The soil used in this study was collected from the topsoil (0–20 cm) of a commercial chili pepper field in Weslaco, Texas, United States, where Fusarium wilt was a consistent and serious problem and where no seed meal amendments were used previously. This soil had a sandy loam texture (sand 69%, silt 14%, and clay 17%) with pH 8.2. The nutrient content of this soil was as follows (Ma et al., 2015): 0.72% organic matter, 10 mg g^-1 nitrate-N, 125 mg g^-1 P, and 277 mg g^-1 K. The soil was passed through a 5-mm mesh before use.

Three types of defatted BSMs, including seed meal of Camelina sativa cv. ‘Crantz,’ provided by Dr. Terry Gentry from Texas A&M University, United States; seed meal of Brassica juncea cv. ‘Pacific Gold,’ provided by Farm Fuel Inc., Freedom, CA, United States; and seed meal of Sinapis alba cv. ‘IdaGold,’ provided by Farm Fuel Inc., Freedom, CA, United States, were used in this study, and these three seed meals are hereinafter called CAME, PG, and IG, respectively. The content and composition of glucosinolate and the content of N, P, and K in these seed meals have been shown previously (Ma et al., 2015). Specifically, PG contained 164 μmol g^-1 of glucosinolate, with 99% being 2-propenyl (allyl) glucosinolate; IG contained 195 μmol g^-1 of glucosinolate, with 97% being p-hydroxylbenzyl glucosinolate; and CAME contained 23.5 μmol g^-1 of glucosinolate, with 51.9% being 10-methyl-sulfinyl-decyl-glucosinolate, 30.2% being 11-methyl-sulfinyl-decyl and 17.9% being 9-methyl-sulfinyl-decyl glucosinolate. The N content was 5.9% in CAME, 6.1% in PG, and 6.2% in IG; the P content was 1.01% in CAME, 1.2% in PG, and 1.18% in IG; and the K content was 1.45% in CAME, 1.5% in PG, and 1.37% in IG. These BSMs were sieved through a 1-mm metal mesh before use.

Three BSMs, CAME, PG, or a mixture of PG and IG (PG:IG (w/w) = 1:1), were prepared. This preparation resulted in the establishment of three levels of 2-propenyl (allyl) glucosinolate in these seed meals: 0 μmol g^-1 in CAME, 164 μmol g^-1 in PG, and 84 μmol g^-1 in PG+IG. Then, these seed meals were added separately at an application rate of 0.34% (w/w, equivalent to 6000 kg ha^-1) and mixed thoroughly with the soil. The soil that was amended with no seed meal was used as a control (CK). To bring the N, P, and K content onto a common scale, urea (46-0-0) and compound fertilizer (0-15-15) were added to the CK treatment so that the nutrition level was consistent across all treatments. Soil moisture was then adjusted to approximately 70% of the SWHC (soil water holding capacity) by adding sterilized water. The soil from each treatment was put into three Ziploc^® plastic bags, with 7.5 kg soil in each bag. The bags were left open and incubated at room temperature (approximately 24°C) for 25 days. During the incubation period, sterile water was replenished every 2 days to maintain stable soil moisture based on the weight loss. At the end of the incubation period, 10 g soil was collected for DNA extraction after the entire soil in each bag was mixed thoroughly. These incubated soil samples hereinafter are referred to as the pre-planting soil samples.

After collecting pre-planting soil samples, the rest of the soil from each bag was divided into six subsamples of equal weight and then put into six pots. Pepper seedlings (Capsicum annuum ‘Ben Villalon,’ 4-leaf stage) were transplanted into the pots at a density of two plants pot^-1. All the plants grown in these six pots were classified as a replicate group within a treatment. The disease severity of each replicate group was measured at days 7, 15, 25, and 35 after transplanting. The disease severity of pepper was ranked into five levels according to a previous method (Ma et al., 2015): 1, healthy; 2, 1/4 leaves turned yellow and fell off; 3, 1/3 leaves turned yellow and fell off; 4, 1/2 leaves turned yellow and fell off; and 5, dead. The disease index was then calculated according to the formula: Σ(number of diseased plants × corresponding level) ÷ total number of investigated plants × 100. At the end of the pot experiment (35 days after transplanting), rhizosphere soil from each treatment was sampled from three replicate groups for DNA extraction. These samples hereinafter are referred to as the post-planting soil samples. All the pot experiments were conducted in the greenhouse of the Horticulture Department of Texas A&M University.

