Female oviposition preference
Bruchid females deposited significantly decreasing number of eggs/female on seeds in the order of host (H), acceptable non-host (ANH) and non-host (NH) species groups (see definitions of these in the Materials and methods). Females laid 38.6 eggs on H species (22 SCA), 18.6 eggs on ANH species (55 SCA), and 13.5 eggs on NH species (63 SCA), and significantly more eggs on larger seeds (Table 1). Mean mass of H seeds was 340.8 mg, that of ANH 235.6 mg, and that of NH 70.3 mg (Table 1). All 66 plant species (62 leguminous and 4 outgroup species) received eggs, but females laid less than 15 eggs on 55%, less than 30 eggs on 35%, and between 30-45 eggs on 10% of plant species. Fig. 1 shows the distribution of eggs among leguminous tribes. Not surprisingly, A. obtectus females laid the highest number of eggs on members of the tribe Phaseoleae, where the main hosts are also found. Within Phaseoleae, soybeans (Glycine) were the least preferred (ESM Table 1). Comparable responses were noted to some species within Caesalpinieae, Genisteae, Robinieae, Cicereae and Fabeae (ESM Table 1). Although non-host species for the bean weevil, seeds of species such as Gleditsia delavayi Franch. (Caesalpinieae), Laburnum alpinum (Mill.) Bercht. & J. Presl, L. anagyroides Medik. (Genisteae), and Robinia viscosa Vent. (Robinieae) received high numbers of eggs. C. arietinum (Cicereae) and Vicia faba L. (Fabeae) are known as occasional hosts. The number of eggs laid/female on them fell into the medium and high categories, respectively. A mean number of less than 10 eggs/female were laid on Vicia tenuifolia Roth, Robinia pseudoacacia L. and Amorpha fruticosa L., and on some other members of the Fabeae tribe (ESM Table 1).
Egg-laying on artificial beans. Females accepted the artificial seeds incorporated with seed coat as an oviposition substrate and laid comparable number of eggs to the control (Table 2).
Mortality of larvae. The major stages of larval performance were: (a) entering the seeds that was constrained by the thickness of seed coat, resulting in L1 mortality outside seeds, and (b) within-seed mortality of various developmental stages. Both were modulated by the intact or drilled status of seeds. The seed coats of host (H) and acceptable non-host (ANH) species were significantly thinner (Table 1). L1 mortality outside intact seeds was different among plant species groups (Table 1). It was significantly lower for the H (24.75%) and ANH groups (93.3%), in comparison with the NH group (100%, all are medians). For the two H and 16 ANH species Table 3 provides data of seed coat thickness, whereas ESM Tables 2 and 3 give similar information for NH, and SCA of H and ANH groups. Remarkably, seed coat constitutes a barrier even on the primary host, beans (Ph. vulgaris), for the bean weevil. Contrary to the very low larval mortality inside seeds, there was substantial mortality outside intact seeds of bean cultivars (27.3%, 13.6-53.3, median, lower and upper quartiles, N=21), but only 4.4% (0-11.4) for drilled seeds. Similar values can be given for other legume species from which adults emerged, with the difference that the upper level of mortality usually reached 100% with intact seeds, with some exceptions such as Vigna unguiculata (L.) Walp. and V. angularis (Willd.) Ohwi & H. Ohashi, where seed coats were extremely thin (Tables 3 and 4). A remarkable case is the runner beans (Ph. coccineus) with 0% (intact seed) and 2.2% (drilled seed) larval mortality outside seeds.
