Winter survival in dormant animals depends on conserving finite energy reserves, yet winter temperatures fluctuate around shifting means. In ectotherms, metabolic rate increases exponentially with temperature, so thermal variability is expected to accelerate energy loss, with consequences for overwinter survival under climate change. Whether dormant ectotherms can compensate physiologically, however, remains unclear. We overwintered Bombus impatiens queens under constant (2, 3, 4°C) or variable (2 ± 6°C or 4 ± 6°C) temperature regimes for six weeks, then measured metabolic rates across the temperature range. Thermal experience altered the temperature dependence of metabolism, but responses depended on mean temperature: variability centered on 2°C elevated metabolic rate and increased thermal sensitivity relative to all other conditions, whereas variability centered on 4°C reduced metabolic rate and dampened thermal sensitivity relative to constant 4°C. Using these responses, we simulated subsequent lipid depletion from a standardized energetic state. Variability around 2°C was predicted to shorten the time to lipid depletion, whereas variability around 4°C produced trajectories similar to those predicted for constant 4°C. Thus, identical thermal variance produced opposite physiological and predicted energetic outcomes depending on the mean temperature around which fluctuations occurred. Integrating both temperature means and variability is, therefore, essential for predicting overwintering energetics in a changing world.

Animals

We kept two commercial Bombus impatiens colonies (Natupol Excel, Koppert Biological Systems Inc., Howell, MI, USA) under controlled conditions in incubators (28°C, 12:12 light-dark cycle, 63% RH; I36VL, PERCIVAL Scientific, Inc., Perry, IA, USA) and supplied them with ground pollen (BZ Bodies, Clovis, CA, USA) and ad libitum access to Koppert’s supplied nectar.

Experimental Setup

We marked newly produced, callow bumble bee queens (easily identified by their silvery pile and curled wings) with color-numeric tags within two days of eclosion, allowing us to track and standardize age across treatments, and returned them to their natal colonies. After marking 120 queens, we housed them in colony-specific boxes supplied with ad libitum nectar and returned them to incubators. Queens were 14-21 days old when we induced dormancy by gradually cooling the incubators from 28°C to 4°C and shortening the photoperiod to 8:16 L:D over seven days, followed by four days at 4°C in darkness (Waybright & Dillon, 2024). This protocol produced an overwintering-like dormant state characterized by very low activity, survival for six weeks without food, and significantly reduced metabolic rates consistent with previous measurements of dormant queens using similar or identical induction protocols (Rondot & Darveau, 2024; Waybright & Dillon, 2024). We then placed queens individually in 5-dram plastic vials (Thornton Plastics, Salt Lake City, UT, USA); each vial was connected via holes in the lid to a second 5-dram vial filled with dampened sphagnum moss to maintain humidity. Queens were haphazardly assigned to one of five temperature treatments, maintained in incubators: constant 2°C (2.01 ± 0.47°C), constant 3°C (3.0 ± 0.27°C), constant 4°C (4.02 ± 0.54°C), or two oscillating treatments. The first oscillating treatment ranged from -1 to 11°C (“4 ± 6°C”; 4.14 ± 5.74°C), while the second ranged from -5 to 7°C (“2 ± 6°C”; 1.77 ± 6.51°C; see Fig. 1C, Supplemental Figure 3). Because queens came from only two source colonies, ‘colony’ was not included as a random effect. This limits our ability to separate colony-level variation from other sources of biological variation, although haphazard assignment of queens among treatment groups should have minimized systematic effects of colony on treatment comparisons.

Metabolic Test Temperatures

After six weeks in the overwintering treatments, we measured the metabolic rate of each queen (n = 120; 24 per treatment) at each of four test temperatures (480 total metabolic measurements): -5, 2, 4, and 11°C. These temperatures represented mean holding temperatures and the extreme of the oscillating treatments (Fig 1C). Test-temperature order was randomized. After each measurement, queens were reweighed and moved directly into acclimation for the next test temperature rather than being given a separate recovery period. We monitored temperature during metabolic trials with T-type thermocouples connected to loggers recording every 10 s (Onset HOBO 4-channel thermocouple data logger, Model: UX120-014M; Onset Computer Corporation, Bourne, MA, USA).

