Learning performance and brain structure of artificially-reared honey bees fed with different quantities of food

Artificial rearing

We reared 853 worker bees artificially under four different feeding regimes (Table 1) that varied in total quantity of artificial diet fed; 150, 160, 170, and 180 µl of total food volume. Rearing took place from the 10th of April 2014 until the 2nd of May 2014 (date of eclosion). Worker bees were obtained from two hives in our institutional apiary at the Biocenter, University of Würzburg. We controlled for age of the bees by enclosing queens on an empty brood comb for 24 h (April 10, 2014) using a wire mesh cage that allowed nurse bees to freely pass, but kept queens confined. Upon hatching of the eggs (April 14, 2014), combs were transferred to the lab, where individual larvae were grafted and placed in a small plastic cup (Nicotplast, Maisod, France) containing 20 µl of artificial diet, situated in a sterile 48-well plate. Well plates and content were pre-heated to 35 °C and grafting took place on a thermal mat (ThermoLux, Murrhardt, Germany), to avoid hypothermia. Well plates were kept in a brood stove at 35 °C and 95% RH. Larvae were reared according to the protocol of (Aupinel et al. (2005); Aupinel et al. (2007) and Aupinel et al. (2009) also see Crailsheim et al., 2013) with some adjustments in regard to the total quantity of artificial diet being fed. We fed four different quantities of artificial diet; 150 µl, 160 µl (according to Aupinel), 170 µl and 180 µl. We varied the quantity of diet given on the last two days to obtain the described gradient (Table 1). Each treatment was represented on each well plate. After pupation well plates were placed in another brood stove with 35 °C and 80% RH. Prior to eclosion plastic cups containing pupae were reordered so that well plates contained only pupae with the same hive origin, that received the same treatment, and these were placed in sealed hatching boxes. From each hive we collected two capped brood combs and also placed them in the incubator one day before artificially-reared bees were starting to eclose (May 2, 2014), we marked hatching adults from these combs and placed them back in their hive for the in-hive control. Hatched in-lab bees were collected and placed in flight cages, kept in a brood stove (28 °C, 70% RH) until the olfactory conditioning experiment 11 days after hatching. Upon eclosion we took a subsample of 10 bees from each treatment (the in-lab reared treatments, as well as the in-hive control group) for brain reconstruction.

