The pot experiment was conducted from May 2023 to April 2024 in the greenhouse of the Spice and Beverage Research Institute at the Chinese Academy of Tropical Agricultural Sciences. Greenhouse conditions were maintained at 22–25°C during the daytime, 20–25°C at nighttime, with 72% relative humidity and 75% light transmittance. The soil used in the pot experiments was naturally infested soil collected from a vanilla plantation at the Spice and Beverage Research Institute, located in Longxing Town, Wanning City, Hainan Province, China (110°19′E−110°22′E, 18°12′N−18°16′N). This site has been continuously cultivated with vanilla for over 10 years, with an annual Fusarium wilt disease incidence of 65%. The soil texture was classified as sandy loam, with the following physicochemical properties: pH 5.58, organic matter (OM) 25.82 g-kg⁻¹, alkali-hydrolyzable nitrogen (AN) 100.32 mg-kg⁻¹, available phosphorus (AP) 20.46 mg-kg⁻¹, and available potassium (AK) 150.46 mg-kg⁻¹.

Prior to the experiment, all soil was transported to the greenhouse and sieved through a 2-mm mesh. The soils were under rotation with black pepper, pandan, and sweet rice tea. After 1 year of crop rotation, the soils were collected and vanilla was planted. The soils cropped with black pepper, pandan, and sweet rice tea were designated as black pepper–vanilla rotation (H), pandan–vanilla rotation (B), and sweet rice tea–vanilla rotation (C), respectively. Soils with vanilla monoculture were labeled as X, while fallow soils served as controls (CK). Notably, this study specifically focused on evaluating the disease suppression capacity of vanilla replanted after crop rotation. Therefore, the post-rotation soils were excluded from subsequent analysis.

The vanilla variety tested was Mexican vanilla (Vanilla planifolia Andrews). Vanilla is propagated asexually, so we planted 60 cm long mature stems from high-quality parent plants. The experiment was based on a randomized block design. Each treatment was replicated three times to ensure the reliability of the results. Each replicate consisted of six pots, with three plants per pot. Each pot was filled with 15 kg of soil, and the pot dimensions were 32 cm × 25 cm, ensuring sufficient space between each plant and independence to avoid potential cross-contamination. Each pot received 90 g of organic fertilizer, which was carefully mixed with the soil using a small shovel to ensure even distribution. The organic fertilizer had a pH of 7.51 and contained 1.33% nitrogen (N), 2.54% P₂O₅, and 1.23% K₂O. The chemical fertilizer application was based on traditional agricultural practices. Soil samples were collected from the rhizosphere compartment after 12 months of cultivation. Notably, this experiment was focused on evaluating the disease suppression effects of vanilla replanted after crop rotation. Therefore, the soils after crop rotation were not subjected to further research. The disease incidence (DI) of vanilla due to Fusarium wilt was statistically assessed on the basis of typical symptoms and was calculated as the percentage of affected plants out of the total surveyed (Xiong et al., 2016).

Seven pots were randomly selected from each treatment, and the three vanilla plants in each pot were combined to form one sample. Each treatment resulted in seven composites soil samples. The vanilla plants were carefully removed from the soil, immediately preserved in an ice box at 4°C, and transported promptly to the laboratory. The soil closely associated with the root system was dislodged by gently shaking the plant roots, and this soil was designated rhizosphere soil (Chen et al., 2020). In total, 35 rhizosphere soil samples were obtained. Each sample was accurately weighed to 0.4 grams for DNA extraction using the PowerSoil DNA Isolation Kit from QIAGEN/MoBio (MoBio Laboratories, Carlsbad, CA, USA). The extraction was performed according to the manufacturer's protocol, ensuring meticulous and error-free execution of the procedure. The extracted DNA samples were subsequently stored at −70°C for future analysis.

