Female fecundity variation affects reproducibility of experiments on host plant preference and acceptance in a phytophagous insect

Abstract

Reproducibility is a scientific cornerstone. Many recent studies, however, describe a reproducibility crisis and call for assessments of reproducibility across scientific domains. Here, we explore the reproducibility of a classic ecological experiment—that of assessing female host plant preference and acceptance in phytophagous insects, a group in which host specialization is a key driver of diversification. We exposed multiple cohorts of Pieris napi butterflies from the same population to traditional host acceptance and preference tests on three Brassicaceae host species. Whereas the host plant rank order was highly reproducible, the propensity to oviposit on low-ranked hosts varied significantly even among cohorts exposed to similar conditions. Much variation could be attributed to among-cohort variation in female fecundity, a trait strongly correlated both to female size and to the size of the nuptial gift a female receives during mating. Small males provide small spermatophores, and in our experiment small females that mated with small males had a disproportionally low propensity to oviposit on low-ranked hosts. Hence, our results provide empirical support to the theoretical prediction that female host utilization is strongly affected by non-genetic, environmental variation, and that such variation can affect the reproducibility of ecological experiments even under seemingly identical conditions.

1. Introduction

Recent studies raise concerns that the pressure to ‘publish or perish’ forces premature publication of non-replicated, and sometimes non-reproducible, scientific discoveries [1–6]. Often, difficulties in reproducing previous results can be linked to early publication of studies based on moderate samples and large effect sizes [7,8]. In ecology, findings are often at best ‘quasi-replicated’ [7] where similar experiments are made on different species under different conditions, rather than on the same genetic material under similar conditions.

Here, we address the issue of reproducibility and result consistency in the study of host plant utilization by phytophagous insects. Such studies are the foundation for inferences of ecological specialization, host plant-driven diversification and the birth and development of coevolutionary theory [9–15]. Typically, female decision-making determines the larval feeding site and the range of host plants that females lay eggs upon is often narrower than the larval feeding range [16–18], implying that host specialization is largely determined by female preference evolution. Indeed, many studies report a strong impact of local evolution on female host preference [19–25].

In order to qualify as a driver of evolutionary specialization, any identified inter- or intra-population variation in host utilization needs to be reproducible and reflect a genetic difference in host plant preference/acceptance within or among populations. Therefore, it is important to identify how, and to what extent female host plant choice is impacted by environmental effects or life-history status. If such effects are found, the importance of host utilization as a driver of evolutionary specialization decreases. In nature, female habitat selection can precede host plant choice [26,27], so that the ultimate female host choice is relevant only within a particular habitat. Likewise, the prior experience [28,29], fecundity and egg load [30–35] of individual females may all affect the female–host interactions.

The choice of bioassay also can affect the determination of a female's, population's or species' host range, because the female acceptance, here defined as her propensity to oviposit on a particular host when this is the only available alternative, is typically broader than her preference, here defined as the female propensity to oviposit on a host in relation to her propensity to oviposit on other available hosts (cf. e.g. [36,37] for reviews of alternative definitions). Female acceptance is typically measured in no-choice-trials, which come with the risk that the ecological importance of a host can be overemphasized if females oviposit on this plant only when it is the only available host. Conversely, multiple-choice experiments measure female preference, and might underestimate the fundamental host plant niche, if females strongly tailor egg-laying toward the most preferred host (cf. [16]). Here, we study the relationship between female host preference and acceptance, and investigate whether inter-cohort variation could have a significant effect upon the quantification of these variables. Such a finding would have important implications for the reproducibility of studies of evolutionary specialization.

In ecology in general [7], and studies of female oviposition behaviour in particular, minor attention has traditionally been paid to the reproducibility of experiments. Often, the host use of a population or species is determined by a single experiment or field observation, and different populations can sometimes be scored for host preference or acceptance at different times, and, hence, under different experimental circumstances. We approach this issue by evaluating the reproducibility of host plant preference and acceptance in similar experimental set-ups on multiple cohorts of a population of the butterfly Pieris napi with the goal of identifying robust and variable aspects of inferred host use dynamics.