Olfactory conditioning

We performed a classical differential olfactory conditioning experiment (see Matsumoto et al., 2012). Conditioning took place from the 12th until the 16th of May and again from the 20th until the 22nd of May 2014. Mean age of the conditioned bees was 16 (±5) days. On the day prior to conditioning bees were immobilized on ice and placed in a plastic harness tube. We let the bees get accustomed to the new situation for about half an hour and fed them until satiation with a 50% (w/w) sucrose solution. Bees were kept in a brood stove (23 °C, 70% RH) and starved overnight until the conditioning trials the following day, in order to establish a uniform hunger level. Half an hour prior to the first conditioning trial bees were taken out of the brood stove and tested for a sucrose response by touching their antennae with a 50% sucrose solution. The proportion of bees that failed to produce PER at the start of the conditioning trials was used as a proxy for sucrose responsiveness which is known to correlate with learning performance in honey bees (Scheiner, 2004; Scheiner, 2012; Scheiner et al., 2005). Bees that showed proboscis extension reflex (PER) were placed in compartments on a rail that slides in front of a ventilation system. We used two different odors as conditioned stimulus (CS); 1-nonanol as the rewarded stimulus (CS+), and 1-hexanol as the unrewarded stimulus (CS−). The reward, or unconditioned stimulus (US), was a 50% (w/w) sucrose solution. Harnessed bees were subjected to twelve conditioning trials in which CS+ and CS− were presented in a haphazard order (each odor six times), followed by two retention tests; one hour, and 48 h after the last trial. Retention tests were carried out on the same bees. We conducted conditioning (and retention tests) as a blind study. To present the CS we used two 20 ml syringes, each with a piece of filter paper that contained 4 µl of either odor. The syringe was held about 15 mm from the bees’ antennae and CS was presented for four seconds. Two seconds after the onset of CS+ the US was presented for 1 s with a toothpick to the antennae to elicit PER, after which the toothpick was held to the proboscis for two seconds. When using CS- a toothpick was held in front of the bee, so that the visual cues were the same as with CS+, but bees were not rewarded. We waited some seconds to allow residual odors to be removed by the ventilation system behind the bee, then moved the next bee in front of the ventilator. After a 10 s pause conditioning was performed on the next bee. Between trials was a 10 min interval. If the bee showed PER to the CS+ the response was recorded as 1, no response was recorded as 0. If the bee showed no PER upon antennal contact it was recorded as n, so that bees that did not show PER three times upon antennal contact could be excluded from the experiment. For the statistical analysis n was considered as a 0. Bees that showed spontaneous PER in the first CS+ trial within the first 2 s were also excluded from the experiment. One hour after the 12th trial, response to CS+, CS−, and US were tested (1 h retention test). CS+ and CS− were tested in a similar fashion as in the conditioning phase but without the reward, US was presented to assure that bees were still capable of PER. If not, those bees were also excluded from the experiment. Bees were kept confined in their tubes and fed to satiation regularly. This retention test was then repeated one more time after 48 h after last conditioning trial. After the last retention test bees were frozen and stored at −20 °C for further measurements. To compare bee size over feed treatments we measured head size and intertegulae span (ITS) using an imaging software program (ImageJ; US National Institutes of Health, Bethesda, MD, USA). To determine head size the shortest distance between the compound eyes was measured, and for ITS the shortest distance between the tegulae was taken (Cane, 1987). As an additional control, we conducted a reciprocal conditioning experiment to show that the bees can associate either odor with the presence or absence of a reward. Conditioning took place from the 22nd until 28th of October 2015. Another 168 bees were taken from the same hives as we used for the initial olfactory conditioning experiment. Half of the bees (82 bees) were conditioned using 1-nonanol as CS+ and 1-hexanol as CS−, for the other half (86 bees) 1-hexanol was used as CS+ and 1-nonanol as CS−.

Wholemount preparations

To investigate possible effects of diet on neuronal development, we compared volume of brain neuropils of freshly eclosed bees from all five treatment groups. Brains were prepared according to the protocol by Muenz et al. (2015). In brief, the bee was anaesthetized on ice, the head was removed and transferred to a wax dish. The head was then covered with physiological saline (130 mM NaCl, 5 mM KCl, 4 mM MgCl₂, 5 mM CaCl₂, 15 mM Hepes, 25 mM glucose, 160 mM sucrose; pH 7.2) and a small window was cut into the frons between the antennae. After removing glands and tracheae the brain tissue was immediately transferred into ice-cold 4% formaldehyde (FA; methanol free, 28908; Fischer Scientific, Schwerte, Germany) in 0.01 M phosphate-buffered saline (PBS; pH 7.2) and fixed at 4 °C overnight. For visualization of different neuropils, we applied an anti-synapsin I antibody in whole mount preparations following the protocol by Groh et al. (2012). The fixed brain was treated with PBS and PBS Triton X-100, before being incubated with the primary mouse antibody SYNORF1 (1:10; kindly provided by E. Buchner, University Würzburg, Germany). A CF488A conjugated goat anti-mouse antibody (1:250; Biotrend, Köln, Germany) was applied for fluorescence labeling and the brain was then dehydrated in an ascending ethanol series and mounted on custom made metal slides in methyl salicylate (M-2047; Sigma Aldrich, Steinheim, Germany).