In accordance with our previous report (Hong et al., 2020), soil pH was measured in a soil–water suspension (1:2.5 w/v) via a glass electrode attached to a pH meter after shaking for 30 min. The organic matter (OM) content was determined using the potassium dichromate titration method, where potassium dichromate oxidizes organic carbon, and the organic matter content is calculated by titrating the dichromate remaining. Available nitrogen (AN) was measured using the alkaline diffusion method, where sodium hydroxide is used to treat the soil, and the resulting ammonia is absorbed by boric acid, followed by titration with standard acid. Available phosphorus (AP) was extracted with sodium bicarbonate and determined by the molybdenum blue method. Available potassium (AK) was extracted with ammonium acetate and measured using flame photometry.

The overall abundance of F. oxysporum in the rhizosphere soil of vanilla was ascertained via real-time quantitative PCR (qPCR), employing the pathogen-specific primer sets AFP308R (CGAATTAACGCGAGTCCCAAC) and ITS1F (CTTGGTCATTTAGAGGAAGTAA) as previously described (Xiong et al., 2017). A standard curve was constructed via a serial dilution method using a plasmid harboring the internal transcribed spacer (ITS) region of F. oxysporum. Quantitative PCR was performed on the Applied Biosystems QuantStudio 7 Flex, following a standardized protocol for template amplification. The PCR mixture had a final volume of 20 μl, containing 10 μl of SYBR^® Premix Ex Taq™ (2 × ), 0.4 μl of each primer (final concentration of 10 μM), 0.4 μl of ROX Reference Dye II (50 × ), 1 μl of DNA template (~20 ng/μl), and nuclease-free water to the full volume. The performance of PCR amplification was assessed by examining the dissociation curve and determining the amplification efficiency. Triplicate analyses were conducted for each sample, and the data were log-transformed to express the results as log-transformed copy numbers per gram of dry soil.

The V4 region of the 16S rRNA gene from soil bacteria was amplified using the primers 520F (AYTGGGYDTAAAGNG) and 802R (TACNVGGGTATCTAATCC; Caporaso et al., 2011). For soil fungi, the ITS1 region was amplified with the primers ITS5F (GGAAGTAAAAGTCGTAACAAGG) and ITS1R (GCTGCGTTCTTCATCGATGC; Schoch et al., 2012). Unique barcodes/linkers and adapters were incorporated during the amplification process to prepare samples for sequencing. The PCR mixture, totaling 25 μl, included 5 × reaction buffer (5 μl), 5 × GC buffer (5 μl), 10 μM primers (1 μl), template DNA (2 μl), 100 mM dNTPs (5 μl), and nuclease-free water to the full volume. The PCR amplification parameters consisted of initial denaturation at 98°C for 2 min, followed by 28 to 30 cycles of denaturation at 98°C for 15 s, annealing at 55°C for 30 s for bacteria and at 50°C for 30 s for fungi, with extension at 72°C for 30 s.

Following PCR amplification, the resulting products were purified using the QIAquick PCR Purification Kit (QIAGEN, Germany). The concentration of the purified DNA samples was measured using a Qubit^® 2.0 fluorometer (Invitrogen, USA). The sequencing libraries were prepared from equimolar pooled samples using the NEBNext^® Ultra™ DNA Library Prep Kit for Illumina (New England Biolabs, UK), after which they were read for sequencing. The quality of the constructed libraries was assessed and verified via an Agilent 2,100 Bioanalyzer (Agilent Technologies, USA) and a KAPA Library Quantification Kit (KAPA Biosystems, USA). After library construction, the library was loaded onto the flow cell of the Illumina MiSeq sequencer. The flow cell surface underwent bridge PCR amplification to form numerous DNA clusters. During sequencing, the sequencer sequentially added nucleotides labeled with fluorescent markers to the DNA chains. Each incorporated nucleotide emitted a specific fluorescent signal, and the DNA sequence was determined by detecting the order of these signals. The entire process from library construction to sequencing was executed and completed by Shanghai Personal Biotechnology Co., Ltd.

