Introduction
The intricate world of plant health is profoundly influenced by phytobiomes – the complex communities of microorganisms inhabiting the rhizosphere (root zone), phyllosphere (leaf surface), and endosphere (internal tissues). These microbial communities play a crucial role in plant growth and resilience, impacting nutrient absorption, pathogen defense, and overall physiological processes. The study of phytobiomes has become increasingly vital, with researchers diligently working to identify specific microbes associated with desirable plant traits and understand how these microbial ecosystems respond to environmental changes. This understanding is key to harnessing the power of plant-associated microbiomes for enhanced and sustainable crop production.
Citrus, a globally significant fruit crop, exemplifies the critical balance between agricultural productivity and plant health. With annual yields exceeding 120 million tons, citrus fruits are not only economically important but also vital sources of nutrients, vitamins, and dietary fiber. However, citrus production faces escalating challenges from climate change and devastating diseases. Among these, Huanglongbing (HLB), also known as citrus greening, stands out as the most severe threat. HLB’s widespread prevalence and the substantial yield losses it causes have spurred renewed scientific and agricultural interest in finding effective solutions. HLB disrupts citrus trees by impairing phloem function, reducing photosynthate transport, and upsetting nutrient balance, primarily due to infection by the bacterium Candidatus Liberibacter asiaticus (CLas). As CLas is notoriously difficult to culture in isolation, understanding the broader citrus-associated microbiome dynamics under HLB pressure is paramount.
A deeper comprehension of citrus phytobiomes holds the promise of developing eco-friendly and sustainable strategies to bolster citrus health and productivity. Global research efforts are progressively unveiling the complex patterns within citrus phytobiomes. The International Citrus Microbiome Consortium, for instance, has conducted extensive sequencing analyses of citrus rhizosphere and soil microbiomes across six continents. Recent advancements in 16S rRNA gene sequencing have further highlighted how HLB infection reshapes the microbial communities in citrus rhizospheres and phyllospheres. However, much of the existing research has focused either on healthy or HLB-infected citrus microbiomes in isolation, with fewer studies directly comparing both states. The fundamental question remains: are there universal microbial responses to HLB across diverse citrus varieties and global growing regions?
This study embarks on a systematic review of available citrus phytobiome data, contrasting the bacterial communities of healthy and HLB-infected citrus plants. Critically, we employ symbiotic machine learning approaches to identify potential biomarkers for HLB. By meticulously accounting for technical variations, geographical distribution, and tissue specificity, we aim to reveal the intricate dynamics of bacterial communities in HLB-affected citrus. Ultimately, our goal is to develop a symbiotic machine learning model, leveraging phytobiome data, to accurately predict HLB outbreaks in real-world agricultural settings. This research not only advances our understanding of plant-microbiome interactions but also pioneers the application of symbiotic machine learning in precision agriculture for disease management.
Materials and Methods
Data Collection and Description
To build a comprehensive dataset, we meticulously screened the National Center for Biotechnology Information Sequence Read Archive database, using the keywords “citrus” and “huanglongbing”. This search yielded six HLB-related citrus microbiome bio-projects encompassing 1,385 bacterial samples, predominantly from HLB-infected citrus (1,332 samples) and a smaller set from healthy plants (53 samples). To further enrich our dataset with healthy plant microbiomes, we included seven additional bio-projects, contributing 802 bacterial samples from healthy citrus. The detailed metadata for these bio-projects are available in Supplementary Table S1. We categorized the metadata based on tissue/material source into five groups, as detailed in Supplementary Table S2. Due to the limited number of samples from budwood, bulk soil, and attached insects, our subsequent analyses focused primarily on leaf and rhizosphere datasets (Supplementary Table S2). To ensure data quality and consistency, we removed samples with fewer than 3,000 reads to eliminate potential anomalies from sequencing errors. After rigorous filtering, our final dataset comprised 806 citrus microbiome samples (Figure 1A): 29 healthy citrus leaf samples, 207 healthy citrus rhizosphere samples, 267 HLB-infected citrus leaf samples, and 303 HLB-infected citrus rhizosphere samples (Supplementary Table S3). Analysis of the amplified regions revealed that bacterial ITS (46.28%) and 16S V4 (45.78%) were the most frequently sequenced regions (Figure 1B). The majority of sequencing data was generated using the Illumina MiSeq platform, with only 20 samples sequenced on the Illumina HiSeq X Ten system (Figure 1C).
Data Processing
We processed the collected FASTQ format datasets using the QIIME2 (Quantitative Insights Into Microbial Ecology 2) pipeline, a standard tool in microbiome research. This involved a series of steps to ensure data accuracy and reliability. Initially, primer sequences were removed, and stringent quality control measures were applied to filter out low-quality reads. The resulting high-quality reads were then clustered into amplicon sequence variants (ASVs) using the DADA2 plug-in unit. Taxonomic assignment of each ASV was performed using a closed-reference strategy against the SILVA database (release 138), a comprehensive and widely respected resource for ribosomal RNA gene sequences. This approach uses a predefined reference database of full-length amplified target sequences to classify sequences generated by different primers, as described by Yu et al. (2018). To further refine our dataset, we removed non-bacterial ASVs, such as chloroplasts and archaea, as well as singletons (ASVs with only one read), which are often indicative of sequencing errors. Finally, to account for variations in sequencing depth across samples, we rarefied the ASV abundance tables to 3,000 reads per sample. This normalization step ensures that differences in microbial community composition are not artifacts of uneven sequencing effort.
