Gene silencing of Diaphorina citri candidate effectors promotes changes in feeding behaviors

Insect effectors are mainly secreted by salivary glands, modulate plant physiology and favor the establishment and transmission of pathogens. Feeding is the principal vehicle of transmission of Candidatus Liberibacter asiaticus (Ca. Las) by the Asian citrus psyllid (ACP), Diaphorina citri. This study aimed to predict putative ACP effectors that may act on the Huanglongbing (HLB) pathosystem. Bioinformatics analysis led to the identification of 131 candidate effectors. Gene expression investigations were performed to select genes that were overexpressed in the ACP head and modulated by Ca. Las. To evaluate the actions of candidate effectors on D. citri feeding, six effectors were selected for gene silencing bioassays. Double-stranded RNAs (dsRNAs) of the target genes were delivered to D. citri adults via artificial diets for five days. RNAi silencing caused a reduction in the ACP lifespan and decreased the salivary sheath size and honeydew production. Moreover, after dsRNA delivery of the target genes using artificial diet, the feeding behaviors of the insects were evaluated on young leaves from citrus seedlings. These analyses proved that knockdown of D. citri effectors also interfered with ACP feeding abilities in planta, causing a decrease in honeydew production and reducing ACP survival. Electrical penetration graph (EPG) analysis confirmed the actions of the effectors on D. citri feeding behaviors. These results indicate that gene silencing of D. citri effectors may cause changes in D. citri feeding behaviors and could potentially be used for ACP control.

Despite the existence of the ACP saliva proteome 33 , ACP effectors or salivary molecules that interact with host plants have not been identified until now 34 . In the present work, we predict secreted D. citri effectors using bioinformatics tools. Gene silencing through dsRNAs combined with an EPG technique was performed to verify the actions of the effectors on D. citri feeding behaviors.

Results
Candidate effector prediction. To generate a list of D. citri putative effectors, we used Myzus persicae Sulzer (Hemiptera: Aphididae) bioinformatics approach as a model for effector prediction 25 . With the intent of obtaining an accurate effectoromics dataset, proteins sequences used in these analyses were obtained from the following two distinct transcriptome and genomic sources: a transcriptome predicted from Arizona University studies 35 and genomic sequences from the HLB consortium (citrusgreening.org).
Considering that insect effectors are secreted through the classical eukaryotic endoplasmic reticulum (ER)-Golgi pathway through salivary glands, the initial predicted D. citri secretome was composed of proteins presenting signal peptides and the absence of transmembrane domains. SignalP analysis allowed the identification of 2,099 proteins with signal peptides. Furthermore, screening using the TMHMM, Phorbius and GPI-SOM tools identified 670 proteins that contained transmembrane domains or GPI-anchor signals, and those proteins were removed from further analysis. Thereafter, the ACP secretome was composed of 1,443 proteins.
It has been reported that insect effectors are mainly species-or genus-specific 36 . Thus, we predicted conserved domains and identified characterized proteins through multiple alignments against public protein databases (Pfam and Uniprot). All proteins that showed ordinary domains were excluded. To select proteins that are only present in salivary glands, we also performed tBlast-X against the arthropod gut and salivary ESTs, removing proteins that may be secreted in the ACP gut. Thus, were identified 131 candidate effectors (Table S1) composed of 78% exclusively D. citri proteins and 22% showed similarity with arthropod saliva proteins (Table S2). Interestingly, many of the sequences showed intrinsically disordered regions (IDRs) ( Table S1). These regions are abundant in phytopathogen effector proteins, interfere with protein secondary structure and allow protein folding in a stimulus-dependent manner 37 . The presence of enriched IDR proteins in the D. citri effectorome suggests a positive evolutionary selection of these proteins and potential functions through host interactions. Additionally, the subcellular localization of the predicted D. citri effectors suggests that the majority of the identified proteins act on the extracellular space (Table S1; Fig. S1). However, these proteins may also act on important host organelles such as the nucleus, chloroplast and mitochondrion (Table S1; Fig. S1).

Selection of effectors candidates of D. citri by RT-qPCR.
