Comparison of Different Pollen Substitutes for the Feeding of Laboratory Reared Bumble Bee (Bombus Terrestris) Colonies

Colonies were reared from commercial colonies (Bioplanet S.c.a., Cesena, Italy) following the procedure described by Bogo et al. (2017). Obtained colonies were reared in a climate room (28 ± 1 °C and 60 ± 10% RH, continuous darkness) and fed on fresh frozen pollen and sucrose solution (1:1 w/w). When the first brood emerged, the colonies were moved to bigger plastic boxes (25 × 15 × 14 cm) and the protein source was shifted to the test diets. Before diet shifting, we estimated the colony size by counting the number of egg cells, larvae, pupae and adults. We assigned the colonies to the different test diet groups making sure that all the groups were composed by colonies similar in size. Three times a week the colony boxes were cleaned, and food was refreshed (both protein source and sucrose solution).

Three times a week we recorded test diet consumption (in weight), number of dead adults, number of dead larvae, date of colony ending (i.e. no workers left and/or complete absence of brood), date of gyne and male emergence and number of emerged gynes. After colony ending, we calculated the number of emerged workers in the period of food substitution as the number of alive adults at the end of the trial plus the number of dead adults during the trial, minus the number of pupae and adults at the beginning of food substitution. Finally, we calculated the colony lifespan, from the beginning of the test to colony ending.

In order to compare colony output in all the groups, we considered the above parameters only during the first twenty-four days of the assay trial and during the first twenty-five days of the second assay (see below), since these were the lifespans reached by the shortestlasting colonies, respectively in the two assays.

We performed the first assay with eight commercial pollen substitutes sold for honey bees colonies during periods of pollen shortages (Karya Ari Keki^®, Calcalar Ltd, Muğla, Turkey; Bee Food^®, Melissokomiki Athinon S.A., Schimatari, Greece; Candipolline^®, Enolapi Srl, Verona, Italy; Feedbee^® (two types: powder and paste) and Nutri-Bombus^® (three types: A, B and C; Nutrifeed Canada Inc., Ajax, Canada) and two natural protein sources (brewer's yeast and chestnut flour) (Tab. 1). The powdery-formulated diets were mixed with sugar syrup, obtained by dissolving the sugar in water in a proportion of 1: 2 (1 kg sugar in 2 litres water). Fresh frozen pollen was used as a positive control, administered in either honey bee pollen pellets alone or mixed with sugar syrup (pollen paste).

In the second assay, we tested again the pollen substitute which had given in the first assay the best result and took into consideration also costs and market availability, namely Feedbee powder, mixed with 50% and 75% of pollen paste. The characteristics of the different diets are described in Tab. 1.

In order to determine the nitrogen concentration (N), the Kjeldahl method (Bradstreet, 1954) was applied, following Conti et al. (2016). A minimum quantity of 300 mg per sample (one for each diet) was used. Crude protein (CP) was estimated by multiplying N concentration by 6.25 (Jones, 1931). For the determination of sugar (summation of sucrose, glucose and fructose) concentrations, one sample (15 g) for each diet was processed according to Conti et al. (2016).

Statistical analyses were performed using multivariate analysis of variance (MANOVA), one-way analysis of variance (ANOVA) and discriminant analysis (DA) (Podani, 2000; Sokal & Rohlf, 2010). In order to control type I error for the subsequent separate univariate ANOVAs following MANOVA, we corrected the experiment-wise error rate by the sequential Dunn-Šidak method (Sokal & Rohlf, 2010). The DA was used to investigate the nature of multivariate effects of different diets and how the dependent variables together discriminate the treatments (Podani, 2000). The DA was carried out using Canoco for Windows 4.5 (Lepš & Šmilauer, 2003; Ter Braak & Šmilauer, 2002) and a manual forward selection of variables, corrected by the sequential Dunn-Šidak method, was used to choose the best explanatory ones after 9999 permutations under a reduced model (Anderson & Legendre, 1999). Furthermore, the selected variables were evaluated for the performance in the correct classification of different bumble bees into their respective diets by estimating the probabilities of misclassification with cross validation.

Effect sizes within MANOVA, ANOVAs and DA were estimated, respectively, with adjusted ξ², Cohen's ω² and R². Given an evidently different number of replicates per treatment, a posteriori comparison of individual means was based on the minimum significant difference (MSD) method obtained from the Hochberg's GT2 statistic (Sokal & Rohlf, 2010).