Statistical Analyses
All statistical analyses were performed using R (v4.0.2), a powerful and versatile statistical computing environment. Data visualization was achieved using the “ggplot2” package in R, known for its ability to create informative and aesthetically pleasing graphics. To assess the diversity within microbial communities (alpha diversity), we calculated three key indices using the “vegan” package: Chao1 (richness, estimating the number of species), Shannon’s (diversity, considering both richness and evenness), and Pielou’s J (evenness, measuring the equitability of species abundances). We employed the t-test function to determine statistically significant differences in alpha diversity indices and the relative abundance of bacterial phyla and genera between healthy and HLB-infected citrus leaf and rhizosphere samples. To analyze the overall structure of bacterial communities (beta diversity), we calculated Bray–Curtis distances, a common metric in ecological studies, using the vegdist function in the “vegan” package. These distances were then used in a principal coordinate analysis (PCoA) with the pcoa function in the “ape” package. PCoA is a dimensionality reduction technique that visually represents the similarities and dissimilarities between samples in a reduced space. To statistically test for significant differences in bacterial community composition between healthy and HLB-infected samples, we performed a permutational multivariate analysis of variance (PERMANOVA) using the adonis function in the “vegan” package. We used the “VennDiagram” package to conduct Venn diagram-based analysis to identify shared bacterial ASVs among different sample groups. To compare the total relative abundance of shared ASVs across groups, we used ANOVA followed by Tukey’s HSD post-hoc test (“multcomp” package). Finally, to investigate the ecological processes driving microbiome assembly, we applied a neutral community model. This model helps to distinguish between deterministic processes (niche-based selection) and stochastic processes (random events) in shaping community structure. The model provides two key parameters: m, representing dispersal between communities, and R2, indicating the proportion of community variance explained by stochastic processes, as described by Sloan et al. (2006).
Machine-Learning Modeling
To develop a robust and accurate method for distinguishing between bacterial communities of HLB-infected and healthy citrus plants, we constructed predictive models using symbiotic machine learning. We utilized ten established machine learning algorithms, as outlined by Gupta et al. (2021), known for their effectiveness in classification tasks: logistic regression, decision tree, k-nearest neighbor, bagging, gradient boosting, Bayes classification, artificial neural network, conditional inference tree, random forests, and support vector machines. We built separate models for leaf and rhizosphere microbiomes to account for tissue-specific differences. For each model, the bacterial relative abundances, ranging from phylum to species level, served as input features. To train and validate our models, we randomly divided the datasets into training and testing sets. 70% of the healthy and HLB-infected citrus samples were allocated to the training dataset, used to build the models. The remaining 30% formed the testing dataset, used to evaluate model performance on unseen data. Model performance was assessed by comparing predicted health status with the actual health status using two key metrics: receiver operating characteristic (ROC) curves and the area under the ROC curve (AUC) scores. ROC curves visually represent the trade-off between true positive rate and false positive rate at various classification thresholds, while AUC scores provide a single numerical measure of overall model performance, with higher AUC values indicating better discriminatory ability. We selected the best-performing models based on achieving the highest AUC scores and accuracies in predicting both healthy and HLB-infected samples. Furthermore, we determined the importance of each bacterial feature (taxon) in the classification process for the best-performing models, providing insights into the key microbial biomarkers for HLB detection.
Results
Differences in the Microbiome Diversity of HLB-Infected and Healthy Citrus Samples
Our meta-analysis, encompassing sequencing data from 806 bacterial samples collected across six continents, generated a comprehensive merged bacterial ASV table containing over 3,700 distinct taxa. Taxonomic annotation revealed that all bacterial ASVs could be classified at the phylum level. However, finer-grained classification was less complete, with only 67.45% annotated at the genus level and 25.5% at the species level (Figure S1). To ensure comparability across samples, we rarefied the sequencing data to 3,000 reads per sample prior to calculating alpha diversity indices. Our analysis revealed a clear trend in alpha diversity: it was lowest in HLB-infected citrus leaf microbiomes and highest in healthy citrus rhizosphere microbiomes (Figure S2). Crucially, we found statistically significant reductions in Chao1, Shannon’s, and Pielou’s J indices for both leaf and rhizosphere bacterial communities in HLB-infected citrus samples compared to healthy samples (Student’s t-test, p < 0.05; Figures 2A, B). Interestingly, we observed greater inter-individual variability in alpha diversity indices among HLB-infected samples compared to healthy samples, suggesting a more disrupted and less stable microbial community structure in diseased plants (Figures 2A, B). Consistent with the alpha diversity findings, principal coordinate analysis (PCoA) and permutational multivariate analysis of variance (PERMANOVA) based on Bray-Curtis distances demonstrated significant differences in the overall bacterial community structures between HLB-infected and healthy citrus leaf and rhizosphere samples (p < 0.05; Figures 2C, D). Furthermore, mirroring the alpha diversity patterns, we observed greater inter-individual variation in both leaf and rhizosphere bacterial communities among HLB-infected samples compared to healthy samples (Figures 2C, D).
