Effects of the different farming modes on the chemical properties of soil
The different farming modes had particular effects on the chemical properties of soil. Compared with those under the conventional farming mode, the changes in soil pH under the organic farming mode were smaller during 2015-2017 (Fig. 1A). The content of soil organic matter (SOM) was higher under the organic farming mode than that under the conventional farming mode in 2016 and 2017(Fig. 1B), while the amount of readily oxidized organic carbon (ROOC) under the organic farming mode was lower than that under the conventional farming mode in 2016 and 2017 (Fig. 1C). The amounts of total nitrogen (TN),total phosphorus (TP) and total potassium (TK) were not significantly different between the organic farming mode and the conventional farming mode in 2015, 2016 and 2017 (Fig. 1D, E and F). The contents of available nitrogen (AN) and rapidly available phosphorus (AP) under the organic farming mode were lower than those under the conventional farming mode in 2014 and 2015, while they were higher under the organic farming mode than those under the conventional farming mode in 2016 and 2017. The rapidly available potassium (AK) under the organic farming mode was higher than or nearly equal to that under the conventional farming mode in 2015-2017 (Fig. 1G, H and I).
Reads and OTU statistics
According to sequencing analysis of the V4 regions of the 16S rRNA gene for bacteria in the soil samples under the different farming modes, we obtained paired sequences with barcode and primer sequences. After removing the short reads and trimming the low-quality regions, 1, 339, 599 optimized sequences (clean reads) were identified from the 42 soil samples with an average length of 412-413 bp (Fig. S1, Table S1). The coverage index was used to reflect the coverage scale of the sample libraries. The coverages of all 42 samples were larger than or equal to 0.97, indicating that sequencing depths could meet the experimental requirements (Table 1). The rarefaction curve was flat when the number of sequences was between 30, 000-40, 000, suggesting that the sequencing data amount was reasonable to obtain the real richness of the community of soil bacteria (Fig. 2A). Based on the DNA sequence similarity (97%), a total of 3, 919 OTUs were identified by aligning the sequence to the Greengenes database (v13_5).
Soil bacteria identification and Venn diagram analysis
Analysis of the taxonomic distributions of bacterial communities at different levels of classification showed that 3, 919 operational taxonomic units (OTUs) belonged to 26 phyla, 42 classes, 78 orders, 120 families, 281 genera, and 340 known species(Table S2). The distributions of common and unique OTU subsets of soil bacteria under the organic farming mode and the conventional farming mode were analyzed through a Venn diagram in April 2015 and October 2017 (Fig. 2B). A total of 1, 507 bacterial OTUs were identified as common to all the soil samples. O6 had the highest number of unique OTUs (177), followed by O1 (169), CK1 (145), and CK6 (115). In addition, There were 2, 225 common OTUs between CK1 and O1 in in April 2015, but there were 1, 926 common OTUs between CK6 and O6 in October 2017. Furthermore, there were 1,889 common OTUs between CK1 and O6 in October 2017. Hence, on the basis of OTUs, different farming mode led to the difference of common and unique OTU subsets of soil bacteria.
Community structure of bacteria
Species accumulation curves are widely used to evaluate the sampling adequacy and the estimates of species richness in biodiversity and community surveys. In this study, the species accumulation curve first rose sharply and then rose slowly, indicating that the sampling was sufficient and that many species were present (Fig. 3A). Based on the shared OTUs of the sample and the species represented by the OTUs, the analysis of the core bacteria showed that the shared OTU number of all the soil samples was more than 300 (Fig. 3B). The relative abundances of the dominant phyla were similar among the 42 soil samples under the different farming modes. Proteobacteria, Acidobacteria, Bacteroidetes, Gemmatimonadetes, Chloroflexi, Actinobacteria, Planctomycetes and Nitrospirae were the most abundant phyla in the 42 soil samples (Fig. 4A). The dominant genera were Lysobacter, Pseudomonas, Massilia, Pseudarthrobacter, Ferruginibacter, Flavobacterium, Flavisolibacter, Brevundimonas, Haliangium and Sphingomonas. The histogram of the microbial species distribution at genera levels showed that Massilia and Lysobacter increased quickly in soil under the organic farming mode in comparison with their performance under the conventional farming mode (Fig. 4B, Fig. 5A, B). To evaluate differences in the composition of the soil sample bacteria under different farming modes during 2015-2018, principal component analysis (PCA) was conducted. The results showed significant clustering of bacterial communities (Fig. 6). PC1 and PC2 explained 47.57% and 24.57% of the global variation in the communities, respectively. O4, O5, O6 and O7 were clearly distinguished from CK1 and O1, indicating the significant influence of different farming modes on soil bacteria.
Analysis of bacterial diversity
OTUs with 97% similarity were used to calculate Shannon, Simpson, Chao and Ace values under different farming modes by the mothur and R (3.4.1) language tools. We found that the Ace and Chao index first increased, then decreased and increased under different farming modes (Fig. 7A, B), indicating that different farming modes had different effects on soil bacterial biodiversity. The Shannon index first decreased and increased under the different farming modes, but the conventional farming mode had larger effects on soil bacterial diversity than the organic farming mode (Fig. 7C). The Simpson index first increased under different farming modes during 2015-2016, but decreased under the conventional farming mode and did not change under the organic farming mode during 2017 (Fig. 7D).
LEfSE of the bacterial community
LEfSE (linear discriminant analysis effect size) is a valid statistical method for high-efficiency biomarker discovery that enables the comprehensive identification of richness characteristics by describing potential differential taxa between different biological samples. A linear discriminant analysis (LDA) score higher than 2 was employed to distinguish only the bacterial groups that were clearly different among the four samples (CK1, CK6, O1, O6). A cladogram depicted all the taxa with LDA values higher than 2.5 by treatment (Fig. 8). The major bacterial groups identified in CK1 were Proteobacteria, Gammaproteobacteria, Oxalobacteraceae and Massilia. CK6 had the highest abundance of Deltaproteobacteria. O1 had a higher level of Acidobacteria and Actinobacteria, while O6 had a higher abundance of Gemmatimonadales, Chloroflexi, Gemmatimonadaceae, and Gemmatimonadetes.
Correlation between soil chemical features and Bacterial Communities
The relationships between environmental factors and bacterial community at the phylum level were elucidated using redundancy analysis (RDA). The 5 most dominant genera were used for the RDA (Fig. 9). The results showed that the first two axes explained approximately 96.73% of the variation in the bacterial composition, with 89.31% explained by the first axis and 7.42% by the second axis. AN was the most important factor mediating the bacterial community under the organic farming mode (p< 0.05), followed by soil AP and AK content. The Massilia, Pseudomonas, Lysobacter and Pseudarthrobacter displayed a strong positive correlation with the AN content. A strong negative correlation between AN and Haliangum was found. A significant separation among the four treatments, which were almost divided into three groups, was found along the first two axes. Additionally, the long distances between conventional farming and organic farming in 2017 further suggested that organic farming and conventional farming significantly affected soil bacterial communities and that there were significant differences between them.