Spatio-temporal dynamics of soil bacterial communities as a function of Amazon forest phenology

Most tropical evergreen rain forests are characterised by varying degrees of precipitation seasonality that influence plant phenology and litterfall dynamics. Soil microbes are sensitive to soil water:air ratio and to nutrient availability. We studied if within-year seasonality in precipitation and litterfall-derived nutrient input resulted in predictable seasonal variation in soil bacterial diversity/microbial functional groups in an Amazonian forest. We characterised the spatio-temporal dynamics of microbial communities from the plot to the stand scales and related them to precipitation seasonality and spatial variability in soil characteristics. Community composition and functional diversity showed high spatial heterogeneity and was related to variability in soil chemistry at the stand level. Large species turnover characterised plot level changes over time, reflecting precipitation seasonality-related changes in soil nutrient and moisture regimes. The abundance of decomposers was highest during the rainy season, characterised also by anaerobic saprophytes and N2-fixers adapted to fluctuating redox conditions. In contrast, Beijerinckiaceae, likely derived from the phyllosphere, were found at higher abundances when litter inputs and accumulation were highest. We showed that in a mildly seasonal rain forest, the composition of soil microbial communities appears to be following canopy phenology patterns and the two are interlinked and drive soil nutrient availability.


Results
Bacterial community characterization assessment. Sequence read numbers after quality control and removal of all OTUs that had less than six occurrences across all samples averaged 14,075 (range 5,555-36,690) per sample and resulted in a total 3,714 OTUs over the three sampling dates. Species richness (SR), Shannon Index (H') and the abundance-based coverage estimator (ACE) per plot averaged 723 ± 24 (range, 498-995), 5.3 ± 0.05 (range, 4. 8-5.8) and 937 ± 50 (range, 522-1582), respectively. Phylogenetic diversity (PD) per plot averaged 163 ± 3 (range, . No significant differences were found for SR, H' and PD, for any paired comparisons between sampling dates (data not shown). Rarefaction curves tended to saturation for almost every sample indicating that OTU's were representative of the total bacterial community (Fig. 2).

Spatio-temporal changes/pattern in bacterial diversity. ADONIS and PERMANOVA analyses car-
ried out on UniFrac distance matrices calculated on both presence/absence and relative abundance data, showed significant effects of both seasonality and site on bacterial community composition (Table 1). Relationships between UniFrac distances and explanatory variables were observed for pH (F Model, 2.59; R 2 , 0.06; p ≤ 0.001), soil organic matter (OM, F Model, 2.22; R 2 , 0.05; p ≤ 0.01), PO 4 3− (F Model, 1.56; R 2 , 0.03; p ≤ 0.05), Ca 2+ (F Model, 2.36; R 2 , 0.05; p ≤ 0.001) and precipitation in the 30-day period preceding sampling (F Model, 1.95; R 2 , 0.04; p ≤ 0.01). The same variables were selected as significant in the canonical correspondence analyses (CCA) made using relative abundance of OTU data ( Fig. 3; for statistical significance of individual variables see Table S1). The majority of taxa represented from order to genus level, belonged to three phyla: Actinobacteria, Alpha-Proteobacteria and Firmicutes (see Table S2 for significance of correlation). Relative abundances of different taxa belonging to Actinobacteria were positively related with the precipitation in the 30 days prior soil sampling and to soil Ca 2+ and PO 4 3− concentration. The abundance of Firmicutes was positively related to pH, whereas Hyphomicrobiaceae, Acetobacteraceae and Rhodospirillaceae (Alpha-Proteobacteria) were related to OM, and Beijerinckiaceae and Bradyrhizobiaceae increased in abundance in samples collected at Times 2 and 3 relative to Time 1.
Mantel correlograms calculated on all samples (i.e. ß sor vs. geographical distance) showed that the composition of bacterial samples was positively autocorrelated at the plot level on both taxonomic (r = 0.32; corrected p-value < 0.001) and phylogenetic data (r = 0.19; corrected p-value < 0.001), indicating that samples collected at different sampling times at the same plot were not spatially independent (i.e. our sampling design allowed valid temporal comparisons), while the minimum distance between plots was greater than the scale of sample autocorrelation, suggesting that our plots represented spatially independent samples (Fig. 5).
Temporal changes in bacterial community composition at the plot level were strongly related to species turnover for all paired comparisons of sampling dates (ß sim = 0.43, ß sne = 0.05, for Time 1 and Time 2; ß sim = 0.42, ß sne = 0.05 between Time 2 and Time 3; ß sim = 0.40, ß sne = 0.07 between Time 1 and Time 3). The relative magnitude of variability at the stand level (Whittaker's β space , 4.02) was, on average, 2.2 times higher than seasonal variability (β time , 1.85) at the plot level (Fig. S1).
