1 Introduction

Melipona quadrifasciata Lepeletier (“mandaçaia”) is a stingless bee species that occurs in most of Brazil, from the states of Paraíba to Rio Grande do Sul (Moure and Kerr 1950). The higher prevalence of subspecies M. quadrifasciata anthidioides is in hot and dry regions, especially in the semi-arid region of the state of Bahia (Nunes et al. 2008). Honey produced by M. q. anthidioides is highly appreciated and its colonies present good production when managed rationally, being of great socioeconomic and environmental importance.

In recent decades, bee colonies of Apis mellifera (vanEngelsdorp et al. 2007; Neumann and Carreck 2010) and populations of stingless bees have decreased worryingly (Steffan-Dewenter et al. 2005; Slaa et al. 2006; Freitas et al. 2009). There are numerous reasons for these reductions, such as the destruction or changes of habitats, pesticide overuse in agricultural crops, and diseases that affect bees (vanEngelsdorp and Meixner 2010; Reyes-González et al. 2016; Sanchez-Bayo and Goka 2016 and references therein). In the case of diseases, a practical and desirable strategy is that the bee species express high levels of natural resistance, such as the hygienic behavior (HB) (Evans and Spivak 2010; Bigio et al. 2013).

In A. mellifera, HB has been described as a process in which bees detect and uncap dead, parasitized or infected brood cells (5th instar, larvae and pupae) and remove them from the nest (Rothenbuhler 1964; Rosenkranz et al. 2010). When quickly performed, this process may prevent the spread of the disease causal agents to healthy brood (Rothenbuhler 1964). HB is a significant general resistance mechanism to a number of pathogens, including Ascosphaera apis (causal agent of chalk brood) (Gilliam et al. 1988), Paenibacillus larvae (causal agent of American foulbrood) (Spivak and Reuter 2001) and the parasitic mite Varroa destructor (Rinderer et al. 2010; Rasolofoarivao et al. 2015).

Hygienic colonies have large economic interests in apiculture, because they have been reported to produce more honey and pollen than non-hygienic colonies (Nicodemo et al. 2013). Although HB is a very important feature, it rarely manifests with intensity in populations (Pérez-Sato et al. 2009). Thus, the selection of queen bees related to this characteristic has been intensified in breeding programs (Espinosa-Montaño et al. 2008; Costa-Maia et al. 2011).

Most studies on HB are carried out with A. mellifera (Gramacho and Gonçalves 2009; Pérez-Sato et al. 2009; Wilson-Rich et al. 2009; Morais et al. 2010; Stevanovic et al. 2011; Pinto et al. 2012; Bigio et al. 2013; Nicodemo et al. 2013; Pereira et al. 2013; Rasolofoarivao et al. 2015). Few species of stingless bees have been investigated in terms of HB, namely Melipona beecheii, Scaptotrigona pectoralis (Medina et al. 2009) and Plebeia remota (Nunes-Silva et al. 2009).

Therefore, HB investigations in other species of stingless bees is necessary, mainly because of the diversity and specific characteristics of each species, extinction risks for the Meliponini and the growth of meliponiculture which can increase the risks of pathogen transmission (Venturieri et al. 2012), especially to species of greater economic interest. In addition, no study on HB in M. q. anthidioides has been conducted.

Our hypothesis is that HB in M. q. anthidioides occurs in a similar manner to that of bees of the genus Apis and that there is difference in the removal time of dead larvae and pupae by worker bees. The objective was to use M. q. anthidioides as a model to investigate HB in short time intervals until the complete removal of all the brood (larvae and pupae) killed by freezing in the colonies.

2 Material and methods

The study was conducted from May to August, 2015 on 40 colonies of M. q. anthidioides housed in standard boxes, INPA model, consisting of first deep and second deep (13 × 13 × 6 cm), honey super (13 × 13 × 3.5 cm), outer cover and bottom board (18 × 18 × 2 cm) and internal space to access the compartments (6 × 6 cm).

HB was quantified at the level of the colony using the freezing method with liquid nitrogen (N2) (described in Spivak and Reuter 1998), adapted to the nest dimensions of the species under study. For that purpose, an area with seven cells containing black-eyed pupae and seven larvae in the last instar (both capped cells) was delimited and marked. A control area was also delimited and marked to discount the natural removal rate of diseased, dead or parasitized pupae and larvae. A PVC cylinder (2 cm diameter × 3 cm high) was positioned over the areas to be tested and 12 mL of liquid nitrogen were poured to kill the brood. Each colony was monitored until the worker bees performed the uncapping and totally removed the brood.

