Varroa destructor and deformed wing virus interaction increases incidence of winter mortality in honey bee colonies

Abstract

Winter mortality of honey bee colonies represents a major source of economic loss for the beekeeping industry. The objectives of this prospective study were to estimate the incidence risk of winter colony mortality in southwestern Quebec, Canada and to evaluate and quantify the impact of the associated risk factors. A total of 242 colonies from 31 apiaries was selected for sampling in August 2017. The presence of Varroa destructor, Vairimorpha (Nosema) spp., Melissococcus plutonius, deformed wing virus (DWV), and viruses of the acute-Kashmir-Israeli complex (AKI complex) was investigated in each colony. Management practices of the various colonies were obtained from a questionnaire. The incidence risk of colony mortality during the winter of 2017–2018 was estimated to be 26.5% [95% confidence interval (CI): 15.4 to 40.3]. In logistic regression modeling of winter mortality in colonies, an interaction was discovered between V. destructor and DWV; the detection of ≥ 1 V. destructor mites per 100 bees was associated with higher odds of mortality (3.46, 95% CI: 1.35 to 8.90) compared to colonies with < 1 mite per 100 bees, but only in DWV-positive colonies. There were more colony losses in apiaries from beekeepers owning 1 to 5 colonies than in apiaries from beekeepers owning over 100 colonies, which suggests that beekeeper experience and/or type of management are important contributors to winter colony mortality. Assuming a causal relationship, the results of this study suggest that up to 9% of all colony mortalities in the population could have been prevented by reducing the level of V. destructor to < 1 mite per 100 bees in all colonies.

Introduction

Winter mortality of honey bee colonies is a major cause of stress and economic loss for beekeepers and leads to increased efforts to maintain production in the subsequent year. The Canadian Association of Professional Apiculturists (CAPA) suggests a long-term objective of less than 15% for winter losses (1). In Canada, winter mortality has consistently exceeded that threshold since 2007, averaging 26% per year at the country level and 23% per year for the province of Quebec over the last 15 y (2).

Winter mortality is the final result of a complex phenomenon. Many infectious agents are considered to contribute to bee mortality, both on their own and through interactions with other pathogens or environmental factors (3–5). In particular, Varroa destructor has been identified as a major cause of colony losses around the world (6–8). An Ontario study concluded that varroosis, which is associated with > 85% of mortality in 400 colonies, was the leading cause of winter mortality in this Canadian province (9).

Other incriminated pathogens include Vairimorpha (Nosema) ceranae; deformed wing virus (DWV); acute bee paralysis virus (ABPV), Kashmir bee virus, and Israeli acute paralysis virus (IAPV), all of which make up the AKI complex; Paenibacillus larvae (American foulbrood); and Melissococcus plutonius (European foulbrood) (4,5,8,10).

Beekeeping management practices and migratory beekeeping may also have various impacts on bee health, through oxidative stress, food scarcity, or pathogens (11,12). In eastern Canada, low-bush blueberry pollination reduces brood population because of nutritional pollen deficiency and cranberry reduces weight gain in colonies due to a lack of nectar availability (13). A Canadian study conducted in the province of Manitoba reported a detrimental impact of outdoor wintering of colonies compared to indoor wintering on bee population score, even though pathogen loads were similar in the 2 groups (14).

A study conducted in southwestern Quebec, Canada in August 2017 reported that V. destructor, Vairimorpha (Nosema) spp., Melissococcus plutonius, and some other viruses were highly prevalent in colonies (15). The objectives of the present study were to: i) estimate the incidence risk of winter colony mortality, and ii) evaluate and quantify the impacts of the risk factors for winter mortality at various levels (colony, apiary, and operation), with focus on colony status for these major bee pathogens at the start of the study and management factors.

Materials and methods

Study design and sampling of apiaries

A prospective study was conducted from a random sample of 242 hives, 31 apiaries, and 15 beekeepers initially selected for a prevalence study (15). Sampling was conducted from August 7 to September 1, 2017 in 2 zones located within the regional county municipalities of Vaudreuil-Soulanges, Haut-Saint-Laurent, Jardins-de-Napierville, Haut-Richelieu from the Montérégie and Pontiac regions (15). The average minimum and maximum monthly temperatures and total precipitation over the study area are shown in Figure 1 (16).