After the bioinformatic predictions, the screened candidate effectors were analyzed for their gene expression. Twelve putative effectors were evaluated (Tables 1 and S3) to identify genes that were expressed in different life stages, upregulated in the ACP head and showed expression modulated by Ca. Las.
First, the mRNA abundance of the candidate effector genes was evaluated in two different ACP life stages (nymph and adult). For both analyzed life stages, similar mRNA patterns were observed for the most candidates of the effectors (Fig. S2). DCEF22, DCEF28 and DCEF33 showed elevated mRNA levels in the nymph stage compared with the adult phase (Fig. S2).
Our goal was to select effectors that are present in the D. citri salivary gland or could be secreted by saliva. To achieve this aim, we evaluated the gene expression of candidate effectors from different body parts (head and body). The mRNA abundance of ten of twelve candidate effector genes was higher in the head than in the ACP  1). Candidate effectors that were not majorly expressed in the ACP head (DCEF10 and DCEF11) were excluded from further analysis. Furthermore, Ca. Las regulation of the gene expression of the ten candidate effectors was evaluated in nymphs and adults. Except for DCEF22, all the selected genes were significantly modulated by Ca. Las infection conditions in at least one of the analyzed life stages (Fig. 2). DCEF08 and DCEF23 were overexpressed in Ca. Las-infected insects compared with healthy insects in both the nymph and adult stages. The presence of Ca. Las contributes to increased DCEF19, DCEF26 and DCEF27 mRNA levels in ACP adults. DCEF33 was downregulated in both nymph and adult Ca. Las-infected ACPs. The DCEF32 and DCEF35 genes were downregulated in the Ca Las-infected adults. DCEF28 mRNA levels were reduced in Ca Las-infected nymphs compared with healthy nymphs (Fig. 2). Based on these results, the following six effectors were selected for the RNAi experiments: DCEF26, DCEF27, DCEF28, DCEF32, DCEF33 and DCEF35.
Delivery of dsRNA by artificial diet. DsRNA was delivered to ACP adults using a 30% sucrose-artificial diet with 100 ng/µL of dsRNA. After five days of continuous dsRNA delivery, we observed 46, 47 and 61% mortality for the DCEF26, DCEF27, and DCEF32 gene-fed insects, respectively (Fig. 3). DCEF28 treatment caused significant mortality during the first 72 hours of the experiment, reaching 48% mortality after five days. DCEF33 and DCEF35 showed fewer expressive effects on ACP mortality, causing 35 and 26% mortality, respectively (Fig. 3).
To verify the interference of silencing the effectors on ACP feeding, salivary sheath evaluation was performed through microscopy analysis. No structural differences in the salivary sheaths were observed between the treatments and control (Fig. S3). However, the salivary sheath sizes of dsDCEF32-, dsDCEF33-and dsDCEF35-treated insects were 20, 32 and 27% lower than the sucrose control, respectively. dsDCEF28-treated insects exhibited a 45% increase in salivary sheath size compared with the control (Fig. 4).
Moreover, the effects of effector silencing on ACP ingestion abilities were evaluated by quantification of the excretions deposited on filter paper at the bottom of the cages. A significant reduction in honeydew was observed in five of six treatments (Fig. 4). The DCEF32, DCEF33 and DCEF35 treatments caused the strongest reduction in honeydew production, presenting tinted areas that were 72, 82 and 69% lower than the sucrose control, respectively. A 38 and 37% reduction in the tinted area was observed for the DCEF27 and DCEF28 treatments, Figure 1. Relative gene expression of the 12 D. citri candidate effectors in the head and body. These data consist of normalized target mRNA levels to the mRNA abundance of the housekeeping genes S13 and S20. The results are the mean of three biological replicates per set. *corresponds to statistically significant differences amongst the treatments (t-test, P-value < 0.05).  Mortality rates of D. citri fed an artificial diet containing 30% sucrose and 100 ng/µL of dsRNA for GFP or dsRNA for the effectors DCEF26, DCEF27, DCE28, DCEF32, DCEF33 or DCEF35. The mortality rates were evaluated daily during a 120-hour period. The data correspond to the average of three independent experiments. Different letters correspond to statistically significant differences amongst the treatments (Tukey test, P-value < 0.05).
respectively. No differences in honeydew production were observed between the DCEF26 treatment and control. These results corroborate the hypothesis that the silencing of selected candidate effectors may affect D. citri feed performance.