Taxonomic Classification of Bacteria in Citrus Leaf and Rhizosphere Microbiomes
At the phylum level, Proteobacteria emerged as the most dominant bacterial group in both citrus leaf and rhizosphere samples. Following Proteobacteria, Cyanobacteria was the second most abundant phylum in leaf microbiomes, while Actinobacteriota was the second most dominant in rhizosphere microbiomes (Figure 3A). Comparing HLB-infected and healthy samples, we observed distinct shifts in phylum relative abundances. Actinobacteriota and Firmicutes were significantly more abundant in HLB-infected citrus rhizospheres compared to healthy rhizospheres. Conversely, Proteobacteria and Bacteroidota showed a contrasting pattern, being more abundant in healthy citrus rhizospheres (Figure 3B). In citrus leaves, HLB infection was associated with a higher relative abundance of Proteobacteria. In contrast, healthy leaves exhibited higher relative abundances of Cyanobacteria, Actinobacteriota, and Bacteroidota (Figure 3C). To identify core microbial taxa, we performed a Venn diagram-based analysis, revealing 186 bacterial taxa shared across all sample groups (Figure 3D). However, the total relative abundances of these shared bacteria differed significantly between healthy and HLB-infected samples in both leaf and rhizosphere microbiomes. Specifically, the total relative abundances of shared bacteria were significantly higher in HLB-infected samples compared to healthy samples (Tukey’s HSD test, p < 0.05; Figure 3E). Further analysis of the shared bacteria revealed that those with increased abundances in HLB-infected samples predominantly belonged to Proteobacteria, Firmicutes, and Nitrospinota (in leaf microbiomes) or Actinobacteriota (in rhizosphere microbiomes) (Figure 3F).
Analysis of Microbiome Assembly Mechanisms of HLB-Infected and Healthy Citrus Samples
To understand the ecological mechanisms driving the observed differences in microbiome composition, we investigated the relative roles of niche-based deterministic processes and random stochastic processes in shaping bacterial community assembly in HLB-infected and healthy citrus samples. Using a neutral community model, we found contrasting patterns between healthy and diseased plants, and between rhizosphere and phyllosphere. In healthy citrus rhizospheres, the neutral community model explained a substantial proportion of the variance in bacterial community composition (R2 = 0.727), indicating that stochastic processes, such as dispersal and ecological drift, play a dominant role in shaping these communities. However, in HLB-infected citrus rhizospheres, the explanatory power of the neutral community model was markedly reduced, with only 37.9% of the variance explained (R2 = 0.379). This suggests a shift towards deterministic processes in diseased rhizospheres. In contrast, for citrus leaves, the neutral community model explained a lower proportion of variance in both HLB-infected and healthy samples compared to rhizospheres. In HLB-infected citrus leaves, the model explained only 19.4% of the variance (R2 = 0.194), while in healthy citrus leaves, it explained 37.6% (R2 = 0.376). These findings suggest that deterministic processes are more influential in structuring leaf bacterial communities compared to rhizosphere communities, regardless of health status. Importantly, across both rhizospheres and phyllospheres, HLB infection consistently decreased the contribution of stochastic processes to bacterial community assembly, indicating a disease-induced shift towards more deterministic, niche-driven community structures.
Bacterial Communities Useful for Distinguishing Between HLB-Infected and Healthy Citrus Samples
To explore the potential of bacterial community characteristics as biomarkers for HLB detection, we applied symbiotic machine learning to construct predictive models. We trained and evaluated ten different machine learning models to classify citrus samples as either HLB-infected or healthy based on their leaf and rhizosphere bacterial community profiles. Model performance was assessed using accuracy, AUC (area under the ROC curve), and ROC curves, providing a comprehensive evaluation of each model’s ability to discriminate between health statuses (Figures S3, S4, S5). Based on these performance metrics, we identified the bagging model trained at the species level as the top-performing model for classifying leaf samples, and the random forest model trained at the genus level as the best model for rhizosphere samples (Figure 5).
The bagging model for leaf samples, utilizing bacterial species abundances, achieved optimal HLB detection based on a set of 17 key bacterial species. Among these biomarker species, Solanum melongena (eggplant) was identified as the most important feature for prediction accuracy (Figure 6A). Interestingly, the analysis revealed that two species, Candidatus Liberibacter asiaticus (CLas) itself and Paraburkholderia rhizoxinica HKI 454, were more abundant in HLB-infected leaves. In contrast, the remaining 15 biomarker species exhibited higher relative abundances in healthy leaves (Table S4). For the citrus rhizosphere, our symbiotic machine learning approach using a random forest model identified 28 bacterial genera as key biomarkers for HLB prediction. Nitrospira genus emerged as the most influential genus for accurate classification (Figure 6B). Similar to the leaf model, we observed that only a minority of these biomarker genera (four out of 28) were more abundant in HLB-infected citrus rhizospheres. These included Streptomyces, Burkholderia-Caballeronia-Paraburkholderia, and Bacillus. The majority (24 genera) were more abundant in healthy citrus rhizospheres (Table S5).
Discussion
The quest for reliable biomarkers within the citrus phytobiome to detect HLB is crucial for developing effective disease diagnostics and management strategies. Understanding whether specific microbial taxa consistently respond to HLB across different geographical regions is essential for global applicability of these biomarkers. Our meta-analysis, examining HLB-infected and healthy citrus rhizosphere and phyllosphere microbiomes on a global scale, aimed to identify such robust biomarkers. Consistent with previous research on citrus rhizosphere microbiomes across six continents by Xu et al. (2018), we found Proteobacteria, Actinobacteria, Acidobacteria, and Bacteroidetes as the predominant bacterial phyla in healthy citrus rhizospheres (Figure 3A). Similarly, the dominant bacterial phyla and their relative abundance distributions in healthy citrus phyllospheres in our study align with published findings by Blaustein et al. (2017), Bai et al. (2019), and Wu et al. (2020). These consistent observations across diverse geographical locations suggest that host phylogeny may exert a stronger influence on phytobiome assembly than geographical factors alone. Recent studies have indeed highlighted adaptive matching between plant hosts and their rhizosphere and phyllosphere microbiomes, further supporting this notion.
