The Lolium grasses used in this study were Italian tetraploid L. multiflorum (cv. Grasslands Tama), diploid L. perenne (cv. Grasslands Samson) and diploid hybrid L. perenne ×L. multiflorum (cv. Grasslands Manawa). For clarity and brevity these grasses are referred to as ‘Italian’ grass, ‘perennial’ grass and ‘hybrid’ grass, respectively; all were endophyte free. Endophytes were excluded because there is now a wide range of differently acting novel endophyte strains in use in New Zealand pasture grasses. Also endophytes may have subtle effects on parasitoid behavior although this has not shown up in a recent field study (e.g., Goldson et al., 2015). Finally endophytes often do not perform very well in the tetraploid L. multiflorum varieties, so in general it seemed prudent to exclude the endophyte variable from the experiment. The grass types that were chosen represent the typical pasture types used in New Zealand farming since the release of the parasitoid.

All experimental work was conducted at ambient laboratory temperatures (23 ± 2°C) and 16:8 L:D photoperiod. Weevil adults were collected from mid-Canterbury ryegrass pastures using a modified leaf blowing machine (Goldson et al., 2000) between January 11, 2016 and January 22, 2016. They were then purged of egg and larval parasitoids by storing them for a minimum of 40 days and a maximum of 55 days with the remaining unparasitised population used for the experiment. The M. hyperodae pupae that emerged from these weevils were reared to obtain adult parasitoids for this study. Overall as detailed below, the experiment comprised three main treatments (grass types) with two subtreatments (plant positioning). These were replicated four times. There we also four grass-free control cages making 28 cages in total.

The experiment was established on March 17, 2016 using 305 mm × 205 mm × 130 mm translucent plastic cages with gauze lids. The four grass main treatment replicates were the minimum required to deal with pseudoreplication (Johnson et al., 2016). All cages were stocked with 23 L. bonariensis and two M. hyperodae. Each cage contained one of the three grass species treatments in the form of two 150 mm long bouquets with their moistened roots and associated soil in tightly sealed small polythene bags at the base of the plants. This resulted in at least 40 tillers per box (Supplementary Figure S1). Each grass treatment comprised two subtreatments, in separate cages, whereby the bouquets were positioned either horizontally or vertically; thus the cages were, respectively, positioned either on one end, or lying flat (Supplementary Figure S1). This different positioning of the plants was specifically to gain an initial indication of whether departure from vertical plant architecture had an effect on weevil parasitism in any of the three grass types. Gerard (2000) noted that weevils tend to leave the upright foliage in the presence of the parasitoid and Phillips (2002) suggested that plant orientation in a cage may influence parasitoid efficacy. There was also a control treatment comprising four cages (two horizontal and two vertical) containing 23 L. bonariensis, two M. hyperodae and two water-soaked dental wicks to maintain humidity. All paired treatments were placed randomly in the laboratory (Supplementary Figure S1). Parasitoids were removed from the cages after 48 h. Thereafter the weevils were maintained in the same ambient conditions for another 3 days until March 22, 2016 when they were frozen at -20°C prior to being dissected to assess parasitism rates (i.e., number of parasitized weevils per total number of weevils dissected).

In this study, the opportunity was taken to use the same Lolium grass types as were used throughout the 1990s during general research into M. hyperodae including the parasitoid’s mass-rearing for release (Goldson et al., 1993; McNeill et al., 1999, 2002). This allowed us to directly compare the results obtained from this experiment with both published and unpublished work conducted in the 1990s. Notably, while the exposure periods of the weevils in some of the comparator experiments were sometimes longer than 48 h this was of minor importance as Phillips et al. (1996) have shown that parasitoid ovipositional effort declines rapidly after the first 48 h.

To test for statistical significance between parasitism rates in the treatments and control, non-parametric complete random permutation tests (n cycles = 10000) were run for a one-way analysis of variance (ANOVA) via the package “lmPem” (Wheeler, 2010) and subsequently we tested the statistical significance via post hoc Tukey’s HSD pairwise permutation tests embedded in the package “stats” in R 3.2.1 (R Development Core Team, 2016). This approach implements the methods for permutation tests described by Kabacoff (2011).

The rate of parasitism of L. bonariensis by M. hyperodae in the presence of the experimental grass types and in the controls are presented in Supplementary Table S1 and the effects of grass type in Figure 1. In addition, Tables 1–3 show these results in the context of other studies yielding parasitism rates both in the 1990s and recently, in the presence of the Italian, diploid, and hybrid grasses, respectively. For purposes of comparison, all data have been normalized to be expressed as the effect of one parasitoid per population of weevils.

In the 1990s experiments, only horizontal treatments were used, therefore only the data from the horizontal subtreatments in this study were used for direct comparisons.

Descriptive analysis of the horizontal data in Tables 1–3 show that in the 1990s, the mean parasitism rate in rearing cages containing perennial grass was 74 ± 4% as opposed to 39 ± 5% in the current study. Similarly the parasitism rate associated with the hybrid grass was 73 ± 6% in the 1990s compared with 45 ± 5% in this study. Conversely, the 73 ± 8% parasitism found in the presence of the Italian grass is very similar to that in the 1990s (70 ± 5%).

The observed 42% decline in M. hyperodae parasitism observed in this cage study in the diploid and hybrid grasses (Tables 2 and 3), compared to 1990s laboratory data, conforms to the findings of recent field-based studies that have indicated a similar c. 50% decline in parasitism rates in diploid grasses since the 1990s (Goldson et al., 2015). That such a reduction in parasitism did not occur in the presence of tetraploid Italian grass either in this study (Table 1) or in the field (Goldson et al., 2015) suggesting that whatever factor(s) reduced parasitism rates in the perennial and hybrid grasses (Tables 2 and 3) did not occur in the presence of Italian grass (Table 1). It is also significant that the laboratory parasitism rates in the Italian grasses in this study were typical of those previously occurring in both the perennial and hybrid grass types in the 1990s (Tables 2 and 3).