Mortalities of various developmental stages inside intact seeds were substantially different from those of drilled ones, however, the critical event for development in the seeds is invariably the survival of first instar larva. Tables 1 and 4 provide results for host (H) and acceptable non-host (ANH) species, and further data are available in ESM Tables 2 and 3 for bean, pea and soybean varieties, as well as for non-host (NH) species, showing differences among host-types. Mortality inside intact NH seeds was practically nil due to the inability of L1s to penetrate the seed coat, whereas for drilled seeds it was 75% (44-93%, median and quartiles). Many L1s first entered the seeds, then left them again, and died outside of starvation or by toxins taken up from the cotyledon. Seed testa frequently bore several shallow pits, where L1 attempted to bore in (e.g. all Gleditsia japonica Miq. seeds, both intact and drilled, had such traces). It is worth remarking that ca. 10% of L1 entered the intact G. delavayi seeds, but inside the seeds many died by various manners: on the surface of the cotyledon, or after burrowing in the radicle by various lengths. Although it is well documented that additional A. obtectus larvae may enter through the hole made by a pioneer larva , the number of L1-made and artificial holes was 1.5:1 on the most preferred bean seeds (Ph. coccineus), i.e., many L1 larvae did not use the pre-prepared holes on this host. In some instances (e.g., Caragana or Onobrychis genera) L1s entered the seeds through the hilum. First instar larvae entering Gleditsia seeds through an artificial hole made at the embryo area invariably died within the embryo. In cases where the cotyledons were soft (several Glycine cultivars/accessions and Caragana), larvae made longer tunnels before dying.
Development in artificial seeds. This experiment proved that, besides being a physical barrier, seed coat also inhibited larval development at the lowest concentration incorporated into artificial seeds (Table 2). To the contrary, controls (without any seed coat content) were fully suitable for development.
Adult emergence. Significantly more adults emerged from H species/cultivars than either from ANH species/cultivars or NH species, and whether intact or drilled (Table 1). Of the 62 legume species A. obtectus larvae developed into adults in 18 (29%) species in four tribes, if the seed coat was drilled, and from only nine species, if they had intact seed coat (Table 4). However, the picture varied considerably concerning SCA (ESM Table 3). Although adult emergence in Ph. vulgaris was generally high, at the cultivar level it ranged between 51 and 100%. In Ph. coccineus all larvae developed into adults. Whereas cowpea (Vigna) species supported larval development to adults, white lupin (Lupinus), soybean (Glycine) varieties and hyacinth bean (Lablab) did so only sporadically, and especially in drilled seeds. This is also paralleled by the length of developmental time needed until adult emergence (Table 4). In some cases (G. max (L.) Merr., V. angularis and P. sativum L.) it was two to three times longer in comparison with beans. Of the 27 cultivars of garden peas, adults emerged from 24 (88.9%), however, only from 13 of these, if the testa was intact. Similar values occurred for 17 G. max cultivars/accessions: adult emerged from six (35.3%), but only from four of these with an intact seed coat. A surprising feature is the asymmetric distribution of adult emergence within the Fabeae tribe. Whereas there was adult emergence from six Lathyrus species (one, L. sativus L., from intact, the rest from drilled seeds), there was only one such case among vetches: the faba bean (Vicia faba). Both intact and drilled seeds yielded adults from faba beans, although it had a relatively thick testa. Although larval development proceeded in some NH reaching as far as L3, no adult emerged from these (ESM Table 2).
In the 18 adult-yielding legume species five (28%) produced malformations and some adults could not leave the seed in 10 of these. Typical malformation was a substantial decrease in elytra width and length: the elytra became shorter and triangular in shape. There were 0.1% malformed adults in beans, 33.0% in peas, 31.6% in L. tuberosus L., 6.7% in L. sativus, and 0.7% in V. faba.
Female preference versus larval performance
Correlations. ESM Table 5 provides the most important nonparametric correlation coefficients referring to the overall relationship between plant traits and insect responses. Accordingly, only in the ANH group there was a significant positive relationship between preference and performance, i.e., between the number of eggs laid/female and the adult emergence (Kendall’s τ=0.3088, intact seeds, N=55). The correlations between seed mass and the number of eggs laid were extremely low in all three plant groups, but were significant with ANH and NH species. The thicker the seed coat, the higher was the first instar larval mortality outside on ANH and NH seeds, but seed coat thickness did not affect L1 larval mortality on H seeds.