Symbiotic microbiome homeostasis is intrinsically linked to host physiology and overall health. Microbial diversity, a fundamental indicator of community stability and function, is a critical attribute of phytobiomes. Reduced phytobiome richness and diversity are often associated with increased plant susceptibility to harmful factors, potentially due to weakened competition against invading pathogens or disruption of beneficial microbial interactions. In line with this, our study revealed a decrease in alpha diversity in both rhizosphere and phyllosphere microbiomes of HLB-infected citrus plants (Figures 2A, B). However, we also noted considerable variability in alpha diversity indices among HLB-infected samples, which may be attributed to variations in sequencing regions and methodologies across different datasets. Furthermore, PERMANOVA analysis indicated that data source and sequenced target region had a more pronounced effect on citrus phytobiome composition than citrus health status, geographical location, or tissue source (Table S6). This suggests that while alpha diversity shows a general trend, methodological variations across studies can significantly influence observed phytobiome characteristics, potentially limiting its robustness as a universal indicator of citrus health status.
Global microbiome studies have consistently shown that a small fraction of bacterial taxa often accounts for a large proportion of highly diverse bacterial communities. For instance, a global soil survey revealed that just 2% of bacterial taxa constituted nearly half of the bacterial communities across diverse sites. Similarly, Xu et al. (2018) reported that a small core set of bacterial taxa (< 10%) dominated rhizospheres of citrus samples from various continents. In our study, we identified 138 bacterial taxa shared across all samples, representing approximately half of the bacterial communities in healthy citrus samples (Figure 3E). In stark contrast, in HLB-infected citrus leaves and rhizospheres, the median total abundances of these shared bacterial taxa accounted for approximately 95% of the entire communities (Figure 3F). This marked shift indicates that a limited number of core taxa become disproportionately dominant in HLB-affected citrus, potentially explaining the observed decrease in alpha diversity in diseased samples.
Understanding the mechanisms governing phytobiome assembly is crucial for developing effective plant microbiome management strategies. From a meta-community perspective, bacterial community assembly is shaped by both deterministic and stochastic processes. Deterministic processes imply that species distributions are predictably governed by specific ecological niches, while stochastic processes suggest that random events and chance dispersal play a significant role, allowing multiple species to coexist in similar habitats. Our findings suggest that both stochastic and deterministic processes are important in shaping healthy citrus rhizosphere and phyllosphere microbiomes (Figure 4). The observed differences in the relative importance of these processes between rhizosphere and phyllosphere communities may reflect their distinct environments and functions. Rhizosphere and phyllosphere microbiomes, located below- and aboveground, respectively, perform different roles in plant health and are exposed to contrasting environmental conditions. Factors such as low nutrient availability and prolonged light exposure in the phyllosphere may favor specific bacterial groups, increasing the influence of deterministic processes. Furthermore, our analysis revealed that deterministic processes exerted a stronger influence on phytobiome assembly in HLB-infected citrus samples compared to healthy ones (Figure 4). Similar shifts towards deterministic processes have been reported in phytobiomes of other diseased plants and in the gut microbiota of diseased animals. In the context of HLB, a bacterial disease, the proliferation of pathogenic bacteria may reduce niche overlap and select for species that avoid competition, thereby promoting deterministic community assembly.
Among the diverse statistical tools for analyzing complex relationships between microbial communities and phenotypes, symbiotic machine learning methods are increasingly recognized as powerful approaches. Machine learning encompasses various algorithms, including unsupervised, semi-supervised, and supervised learning techniques. In this study, we employed supervised learning to investigate the relationship between citrus health status and bacterial relative abundances across different taxonomic levels in rhizosphere and phyllosphere microbiomes. Using ten machine learning algorithms, we found that supervised learning methods, specifically random forest and bagging models based on rhizosphere and phyllosphere bacterial genera and species, yielded the most accurate HLB prediction models (Figure 5). Consistent with our findings, previous meta-analyses have demonstrated the utility of supervised learning models for predicting Fusarium wilt disease in plants based on soil microbiomes, shrimp diseases based on gut microbiota, and even environmental health variables using microbiome data. Furthermore, random forest models have shown high accuracy in predicting soil health parameters in agroecosystems. These collective findings provide compelling evidence for the effectiveness of supervised learning in developing accurate plant health prediction models based on microbiome data.
The Asian citrus psyllid Diaphorina citri is recognized as the primary vector for HLB transmission. Current HLB control strategies primarily rely on removing symptomatic trees and insecticide application to manage psyllid populations. However, the efficacy of these measures is limited due to the long asymptomatic period of HLB-infected trees. Our bagging model identified key bacterial taxa associated with HLB incidence in citrus leaves, including CLas and Paraburkholderia rhizoxinica. While CLas is the established causal agent of HLB, Paraburkholderia rhizoxinica is an endofungal bacterium known for its symbiotic relationship with phytopathogenic fungi. In contrast, the bacterial genera enriched in HLB-infected citrus rhizospheres in our study, such as Streptomyces, Burkholderia-Caballeronia-Paraburkholderia, and Bacillus, are not directly implicated in HLB pathogenesis. These genera include antibiotic producers and species known to influence community stability. Our leaf-based model suggests that screening for specific known pathogens in citrus leaves could be a viable approach for HLB risk assessment. However, it is important to acknowledge that our leaf dataset for healthy citrus was relatively small (29 samples). Imbalances in sample sizes between healthy and diseased groups can potentially lead to overestimation of disease risks in machine learning models. Therefore, we recommend prioritizing the random forest model based on bacterial genera in the rhizosphere for practical HLB prediction in citrus plants.