Soil analyses. Soil pH, mineral N, K + , exchangeable acidity (hydrogen, H + plus aluminium, Al 3+ ) and cation exchange capacity (CEC) were significantly different among sampling dates ( Table 2). Values of pH increased successively from Time 1 to Time 3. Concentrations of both NH 4 + and NO 3− were significantly less at Time 2 than at Time 1 and higher at Time 3 than Time 2 or Time 1. K + was significantly higher at Time 3 than Times 1 and 2 while exchangeable acidity and CEC had an opposite pattern and both decreased from Time 1 to Time 3 and from Time 2 to Time 3. Fine root biomass was significantly higher at Time 1 when compared with Times 2 and 3. For correlation among variables of soil chemical properties see Table S5.

Discussion
This is the first study that provides insights into the spatio-temporal dynamics of soil bacterial communities in a tropical lowland evergreen rain forest (LTERF). Bacterial community composition and functional diversity were, on the one hand, related to soil resource patterns, and on the other, to seasonality-related changes in soil nutrients and moisture regimes. Community assemblages were largely structured by spatial turnover. Patterns of ß-diversity at the plot level were overwhelmingly attributable to species replacement -either in space (within a sampling date) or in time across three sampling dates, of the same plot. The relative magnitude of variability, calculated by Whittaker´s beta was twice as large at the stand level than seasonal variability at the plot level. Spatial heterogeneity in soil chemical properties is determined by abiotic and biotic controls. Their relative importance in LTERF is expected to shift from abiotic to biotic control from regional to landscape and local   34 . Soil microbial community structure seems to follow identical pattern, being influenced by broad-scale environmental drivers at the regional/landscape scale, e.g. soil pH 35 , soil moisture 36 and temperature 37 , while local-scale relationships are more likely to be influenced by plant-soil interactions 38,39 . While disentangling the effects of plant species on bacterial composition was outside the scope of the present study, we did observe that bacterial communities collected at different dates were spatially structured and positively autocorrelated at the plot level. This confirms the existence at the local scale of spatial horizontal patterns likely linked to the effects of plant species. Different species affect soils both directly through the chemistry of their litter 7,40,41 and indirectly through the effect on detritivores, and litter Ca appears to be an important general agent in these processes 42 . Bacterial community composition in our study was significantly related to Ca 2+ , PO 4 3−3 and soil organic matter (OM) that did not follow a seasonal pattern and which are likely to reflect plant biotic control on spatial soil variability, and by pH, which did show a seasonal pattern with increasing values from the rainy to the dry season. In the light of these results we can reject both hypothesis H 1 (bacterial communities are not affected by spatial variability of resources and by seasonality patterns) and hypothesis H 3 (bacterial community composition is affected by seasonality patterns and this change overrides any spatial variability) and at this point, partially accept hypotheses H 2 and H 4 whereby in an ecosystem that lacks a restriction of the growing season by climate, bacterial community composition is influenced by spatial variability of resources but not by seasonality (H 2 ), and that bacterial community composition reflects both spatial and temporal heterogeneity (H 4 ).
The rainy season was characterized by a relative high abundance of Actinobacteria (i.e. Actinomycetales and Solirubrobacteriales) and a group of indicator species mainly represented by Actomycetales (e.g. Streptomycetaceae) and Firmicutes (e.g. Paenibacillus and Clostridium), that include major decomposers of complex polymers, anaerobic saprophytes and anaerobic N 2 -fixers adapted to fluctuating redox conditions 43 . The temporal changes in the abundance of Actinobacteria could have been partially driven by interactions between seasonal changes in nutrient and water availability. We found Actinobacteria to be correlated with precipitation in the month preceding sampling. This finding is in agreement with a study conducted in agricultural and temperate grassland ecosystems 8 where the authors have suggested that the pattern could be attributed either to the taxa preference for moisture or, alternatively to a presumed late seasonal peak in plant litter and root exudates that could favour Actinobacteria over more oligotrophic taxa. With the start of the rainy season abundant daily rainfall promotes the movement of readily decomposable dissolved organic C from the litter layer 21,44 , driving high annual rates of OM decomposition 16,22 . Since different bacterial taxa process OM at different rates, even under similar abiotic conditions, shifts in OM resources could drive changes in microbial community composition 45 .
