plant defence syndromes
The observed variation of plant traits was higher across species than within species between leaves of different age. Nevertheless, leaf age nested within species had a significant effect on variation of the highly relevant plant traits water content, SLA, myrosinase activity and abaxial and adaxial trichome densities (see Tables S1 and S2 in Supporting Information). Correlation coefficients for mean values of young or old leaf subsets were very similar (data not shown). There were significant positive correlations between protein and nitrogen content as well as between adaxial and abaxial trichome densities, whereas nitrogen and water content were negatively correlated. Several further coefficients were at least r > 0·7 or higher. Thus, two major groups of correlations with high coefficients (r > 0·7) were used for principal component analysis using mean values of young and old leaf data, respectively. Water, soluble protein and nitrogen content of fresh weight and SLA were combined to the factor ‘nutrition’ (PC-A, n = 14; Eigen value: 3·48; communalities – water: 0·868; protein: 0·899; nitrogen: 0·959; SLA: 0·751). An 87% of original variability between species and leaf ages of the four variables was incorporated in the factor values based on regression (component loadings – water: –0·931; protein: 0·948; nitrogen: 0·979; SLA: –0·867). Trypsin inhibitor activities as digestibility reducers and trichome densities of both leaf sides were combined to obtain a second factor (PC-B, n = 14; Eigen value: 2·69; communalities – trypsin inhibitor activity: 0·914; abaxial trichome density: 0·893; adaxial trichome density: 0·88). A 90% of original variability between species and leaf ages of the three variables was incorporated in the factor values based on regression (component loadings – trypsin inhibitor activity: 0·956; abaxial trichome density: 0·945; adaxial trichome density: 0·938). For reasons of clustering efficiency and to avoid incorporating redundant information, these factor values were used for further analyses.
The hierarchical cluster analysis using Ward's method revealed three clusters (Fig. 1). In the three clusters plant species are represented which have combinations of: (i) high nitrogen content, intermediate C : N ratio together with high glucosinolate concentrations, (ii) low nitrogen content, high C : N ratio and low glucosinolate concentrations, (iii) high nitrogen content, low C : N ratio together with high densities of trichomes and high proteinase inhibitor activities. Unifactorial anova on single variables revealed significant differences for all plant traits except for water content, SLA and myrosinase activity concentration between the clusters (Table S3). In a subsequent discriminant function analysis all traits contributed substantially to the separation of clusters, except for myrosinase activity concentration (Table 2). According to these functions plant clusters were effectively separated with a minimum of 98% correct classification.
Figure 1. Plant clustering according to the two discriminant functions; C1, centroid of cluster 1; C2, centroid of cluster 2; C3, centroid of cluster 3; three letter abbreviations refer to plant species means for young and old leaves (see Table 1).
Download figure to PowerPoint
Table 2. Coefficients of discriminant function analysis of clustering results using mean plant traits of young and old leaves of seven species of Brassicaceae
|C : N ratio||–0·696||–0·077|
|Myrosinase activity concentration||0·223||0·648|
|Trichomes and digestibility reducers (PC-B)||1·391||–0·214|
|Eigenwert of function||19·062||2·589|
|Explained variability (%)||88||12|
insect performance and preference
Means of insect developmental parameters varied more distinctly between host plant species than between leaf age classes. Nevertheless, leaf age nested within species contributed significantly to the variation of almost all insect traits (Tables S4 and S5). Particularly, mortality rates of male larvae were on average higher when insects were reared on young leaves compared to old leaves for most species. Larval developmental times of females were a little longer, but their pupal times were about 1 day shorter than those of males. Females were about 30% heavier as eonymphs and as adults compared to males (Table S4). However, male and female mean parameters were highly correlated (Pearson-product moment correlations; mortality rates: r = 0·94, n = 14, P < 0·001; developmental times: r = 0·93, n = 11, P < 0·001; adult masses: r = 0·94, n = 11, P < 0·001).
Correlation coefficients of mean values for insects reared on young or old leaves were very similar (data not shown). Developmental parameters of larval stages determined the adult performance parameters (Table S4). Larval mortality rates were highly correlated with total mortality rates (n = 14; males: r = 0·99, P < 0·001; mixture of sexes: r = 0·99, P < 0·001). Larval developmental times were strongly correlated with total developmental times (males: n = 488, r = 0·86, P < 0·001; females: n = 196, r = 0·90, P < 0·001) and eonymph mass determined adult mass (males: n = 487, r = 0·86, P < 0·001; females: n = 196, r = 0·93, P < 0·001).