Compared with in vitro conditions, plant responses could alter insect feeding performance 31 . To confirm the silencing interference on ACP feeding in planta, we evaluated the ACP behaviors on citrus plants after delivery of dsRNAs. This analysis was performed using the following three candidate gene effectors: DCEF27, DCEF28 and DCEF32. ACP adults were treated with dsRNA via artificial diet for five days. Then, insects were transferred to 'Rangpur lime' seedling leaves for five days. Mortality was evaluated daily and the quantification of honeydew droplets and gene expression analysis were performed at the end of the assays. After the first 48 hours, all treatments showed elevated mortality compared with the sucrose and GFP controls (Fig. 5), presenting 44, 41 and 47% mortality for the DCEF27, DCEF28 and DCEF32 gene treatments after five days, respectively. The honeydew analysis demonstrated a reduction in excreta production by dsRNA-treated insects (Fig. 5). A decrease of 10 and 38% for the tinted area was observed for the DCEF28 and DCEF32 treatments, respectively, compared with sucrose. dsDCEF27-treated insects showed the strongest reduction in honeydew production, showing tinted areas that were 70% lower than the controls (Fig. 5). These results demonstrated that knockdown of D. citri effectors could also interfere with ACP feeding in planta.

Evaluation of D. citri feeding behaviors.
To confirm the changes to D. citri feeding behaviors caused by the silencing of candidate effectors, EPG analysis was applied to reveal the differences by the treatments. EPG recordings were performed on adult females after dsRNA delivery via an artificial diet for five days. The DCEF28 and DCEF32 treatments showed long feeding periods on xylem (waveform G) (Fig. 6A) compared with the controls (sucrose and GFP). Moreover, the proportion of individuals that produced waveform G (PPW) was lower for the treatment GFP than in the other treatments (Table 2).
Sustained phloem ingestion (E2 > 10 minutes) was significantly reduced for the DCEF32 treatment compared to the other treatments (Fig. 6B). Furthermore, the proportion of individuals that underwent ingestion in the phloem (E2 and E2s) was lower for the DCEF32 treatment when compared to the other treatments ( Table 2).
The mean duration per event (WDE) of non-probing was significantly longer for the DCEF32 treatment than for the sucrose and GFP treatments (Fig. 6C). The WDE of C (stylet pathway) and probe were significantly shorter for the sucrose treatment control compared with the other treatments (Fig. 6D,E). The WDE of D was significantly longer for the DCEF27 treatment when compared with the GFP treatment (Fig. 6F).

Gene expression analysis.
After five days of dsRNA delivery via an artificial diet, gene expression analysis showed a reduction in the mRNA levels of all the effector genes, confirming the knockdown of the candidate effectors (Fig. 7). The strongest mRNA reduction was observed for the DCEF33 and DCEF32 genes, which presented 63 and 59% inhibition in their mRNA levels, respectively. The DCEF26, DCEF27 and DCEF28 genes www.nature.com/scientificreports www.nature.com/scientificreports/ showed a reduction of 34, 39 and 43% of their mRNA levels, respectively, compared with the GFP control. The DCEF35 effector presented an 18% inhibition in gene expression (Fig. 7).
To confirm the maintenance of effector silencing in the in planta experiments, gene expression analysis was performed 24 and 120 hours after the transfer of the dsRNA-treated insects from an artificial diet to 'Rangpur lime' plants ( Fig. 8). During the first 24 hours when the dsRNA-treated ACPs were fed plants, all the analyzed genes showed a reduction in their mRNA levels, proving that the D citri effectors still silenced (Fig. 8). However, the dsRNA-treated D. citri showed restoration of effector gene mRNA levels after 120 hours of feeding with citrus plants (Fig. 8).