Confocal laser-scanning microscopy and 3D-reconstructions

Whole mount preparations of synapsin labeled brains were scanned using a laser-scanning confocal microscope (Leica TCS SP2 AOBS; Leica Microsystems AG, Wetzlar, Germany) equipped with an argon/krypton and three diode lasers. A HC_PL_APO objective lense (10x/0.4 NA imm) was used for image acquisition and optical sections were taken in 5 µm steps. Image processing was performed using 3D image software AMIRA 5.3 (FEI Visualization Group, Mérignac, France, see Fig. 1). Based on outline and interpolation reconstructions the volumes of the following neuropils (without cell bodies) were measured (also see Fig. 1): antennal lobe (AL), optic lobe (OL; comprising the medulla and lobula), mushroom body (MB, comprising calyx and pedunculus), central complex (CX, comprising upper and lower central body and protocerebral bridge) and protocerebrum (PC, without suboesophageal ganglion). Volumetric data were calculated in Amira 5.6 and exported to Microsoft Excel 2007 (Microsoft Corporation, Redmond, WA, USA) for further analysis.

Statistical analysis

All statistics were performed using R statistical software (R Core Team, 2014). We evaluated whether there is a difference in the proportion of bees among treatment groups that failed to show PER at the start of the conditioning trials using a Chi-squared test. The responses during the conditioning trials (learning phase) were analyzed with a generalized linear mixed effects model (glmer function, binomial family) using the lme4 package (Bates et al., 2014). Preliminary analyses showed that on one of the conditioning days (14th of May 2014) bees were not able to learn well, probably due to a defective sink in the lab that caused a pungent smell. We decided to exclude these bees from further analysis. In the full model we specified PER as a binary variable in response to the fixed factors trial, CS (CS+ or CS−), and treatment (150 µl, 160 µl, 170 µl, 180 µl, and the control; in-hive), with their interactions (also see Kirkerud et al., 2013). To account for the differences in age of the conditioned bees as the experiment progressed, we included date (of conditioning) as random factor. Colony background was also included as random factor to account for potential differences between the two colonies that were used. Finally also bee ID was included as random factor to correct for the repeated measurements. Model simplification was carried out via stepwise model simplification using Likelihood Ratio Tests (LRT). Trial was no longer taken into account. The 1 h and 48 h retention tests (memory formation) were analyzed in a similar way; whether or not bees showed PER as the binary response variable, in relation to the factors CS and treatment, and their interaction, with colony background and date of conditioning in the random structure. Not all bees survived the 48 h period being harnessed in the plastic tubes, in two of the treatment groups (160 µl and 180 µl) none of the remaining bees responded to the CS−, resulting in unbalanced data. Thus for the 48 h retention test we used the function bglmer from the package blme (Dorie, 2014) which is able to work with unbalanced data. All models were checked for overdispersion. For the 1 h retention test we had overdispersion, which was dealt with by including an observation level random factor (Pirk et al., 2013).

To compare ITS and head width among treatment groups we carried out linear mixed effects models (function lme), including colony as random factor (Pinheiro et al., 2014). Visual evaluation of the model fits showed that residuals were normally distributed and the assumption of heteroscedasticity was met. For post hoc comparisons among treatment levels we carried out a Tukey test with Benjamini Holmes correction (Benjamini & Yekutieli, 2001).

Total brain volume, as well as the different neuropils (see previous paragraph) were analyzed with a regression type analysis (function lm, package stats—R Core Team, 2014) with treatment as only fixed factor. As all bees in the subsample were taken on the same day (right after eclosion) and originated from the same colony there was no need to specify a random structure here. Model validation was carried out via visual evaluation of the model residuals. All models had met the assumption of normal distribution of residuals, however the variances of all but one model (CB) were heterogeneous. For those we allowed for unequal variances across treatment groups using the function varIdent from the package nlme (Pinheiro et al., 2014; Zuur et al., 2009).