To define clusters, again only a subset of mean values of insect parameters was used: overall mortality rate, female adult masses, egg numbers of naïve females, as well as fat content and glucosinolate concentration of males. The hierarchical cluster analysis using Ward's method revealed three clusters (Fig. 2). Cluster 1 was formed from insects with high adult masses and high glucosinolate concentrations and nitrogen contents in males. In contrast cluster 2 included insects with low adult masses and low glucosinolate concentrations and nitrogen contents in males. Insects of both clusters showed low mortality and developmental times. In cluster 3 insects with high mortality rates were grouped. Surviving insects showed prolonged developmental times and lower adult masses compared to insects from clusters 1 and 2. Furthermore, males contained remarkably more fat per mass than in the other two clusters. Unifactorial anova revealed significant differences depending on the clusters for mortality rates, as well as for female developmental times and adult masses (Table S6). T-tests comparing all other traits between clusters 1 and 2 revealed differences for most other parameters except for male fat and non-fat dry matter content and male developmental times. A subsequent discriminant function analysis revealed that all parameters contributed meaningfully to the separation of clusters (Table 3). According to these functions, insect clusters were effectively separated with 100% correct classification of rearing groups into the three clusters.
Figure 2. Insect clustering according to the two discriminant functions; C1, centroid of cluster 1; C2, centroid of cluster 2; C3, centroid of cluster 3; three letter abbreviations refer to plant species means for young and old leaves (see Table 1); 1three or 2two missing values were substituted with the parameter mean (n = 11/12) for female mass in two groups (Boo and Boy) and fat content and glucosinolate concentration of males in three groups (Boo, Boy and Lry).
Download figure to PowerPoint
Table 3. Coefficients of discriminant function analysis of clustering results using mean parameters of insects reared on young and old leaves of seven species of Brassicaceae
|Overall mortality rates||1·444||–0·241|
|Egg numbers of naïve females||–1·110||0·011|
|Female adult mass||0·139||0·861|
|Male fat content||1·981||0·320|
|Male glucosinolate concentration||1·083||0·240|
|Eigenwert of function||25·466||5·443|
|Explained variability (%)||82·4||17·6|
Plant species had a stronger influence on oviposition preference of naïve females than leaf age, that is, 60% of variance was explained by the first and only 6% by the latter. The covariate female mass was not significant as a predictor (n = 238, ancova: d.f. = 8, F = 48·1, η2 = 0·63; species effect d.f. = 6, F = 57·5, P < 0·001, = 0·60; leaf age effect d.f. = 1, F = 13·5, P < 0·001, = 0·06; covariate female mass d.f. = 1, F = 0·2, P = 0·669, = 0·001; variances were not homogenous).
For experienced females, the total numbers of eggs that were laid in seven consecutive bioassays were significantly influenced by the species on which larvae were raised, but not by female mass (n = 123, ancova: d.f. = 6, F = 6·1, η2 = 0·24; rearing species effect d.f. = 5, F = 6·3, P < 0·001, = 0·21; covariate female mass d.f. = 1, F = 0·1, P = 0·752, = 0·001; variances were not homogenous). The numbers of accepted host plants, a measure of host selectivity, were influenced by the species on which larvae were raised, and even five times stronger by the total numbers of eggs laid within the bioassays, but not by female mass (n = 123, ancova: d.f. = 7, F = 30·7, η2 = 0·65; rearing species effect d.f. = 5, F = 2·8, P = 0·020, = 0·11; covariate total numbers of eggs d.f. = 1, F = 124·6, P < 0·001, = 0·52; covariate female mass d.f. = 1, F = 2·8, P = 0·098, = 0·024; variances were homogenous).
multiple regression analyses on plant and insect traits
Conducting probit analyses on total mortality rates of insects revealed that the best results were obtained when the plant factor PC-B was replaced by its individual plant traits, trichome densities and proteinase inhibitors, in a model also including plant C : N ratio, glucosinolate, myrosinase activity concentrations, and the factor ‘nutrition’ (Table 4). The coefficients for adaxial trichome density, trypsin inhibitor activity and the factor ‘nutrition’ were positive. These traits should thus increase the probability of death of an insect. C : N ratio, glucosinolate concentration and abaxial trichome density coefficients were negative and therefore these traits likely lead to a decrease of mortality probability. Myrosinase activity concentration had no significant influence on total mortality rates.