Figure 1.

Average minimum and maximum monthly temperatures and total precipitation in the study area in relation to sampling for pathogens and evaluation of winter mortality for a cohort study on winter bee mortality in southwestern Quebec, Canada in 2017, estimated from thin-plate spline-based historical climate models generated by Natural Resources Canada (16).

If there were 10 hives or less in a selected apiary, all colonies were sampled; otherwise, 10 hives were systematically selected for sampling. In this case, starting from a randomly selected colony, the other chosen colonies were evenly distributed over an ordered list of all colonies based on their location in the apiary, using a fixed interval for selection.

If the selected colonies were not uniquely identified, a tag with a unique identification number was stapled under the top cover of the hive and this tag was kept for follow-up. Colonies in which more than 30% of the surface of the brood frames was covered by bees were considered strong; otherwise, colonies were considered weak.

The 3 oldest looking brood frames of the hive were thoroughly examined for the presence of dead larvae and scattered brood. Samples for pathogen detection were collected. Samples were either kept in alcohol or frozen until the time of analysis, as described in the previous study (15).

During their visit, the research team asked beekeepers whether the colonies were managed as a sedentary apiary, i.e., no movement of the colonies during the current production season, or were moved for transhumance, e.g., to provide pollination services or to collect a specific type of honey.

The research team also evaluated the number of colonies in the apiary and the immediate surrounding land cover during their visit. The “presence of trees” referred to hives located on a wooded lot, “by tree line” referred to hives placed on a cleared lot adjacent to woodland, and “no trees” referred to hives on a cleared lot.

Pathogen detection

Laboratory testing for pathogen detection was conducted as described in the previous study (15). Briefly, V. destructor mites were counted on approximately 300 nurse bees per colony using the alcohol wash technique. Vairimorpha (Nosema) spp. spores were counted in the guts of 60 foraging bees per colony. A total of 5 bees from spore-positive colonies were tested with polymerase chain reaction (PCR) to identify the Vairimorpha (Nosema) species present.

For the culture of M. plutonius (European foulbrood), a homogenate from 30 nurse bees per colony was plated on basal media plates and incubated anaerobically for 7 d at 35°C, as described in a previous study (17). Melissococcus plutonius was then identified by PCR on bacterial cultures. Deformed wing virus (DWV) and viruses in the AKI complex were identified by real-time PCR on a sample of 10 adult bees.

Questionnaire

A questionnaire on wintering practices in 2017–2018 was developed and pre-tested by 2 professional beekeepers (18). The questionnaire covered the total number of colonies owned by the beekeeper during the summer of 2017, the 2017–2018 wintering method (indoors versus outdoors) for each apiary, as well as the outcome (winter survival or mortality) of each selected colony.

The beekeepers were asked to evaluate colony mortality using the following specific definition for mortality: a colony with fewer than 4 frames covered at 75% by bees as seen from above on May 15, 2018. This definition included dead and nonviable colonies. The questionnaire was administered during a telephone interview with beekeepers from May 15 to May 28, 2018.

Statistical analyses

All statistical analyses were carried out using SAS software, Version 9.4 (SAS Institute, Cary, North Carolina, USA). The incidence risk of winter mortality with modified Clopper-Pearson exact confidence intervals (CI) was estimated with the SURVEYFREQ procedure, considering strata by region, clusters by apiaries, and sampling weight.