Discussion
Candidate effectors prediction and validation. Salivary proteins are important for proper insect feeding and promoting lubrification, digestion and penetration 38 . These proteins also play important roles in plant-insect interactions 34 . For example, the potato aphid Macrosiphum euphorbiae Thoma (Hemiptera: Aphididae) presents an effector that interacts with two plant defense compounds to enhancing aphid fecundity 39 . The Hessian fly, Mayetiola destructor (Say 1817) (Diptera: Cecidomyiidae), phosphatase 2 C effector family elicits effector-triggered immunity in wheat by interacting with signal transduction pathways 40 . Furthermore, feeding is the principal vehicle of Ca. Las acquisition by ACP. Studies have demonstrated that long phloem ingestion by ACP nymph instars is responsible for the high acquisition efficiency of Ca. Las by immature D. citri instars 14 . Based on this, we proposed the identification of D. citri candidate effectors, looking for proteins important for interactions between ACP, citrus and Ca. Las.
Bioinformatics tools are the first line for the prediction of prokaryote or eukaryote effectors 41,42 . Independent of species, effector prediction is based on the following same criteria: amino acid composition and primary protein sequences, such as the presence of signal peptides, absence of anchor membrane domains, and secondary characteristics such as species-specific sequences, nuclear localization signals (NLSs), three-dimensional structure, positive selection, etc 41 . Thus, we use a Pea aphid bioinformatics pipeline 25 as a model to predict ACP candidate effectors. Pea aphid effectors are some of the best characterized insect effectors 28,43,44 , and these tools allowed the identification of 131 ACP candidate effectors (Table S1). Saliva or salivary gland protein sequences are very often applied for the prediction of insect effectors [45][46][47] . However, in the absence of this kind of sequence source, the Pea aphid pipeline was successfully applied for mite effector prediction using full body genomics and transcriptomic data 48 , as was also applied in this work. These studies support the legitimacy of the results obtained here. Seeking to ensure that the predicted candidate effectors are manly secreted in ACP saliva, all proteins that presented similarity with arthropod gut proteins were excluded. We identified twenty-nine D. citri candidate effectors showing similarity with arthropod salivary proteins (Table S2), supporting the hypothesis that these effectors may be produced and secreted by D. citri salivary glands. www.nature.com/scientificreports www.nature.com/scientificreports/ The major ACP candidate effectors are composed of D. citri-unique unknown function proteins (Table S1). Prediction of aphid effectors also showed a large number of unknown functional proteins 28,29 . Insect effectors are mostly species-or genus-specific 36,43,49 . An elevated number of unique uncharacterized proteins may indicate  www.nature.com/scientificreports www.nature.com/scientificreports/ that the effectors are highly host-specific, resulting from strong selection pressures to adaptation between insects and their host plant 28,29 . Bioinformatics analysis revealed nuclear localization signals (NLSs) and chloroplast transit peptides (cTP) in several ACP effectors (Table S1; Fig. S1), showing that these putative effectors may act on important host organelles (Table S1; Fig. S1). The VPS52 effector protein from M. persicae is relocated to post-Golgi and pre-vacuolar compartments in Nicotiana benthamiana plants and interferes with aphid virulence 50 . The B. tabaci effector Bsp9 acts on the host cell cytoplasm, interacting with plant WRKY33 transcription factors to suppress defenses 51 .
D. citri effectors genes are expressed during different ACP life stages (Fig. S2). These candidate effectors genes are also overexpressed in D. citri head compared with their body, supporting the hypothesis that these proteins may be secreted by ACP saliva (Fig. 1). Over-expression in the insect head was also a criterion applied to discriminate effectors from aphid and mite species 29,48 . Moreover, gene expression analysis of Ca. Las-infected and Ca. Las-free nymphs and adults showed that the D. citri effectors genes are modulated in response to these bacteria (Fig. 2), which is indicative that these genes could interfere with the Ca. Las transmission process by D. citri.