Pieris butterflies (Pieridae) are suitable targets for these kinds of experiment, because a previous study on P. napi found variation in host plant acceptance among females of the same Swedish population of P. napi [38]. Whereas all females readily laid eggs on Barbarea vulgaris, about half of the females tested refused certain hosts in long-term no-choice trials (e.g. Armoracia rusticana or Brassica napus) [38]. Investigating the basis of this variation is of interest as a route to uncovering the genetic basis of host plant choice. However, before spending extensive functional genomic resources, we investigated whether the dramatic variation in acceptance of these lower-ranked hosts was reproduced across cohorts of the same stock population, which offers an opportunity to study the reproducibility of egg-laying experiments and evaluate how variation in female status may affect her host utilization.

We first performed two preference experiments, and multiple acceptance experiments under similar conditions to assess the variability and reproducibility of results. We then assessed to what extent variability in host use within and among cohorts was driven by variation in female fecundity. In P. napi, as in many holometabolous insects [39], fecundity is expected to depend on the resources a female is able to obtain as a larva [40] and thus increases with adult size. Furthermore, P. napi females are provided with a nutrient-rich spermatophore during mating, which functions as a male-derived nuptial gift that can be converted into egg tissue [34,41,42]. The spermatophore size produced by virgin males is tightly correlated to male body size [43], and female fecundity variation could thus be linked to both her own size, and the size of her mate(s). Hence, we (i) experimentally assessed the relationship between female host acceptance and host preference in a replicated experiment, (ii) determined the reproducibility of host acceptance trials across multiple cohorts, and (iii) investigated how the detected variability in host acceptance related to female size and fecundity variation within and among cohorts.

2. Methods

(a). Study system

The green-veined white butterfly P. napi uses numerous crucifers (Brassicaceae) as hosts [44]. We compared female host use of the three host plants B. vulgaris (winter cress), B. napus (rapeseed) and A. rusticana (horseradish). All these plants are highly suitable for larval growth [38]. Across experiments, we tried to minimize effects of phenological state or genetic variation in the host plants, by continuously growing new plants of the same seed/root stock, and picking leaves of the same physiological stature, size and freshness.

About 200 larvae were collected from wild plants of B. vulgaris, Alliaria petiolata and Cardamine pratensis at two sites (approx. 20 km apart) in Skåne southern Sweden (Kullaberg; 56°18′ N, 12°27′ E and Vejbystrand; 56°18′ N, 12°46′ E) during summer/autumn 2013 and 2014. Individuals were brought to the laboratory and reared, in a common laboratory population, between one and five generations on A. petiolata or A. rusticana before being used in the experiment. Extensive unpublished data from two master's theses [45,46] strongly indicate that larval host has no influence on female preference in P. napi.

(b). Experimental cohorts and rearing conditions

We report data on no-choice host acceptance experiments, performed during 2015, using individuals from a total of 10 different cohorts (a–j) that all descended from the Skåne population but differed in certain aspects (electronic supplementary material, table S1). Cohorts a–e were reared and tested at the Department of Zoology at Stockholm University, and cohorts f–j were reared and tested at the Department of Plant Ecology and Evolution at Uppsala University. Cohorts a, c and d had undergone a pupal winter diapause, and spent 4–12 months at 2°C. Cohorts b, and e–j had been induced to develop directly, with only 7–10 days in the pupal stage. Cohorts a–e (the Stockholm cohorts) met three different host plant species in a randomized sequential order (electronic supplementary material, table S2), whereas females of cohorts f–i were allowed to first meet B. napus and then B. vulgaris (cohort g was also presented to the inferior host Erysimum cheiranthoides—data not shown). The acceptance by these females for B. napus and B. vulgaris could thus be compared across nine (a–i) cohorts. Females of cohort j were only presented to B. napus throughout their entire lifespan, and were presented with a new host individual every third day. Females of cohorts a–e and j were dissected post mortem, and mature eggs left in their abdomens were counted. The discal cell length (measured on the dorsal side of the right forewing) was measured for all females of cohorts a–e. The pupal weight of females in cohort a and b was taken 2 days after pupation and correlated with the discal cell length to certify that the wing trait was indeed useful as a measure of female size (r = 0.80, n = 78, p < 0.001).