The raw sequencing data were subjected to initial quality filtering with Trimmomatic (version 0.33) to exclude low-quality sequences. The primer sequences were removed via Cutadapt (version 1.9.1) (Martin, 2011). USEARCH (version 10) was used to merge paired-end sequences (Edgar, 2013), and UCHIME (version 8.1) was used to remove chimeras (Edgar et al., 2011), ensuring high-quality sequences for further analysis. Denoising of quality-controlled data was performed using the DADA2 algorithm (Callahan et al., 2016) in QIIME2 (2020.6 version) (Bolyen et al., 2019), with a 0.005% threshold to filter anomalous sequences. Operational taxonomic units (OTUs) were clustered at a 97% sequence similarity threshold and taxonomically classified using the RDP classifier against the RDP Bacterial 16S rRNA (Wang et al., 2007) and UNITE Fungal ITS (Kõljalg et al., 2013) databases.

After quality control, 12,250 bacterial and 3,756 fungal OTUs were identified. The Chao 1 and Shannon indices were used to assess species richness and community diversity, respectively. Beta diversity was analyzed using principal coordinate analysis (PCoA), multiple regression tree (MRT; De'Ath, 2002), and heatmap clustering on the basis of the Bray–Curtis distance. Permutational multivariate analysis of variance (PERMANOVA) and analysis of similarity (ANOSIM) were used to test for differences in community structure. Hellinger standardization was performed on the rarefied OTU table before beta diversity was analyzed. The relative abundance of each OTU per sample was calculated as the number of sequences affiliated with the target OTU divided by the total number of sequences. Structural equation models (SEMs), constructed with the GGally, lavaan, and semPlot packages, were used to evaluate the influence of bacterial and fungal communities on pathogens (Mamet et al., 2019). In the structural equation model, soil microbial community structure (PCoA1 scores) and diversity (Shannon index) were included as measured variables, with the abundance of F. oxysporum as the response variable.

We first applied linear discriminant analysis effect size (LEfSe) with a threshold of scores ≥3 to identify differential microbial biomarkers across various spice crops (Hong et al., 2020). Subsequently, the predictive importance of these biomarkers in suppressing F. oxysporum was evaluated using a random forest model implemented through the R package randomForest (Wen et al., 2023b). We subsequently identified hub OTUs among the rotated crops. The psych package in R was used for graphical representation, and data points with correlation coefficients above 0.8 and P < 0.05 were selected to construct a bacterial–fungal co-occurrence network (Revelle and Revelle, 2015). The topological properties of this network were assessed using GEPHI software for visualization (Bastian et al., 2009). A high average degree of nodes indicated increased network complexity, implying a more intricate interplay among microbial species. Hub OTUs was identified on the basis of a customized threshold reflecting node characteristics: a degree > 7 and closeness centrality > 0.6, which accounted for <15% of all nodes in the network (Costello et al., 2012; Agler et al., 2016; Banerjee et al., 2018). Ultimately, the core microbiome was identified via the intersection of biomarker microbes identified through LEfSe and random forest analysis with hub microbes identified through network analysis (Hong et al., 2023).

A comparison of pathogen abundance, microbial diversity, taxonomic composition, and soil physicochemical factors was conducted using Duncan's multiple range test in the agricolae package of R. We employed the vegan package in R to conduct both linear and nonlinear regression analyses, thereby investigating the relationship between pathogen abundance and microbial diversity. The rdacca.hp R package was employed to evaluate the relative importance of soil physicochemical factors as explanatory variables influencing bacterial and fungal microbial communities (Lai et al., 2022). Furthermore, we applied a general linear model (GLM) using the relaimpo package in R to evaluate the contributions of bacterial and fungal communities, core microbes, and physicochemical factors to pathogen suppression (Shen et al., 2019).

The sequences have been submitted to the Sequence Read Archive (SRA) of the National Center for Biotechnology Information under the accession number PRJNA1126394.

After a decade of continuous vanilla cultivation, a rotation with black pepper, pandan, and sweet rice tea was performed before replanting with vanilla. We found that the DI of vanilla was 55% with continuous cultivation, but it significantly decreased to 27% after rotation with black pepper, 22% with sweet rice tea, and 17% with pandan (Figure 1A). F. oxysporum abundance was significantly lower in the pandan–vanilla and sweet rice tea–vanilla rotation systems (P < 0.05), with no significant difference between the two treatments (Figure 1B).