Silencing of candidate effectors on D. citri feeding behavior. Effectors are strongly related to insect feed performance. Salivary compounds may elicit or suppress plant defenses, increasing or inhibiting insect feeding on its host plants 52 . Thereby, feeding performance and fecundity are the main phenotypes evaluated in insect effector studies. Silencing of angiotensin-converting enzymes (ACEs) genes from A. pisium interferes with phloem sap ingestion, enhancing the aphid mortality 23 . The knockdown of salivary endo-b-1,4-glucanase from N. lugens decreased its feeding performance in rice 53 . To ensure that the predicted D. citri effectors are feeding-related, we evaluated the ACP feeding performance after gene silencing. Gene expression evaluation occurred on ACP adults after feeding with an artificial diet containing dsRNAs for the effectors or dsRNAs for GFP for five days. The data correspond to the mean of three independent experiments. *corresponds to statistically significant differences amongst the treatments (t-test, P-value < 0.05). , and DCEF32 (C) compared with GFP control observed 24 hours (light grey) and 120 hours (dark gray) after ACP adults were treated with dsRNAs for effectors and transferred for feeding on 'Rangpur lime' leaves. The data correspond to the average of four independent experiments. Different letters correspond to statistically significant differences amongst the treatments (Tukey test, P-value < 0.05).

Scientific RepoRtS |
(2020) 10:5992 | https://doi.org/10.1038/s41598-020-62856-5 www.nature.com/scientificreports www.nature.com/scientificreports/ Classical strategies for evaluating insect feeding include monitoring the salivary sheath and excreta production 54,55 . We observed that knockdown of ACP effectors in vitro assays promoted changes in the salivary sheath size and a strong reduction in honeydew production (Fig. 4), which indicates that the size of the salivary sheath may affect ACP ingestion.
The salivary sheath is one of two common salivary secretions produced by phytophagous hemipterans. Resulting from the solidification of insect gelling saliva, the salivary sheath is a solid structure that covers insect stylet providing orientation and stability during the piercing process and also protects the insect stylet from plant defenses [56][57][58] . The production of this structure is tightly related to appropriate insect feeding. Even so, excretion production is directly associated with insect ingestion 59,60 . Similar phenotypes were observed upon knockdown of salivary sheath proteins in the brown planthopper N. lugens, which also resulted in a reduction of honeydew amounts 61 .
However, these alterations did not directly affect mortality rates. The DCEF33 and DCEF35 treatments, which showed the strongest reduction in salivary sheath size and honeydew droplets, presented no significant mortality in the artificial diet experiments (Fig. 3). Similar patterns were observed when silencing the NlShp gene from N. lugens, which cause changes in insect feeding behaviors in vitro; however, knockdown of this gene did not cause significant insect mortality 31 .
Additionally, knockdown of ACP effectors also interferes with ACP feeding in planta. An expressive decrease in excreta production in dsDCEF27 and dsDCEF32-treated insects and elevated mortality were observed when these insects were fed the 'Rangpur lime' leaves (Fig. 5).
Interference of feeding performance caused by effector silencing was proven by evaluation of D. citri feeding behaviors through EPG analysis, which showed a significant reduction in phloem ingestion by the dsD-CEF32-treated insects. Only 50% of the dsDCEF32-treated insects showed E2 and E2 > 10 min waveforms (Table 2). Moreover, the number (NWEI) of E2 > 10 min waveforms was reduced by 50% for the DCEF32 treatment compared with the other treatments (Fig. 6B). These results indicate that dsDCEF32-treated insects seldomly performed passive phloem ingestion. Since Ca. Las is a phloem-inhabiting bacteria 62 , these results allow the hypothesis that DCEF32 may be important for the Ca. Las transmission process. Luo and collaborators (2015) showed that ACP adults that displayed an E2 waveform (passive phloem-sap uptake from sieve elements) were subsequently detected as Ca. Las-positive, demonstrating these EPG waveforms are strongly associated with Ca. Las acquisition 63 . In contrast, the strong correlation between the duration of phloem salivation (E1 waveform) and Ca. Las inoculation was described by Wu and collaborators (2016). These studies demonstrated that phloem phases (E1 and E2) are a critical point for the transmission of Ca. Las by D. citri [63][64][65] .