Females of cohort a and b were tested also for host plant preference in choice trials. These trials took place during the first 2 days after mating in flight cages measuring 0.8 × 0.8 × 0.5 m. In these cages, females could choose among a fresh cutting of each of the three alternative host plants (A. rusticana, B. vulgaris and B. napus). The cuttings were placed in water in bottles and were positioned in a triangle and close to the roof of each cage. All females had access to a nectar plant (Kalanchoe sp.) on which we sprayed sugar water (20%). Eggs were counted at the end of day 2, when females were transferred to the no-choice treatment (see below).

Butterfly rearing and mating took place under similar conditions in the two laboratories. Diapausing pupae were kept at 2°C and transferred prior to the experiment to 23 ± 2°C for hatching. Eclosing males were transferred to mating cages. In Stockholm, these measured 0.8 × 0.8 × 0.5 m and were lit with a 400 W HQIL lamp placed 0.5 m above the cage roof. In Uppsala, cages measured 0.8 × 0.7 × 0.4 m, and were placed under an Agrilight AL2700 lamp (400 W) in a glasshouse. Lights were on between 9.00 and 17.00. Females were kept at 8°C until a sufficient number of females had eclosed. Typically, eclosion was synchronized, and only very few females waited longer than 2 days under cold conditions. Females were then released in the mating cage and mating pairs were isolated and marked individually. Males and females were allowed to mate only once.

The day after mating (3 days after mating for cohorts a and b, see above) females were confined in individual 1 l plastic jars provided with a leaf of one of the three host plants. The host plants were grown from seeds (B. napus, B. vulgaris) or were propagated from root chunks (A. rusticana). Brassica napus seeds were commercially available, and kindly donated by Olssons Frö AB (Helsingborg, Sweden), whereas the B. vulgaris seeds were collected from a patch close to Stockholm University (59.3665°N, 18.0750°E; approx. 50 seed families). Armoracia rusticana were purchased at the local supermarket (approx. 50 roots), divided, planted and later propagated.

Cohorts a–e were allowed to lay eggs for 3 days on each host species (9 days in total), with the succession order following one of six possible combinations (electronic supplementary material, table S2). Females of cohorts f–i always met B. napus during day 1–3 and B. vulgaris during day 4–6. In Stockholm (a–e), females were kept at approximately 26°C from 9.00–17.00 and at approximately 17°C from 17.00–09.00. In Uppsala (cohort f–i), females were placed in a climate cabinet (Bio Chambers-SPC-56, Winnipeg Manitoba) with a 9 h (20°C) light and a 15 h (15°C) dark photoperiod. Females were fed from cotton balls moistened with 20% sugar solution through the gauze top of the 1 l plastic jar. Eggs were counted daily.

(c). Size, fecundity and host acceptance

Cohorts a and b were targets of an experiment that tested how female fecundity level affected host acceptance. In cohort a, all pupae spent 12 months in diapause at 2°C until being transferred to 23°C and a photoperiod of 22 : 2 h L : D. All pupae were sexed and weighed individually. Based on these weights, eclosing adults were divided into one of four groups (large/small, males/females) (electronic supplementary material, table S3). In the second cohort (b), we used directly developing individuals. These were reared in groups of five larvae in 1 l plastic cups with ad libitum access to A. rusticana. In order to generate a large size range, 98 of the 168 larvae were removed prematurely from the host plant in the latter part of the ultimate larval instar. At this larval stage, food deprivation induces pupation, which meant that some larvae were forced to pupate at smaller sizes. All individuals were weighed as pupae and divided into one of four groups (large/small, males/females) (electronic supplementary material, table S3).

Upon eclosion all individuals were marked individually and ‘small’ and ‘large’ males were transferred to one of two separate mating cages. Then, a similar proportion of ‘small’ and ‘large’ females were released in each cage. Following mating, females were first tested for egg-laying preference and transferred to individual cages provided with a leaf of each of the plants A. rusticana, B napus and B. vulgaris, and allowed to lay eggs for 2 days between 9.00 and 17.00. Following this choice experiment, females were tested for host acceptance (see above).