Data were subjected to a logarithmic transformation (Sokal & Rohlf, 2010) before analysis, which effectively homogenized variances (Levene's test) and produced normal distributions (Shapiro-Wilk test) (Sokal & Rohlf, 2010). Consequently, we chose to present the median values instead of the arithmetic means. Covariance was homogenous as well (Box's M test) (Podani, 2000). In order to compare the different diets for their similarity in sugar and CP concentration, a principal coordinate analysis (PCoA) was performed on a matrix obtained by Goodall probabilistic similarity indices (Goodall, 1966; Podani, 2000).

Only diets consisting of pollen (either pellets alone or pellets with syrup) allowed the normal development of bumble bee colonies and their subsequent production of gynes and males. The numerical parameters for these diets were not included in the analysis because they would have been out of scale compared to those of the other diets. With all the other test diets, the colonies ceased their growth at an early stage of development and did not produce males or queens.

Using Pillai's trace, food type significantly affected the colony duration, number of emerged workers and number of dead larvae and adults after twenty-four days (V = 1.49, F_{36, 364} = 5.99, p ≤0.001). The dependent variables had a large effect size being the adjusted ξ² = 0.31. The separate univariate ANOVAs on the outcome variables revealed that food type significantly affected the colony duration (F_{9, 91} = 5.59, p ≤ 0.001), dead adults (F_{9, 91} = 16.0, p ≤ 0.001), dead larvae (F_{9, 91} = 4.88, p ≤ 0.001) and emerged workers (F_{9, 91} = 2.18, p ≤ 0.05). Colonies fed with Feedbee had a significantly longer duration than those fed with Bee Food, Karya Ari Keki, Feedbee paste and chestnut flour but not from the others. Colonies fed with chestnut flour had a significantly shorter duration than all the others except for those fed with Bee Food, Karya Ari Keki, Feedbee paste and Nutri-Bombus A. The effect size was moderately large (ω² = 0.29) as colonies fed with Feedbee whose longest duration lasted 1.5 times longer than those fedwith y chestnut flour which lasted the least (Tab. 2).

Given high variation in the number of dead adults, the effect size was remarkably high for this variable (ω² = 0.57). The number of dead adults within twenty-four days under Karya Ari Keki, Bee Food and Candipolline feeding was the highest and significantly different from all other treatments (Tab. 2). Candipolline treatment was, however, not significantly different from Fedbee powder, Chestnut flour and Feedbee paste. All three Nutri-Bombus treatments had the least negative effects on adults survival and were significantly different from all the other treatments except for Feedbee powder, Brewers’ yeast and Chestnut flour.

The number of dead larvae within twenty-four days in colonies fed by Feedbee paste was the highest and significantly different from all the treatments except from Nutri-Bombus A and C. All the other treatments were not significantly different between them. Nevertheless, the effect size was moderately large (ω² = 0.26) mainly because the number of dead larvae in colonies fed by Feedbee paste was from 1.5 to 3 folds higher than in other colonies. Due to considerably high variation in number of emerged workers within each treatment, only Candipolline and Nutri-Bombus B treatments were significantly different (Tab. 2). Consequently, also the effect size was much smaller (ω² = 0.10).

Although MANOVA was highly significant and with large effect size, the single ANOVAs, separately for each variable, were generally unable to discriminate well among the different pollen substitutes. Therefore, as a follow up MANOVA we applied discriminant analysis (DA), a multivariate constrained ordination technique, that takes account of the correlation between the dependent variables. Discriminant analysis revealed four discriminant functions of which the first two explained 84% of the variance (Fig. 1A). The canonical R² of the first and second discriminant functions were, respectively, 0.68 and 0.39 in accordance with the high value of the adjusted ξ² from MANOVA. The correlations between outcomes and the discriminant functions revealed that the number of dead adults loaded significantly (p ≤ 0.001) highly in the first function (R = 0.84) (Fig. 1A). The number of dead larvae and colony duration loaded significantly (p ≤ 0.001) and moderately evenly in the the second function with opposite effect, with canonical variate correlation coefficients of −0.61 and 0.69, respectively (Fig. 1A). On the contrary, the number of emerged workers loaded negligibly but significantly (p ≤ 0.05) on both functions, with canonical variate correlation coefficients of −0.12 and −0.06, respectively (Fig. 1A).

Fig. 1

Relatively to the centroids of the different diets, the discriminant combined-treatment plot showed that the first function discriminated Feedbee powder, Feedbee paste, brewer's yeast, and all three Nutri-Bombus substitutes from all the other treatments (Fig. 1B). The second function differentiated Feedbee powder, brewer's yeast and Nutri-Bombus B from Nutri-Bombus A and C, and Feedbee paste. Finally, the second function differentiated also Candipolline from Karya Ari Keky, Bee Food and Chestnut flour (Fig. 1B). The discriminant function based on significant variables on the four group of diets, defined as above, displayed overall 67% classification success after cross-validation. More precisely, 84%, 67%, 61% and 44% classification success were displayed, respectively, for single groups comprising (i) Karya Ari Keky, Bee Food and Chestnut flour, (ii) Feedbee powder, brewer's yeast and Nutri-Bombus B, (iii) Nutri-Bombus A and C, and Feedbee paste, and (iv) Candipolline alone (Fig. 1B). Thus a partial overlap of some diets was present (Fig. 1).

