Amplicon sequencing and metagenomics depict similar pictures of the structure of bacterial and fungal communities
The structures of bacterial and fungal communities obtained with amplicon sequencing25 and metagenomics displayed similar patterns in both control and amended treatments regardless of land-use history (Fig. 1). In addition, these two approaches provided similar results in terms of reproducibility, with the three field replicates showing close similarity (Fig. 1; Supplementary Fig. 1). Control grassland and cropland communities were discriminated on the factorial map, evidencing the impact of land-use history on the diversity of soil microbial communities25,26. In addition, while bacterial and fungal communities did not show any significant differences in their dynamics between the four sampling dates in the control, highlighting their stability along the seasons, the input of wheat residues strongly impacted these two communities, regardless of land-use history (Permutational Multivariate Analysis of Variance [PERMANOVA], pvalue < 0.001) (Supplementary Fig. 1). Thus, a strong population shift was observed after three days for both fungi and bacteria (Fig. 1). In the latter phase (days 51 and 125), bacterial communities but not fungal communities were resilient (Fig. 1).
PCoA of communities separated by taxonomic group. Both metabarcoding (a) and metagenomics (b) separate communities according to the land-use history, and bacterial and fungal communities display similar patterns with both approaches. (a) PCoA of metabarcoding CLR transformed data. (b) PCoA of metagenomics CLR transformed taxonomic data. Green samples denote grassland, yellow crops, pale colours controls, deep colours amended samples. The percentage of explained variability for each axis is indicated in square brackets. The p-values of PERMANOVAs computed between cropland and grassland samples are indicated in parentheses on Axis.1.
Compared to Bacteria and Fungi, patterns of archaeal communities differed between the two sequencing approaches (Fig. 1). This discrepancy could be due to the lack of archaeal sequence recovered in the amplicon sequencing dataset. For this specific group, metagenomics proved to be far more resolutive than the metabarcoding dataset (probably due to its limited 4000 bacterial and archaeal sequences) and was able to detect a significant difference between cropland and grassland archaeal communities with PERMANOVA and Redundancy Analysis (RDA) (Supplementary Fig. 1 & Supplementary Table 1).
When looking at the relative abundance of the most abundant phyla, there was no major difference for bacterial and fungal phyla between amplicon sequencing and metagenomics (Supplementary Fig. 2), except for Cyanobacteria and Basidiomycota that were underrepresented in the amplicon sequencing datasets. In contrast, Becker and Pushkareva20 found that Cyanobacteria were overrepresented with their amplicon sequencing primers. The poor detection of Cyanobacteria in the amplicon sequencing dataset was also in apparent contradiction with previous in silico analyses27 of the PCR primers used in25, which predicted high hit frequency for Cyanobacteria. This could be explained by the change in taxonomic classification since the in silico analysis or by PCR biases in the in vivo application. This illustrates the potential drawback of amplicon sequencing whose representativeness for specific groups is highly dependent on the primers used and the PCR conditions. Moreover, thanks to its considerable depth of coverage, metagenomics identified many low abundance bacterial phyla that could not be detected with amplicons, among which were numerous candidate phyla (Supplementary Fig. 3).
At a finer taxonomic level, we compared the variations of raw count abundances of genera previously highlighted in25 during the experiment, with those obtained with the global metagenomic approach (Supplementary Fig. 4). For bacterial genera, the abundance patterns were consistent between the two methods (Supplementary Fig. 4a. and c.), with genera such as Pseudomonas and Massilia highly stimulated by straw amendment, regardless of land-use history. However, metagenomics appeared more sensitive than amplicon sequencing for some genera such as Bulkholderia and Lysobacter. Indeed, it showed a significant influence of straw amendment in both land-use histories for these taxa, whereas amplicon sequencing only detected a significant influence in one soil management history (Supplementary Fig. 4a. and c.). For fungal genera, both approaches also exhibited close abundance profiles (Supplementary Fig. 4b. and d.). Again, metagenomics appeared more sensitive than amplicon sequencing for certain genera such as Rhizopus and Mucor, for which hardly any sequences were detected in the control samples with amplicon sequencing. This could be explained easily by the much higher sequencing depth in metagenomics than in amplicon sequencing, as previously suggested20.
Overall, these results provide evidence that amplicon sequencing and metagenomics identified similar taxonomic patterns at the community, phyla and genus levels, highlighting that metagenomics is suitable for assessing the dynamics of the soil microbial communities in situ. Moreover, metagenomics was more sensitive than the amplicon dataset, showing the huge potential of this approach for the fine analysis of population dynamics. The similarity of the results observed between the two approaches for bacteria and fungi makes us confident about the robustness of the patterns observed for the other microbial groups (i.e., Viruses and protists) through metagenomics.