An interesting observation was the increased frequency and time of waveform G, which represents xylem ingestion, for the DCEF28 and DCEF32 treatments (Fig. 6A). In general, the G waveform presents low frequency and duration during D. citri feeding behaviors. Previous analysis demonstrated that the G waveform occurs on approximately 30% EGP recordings for D. citri adults, with durations ranging from 9 to 18% of the recording time 14,66,67 . Surprisingly, our results showed a proportion of individuals that had G waveforms (PPW) greater than 70% for the dsRNA-effector treatments ( Table 2). The dsDCEF28-and dsDCEF32-treated insects showed longer feeding periods on xylem vessels ( Fig. 6A; Table S4). G waveform formation for phloem sap feeders is frequently associated with dehydration or osmotic balance 68,69 . During citrus-ACP interactions, an increase in the G waveform duration was reported when ACP feeds on CLas-infected plants 14,64 . D. citri that performed xylem ingestion more often and for long periods were also observed on mature leaves compared with young ones due to the more pronounced sclerenchymatous ring around phloematic vessels, which difficult for phloem ingestion 67 . In this case, we hypothesize that xylem ingestion occurs to compensate for inefficient phloem ingestion 67 .
Effect of silencing on candidate effectors gene expression. Knockdown of D. citri effectors was proved by the significant reduction of the mRNA levels of these genes after delivery of dsRNA via an artificial diet and 24 hours after transferring the dsRNA-treated insects to Rangpur plants ( Figs. 7 and 8). Nevertheless, after 120 hours of constant feeding of dsRNA-treated insects with citrus plants, the mRNA levels of D. citri effector genes were restored (Fig. 8), demonstrating that the silencing effects of the analyzed genes are transient in the absence of a constant dsRNA source. The lack of an efficient silencing signal amplification system and the presence of nucleases that degrade dsRNAs may be factors that could explain the observed phenomena, though none of these observations have been previously described for ACP 70 . Several hemipterans species have been reported to have nucleases that quickly degrade dsRNAs in different body fluids 71 . Differences in nuclease activity in different tissues were also described as an important factor for determining dsRNA delivery strategies for some insects [72][73][74] .

Conclusion
Most of the RNAi studies applied to D. citri control are focused on identifying candidate genes that promote significant mortality or insecticide susceptibility as well as new efficient strategies for dsRNA delivery, however, no practical use of this technology against D. citri has emerged. Studies focused on D. citri-Citrus-Ca. Las are recent, and many gaps regarding the genetic and molecular aspects of these interactions need to be fulfilled. Thus, the increase of the knowledge about D. citri biology and its interactions with citrus and Ca. Las is essential to develop new and efficient control strategies for HLB. Despite of recent identification of the D. citri saliva proteome, ACP effectors or salivary molecules that interact with plant hosts have not been identified to date. This study predicted D. citri effectors using a classical bioinformatics pipeline. Gene expression analysis combined with RNAi and EPG techniques allowed us to prove that the predicted effectors are directly related to ACP feeding performance, which is similar to the majority of insect effectors identified to date. The DCEF32 gene stands out amongst the identified effectors due to its action on D. citri phloem ingestion, which is the main colonization site of Ca. Las in plants. Moreover, gene expression analysis showed that the DCEF32 gene is modulated in response to Ca. Las Scientific RepoRtS | (2020) 10:5992 | https://doi.org/10.1038/s41598-020-62856-5 www.nature.com/scientificreports www.nature.com/scientificreports/ during both the nymph and adult stages. These results may indicate that the DCEF32 effector could participate in Ca. Las-D. citri interactions. However, how D. citri effectors interact with host plants and/or Ca. Las still needs to be investigated. Understanding the participation of D. citri effectors in the HLB pathosystem can elucidate the molecular mechanisms of Ca. Las transmission by D. citri and its interaction with citrus plants. This knowledge is essential to develop novel and efficient control strategies for HLB.

Materials and Methods
Sequence database. Transcripts sequences from Arizona University transcriptome studies 35 were initially used as a database. Complementary version 0.9 of the transcriptome and version 2.0 of the D. citri genome available on Citrus greening Solutions website (citrusgreening.org) were added to the bioinformatics analysis.
