Non-Target Effects of dsRNA Molecules in Hemipteran Insects

Insect pest control by RNA interference (RNAi)-mediated gene expression knockdown can be undermined by many factors, including small sequence differences between double-stranded RNA (dsRNA) and the target gene. It can also be compromised by effects that are independent of the dsRNA sequence on non-target organisms (known as sequence-non-specific effects). This study investigated the species-specificity of RNAi in plant sap-feeding hemipteran pests. We first demonstrated sequence-non-specific suppression of aphid feeding by dsRNA at dietary concentrations ≥0.5 µg µL−1. Then we quantified the expression of NUC (nuclease) genes in insects administered homologous dsRNA (with perfect sequence identity to the target species) or heterologous dsRNA (generated against a related gene of non-identical sequence in a different insect species). For the aphids Acyrthosiphon pisum and Myzus persicae, significantly reduced NUC expression was obtained with the homologous but not heterologous dsRNA at 0.2 µg µL−1, despite high dsNUC sequence identity. Follow-up experiments demonstrated significantly reduced expression of NUC genes in the whitefly Bemisia tabaci and mealybug Planococcus maritimus administered homologous dsNUCs, but not heterologous aphid dsNUCs. Our demonstration of inefficient expression knockdown by heterologous dsRNA in these insects suggests that maximal dsRNA sequence identity is required for RNAi targeting of related pest species, and that heterologous dsRNAs at appropriate concentrations may not be a major risk to non-target sap-feeding hemipterans.


Introduction
Ground-breaking studies on corn root worm and cotton bollworm conducted over a decade ago [1,2] provided proof-of-concept for the use of RNA interference (RNAi) in insect pest control. Since then, major advances have been made to understand the mechanism of RNAi in insects; to document among-species variation in RNAi efficacy; and to elucidate the genetic, molecular and biochemical bases of barriers to RNAi-triggered gene silencing [3][4][5]. RNAi, unlike most traditional insect control technologies, offers the potential for exquisite control over the specificity of control agents [6,7]. This is because the enzyme of the Argonaute family in the RNA-induced silencing complex (RISC) is a guide-dependent RNase, requiring perfect sequence complementarity between the target transcript and the 21 nt guide RNA, a small interfering RNA (siRNA) generated by Dicermediated cleavage of double-stranded RNA (dsRNA) [8]. There is a general expectation that by careful selection of the length and sequence of dsRNA, RNAi can be limited to target insect pest(s) without any direct, deleterious effect on non-target species, including beneficial insects such as pollinators and biological control agents [9,10].
The motivation for our research was the increasing recognition that empirical studies on the taxon specificity of RNAi are essential for the successful application of this technology to insect pest control. Our experimental design investigated the gene silencing hemipterans in agroecosystems would not be particularly susceptible to heterologous dsRNAs, and that dsRNAs designed against multiple hemipteran pests should include high numbers of 21-mer siRNAs that match the sequences of all the target species.

Insects
Acyrthosiphon pisum clone CWR09/18 [43] and clone SC_37 [44], each derived from a single parthenogenetic female collected from alfalfa crop in Freeville, NY and Ithaca, NY (USA), respectively, and were maintained on pre-flowering Vicia faba cv. Windsor at 20 • C with 16L:8D light cycle. The New York GPA clone of Myzus persicae [38] was reared at 25 • C and 16L:8D light cycle on Brassica juncea cv. Florida Broad Leaf. Bemisia tabaci MEAM1 derived from a collection from Euphorbia pulcherrima Willd. Ex Klotzsch in Ithaca, NY, USA in 1989 was maintained on Solanum lycopersicum cv. Florida Lanai at 25 • C and 14L:10D light. A culture of Pseudococcus maritimus were generated from three collections from Vitis vinifera cv. Chardonnay in Seneca County, NY in July-August, 2018 and maintained on V. vinifera cv. Pixie at 21 • C with 17L:7D [39].