(d). Statistical analysis

Statistical analysis was performed in the software Statistica 13 [47]. To test the relationship between host plant acceptance (no-choice) and host plant preference (choice) we performed a repeated measures ANOVA (III) on the female average egg-laying rate (log-transformed number of eggs/day) in cohorts a and b. We used cohort and treatment (choice/no choice) and their interaction as factors. Non-significant interactions were stepwise removed from this and subsequent ANOVA models. When applicable, significant differences among groups were further explored with Tukey's HSD test.

The effect of cohort on female host acceptance was further assessed using a chain of analyses. First, we tested the effect of cohort on female egg-laying in all nine cohorts, using the daily rate of egg laying (log-transformed) on B. napus and B. vulgaris in cohorts (a–i) as the repeatedly measured response variable (ANOVA III) with cohort as factor. A significant main effect of cohort would indicate cohort-specific fecundity variation, and a significant interaction between cohort and host plant would indicate that different cohorts show different relative acceptance of the different hosts. In similar models, we used site (Uppsala) or generation (direct/diapause) as factors.

We asked whether variation in effect size among cohorts, in this case a difference in acceptance between hosts, was negatively related to variation in sample size among cohorts (as predicted by e.g. [7]). We subtracted the mean daily number of eggs laid on B. napus from the mean daily number of eggs laid on B. vulgaris for all nine cohorts. For cohorts a–e, we generated similar differences for B. vulgaris–A. rusticana, and A. rusticana–B. napus. These values were standardized by dividing them by the average daily fecundity of each cohort. The resulting effect sizes were separately regressed against the cohort sample size.

Then we performed more detailed analyses on the five cohorts (a–e) that included all three host plants to further explore potential causes behind cohort-level variation. Again, we used a repeated measures ANOVA with the number of eggs laid on each host per day (log-transformed to approach similar variances) as the repeatedly measured response variable, with cohort (a–e) and presentation order (electronic supplementary material, table S2) as factors. We evaluated the relationship between total fecundity (number of eggs laid + number of mature eggs left in the abdomen post mortem) and female size (discal cell length) in a mixed model (ANOVA III) with cohort as a categorical factor. Thereafter, we focused on the female propensity to oviposit on the least preferred host plant by calculating the proportion of eggs laid on B. napus by each female in the no-choice experiment. This variable was then used as a response variable in two separate mixed models (ANOVA III) with cohort (a–e) as a categorical factor and either the total fecundity or female size as continuous variables. For the females in cohort j, that spent their entire life in close proximity to B. napus, we performed a Spearman rank correlation between the number of eggs still in the abdomen post mortem and the number of eggs laid during female life.

Finally, we used the datasets of cohorts a and b to experimentally assess the impact of size (of both males and females) and its link to fecundity as a potential explanation for cohort-specific variation in host acceptance. We divided females into four groups depending on their size (large/small) and the size of their mating partner (large/small). We then performed a repeated measures ANOVA, again with the number of eggs laid on each of the three hosts (log-transformed) as the repeatedly measured response variable, and with cohort (a,b) and size group (large females and males, LL; large females and small males, LS; small females and large males, SL; and small females and males, SS) as factors. We further evaluated the effect of male and female size on average fecundity (log eggs/day) in a linear model with cohort (a,b), male size (large/small) and female size (large/small) as factors.

3. Results

(a). Host plant preference and acceptance

Females of cohorts a and b had a similar host plant rank order (B. vulgaris > A. rusticana > B. napus) both in the no-choice trials and in the choice experiment (figure 1a). However, significantly more eggs were laid on the intermediate (A. rusticana) and lower-ranked (B. napus) host species in the no-choice experiment than in the choice trials, where B. vulgaris often was the single host species receiving eggs (table 1 and figure 1b). There was also a general difference between the two cohorts in the propensity to oviposit on the least preferred host, B. napus (table 1), with cohort b laying a higher proportion of eggs on this plant than cohort a (figure 1a). There was no significant relationship between the proportion of eggs a female laid on a certain host plant in the choice experiment, and the proportional distribution of eggs laid by that female on that host in the no-choice experiment (A. rusticana r = 0.001, p = 0.99; B. vulgaris r = −0.067, p = 0.55; B. napus r = −0.045, p = 0.69). Hence, females that oviposited on lower-ranked hosts in the choice experiment were not necessarily the same females that oviposited on lower-ranked hosts in the no-choice experiment and vice versa (figure 1b).