Compared with the vanilla monoculture, crop rotation with black pepper, sweet rice tea, or pandan did not significantly affect soil bacterial richness or diversity, but both richness and diversity were significantly greater in the rotations than those in the fallow treatment (Supplementary Figures S1A, B). The richness and diversity of the fungal community significantly increased with rotation (Supplementary Figures S1C, D) and were significantly negatively correlated with pathogen abundance (Supplementary Figures S1G, H). Principal coordinate analysis (PCoA) based on Bray–Curtis distances clearly revealed significant differences in the soil fungal communities among the fallow, vanilla monoculture, and crop rotation treatments (Figure 1C, left). However, no distinct separation was observed among the rotations of black pepper, pandan, and sweet rice tea. This finding was confirmed by PERMANOVA (Figure 1C, R² = 0.38, P = 0.001) and ANOSIM (Supplementary Table S1, P = 0.001). Moreover, the fungal community (PCoA1) was significantly negatively correlated with the abundance of F. oxysporum (Figure 1C, middle). Further cluster heatmap analysis indicated that fallow and vanilla monoculture could be clustered separately, whereas the rotations of pepper, pandan and sweet rice tea had similar community structures and were clustered together (Supplementary Figure S2B). The trend in the soil bacterial community structure was consistent with that of the fungal community structure (Figure 1D).

ITS sequencing revealed that, compared with vanilla monoculture, crop rotation with pepper, pandan and sweet rice tea significantly increased the relative abundance of the phylum Ascomycota (Figure 1C, right) and was significantly negatively correlated with F. oxysporum abundance (Supplementary Table S2). The genera Thermomyces and Arthrobotrys were significantly more abundant after rotation, with Arthrobotrys increasing 9.8-fold after pandan rotation, and both genera were negatively correlated with the pathogen (Supplementary Table S3). 16S rRNA gene sequencing revealed that crop rotation significantly increased the abundance of Proteobacteria, Gemmatimonadetes, and Latescibacteria (Figure 1D, right), which was also negatively correlated with F. oxysporum abundance (Supplementary Table S2). At the genus level, Nitrosospira, Gemmatimonas, and Gp17 were significantly more abundant in rotation systems, particularly those involving pandan and sweet rice tea, and were negatively associated with the pathogen (Supplementary Table S4).

MRT analysis revealed that cropping system was the main factor shaping rhizosphere fungi, explaining 26.81% of the variance, whereas crops in rotation were secondary factors, accounting for 9.29% of the variance (Figure 2A). For bacteria, crop rotation was the primary factor, explaining 61.23% of the variance (Figure 2B).

A general linear model (GLM) was used to assess the impact of fungal and bacterial community structure and diversity on the suppression of F. oxysporum abundance, quantified via qPCR-based absolute quantification (Figure 2C). Both the fungal and bacterial principal coordinate axes significantly reduced pathogen abundance, with the fungal community being the major factor, accounting for 24.95% of the variance. The second axis of the bacterial community also significantly contributed, accounting for 13% of the variance. SEM revealed that bacterial community structure and diversity positively influence fungal community structure (r = 0.22 for structure, r = 0.40 for diversity), which in turn significantly decreased the abundance of F. oxysporum quantified via qPCR (r = −0.79; Figure 2D).

We used a more stringent LDA score threshold of ≥3 to identify biomarkers between groups. We then displayed the top 20 fungal (Figure 3A) and bacterial (Figure 3B) OTU taxa by relative abundance. Among fungi, 20 biomarkers were identified (Supplementary Table S5). Specifically, the fallow treatment enriched 5 biomarkers, mainly from the phyla Ascomycota and Basidiomycota; vanilla monoculture enriched 1 Ascomycota biomarker; black pepper rotation enriched 8 biomarkers, primarily from the phyla Ascomycota and Zygomycota; pandan rotation enriched 3 Zygomycota biomarkers; and sweet rice tea rotation enriched 3 biomarkers from the phyla Ascomycota and Zygomycota. Notably, black pepper rotation significantly enriched OTU1_Thermomyces (F = 4.682, P = 0.005) and OTU37_Arthrobotrys (F = 4.188, P = 0.008), pandan rotation significantly enriched OTU18_Arthrobotrys (F = 3.326, P = 0.023), and sweet rice tea rotation significantly enriched OTU16_Mortierella (F = 2.800, P = 0.044). The relative abundances of these biomarkers were significantly negatively correlated with F. oxysporum abundance (P < 0.05).