The experiment was installed in a complete randomized block design with 14 treatments repeated 40 times (colonies) in each block, totaling 1680 records. The blocks were made of three different observation periods for each colony, with intervals of 30 days between each observation. The treatments or measurement times occurred up to 264 h, as follows, 3, 6, 12, 24, 48, 72, 96, 120, 144, 168, 192, 216, 240 and 264 h after liquid nitrogen application. The variables analyzed were: accumulated average proportion of dead larvae removal (LR) and dead pupae removal (PR), and accumulated average proportion of uncapped cells with not yet removed dead larvae (UL) and pupae (UP). All the colonies assessed for HB had the population size estimated before each repetition according to Ihering (1932) and after measuring the number of brood cells (Aidar 2010).

Generalized linear models (GLM) with binomial error structure and logit link function expressed by g (μ) = ln (μ/1-μ) were used to test the effect of time and population of the colonies on HB. The model parameters were estimated using the method of maximum likelihood by maximizing the log-likelihood function through generalized estimating equations (GEE). The GEE considered the dependence between the observations measured within the times (hours), which characterized repeated measures in the experimental units (colonies). The responses obtained between experimental units from different periods (blocks) were assumed to be statistically independent. The deviance analysis (ANODE) was used for the GLM adjustment, from a maximal model, represented by systematic portion η = g(μ) = μ + Ti + Bj + P + εij, where μ is the general average effect, Ti is the time levels effect in hours (i = 1, 2, …, 14), Bj is the blocks effect (j = 1, 2 and 3), P is the effect of the ‘colonies population’ covariate ,and εij is the random error.

The adjustment quality of the models to the observed data and the selection of the best model were based on the higher value of the maximum likelihood function logarithm (LogLik). Dispersion parameters were adjusted when overdispersion was found, correcting the standard errors using a quasibinomial model, for subsequent comparison between adjusted LogLik values of different models. After defining the most appropriate model, the significances of time (T), block (B) and population (P) on the variables were recorded in the type 3 GEE analysis. The effects of time and population on the variables were evaluated by means of the logistic regression models.

After replacing the time values (hours after brood death by freezing) in the models, the identity intercepts and coefficients between the logistic regression models of LR and PR and predictive models of UL and UP was carried out initially by the linear regression model adjustment with 1st degree (yi = b0 + b1xi + εi) of the average values of larvae removal and uncapping (y), estimated at each time (n = 14), based on their estimated average removal and uncapping of pupae values (x), using the method of ordinary least squares.

The existence of a linear relationship between the estimated average values of LR and PR, and of UL and UP, was assessed by detecting significance of estimated parameter β1 (angular coefficient), verified by the partial t test application to test the invalidity hypothesis of H0:βi = 0. The non-acceptance of the invalidity hypothesis to the angular coefficient suggested the influence of the PR and UP estimated average values to explain the variation in the respective LR and UL estimated average values. At the same time, the F test was applied (Montgomery et al. 2006) to test the hypothesis of joint nullity for the linear regression parameters (H0:β0 = 0 and β1 = 1). The rejection of H0 indicates an absence of similarity between estimates of average accumulated proportions of LR (y) and PR (x), and UL (y) and UP (x), that is, high magnitude occurs in the residual values (εi = y−ŷ).

The significance level of 0.05 was adopted in all the analyses. All statistical analyses were performed using the software R v.3.0.2 (R Core Team 2015).

3 Results

We used recommendations of Spivak and Downey (1998) in A. mellifera for the parameters to compare HB of the colonies of M. q. anthidioides investigated. These authors recommended for hygienic colonies of A. mellifera a removal greater than 95% within 48 h in at least two tests using the method of killing the brood by freezing.

For LR, of 120 measurements, 78 (65%) were above 95% and 30 (75%) of 40 colonies showed, in at least two periods, removal greater than 95% within 48 h. For PR, of 120 measurements, 40 (33%) were above 95% and 6 (15%) of 40 colonies showed, in at least two periods, removal greater than 95% within 48 h. The average number of days for the removal of all larvae killed by freezing was 2.90 and for pupae 5.02, with a range of variation of 0–11 days.