Logistic regressions were used to model the association between winter mortality of colonies as the outcome according to the detected pathogens and clinical signs (Table I). All models were built with the GENMOD procedure, accounting for clustering by apiaries with a compound symmetry covariance structure. For V. destructor, the recommended threshold of 1 mite/100 bees for treatment in late summer in Quebec was used for variable categorization (19). For Vairimorpha (Nosema) spp. and Vairimorpha (Nosema) ceranae, categories were defined based on the reported distribution (minimum, mid-range, and 95% percentile) of spore counts per bee in these apiaries (15). Each variable was first screened using univariable models. As they were biologically expected, the 2-way interactions between V. destructor and each virus (DWV, AKI complex viruses) were also tested in separate models that only included the 2 variables and their interaction. Variables with P < 0.1 in the univariable models, as well as variables involved in 2-way interactions with P < 0.1, were included in a full multivariable model. A backward selection procedure was then carried out to keep only the variables with a P-value of < 0.05 and those involved in a significant interaction (P < 0.05) in the model.

Table I.

Descriptive statistics and odds ratios from univariable logistic regression modeling of winter mortality of honey bee colonies according to the detection of pathogens and clinical signs in colonies from 31 apiaries in southwestern Quebec in 2017.

Characteristics (number of coloniesa)	Number of colonies	Percent mortality (%)	Odds ratio		P-value
			Estimate	95% CI
V. destructor (n = 242)					0.046
0 mites/100 bees	108	30	Ref.
> 0, < 1 mite/100 bees	81	25	0.77	0.43 to 1.37
≥ 1 mites/100 bees	53	42	1.63	0.85 to 3.13
Vairimorpha (Nosema) spp. (n = 242)					0.43
0 spores/bee	154	29	Ref.
> 0, < 0.5 × 106 spores/bee	56	30	1.12	0.62 to 2.05
≥ 0.5 × 106, < 1 × 106 spores/bee	19	37	1.47	0.75 to 2.87
≥ 1 × 106 spores/bee	13	46	2.17	0.73 to 6.44
Vairimorpha (Nosema) ceranae (n = 224)b					0.74
0 spores/bee	156	30	Ref.
> 0, < 0.5 × 106 spores/bee	45	29	0.96	0.49 to 1.92
≥ 0.5 × 106, < 1 × 106 spores/bee	15	40	1.29	0.59 to 2.84
≥ 1 × 106 spores/bee	8	50	2.02	0.51 to 8.06
M. plutonius (n = 242)					0.48
Not detected	185	32	Ref.
Detected	57	25	0.79	0.40 to 1.56
DWV (n = 182)					0.51
PCR-negative	121	25	Ref.
PCR-positive	61	38	1.31	0.59 to 2.93
AKI (n = 182)					0.82
PCR-negative	163	28	Ref.
PCR-positive	19	42	1.09	0.49 to 2.44
Strength in August (n = 239)					0.31
Strong	206	30	Ref.
Weak	33	33	1.36	0.75 to 2.49
Dead larvae (n = 241)					0.37
No	212	29	Ref.
Yes	29	42	1.41	0.66 to 3.03
Scattered brood (n = 241)					0.32
No	205	29	Ref.
Yes	36	39	1.48	0.68 to 3.25

^aNumber of colonies with non-missing data for each characteristic. Of the 242 colonies selected for the study, 182 were tested for viruses.

^bExcluding cases in which the species was not identified.

CI — Confidence interval; DWV — Deformed wing virus: AKI — Acute-Kashmir-Israeli complex; Ref. — Reference category.

For pathogens identified as significant risk factors, the population-attributable risk and population-attributable fraction were calculated. The formulas described by Dohoo et al (20) were used, using the estimated predicted risks from the models for the exposed and unexposed groups, the estimated incidence risk of colony mortality, and the estimated prevalence of pathogens from our previous prevalence study (15).

We modeled the risk factors for winter colony loss at the apiary level using negative binomial regressions with the GENMOD procedure, with the number of dead colonies per apiary as the outcome variable. The assessed risk factors were transhumance (yes or no), wintering method (outdoors or indoors), land cover (trees, by tree line, or no trees), and number of colonies in the apiary (1 to 5, 6 to 25, or 26 to 39). The number of followed-up colonies per apiary was included as an offset. All variables were screened using univariable models. A full multivariable model was built, including variables with P < 0.1, followed by a backward selection as described for the colony-level model.

