Sucrose syrup preparation. To produce syrup, sucrose was added to distilled water in a 20 L bucket and mixed until sugar was dissolved to produce a solution of 1:1 sucrose weight to water volume. To produce syrup with imidacloprid, 1 kg sugar was dissolved in 900 mL distilled water to produce 1.9 kg “short” solution. A spike solution was prepared by first dissolving 1.0 mg imidacloprid (Pestanal, CAS # 138261-41-3) in 100 mL distilled water, then mixing 1 mL of that solution into 99 mL distilled water to achieve a 100 mL spike solution. The spike was then added to 1.9 kg short sucrose solution to achieve 2 kg of 5 ppb imidacloprid syrup.
Field experiments. The 2014 experiment16 and 2016 experiment17 have been described previously and were conducted at apiaries in the Santa Rita Experimental Range (31°46'39"N, 110°51'46"W). The 2017, 2018 and 2019 experiments were conducted <1 km north of the 2014 site; data for the control treatment in the 2017 and 2018 studies have also been described previously45. The apiaries were surrounded by native mesquites (Prosopis spp.), cacti (e.g., Opuntia spp.), creosote (Larrea tridentata) and other wildflowers.
Honey bee colonies were purchased as 1kg bee packages (C.F. Koehnen & Sons, Inc., Glenn, CA 95943) for the first three experiments, and as colonies (Marquette Apiaries, Winchester, CA 92596) for the last two. Colonies were kept in painted, 10-frame, wooden Langstroth boxes (43.7 l capacity) with migratory wooden lids. Each package was given 4 frames of capped honey, 2 frames empty drawn comb, 3 frames of foundation and a frame feeder. Colonies established as packages were all immediately fed 2 kg sugar syrup (1:1 w:w) and 250 g pollen patty, made at a ratio of 1:1:1 corbicular pollen (Great Lakes Bee Co.): granulated sugar: drivert sugar (Domino Foods). Hives were placed on electronic scales (Tekfa model B-2418 and Avery Weigh-Tronix model BSAO1824-200, max. capacity: 100 kg, precision: ±20g; operating temperature: -30ºC to 70ºC), powered by deep-cycle solar powered batteries, and linked to dataloggers (Hobo UX120-006M, Onset Computer Corporation, Bourne, MA, USA) with weight recorded every 5 minutes (every 15 min in 2014). Hives were organized in groups of 4-6 hives; groups were >3 m from each other and within groups hives were at least 0.5m apart.
In early July temperature sensors (iButton Thermochron, precision ±0.06°C) enclosed in plastic cassettes (Thermo Fisher Scientific, Waltham, MA) were stapled to the center of the top bar on the middle frame in the bottom box and set to record every 15 min. In the 2018 and 2019 experiments, CO2 probes (Vaisala Inc., Helsinki, Finland), calibrated for 0-20% concentrations, were placed on top of the center frames in the top box of five hives in each treatment group and set to record every 5 min. At that time, paperboards coated with petroleum jelly and covered with mesh were inserted onto hive bottom boards and left 2-3 days to monitor Varroa mite fall46.
Hives were assessed before assignment to treatment groups using a published protocol26,47. Briefly, each frame was gently shaken to dislodge bees, photographed (Rebel SL1, Canon USA, Inc., Melville, NY), weighed (EC15, OHaus), and replaced. During this first assessment, all hive components (i.e. lid, inner cover, boxes, bottom board, etc.) were also shaken free of bees and weighed. The total adult bee population weight was calculated by subtracting the combined weights of hive components free of bees from the total hive weight with bees recorded at midnight prior to the inspection. In addition, 3-5 g of honey were collected from each hive into 50 ml tubes and stored at -80ºC; samples collected in July before treatment were pooled and subjected residue analysis of pesticides and fungicides by the Laboratory Approval and Testing Division, USDA.