RNA Extraction
Insects were homogenized in 300 µL RNAzol (catalog number R4533, Millipore Sigma, Burlington, MA) using Lysing matrix D beads (catalog number 116913050, MP Biomedicals, Santa Ana, CA, USA) and a FastPrep homogenizer (Santa Ana, CA, USA). Following addition of 120 µL water, the homogenate was vortexed vigorously for 15 s and incubated at room temperature for 15 min. The mixture was centrifuged at 12,000× g and 4 • C for 15 min and 360 µL supernatant was transferred to a new microcentrifuge tube. The supernatant was combined with 1.8 µL 4-bromoanisole and vortexed vigorously for 15 s followed by incubation at room temperatures for 5 min, and centrifugation at 12,000× g and 4 • C for 10 min. The supernatant was transferred to a clean microcentrifuge tube and an equal amount of isopropanol and 1 µL linear acrylamide was added and mixed by vortexing. The mixture was incubated for 10 min at room temperature for the aphids, and overnight at −20 • C for whiteflies and mealybugs. Following incubation, the samples were centrifuged, and the RNA pellet was washed twice with 75% alcohol and suspended in 25 µL nucleasefree water. The RNA was free of genomic DNA, as indicated by the absence of detectable product in qPCR assays of cDNA samples generated with no reverse transcriptase (using primers predicted to yield the same product for cDNA and genomic DNA).

cDNA Synthesis
cDNA was synthesized from 500 ng RNA in a 20 µL reaction using Superscript™ II kit (catalog number 18064014, ThermoFisher Scientific, Waltham, MA, USA). A 12 µL reaction containing 1 µL random primers, 1 µL dNTPs and 500 ng RNA was incubated at 65 • C for 5 min and transferred to ice. Two µL dithiothreitol, 4 µL 5× buffer and 1 µL water were added to the mixture and incubated at 25 • C for 2 min, followed by a transfer to ice, and addition of 1 µL reverse transcriptase. The final reaction was incubated at 25 • C for 10 min, 42 • C for 50 min and 70 • C for 15 min. The cDNA was stored at −20 • C.

Quantitative Real Time-PCR (qRT-PCR)
The 10 µL reaction mixtures were run in a C1000 Thermal Cycler with CFX96 TouchTM Real-Time PCR Detection System (Bio-Rad, Hercules, CA, USA). The reaction mixture contained 5 µL iQ SYBR Green supermix (catalog number 1708862, Bio-Rad), 1 µL cDNA, 3 µL water and 0.5 µM forward and reverse primers (Supplementary Table S1A). The template cDNA was dissociated at 95 • C for 3 min followed by 40 cycles of 95 • C for 10 s and 60 • C for 30 s. A dissociation curve, from 65 • C to 95 • C in 0.5 • C increment per 0.05 s, was generated for every sample and confirmed the specificity of the PCR reactions. β-tubulin and RPL32 were used as reference genes for A. pisum and M. persicae, and β-tubulin was used as reference gene for B. tabaci and P. maritimus, following published protocols [37,39,40]. No-template reaction was used as negative control. Average Ct-values of two technical replicates were calculated and the difference between treatment delta Ct and control delta Ct was used to calculate fold changes [45].

dsRNA Synthesis
Target sequences of NUC genes were amplified using primers (Supplementary Table  S1B) with the T7 promoter sequence at the 5 end of both forward and reverse primers and cloned in pGEMT Easy plasmid (catalog number A1360, Promega, Madison, WI, USA). The plasmid was transformed into Escherichia coli DH5α (catalog number 18258012, ThermoFisher Scientific) using the heat shock method. The dsGFP template comprised a 370 bp fragment of the GFP gene (coding the green fluorescent protein of the jellyfish Aequorea victoria) previously cloned into pGFP2 plasmid in E. coli DH5α, as in [37,40]. The dsRNA was purified with the Zymo DNA Clean and Concentrator kit (catalog number D4033, Irvine, CA, USA) following the manufacturer's protocol, and then quantified using a Nanodrop spectrophotometer.
A 20 µL reaction with 1 µg template DNA was used to synthesize dsRNA using AmpliScribe T7-Flash Transcription kit (catalog number ASF3507, Lucigen, Middleton, WI, USA). The template DNA was then removed by addition of 1µL RNase-free DNase at the end of the reaction. To remove the NTPs and other impurities, 79 µL water was added for a final volume of 100 µL, followed by the addition of an equal volume of 5 M ammonium acetate (catalog number AM9070G, ThermoFisher Scientific) and 250 µL ethanol. The mixture was incubated at −20 • C overnight, and then centrifuged at 12,000× g at 4 • C for 20 min. The resulting RNA pellet was washed twice with 75% ethanol and resuspended in 30 µL nuclease-free water.