Table 4. Probit analysis of mortality rates using the sum of dead larvae and pupae of male insects only and in mixture with females in rearing groups depending on plant traits (Z-scores); whole model predictability male χ2 = 12·99, d.f. = 6, P = 0·043; both sexes χ2 = 9·01, d.f. = 6, P = 0·173; Test for parallelism: χ2 = 13·778, d.f. = 1, P < 0·001
| C : N ratio||–0·40||0·13||–3·01||0·003|
| Glucosinolate concentration||–1·07||0·30||–3·60||< 0·001|
| Myrosinase activity concentration||0·10||0·12||0·83||0·406|
| Adaxial trichome density||2·43||0·82||2·96||0·003|
| Abaxial trichome density||–1·81||0·54||–3·37||0·001|
| Trypsin inhibitor activity||0·50||0·34||1·45||0·147|
| ‘Nutrition’||1·06||0·23||4·58||< 0·001|
|(b) Both sexes|
| C : N ratio||–0·50||0·09||–5·41||< 0·001|
| Glucosinolate concentration||–0·76||0·13||–6·01||< 0·001|
| Myrosinase activity concentration||0·02||0·07||0·27||0·787|
| Adaxial trichome density||1·48||0·30||4·90||< 0·001|
| Abaxial trichome density||–1·63||0·22||–7·28||< 0·001|
| Trypsin inhibitor activity||0·75||0·21||3·61||< 0·001|
| ‘Nutrition’||0·52||0·11||4·81||< 0·001|
Linear regression models for five important insect parameters were fitted on mean plant traits (Table 5). Longer developmental times could be attributed mostly to higher trichome and proteinase inhibitor levels and a little less to lower C : N ratios of plants for both sexes. Higher myrosinase activity was found to correlate with developmental times at least for males. Adult masses of insects regressed strongly on C : N ratio and also slightly on the factor ‘nutrition’. Whereas for males only the coefficient for plant C : N ratio was significantly different from zero, for females both coefficients revealed significant influences. The multiple linear regression models displayed a highly negative influence of myrosinase activity concentration and a slightly positive one of trichomes and proteinase inhibitors on fat accumulation of male adults. The glucosinolate concentration of adult males was strongly influenced by plant C : N ratio, and a little less by trichomes and proteinase inhibitors as well as ‘nutrition’. Egg numbers of naïve females regressed negatively on myrosinase activity and the factors trichomes and proteinase inhibitors as well as ‘nutrition’. Glucosinolate concentration of the plant tissue never had a significant influence on insect development or egg deposition.
Table 5. Linear regression for developmental parameters depending on plant traits (Z-scores).
|Insect parameter||Model||R2||anova||Standardized coefficients||‘Nutrition’|
|F||d f||P||C : N ratio ||Myrosinase||Trichomes and digest. reducers|
|Total developmental time||Complete||0·98||41·41||5||< 0·001||–0·39**||0·42**||0·67**|| |
| ||Reduced||0·97||87·34||3||< 0·001||–0·38**||0·43**||0·69***|| |
|Adult mass||Complete||0·73||2·63||5||0·156||0·76 (P = 0·05)|| || ||0·27 (P = 0·44)|
| ||Reduced||0·71||9·98||2||0·007||0·80**|| || ||0·36 (P = 0·10)|
|Fat content of adults||Complete||0·85||5·74||5||0·039|| ||–0·73*||0·36 (P = 0·15)|| |
| ||Reduced||0·82||18·72||2||0·001|| ||–0·76**||0·35 (P = 0·05)|| |
|Glucosinolate concentration of adults||Complete||0·89||8·20||5||0·019||0·82**|| ||0·49*||0·53 (P = 0·05)|
| ||Reduced||0·86||14·02||3||0·002||0·81**|| ||0·50*||0·69**|
|Total developmental time||Complete||0·93||14·79||5||0·003||–0·38*||0·29 (P = 0·08)||0·68**|| |
| ||Reduced||0·91||27·69||3||< 0·001||–0·41*||0·26 (P = 0·07)||0·67**|| |
|Adult mass||Complete||0·89||8·18||5||0·019||0·77**|| || ||0·48 (P = 0·07)|
| ||Reduced||0·87||26·67||2||< 0·001||0·80***|| || ||0·55**|
|Egg numbers of naïve females||Complete||0·68||3·32||5||0·065|| ||–0·57*||–0·43 (P = 0·16)||–0·49 (P = 0·14)|
| ||Reduced||0·62||5·44||3||0·018|| ||–0·57*||–0·53*||–0·35 (P = 0·10)|