At the beekeeper level, the association between the percentage of winter mortality in sampled colonies and the categorized number of beekeeper-owned colonies was assessed using a Kruskal-Wallis exact test, followed by post-hoc pairwise comparisons in the presence of a global association. The association between the percentage of winter mortality in sampled colonies and in the region was also assessed using a 2-sided Wilcoxon exact test. The NPAR1WAY procedure was used.

Results

Incidence risk of winter mortality

Winter mortality was reported in 74 of the 242 sampled colonies. Median intra-apiary mortality in sampled colonies was 30%, ranging from 0 to 100%. At the beekeeper level (n = 15), the median mortality in sampled colonies was 37%. The incidence risk of winter mortality in colonies for 2017–2018, adjusted for the sampling design, was estimated at 26.5% (95% CI: 15.4 to 40.3).

Risk factors at the colony level

Varroa destructor was the only statistically significant variable in univariable analyses (Table I). As no significant difference was noted between V. destructor-negative colonies and V. destructor-positive colonies with less than 1 mite (OR = 1.63, 95% CI: 0.85 to 3.31), these 2 categories were merged for multivariable regressions. Colonies with ≥ 1 mites per 100 bees had higher odds of winter mortality [odds ratio (OR) = 2.13, 95% CI: 1.31 to 3.46] than V. destructor-positive colonies with less than 1 mite.

In the multivariate model, the effect of the interaction between V. destructor and DWV on colony mortality was statistically significant (P = 0.034). The DWV-positive colonies with ≥ 1 Varroa mites per 100 bees had 3.46 higher odds of mortality (95% CI: 1.35 to 8.90) than DWV-positive colonies with less than 1 mite per 100 bees. No other pairwise comparison between categories of the 2 variables was statistically significant. The predicted values of mortality illustrating this interaction are presented in Table II.

Table II.

Odds ratios and predicted probabilities from logistic regression modeling of winter mortality of honey bee colonies according to V. destructor level, DWV detection, and their interaction in 182 colonies from 28 apiaries in southwestern Quebec in 2017.

Characteristics	Number of colonies	% of mortality	Odds ratio		P-value	Predicted mortality (%)
			Estimate	95% CI		Estimate	95% CI
DWV-negative
V. destructor
< 1 mite/100 bees	95	24	Ref.			28a,b	18.6 to 38.7
≥ 1 mites/100 bees	26	27	1.34	0.70 to 2.58	0.25	34a,b	18.8 to 52.9
DWV-positive
V. destructor
< 1 mite/100 bees	34	24	Ref.			25b	13.4 to 41.3
≥ 1 mites/100 bees	27	56	3.46	1.35 to 8.90	< 0.01	53a	32.6 to 72.9

^a,bMortality percentages with different superscripts are statistically significant (P < 0.05) in post-hoc pairwise comparisons with Tukey-Kramer adjustment for multiple comparisons.

CI — Confidence interval; DWV — Deformed wing virus; Ref. — Reference category.

The population-attributable fraction and population-attributable risk were estimated using predicted probabilities of mortality from the model including only V. destructor, which was 44.3% for colonies with ≥ 1 mites per 100 bees and 29.4% for colonies with < 1 mite per 100 bees. The population-attributable fraction was 9%, which estimates the proportion of all colony mortalities in the population that could have been prevented by reducing the level of V. destructor to < 1 mite per 100 bees in all colonies, assuming a causal relationship. The population-attributable risk was estimated as 2.4%, which represents the reduction in the incidence of mortality in the population that could have been similarly prevented, i.e., a reduction of winter colony mortality from 26.5 to 24.1%.

Risk factors at the apiary level

As the “number of colonies in the apiary” was the only selected variable from univariable analyses (P < 0.1), no multivariable model was conducted. None of the tested variables had an overall statistically significant effect on winter mortality of colonies at the apiary level (Table III).

Table III.

Descriptive statistics and odds ratios from univariable binomial regression modeling of winter mortality according to characteristics at the apiary level (31 apiaries in southwestern Quebec in 2017).