For example, Actinobacteria have been shown to be able to survive when resources are limited or in the presence of complex, recalcitrant OM 46 . Actinomycetales are major decomposers of complex polymers such as lignocellulose and chitin 47,48 and their higher abundance toward the end of the rainy season seems to be linked to changes in soil moisture and nutrient availability that promote optimal conditions for the decomposition of recalcitrant C. Seasonal changes in soil moisture are linked to fluctuations in the availability of soil N, which appears to be related to the delicate balance between the processes of immobilization and mineralisation 49 . Luizão, et al. 20 have observed in Amazonia that rewetting of the soil after a dry period induced net N immobilization, whereas net N mineralisation during dry periods resulted in mineral N accumulation in the soil. Net N mineralisation and nitrification in the study area were also higher in the dry season compared with those in the rainy season, with highest N mineralisation at the peak of the dry season 20,50 . In our study, we detected a higher mineral N concentration at the end of the dry season, when compared with values at the other sampling dates.
In tropical rain forest soils, microbial communities under fluctuating moisture regime appear to be characterised primarily by specialist taxa that are able to maintain several processes under fluctuating conditions 24,43 . For example, DeAngelis, et al. 43 have shown that taxa which are significantly more active under fluctuating rather than static oxic conditions are mainly represented by Clostridia, Flavobacteria and Actonomycetales. This could have also been the case at our site at Time 1, characterized by highest abundance of Actinomycetales and by a high number of indicator species belonging to Firmicutes. These include taxa involved in the anaerobic metabolism of complex carbon, such as aerobic/facultative anaerobic Paenibacillaceae, as well as obligate anaerobic taxa involved in N 2 -fixation 51 .
Conversely, soil samples at the transition between rainy and dry seasons (Time 2) and at the end of the dry season (Time 3) were characterized by higher values of pH and K + and a significantly higher relative abundance of Beijerinckiaceae belonging to Rhizobiales, when compared with their values in the rainy season. Times 2 and 3 shared also a relatively high number of indicator species belonging to Alpha-Proteobacteria. While species belonging to Beijerinckiaceae are known to be free-living N 2 -fixers 52 , we suspect that their higher relative abundance at Times 2 and 3 could not be related to potential higher rates of N 2 fixation at these times of the year. Some studies conducted in LTERF have indeed observed higher rates of free-living N 2 -fixation in the rainy season when compared to the dry one, and attributed them to the co-occurrence during this period of C availability and favourable leaf litter N:P ratios 31,32 . To investigate the effects of OM content on soil microbial community structure, Nemergut, et al. 45 manipulated litter inputs in situ. Interestingly, they found that the double-litter plots were characterized by a higher abundance of Rhizobiales than the control plots and suggested that the observed shifts could have been related to changes in N 2 -fixation. However, we suggest that another plausible explanation for the high abundance in their double-litter plots, and at our study site, both at Times 1 and 3, could perhaps be attributed to an increased input of these bacteria from the phyllosphere 53 through increased litterfall 2,16 . Recent studies conducted in lowland tropical rain forests during the dry season 54 and at the transition between dry and rainy season 55 , have shown that the most dominant taxa in the phyllosphere were taxa of Beijerinckiaceae and within them genera known to associate closely with plant hosts, including N 2 -fixing and methylotrophic taxa 54 . In the present study, the significant higher abundance of Beijerinckiaceae at the end of the dry season when compared with the rainy season, and almost significant at the transition between rainy and dry season would support the possibility of a higher abundance of bacteria in the soil linked to the phyllosphere when litter inputs and accumulation are at the highest 56 . Since our results indicated that bacterial community composition is influenced by both spatial and temporal variability of resources we are able to reject hypothesis H 2 in relation to seasonality, and accept hypothesis H 4 , whereby both spatial and temporal heterogeneity contribute significantly in shaping the composition of bacterial communities.
In conclusion, soil bacterial community composition and functional diversity were related to soil resource patterns and to seasonality-related changes in soil nutrient and moisture regimes in an Amazonian LTERF. The highest relative abundance of decomposers of complex polymers were found during the rainy season that was also characterised by indicators species that included anaerobic saprophytes and anaerobic N 2 -fixers adapted to fluctuating redox conditions. Beijerinckiaceae likely derived from the phyllosphere, were found at higher abundances when litter inputs and accumulation were highest. The composition of bacterial communities showed high spatial heterogeneity due to spatial variability in soil chemistry at the stand level and temporal changes that reflected seasonality at the plot level. To corroborate whether some of the reported changes are associated with large inputs of phyllosphere bacteria during the litterfall peak or are instead due to internal soil community dynamics requires further work to establish a functional link between canopy phenology and corresponding temporal patterns in microbial community composition and functioning.