After the first assessment, hives were assigned to groups based on adult bee mass and location (to avoid clumping) and each group randomly assigned a treatment. Just prior to treatment all broodless frames with food stores were replaced with empty drawn comb. Colonies were then fed 2-3 kg treatment syrup twice per week for six weeks, from mid-July onwards. Syrup consumption per colony was recorded. Hives were assessed every 5-6 weeks after the treatment period until Nov., and once more in Feb. At each assessment, hive lid and inner cover weights were compared to previous values to correct wooden parts for moisture content and improve adult bee estimates.
We sampled NEBs before and after the treatment period by pressing an 8 cm x 8 cm wire cage into capped brood, collecting trapped NEBs into 50 mL centrifuge tubes 24 h later, and then storing them at -80°C. To determine dry weight, five bees per hive per date were placed in Eppendorf tubes and weighed before and after drying for 72 h at 60°C. Varroa mite fall was measured within a few weeks after treatment and colonies were treated in Oct. or Nov. using an amitraz-based product. Ambient weather data was obtained from AmeriFlux US-SRM Santa Rita Mesquite, https:// doi.org/10.17190/AMF/1246104.
Mass values were converted to adult numbers by assuming an average of 0.11 g per adult. The area of sealed brood per frame was measured from photographs using ImageJ version 1.47 software (W. Rasband, National Institutes of Health, USA) or CombCount48 and multiplied by 4.1 cells per cm2 47 to estimate capped cell number. Stored food mass was calculated thus:
- Subtracting the weight of an empty drawn comb, about 555 g47, from the weight of each frame (negative values counted as zero) to estimate brood + food weight;
- Adding the brood + food frame weights for each hive;
- Multiplying the brood surface area by 0.77 to estimate g brood mass47 and subtracting that value from the brood + food weight.
- Dividing the change in the stored food mass between the Oct. and Feb. hive assessments by the number of days between those assessments.
Adult bee, brood cell and food resources were subjected to repeated-measures MANOVA with treatment, year, sampling date, and all 2-way interactions as fixed effects and with pre-treatment values as covariates, where possible, to control for pre-existing differences (Proc Glimmix, SAS Inc. 2002). Post-hoc contrasts with Bonferroni correction were reported for significant main effects. Analyses of NEB dry weight and Varroa fall were limited to a single sampling occasion immediately after the end of treatment application. Temperature and CO2 data were transformed into daily average and within-day detrended data, calculated as the difference between the 24 hour running average and the raw data. Sine curves were fit to 3-day subsamples of detrended data47 and MANOVA analyses conducted on daily averages and sine amplitudes. To improve the detection of treatment effects across seasons, continuous data were divided into different time periods for analysis (see Table 1). Hive CO2 concentrations were also analyzed using a second method: CO2 concentrations at the time of hive evaluations were estimated by averaging data over three days prior to evaluations (July concentrations at start of study obtained three days post evaluation) and subjected to the MANOVA analysis described above with adult bee mass as covariates. Figures were made in R v4 using tidyr v1.1.0, dplyr v0.8.5, cowplot v1.0.0 and ggplot 2 3.3.1.
Continuous hive weight data per day were analyzed by piecewise regression using a bootstrapping procedure49 after subtracting the hive weight at midnight from each subsequent weight value over the next 24 h50. Data were fit with piecewise regression, yielding estimates for 4 break points, 5 slope values and the adjusted r2. Four parameters were used: 1) night slope SN (rate of hive weight change due to moisture loss from midnight until dawn); 2) dawn break point tD (start of daily foraging activity); 3) slope of the 1st segment after dawn SM (rate of morning hive weight change); and 4) dusk break point (end of daily foraging activity). SM was attributed to both forager departure and moisture loss (i.e. nectar drying, respiration). To estimate weight change due to forager departure, the effect of moisture weight change was removed thus:
ΔF = SM(tD+1 – tD) – SN(tD+1 – tD) 
with ΔF the hive weight change due to forager departure, and tD+1 the time of the break point following the dawn break point tD. ΔF, tD, and the dusk break point were used as response variables in MANOVA analyses (see above). Days with rainfall >3 mm, slopes <-0.4kg/min, and forager weight change > 0 (indicating hive weight gain) were excluded. Analyses of hive weight data were limited to approximately 2 months after the end of treatment to focus on the active season.