Junctions at 1 cm length between the diseased and healthy tissue on stems of the wilted plants were surface-sterilized and incubated on potato-dextrose agar (PDA) containing 50 μg mL^-1 rifampicin. When mycelium appeared around the stem, they were further cultured on PDA medium and were identified based on microscopic observation of biological characteristics. Fusarium spp. was the only fungus based on the microscopic observation results. To follow Koch’s postulates, this isolate was inoculated in sterilized soil where healthy pepper plants were grown. Compared to the plants in the uninoculated soil, plants in the Fusarium-inoculated soil developed the same wilt symptoms. This result further confirmed that the isolated Fusarium spp. was the only pathogen.

Total genomic DNA was extracted from 10 g pre-planting soil samples and post-planting soil samples collected after 25 days of incubation and after 35 days of pepper plant growth, respectively, using a PowerSoil DNA Isolation Kit (Mo Bio Laboratories, Inc., Carlsbad, CA, United States) following the manufacturer’s instructions. Extracted DNA was purified using an Illustra MicroSpin S-400HR Column (GE Healthcare Bio-Sciences Corp., Piscataway, NJ, United States). The quality of the purified DNA was examined by electrophoresis in a 0.8% agarose gel. DNA quantity and purity were then determined with a Nanodrop ND-1000 spectrophotometer (NanoDrop Technologies Inc., Wilmington, Delaware, United States). All DNA samples were stored at -20°C until use.

The real-time quantitative PCR technique was applied to quantify the bacterial 16S rRNA gene on a Rotor-Gene 6000 series thermal cycler (Qiagen, Valencia, CA, United States). The forward and reverse primers used were Eub338 (5′- ACT CCT ACG GGA GGC AGC AG-3′) and Eub518 (5′-ATT ACC GCG GCT GCT GG-3′), respectively (Fierer et al., 2005). The standard curve for the bacterial 16S rRNA gene was generated via 10-fold serial dilution of plasmid DNA harboring the target gene. The reaction mixture (15 μL) contained 6.75 μL Real Master Mix with 20 × SYBR solution (5 Prime, Inc., Gaithersburg, MD, United States), 0.5 μL each primer (10 μM), 1.5 μL BSA (10 mg mL^-1), 1 μL template DNA, and 4.75 μL ddH₂O. The thermocycling steps were as follows (Fierer et al., 2005): denaturing at 95°C for 15 min; followed by 40 cycles of 1 min at 95°C, 30 s at 53°C, and 1 min at 72°C; and 30 s with a plate read. The negative controls with water as the template instead of soil DNA were always run when the real-time quantitative PCR for soil samples was carried out. Real-time quantitative PCR was performed with three technical replications for each DNA sample.

For each soil DNA, the 16S universal bacterial primer set 27F (5′-AGR GTT TGA TCM TGG CTC AG-3′) and 519R (5′-GTN TTA CNG CGG CKG CTG-3′) was used for amplifying the ∼ 500 bp region of 16S rRNA genes. To resolve different samples, an 8-bp barcode was fused to the forward 27F primer. A HotStarTaq Plus Master Mix Kit (Qiagen, Valencia, CA, United States) was used for PCR under the following conditions: 94°C for 5 min, followed by 32 cycles of 94°C for 30 s, 55°C for 30 s, and 72°C for 45 s; and a final elongation step at 72°C for 5 min. The PCR products was visualized and confirmed on a 1.8% agarose gel. All PCR amplicons from different samples were then purified using Agencourt AMPure beads (Agencourt Bioscience Corp., Beverly, MA, United States). The concentration of each purified PCR amplicon was determined using a Nanodrop ND-1000 spectrophotometer (NanoDrop Technologies Inc., Wilmington, Delaware, United States). Equimolar concentrations of the amplicons were then merged into a single tube and subjected to pyrosequencing at Molecular Research Laboratory (Shallowater, TX, United States) using 454 GS FLX titanium technology (454 Life Sciences, Branford, CT, United States). The amplicons were sequenced in the forward direction.