Bioinformatics analysis. The bioinformatics for the identification of D. citri candidate effectors was based on a pipeline for the prediction of aphid effectors 25 . First, redundant sequences and proteins with 1000 > amino acids were removed from the analysis. Redundant sequences were those that presented a similarity > 95% and an e-value < 10 −10 . Secreted proteins were predicted using SignalP 4.0 75 . Proteins that had transmembrane domains were predicted using the TMHMM and GPI-SOM software 76,77 and removed from future analysis. InterPro, Uniprot and Prosite 78,79 were used to identify and exclude ordinary proteins. Proteins that displayed similarity with arthropod gut proteins were removed through blastX against arthropod gut ESTs. To identify potential D. citri salivary proteins, blastX against arthropod salivary proteins was performed. Additionally, prediction of ACP candidate effector subcellular localization was performed using the NetNES 1.1 Server, Localizer and LocTree tools 80-82 . Asian citrus psyllid colony. Mesh cages containing Murraya paniculata were used to rear healthy D. citri at 25 ± 2 °C under a 14:10 h (light:dark) photoperiod and 60 to 70% relative humidity (RH). CLas-infected ACP colonies were reared on Ca. Las-infected Pera sweet orange trees (symptomatic and PCR positive) were grown under greenhouse conditions for light, temperature and humidity. Both ACP populations were maintained at the biotechnology laboratory at the Centro de Citricultura 'Sylvio Moreira' .
Nuclei acid isolation and cDNA synthesis. RNA extraction was performed using a Direct-zol RNA MiniPrep and DNAse I Set (Zymo Research, Irvine, CA, USA) according to the manufacturers' instructions. A NanoDrop ND 8000 spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA) was used to measure the RNA integrity and quality. One microgram of extracted RNA was applied for cDNA synthesis using the GoScript reverse transcription system (Promega) according to the manufacturer's protocol. After the reaction, the cDNA was diluted 1:25 for further analysis.
Double-strand RNA synthesis. The target sequences of ACP candidate effectors were amplified by RT-PCR using specific gene primers conjugated with 19 bases of the T7 RNA polymerase promoter (Table S3). The GFP sequence was amplified from the pIG1783f plasmid 83 . The PCR reaction was performed using 5 μL of cDNA, 25 μL of Gotaq colorless MasterMix (Promega), and 100 nM of each primer pair in a final volume of 50 μL. Thermal profiles were determined using 10 cycles of 94 °C for 30 s, 1 min at 10 °C above the optimal primer annealing temperature, and 72 °C for 1.3 min, followed by 35 cycles of 94 °C for 30 s, 1 min at the optimal primer annealing temperature, and 72 °C for 40 s. DsRNAs were synthesized from purified PCR products using the MEGAscript RNAi kit (Ambion, catalog no. AM1626) according to the manufacturer's protocol. The purified dsRNAs were quantified spectrophotometrically at 260/280 nm and the integrity was examined by agarose gel electrophoresis.
Delivery of dsRNA and bioassay in vitro. Ten-day-old adults were used for the RNAi experiments. The insects were maintained in plastic cages containing a sachet composed of 100 µL of an artificial diet (30% of sucrose and 0.1% green and 0.4% yellow food dyes) homogeneously distributed between two parafilm layers 84 and filter paper on the bottom of the cage. Feeding assays were performed for a period of five days at room temperature under a 14:10 h (light: dark) photoperiod and 60 to 70% HR. Each treatment consisted of five biological replicates (10 insects per replicate). Insect mortality was evaluated daily. Three independent bioassays were performed.
