Ideally, we would have liked to collect queens at the beginning of April, 1–2 weeks before B. impatiens queens are normally observed; however, the timing of field work in 2020 had to be shifted due to the onset of Covid-19 pandemic restrictions to field work, and we chose to use the same timing in 2021. In the spring, we used similar methods to determine whether queens had survived over the winter, though dead queens were far more degraded in the spring than they had been in the fall . As queens began overwintering sometime between August and mid-October, there was some variability in the amount of time queens had spent overwintering. However, most queens had been overwintering for about two months when they were checked for the first time and for 6–7 months when they were checked for the second time. For simplicity, we will refer to bee fates at these time points as survival after “two” and “six” months.We estimated diapause survival rates using generalized linear models . Statistical analyses were performed in R, version 4.0.2 . Survival rates were estimated at two time points . At two months, survival was coded based on whether the queen was found live or dead ; we assumed that the site had been abandoned if no queen was recovered when the overwintering site was excavated, as queens may desert burrows if an obstruction is encountered while digging . At six months, we analyzed survival in two ways. Our first analysis used only known-fate individuals, i.e.,flower bucket we coded survival as whether or not the queen was found live in the spring or dead either during the fall or the spring ; queens that were not recovered were treated as missing data. Our second analysis included all individuals, i.e., survival was coded as surviving if found live in spring or as dead if the queen was not recovered in the spring or was found dead in either the spring or fall.

This second way of coding the data leads to a lower estimate of survival and was of interest due to the wide difference between field and lab estimates of survival . To estimate average survival, we ft an intercept-only model to each of the three response variables: survival to two months, survival to six months and survival to six months . We tested whether survival differed among aggregations by fitting a model with aggregation ID included as a categorical predictor variable to each of the three response variables. We evaluated statistical significance using Wald chi-square tests implemented with the Anova function in the package car . We tested if queen body condition and hibernaculum depth differed across aggregations using linear models. Linear models ft to each response variable included only aggregation ID as a predictor. Similarly, we tested if rates of abandonment of hibernaculum differed across aggregations using a univariate generalized linear model . We coded aggregation ID as a predictor variable and the presence/absence of queens during the first excavation of hibernacula in the fall as the response. We evaluated statistical significance using Wald chi-square tests implemented with the Anova function in the package car .To obtain estimates of bumblebee overwintering survival from other studies, we conducted a systematic literature search on January 26th, 2022, using Web of Science and Open Access Theses and Dissertations. A total of 73 research articles and 2 theses/dissertations were obtained using the search terms: AND AND . Of these, we retained 32 studies that provided direct estimates of bumblebee overwintering survival . We scanned the introduction and discussions of each of these retained manuscripts for additional relevant studies, locating an additional 9 papers. We excluded 6 of the studies we obtained that were not published in English. A few authors presented the same data sets in multiple publications; 2 studies were excluded as duplicate data. A final study was excluded because queen survival rates could not be calculated from the data as presented.

Thus, a total of 32 studies were included in our literature review, nearly all of which were conducted in the lab. Nearly all of the studies included in our meta-analysis tested several different diapause regimes and reported queen survival rates at monthly intervals. Thus, for each set of experimental conditions reported by each study, we recorded 1) the proportion of queens to survive, 2) the sample size, 3) the length of the diapause regime, 4) the species, mating status, origin , and approximate age of queens used in each study, 5) the temperature and relative humidity that the queens were exposed to during diapause, and 6) any other details related to the experimental design . We assumed that all queens were alive at the start of each experiment; thus, we recorded survival rates of 100% for each group of queens at month 0. For papers that did not report monthly survival estimates directly in the text of the manuscript, we used the digitize package in R to extract data from figures. For studies that monitored diapausing queens continually and reported the length of time queens survived rather than monthly survival rates , we estimated monthly survival rates manually. Prior to statistical analysis, we converted survival rates recorded from manuscripts to a binomial data set of the number of successes and failures from sample sizes and percent survival at each time interval. Many of the studies included in this literature review performed treatments on queens that we thought might lead to lowered rates of survival: inoculating queens with parasites, exposing queens to chemicals, starving queens, etc. We excluded all data from treatments we deemed obviously harmful, and included only measurements for queens that were overwintered in continuous darkness at constant temperatures between 1 and 5 °C . Queens from excluded treatments had lower survival than those included in our meta-analysis . We used a generalized linear mixed-effects model , with queen survival coded as the response to estimate diapause mortality rates. GLMMs were ft using the command glmer in R package lme4 .