Experimental Designs
For all experiments, the insects were administered dsRNA over two days by feeding from a liquid artificial diet aseptically enclosed within stretched Parafilm sheets, as described previously [46]. The diet comprised 0.5 M sucrose, amino acids at a total concentration of 0.15 M, micronutrients with potassium phosphate buffer, pH 7.0 [46], and dsRNA is stable in this diet [37,40]. For each species, all diet experiments were conducted under the same temperature and light regime as used for the plant-reared insects. The experiments on aphids included a pre-treatment of 2-to-5-day-old nymphs on dsRNA-free diets because A. pisum reared on plants to ages > 2 days do not feed well from diets.
The first experiment quantified food consumption of A. pisum clone CWR09/18 administered dsGFP. Thirty replicate groups of ten 2-day-old plant-reared nymphs were administered dsRNA-free artificial diet [46] ( Figure 1A). When 5-days-old, five replicate groups of 10 nymphs each were transferred to the artificial liquid diet supplemented with dsGFP at each of 0, 0.1, 0.2, 0.5 and 1 µg µL −1 or water only in diet cages (3.5 cm diameter × 0.5 cm height). A pre-weighed circle of aluminum foil was placed under the feeding aphids in each diet cage to collect the honeydew. Two days later (day 7), the number of surviving aphids per cage was scored, and the pooled weight determined. The honeydew accumulation was quantified by subtracting the final weight of the aluminum foil from the initial weight. All weights were determined on a Mettler MT5 microbalance to the nearest µg. All the aphids in every diet cage survived the experiment, apart from one replicate containing 0.2 µg µL −1 diet, and this replicate was removed from the analysis. The second experiment determined the effect of homologous dsNUC on NUC expression in aphids. Each replicate group of five 5-day-old A. pisum that had been raised on dsRNA-free diet from day 2, were administered dsApNUC (dsRNA against the nuclease gene NUC of A. pisum NCBI accession number XM_003242604.4, described in [40] as NUC1) at 0, 0.05, 0.1, 0.2 and 0.4 µg µL −1 diet ( Figure 2A). The same experimental design was adopted for M. persicae, using dsMpNUC (dsRNA against the M. persicae NUC gene, NCBI accession number XM_022327343.1) (Figure 2A). The aphids were harvested on day-7 and the group of aphids from each cage (i.e., each replicate) was stored at −80 • C prior to RNA extraction. As the supply of insects was limited, the experiment was conducted twice for each species. In the first iteration, five replicates were used for both species. In the second iteration, five replicates were used for A. pisum and four for M. persicae. The data did not differ significantly between the two iterations (p = 0.43 for A. pisum and p = 0.90 for M. persicae), and the results for the two iterations for each species were pooled, yielding 10 and 9 replicate cages for A. pisum and M. persicae, respectively. The final experiments investigated the effect of heterologous dsNUCs on NUC expression. For the analysis of the two aphid species, dsApNUC and dsMpNUC at 0.2 µg µL −1 was administered to 5-day-old aphids of A. pisum clone SC_37 and M. persicae, with dsRNA-free diet and dsGFP as negative controls ( Figure 3A). Five replicate cages, each containing 5 aphids, were used. The analysis of dsNUC cross-reactivity was then extended to the mealybug P. maritimus (ca. 30 days old) and the whitefly B. tabaci (3 -days post-eclosion) ( Figures 4A and 5A). The homologous P. maritimus dsNUC (NCBI accession number MT187988.1, referred to as NUC1 in [39]), the two homologous B. tabaci NUC1 and NUC2 (NCBI accession numbers KX390872.1 and KX390873.1, respectively [37]) and the heterologous dsNUC from both aphids, were administered at 0.2 µg µL −1 over 2 days. The experiments included two negative controls: dsRNA-free diet and dsGFP. Twenty insects were administered to each diet cage, and the harvested insects were stored at −80 • C prior to analysis. Due to limited supply of insects, the experiments comprised two iterations: three replicates for the first iteration and two replicates for the second iteration for P. maritimus; and four replicates in each of the iterations for B. tabaci. The data for the two iterations for each insect species were pooled, following confirmation that the data did not differ significantly between two iterations (p = 0.26 for P. maritimus; p = 0.35 and 0.69 for B. tabaci NUC1 and NUC2, respectively).

