The plant phytobiome, consisting of the rhizosphere, phyllosphere, and endosphere, harbors diverse microbes that affect plant growth and health (Berendsen et al., 2012; Liu et al., 2019; Liu et al., 2020). Considerable research has been carried out on the diversity, composition, and function of the phytobiome of various plant species, including rice (Edwards et al., 2015), maize (Peiffer et al., 2013), corn (Jat et al., 2021), soybean (Zhang et al., 2018), cucumber (Zhou et al., 2022a), citrus (Xu et al., 2018), and tomato (Oyserman et al., 2022). Microbes inhabiting plant surfaces or internal tissues produce metabolites that support plant growth by regulating physiological processes (e.g., nutrient absorption and pathogen suppression) (Trivedi et al., 2020). The identification of microbes associated with specific phenotypes is currently a fundamental objective of phytobiome researchers. Moreover, changes to the plant microbiome are influenced by diverse biotic and abiotic factors of the host and surrounding environment (Dastogeer et al., 2020). Niche and neutral theory-based approaches have been used to explore the mechanisms modulating the microbiome assembly, with all factors classified into deterministic or stochastic processes (Chen et al., 2019). These studies have expanded our understanding of phytobiomes, with implications for managing plant-associated microbiomes to enhance crop production (French et al., 2021).

Citrus is an economically important fruit crop comprising several widely cultivated species initially domesticated more than 1,000 years ago (Gmitter and Hu, 1990). Because their fruits contain an abundance of diverse nutrients, vitamins, and dietary fiber, citrus species are extensively cultivated worldwide, with annual yields exceeding 120 million tons (Mahato et al., 2021). Nevertheless, citrus production has been restricted by global climate change and the prevalence of diseases (Wang, 2019; Wang, 2020). Huanglongbing (HLB), a fatal disease, is the biggest threat to citrus production. The high prevalence of HLB and associated, considerable yield losses have recently renewed interest in this disease (Das et al., 2019). HLB leads to a loss of citrus root carbonaceous compounds, malfunctioning phloem tissues, and decreased release of photosynthates, all of which impair the transport of photoassimilates (Trivedi et al., 2020). In addition, HLB upsets the nutrient balance by altering the ability of roots to absorb and transport nutrients and water (Wang, 2020). HLB is mainly caused by the bacterium Candidatus Liberibacter asiaticus (CLas) (Li et al., 2017). Because CLas cannot be monocultured (Killiny-Mansour, 2019), clarifying citrus-associated microbiome dynamics due to HLB pressure is of serious interest.

A comprehensive understanding of citrus-associated microbiomes may lead to the development of sustainable, environmentally friendly methods for increasing citrus plant health and productivity. Global patterns in citrus phytobiomes are already being revealed (Xu et al., 2018; Zhang et al., 2021; Penyalver et al., 2022). For instance, the International Citrus Microbiome Consortium, established in 2015, has performed sequencing analyses of citrus rhizosphere and soil microbiome samples from major citrus-producing regions on six continents (Wang et al., 2015). In addition, recent Illumina-based sequencing of 16S rRNA genes has revealed that HLB alters the microbiome of citrus rhizospheres and phyllospheres (Srivastava et al., 2022). Most related research has only focused on healthy citrus microbiomes or HLB-infected ones, however, with relatively few comparative studies performed on both types of microbiomes (Li et al., 2017; Ginnan et al., 2018; Blacutt et al., 2020). Whether some citrus taxa in worldwide cultivation are robust and universally responsive to HLB remains unclear.

In this study, we systematically reviewed available data on the citrus phytobiome and compared the bacterial communities of healthy and HLB-infected citrus plants. In addition, machine learning approaches were applied to identify potential biomarkers for HLB occurrence after technical biases, geographic distribution, and tissue specificity were taken into account. These analyses allowed us to reveal the diversity, composition, and mechanisms underlying the bacterial community assembly of HLB-infected citrus plants. Finally, we developed a phytobiome-based model to predict HLB outbreaks under field conditions.