Materials and Methods
Study site. The study was carried out in a 10-km 2 forest reserve (Adolpho Ducke Forest Reserve, RFAD), 26 km NE of Manaus (03°00′00″ − 03° 08′00″ S; 59°52′40″ − 59° 58′00″W), Brazil, in 2013. The RFAD has a dissected undulating topography at 40-140 m a.s.l. and it is associated with different soil types: oxisols on plateaux, ultisols on slopes and hydromorphic spodosols in the valleys 57 . The plateau soils are well-drained and are not affected by waterlogging or flooding during the rainy season. The dominant vegetation type is lowland evergreen rain forest 58 . For further details see Oliveira Freitas, et al. 59 .
The mean annual temperature is 26 °C and mean annual precipitation is 2,550 mm/year (range 1840-3385 mm/year for the period 1966-2014, measured at the Reserve's meteorological station). Periods with monthly precipitation <100 m usually occur between September and November. Although the year of the study, 2013, was much wetter than the average and had no month with a perceived negative hydrological balance, the overall precipitation pattern did not deviate from the long-term mean (Fig. 6).
Litterfall patterns were established based on 3 years of data 1979-1982 16 . The study was designed to capture soil microbial community dynamics during a year, (1) in the wet season, (2) in a transition period during the displacement of the Inter-tropical Transition Zone to the north and the subsequent diminution of rainfall, and (3) at the end of the 'dry season' . Data Accessibility. This Targeted Locus Study project has been deposited at DDBJ/EMBL/GenBank under the accession KBUV00000000. The version described in this paper is the first version, KBUV01000000.
Sampling. Twenty points were chosen randomly along 1.5 km section in a plateau area of a 6-km trail that runs east to west in the RFAD. The minimum distance between any two points was set to be of 40 m for full sampling design see 60 . Soil was sampled on 13 April 2013 (wet season, Time 1), on 28 July (transition into dry season, Time 2) and on 6 November 2013 (dry season, Time 3). At every point, one 1 m × 1 m plot was marked permanently and in every plot three sub-samples were collected to a depth of 5 cm, after the removal of litter, using a cylindrical metal corer of 3 cm of diameter. Each sub-sample was sampled for DNA extraction by using a sterile needle and pooled by plot at each sampling in a 2.5-ml Eppendorf vial. The samples were maintained at −20 °C until molecular analyses. The remaining soil was pooled per plot for soil analyses and for the determination of fine root biomass at each sampling date.  61,62 were used for amplification of bacterial 16 S rDNA. The forward primer was linked through a barcode adaptor with the Ion-Xpress barcode and the Ion adaptor ' A' sequence, while the reverse primer was linked with the P1 sequence (Ion Torrent adaptors, primers and barcodes used in the present study are available in Table S6 of Supporting information). For details on PCR settings see Buscardo, et al. 60 . Negative controls consisting of nuclease free water instead of DNA were made for every barcode and subjected to PCR. Amplicon libraries were quantified using a Qubit 3.0 Fluorometer using a Qubit HS dsDNA Assay Kit (Life Technology). Equimolar amounts of each library were pooled (24 barcode per chip), purified with the Agencourt AMPure XP Kit (Beckam Coulter) and diluted to 100 pM. The final solution for emulsion PCR was obtained by diluting 3 μl of the solution 100 pM in 22 μl of nuclease free water according to the Ion PGM Template OT2 400 Kit manual for further details see 60  Sequence processing. The primers were removed and poor-quality ends were trimmed off based on 0.02 error probability limit in Geneious Pro 5.6.1 (BioMatters, New Zealand). Sequences were then filtered using USEARCH v.8.0 63 as described in Buscardo, et al. 60 . After excluding singletons and discarding OTUs that were also present in any of the five negative samples, the data set contained 588,168 sequences in 7,888 OTUs. Since the reliability of OTUs with a low number of sequences may also be questioned and given the high total number of sequences per sample (between 195,000 and 370,000), we choose to follow the conservative approach suggested by 64 and removed all OTUs that had less than six occurrences across all samples. For the final analyses the total number of sequences was thus reduced to 576,867 (3,714 OTUs). Sequences were assigned to taxonomic groups by RDP Naïve Bayesian Classifier Version 2.10 65 .