Figure 1.

(a) The mean daily number of eggs (log-transformed) laid on A. rusticana, B. napus and B. vulgaris (±95%CI) by females of cohorts a (left panel) and b (right panel) in the no-choice set-up (dark bars) and the choice set-up (light bars); (b) the proportion of eggs laid on each plant by each individual in each experiment. Note the larger variation in host acceptance (x-axis) than in host preference (y-axis).

Table 1.

Repeated measures ANOVA (III) reporting the effects of cohort (a,b), test type (choice/no choice) and their interaction on the willingness to oviposit on the three different experimental plants. Non-significant interactions (in italics) were removed from the final model.

	d.f.	F	p
cohort (C) test type (TT) C × TT	1	5.98	0.015
	1	49.0	<0.001
	1	3.55	0.057
error	161
host plant (HP)	2	256.5	<0.001
C × HP	2	4.07	0.018
TT × HP	2	12.9	<0.001
C × TT × HP	2	0.50	0.61
error	322

(b). Host plant acceptance

The egg-laying rate on B. vulgaris and B. napus varied significantly among cohorts (a–i; table 2a, figure 2a), but did not vary significantly with site (Uppsala/Stockholm; Site. × Host pl. F_1,259 = 0.28, p = 0.60; electronic supplementary material, figure S1), nor generation (diapausers/direct developers; Gen. × Host pl. F_1,259 = 0.47, p = 0.49; electronic supplementary material, figure S1). Cohorts d and e laid substantially more eggs than the other cohorts, and there was a significant interaction between the number of eggs laid on each host plant (B. vulgaris/B. napus) and the cohort, indicating that the relative acceptance of B. napus varied among cohorts. Still, females of all cohorts accepted B. vulgaris to a larger extent than B. napus (figure 2a). The mean number of eggs laid by each cohort was not significantly related to variation in cohort sample size (linear regression: b = −0.88; t₇ = −1.02; p = 0.34), and cohort sample size did not affect the average effect size (difference in acceptance of different hosts) significantly (figure 2b).

Table 2.

Output from repeated measures ANOVAs (III) testing the effect of cohort on host plant acceptance for (a) the five cohorts meeting all three host plants, and (b) all nine cohorts that met both B. napus and B. vulgaris. Non-significant interactions (in italics) were removed from the final model.

all nine cohorts				cohorts with A. rusticana
(a)	d.f.	F	p	(b)	d.f.	F	p
				order (O)cohort (C)	5	0.98	0.43
cohort (C)	8	11.0	<0.001		4	20.2	<0.001
				O × C	20	1.18	0.28
error	252			error	146
host plant (HP)C × HP	1	369.1	<0.001	host plant (HP)O × HPC × HP	2	121.3	<0.001
					10	2.45	0.0078
	8	3.21	0.0017		8	4.68	<0.001
				O × C × HP		1.08	0.35
error	252			error	292

Figure 2.

(a) Mean (log) daily number of eggs laid on A. rusticana (five cohorts), B. napus and B. vulgaris (±95%CI) across nine P. napi cohorts (* = diapausing cohorts; u = Uppsala cohorts). (b) The non-significant relationships between effect size and sample size for B. vulgaris and B. napus (light circles; b = −0.002, t₇ = 0.29, p = 0.78), B. vulgaris and A. rusticana (b = 0.0079, t₃ = 1.78, p = 0.17; dark squares), and for A. rusticana and B. napus (b = −0.0001, t₃ = −0.21, p = 0.84; grey triangles). (c) The number of eggs laid on B. napus by females enclosed with this plant throughout their life (dark bars) and the number of mature eggs left in the abdomen at the point of death (white bars). Also shown is the negative relationship between these two variables (top left corner). Analyses of cohort a–e show significant positive relationships between the proportion of eggs that were laid on the lowest ranked plant B. napus and (d) female total fecundity (eggs laid + eggs left in abdomen post mortem) and (e) female size measured as the length of the forewing discal cell (mm) (filled circles = cohort a, open circles = cohort b, filled squares = cohort c, open squares = cohort d and triangles = cohort e).

