Host plant affects morphometric variation of Diaphorina citri (Hemiptera: Liviidae)

Host plants and insect rearing

Plants were grown from seed starting on February 19, 2013. Both plants and ACP were reared in a conventional greenhouse supplied with an evaporative cooling system at the USDA-ARS Horticultural Research Laboratory in Ft. Pierce FL (Hall & Hentz, 2016). ACP colonies were maintained on six different host plant species: Bergera koenigii L. (curry tree), Citrus aurantifolia (Christm.) Swing (Mexican lime), Citrus macrophylla (alemow), Citrus maxima (Berm. F.) Merr. (pomelo), Citrus taiwanica Tan and Shim. (Nanshao Daidai sour orange), and Murraya paniculata (orange jasmine). All six species were known to support all life stages of ACP (Westbrook et al., 2011). Plants were grown from seed in steamed potting mix (Pro-Mix BX, Premier Horticulture, Inc., Quakertown, PA, USA) and fertilized weekly with water-soluble fertilizer mix (20N-10P-20K) (Peters Professional, The Scotts Company, Marysville, OH, USA). ACP were originally obtained from a USDA-ARS colony routinely maintained on C. macrophylla. Colonies were established by trimming plants to stimulate flush for oviposition and nymphal development. Two plants with flush were placed into each of several cages (BugDorm-2, BD2120F, MegaView Science Education Services Co., Ltd, Taichung, Taiwan), into which some adults were introduced and allowed to oviposit for 2 days and then removed. Resulting immature ACP were allowed to develop to the adult stage. The ACP colony was started on August 19, 2014. Adult ACP began emerging September 2, 2014. Adults for analysis were collected upon emergence from the colonies starting on September 5, 2014 through September 24, 2014. During the development time of the ACP the mean daily temperature in the greenhouse ranged from 25.8–28.2 °C, placed into labeled microcentrifuge tubes and held in a freezer until processed.

Preparation and digitization of specimens

Only female ACP were utilized for morphometric analysis to avoid gender-related differences confounding the data analysis as male traits are generally shorter in length than females (García-Pérez et al., 2013). Color morphs of ACP were classified as either blue/green or gray/brown based on the abdominal color of newly emerged adults (Wenninger & Hall, 2008). Many ACP possessed yellow ova or reproductive structures on the blue/green or gray/brown abdomens, yet this color was not considered in the analysis as it occurs on both male and females and is thought to be related to the age of the ACP (Wenninger & Hall, 2008). All ACP samples were measured after mounting the body, right forewing and tibia onto glass slides using 10% bovine serum albumin solution. Slide-mounted specimens were digitally photographed using ultra-small high-performance zoom lens (Model: VH-Z100R; Keyence, Osaka Japan) at 100X magnification and a free angle observation system (VH-S30K; Keyence). A 1 mm scale was used to calibrate body length measurements. Measurements of the digital images were made on a computer using ImageJ software (Version 1.47) (Schneider, Rasband & Eliceiri, 2012).

Traditional morphometrics

Wing and vein nomenclature are based on Hodkinson & White (1979). Measurements of twelve standard morphological traits were obtained from the forewing, tibia, and genal process (Fig. 1): (1) wing length measured as the distance between the proximal end of the C+Sc vein to the wing apex, (2) wing width measured as the distance between the apex of the anterior to the apex of the posterior forewing, (3) tibia length (T) measured from the apex of tibia where it connects to the femur to the most distal part of the femur where it connects to the tarsus, (4) length of genal process (GCL) measured from the base to the tip of the apex, (5) width of genal process (GCW) measured at the widest part of the genal process, (6) length of the M+Cu₁ vein, (7) length of the Cu₁ vein (8) length of the Cu_1b vein, (9) length of the Cu_1a vein, (10) length of the M vein, (11) length of the M₁₊₂ vein, and (12) length of the R_s vein. Vein measurements (6–12) were obtained by calculating the interlandmark distance used in geometric morphometrics (see below). The distances were calculated using the Pythagorean theorem.

Wing aspect ratio was calculated using the ratio of wing length to wing width. Wing loading was calculated by the ratio of tibia length to wing length. Tibia length is considered a correlate of the overall size of the organism (Kjaersgaard et al., 2015). Each measurement was repeated twice, and the average was used to reduce measurement error.

Geometric morphometrics

Wing shape was quantified based on a set of 11 homologous landmarks (x and y coordinates in a Cartesian space) consisting of intersections between wing veins or wing veins and the wing margin recorded from the right forewing of each specimen (Fig. 1) (Bookstein, 1991; Rohlf & Marcus, 1993). Landmarks were digitized using ImageJ software with the software plugin PointPicker (http://bigwww.epfl.ch/thevenaz/pointpicker/) (Thévenaz, 2013). Measurement error as a result of landmark placement was quantified for each specimen by making two sets of landmarks and using the average for comparisons of ACP reared on different host plants.