Differential olfactory conditioning

A total of 229 bees lived after being harnessed and starved for a night prior to the conditioning experiment. We compared the proportions of bees that failed to show PER at the start of the experiment among treatment groups as a proxy for sucrose responsiveness. There was no difference in sucrose responsiveness among treatment groups (χ² = 7.472, df = 4, p = 0.113). A total of 140 bees were conditioned, seven of which had to be excluded because their behavior was influenced by a disturbance of the experiment (on 14th of May 2014). A further five bees that showed spontaneous PER upon the first trial were also removed from the data set, leaving a total of 128 conditioned bees to do the analysis with. In all artificially-reared groups, regardless of treatment, a majority of bees was able to differentiate between CS+ and CS− over the course of the conditioning period (GLMM: z = 12.200, p < 0.001). Upon first glance it seems that artificially-reared bees performed slightly better than bees that were reared in the hive (Fig. 2); the percentages of bees that showed PER to CS+ at the last trial were 83% (150 µl), 77% (160 µl), 67% (170 µl), 63% (180 µl), compared to 48% of the in-hive bees. However, statistically all groups that were reared artificially (regardless of the quantity of artificial food they received) performed equally well as the in-hive bees (LRT: χ² = 6.600, p = 0.159). When we disregard trial, we find that those bees that received the highest quantity of food in the lab performed marginally better than the in-hive bees (GLMM: z = 1.796, p = 0.073), which can also be easily seen in the graph as none of those bees respond to the CS− as from trial 5. After one hour (mid-term memory; MTM) all bees that were reared in the lab perform better at differentiating CS+ and CS− than the in-hive control (GLMM: (150 µl) z = 1.982, p = 0.048; (160 µl) z = 2.076, p = 0.038; (170 µl) z = 4.662, p = 0.023, (180 µl) z = 1.999, p = 0.046). During the 48 h retention test (early long-term memory, eLTM) all treatment groups performed the same again (LRT: χ² = 2.950, p = 0.566). During the reciprocal conditioning bees showed to be able to associate either odor equally well with the presence or absence of a reward (GLMM (CS+ = 1-nonanol): z = 9.682, p < 0.001; (CS+ = 1-hexanol): z = 9.819, p < 0.001), as well as distinguish between odors regardless which odor was used as CS + during conditioning after 1 h (GLMM: (CS+ = 1-nonanol): z = 9.184, p < 0.001; (CS+ = 1-hexanol): z = 7.766, p < 0.001) and after 48 h (GLMM: (CS+ = 1-nonanol): z = 18619, p < 0.001; (CS+ = 1-hexanol): z = 3.486, p < 0.001). For a graphical depiction of these outcomes see the Supplemental Information.

Morphometrics

We measured head and thorax width (ITS) of all the bees that were used for the olfactory conditioning experiment, plus some additional bees (that were not suitable for conditioning, or had died before start of the experiment) resulting in a total number of 263 bees. When fed with 150 or 160 µl bees were smaller than those of the control in terms of ITS (Linear regression: (150 µl) t =  − 6.687, p < 0.001; (160 µl) t =  − 4.493, p < 0.001) as well as head width (Linear regression: (150 µl) t =  − 3.073, p = 0.002; (160 µl) t =  − 2.954, p = 0.003). But bees fed with only 150 µl tend to be even smaller (ITS) than bees fed with 160 µl (Fig. 3—not significant; Tukey post hoc test: z = 2.219, p = 0.170), yet their heads had the same size (Tukey post hoc test: z = 0.008, p = 1.000). There seems to be a tradeoff; with limiting resources a bee invests in the development of the head, at cost of body size (ITS). Likewise, when fed with a surplus diet (180 µl treatment) bees’ heads were significantly bigger than the in-hive control (linear regression: t = 2.343, p = 0.020) yet in terms of ITS, bees had the same size as the control (linear regression: t =  − 1.107, p = 0.269).

On the first day of eclosion, 10 bees per treatment group were taken as a subsample for brain volume reconstruction. Out of those 50, 45 bee brains were successfully reconstructed (Fig. 4). Looking at the overall brain volume, we did not detect any differences (Table 2). However the brain sizes of bees that were fed between 150 and 170 µl were marginally smaller than the control. Looking in more detail, we find that artificially-reared bees had smaller calyces. For the 150 and 170 µl groups medial calyces were smaller than the in-hive control and lateral calyces were consistently smaller for all treatment groups (Table 3). This indicates that higher central neuropils (MB) but not the periphery (OL and AL) have been affected.