The exploration of biomarkers common to the phytobiome of citrus plants infected with HLB is critical for developing improved methods for diagnosing plant diseases caused by bacteria and for determining optimal treatments. Nevertheless, knowledge of whether certain microbial lineages consistently respond to HLB across global biogeographic regions is unclear. In this study, we performed a meta-analysis of HLB-infected and healthy citrus rhizosphere and phyllosphere microbiomes on a global scale to screen for biomarkers useful for detecting HLB. By combining global datasets, we determined that Proteobacteria, Actinobacteria, Acidobacteria, and Bacteroidetes were the predominant bacterial phyla in healthy citrus rhizospheres ( Figure 3A ). This result is consistent with the findings of an earlier study on citrus rhizosphere samples from six continents (Xu et al., 2018). Furthermore, the dominant bacterial phyla detected in healthy citrus phyllospheres as well as the distribution of their relative abundances in this study are in accordance with published results (Blaustein et al., 2017; Bai et al., 2019; Wu et al., 2020). These observations revealing similarities in citrus phytobiomes from various geographical regions imply that host phylogeny may influence phytobiome assembly more than geographical factors. Recent reports have described adaptive matching between rhizosphere and phyllosphere microbiomes and plant hosts (Lajoie et al., 2020; Escudero-Martinez et al., 2022).

Symbiotic microbiome homeostasis is closely associated with host physiological features and health (Paasch and He, 2021). As a fundamental indicator of the stability and performance of microbial communities, diversity is a crucial index for phytobiomes (Sare et al., 2020). A decrease in phytobiome richness and diversity is often responsible for the increased susceptibility of plant hosts to potentially harmful factors (Agler et al., 2016). In addition, decreased phytobiome diversity may be due to insufficient competition between resident commensals and invading pathogens (Thoms et al., 2021). In the current study, alpha diversity decreased in the rhizosphere and phyllosphere microbiomes of citrus plants infected with HLB ( Figures 2A, B ). Alpha diversity indices varied considerably among HLB-infected samples, however, which may be related to the features of the sequenced regions and the methods used. Moreover, the PERMANOVA results revealed that data source and sequenced target region had larger effects on citrus phytobiome composition than did citrus health status, geographic location, and tissue source ( Table S6 ). Alpha diversity and citrus phytobiome composition were thus strongly affected by the methods used in different bio-projects and may therefore not be robust indicators of citrus health status.

Most research conducted on a global scale has suggested that relatively few bacterial taxa represent a large proportion of highly diverse bacterial communities. For example, a study examining global soil samples found that 2% of bacterial taxa accounted for nearly half of bacterial communities at various sites (Delgado-Baquerizo et al., 2018). Notably, Xu et al. (2018) reported that a small number of bacterial taxa (< 10%) were the core taxa in rhizospheres of citrus samples collected on various continents. In the present study, we detected 138 bacterial taxa common to all samples; these taxa represented approximately half of the bacterial communities in healthy citrus samples ( Figure 3E ). In contrast, the median total abundances of shared bacterial taxa in HLB-infected citrus leaves and rhizospheres corresponded to approximately 95% of the entire communities ( Figure 3F ). Hence, a limited number of core taxa were associated with the occurrence of HLB in citrus plants. This result may help to explain the decrease in alpha diversity index values among HLB samples.

The mechanisms mediating the assembly of phytobiome communities must be considered when designing plant microbiome management strategies (Trivedi et al., 2020). From a meta-community perspective, bacterial community assembly is governed by both deterministic and stochastic processes (Tian et al., 2022). If communities are controlled by deterministic processes, species will occupy specific ecological niches in a predictable fashion (Vanwonterghem et al., 2014). In contrast, multiple species can exist in similar or overlapping habitats in communities affected by stochastic fluctuations (Sloan et al., 2006). Our findings suggest that stochastic and deterministic processes are critical for shaping healthy citrus rhizosphere and phyllosphere microbiomes ( Figure 4 ). Inconsistencies between citrus rhizosphere and phyllosphere community assemblies may be attributed to their different lifestyles and functions. Plant leaves and roots are located above- and belowground, respectively, which allows phyllosphere and rhizosphere microbiomes to perform different functions (Trivedi et al., 2012). In addition, low nutrient levels and long-term illumination may lead to changes in the abundances of specific bacteria in the phyllosphere microbiome (Carvalho and Castillo, 2018), thereby increasing the importance of deterministic processes. Moreover, we observed that deterministic processes mediated phytobiome assembly more substantially in HLB-infected citrus samples than in healthy ones ( Figure 4 ). Similar observations have been reported for phytobiomes of other diseased plants and the gut microbiota of diseased animals (Yao et al., 2018; Liu et al., 2022; Zhang et al., 2022). HLB is a disease caused by pathogenic bacteria, and the enriched pathogens will decrease the abundance of species with an overlapping niche and select for species without niche conflicts through competition (Chase, 2011). This phenomenon may explain why deterministic processes shaped the HLB-infected citrus microbiome.