After the feeding period, live insects were collected for further gene expression analysis. Salivary sheaths present on the parafilm membrane were dyed using 0.1% safranin, quantified and photographed with a light microscope using a 40x objective. The sizes of the salivary sheaths were measured using the ImageJ software 85 . Honeydew droplets present on the filter paper were dyed with 2% ninhydrin, counted and measured using the Quant software 86 . The remaining artificial diet was collected and examined by electrophoretic roll to observe the dsRNA stability (Fig. S4). Statistical analyses were performed using one-way ANOVA (P < 0.05) with the GraphPad Prism software 87 . In vivo bioassay for silencing effect evaluation. After the in vitro dsRNA delivery described above, living insects were confined to 'Rangpur lime' seedling leaves using clip cages. The clip cages were made from 90-mm in diameter plastic Petri dishes. A hole (7 mm) was made on the lateral side of the Petri dishes to facilitate the accommodation of a leaf blade and petiole. The leaf was attached to the Petri lid using adhesive tape. The filter paper was placed in the bottom of the clip cage to collect the honeydew droplets. The plates were sealed using a cotton and PVC film. Mortality was evaluated for five days and the feeding abilities were analyzed based on the quantity of honeydew droplets present on filter paper as described before. The experiment was repeated four times using four biological replicates for each treatment (ten adults ACP per biological replicate). Statistical analyses were performed using one-way ANOVA (Tukey test) and the t-test (P < 0.05) using the GraphPad Prism software 87  www.nature.com/scientificreports www.nature.com/scientificreports/ Feeding behavior evaluation of D. citri by electron penetration graph (EPG). After delivery of the dsRNA via the artificial diet as described above, we performed feeding behavior assays using an EPG technique, which allows for the real-time study of an insect's stylet activities in plant tissues. Females ACPs were anesthetized for 5 seconds with CO 2 and immobilized using a vacuum chamber under a dissecting microscope. Then, a gold wire (3 cm length, 20 µm in diameter; EPG Systems, Wageningen, The Netherlands) was attached to the ACP pronotum with a small droplet of water-based silver glue. The opposite end of the gold wire was glued to a thin copper wire (2 cm length), which was connected to the EPG probe. Another copper electrode (10 cm long, 2 mm wide) was inserted into the plant soil.
The insects were placed on the abaxial surface of a young 'Rangpur lime' leaf. The EPG waveforms were recorded using a Direct Current 8-channel EPG device model Giga-8d, with Stylet+ for Windows software (EPG Systems). The recordings were carried out in a room at 25 ± 1 °C inside a Faraday cage (for electrical noise isolation) for 8 hours. The waveforms recorded for D. citri feeding behaviors were characterized according to previous reports 54 . The output is given in an Excel workbook 88 to calculate the treatment mean for each EPG variable. Fifteen individual recordings were performed for each treatment.
The EPG waveforms previously described for D. citri 66 were identified as follows: waveform np (non-probing behavior); waveform C (which indicates movement of the stylets in the intercellular apoplastic space); waveform D (a short waveform always observed between waveforms C and E1); E1 waveform (indicates salivation into phloem sieve elements); waveform E2 (correlated with passive phloem-sap uptake from sieve elements); and waveform G (active intake of xylem sap).
The experimental design was entirely randomized. The EPG data were transformed when necessary with ln (x + 1) or √(x + 1) to reduce the heteroscedasticity and improve the normal distribution. All parameters were analyzed with Tukey's test (P < 0.05). If the data did not follow a normal distribution according to the Shapiro-Wilk normality test, a nonparametric Kruskall-Wallis test (P < 0.05) was performed. The proportion of individuals that produced a specific waveform type (PPW) was compared among the different treatment groups using the chi-square test. All data were analyzed using the IBM Statistics SPSS 22.0 software 89 .
Gene expression analysis. To select candidate effectors by RT-qPCR, pools of 10 adults and 50 nymphs were employed for RNA extraction as described before. Gene expression analysis of different body parts was conducted using 10 heads and 10 bodies from ACP adults. The measurement of gene expression in insects silenced by dsRNA was performed using RNA extracted from a pool of four ACP adults per biological replicate from both the artificial diet and plant experiments.
RT-qPCR reactions were performed using 6.5 µL of GoTaq qPCR MasterMix (Promega), 120 nM of each gene-specific primer pair and 3 μL of the diluted cDNA in a final volume of 12.5 μL. The amplification cycles were performed on a 7500 Fast Real-Time PCR System device (Thermo Scientific, Waltham, MA, USA) using the following standard thermal profile: 95 °C for 20 s followed by 40 cycles of 95 °C for 3 s and 60 °C for 30 s. Five technical replicates were analyzed for each sample. The Cq values and the primer efficiency were estimated using the Miner software (http://miner.ewindup.info). The normalized gene expression analyses were performed with the 2 ΔCq method and relative gene expression was estimated using the 2-ΔΔCq method using the ribosomal genes S20 and S13 as reference genes. For statistical analysis, one-way ANOVA (Tukey test) and the t-test (P < 0.05) were performed using the GraphPad Prism software 87,90 .