Data analysis

Linear measurements using traditional morphometrics were analyzed with the null hypothesis that there were no significant differences among measurements of ACP reared on different plants. All linear measurements were log-transformed for analysis. To compare each trait between host plants, data were analyzed using one-way analysis of variance (ANOVA) followed by mean separation with Tukey HSD test (P ≤ 0.05) contingent on a significant effect using R Studio (Racome, 2011) and R Statistical software (R Development Core Team, 2009). The following R packages were used for the data analysis: Plotrix (Lemon, 2006), Agricolae (de Mendiburu, 2015), Reshape (Wickham, 2007), Lattice (Sarkar, 2008), and Vegan (Oksanen et al., 2015). Ordination of the data using principal components analysis was used to determine which traits contributed the most variability. A second ordination technique, canonical variate analysis, was used to examine group differences and included a cross-validation confusion matrix. Group means were compared via multivariate analyses using Paleontological Statistics (PAST) v. 3.04 (Hammer, Harper & Ryan, 2001). A posteriori cross-validation and jackknife procedures were applied to determine the ability of canonical variate analysis to assign individuals to the correct group (Viscosi & Cardini, 2011). Dendrograms depicting squared Euclidean distances of size data between groups were plotted using unweighted pair group method with arithmetic mean (UPGMA) hierarchical cluster analysis (Sneath & Sokal, 1973) using PAST software (Hammer, Harper & Ryan, 2001). There were a total of 122 specimens used for the size and shape analysis (M. paniculata = 20, C. taiwanica = 20, C. macrophylla = 19, C. maxima = 23, C. aurantifolia = 21, B. koenigii = 19), however, some tibia and genal combs were lost and less samples were available to conduct the multivariate analysis of the linear measurements (M. paniculata = 13, C. taiwanica = 15, C. macrophylla = 9, C. maxima = 16, C. aurantifolia = 16, B. koenigii = 16).

Shape data (as Procrustes coordinates) were obtained using Procrustes superimposition, which removed size, position and orientation data and only extracted variation from shape (Dryden & Mardia, 1998; Rohlf & Slice, 1990). Geometric shape variation was analyzed based on Procrustes coordinates using multivariate techniques in MorphoJ v.1.04a software (Klingenberg, 2011). Shape variation from allometry (Gould, 1966; Klingenberg, 1996) was removed by using the residuals of the regressed log-transformed standard size measurement (the centroid) in shape analysis. The centroid is an estimator of size based on the square root of the sum of the distances of each landmark from the center of the landmark grouping (Parsons, Robinson & Hrbek, 2003). Ordination technique principal component analysis was undertaken to analyze patterns in the data. Discriminant function analysis was used for groups of two and canonical variate analysis for larger groups (Campbell & Atchley, 1981) with significant differences determined by permutations (10,000 rounds) to determine Mahalanobis distances between means. Cross validation of the discriminant functional analysis or canonical variate analysis was done using a confusion matrix that determines the accuracy of classifying the individuals to the proper group based on the discriminant function analysis (Viscosi & Cardini, 2011). Dendrograms depicting Procrustes distance of shape data between groups were plotted using UPGMA hierarchical cluster analysis (Sneath & Sokal, 1973) using PAST software (Hammer, Harper & Ryan, 2001). Differences between the average shape of ACP reared on all plants and the shape of ACP reared on a particular species were constructed using the MorphoJ software and provided a visual comparison of effects of each treatment on landmark values (Klingenberg, 2011).

Traditional morphometrics

Most ACP obtained from C. taiwanica were smaller than ACP obtained from other plant species. All traits measured differed significantly among host plant species except for wing width, and the lengths of the M+Cu₁ and Cu_1b veins (Table 1). Adult psyllids reared on C. aurantifolia were consistently larger compared to psyllids raised on C. taiwanica but similar to those reared on all other hosts tested. ACP reared on C. taiwanica were similar to those reared on B. koenigii except for a smaller genal comb length. The greatest range of size difference was observed with the genal comb length which was largest for ACP reared on M. paniculata, intermediate from B. koenigii and smallest from C. taiwanica. When all ACP reared from host plants were combined and only color morph considered, the only traits that differed were M+Cu₁ (t = −2.34, P = 0.02) and M (t = −1.99, P = 0.05) veins of the gray/brown colored ACP which were significantly smaller than the blue/green colored ACP (data not presented). There were also significant differences among ACP reared on different plants in terms of wing aspect ratio (wing length/wing width) (F = 2.51; df = 5.89; P < 0.04), but not wing loading (tibia length/wing length) (F = 0.80; df = 5.89; P = 0.55). ACP reared on C. taiwanica had significantly lower wing aspect ratio compared to ACP reared on C. maxima (Fig. 2). All other host plants produced ACP in the intermediary range.

