Engineered yeast provides rare but essential pollen sterols for honeybees

Strains, culture conditions and chemicals

Escherichia coli strain DH5α was used for plasmid construction. E. coli was grown at 37 °C and 300 r.p.m. in lysogeny broth (LB) liquid medium and at 37 °C on plates of LB solid medium supplemented with 20 g l^–1 agar. Ampicillin was supplemented at a concentration of 100 mg l^–1 for plasmid selection.

The Y. lipolytica W29 strain (MATa, Y-63746 ARS Culture Collection, The National Center for Agricultural Utilization Research) and the W29-derived platform strain ST9100 (MATa ku70∆::PrTEF1-cas9-TTef12::PrGPD-dsdA-TLip2 IntC_2-HMG<-PrGPD-PrTefInt->ERG12 pCfB8823 IntC_3-SeACS<-PrGPD-PrTefInt->YlACL1 IntD_1-IDI1<-PrGPD-PrTefInt->ERG20, a mevalonate-upregulated strain) have been previously described²³. The platform strain ST9100 was used to construct the sterol-producing strains. Details for all of the strains used in this study are provided in Supplementary Table 2.

Y. lipolytica was grown at 30 °C on yeast extract peptone dextrose (YPD) medium containing 10 g l^–1 yeast extract, 20 g l^–1 peptone and 20 g l^–1 glucose, supplemented with 20 g l^–1 agar for preparation of solid medium. For selection, either nourseothricin (250 mg l^–1) or hygromycin (400 mg l^–1) was added to the medium. Cultivation of strains for sterol production was performed in YPD medium containing 80 g l^–1 glucose. Chemicals were obtained, unless indicated otherwise, from Sigma-Aldrich or Merck. Nourseothricin was purchased from Jena BioScience.

Plasmid construction

The following coding sequences for enzymes were codon-optimized for Y. lipolytica and synthesized as GeneArt Strings DNA fragments by Thermo Fisher Scientific: Δ⁷-sterol reductase from S. tuberosum (StDWF5, GenBank accession: BAQ55276.1), D. rerio (DrDHCR7, accession: NP_958487.2), L. drancourtii (LdDWF5, accession: FJ197317.1), E. siliculosus (EsDWF5, accession: CBN77313.1), ‘Candidatus Protochlamydia amoebophila’ (CPaDWF5, accession: KIC71363.1), C. subellipsoidea (CsDWF5, accession: XM_005650286.1), M. vertcillata (MvDWF5, accession: KFH65691.1), G. soja (GsDWF5, accession: XP_028244742.1); Tetraselmis sp. GSL018 (TspDWF5, accession: JAC78771.1) and W. chondrophila (WcDHCR7, accession: ADI39181.1); squalene-tetrahymanol cyclase from T. thermophilia (TtSTC1, accession: XP_001026696.2); Δ²⁴⁽²⁵⁾-sterol reductase from S. lycopersicum (SlSSR2, accession: BAQ55273.1); C-28 sterol methyl transferase from C. quinoa (CqSMT, accession: XP_021737090.1); and Δ²⁴⁽²⁸⁾-sterol reductase from S. tuberosum (StSSR1, accession: AB839749.1). The codon-optimized sequences are listed in Supplementary Table 3.

The plasmids, BioBricks and primers used in this study are listed in Supplementary Tables 4–6. BioBricks were amplified by PCR using Phusion U polymerase (Thermo Scientific). BioBricks were assembled into EasyCloneYALI vectors with uracil-specific excision reagent (USER) cloning³¹. For marker-mediated gene deletion, upstream and downstream homology arms for relevant genes were synthesized as BioBricks by PCR amplification from the genomic DNA of the platform strain ST9100. Knockout constructs were assembled from BioBricks through USER reactions as detailed in Supplementary Table 5. USER reactions were transformed into E. coli, and correct assemblies were verified by Sanger sequencing (Eurofins).

Yeast transformation

The yeast vectors were integrated into different previously characterized intergenic loci in the Y. lipolytica genome as previously described³¹. Integration vectors were digested with NotI enzyme (New England BioLabs) before lithium acetate transformation, as previously described³¹. Correct integration was verified by colony PCR using Taq DNA polymerase master mix RED (Ampliqon) with vector-specific primers and primers complementary to the genomic region adjacent to the integration site³¹.

For marker-mediated gene deletion, transformants were selected on YPD plates supplemented with antibiotic, and correct transformants were confirmed by colony PCR. Marker removal was performed by transformation of the strains with a Cre-recombinase episomal vector³¹. Marker removal was confirmed by colony PCR.

Yeast cultivation

Yeast strains were inoculated into 2.5 ml YPD in 24-deep-well plates with air-penetrable lids (EnzyScreen). The plates were incubated at 30 °C with 300 r.p.m. agitation for 24 h. The optical density at 600 nm (OD₆₀₀) was measured with an Implen P300 NanoPhotometer. The cultures were then diluted to an OD₆₀₀ of 0.1 in 2.5 ml fresh YPD medium with 80 g l^–1 glucose and grown for a further 72 h at 30 °C with 300 r.p.m. agitation. All cultivations were performed in triplicate. DCW was measured at the end of cultivation, whereby 1 ml of culture broth was transferred to a preweighed 2 ml microcentrifuge tube, centrifuged (3,000g, 5 min) and the supernatant was discarded. The cells were then washed twice with deionized water (1 ml). The cell pellet was dried at 60 °C for 7 days before the final weight was measured.