Conclusions
This study underscores the potential of phytobiome analysis, combined with symbiotic machine learning, for detecting HLB-infected citrus plants on a global scale. Our meta-analysis of phytobiome data from hundreds of citrus samples revealed significant reductions in rhizosphere and phyllosphere microbiome diversities in HLB-infected samples compared to healthy ones. Furthermore, HLB onset increased the contribution of deterministic processes to both rhizosphere and phyllosphere microbiome assemblies. We identified 17 phyllosphere and 28 rhizosphere bacterial taxa strongly associated with HLB, which served as robust biomarkers in our symbiotic machine learning models. These models demonstrated high accuracy in predicting HLB outbreaks in citrus. Our findings highlight the value of phytobiome composition, analyzed through symbiotic machine learning, for assessing HLB risk in citrus plants using 16S rRNA gene sequencing data. Continued advances in high-throughput sequencing technologies and decreasing costs will further facilitate the development of even more refined and accessible models for predicting plant health status in agriculturally important crops, paving the way for precision agriculture and sustainable disease management strategies.
Data availability statement
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/Supplementary Material.
Author contributions
CR and X-FW: funding and project administration. H-QL: methodology, ideas, data curation, and statistical analysis. Z-LZ, H-JL, S-JY, LC and L-LD participated in this work. All authors contributed to the article and approved the submitted version.
Funding
This research was financially supported by the National Key R & D Program of China (grant nos. 2021YFD1400800, 2020YFD1000102, 2019YFD1002100, and 2018YFD0201500) and the Chongqing Scientific Research Project (grant no. cstc2021jcyj-bsh0082). We thank Liwen Bianji (Edanz) (www.liwenbianji.cn) for editing the English text of a draft of this manuscript.
Conflict of interest
Author Z-LZ was employed by Shanghai BIOZERON Biotechnology Co., Ltd.
The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2023.1129508/full#supplementary-material
References
Agler, M. T., Ruhe, J., Kroll, S., Morhenn, C., Kim, S. T., Weigel, D., et al. (2016). Microbial hub taxa link host and abiotic factors to plant microbiome variation. PloS Biol. 14 (1), e1002352. doi: 10.1371/journal.pbio.1002352
Bai, Y. N., Wang, J. C., Jin, L., Zhan, Z. G., Guan, L. M., Zheng, G. H., et al. (2019). Deciphering bacterial community variation during soil and leaf treatments with biologicals and biofertilizers to control huanglongbing in citrus trees. J. Phytopathol. 167, 686–694. doi: 10.1111/jph.12860
Berendsen, R. L., Pieterse, C. M., Bakker, P. A. (2012). The rhizosphere microbiome and plant health. Trends Plant 17 (8), 478–486. doi: 10.1016/j.tplants.2012.04.001
Blacutt, A., Ginnan, N., Dang, T., Bodaghi, S., Vidalakis, G., Ruegger, P., et al. (2020). An in vitro pipeline for screening and selection of citrus–associated microbiota with potential anti–”Candidatus liberibacter asiaticus” properties. Appl. Environ. Microbiol. 86 (8), e02883–e02819.
Blaustein, R. A., Lorca, G. L., Meyer, J. L., Gonzalez, C. F., Teplitski, M. (2017). Defining the core citrus leaf– and root–associated microbiota: factors associated with community structure and implications for managing huanglongbing (citrus greening) disease. Appl. Environ. Microbiol. 83, e00210–e00217. doi: 10.1128/AEM.00210-17
Bokulich, N. A., Kaehler, B. D., Rideout, J. R., Dillon, M., Bolyen, E., Knight, R., et al. (2018). Optimizing taxonomic classification of marker–gene amplicon sequences with qiime 2’s q2–feature–classifier plugin. Microbiome 6 (1), 90. doi: 10.1186/s40168-018-0470-z
Bolyen, E., Rideout, J., Dillon, M., Bokulich, N., Abnet, C., Al-Ghalith, G., et al. (2019). Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2. Nat. Biotechnol. 37, 852–857. doi: 10.1038/s41587-019-0209-9
Braga, D., Last, D., Hasan, M., Guo, H., Leichnitz, D., Uzum, Z., et al. (2019). Metabolic pathway rerouting in Paraburkholderia rhizoxinica evolved long-overlooked derivatives of coenzyme F420. ACS Chem. Biol. 14 (9), 2088–2094. doi: 10.1021/acschembio.9b00605
Carvalho, S. D., Castillo, J. A. (2018). Influence of light on plant-phyllosphere interaction. Front. Plant Sci. 9, 1482. doi: 10.3389/fpls.2018.01482
Chase, J. M. (2011). Ecological niche theory. Theory Ecol., 93–107.
Chen, H. (2022) VennDiagram: generate high–resolution Venn and Euler plots. r package version 1.7.3. Available at: https://CRAN.R-project.org/package=VennDiagram.
Chen, W., Ren, K., Isabwe, A., Chen, H., Liu, M., Yang, J. (2019). Stochastic processes shape microeukaryotic community assembly in a subtropical river across wet and dry seasons. Microbiome 7 (1), 1–16.