The 454 pyrosequencing data were analyzed using the Quantitative Insights Into Microbial Ecology (QIIME) Pipeline¹. Specifically, the low-quality sequence reads (reads lengths < 150 bp, ambiguous bases > 0, homopolymers > 6, barcode mismatches, and average quality scores < 25) were discarded, and the 8-bp barcode was examined to distribute the sequences to proper samples. Then, chimeric sequence reads were identified and filtered using the Uchime algorithm based on a chimera-free reference database (Edgar et al., 2011) through the Usearch tool. The operational taxonomic units (OTUs) were generated at a 97% sequence-similarity level (Edgar, 2010). A representative sequence of each OTU was aligned with the PyNAST tool (Caporaso et al., 2010). The taxonomic identification of OTUs was obtained using the RDP Classifier with RDP as the reference database (Wang et al., 2007).

To standardize sampling efforts and to bring the pyrosequences of each sample onto a common scale, all samples were rarefied to an equal sequence depth (866 sequences per sample in this study). The relative abundance of all the phylotypes at each taxonomic level (phylum, class, order, family, and genus) was then summarized. Principal coordinate analysis (PCoA) was performed using the subsampled sequence data to determine the differences in the microbial community structures based on the weighted UniFrac distances, a method that accounts for the phylogenetic relationship between sequences and is thus more powerful than taxon-based measures (Lozupone and Knight, 2005; Hamady et al., 2010). Samples were clustered using UPGMA (unweighted pair group method with arithmetic mean) on weighted UniFrac distances. Three different complementary non-parametric analyses for multivariate data (Zhou et al., 2012), namely, the analysis of similarities (ANOSIM) (Clarke, 1993), non-parametric multivariate analysis of variance using distance matrices (adonis) (Anderson, 2001), and a multi-response permutation procedure (MRPP) (Mielke and Berry, 2001; McCune and Grace, 2002), were used to test for differences in the community structure between treatments. Weighted UniFrac distances were exploited for ANOSIM, Adonis, and MRPP analyses, and a Monte Carlo permutation was used to determine the statistical significance. ANOSIM, adonis, and MRPP were performed using the “vegan” package in the R environment.

For the relative abundance data of each of the taxa, the linear discriminant analysis (LDA) effect size (LEfSe) method² was used to test significant differences between treatments. Furthermore, Pearson correlation analysis was performed to understand the possible relationship between the bacterial taxa and the disease index. The Pearson correlation analysis were processed with the software SPSS.

The original sequence data have been deposited in the European Nucleotide Archive under accession number PRJEB19197.

Based on the disease index, a parameter used to assess the disease severity, reported on days 7, 15, 25, and 35 after transplanting (Figure 1), PG-containing (i.e., PG and PG+IG) treatments significantly (P < 0.05) decreased the wilt index compared to the control during 35 days of cultivation (Figure 1). PG exhibited the most suppressive effect and completely suppressed the wilt throughout the 35-day period after transplanting. PG+IG treatment also significantly decreased the wilt index by 72.8% (Figure 1). However, the addition of CAME did not show an apparent suppressive effect. In addition, a significant negative correlation (r = -0.959, P = 1.68e-4) was found between 2-propenyl glucosinolate content in BSMs and the disease index on day 35. In other words, the suppressive efficacy was positively correlated with the 2-propenyl glucosinolate contained in the BSMs.

The 16S rRNA gene-based pyrosequencing technique was used to understand the bacterial community changes resulting from BSM amendment and to explore its relationship with plant disease severity. After applying all quality filters, a total of 53,564 high-quality sequences across all samples or an average of 2,060 sequences per sample were obtained (Supplementary Table S1). All BSM treatments significantly decreased the bacterial richness and diversity before pepper planting based on the observation that the observed OTU number and phylogenetic diversity (PD) value were significantly (P < 0.05) lower under CAME, PG, and PG+IG treatments in pre-planting soil samples (Figures 2A,B). Although the significant decrease still existed after 35 days of pepper growth in PG and PG+IG treatments (Figures 2A,B), it was greatly weakened under all BSM treatments, as the difference between all BSM treatments and CK in terms of the observed OTU number and PD value in post-planting soil samples became smaller than that in pre-planting soil samples (Figures 2C,D).