Metagenomics revealed the temporal response to wheat input of the whole soil microbiota
Metagenomics gave access to the whole soil microbiota, including Viruses, which is difficult to achieve with amplicon sequencing28. As with Archaea, Bacteria and Fungi, viruses and protists presented distinct structures between grasslands and croplands (Fig. 1b., PERMAVOVA, p-value < 0.001), showing a strong influence of land-use history. In addition to determining the structure of all microbial groups, land-use history also impacted the complexity of the whole soil microbial community, as evidenced by the results of the multigroup cooccurrence network analysis that revealed an increase of the total number of links and connectivity in cropland control compared to grassland control (Fig. 2). This result contradicts previous reports from national29 and regional scale studies30 that showed higher complexity of microbial networks in semi-natural systems such as forest and grassland compared to cropland. However, it must be kept in mind that our study deals with the effect of land-use history rather than with the land-use itself, since plant cover was removed and the soil was left bare by manual weeding for 5 months to stabilise before applying wheat residue inputs (see material and methods section). It is likely that such elimination of plant cover may have deeply affected the complexity of the soil microbial network, in agreement with31 who observed a high overall network complexity in the rhizosphere of oat that increased as plants grew. In other respects, this highlights the importance of plant cover for stimulating the complexity of soil microbial interactions and suggests that this effect is transient and disappears quickly following plant removal. In absence of plant cover, the lasting differences in soil properties (i.e. lower pH and phosphorous content; higher values of SOM, soil organic carbon content, total nitrogen, Soil C/N and cation exchange capacity in grassland soil) probably accounted for a large part of the microbial community’s differentiation between grassland and cropland soils (25).
Cooccurrence networks statistics according to treatment and land-use history. Complexity of the whole microbial communities was impacted by land-use history and amendment, with more complexity in the cropland soil than in the grassland soil and a significative increase in complexity for both land-uses after wheat straw amendment. (a) Number of links. (b) Connectance (proportion of links formed out of all possible links). Kruskal Wallis tests with Bonferroni corrections (alpha 0.05) were performed on the metrics to assess their difference between different treatments and land-uses.
As observed in the previous section for Bacteria and Fungi, wheat residue input strongly impacted the structure of Viruses and protists (Fig. 1 & Supplementary Table 1). Interestingly, it also significantly increased the complexity of the whole soil microbial community, as evidenced by the higher number of links and higher connectance of the multigroup microbial co-occurrence networks observed in both land-use histories (Fig. 2a, b.). This increase in the complexity of the soil microbial community by crop residue input had been reported previously for Bacteria and Fungi32, but to our knowledge, this is the first time it has been evidenced at the whole soil microbial community level. It points to the overall high network complexity that increases as C-inputs feed the soil microbiota, reflecting extensive interactions such as mutualism, competition, predation and parasitism that occur among microbial groups during the decomposition process.
The rapid and highly significant changes in the structure of both bacterial and fungal communities on day 3 after straw input (Fig. 1) reflected that not only bacteria but also fungi can act as pioneer decomposers of wheat straw25,33. The resilience of bacterial populations in the late phase of straw decomposition, where easily decomposable C-substrate was depleted, could be attributed to the decrease of copiotrophic bacteria populations (Fig. 1). In contrast, the different trajectories of the fungal communities implied that other fungal populations rose at days 51 and 135. Indeed, in this latter stage of straw decomposition, mainly recalcitrant organic carbon remained, thus fungal oligotrophs were likely preponderant, as was suggested before25,33 due to improved abilities to decompose more complex C-substrates compared to bacteria.
In amended plots, viral communities experienced transient but strong modifications on day 3 (Supplementary Fig. 1). Since Viruses are obligate parasites, such modifications can result only from the increase of their hosts abundance. Indeed, Virus would multiply through lysogenic or lytic infections. In oceans, viral lysis is estimated to turnover ~ 20% of microbial biomass every day and is a major actor of biogeochemical cycles by liberating nutrients and carbon34. In soil, Viruses have been studied less so far and very little is known about their roles in regulation of other microbial groups and carbon decomposition35. In our study, their short-lived response following wheat straw input might indicate a strong link with copiotrophic populations. They might be in part responsible for the rapid resilience of bacterial population structures. As such, these results suggest that Viruses could act as moderating agents by targeting the most active populations of soil microorganisms in a “kill-the-winner” strategy36.