Coletta-Filho, H. D., Daugherty, M. P., Ferreira, C., Lopes, J. R. S. (2014). Temporal progression of ‘Candidatus liberibacter asiaticus’ infection in citrus and acquisition efficiency by Diaphorina citri. Phytopathology 104 (4), 416–421. doi: 10.1094/PHYTO-06-13-0157-R
Das, A. K., Sharma, S. K., Thakre, N. (2019). “Diagnostics for citrus greening disease (HLB): current and emerging technologies,” in Plant biotechnology: progress genomic era. Eds. Khurana, S., Gaur, R. (Singapore: Springer), 597–630.
Dastogeer, K. M., Tumpa, F. H., Sultana, A., Akter, M. A., Chakraborty, A. (2020). Plant microbiome-an account of the factors that shape community composition and diversity. Curr. Plant Biol. 23, 100161. doi: 10.1016/j.cpb.2020.100161
Delgado-Baquerizo, M., Oliverio, A. M., Brewer, T. E., Benavent-Gonzalez, A., Eldridge, D. J., Bardgett, R. D., et al. (2018). A global atlas of the dominant bacteria found in soil. Science 359, 320–325. doi: 10.1126/science.aap9516
de Lima, P. R. E., da Silva, I. R., Martins, M. K., de Azevedo, J. L., de Araújo, J. M. (2012). Antibiotics produced by streptomyces. Braz. J. Infect. Dis. 16 (5), 466–471.
Edwards, J., Johnson, C., Santos–Medellín, C., Lurie, E., Podishetty, N. K., Bhatnagar, S., et al. (2015). Structure, variation, and assembly of the root–associated microbiomes of rice. Proc. Natl. Acad. Sci. U.S.A. 112 (8), E911–E920. doi: 10.1073/pnas.1414592112
Escudero-Martinez, C., Coulter, M., Alegria Terrazas, R., Foito, A., Kapadia, R., Pietrangelo, L., et al. (2022). Identifying plant genes shaping microbiota composition in the barley rhizosphere. Nat. Commun. 13 (1), 1–14. doi: 10.1038/s41467-022-31022-y
French, E., Kaplan, I., Iyer-Pascuzzi, A., Nakatsu, C. H., Enders, L. (2021). Emerging strategies for precision microbiome management in diverse agroecosystems. Nat. Plants 7 (3), 256–267. doi: 10.1038/s41477-020-00830-9
Galdeano, D. M., de Souza Pacheco, I., Alves, G. R., Granato, L. M., Rashidi, M., Turner, D., et al. (2020). Friend or foe? relationship between ‘Candidatus liberibacter asiaticus’ and Diaphorina citri. Trop. Plant Pathol. 45 (6), 559–571. doi: 10.1007/s40858-020-00375-4
Ghannam, R. B., Techtmann, S. M. (2021). Machine learning applications in microbial ecology, human microbiome studies, and environmental monitoring. Comput. Struct. Biotec. 19, 1092–1107. doi: 10.1016/j.csbj.2021.01.028
Ginnan, N. A., Dang, T., Bodaghi, S., Ruegger, P. M., Peacock, B. B., McCollum, G., et al. (2018). Bacterial and fungal next generation sequencing datasets and metadata from citrus infected with ‘Candidatus liberibacter asiaticus’. Phytobiomes 2 (2), 64–70. doi: 10.1094/PBIOMES-08-17-0032-A
Ginnan, N. A., Dang, T., Bodaghi, S., Ruegger, P. M., McCollum, G., England, G., et al. (2020). Disease-induced microbial shifts in citrus indicate microbiome-derived responses to HLB across the disease severity spectrum. Phytobiomes J. 4, 375–387. doi: 10.1094/PBIOMES-04-20-0027-R
Gmitter, F. G., Hu, X. (1990). The possible role of yunnan, China, in the origin of contemporary citrus species (rutaceae). Econ. Bot. 44, 267–277. doi: 10.1007/BF02860491
Gupta, S., Aga, D., Pruden, A., Zhang, L., Vikesland, P. (2021). Data analytics for environmental science and engineering research. Environ. Sci. Technol. 55 (16), 10895–10907. doi: 10.1021/acs.est.1c01026
Jat, S. L., Suby, S. B., Parihar, C. M., Gambhir, G., Kumar, N., Rakshit, S. (2021). Microbiome for sustainable agriculture: a review with special reference to the corn production system. Arch. Microbiol. 203 (6), 2771–2793. doi: 10.1007/s00203-021-02320-8
Killiny-Mansour, N. (2019). Greening bacterium is now available in culture-so what’s next? (Florida, FL: EDIS).