Soil analyses. Ammonium (NH 4 + ), nitrate (NO 3− ) and soil moisture were quantified in the Soil Laboratory of the National Research Institute of Amazonia (INPA), Manaus, Brazil for each soil sample (a total of 60 samples), within 48 h after collection for details see 60 . The remaining soil was air dried and pH, total soil organic matter (OM), extractable phosphorous (PO 4 3− ), potassium (K + ), calcium (Ca 2+ ), magnesium (Mg 2+ ), and exchangeable acidity (hydrogen, H + plus aluminium, Al 3+ ), were determined in the Laboratory of Soil Analysis, Escola Superior Agraria Luiz de Queiroz, University of São Paulo (ESALQ/USP; Piracicaba, Brazil), according to van Raij, et al. 66 . Fine (≤2 mm) and coarse roots were separated in each soil sample and their biomass determined by oven drying at 65 °C dry to constant mass. Data analyses. All analyses were conducted on data rarefied to 5,555 sequences (corresponding to the smallest number of sequences encountered in a sample). All analyses based on phylogenetic information were based on a phylogenetic tree built with an approximate maximum-likelihood approach with FastTree v.2.1 67 , built on a code edited with the -DUSE_DOUBLE parameter. The phylogeny was midpoint-rooted, using the Archaeopteryx interface 68 . The generalized UniFrac procedure 69 implemented in the R package 'GUniFrac' v.1, was employed to compute UniFrac distances 70 . Rarefaction curves, α-diversity (SR), Shannon index (H') and the abundance-based coverage estimator (ACE) were computed in the 'vegan' R package v.2.3-25 71 . Phylogenetic diversity was calculated as Faith's phylogenetic diversity (PD) using the R package 'picante' v.1.6-2 72 . Taxonomic and phylogenetic ß-diversity were computed R package 'betapart' v.1.3 73 to characterize spatial turnover (ß sim ) and nestedness (ß sne ) at each sampling date. Differences in diversity indices among community at different sampling dates were tested using pair-wise differences with the t-test.
To determine the relative magnitude of spatial and temporal variability in OTU richness, we calculated ß-diversity by using Whittaker's orginal formula (β = γ/α), where γ is the total number of OTUs at each of the sampling ocassion (total of 15 plots for space and of three samplings per plot for time) and α is the number of OTUs per plot at a given time. We did this (a) for spatial variability (β space ) at the stand level (full extent of the sampling transect) and (b) for temporal variability (β time ) at the individual plot level.
Spatial autocorrelation between bacterial community composition (ß sor ) at the sample level (within plot at different sampling dates and between plots) and geographical distances were assessed with a Mantel correlogram in the 'vegan' R package, using the Pearson's correlation method with 9999 permutations. Distance classes were chosen based on the sampling distances within our sampling design. Significance values across distance classes were corrected for multiple comparisons using the progressive Holm method.
Temporal changes in taxonomic community composition were assessed on presence/absence data, by computing dissimilarities ('betapart' v.1.3 R package) for each sampled plot, between different sampling dates, and considering the overall change as well as the single components (i.e. ß sim and ß sne ) over time.
Spatial and temporal differences in phylogenetic community composition considering repeated-measures were assessed on rarefied presence/absence and relative abundance data with two different analyses, ADONIS, with sampling date nested in site to evaluate sampling date effect ('vegan' R package v.2.3-2), and nested PERMANOVA 'BiodiversityR' R package 74 ; to evaluate the site effect, both using 10,000 permutations. ADONIS was also employed to quantify the relationship between UniFrac distances calculated on rarefied relative abundance data and soil explanatory variables. The GUniFrac distance matrix used in ADONIS was made by setting alpha to 0.5 to provide the best overall power 69 .
Canonical correspondence analyses (CCA) were carried out on relative abundance taxonomic data to visualize differences in OTU-based community composition. Only significant variables were included in the ordination following a sequential ANOVA significance assessment of each single term. Vectors representing bacterial taxa at SCIENTIFIC REPORtS | (2018) 8:4382 | DOI:10.1038/s41598-018-22380-z phylum and lower taxonomic levels were fit into the ordination by using the 'vegan' 'envfit' function (only relative abundances of the first 300 most abundant OTUs were considered) and their significance assessed under 999 permutations.
Differences between sampling dates, in relative abundances of the first 300 most abundant OTUs (corresponding to 77% of the total abundance) were computed on data at phylum and at lower taxonomic levels and evaluated with the Wilcoxon signed-rank test. Significances were corrected for multiple comparisons using the false discovery rate (FDR) method 75 .
An indicator species analysis was carried out using 'indicspecies' v.1.7-6 in R 76 . Differences in soil parameters at different sampling dates were evaluated with paired t-test or with the Wilcoxon signed-rank test when normality assumptions were not satisfied. Correlations between soil parameters were assessed using the Pearson's correlation method.