Characteristics	Number of apiaries	Mean colony mortality in apiaries	Incidence risk ratio		P-value
			Estimate	95% CI
Transhumancea					0.23
No	25	37	Ref.
Yes	4	50	1.69	0.72 to 3.95
Wintering method					0.15
Indoors	11	24	Ref.
Outdoors	20	46	1.63	0.83 to 3.18
Land cover					0.14
Trees	2	10	Ref.
By tree line	21	37	2.92	0.58 to 14.7
No trees	8	46	4.65	0.88 to 24.6
Number of colonies in apiary					0.10
26 to 40	10	22	Ref.
6 to 25	11	33	1.61	0.76 to 3.41
1 to 5	10	58	2.79	1.12 to 6.94

^aExcluding 2 colonies with missing data.

CI — Confidence interval; Ref. — Reference category.

Risk factors at the beekeeper level

The more colonies that were owned by the beekeeper, the lower the probability of winter mortality in their colonies (Table IV). In post-hoc comparisons, the median colony mortality was significantly higher for beekeepers who owned 5 colonies or less than for beekeepers with more than 100 colonies (P = 0.02). No association was observed with the region where the beekeeper’s apiaries were located (Table IV).

Table IV.

Median winter mortality of colonies according to characteristics at the beekeeper level (15 beekeepers, southwestern Quebec, 2017).

Characteristics	Number of beekeepers	Median colony mortality for beekeepers	P-value
Number of colonies owned
1 to 5	6	58	0.038
6 to 100	6	35
> 100	3	10
Region
Pontiac	3	33	0.87
Montérégie	12	38

Discussion

We observed a 26.5% winter mortality in bee colonies in southwestern Quebec. This result is similar to the 30.7% winter mortality reported among all Quebec beekeepers owning more than 50 colonies during the same 2017–2018 winter (1). Using the same definitions, the overall winter mortality of colonies in Quebec varied from 15 to 30% in the last decade (2).

We identified V. destructor as a significant contributor to winter losses, as has previously been reported in the neighboring province of Ontario and in other countries (3,6–9,21).

After sampling took place in August, all beekeepers participating in this study treated their colonies against V. destructor using different acaricides, but did not necessarily follow recommended protocols (18). Despite these treatments, a higher V. destructor level in August was a risk for winter mortality. This agrees with the results of an experimental study that suggest that treating for V. destructor in early fall is too late to prevent winter losses caused by the parasite due to early physiological alterations in bees during winter (22).

Overall, our results suggest that improved control of V. destructor earlier in the summer could have prevented 9% of observed winter colony losses, assuming a causal relationship. This figure is relatively low, as only 16% of colonies had ≥ 1 Varroa mites per 100 bees. It is noteworthy that the predicted winter mortality in colonies with low (< 1) mite infestation was quite high at 29.4%, which suggests that there were other important, but undetected factors, involved in the observed winter losses.

Neither DWV nor AKI complex viruses were associated with winter mortality on their own. This differs with previous studies reporting that the presence or load of DWV has an impact on winter mortality (4,14,23). The presence of DWV from November to January was associated with a reduction of the life span of bees during winter, thus possibly contributing to overwintering losses (7). The viral load of DWV might not have been sufficient to directly cause winter colony mortality in our study. As recently reported, differences in the regional distribution and possibly virulence among DVW genotypes, mainly DVW-A versus DVW-B, might also be involved in conflicting results among studies (24). Although the Israeli acute paralysis virus, which is the only virus of the AKI complex identified by sequencing in this study (15), has also previously been linked to the colony collapse disorder phenomenon in the United States (25), a subsequent study found no evidence supporting this association (26).

Although no impact of DWV by itself was noted on winter mortality, its presence combined with V. destructor was detrimental (Table II). This interaction concurs with previous evidence that V. destructor supports DWV replication and suppresses the immunity of the parasitized bee (27). Furthermore, in the United States, Spain, and Denmark, the severity of DWV infection was positively correlated with Varroa mite infestation levels (28–31). The origin and age of the queen bee might also impact both the presence of a virus (through genetic resistance or transovarial transmission) and colony survival (32). These potential confounding variables were not accounted for in our model.