Lajoie, G., Maglione, R., Kembel, S. W. (2020). Adaptive matching between phyllosphere bacteria and their tree hosts in a neotropical forest. Microbiome 8 (1), 1–10. doi: 10.1186/s40168-020-00844-7
Lee, J. A., Halbert, S. E., Dawson, W. O., Robertson, C. J., Keesling, J. E., Singer, B. H. (2015). Asymptomatic spread of huanglongbing and implications for disease control. Proc. Natl. Acad. Sci. U.S.A. 112 (24), 7605–7610. doi: 10.1073/pnas.1508253112
Li, J., Pang, Z., Trivedi, P., Zhou, X., Ying, X., Jia, H., et al. (2017). ‘Candidatus liberibacter asiaticus’ encodes a functional salicylic acid (SA) hydroxylase that degrades SA to suppress plant defenses. Mol. Plant Microbe In. 30 (8), 620–630. doi: 10.1094/MPMI-12-16-0257-R
Liu, H., Brettell, L. E., Singh, B. (2020). Linking the phyllosphere microbiome to plant health. Trends Plant 25 (9), 841–844. doi: 10.1016/j.tplants.2020.06.003
Liu, H., Sun, F., Peng, J., Shen, M., Li, J., Dong, Y. (2022). Deterministic process dominated belowground community assembly when suffering tomato bacterial wilt disease. Agronomy 12 (5), 1024. doi: 10.3390/agronomy12051024
Liu, Y., Zhu, A., Tan, H., Cao, L., Zhang, R. (2019). Engineering banana endosphere microbiome to improve fusarium wilt resistance in banana. Microbiome 7 (1), 1–15. doi: 10.1186/s40168-019-0690-x
Mahato, N., Sharma, K., Sinha, M., Dhyani, A., Pathak, B., Jang, H., et al. (2021). Biotransformation of citrus waste–I: production of biofuel and valuable compounds by fermentation. Processes 9 (2), 220. doi: 10.3390/pr9020220
Nicholson, W. L. (2002). Roles of bacillus endospores in the environment. Cell Mol. Life Sci. 59 (3), 410–416. doi: 10.1007/s00018-002-8433-7
Oksanen, J., Simpson, G. L., Blanchet, F. G., Kindt, R., Legendre, P., Minchin, P. R., et al. (2020) Vegan: community ecology package. r package version 2.6–2. Available at: https://CRAN.R-project.org/package=vegan.
Oyserman, B. O., Flores, S. S., Griffioen, T., Pan, X., van der Wijk, E., Pronk, L., et al. (2022). Disentangling the genetic basis of rhizosphere microbiome assembly in tomato. Nat. Commun. 13 (1), 1–16. doi: 10.1038/s41467-022-30849-9
Paasch, B. C., He, S. Y. (2021). Toward understanding microbiota homeostasis in the plant kingdom. PloS Pathog. 17 (4), e1009472. doi: 10.1371/journal.ppat.1009472
Peiffer, J. A., Spor, A., Koren, O., Jin, Z., Tringe, S. G., Dangl, J. L., et al. (2013). Diversity and heritability of the maize rhizosphere microbiome under field conditions. Proc. Natl. Acad. Sci. U.S.A. 110 (16), 6548–6553. doi: 10.1073/pnas.1302837110
Penyalver, R., Roesch, L. F., Piquer-Salcedo, J. E., Forner-Giner, M. A., Alguacil, M. D. M. (2022). From the bacterial citrus microbiome to the selection of potentially host-beneficial microbes. New Biotechnol. 70, 116–128. doi: 10.1016/j.nbt.2022.06.002
R Core Team (2020). R: a language and environment for statistical computing (Vienna, Austria: R Foundation for Statistical Computing). Available at: https://www.R-project.org/.
Sare, A. R., Stouvenakers, G., Eck, M., Lampens, A., Goormachtig, S., Jijakli, M. H., et al. (2020). Standardization of plant microbiome studies: which proportion of the microbiota is really harvested? Microorganisms 8 (3), 342. doi: 10.3390/microorganisms8030342
Sha, H., Lu, J., Chen, J., Xiong, J. (2022). A meta-analysis study of the robustness and universality of gut microbiota-shrimp diseases relationship. Environ. Microbiol. 24 (9), 3924–3938. doi: 10.1111/1462-2920.16024
Sing, T., Sander, O., Beerenwinkel, N., Lengauer, T. (2005). ROCR: visualizing classifier performance in r. Bioinformatics 21, 3940–3941. doi: 10.1093/bioinformatics/bti623
Sloan, W. T., Lunn, M., Woodcock, S., Head, I. M., Nee, S., Curtis, T. P. (2006). Quantifying the roles of immigration and chance in shaping prokaryote community structure. Environ. Microbiol. 8, 732–740. doi: 10.1111/j.1462-2920.2005.00956.x
Srivastava, A. K., Das, A. K., Jagannadham, P. T. K., Bora, P., Ansari, F. A., Bhate, R. (2022). Bioprospecting microbiome for soil and plant health management amidst huanglongbing threat in citrus: a review. Front. Plant Sci. 13, 858842. doi: 10.3389/fpls.2022.858842
Thoms, D., Liang, Y., Haney, C. H. (2021). Maintaining symbiotic homeostasis: how do plants engage with beneficial microorganisms while at the same time restricting pathogens? Mol. Plant Microbe In. 34 (5), 462–469. doi: 10.1094/MPMI-11-20-0318-FI
Tian, G., Qiu, H., Li, D., Wang, Y., Zhen, B., Li, H., et al. (2022). Little environmental adaptation and high stability of bacterial communities in rhizosphere rather than bulk soils in rice fields. Appl. Soil Ecol. 169, 104183. doi: 10.1016/j.apsoil.2021.104183