The 2, 4, and 11°C measurements were conducted in incubators (Percival Scientific, Perry, IA, USA), and the -5°C measurements were conducted in a chest freezer outfitted with a temperature controller (TIC-308, INKBIRD Tech. C.L., Luohu District, Shenzhen, Guangdong Province, China). We chilled the air used to flush test chambers by first flowing it through copper coils (internal diameter = 7.82 mm, outer diameter = 9.6 mm, length = 1.25 m) submerged in large containers of windshield fluid (3.5 L per container, non-freezing) kept in each chamber. This chilled the air appropriately before it reached a homerun manifold through which we flushed 12 syringes simultaneously (10 containing queens and two left empty to serve as calibration measurements before and after all 10 queens in a trial).

Respirometry

Each 30-mL syringe had a small hole drilled near the plunger opening and a two-way valve attached to the tip. By opening all two-way valves connected to the manifold and drawing the plungers out past the drilled holes, we flushed all syringes for two minutes with dry, CO₂-free air (2 L min^-1) from an FT-IR Purge Gas Generator (Parker-Hannifin Corporation, Cleveland, OH, USA) conditioned at chamber temperature.

After flushing, we depressed each plunger past the drilled hole, closed each valve to seal the syringe, and noted the initial air volume. We conducted preliminary trials to determine the maximum duration we could seal bees without inducing hypoxia or hypercapnia, while still ensuring a measurable respiratory signal. For example, at 2°C, syringes sealed for two hours had oxygen depletion less than 2% and CO₂ accumulation below 1600 ppm. At 11°C, we limited sealing time to 60 minutes to avoid exceeding these thresholds. Based on these trials, we sealed syringes for 90 minutes at -5, 2, and 4°C, and for 60 minutes at 11°C.

We measured CO₂ production and O₂ consumption using stop-flow respirometry with an integrated respirometry system (FMS-3; Sable Systems International Inc., Las Vegas, Nevada, USA). We zeroed (using pure N₂ gas) and spanned the CO₂ and O₂ sensors daily with a primary standard (1011.11 ppm) and outside air  (assumed value: 20.95%), respectively. For measurements of syringe gases, we flowed nitrogen (N₂) from a gas cylinder through a column of DRIERITE–Ascarite II–DRIERITE to scrub residual water vapor and CO₂. Flow rate was regulated at 150 mL min^-1 by the FMS-3’s built-in mass-flow controller before the dry, CO₂-free air passed sequentially through the flow meter, CO₂ sensor (fractional concentration), and O₂ sensor (%). Air samples were injected into a three-way valve located between the flow meter and the WVP sensor. To prevent dilution effects from residual moisture or CO₂ from each syringe injection (Lighton, 2018), we plumbed a 30-mL column of magnesium perchlorate (Mg(ClO₄)₂) before the CO₂ sensors to remove water vapor, followed by a 30-mL column of Ascarite II–Mg(ClO₄)₂ between the CO₂ and O₂ sensors to scrub CO₂.

We first injected an empty chamber sealed just prior to the first chamber holding an animal, followed by sequentially injecting all 10 chambers containing animals, followed by an empty chamber sealed immediately after the last animal chamber. For each injection, we slowly injected 10 mL of syringe air into the N₂ stream via the three-way valve.