Furthermore, after 35 days of pepper growth, in the treatments where the plant disease index was similar, the bacterial richness and diversity were also similar. Specifically, in the CK and CAME soil where the plant disease index was 89.2 and 100, respectively (Figure 1), there was no significant difference (P > 0.05) in terms of the observed OTU number and the PD value between these two treatments after 35 days of plant growth (Figures 2A,B), while in the PG and PG+IG treatments where the plant disease index was obviously decreased but was still similar (PG: 0; PG+IG: 24.2) (Figure 1), there was still no significant difference (P > 0.05) in terms of the observed OTUs number and the PD value (Figures 2A,B). Intriguingly, we also found that 35 days of pepper growth exerted no significant influence on the bacterial richness and diversity when BSM was not amended, while it had significant effect in BSM-amended soil (Figures 2A,B).

The bacterial community structure is an overall reflection of the composition and abundance of different taxonomic population in a microbial community. Changes in bacterial community structure as affected by BSMs were thus assessed with the PCoA and UPGMA method based on the weighted UniFrac distances.

Twenty-five days of incubation with BSMs significantly changed the bacterial community structure. Specifically, samples from the BSM-amended treatments and unamended control were distributed in different regions of the PCoA space (Figure 3A) or grouped in different clusters in the UPGMA tree (Figure 3B). Three non-parametric multivariate statistical tests (ANOSIM, adonis, and MRPP) further indicated the significant influence (P < 0.05) of the BSMs on the bacterial community structure (Table 1). However, the influence of the BSMs on the bacterial community structure was weakened after 35 days of pepper growth based on the observation that the UniFrac distances, a measurement on the dissimilarity in the bacterial community structure, between BSM-amended treatments and CK in the post-planting soil samples were significantly (P < 0.05) smaller than in the pre-planting soil samples (Figure 3C).

Furthermore, in the treatments where the plant disease index was similar, the dissimilarity in the bacterial community structure was smaller based on the observation that the UniFrac distances between CAME and CK (Figure 3C), where the plant disease index on day 35 was similar (Figure 1), was significantly (P < 0.05) smaller than that between PG and CK and between PG+IG and CK (Figure 3C), where the plant disease index was distant (Figure 1) after 35 days of plant growth. Another interesting result was that 35 days of pepper growth had no significant influence on the bacterial community structure when no BSM was added but that it exerted a significant influence on the bacterial community structure in BSM-amended soil (Figures 3A,B and Table 1).

The changes in some specific bacterial populations resulting from BSM application may be related to differences in plant disease severity. To further understand the specific bacterial population that may be associated with the plant disease severity, the relative abundance of the phylotypes was summarized at the phylum or lower taxonomic level. Then, the correlation between the relative abundance of specific populations at the phylum or lower taxonomic level and observed disease index was analyzed.

The shift in the relative abundance of the main phyla/classes (those with relative abundance > 1% under at least one treatment were included) are shown in Figure 4. Of these main phyla/classes (a total of 15), 12 and 11 phylotypes were significantly changed in relative abundance by at least one BSM treatment in pre-planting soil and post-planting rhizosphere soil samples, respectively (Figure 5). Pearson correlation analysis revealed that the relative abundance of Actinobacteria and Gammaproteobacteria in pre-planting soil samples was negatively correlated (P < 0.05) with the day-35 disease index, with the correlation being strongest in Actinobacteria (r = -0.953, P < 0.05), while that of Chloroflexi and Gemmatimonadetes was positively correlated with the day-35 disease index, with the correlation being strongest in Chloroflexi (r = 0.760, P < 0.05) (Table 2). The significant (P < 0.05) correlation between these four phylotypes and the day-35 disease index still existed after 35 days of pepper growth, indicating their possible function in manipulating plant disease (Table 2). As the bacterial biomass, indicated by the abundance of the 16S rRNA gene copies was not significantly influenced by BSM treatment (Supplementary Figure S1), changes in relative abundance could then reflect the absolute abundance of specific phylotypes.