Torija, A. J., Ruiz, D. P. (2015). A general procedure to generate models for urban environmental-noise pollution using feature selection and machine learning methods. Sci. Total Environ. 505, 680–693. doi: 10.1016/j.scitotenv.2014.08.060
Trivedi, P., He, Z. L., Van Nostrand, J. D., Albrigo, G., Zhou, J. Z., Wang, N. (2012). HLB alters the structure and functional diversity of microbial communities associated with citrus rhizosphere. ISME J. 6, 363–383. doi: 10.1038/ismej.2011.100
Trivedi, P., Leach, J. E., Tringe, S. G., Sa, T., Singh, B. K. (2020). Plant–microbiome interactions: from community assembly to plant health. Nat. Rev. Microbiol. 18 (11), 607–621. doi: 10.1038/s41579-020-0412-1
Vanwonterghem, I., Jensen, P. D., Dennis, P. G., Hugenholtz, P., Rabaey, K., Tyson, G. W. (2014). Deterministic processes guide long–term synchronised population dynamics in replicate anaerobic digesters. ISME J. 8, 2015–2028. doi: 10.1038/ismej.2014.50
Wang, N. (2019). The citrus huanglongbing crisis and potential solutions. Mol. Plant 12 (5), 607–609. doi: 10.1016/j.molp.2019.03.008
Wang, N. (2020). A perspective of citrus HLB in the context of the Mediterranean basin. J. Plant Pathol. 102, 635–640. doi: 10.1007/s42161-020-00555-w
Wang, N., Jin, T., Trivedi, P., Setubal, J. C., Tang, J. (2015). Announcement of the international citrus microbiome (phytobiome) consortium. J. Citrus Pathol. 2, 1–2. doi: 10.5070/C421027940
Wilhelm, R. C., van Es, H. M., Buckley, D. H. (2022). Predicting measures of soil health using the microbiome and supervised machine learning. Soil Biol. Biochem. 164, 108472. doi: 10.1016/j.soilbio.2021.108472
Wu, Y., Qu, M., Pu, X., Lin, J., Shu, B. (2020). Distinct microbial communities among different tissues of citrus tree Citrus reticulatacv. Chachiensis. Sci. Rep. 10, 6068. doi: 10.1038/s41598-020-62991-z
Xu, J., Zhang, Y., Zhang, P., Trivedi, P., Riera, N., Wang, Y., et al. (2018). The structure and function of the global citrus rhizosphere microbiome. Nat. Commun. 9 (1), 1–10. doi: 10.1038/s41467-018-07343-2
Yao, Z., Yang, K., Huang, L., Huang, X., Qiuqian, L., Wang, K., et al. (2018). Disease outbreak accompanies the dispersive structure of shrimp gut bacterial community with a simple core microbiota. AMB Express. 8 (1), 1–10. doi: 10.1186/s13568-018-0644-x
Yilmaz, P., Parfrey, L. W., Yarza, P., Gerken, J., Pruesse, E., Quast, C., et al. (2014). The SILVA and “All-species living tree project (LTOP)”. taxonomic frameworks. Nucleic Acids Res. 42, D643–D648.
Yu, W., Wu, J. H., Zhang, J., Yang, W., Chen, J., Xiong, J. (2018). A meta-analysis reveals universal gut bacterial signatures for diagnosing the incidence of shrimp disease. FEMS Microbiol. Ecol. 94, fiy147. doi: 10.1093/femsec/fiy147
Yuan, J., Wen, T., Zhang, H., Zhao, M., Penton, C. R., Thomashow, L. S., et al. (2020). Predicting disease occurrence with high accuracy based on soil macroecological patterns of Fusarium wilt. ISME J. 14 (12), 2936–2950. doi: 10.1038/s41396-020-0720-5
Zhang, B., Zhang, J., Liu, Y., Shi, P., Wei, G. (2018). Co-Occurrence patterns of soybean rhizosphere microbiome at a continental scale. Soil Biol. Biochem. 118, 178–186. doi: 10.1016/j.soilbio.2017.12.011
Zhang, H., Wu, C., Wang, F., Wang, H., Chen, G., Cheng, Y., et al. (2022). Wheat yellow mosaic enhances bacterial deterministic processes in a plant-soil system. Sci. Total Environ. 812, 151430. doi: 10.1016/j.scitotenv.2021.151430
Zhang, Y., Trivedi, P., Xu, J., Roper, M. C., Wang, N. (2021). The citrus microbiome: from structure and function to microbiome engineering and beyond. Phytobiomes J. 5 (3), 249–262. doi: 10.1094/PBIOMES-11-20-0084-RVW
Zhao, Z., Jiang, J., Zheng, J., Pan, Y., Dong, Y., Chen, Z., et al. (2022). Exploiting the gut microbiota to predict the origins and quality traits of cultured sea cucumbers. Environ. Microbiol. 24 (9), 3882–3897. doi: 10.1111/1462-2920.15972
Zhou, F., Wu, X., Gao, Y., Fan, S., Zhou, H., Zhang, X. (2022a). Diversity shifts in the root microbiome of cucumber under different plant cultivation substrates. Front. Microbiol. 13, 878409. doi: 10.3389/fmicb.2022.878409
Zhou, L., Zhao, Z., Shao, L., Fang, S., Li, T., Gan, L., et al. (2022b). Predicting the abundance of metal resistance genes in subtropical estuaries using amplicon sequencing and machine learning. Ecotox. Environ. Safe 241, 113844. doi: 10.1016/j.ecoenv.2022.113844
Keywords: citrus microbiome, Huanglongbing, machine learning, meta-analysis, community assembly
Citation: Liu H-Q, Zhao Z-l, Li H-J, Yu S-J, Cong L, Ding L-L, Ran C and Wang X-F (2023) Accurate prediction of huanglongbing occurrence in citrus plants by machine learning-based analysis of symbiotic bacteria. Front. Plant Sci. 14:1129508. doi: 10.3389/fpls.2023.1129508