From the results of the first assay, we selected Feedbee as the pollen substitute which gave the best results, taking into account all the observed parameters. The only other substitutes which gave a similar effect were brewers’ yeast and Nutri-Bombus B, but the former is more expensive while the latter at the moment was not yet available on the market. Therefore, in the second assay we mixed Feedbee with different amounts of pollen paste (50 and 75%), to evaluate the optimal amount needed to achieve whole colony development, in comparison to pollen alone.

The difference in larvae mortality was not significant among treatments (F_{2, 25} = 0.08) (Tab. 3). On the contrary, adult mortality was significantly affected by different supplements (F_{2, 25} = 4.28, p ≤ 0.05) and the effect size was moderately large (ω² = 0.19). Indeed, colonies fed by pollen paste had significantly (p ≤ 0.05) fewer dead adults (on average three folds) than those fed by Feedbee mixed with only 50% pollen paste (Tab. 3). However, there was no significant difference in dead adults between colonies fed by pollen paste and Feedbee mixed with 75% pollen paste, and between the two Feedbee treatments.

All three supplements allowed the whole development of bumble bee colonies and their subsequent production of sexuals. We then further compared the number of produced gynes, the pre-switch point period and the pre-gyne point period (i.e. the days elapsed between the deposition of the first egg cell and the switch point and gyne point respectively; Bogo et al., 2018) among these three test diets (Tab. 3) and found no significant differences after ANOVA tests (F_{2, 25} = 1.30, 1.99 and 0.01 for, respectively, produced gynes, the pre-switch point period and the pre-gyne point period).

The fourteen different diets were clearly distinguished into three clusters by their mean CP-to-sugar ratios (Fig. 2). Candipolline and Karya Ari Keky belonged to the cluster with the smallest mean CP-to-sugar ratio (0.005) while Brewer's yeast, Feedbee 25% and 50%, Feedbee paste and powder, Nutri-Bombus B and C, and Pollen pellets to the cluster with the highest mean CP-to-sugar ratio (0.7). Finally, Bee Food, Chestnut flour, Nutri-Bombus A and Pollen paste belonged to the cluster with an intermediate mean CP-to-sugar ratio (0.2) (Fig. 2).

Fig. 2

The mean total daily consumption significantly (p ≤ 0.001) differed between the test diets (Fig. 3) and with a very large effect size (ω² = 0.71). The most consumed diet was Feedbee 25% although not significantly more than Feedbee 50%, the two diets containing only pollen and Candipolline. The other test diets were significantly lower from Feedbee 25% and Pollen paste and had a daily consumption lower than one gram and from 2 to 38 folds lower than the Feedbee 25 and 50%, Pollen and Candipolline diets (Fig. 3). However, Karya Ari Keky was not significantly different from Feedbee 50%, Pollen Pellets and Candipolline, and Bee Food was not significantly different from Pollen Pellets and Candipolline (Fig. 3).

Fig. 3

When we combined total mean daily consumption with the percent of protein and sugar in each diet, calculating the amount of CP and sugar respectively achieved with each diet, we also obtained significant (p ≤ 0.001) differences between test diets and with a very large effect size (ω² = 0.80 and 0.74 for CP and sugar, respectively) (Figs 4 and 5). The significantly (p ≤ 0.05) highest amount of protein was achieved by diets containing Feedbee 25% and 50% and the two diets containing only pollen which had from 7 to 229 folds higher amount of protein than other diets (Fig. 4). Among the remaining diets, Candipolline, Karya Ari Keky and Chestnut flour had the lowest amount of protein, although not always to a significant extent (Fig. 4).

Fig. 4

Fig. 5

Sugars obtained by Feedbee 25%, Pollen paste and Candipolline were the highest and significantly (p ≤ 0.05) different from all the others diets except for Feedbee 50%, Pollen pellets and Karya Ari Keky (Fig. 5). Feedbee 50%, Pollen paste and Karya Ari Keky were also not significantly different from Bee Food, which was significantly (p ≤ 0.05) different from all the remaining diets except Feedbee powder (Fig. 5). Feedbee 25% and 50%, the diets containing only pollen, Candipolline and Karya Ari Keky had 3 to 53 folds differences from other significantly different diets (Fig. 5).