Respirometry analyses

We determined the syringe seal time as the fractional hours from sealing to injection (± 2 seconds). We adjusted the initial and final volumes for each queen’s body volume, estimated from bee mass measured immediately before respirometry and an assumed tissue density of 1 g cm^-3. Injection periods extended from the start of the injection to the opening of the three-way valve for the next injection, with the final peak ending 2.5 minutes after injection (the time-constant for washout, given the volume of the respirometry setup and flow rate, was approximately 2 minutes). We corrected the baseline drift of both O₂ and CO₂ traces using the slope between the gas concentrations before and after each peak. We converted the drift-corrected values to fractional proportions, multiplied by flow rates (mL min^-1), divided by 60 to convert to mL sec^-1 for second-by-second readings, integrated across each peak, and adjusted for injected volume and syringe seal time to calculate whole-organism O₂ consumption (V̇O₂)  and CO₂ production (V̇CO₂). We calculated the respiratory exchange ratio (RER) as V̇CO₂/V̇O₂.  The RER describes the ratio measured at the organismal level and may differ from the tissue-level respiratory quotient under non-steady-state conditions.

We report metabolic rates in two forms. For treatment comparison, we analyzed whole-organism V̇CO₂ and V̇O₂ with body mass included as a covariate. We refer to estimates from these models as mass-adjusted rates because they compare treatments at a common body mass while keeping the units of whole-organism metabolism (mL h^-1). For visualization and individual-level Arrhenius analyses, we divided whole-organism rates by body mass to calculate mass-specific rates (mL h^-1 g^-1). Accordingly, figures primarily show mass-specific rates, whereas tables report mass-adjusted estimates and treatment comparisons. We also used mass-specific rates for the Arrhenius fits and subsequent lipid-depletion simulations. RER, calculated as V̇CO₂/V̇O₂, was not adjusted or divided by body mass.

For queens with complete O₂ and CO₂ measurements at all four test temperatures, we estimated individual Arrhenius relationships for mass-specific V̇CO₂ and V̇O₂. For each queen and gas exchange measure, we fitted the equation ln(rate) = ln(b₀) - E_a·(kT)^-1, where ln denotes the natural logarithm, rate is mass-specific V̇CO₂ or V̇O₂, b₀ is the normalization constant, E_a is the apparent activation energy (eV), k is the Boltzmann constant (8.617 × 10^-5 eV K^-1), and T is absolute temperature (Kelvin). We fitted the relationship by regressing ln(rate) against inverse absolute temperature (1/T). Accordingly, E_a was calculated as -slope · k, and b₀ as exp(intercept) (Gillooly et al., 2001; Dillon et al., 2010).

Overwintering lipid depletion simulations

To evaluate the potential consequences of temperature-dependent metabolic responses, we simulated subsequent lipid depletion in bumble bee queens under alternative thermal scenarios. We used metabolic parameters measured after the six-week treatment and began all simulations from the same standardized energetic state. The simulations, therefore, compare subsequent depletion among metabolic phenotypes. They do not reconstruct lipid use during the treatment period or estimate total survival across the experiment.

We modeled temperature-dependent CO₂ production for each queen using Arrhenius relationships fit to mass-specific V̇CO₂ measurements (Fig. 2; Supplemental Fig. 2). For each individual, we used the Arrhenius intercept and slope to predict whole-organism CO₂ production for a standardized 0.5-g queen across hourly temperature profiles. We summed the hourly rates to estimate daily CO₂ production, then converted CO₂ production to lipid consumption using an energetic equivalent of 27.9 J mL^-1 CO₂ and an energy density of 37 kJ g^-1 of lipid (Lighton, 2018). This conversion assumes an RER of 0.7, consistent with predominantly lipid-based metabolism during overwintering.

We ran simulations under four future thermal scenarios: constant 2°C and 4°C, and variable 2°C and 4°C (± 6°C; hourly temperatures with a sinusoidal cycle). For each combination of experienced thermal regime during the initial six-week exposure and future thermal scenario, we summed daily lipid depletion from an initial lipid store equal to 10% of body mass until stores reach 2% of body mass, a threshold associated with diapause termination and forced emergence (Keaveny & Dillon, 2022; Waybright & Dillon, 2024). We defined the number of days required to reach this threshold as the predicted time to lipid store depletion. We did not adjust starting lipid stores using treatment-specific mass loss because total mass change can include water, gut contents, lipid, and other tissues, and we do not know when metabolic adjustments started during the six-week exposure. We evaluated both the continuation of the prior thermal regime and switches between constant and variable future conditions.