Statistical Analysis
R software version 3.5.1 was used to analyze the data (R Core team). For the analysis of the response of A. pisum to dietary dsGFP, variation in aphid weight was analyzed by nonparametric Kruskal-Wallis test (following demonstration of heterogeneity of variances), and non-parametric Dunn's test for treatment comparisons to the dsRNA-free control using PMCMRplus package version 1.4.4 [47]. All other datasets were normally distributed with homogeneous variances. Honeydew production by pea aphids was investigated as a linear model [48] followed by Dunnett's post hoc test using emmeans package version 1.3.3 [49]. The log2-fold variation in gene expression was analyzed as a linear model using the lm function in lme4 package version 1.1-2.1 [48] with treatments as categorical predictors.

Optimization of the Dietary Concentration of dsRNA
To identify the optimal concentration of dietary dsRNA for analysis of the reactivity of dsRNA in different insect species, we first investigated the sequence-non-specific effects of dsRNA at different concentrations. The experiments were conducted on A. pisum clone CWR09/18, which has been used extensively in previous feeding analyses and RNAi studies in our laboratory [40,43,44]. The insects were fed for two days on diets containing 0-1.0 µg dsGFP µL −1 (with no sequence homology to insect genomes), and on water as a control. Compared to aphids on the dsGFP-free diet, the final weight of the aphids was significantly depressed on the water-only control but not on the diets containing dsGFP ( Figure 1B). However, the weight of honeydew produced by the insects, which provides a useful index of the amount of food consumed by aphids [50], was significantly depressed while they were on diets containing 0.5 and 1.0 µg dsGFP µL −1 , but not on diets with lower dsGFP concentrations, relative to the dsGFP-free diet ( Figure 1C). The aphids on the water control diet produced negligible amounts of honeydew, reflecting the absence of phagostimulatory sugar. These data suggest that dsGFP at ≥0.5 µg µL −1 has an antifeedant effect on the insects, and that concentrations <0.5 µg µL −1 should be used to avoid sequence-non-specific effects of dsRNA.
We then investigated the expression knockdown in aphids administered dsRNA against the nuclease gene NUC at concentrations in the range 0.05-0.4 µg dsNUC µL −1 . The experiments were conducted on A. pisum clone SC_37 and M. persicae clone GPA, which are highly susceptible to RNAi [38,44]. For both species, NUC expression was significantly reduced, on average by ca. 50% relative to the dsRNA-free diet, on diets containing 0.2 µg µL −1 but no other dsRNA concentration tested ( Figure 2B,C). Subsequent experiments investigating the efficacy of aphid dsNUC against other insect species focused on dsRNA at 0.2 µg µL −1 . Possible reasons for the low efficacy of dsRNA at 0.4 µg µL −1 are considered in the Discussion.

dsNUC Cross-Reactivity
We first compared the gene expression of NUC in A. pisum SC_37 and M. persicae GPA administered dsApNUC or dsMpNUC at 0.2 µg µL −1 . These genes have high sequence identity, and 77 of the 308 predicted 21 nt siRNA sequences derived from the two dsNUCs are perfect matches (Table 1, Figure S1). For both species, the homologous dsNUC significantly reduced NUC expression relative to dsRNA-free control, but the heterologous dsNUC had no significant effect ( Figure 3B,C). We then extended the analysis to investigate the expression of NUC genes in the mealybug P. maritimus and the whitefly B. tabaci administered heterologous dsNUC from the two aphid species. The dsNUCs of these species have low sequence identity with aphid dsNUCs and share no perfectly matching 21-mer sequences (Table 1; Figures S2-S7). We predicted that in the absence of sequence-non-specific effects, the aphid dsNUCs would not significantly affect the expression of NUC genes in these insects, while the homologous dsNUC would mediate NUC expression knockdown. The results for P. maritimus were fully consistent with this expectation ( Figure 4B). Our analysis of B. tabaci took into account that two nuclease genes, BtNUC1 and BtNUC2, are expressed in the gut [40]), unlike the other species with a single gut-expressed NUC gene. BtNUC1 and BtNUC2 have low sequence identity and share no 21-mer sequences (Table 1 and Figure S8). As predicted, the expression of both B. tabaci NUC genes was significantly reduced only in whiteflies administered the homologous dsNUC ( Figure 5B,C). However, a nonsignificant trend of reduced NUC expression was evident in several treatments, with mean transcript abundance being more than halved for BtNUC1 in insects administered dsApNUC, and for BtNUC2 in whiteflies administered dsGFP and dsBtNUC1. These findings raise the possibility of sequence-non-specific effects of dsRNA administered at relatively low concentrations in B. tabaci.

