Genomic Diversity Of "Candidatus Liberibacter Asiaticus" Strains And Classification Study Of Huanglongbing Symptoms Through Artificial Intelligence Approach

Citrus Huanglongbing(HLB)is the one of the most devasting pathogenetic agent to the citrus industry around the world.Complicated yellowing symptoms would eventually develop on citrus hosts by which making field detection and research difficult for researchers.Citrus plants decline and die within one year or several after infection.The well-known agents to HLB were three species of genus Liberibacter: “Candidatus Liberibacter asiaticus”(CLas),“Candidatus Liberibacter africanus”(CLaf)and “Candidatus Liberibacter americanas”(CLam).Of all,CLas is the most widely distributed in globe and causing the greatest loss to the citrus industry.To date,the fact that CLas has not been successfully cultured in vitro result in impossibility to explore its pathogenicity using traditional research techniques for bacteria research.As the technology developing,bioinformatic research based on high throughput sequencing and deep learning has been a hot spot in the field of bacterial pathogen.In this study,we explored the diversity of CLas strain in California with the assistance of genomic analyses,the prophage region and several loci were used to explore the relationship among CLas strain.In addition,metagenomic approach was employed to explore the response of microbial community against the HLB.We also explored the possibility of symptom classification of HLB based on deep learning technique,classifiers were constructed using vast numbers of HLB imgaes and the classifying performance under field conditions was evaluated.The main results and conclusions were as follows,1.Diversity of “Candidatus Liberibacter asiaticus” population in CaliforniaA diversity profiling approach based on prophage region and multiple loci,including the tandem repeat locus trn,miniature inverted repeat transposable elements(MITEs),and a terminase gene ter L,was established,by which the relationship among CLas strains in California was successfully charcterized.Based on prophage type grouping and MITEs,all California strains can be classified to 6 groups.The group ‘PTG1,A’ consisted of 3 strains from San Gabriel,‘PTG2,A’ consisted of 2 strains from Hacienda Heights,‘PTG1-2,B’ consisted of the La Habra strain,‘PTG1-3,B,A’ consisted of the only strain from Cerritos,‘PTG1,B’ consisted of 2 strains from Anaheim,‘PTG1-B,A’ consisted of the only Riverside strain.All these results implied that CLas strains from different geographical origins in California might derived from various sources,the new strains discovered in the past 2 years were not a result caused by natural spread of starins from San Gabriel and Hacienda.In addition,alignment of ter L sequences showed that all California strains were from unknonwn Asian source instead of Florida source.2.Metagenomic analyses of Huanglongbing-infected Asian citrus psyllids and citrus in CaliforniaHigh throughput sequencing and metagenomic analyses were employed to explore the microbial community compositions of the Asian citrus psyllids sample A-AHCA1 and the citrus sample C-AHCA2 from Anaheim,California.Microbiome in the ACP sample A-AHCA1 and C-AHCA2 carried similar microbial taxas,but the ratio of these taxas showed great quantitative variance.Comparative genomic analysis showed that AHCA1 shared at least 99.43% identity to AHCA2,which implied the passage from citrus to plants didn’t change the CLas genome.Significant difference of bacterial community compositions was observed between ACP group and citrus group while little variance was within groups.The bacterial community compositions in citrus was relatively much more stable than those of ACP with Proterbacteria,Firmicutes and Cyanobacteria as dominant taxas.5 genuses was observed to be present in all ACP samples and citrus samples,including Pseudomonas,Acinetobacter,Staphylococcus,Liberibacter and Rickettsia.Significant difference of relative abundance of Rickettsia and Acinetobacter was observed between ACP samples and citrus samples,the abundance of Acinetobacter was much higher in citrus(P<0.05)than ACP while the abundance of Rickettsia significantly lower in citrus samples than ACP samples(P<0.05).3.Comparative virome research of HLB infected citrus and non-symptomatic citrusComparative virome analysis was employed to explore the viral community compositions of HLB-infected citrus and non-symptomatic citrus with the assistance of high throughput sequencing and metagenomic analysis.Result of alpha diversity showed that there’re no significant difference between viral community of HLB group and ACP group.3 viral pathogens,as previously reported to be,Citrus tristeza virus(CTV)、Citrus tatter leaf virus(CTLV),Ctirus bend leaf viroid(CBLVd),were identified in citrus samples with high quality genomes assemblage,the Cucumber mosaic virus(CMV)was also identified in pumello while there was no report in China proving the CMV was observed in citrus in China.In addition,Diaphorina citri densovirus(Dc DNV)was identified as it was the first report in citrus sampels in China.Population investigation showed that CTV,CBLVd and CMV were popular in greenhouse citrus with positive ratios as 56.4%,27.3% and 16.40% repectively,no CTLV incidence was observed in green house citrus population.CTV,Dc DNV and CBLVd were popular in ACP populations with positive rates as 39.50,32.60% and 18.60% repectively.Further investigation revealed different features of quatitative distribution among different virues.Of all Dc DNV positives ACP samples,samples of high titers(CT values <25)and medium titers(25 ≤ CT values ≤ 30)account for 64.29% while tites of all Dc DNV positive samples in citrus were of low CT values(CT value > 30),which implied citrus might not be a host;titers of CBLVd and CTV were all of low CT values(average CT values over 28)in ACP samples and no positive samples of CTV and CTLV were observed in ACP samples,which implied ACP might not be a host for CBLVd,CTV,CMV and CTLV.Result of phylogenetic analyses showed that new strains of CTLV,CTV and CMV discovered in this study belonged to Asian phylogenetic group while the new CMVsat strain belonged to European group instead of other Chinese strains,which implied that CMVsat and CMV may be from different origins.The new CBLVd strain found in this study was also observed to be close to European strains and Mideast strains.4.Classification research of Huanglongbing symptom based on artificial inteligenceBased on Tensorflow framework and two structures of artificial neural networks,Mobile Net V1 and Inception V3,4 classifiers,named M-8-C,I-8-C,M-2-C,I-2-C,respectively were constructed based on vast numbers of HLB images and healthy images through tranfer-learning.The main results were as follows.The classifying performance of M-8-C outstanded in all classifiers with overall accuracy as 88.4%,implying deep convolutional neural networks have great capability in HLB symptom classifying.The overall accuracy of I-2-C was lower than M-2-C while I-8-C was lower than M-8-C,implying that the the Mobile Net structure was better than Inception structure.The 8-class classifiers performed better than 2-class classfiers.Compared with 2-class classfier,8-class classifiers was capable of classifying various types of HLB symptoms with mean accuracy as 87.4%,but the true positive detection rate to whole leave yellow symptom caused by HLB and non-HLB was 44.0% and 64.0%,respectively.Classifiers could be transferred to cell phone and show excellent performance under field conditions.Test results on cell phones showed that accuracy of 8-class classifier(≥68.0%)was lower than 2-class classifier(≥83.5%),but accuracy of 8-class classifier was higher than 2-class classifier(≥89.0%,≥86.4%),which implyed that the classification performance was associated with the computing power and 8-class classifier require more computing resource than 2-class classifier.