The first two principal components accounted for 93.6% of the total variation observed among species (PC1 = 90.1%, PC2 = 3.5%). The greatest contributions to the first principal component were the M₁₊₂ (0.49) and R_s (0.49) veins with moderate correlation (0.39) from the length of the M vein (Table 2). In the second principal component, the Cu_1b vein was strongly correlated (0.78), while the length of the genal comb length was moderately correlated (0.46) (Table 2). Analysis by permutational multivariate analysis of variance (PERMANOVA) revealed significant differences between ACP reared on different host plant species (N_perm = 10,000; P < 0.001; F_5,80 = 5.15).

Visualization of the canonical variate analysis scatterplot did not demonstrate complete separation among plant species. However, ACP reared on C. taiwanica were completely separated from ACP reared on M. paniculata and minimal overlap with those reared on C. aurantifolia, C. macrophylla and C. maxima (Fig. 3). The first two canonical variables (CV1 and CV2) explained 86.4% of the total variance. Lengths of the Cu₁ vein, M vein, M₁₊₂ vein, and R_s vein wing were the highest contributors for the first canonical axes (CV1) (Table 3).

ACP reared on C. taiwanica exhibited larger Malahanobis distances compared to ACP reared on the other plants (Table 4). A posteriori classification summary assigned 61% of ACP into the correct host plant based on traditional morphometric measurements. A subsequent, jacknifed cross-validation resulted in correct assignment of 33% of ACP to the proper host plant using traditional morphometrics (Table 5). ACP reared on C. taiwanica had the lowest misclassification, 23 and 27% for posteriori and jacknife cross validation, respectively. The a posteriori classification error rate was reduced to 8% by randomly averaging the samples into groups of three, while the jacknifed cross-validation was not improved by averaging. A high error rate is expected in a sample of the same species where subtle intraspecific differences are examined. The host plant association is better able to differentiate ACP with respect to host plant species based on the average measurements of the group than to assign individuals to the correct host plant species on which they were reared. Cluster analysis using squared Euclidean showed ACP reared on C. maxima separate from the others in a single branch, while ACP reared on the remaining plants clustered into a second multilevel branch (Fig. 4).

Geometric morphometrics

Measurement error due to landmark placement was negligible as indicated by significantly less error for both the centroid and shape variation compared to the individual female ACP with respect to landmark placement (Table 6). As a result, not all ACP duplicating landmark placement was deemed necessary and more samples were collected instead of double measuring.

Separation of confidence ellipses from ordination of the individuals in morphospace defined by the first two canonical variates demonstrated significant differences among ACP reared on different host plants based on the mean of ACP (Fig. 5). While ACP in the two covariates did not separate into discrete clusters, mean ellipses of wing shape formed three distinct clusters. Wing shape means and confidence ellipses did overlap for C. taiwanica and B. koenigii indicating similarity. There were no differences among wing shapes for ACP reared on C. aurantifolia, C. maxima and C. macrophylla. ACP reared on M. paniculata differed in wing shape from ACP reared on all of the other host plants. Mahalanobis distances indicated that differences in shape from the geometric morphometric analysis were evident (Table 7). Based on confidence ellipses (Fig. 5), ACP reared from M. paniculata differed in shape from ACP reared on all other host plants. ACP reared from C. taiwanensis differed from those reared on C. macrophylla, C. maxima and C. aurantifolia but not B. koenigii.

Wireframe visualizations of the shape change of ACP wings reared on different host plant species provided visualization of differences in shape from each host compared to the overall average wing shape (Fig. 6). Wings from ACP reared on M. paniculata were clearly more narrow than average, whereas those from C. taiwanica and B. koenigii tended to be broader but were distinctly different from ACP reared on C. maxima, C. macrophylla, and C. aurantifolia.

The graphical visualization of the Procrustes distances in a dendrogram indicated that ACP reared from C. macrophylla, C. maxima and C. aurantifolia separated from ACP reared on B. koenigii, C. taiwanica, and M. paniculata (Fig. 7). The most distinct ACP were those reared from M. paniculata and C. macrophylla, which formed a single branch on both trees (Figs. 4 and 7). When color morph of ACP was considered, the shape of gray/brown ACP were not significantly different from blue/green ACP according to Hotelling’s T² test (P = 0.26; 10,000 permutations).