Discussion
The successful application of RNAi for pest control is a balancing act, requiring high efficacy against pest species with minimal deleterious effects on non-target organisms. Achieving this balance requires a firm understanding of both sequence-dependent and sequence-non-specific effects of dsRNA, while recognizing that responses to RNAi triggers can vary widely between different species. This variation is particularly evident for insects, given substantial reported differences in RNAi susceptibility between insect orders [3][4][5], related insect species [11,51] and even within species [44,52,53]. This study on the specificity of gene expression knockdown in plant sap-feeding insects has revealed both sequencenon-specific effects at high concentrations of administered dsRNA (≥0.5 µg µL −1 diet) and sequence-dependent effects on gene expression at a lower concentration (0.2 µg µL −1 ). In this Discussion, we address the implications of these findings, particularly in relation to the application of RNAi for the control of plant sap-feeding crop pests.
Insight into sequence-non-specific effects of dsRNA was obtained from the analysis of pea aphids A. pisum administered dsGFP at different dietary concentrations (Figure 1). The significantly reduced food consumption from diets containing ≥0.5 µg dsGFP µL −1 diet, without significant concomitant reduction in weight or any mortality over two days, is strongly indicative of an antifeedant effect of the dsRNA. Factors contributing to reduced feeding may include high viscosity of the liquid diet [54] arising from the addition of dsRNA at high final concentrations, and possible residual chemical contaminants in the administered dsRNA (despite our use of rigorous protocols for dsRNA purification, see methods). Plant sap-feeding insects may be more susceptible to these effects than chewing insect pests, such as Lepidoptera and Coleoptera. This is because dsRNA can be added to solid foods with minimal effects on the physical properties of the diet, and plant sap-feeders are notoriously intolerant of low concentrations of impurities in artificial diets [55]. Nevertheless, our findings raise the possibility that antifeedant effects may contribute to some of the observed sequence-non-specific effects of dsRNA on other insects, including increased mortality and global changes in gene expression [11,23]. If further research demonstrates that chemical contaminants contribute to the sequence-non-specific effects of dsRNA preparations, this issue should be addressed in the design of commercial preparations using exogenously-applied dsRNA for insect pest control [56][57][58].
An alternative and widely-adopted approach to investigate sequence-non-specific effects of dsRNA is to include dsGFP (or dsRNA against another sequence with no homology to the insect genome) in experiments quantifying dsRNA effects on expression of a target gene [3,5]. For example, the non-significant effects of dsGFP on NUC expression by the insects tested in this study (Figures 3-5) offers a first indication that, at 0.2 µg µL −1 , dsRNA does not induce substantial sequence-non-specific effects on expression of this test gene in these insects. In support of this interpretation, the mortality of all the insect species studied here was negligible in these experiments, although other indices of insect performance, including developmental rate and fecundity, were not investigated. Nevertheless, as highlighted in the Results section, the expression of the B. tabaci nuclease genes NUC1 and NUC2 was reduced two-fold by dsRNAs with no predicted sequence-dependent crossreactivity, including dsGFP for BtNUC2. Although these differences were not statistically significant, they raise the possibility that B. tabaci and other whiteflies may be particularly susceptible to sequence-non-specific effects of dsRNA. These considerations indicate the importance of further research to quantify the patterns and underlying processes of nonsequence dependent effects of dsRNA in different insects, and also to ensure that RNAi studies have the appropriate controls to discriminate between sequence-dependent and sequence-non-specific effects.
An issue arising from these considerations is our finding that expression knockdown of the NUC gene in both aphid species was lower in aphids administered the homologous dsNUC at 0.4 µg µL −1 diet, relative to the lower concentration of 0.2 µg µL −1 . Antifeedant effects of dsRNA at 0.4 µg µL −1 may have contributed to this effect. Additionally or alternatively, the regulatory circuits controlling NUC gene expression may be complex, with a more robust or more rapid compensatory upregulation of expression triggered by dsNUC at 0.4 than at 0.2 µg µL −1 . Further experiments investigating the time course of NUC expression and food consumption in aphids administered different dsNUC concentrations should resolve these issues. More generally, these results demonstrate that the efficacy of RNAi is not necessarily enhanced by increasing the dose of dsRNA, and reinforce the importance of determining the optimal concentration for each target gene [59].