The relative abundance of each phylotype was analyzed at the genus level to understand which bacterial populations were affected by BSM at a finer taxonomic level. Then, the correlation between the relative abundance of the main genera and the disease index was analyzed to explore the possible relationship between bacterial population shifts and plant disease.

Among the main genera (with relative abundance > 0.5% in at least one treatment), a total of 34 genera was significantly affected by at least one BSM treatment in both pre-planting and post-planting soil samples in terms of relative abundance (Figures 6A,B and Supplementary Figure S2). Of these significantly affected genera, the relative abundance of the following 4 genera from pre-planting soil samples, namely, two Streptomycetaceae-affiliated genera within Actinobacteria, i.e., Streptomyces and an unclassified genus; the genus Balneimonas within Alphaproteobacteria, and the genus Pseudoxanthomonas within Gammaproteobacteria, was significantly (P < 0.05) increased in the PG and PG+IG treatments (Figure 6A) where the plant disease was suppressed, but these genera (except for Pseudoxanthomonas) were not affected in CAME treatment (Figure 6A) where the day-35 plant disease index was similar to that observed in CK. Streptomyces and an unclassified genus within Actinobacteria were the strongest PG- or PG+IG-induced genera in both pre-planting and post-planting soil samples. Pearson correlation analysis further revealed a significantly negative correlation (r ranged from -0.822 to -0.864, P < 0.05) between the relative abundance of these 4 phylotypes in pre-planting soil samples and the day-35 plant disease index (Figure 6C). This significant correlation still existed after 35 days of pepper growth (r ranged from -0.938 to -0.819, P < 0.05). These 4 genera were hereinafter called significantly and negatively related phylotypes. In contrast, the relative abundance of the other 2 genera from pre-planting soil samples, namely, an unclassified genus within Acidobacteria and an unclassified genus within Gemmatimonadetes, was significantly decreased in PG-containing treatments (i.e., PG and PG+IG), but their relative abundance was not affected in the CAME treatment (Figure 6B). Pearson correlation analysis further indicated a significantly positive correlation (r = 0.847 for the Acidobacteria-affiliated genus, r = 0.791 for the Gemmatimonadetes-affiliated genus; P < 0.05) between the relative abundance of these 2 pre-planting phylotypes and the day-35 plant disease index (Figure 6C). The significant correlations still existed after 35 days of pepper growth (r = 0.788 for the Acidobacteria-affiliated genus, r = 0.702 for the Gemmatimonadetes-affiliated genus; P < 0.05). These 2 genera were hereinafter called significantly and positively related phylotypes. As no significant changes were found in terms of total bacterial biomass, as assessed by quantifying the 16S rRNA gene using real-time quantitative PCR, under BSM treatments (Supplementary Figure S1), relative abundance data could then reflect the changes in the absolute abundance of specific taxa.

Streptomyces and an unclassified genus within Actinobacteria were the strongest PG- or PG+IG-induced genera in both pre-planting and post-planting soil samples. The unclassified genus within Actinobacteria exhibited approximately 15- and 14-fold increases in relative abundance within 25 days of PG and PG+IG amendment, respectively, with its relative abundance increasing from 0.23% under CK to 3.62% and 3.41% under PG and PG+IG treatments, respectively (Figure 6A). This elevated relative abundance was maintained after 35 days of pepper growth, with its relative abundance increasing from 0.23% under CK to 3.54% under PG and 2.65% under PG+IG (Figure 6B). Similarly, the Actinobacteria-affiliated Streptomyces showed a significant increase (approximately a 2–4 times increase) in its relative abundance in PG and PG+IG in pre-planting and post-planting soil samples, with its relative abundance increasing from 0.65-0.80% under CK to 3.54–3.98% and 2.48–3.11% under PG and PG+IG treatments, respectively, in pre-planting and post-planting soil samples (Figures 6A,B). Correlation analysis further revealed that the relative abundance of these two phylotypes in both pre-planting and post-planting rhizosphere soil samples showed the strongest relationship with the day-35 plant disease index (Figure 6C).