The five cohorts that met all three plant species (a–e) showed similar host plant rank orders, but different acceptance levels for individual hosts (figure 2a). For example, cohorts c and d laid eggs on A. rusticana at similar levels as on B. vulgaris, whereas cohorts a, b and e laid significantly fewer eggs on A. rusticana (figure 2a). Brassica napus was always the least accepted host, but a significant host species×cohort interaction in the statistical model revealed that the relative number of eggs laid on this plant varied among cohorts (table 2b). The order in which plants were presented to the butterfly females had no general effect on egg-laying, but when A. rusticana was presented after B. vulgaris, females tended to lay fewer eggs on A. rusticana as indicated by a significant interaction between host plant and presentation order (table 2a; data not shown).

Several females of cohort j laid few or no eggs on B. napus despite being enclosed with this plant throughout their entire lifespan. These females typically carried multiple mature eggs ready to be laid in their abdomen at the point of death, and the number of eggs left in the abdomen was negatively correlated with the number of eggs laid (Spearman rank correlation; r_s = −0.39, p < 0.05) (figure 2c). Half of the females carried more eggs in their abdomen at death than they had laid during their entire life, whereas other females laid more than 100 eggs on B. napus and died with only few mature eggs left to lay (figure 2c).

Estimated total fecundity (eggs laid + eggs left in abdomen post mortem) was significantly positively correlated with female size (discal cell length) in the five cohorts that were dissected post mortem (cohort a–e), but the strength of this relationship varied significantly among cohorts (table 3a; electronic supplementary material, figure S2). We found a significant effect of both female size and estimated fecundity on the proportion of eggs laid on the low-ranked plant B. napus (table 3b,c and figure 2d,e), but a significant interaction between total fecundity and cohort revealed that the total fecundity affected the proportion of eggs laid on B. napus differently in different cohorts (table 3c; electronic supplementary material, figure S3). In general, however, larger females had higher fecundity, and females with higher fecundity were disproportionally more likely to use the low-ranked B. napus.

Table 3.

The output of linear mixed models (ANOVA III) testing (a) the effect of cohort and female size (discal cell length) on total fecundity (eggs laid + eggs in abdomen), (b) the effect of cohort and females size on the proportion of eggs laid on B. napus and (c) the effect of cohort and total fecundity on the proportion of eggs laid on B. napus. Non-significant interactions (in italics) were removed from the final model.

	(a) total fecundity			(b) prop. laid on B. napus				(c) prop. laid on B. napus
	d.f.	F	p	d.f.	F	p		d.f.	F	p
cohort (C)	4	1.96	0.1	4	4.59	0.016	cohort (C)	4	6.11	<0.001
female size (FS)	1	26	<0.001	1	8.36	0.004	total fec. (TF)	1	16.3	<0.001
C × FS	4	2.64	0.037	4	0.85	0.5	C × TF	4	6.14	<0.001
error	140			144			error	166

(c). Female size, fecundity and host acceptance

The effects of size on host acceptance were further investigated in cohort a and b where data were also available on male size. Both male and female size impacted female fecundity, as indicated by a significant effect of size group combination on egg-laying (table 4a). Subsequent post hoc analyses revealed that large females that mated with large males (LL) had significantly higher fecundity than the other three potential male–female size combinations (LS, SL, SS) (figure 3a). Furthermore, the effect of male and female size on female egg-laying was largest on the non-preferred host B. napus (figure 3a), as indicated by a significant size group × host plant interaction (p = 0.019; table 4a). An analysis trying to further disentangle the effects of cohort, female size and male size on total egg-laying rate (across host plants), revealed that although both male and female size affected female fecundity, the importance of female size varied among cohorts (figure 3b), as indicated by a significant interaction between cohort and female size (table 4b).