As for the other microbial groups; the wheat residue input still had a clear impact on the structure of the protist community on day 3, suggesting that copiotrophic protists (or perhaps predators of copiotrophic microorganisms) might multiply quickly following C-input into the soil. Moreover, we observed a shift of both control and amended communities from the initial state at day 125 (Fig. 1), implying that factors other than amendment might impact this group. It could be, for example, an influence of seasonality, as day 125 samples were collected in January. Indeed, RDA analyses including land use, amendment and meteorological data (i.e. we chose soil temperature at 10 cm depth, as soil humidity was strongly anticorrelated to temperature) demonstrated that the protist community was the only microbial group majorly influenced by climatic conditions (Supplementary Fig. 5 & Supplementary Table 1). These results are in agreement with a global distribution study that evidenced that meteorological data were good predictors of the composition of the protist soil community37.
Microbial heterotrophic successions revealed common and specific patterns according to land-use history
To elucidate the abundance patterns of genera within the reactive part of the community, i.e., genera whose abundance changed significantly following wheat addition, differential analyses were performed on microbial genera with DESeq2, between control and amended conditions. We identified 351 Differentially Abundant Genera (or DAGs) from all domains of life (Supplementary Table 2). Using these DAGs, hierarchical biclustering was performed to identify the succession of populations during the whole kinetics linked with land management history (Fig. 3). In parallel, Principal Component Analyses (PCAs) and PERMANOVAs were performed on each of the genera clusters defined with the hierarchical biclustering to assess their role in community response38. This analysis evidenced different features of the response of the whole soil microbial community to wheat straw residue input.
DAG and sample hierarchical biclustering. Clusters of genera, some specific to one land-use history, some generalist, that multiply in either the early phase or late phase after amendment are observed. Sample hierarchical clustering displayed at top, DAG hierarchical clustering on the left. DAGs clusters numbers are indicated in colours next to the genera, as well as their taxonomic assignation on the left.
Firstly, the clustering of the samples above all showed a separation between the amended samples 3 days after the straw input from all other samples, regardless of land management history (Fig. 3). This highlighted the strong impact of the wheat straw input on soil microbial communities in the early decomposition phase. Notably, many over-represented microbial genera were common to both land use histories during this early phase of degradation. These generalist genera, like Massilia, a bacterial genera which was also found to proliferate in other straw amendment studies25,39, and the fungus Mortierella, were mainly gathered in DAGs cluster 9. Interestingly, cluster 9 also encompassed protists such as Pythium and several viral genera of the class Caudoviricetes (which groups bacterial and archaeal phages) like Amigovirus.
Secondly, amended soils on day 3 were discriminated between croplands and grasslands, suggesting differences between the early responses of the microbial communities to the straw addition according to land-use history. Put simply, these differences stemmed from early responder genera reacting specifically in one land-use history with genera belonging to cluster 11, e.g., Leifsonia and Paraburkholderia, specific to grassland; or to cluster 7, e.g., Serratia and Enterobacter, specific to cropland).
Thirdly, all the other samples were separated according to their land-use history, with unamended soil microbiota remaining close together throughout the whole kinetic. This separation was mainly driven by cluster 1 genera which comprised all the archaeal DAGs which were more represented in croplands, and cluster 8 genera that were more abundant in grasslands. Interestingly, in these clusters, amended day 0 samples were clustered with the control samples. It was obviously due to a lack of time for the communities to respond to wheat addition, but it confirms the robustness of our field experiment and sampling approach.
Finally, days 51 and 125 amended soil samples were clustered together within their respective land-use histories. This can be linked with the response of genera stimulated during the late phase of wheat straw decomposition (days 51 and 125) belonging to DAGs clusters 4, 5, and 10. Again, some responding genera were common to both land-use histories (e.g., Chitinophaga, Fusarium, Syncephalis) while others were specific to grasslands (e.g., Arenimonas, Cytophaga) or croplands (e.g., Hypoxylon, Lysobacter, Rhizopus, Acanthamoeba).
Archaeal genera, did not react strongly to the amendment, suggesting that they were not major players in carbon decomposition. Bacteria represented the bulk of genera (240 out of 351) influenced by wheat straw input. The variety of their responses in the late or early phase, in one land-use or in both, illustrated the considerable breadth of life strategies amongst Bacteria. Fungal genera likewise showed a broad variety of responses after straw input, showing their ability to act as labile or recalcitrant organic matter degraders. Regarding protists, Oomycetes reacted in the early phase of decomposition in both land-use histories, showing their similarities with fungal lifestyles, while other protists did not react quickly to carbon addition. Lastly, most viral genera reacting to the amendment belonged to the Caudoviricetes phages. They responded positively and quickly after the straw input in both land-use histories. This suggests either that their hosts multiplied in the early stage of straw decomposition in both land-uses, or that they might infect multiple hosts.