For bacteria, we identified a total of 147 biomarkers (Supplementary Table S6). Specifically, the fallow treatment enriched 27 biomarkers, mainly from the phyla Proteobacteria and Actinobacteria; vanilla monoculture enriched 22 biomarkers, primarily from the phyla Proteobacteria and Actinobacteria; black pepper rotation enriched 26 biomarkers, mainly from the phyla Proteobacteria and Acidobacteria; pandan rotation enriched 34 biomarkers, primarily from the phyla Proteobacteria and Acidobacteria; and sweet rice tea rotation enriched 38 biomarkers, predominantly from the phylum Proteobacteria.

Random forest models identified significant predictors of F. oxysporum response. In the fungal model (Figure 4A, R² = 0.331, P < 0.01), black pepper–vanilla rotation enriched OTU1_Thermomyces and OTU37_Arthrobotrys, and pandan–vanilla rotation enriched OTU18_Arthrobotrys significantly. For bacteria (Figure 4B), the black pepper–vanilla rotation model (R² = 0.314, P < 0.01) highlighted OTU235_Gp16, OTU145_Gp11, and OTU158_Gp17 as predictors. In the pandan–vanilla rotation model (R² = 0.625, P < 0.01), multiple bacterial OTUs were identified as predictors, including OTU11_Gp6, OTU5794_Magnetospirillum, OTU78_Zoogloea, OTU33_Lacibacterium, OTU22_Nitrosospira, and OTU109_Latescibacteria. The sweet rice tea–vanilla rotation model (R² = 0.465, P < 0.01) also revealed several significant bacterial predictors, such as OTU200_Methylibium, OTU113_Roseisolibacter, OTU178_Actinospica, OTU56_Lacibacterium, OTU18_Nitrospira, and OTU90_Micromonospora. Compared with those in the fallow and vanilla monoculture treatment, the relative abundances of the identified biomarkers significantly increased (Supplementary Figure S3, P < 0.05).

Co-occurrence network analysis of rhizosphere bacteria and fungi revealed distinct structures under fallow, vanilla monoculture, and rotation treatments (Figure 5). The pandan–vanilla rotation network had the highest node and connection counts, with 96% positive correlations, and the highest average node degree was 2.78, indicating strong microbial interactions (Supplementary Table S7). On the basis of node degree and closeness centrality (Table 1), in the pandan–vanilla rotation, 8 bacterial hub taxa, such as OTU31, OTU99, OTU178, OTU10, OTU14, OTU22, OTU56, and OTU130, were uniquely identified. The 8 identified hub taxa had relative abundances >0.2%, with values of 0.53%, 0.27%, 0.28%, 1.26%, 0.87%, 1.01%, 0.27%, and 0.38%, respectively. Notably, OTU22, OTU56, and OTU178, identified as enriched biomarkers by random forest analysis, also qualified as hubs in the network analysis, designating them as core microbiota. OTU22, specifically, emerged as a pandan–vanilla rotation-enriched, host-specific core microbe.

Following crop rotation with spice crops, we measured soil physicochemical properties and found that, compared with those under vanilla monoculture cultivation, soil pH, OM, AN, AP, and AK significantly increased under spice rotation (Figure 6A, P < 0.05). However, soil pH and OM did not differ significantly between the pandan and sweet rice tea rotation systems. Soil pH and OM significantly influenced both fungal and bacterial communities, with soil pH being the primary factor, explaining 10.8% and 42.2% of the variance in fungal and bacterial communities, respectively (Figures 6B, C).

A general linear model was used to evaluate the potential of microbial communities, core microbiota and physicochemical properties in suppressing pathogens. The model explained 71.02% of the variation in pathogen suppression (Table 2, P < 0.001). Key factors included fungal community structure, the core microbial species OTU22, and soil pH, with individual explanatory contributions of 18.71%, 7.59%, and 17.83%, respectively (P < 0.05 for all).