Table 4.

Output table from (a) a repeated measures ANOVA (III). Factors include cohort (a,b), size group (LL, LS, SL, SS), the repeatedly measured variable host plant species and their interactions. (b) The output of a linear model (ANOVA III) on the same dataset, testing the impact of cohort (a,b) female size (large/small), male size (large/small) and their interactions on the female fecundity in the host acceptance experiment (log eggs/day). Non-significant factors and interactions (in italics) were removed from the final model.

(a)	d.f.	F	p	(b)	d.f.	F	p
cohort (C)	1	0.0013	0.97	cohort (C)	1	0.46	0.50
group (G)	3	4.21	0.0081	female size (FS)	1	14.5	<0.001
C × G	3	1.26	0.29	male size (MS)	1	6.22	0.015
error	82			C × FS	1	9.53	0.0028
host plant (HP)	2	98.5	<0.001	C × MS	1	0.51	0.48
C × HP	2	2.69	0.071	FS × MS	1	0.56	0.46
G × HP	6	2.60	0.019	C × FS × MS	1	2.51	0.12
C × G × HP	6	0.81	0.56	error	77
error	164

Figure 3.

(a) The mean daily number of eggs (log-transformed) laid on B. vulgaris, A. rusticana and B. napus (±95%CI) by females of the four size group combinations (large female/males (LL); large females/small males (LS); small females/large males (SL); and small females/males (SS)) of cohorts a and b. (b) The mean daily fecundity of small and large females mated with small and large males (±95%CI).

4. Discussion

Collectively, the results of our set of experiments emphasize that an insect female's propensity to accept a host plant species is based on a complex combination of her innate predisposition and her ability to obtain resources during larval- and adulthood (cf. e.g. [30,31,48,49]). Such effects of female resource level are predicted from theory [30,33], and have previously been tested by linking female egg-laying decisions to her current egg load (e.g. [35,50,51]). However, Agnew and Singer [52] demonstrated that an increasing acceptance for lower-ranked hosts with increasing egg load does not necessarily reflect a causal relationship, because a prolonged search without host encounters could increase both egg load and host acceptance of low-ranked hosts when studied in nature [52]. Here, we used controlled experiments to show how both male and female size variation contribute to variation in female fecundity and how the fecundity variation, in turn, influenced among-cohort variation in host acceptance. The P. napi females maintained their host plant rank order across experiments, but the willingness to accept lower-ranked host varied substantially among cohorts. Hence, our study emphasizes the importance of reproducing behavioural experiments, and the difficulty of making definite statements about host plant utilization patterns based on single experiments on limited numbers of individuals.

The risk of drawing premature conclusions from individual studies can be exemplified by comparing the results of different cohorts. For example, females of cohorts c and d laid similar numbers of eggs on B. vulgaris as on A. rusticana, whereas females of other cohorts discriminated significantly against the latter plant (figure 2a). Also, acceptance for the lowest-ranked species B. napus was highly variable among the nine cohorts that compared acceptance on B. napus and the highest-ranked B. vulgaris. The proportion of females that did not lay a single egg on B. napus varied between 7% (cohort d) and 55% (cohort g). If such variation in host plant acceptance had been identified in comparisons of different populations, it could easily have been interpreted as reflecting local specialization and the potential for future diversification.

Importantly, the host plant rank order was similar across treatments, and largely reproducible both for host plant preference and host plant acceptance. Barbarea vulgaris was the most preferred plant, A. rusticana was often of intermediate interest, and B. napus was consistently the lowest ranked plant, which is congruent with a previous study performed on this population [38]. As predicted, females were more willing to oviposit on lower-ranked hosts in the acceptance assay, than in the preference assay, supporting the well-established hypothesis [30,53–55] that females can alter their use of certain hosts depending on the current availability and abundance of different host species. Such test-specific effects have been commonly discussed also from applied viewpoints when evaluating biological control agents for weeds [56] or insect pests [57], where the importance of the control agent risks being overemphasized in no-choice tests. Our data hint at an additional difficulty of evaluating the results of the two test types, because individual females showed no consistency between the preference and the acceptance experiment, which further emphasizes the difficulty to directly compare results of choice and no-choice bioassays.