While soil bacterial and fungal heterotrophic successions have already been described in previous studies25,39, our metagenomics survey provides a powerful tool for examining the concerted succession of all soil microbial actors. It allows looking further than the regulation of bacterial and fungal populations by resource availability, and investigating potential trophic regulations between all taxonomic groups16.
Biotic interactions suggested
Since we looked at every microbial entity at the same time, we could hypothesise about potential interactions between taxonomic groups and how they might participate in regulating the growth of wheat straw consumers. For instance, the bacterial facultative predator genus Lysobacter was most abundant at day 51. Lysobacter species can lyse Bacteria, Fungi and Oomycetes cells40. Interestingly, all the reactive Oomycetes genera were found in cluster 9 (Supplementary Table 2) and were most abundant on day 3 and then decreased, perhaps in part because of predators such as Lysobacter. Overall, several bacterial genera previously characterised as facultative predator (e.g., Cupriavidus41, members of the Cytophagales42, Ensifer43, Stenotrophomonas44) were influenced positively by the straw input. However, because of their omnivorous diet, the question of whether their multiplication was fuelled mostly by wheat straw decomposition, predation or by a combination of both processes cannot be answered. In contrast, the situation was more straightforward for Acanthamoeba, a ubiquitous amoebozoan genus that feeds on bacterial and eukaryotic prey45. It was significantly more abundant in cropland amended soils at days 51 and 125, suggesting greater nutrient uptake by predation and gradual multiplication (Fig. 4). Predation by this genus has also been observed in a leaf litter decomposition experiment15. Protozoan grazers notably release the ammonium contained by their prey46. They have been suggested to play an important role in nutrient release from Bacteria to soils, where the nutrients could be exploited by other microorganisms, feeding the “microbial loop”47.
Abundance patterns of selected genera and theorized trophic links. Abundance profiles (a) of some genera with trophic interactions described in the literature suggest potential top-down regulation models (b) in our dataset. (a) Raw abundances of several selected genera along the time series. Stars denote differentially abundant counts between the control and the amended treatment at a given time point (b) Models of potential trophic interactions between the selected genera.
In DAGs cluster 9, sequences affiliated to several bacteriophages displayed significant enrichment: Amigovirus, R4virus and Slashvirus. Notably, Amigovirus is a known predator of Arthrobacter48 and sequences affiliated to this virus increased in parallel with Arthrobacter sequences. As Viruses have no means of directly exploiting wheat residues, this could suggest either a passive multiplication of viral sequences in the bacterial host, or an active lytic cycle infection (Fig. 4). Interestingly, another type of virus was found in the DAGs, namely Pandoravirus, a giant virus genus known to infect Acanthamoeba organisms49. These two organisms were most abundant specifically in croplands on day 125. We could hypothesise that populations of the predator Acanthamoeba could in turn be regulated by viral infection (Fig. 4). The haustorial obligate mycoparasite genus Syncephalis that targets Mucorales50 was found in DAGs cluster 5. This parasite might be implicated in the reduction of sequences affiliated to its potential hosts Mucor or Mortierella at days 51 and/or 125 in both land-use histories (Fig. 4). This parasitic genus was also previously shown to incorporate carbon from plant residue, alongside Mortierella51.
In conclusion, in this study we have shown that amplicon sequencing and metagenomics conducted years apart, with different technologies and with large sequencing depth differences, still provide similar assessments of in situ bacterial and fungal communities. Owing to its greater sequencing depth and its without a priori approach, metagenomics was more resolutive and allowed simultaneously assessing the whole soil microbial community (Archaea, Bacteria, Eukaryota and Viruses), hence providing precious data to decipher the interactions between groups in response to fresh plant residue inputs. Organisms from every taxonomic compartment were affected either positively or negatively by the amendment. Land-use histories impacted the soil communities and their responses to amendment, with different microorganisms responding in the two types of soils. Lastly, both inter and intra-domain trophic interactions implicating known consumer genera could be suggested from our dataset. Microbial predators, parasites and Viruses likely regulated soil communities’ response due to their generalist or specific prey/host range, thereby constituting top-down regulation and counterbalancing the bottom-up regulation of plant inputs (exudates, or plant litter)14.
Going further in establishing the functional potential of the soil microbiota as a whole will enable the exploration of carbon cycling pathways. In addition, by analysing the functional repertoire of the microbial community, we might gain insight into why some genera had different responses in the two land-use histories. For instance, it might allow distinguishing different strains with different capacities for carbon decomposition or different defence mechanisms against predators or competitors. Lastly, by analysing viral sequences more thoroughly, it could be possible to predict their potential hosts and assess whether phages are induced or integrated in their hosts’ genome. Thus, we could confirm whether, as in marine environments52, Viruses take part in a “kill-the-winner” strategy in soil environments.