Our model included diapause interval as a fxed effect and random slopes for both study species and paper ID . As a basic check of model ft, we used a linear model to compare the observed and the predicted values, and estimated confidence intervals for the slope and the intercept of this model. Diapause is an important and often under-studied feature of insect populations. Our study suggests that estimates of bumblebee survival during diapause in the lab are lower than survival in the wild. Specifically, past laboratory studies of bumblebee diapause indicate that many queens are unable to survive the natural length of diapause . In contrast, we observed that more than 60% of B. impatiens queens monitored in the field survived a 6-month period of diapause. This estimate is higher than predicted for any laboratory study of B. impatiens, as well as most studies of other bumblebee species. There were a handful of studies that achieved high rates of survival after 6 months , all of which obtained queens from field colonies. Unlike the other labbased studies included in our review, Holm and allowed B. terrestris and B. lapidarius queens to dig themselves into containers of soil and other substrate placed in an outdoor greenhouse. Similarly, Milliron placed B. fervidus queens in large containers of heartwood, providing an opportunity for some queens to bury themselves in the substrate. Our field estimates of survival are also comparable to those of Pouvreau ,square flower bucket who used methods similar to Holm and reported similarly high survival rates for queens after the natural length of diapause. In combination with the findings of overwintering studies under semi-natural conditions, the results of our field study suggest that surviving diapause may be less of an ecological hurdle for queen bumblebees than previously indicated. In many ways, it is surprising that we observed higher rates of diapause survival for overwintering queens in the field compared to the lab. Among other insect species, diapause can be a period of high risk . In temperate regions, bumblebees must cope with less-than ideal temperatures in the winter; some queens may encounter additional environmental challenges during hibernation, including pesticide exposure and pathogen infection . The fact that we observed generally high rates of diapause survival may be related to the location of our focal aggregations: Appleton Farms and Grassrides is located well within the geographic range of B. impatiens, which is found throughout the Eastern United States as well as parts of Southern Canada . We also measured survival only in forests, a habitat type in which colonies produced more new queens than in meadows . It is possible that the environmental stressors experienced by bumblebees during diapause are heightened in marginal habitat types. For example, in areas near the edges of their climatic ranges, bumblebee populations are more vulnerable to decline . The northward expansion of at least one other Bombus species, B. haematurus, has been linked to warmer winter temperatures . Given that we monitored aggregations at one study site, using similar methods to monitor diapause survival rates of B. impatiens in other regions of the United States could be a valuable area of future research. At this point, it is unclear why survival rates of queens overwintered in the laboratory were generally lower than our field estimates. It may be that laboratory conditions are stressful for queen bumblebees; maintaining study organism in artificial settings can be challenging, and the ability of study subjects to thrive in the lab can depend greatly on methods of husbandry implemented by researchers . Among-study variation in survival rates was high, suggesting that some researchers were better than others at maintaining queens in the artificial settings. It is also possible that study subjects used in lab and field experiments differ phenotypically. Insect populations maintained in the lab are known to adapt to artificial conditions ; among other insect taxa , laboratory rearing has been observed to disrupt the ability of other insect species to enter diapause .

Bumblebee queens used in diapause experiments are typically obtained from colonies purchased from commercial suppliers or from colonies reared in the lab . As bumblebee queens can be induced to bypass diapause using carbon dioxide narcotization , it is possible that the selection on commercial bumblebees to undergo long periods of diapause has been relaxed. To date, no study has addressed the impacts of captive rearing on the ability of queen bumblebees to complete diapause. However, other authors have noted differences in the morphologies of wild and lab reared bumblebees , which may impact their ability to undergo diapause . Understanding how outcomes of artificial diapause are impacted by the origin of study subjects and by other husbandry strategies would be a valuable area of future research. Until best practices are better defined, we recommend that researchers leverage the methods of authors who have had relatively high rates of success in maintaining diapausing queens , i.e., by obtaining queens from field colonies and providing queens an opportunity to excavate their own hibernacula. Our field observations broadly corroborate a pattern observed by the many lab-based studies of queen diapause survival , in that survival rates of queens were dependent on colony origin/ aggregation ID. For individuals that originate from the same colony, similarities in survival rates could be explained by a multitude of factors, including relatedness, environment during development, nutritional status, and body condition of adult queens . In the field, differences in environmental conditions around nests could have also contributed to differences in survival rates.Compacted soils may have limited the ability of queens to dig their hibernacula, as queens dug themselves less deeply in the soil and greater number of hibernacula were abandoned by queens at these aggregations. These aggregations also had smaller queens, as measured by both IT span and body mass. Due to low sample sizes, we were unable to statistically separate effects of colony origin and the other ecological or environmental factors which might have influenced diapause survival . In this context, our research points to at least one major advantage of monitoring queen vital rates in the lab, in that extraneous sources of variation can be controlled in the lab , making identifying the environmental factors which impact queen vital rates in the field more straightforward.