Among the many statistical methods for elucidating the complex relationships between microbial communities and specific phenotypes, machine learning-based methods are considered the most promising (Torija and Ruiz, 2015). Machine learning approaches take various forms according to their algorithms (e.g., unsupervised, semi-supervised, or supervised learning) (Ghannam and Techtmann, 2021). In the present study, we assessed the relationship between the health status of citrus samples collected worldwide and the relative abundance of bacteria at different taxonomic levels in the rhizosphere and phyllosphere. We used 10 machine learning approaches and found that the most appropriate models for predicting HLB infections were supervised learning methods (random forest and bagging) that were based on rhizosphere and phyllosphere bacterial genera and species ( Figure 5 ). A previous meta-analysis demonstrated that supervised learning models for soil microbiomes may be used to predict the potential occurrence of Fusarium wilt disease in plants (Yuan et al., 2020). In addition, supervised learning methods have identified robust and reproducible features relevant for diagnosing shrimp diseases according to meta-analyses of gut microbiota (Sha et al., 2022). The origin and quality of sea cucumber cultured in diverse geographic regions has also been accurately predicted using random forest models for gut microbiota (Zhao et al., 2022). Furthermore, supervised machine learning approaches have accurately predicted environmental health variables following analyses of microbiome data (Zhou et al., 2022b). Random forest models for microbial communities have predicted the soil health parameters of agroecosystems, with accuracies exceeding 80% (Wilhelm et al., 2022). These results provide convincing evidence of the utility of supervised learning methods for establishing models that accurately predict the health status of plants.

The vector of HLB, a bacterial infection of citrus trees, is believed to be the Asian citrus psyllid Diaphorina citri (Galdeano et al., 2020). At present, the dominant control strategies for HLB are removal of HLB-symptomatic citrus trees and the spraying of insecticides to restrict the psyllid (Coletta-Filho et al., 2014). Because HLB-infected trees may remain asymptomatic for several months, however, the efficacy of current disease control measures is limited (Lee et al., 2015). Using the bagging model, we identified crucial bacterial taxa related to citrus HLB disease incidence, including CLas and Paraburkholderia rhizoxinica in the phyllosphere and Streptomyces, Burkholderia-Caballeronia-Paraburkholderia, and Bacillus in the rhizosphere. A recent study indicated that CLas is the main pathogen responsible for HLB outbreaks (Ginnan et al., 2020). Paraburkholderia rhizoxinica is an endofungal bacterium that has a symbiotic relationship with phytopathogenic fungi (Braga et al., 2019). In contrast, bacteria enriched in HLB-infected citrus rhizospheres in our study were not directly related to the disease phenotype. These bacteria included antibiotic producers and species with detrimental effects on community stability (Nicholson, 2002; de Lima et al., 2012). Our findings imply that the risk of HLB can be assessed by screening for a few specific known pathogens in citrus leaves. Data for only 29 healthy citrus phyllospheres were included in our analyses, however, and a limited sample size and bias between two classifications can cause machine learning models to overestimate the risks of HLB outbreaks (Gupta et al., 2021). We thus recommend the use of a random forest model based on bacterial genera in the rhizosphere to predict the likelihood of HLB in citrus plants.

In this study, we analyzed the utility of phytobiome examinations for detecting HLB-infected citrus plants on a global scale. Meta-analyses involving the phytobiome data of hundreds of citrus samples revealed significant decreases in rhizosphere and phyllosphere microbiome diversities of HLB-infected samples relative to healthy ones. Furthermore, the onset of HLB increased the contribution of deterministic processes to citrus rhizosphere and phyllosphere microbiome assemblies. We also identified 17 and 28 HLB-related taxa in the phyllosphere and rhizosphere, respectively. These taxa may be exploited to accurately predict citrus HLB outbreaks on the basis of selected machine learning models. The findings of this study are relevant for evaluating the risks of HLB in citrus plants according to phytobiome compositions derived from 16S rRNA gene sequencing data. Advances in high-throughput sequencing technology and decreases in associated costs should enable researchers to further improve models for predicting the health status of agriculturally important plant species.

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 .

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.