Most samples positive for Vairimorpha (Nosema) were infected with V. ceranae only, which has been associated with colony depopulation (33). Even though a monotonic relationship was noted between the percentage of winter mortality and categorized spore counts, no significant association was found between samples positive for Vairimorpha (Nosema) and the risk of colony mortality. This absence of association is consistent with results from the German bee monitoring project (3). Similarly, no negative impact of V. ceranae infection on winter mortality was reported in Ontario when using the threshold of 10⁶ spores/bee, despite the detrimental effects observed on honey bee populations (34). The impact of V. ceranae on colony health and colony mortality is still poorly understood (35).

Colony strength, dead larvae, and scattered brood observed in August were not associated with winter mortality. Even if colony strength was positively associated with higher Vairimorpha (Nosema) spp. spore counts and scattered brood was associated with the presence of M. plutonius at time of sampling (15), neither Vairimorpha (Nosema) spp. nor M. plutonius were determined to have an impact on winter mortality to our knowledge.

We did not identify any risk factors for winter mortality at the apiary level. It has been reported that land use significantly affects winter colony mortality (36). Other studies in northern temperate climates have reported greater losses in colonies wintered outdoors than those wintered indoors (14,37). In the present study, the mean colony mortality was numerically higher in apiaries using transhumance and outdoor wintering methods. Due to the limited sample size, these results should be considered as preliminary and further investigation is warranted.

The number of colonies that a beekeeper owned was an important risk factor for colony mortality. Annual surveys conducted in Quebec have reported this trend over the years (2). The distribution of Quebec’s population of beekeepers at the time of this study is worth noting: about 50 large beekeeping operations own 80% of Quebec’s bee colonies, whereas the remaining 20% of colonies are owned by more than 1200 beekeepers with only a few colonies.

The number of colonies owned by beekeepers is related to whether they are hobbyists or professionals, their amount of experience, and management techniques (18). European studies have shown that the size of the operation and the apiaries, as well as the beekeeper’s experience, significantly affected overwinter losses (5,20,36,38). A reduction of V. destructor infestation, viral infection, and colony mortality was observed under a best management practice regimen among backyard beekeepers in the US (39).

Little information is available on the performance of diagnostic tests for bee pathogens. The ability of these tests to detect pathogens in a colony can be influenced by the level of infection and the strength of the colony at the time of sampling. This lack of information limits any adjustment for potential misclassification bias and further research on the subject is required. Our assessment of colony strength, as well as the evaluation of winter mortality by beekeepers, both of which relied on visual approximations, are also prone to misclassification. Finally, the status of pathogens was evaluated in August, whereas the outcome was measured the following spring. Both the population structure in a bee colony and bee-pathogen interactions vary highly with time (40). The level of infection in a colony could have changed between these 2 seasons, either naturally or after the beekeeper intervened, especially by applying treatment, and this variation was not taken into account. Other likely contributors to colony losses, such as exposure to pesticides or availability of high-quality food, might have played a role in the observed associations, but were overlooked. Finally, at the apiary and beekeeping operation levels, the sample size was relatively small, which limits the statistical power of the study.

In conclusion, 26.5% of the colonies that appeared healthy in late summer did not survive the following winter. The presence of at least 1 Varroa mite per 100 bees in DWV-infected colonies increased the risk of winter mortality, which emphasizes the importance of controlling this parasite throughout the season. There was more winter mortality in colonies operated by beekeepers who owned less than 5 colonies. Our results suggest that there were other important factors that went undetected, which warrants further investigation.

Funding Statement

This study was made possible by a research grant from the Fonds du centenaire of the Faculty of Veterinary Medicine at the Université de Montréal and scholarships provided to Gabrielle Claing by the Fonds de recherche du Québec — Nature et Technologies (FRQNT) and the J.A. DeSève and Bank of Montreal foundations through the Faculty of Graduate Studies at the Université de Montréal.