous dispersion was met for PERMANOVA, we conducted a betadisper analysis using the Bray-Curtis dissimilarity matrix, with the betadisper function in the vegan R packag61. For this, a Bray-Curtis dissimilarity matrix was calculated from the ASV abundance data for both ITS2 and V4 datasets. Dispersion homogeneity was tested across groups defined by site and vegetation type. Statistical significance was determined using an analysis of variance (ANOVA) applied to the distances to group centroid.

The unweighted and weighted UniFrac dissimilarities, that take in consideration phylogenetic distances, were also computed using QIIME2 plugins. Based on the dissimilarity indexes, non-metric multidimensional scaling (NMDS) was used to represent fungal and bacterial dissimilarities between plots. Graphics were realized with the ggplot2 R package65.

Statistical analyses on the phylogenetic composition of the datasets to search for groups that would be significantly represented in a particular site or condition, were performed by Bayesian analyses using the R package Anaconda12,66 Differential groups were kept according to an adjusted p-value cutoff of < 0.05 and a LogFoldChange > |2|.