Results of logistic regressions. The logit-regression provided evidence that L1 mortality outside seeds was due to different seed coat thicknesses. Significantly (15:1) higher number of cases indicated <50% L1 mortality outside seeds, if seed coat was thin (0.08 ± 0.001 mm, mean±SE, N=62), in comparison with thick seed coat (0.15 ± 0.005 mm, N=78; Wald test: 8.2, df=1, p=0.0043; log-likelihood: -48.4; goodness of fit χ2: 17.9, df=1, p<0.001). Approaching the same hypothesis from another angle, i.e., if seed coat was assigned as ’penetrable’ or ‘impenetrable’, significantly higher number of L1 entered seeds with ‘penetrable’ seed coat (Wald test: 27.3, df=1, p<0.001; log-likelihood: -81.4; χ2= 31.1, df=1, p<0.001). Here only those cases were taken into account where L1 larvae entered a seed, then died immediately after this. This result is interesting, because ‘penetrable’ seed coat thickness was 0.0998 ± 0.004 mm (N=64), whereas ‘impenetrable’ showed 0.119 ± 0.005 mm (N=76), a mere 0.02 mm difference.
As for the second hypothesis, larval performance and adult emergence did depend on the ‘quality’ or ‘suitability’ of cotyledons of seeds. Significantly higher number of cases showed <30% larval mortality inside seeds, if cotyledon was ‘suitable’ (4.7%, 2.2-43.2, N=124) in comparison with ‘unsuitable’ cotyledon (61.3%, 12.9-90.7, medians and quartiles, N=88; Wald test: 27.6, df=1, p<0.001; log-likelihood= -131.5; χ2=29.8, df=1, p<0.001). Adult emergence from ‘suitable’ seeds was 37.8% (10-84.1, median and quartiles, N=123) vs. from ’unsuitable’ cotyledons: 0% (N=89). Logistic regression for adult emergence could not be performed, because one cell of the χ2 table contained zero. Nevertheless effect sizes could be computed.
Effect sizes and risk analyses. The first phase of the preference-performance relationship refers to the larval ability to overcome seed coat thickness. Odds ratio (OR) provided 19 times larger chance for L1 larvae to have <50% mortality, and risk difference (RD) indicated 45% higher survival for larvae, if they happened to bore in a seed with seed coat thickness of <0.1 mm, in comparison with seeds having thicker testa (ESM Table 6). However, the regression coefficient (φ2) explained only 10% of variance of seed coat effect, referring to other important factors affecting larval entry. On the other hand, when facing penetrable/impenetrable seed coat (results are not shown in table), L1 larvae had three times higher risk to die with ‘impenetrable’ seed coat (RR = 3.01±1.3, mean±SE, CI95=1.9 & 4.8), whereas chances for larvae to enter a seed with penetrable testa was 7.5 times higher (OR=7.5 ± 1.5, CI95= 3.5 & 16.1), even if they died after the first bites from the cotyledon. Here, the regression coefficient explained a relatively high level (21%) of the variance.
The second phase of the preference-performance relationship is manifested in the adult progeny production as related to the number of eggs laid/female on legume species. The risk of >30% mortality inside seeds for larvae increased ca. 2 times (RR= 1.95 ± 1.13, CI95= 1.52 & 2.49) in cotyledons unsuitable for reaching later developmental stages. The odds for such an outcome was high (OR= 4.85 ± 1.28, CI95= 2.96 & 7.94, results are not shown in table). The chance for reaching adulthood in seeds from which >10% adults emerged was ca. 4 times higher (RR in ESM Table 7), than in cotyledons allowing only 1-2 weevils to successfully complete their development. The odds for adulthood in suitable seeds were extremely high (OR= 523.42 ± 1.23, CI95= 348.74 & 785.60) due to the asymmetry caused by the NH seeds. The regression coefficient (φ2) explained a high portion (54%) of the variance (ESM Table 7).
As expected, there was an interaction between penetrable/impenetrable and intact/drilled states of seed coats: the joint effect (OR11=0.6718) was larger than the multiplied value (0.4820) of their respective effects (OR10=0.7179 and OR01=0.6714). The joint effect of both variables is 1.4 times higher than the combined effect of the variable acting separately.