The percentage of LR and PR killed by freezing within 48 h ranged from 0 to 100% in the observation periods 1 and 2 and from 14 to 100% in period 3, with a variation coefficient of 42% for LR and 89% for PR. When considering the variation in the percentage of colonies that removed more than 95% of brood killed by freezing within 48 h in each period, LR ranged from 25% of the colonies in period 1 to 85% in period 3 and PR ranged from 4% of colonies in period 1 to 26% in period 3.

There was a positive effect of time and block for LR (time: χ 2 = 35.91, P < 0.0001; block: χ 2 = 21.03, P < 0.0001). There was a positive effect of time, population and block for PR (time: χ 2 = 38.93, P < 0.0001; population: χ 2 = 4.92, P = 0.0265; block: χ 2 = 27.95, P < 0.0001). There was a positive effect of time for UL (χ 2 = 9.61, P = 0.0019). There was a positive effect of time and population for UP (time: χ2 = 7.18, P = 0.0074; population: χ 2 = 3.57, P = 0.050).

Figure 1 shows the LR and PR regression models in terms of hours after the brood death by freezing. The LR average estimate ranged from 39.97 to 99.64% within the evaluated times (3–264 h). LR average estimates within 24 and 48 h were 51.97 and 65.32%, respectively. Considering an average population of 806 individuals, the PR average estimate ranged from 14.98 to 98.88% within the measured times (3–264 h). PR average estimates within 24 and 48 h were 22.52 and 33.98%, respectively.

Figure 1.
figure 1

Regression models of the accumulated average proportion of larvae removal (LR) and pupae removal (PR) (average population of PR was 806 individuals) of M. q. anthidioides in terms of time values, expressed in hours after brood death by freezing.

Full size image

Figure 2 shows the regression models for UL and UP in terms of hours after the brood death by freezing. Low percentages of UL and UP with brood killed by freezing not yet removed by worker bees were observed. Only 4.81% UL were estimated within 48 h. The highest UL average value estimated occurred within 264 h (8.82%). When the average estimate of 806 individuals was assessed in the colony, the UP estimated percentage within 48 h was 1.44%. The highest UP estimated average value occurred within 264 h (3.21%).

Figure 2.
figure 2

Regression models of the accumulated average proportion of uncapped larvae cells (UL) and uncapped pupae cells (UP) (average population of UP was 806 individuals) of Melipona quadrifasciata anthidioides in terms of time values, expressed in hours after brood death by freezing.

Full size image

The differences between the LR and PR average estimates and the UL and UP killed by freezing over time could be statistically detected, because the rejection of the hypothesis of joint nullity was observed for linear regressions parameters (H0:β0 = 0 and β1 = 1) when the estimated average values of LR in PR and UL in UP were regressed. In both tests, the significance probability value (P) was lower than 0.0001 (P < 0.0001). The estimated 1st degree equations were LR = 0.360 + 0, 682PR (R 2 = 0.9602) and UL = 0.015, +2.317UP (R 2 = 0.9987). It was observed that the inclination angles (inverse tangent of the angular coefficient) of the straight lines for removal and uncapping were 34.29 and 66.66°, different values (P < 0.0001) of 45°, ideal angle, a situation that characterizes the lack of similarity between pairs of x and y values.

When a certain time is kept fixed after the brood death by freezing, PR and UP estimates increase as the population in the colonies also increases (Figures 3 and 4, respectively).

Figure 3.
figure 3

Response surface model of the accumulated average proportion of pupae removal (PR) of M. q. anthidioides in terms of time values (expressed in hours after brood death by freezing) and population of the colonies. PR model: μ = e(−2.4513 + 0.0238T + 0.0008P) / 1 + e(−2.4513 + 0.0238T + 0.0008P).

Full size image
Figure 4.
figure 4

Response surface model of the accumulated average proportion of pupae uncapped (UP) cells of Melipona quadrifasciata anthidioides in terms of time values (expressed in hours after brood death by freezing) and population of the colonies. UP model: μ = e (−5.7781 + 0.0038T + 0.0017P) / 1 + e (−5.7781 + 0.0038T + 0.0017P).

Full size image

4 Discussion

HB sequences observed in unselected colonies of M. q. anthidioides confirm the hypothesis that this behavior is similar to that of A. mellifera. The main difference is that the M. q. anthidioides worker bees destroy brood cells after removing the brood. In addition, although the behavior of partial removal of dead brood in the test areas, which occurs in A. mellifera, was not observed, body parts of the brood were observed in the colony, which highlights that such behavior does also occur in M. q. anthidioides.