Horizontal versus vertical positioning of grass bouquets within the cages resulted in no significant differences in the rates of parasitism of L. bonariensis across all of the grass treatments (Table 1; Figure 1). Such result indicates that, at least in the absence of soil or detritus, the orientation of plant material does not affect parasitism rates. This is contrary to Phillips (2002) suggestion that plant orientation in a cage may influence parasitoid efficacy. Further, the lower attack rates in the diploid and hybrid grasses were unlikely to have been based on the avoidance of parasitism by the weevils abandoning the foliage in the presence of the parasitoid as discussed by Gerard (2000). In the horizontal treatment, the grass leaves were broadly spread across the floor of the cage obviating the ability of L. bonariensis to drop off. The results of this study also point to the probable incorrectness of the contention of Goldson et al. (2015) who suggested that the higher levels of parasitism in the tetraploid L. multiflorum could have resulted from a difference in the architecture of the tetraploid versus diploid and hybrid perennial plants. This architecture hypothesis would seem to have been possible when considering vertical plants. However, the horizontal plant placement was a gross departure from the natural growth habit, yet there were no differences in the levels of parasitism between plant positioning subtreatments and the upright plants. This suggests that plant architecture was unlikely to be the underlying cause of the observed differences in parasitism between the grass types.

Finally, all treatments in the cages comprised grass bouquets that were bundled at their stem bases where the roots entered the polythene bags thereby providing limited scope for the weevils to ‘hide’ from the parasitoids. This is clearly different from the growth habit of the plants in the field.

The lack of any notable difference in L. bonariensis parasitism rates in the cages containing diploid and endophyte-free hybrid grasses is significant ecologically. At the time of the first parasitoid releases, and in order to expedite its establishment by using areas with plentiful weevils, the work was conducted in either pure hybrid pastures or pastures comprising a mix of diploid and hybrid ryegrass (Goldson et al., 1998; Barker and Addison, 2006) as the hybrid is known to be preferred to the perennial as a host plant of the weevil (Goldson, 1982). As a consequence, some of the early parasitism field data were collected from these hybrid sites. This study has shown no differences in parasitism rates in the perennial and the hybrid grasses (Gaynor and Hunt, 1982; Goldson et al., 1998; Barker and Addison, 2006). This eliminates the prospect of any bias having occurred through possibly higher measured parasitism rates occurring in the limited and very early sampling in the hybrid grasses. This is consistent with the observation that high parasitism rates were typically found in the diploid pasture that surrounded the original release sites during investigation into the parasitoid’s lateral dispersal from the release sites (e.g., Barker and Addison, 1997; McNeill and Goldson, unpublished data).

The attack rates measured in this caging study were very similar to those currently observed in the field (Goldson et al., 2015). This is surprising given the obvious environmental differences between the field and laboratory cages (e.g., no soil or detritus).

It can be hypothesized that the underlying mechanism for the observed general decline in parasitism rates since the 1990s (e.g., Goldson et al., 2014a,b) could have been based on the adoption of novel endophytes. However, none of the grasses in this laboratory study were infected with endophytes. Additionally, Goldson et al. (2015) in a 5-month summer field study, showed no significant field effects of endophyte on L. bonariensis parasitism rates in the mix of Lolium varieties and endophytes.

Contrary to the findings here, the data collected in the 1990s indicated no differences in parasitism rates, irrespective of grass type. At that time weevil parasitism rates in the hybrid and perennial grasses were comparable to those now only found in the tetraploid Italian plants (Tables 1–3).

Barker (1989) observed much higher rates of L. bonariensis feeding and oviposition in the leaves of tetraploid Italian grasses than in the perennial grasses. Related to this Phillips (2002) showed that weevil feeding, walking, grooming defecating, or mating predisposes it to higher levels of parasitoid attack and this therefore could be the reason for higher parasitism rates on the Italian grass. Conversely, Barker (1989) also showed that hybrid ryegrass (cv. Grasslands Manawa) is equally favored as a host by L. bonariensis as the Italian grass. In spite of this, the results here showed significantly less parasitism in the hybrid grasses than in the Italian grasses. This observation suggests that the intensity of weevil feeding and oviposition per se may not entirely be the reason for varied parasitism rates. Significantly, the growth habit of the hybrid grass is much closer to that of the diploids and neither of these ryegrass types support the same levels of leaf-feeding and oviposition as found in the Italian plants (Barker, 1989). It is also of interest that parasitism rates in grass-free control cages, while usually lower than in the cages with the grasses present, still showed substantial parasitism indicating that L. bonariensis remains susceptible to parasitism when not feeding or ovipositing.

The decline in parasitism in the hybrid and diploid grasses since that 1990s has not coincided with any sign of physiological resistance in the weevils. In spite of 1000s of weevils having been dissected by numerous workers since the introduction of the parasitoid, there has never been any observation of M. hyperodae early stages being encapsulated in L. bonariensis (e.g., Goldson et al., 2015).

In general, the results in this study support the contention of Goldson et al. (2015) that if selection pressure has led to an enhancement of some kind of parasitoid-avoiding behaviors amongst L. bonariensis, then such evolution would most likely to have occurred in the country’s extensive diploid pastures rather than in the rare tetraploid Italian L. multiflorum pastures (B.R. Belgrave, Grasslanz Technology Ltd., pers. comm.).