The sequence-dependent effects of dsRNA are also critical for the effective and safe application of RNAi in the control of plant sap-feeding insect pests. As considered in the Introduction, the number and length of exactly matching siRNA sequences derived from a heterologous dsRNA that are required for successful expression knockdown can vary with the target gene and insect species. Although the cross-reactivity of dsRNAs in plant sapfeeding insects has not been investigated extensively, evidence that heterologous dsRNA can confer effective RNAi is provided by a study of CP19, an essential gene coding aphid cuticle protein 19 which protects the insects against cuticular water loss and desiccation [60]. Across the three aphid species studied, CP19 had 94% sequence identity (the number of matching 21-mer sequences was not reported) and dsCP19 matching the CP19 sequence of Aphis citricidus reduced CP19 expression in both A. pisum and M. persicae, with attendant mortality effects. The different result in our study, where no detectable cross-reactivity in expression of NUC genes of A. pisum and M. persicae was obtained, may reflect a critical difference in the number of perfectly matching siRNA molecules (just 25% of the siRNAs met this criterion, despite 92% sequence identity of the dsNUCs constructed for the two species; see Table 1). Other factors may also be important, including choice of gene, aphid genotype, culture conditions and possibly dsRNA concentration. In [60], dsCP19 was administered at 1.5 µg µL −1 by petiole dip, raising the possibility of sequence-non-specific effects, although the concentration ingested by the aphids was not determined. More broadly, the interesting differences between the findings of [60] and this study illustrate how further investigation are required on the species-specificity of dsRNA against plant sap-feeding insect pests.
We conclude with two points. First, developing an understanding of the factors that promote sequence-non-specific effects of dsRNA will be crucial to enable accurate predictions of the deleterious effects of heterologous dsRNAs on non-target, plant sap-feeding insect species. This study adds to the conclusions from prior research on various insects that these effects tend to be mediated by high dsRNA concentrations [3,5,25], and it also demonstrates that dsRNA formulations can have significant antifeedant effects for orallydelivered RNAi. Second, there is a need to extend research on the relationship between the number of matching siRNA sequences and RNAi efficacy of dsRNA molecules [16,61,62], so that effective dsRNAs can be designed against single or closely-related pest species that vary in sequence identity with the target gene. Attention to these issues will facilitate the development and deployment of RNAi-based solutions for the management of plant sap-feeding and other insect pests.
Supplementary Materials: The following are available online at https://www.mdpi.com/2073-442 5/12/3/407/s1, Table S1. Primer sequences. Figure S1. Alignment of A. pisum NUC (NCBI accession number: XM_003242604.4) and M. persicae NUC (NCBI accession number: XM_022327343.1). Red font represents the dsNUC sequence and the yellow background represents perfect matches greater than or equal to 21 nt between A. pisum dsNUC and M. persicae dsNUC. Figure S2. Alignment of A. pisum NUC and P. maritimus NUC (NCBI accession number: MT187988.1). Red font represents the A. pisum dsNUC sequence. Figure S3. Alignment of A. pisum NUC and B. tabaci NUC1 (NCBI accession number: KX390872.1). Red font represents the A. pisum dsNUC sequence. Figure S4. Alignment of A. pisum NUC and B. tabaci NUC2 (NCBI accession number: KX390873.1). Red font represents the A. pisum dsNUC sequence. Figure S5. Alignment of M. persicae NUC and P. maritimus NUC. Red font represents the M. persicae dsNUC sequence. Figure S6. Alignment of M. persicae NUC and B. tabaci NUC1. Red font represents the M. persicae dsNUC sequence. Figure S7. Alignment of M. persicae NUC and B. tabaci NUC2. Red font represents the M. persicae dsNUC sequence. Figure S8. Alignment of B. tabaci NUC1 and B. tabaci NUC2. Red and green represent B. tabaci dsNUC1 and dsNUC2 sequences.  Data Availability Statement: All data generated or analyzed during this study are available from the corresponding author on reasonable request.