Few uncapped larvae cells and uncapped pupae cells killed by freezing were not immediately removed by worker bees, which is important to avoid possible pathogen dissemination in the colonies if the brood were infected.

On average, there was a gradual increase of LR and PR over time, with 75 and 6%, respectively, of the unselected colonies of M. q. antidioides showing removal greater than 95% within 48 h in at least two periods. In A. mellifera, Pérez-Sato et al. (2009) and Bigio et al. (2013) found 3 and 0% of colonies with removal percentage greater than 95% within 48 h, respectively. Medina et al. (2009) observed HB in Scaptotrigona pectoralis and M. beecheii using frozen brood assay and found 100% removal of dead pupae after 2.3 ± 0.6 days and 4.4 ± 2.0 days, respectively. Nunes-Silva et al. (2009) studied HB in Plebeia remota using pin-killed brood and found a 96.4% removal after 48 h. Toufailia et al. (2016) found a removal of freeze-killed brood after 48 h of 99.3% in M. scutellaris, 79.5% in S. depilis and 62% in Tetragonisca angustula.

The workers of M. q. anthidioides seem to achieve LR within 48 h from the examination of seven cells of larvae and seven of pupae, although this small sample may arguably have placed a limitation on our conclusions. Compared with other species of stingless bees, it appears that the removal of dead brood is relatively less effective in M. q. anthidioides, because, during investigations by Medina et al. (2009), Nunes-Silva et al. (2009), and Toufailia et al. (2016), more dead brood were removed within 48 h from a higher number of cells for their respective species. However, none of these studies made distinctions between the stages of brood development. Differences between the bee species for the brood comb diameter, as well as the different methodologies used to investigate HB are factors that hinder direct comparisons between the different studies and our results. Therefore, any comparison between different species of bees for HB should be made with caution. There was variation in HB between the colonies of M. q. anthidioides within each period and between the observation periods. The removal percentage within 6 h, using the method of brood drilling, ranged from 0 to 82% in A. mellifera unicolor (Rasolofoarivao et al. 2015), from 1 to 31% in A. mellifera carnica (Gramacho et al. 1998) and from 0 to 100% in Africanized honeybee A. mellifera using a similar protocol (Fries and Raina 2003). HB is strongly influenced by the time at which the tests are carried out (Uzunov et al. 2014). A. cerana colonies showed different HB in various ecological habitats (Athreya and Reddy 2013). However, it is believed that part of the variation in the observed HB for the colonies between observation periods is due to climatic conditions, although they were not investigated.

Worker bees removed more killed larvae than pupae from brood cells within 48 h and PR takes a longer time than does LR in M. q. anthidioides. In A. mellifera, the behavior used to detect and remove pupae killed by freezing is not necessarily the same as that used to detect and remove mite-infested pupae (Spivak 1996). Differences in olfactory sensitivity between hygienic bees can lead to a partition of the uncapping and removal behavior (Gramacho and Spivak 2003). Pores found in opercula have been reported to affect the effectiveness of olfactory detection of bees (Mathur et al. 2010).

At first, we expected that worker bees removed the dead pupae faster than the dead larvae because of the greater amount of wax and resin in the brood disks for larvae, which does not occur in pupal cells due to its removal by worker bees to facilitate adult emergence. The larger amount of wax and resins in cells with larvae could hinder the workers in detecting dead larvae. However, in M. q anthidioides, the shorter time for LR than for PR was attributed to the brood integument constitution. Landim (2009) found that pupae integument is more complex and sclerotized than larvae integument. Empirical observations showed that a single worker bee of M. q. anthidioides removed dead larvae easily, while for dead pupae, the bees only managed to remove them after tearing the pupa body.

The variation in the estimated population of bees between the colonies did not influence the worker bees in LR and stronger colonies had greater PR. When there is an increase in the colony population, more bees assist in the PR activity of a single cell, reducing removal time.

This is the first study that has investigated HB in M. q. anthidioides. The findings confirmed HB in M. q. anthidioides and showed that its stages are identical to those previously reported for A. mellifera. However, there is a difference in behavior by the worker bees for LR and PR. Also, strong colonies can shorten the time to remove dead pupae in the colony. This information can be useful for genetic improvement programs of M. q. anthidioides colonies on resistance to pests and diseases.