The variation in propensity to oviposit on the low-preferred host was not linked to different developmental pathways (diapause/direct development), laboratory venues (Stockholm/Uppsala) or to the sample size variation across cohorts (cf. [7]). Instead, the propensity to accept the lowest-ranked host plant B. napus, was influenced by the total female fecundity. Female fecundity, in turn, was influenced both by female size and the size of her mating partner, which is tightly correlated to the size of the spermatophore nuptial gift that the male transfers to the female during mating. In the targeted experiment on cohorts a and b, large females that mated with large males laid the highest number of eggs, whereas small females and females that mated with small males tended to lay fewer eggs especially on the low-ranked B. napus. Hence, size variation could partly explain variation in host acceptance both among cohorts and individuals. However, our data further indicate that fecundity alone cannot explain all among-cohort and individual variation in host acceptance. Even when confined with B. napus for their entire life, many females refused to lay a single egg on this plant showing robust discrimination against B. napus despite having multiple mature eggs available to lay at the time of death. This indicates that the refusal to accept this host plant could be tied both to an innate (genetic) aversion against this host and as a plastic response to current fecundity.

The limited sample sizes of many behavioural experiments makes them susceptible to chance events in general, and drift or founder effects in particular, which depend on the genetic variability of the target trait in the source population. In our case, the laboratory population was founded by individuals collected as eggs or larvae from two nearby sites. This means that even though these sites were only 20 km apart, we cannot exclude the risk of having also imposed micro-level variation that, through chance events, could be non-randomly distributed across cohorts. This would be the case if females from the two sites, collected as eggs or young larvae, had consistently somewhat different host preferences. There is still a general lack of knowledge of the extent of such genetically determined host use variation within populations and of studies that evaluate the scale of host plant preference and acceptance variation.

Many recent studies from a variety of scientific fields point out how the increasing pressure to ‘publish or perish’ risks affecting the propensity and opportunity to reproduce and validate both one's own and other scientists' findings [1–6]. Hence, there is an apparent risk that future research programmes, or applied measures such as medical treatments or pest management efforts, are based on non-reproducible foundations. Indeed, recent studies in animal and human behaviour [58,59] point to the often poor reproducibility of many behavioural experiments. Few studies have approached these issues in the study of phytophagous insects, which is a highly diverse group that harbours many pest species and where host utilization is generally considered a key trait for diversification [7,9–13]. Our dataset demonstrates that variation in host use of an insect herbivore can be impacted through plasticity in female fecundity in addition to innate or genetic predisposition at a local scale. In our dataset, host plant rank order was more stable across experiments and treatments than the number of eggs laid per se, and future studies should potentially take this into account when choosing the response variable or experimental set-up for female host preference- and/or acceptance experiments. More generally, this study underlines the necessity to pay increasing attention to assessment of reproducibility also in ecological studies.

Supplementary material

See electronic supplementary material (Schapers_et_al_tablesS1-S3_figuresS1–S3_electronic supplementary material.pdf). Electronic supplementary material, tables S1–S3 provide additional background information about experimental set-up, sample size, etc. Electronic supplementary material, figures S1–S3 provide additional graphics to the results.

Data accessibility

Data are available from the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.8qp7s.

Authors' contributions

M.F., A.S., C.W., C.W.W. and H.P. planned the research; A.S., C.W., H.P. and M.F. performed experiments; M.F., A.S. and C.W.W. performed statistical analysis, M.F. wrote the manuscript, A.S. provided substantial revisions, and H.P., C.W.W. and C.W. commented and provided editorial advice on the text.

Competing interests

The authors declare no competing interests.

Funding

M.F. was supported by a grant from the Swedish Research Council. C.W.W. was supported by grants from the Swedish Research Council, and the Knut and Alice Wallenberg Foundation.