Sterol analysis

For sterol extraction from yeast, 1 ml of culture broth was transferred to a 2 ml microcentrifuge tube, centrifuged and the supernatant was discarded. The cells were washed twice with deionized water (1 ml). The cell pellet was resuspended in 10% w/v methanolic potassium hydroxide (500 μl) and transferred to a 1 ml glass vial for saponification. The suspension was incubated at 70 °C for 2 h with vortexing at 15-min intervals. The saponified samples were then vortexed and spiked with 50 μl of internal standard (1 mg ml^–1 epicoprostanol in absolute ethanol). Next, 500 μl n-hexane was added to each sample for extraction of the free sterol component. Samples were vortexed and the organic phase transferred to a 2 ml microcentrifuge tube. The extraction step was repeated in a further 500 μl n-hexane. The combined hexane phases were left overnight at room temperature for evaporation of the solvent. Sterol crystals remained in the tube.

For sterol analysis of the diets, each diet was sampled three times into preweighed 20 ml glass vials, and the weight of each sample was recorded. For sterol extraction from honeybee tissues, samples were first dried by freeze drying. Samples were dried at −48 °C under vacuum for 4 days. Dried samples were weighed and stored at −80 °C. For saponification, samples were first broken up with a spatula. For gut samples (in 2 ml microcentrifuge tubes), samples were suspended in 500 μl 10% w/v methanolic potassium hydroxide. For honeybee tissue samples (in 20 ml vials; pupae, nurse carcasses and queens), samples were suspended in 2.5 ml 10% w/v methanolic potassium hydroxide. Diet samples (in 20 ml vials) were suspended in 5 ml 10% w/v methanolic potassium hydroxide. Samples were incubated at 70 °C for 2 h in a water bath, with vortexing at 30–60 min intervals. The saponified samples were then spiked with 50 μl (gut samples) or 100 μl (diet, pupae, nurse carcasses and queen samples) of internal standard (1 mg ml^–1 epicoprostanol in absolute ethanol). For extraction of the free sterol component, 500 μl (gut samples), 2.5 ml (pupae, nurse bee carcasses and queen samples) or 5 ml (diet samples) n-hexane was added to each sample. Samples were vortexed and the organic phase transferred to a 2 ml (gut samples) or 7 ml (diet, pupae, nurse bee carcasses and queen samples) glass vial. The extraction step was repeated and the combined hexane phases were left overnight at room temperature for evaporation of the solvent. The resulting extracts were resuspended in 500 μl (gut samples) or 1 ml (diet, pupae, nurse bee carcasses and queen samples) n-hexane and vortexed. From each sample, a subsample of 250 μl was transferred to a 1.5 ml microcentrifuge tube and left at room temperature overnight for final drying.

Sterols were resuspended in 500 μl pyridine that contained 20 μl N,O-bis(trimethylsilyl)acetamide (Merck) and incubated for 4 h at 50 °C and then briefly vortexed before direct injection into an Agilent Technologies 8860 gas chromatograph connected to an Agilent Technologies 5977 MSD mass spectrometer (for gas chromatography–mass spectrometry). Samples were eluted over an Agilent HP-5MS column using a splitless injection at 250 °C with a standard gas chromatography program at 170 °C for 1 min, ramped to 280 °C at 20 °C min⁻¹ and monitoring between 50 and 550 AMU.

Sterols were identified by comparing their retention time relative to CHOL and mass spectra data available from the National Institute of Standards and Technology mass spectral library per a previous study¹⁸. Sterol identity in the final strain ST12178 was confirmed through comparison with authentic standards. Sterols were quantified by calculating the ratio of the peak area of the targeted sterol to that of the internal standard. The mass of each sterol in the sample was obtained by multiplying the ratio with the mass of the internal standard. Compound identification (using target ions) and quantification were carried out using ChemStation Enhanced Data Analysis (v.E.01.00).

Bioreactor fed-batch cultivation

The Ambr 250 system (Sartorius Stedim Biotech) was used to carry out 250 ml fed-batch fermentation in duplicate. The 24-MC strain ST11064 was re-streaked from glycerol stocks stored at −80 °C onto a YPD agar plate and incubated at 30 °C for 48 h. The preculture was prepared by inoculating strain ST11064 biomass from the plate into 50 ml YPD medium in a 250 ml shake flask and incubating at 30 °C for 24 h with 250 r.p.m. agitation. Next, 5 ml of preculture was used to inoculate 115 ml of batch medium to a starting OD₆₀₀ of 0.25. For Ambr 250 cultivation, the batch medium comprised mineral medium supplemented with yeast extract (10 g l^–1) and citric acid (20 g l^–1). The mineral medium was prepared with 0.5 g l^–1 MgSO₄⋅7H₂O, 14.4 g l^–1 KH₂PO₄, 0.1% (v/v) vitamin solution and 0.2% (v/v) trace metal solution as previously described⁴⁵, but with 3.4 g l^–1 NH₄Cl and glycerol as the carbon source (40 g l^–1).

The temperature was held constant at 30 °C. Dissolved oxygen was maintained above 20% by using a cascade of stirring speed ranging from 600 to 3,000 r.p.m. and aeration up to 1 volume air per volume growth medium per minute. The pH was maintained at 5.5 through the automatic addition of 1 M NaOH and 2.6 M H₃PO₄. Antifoam 204 (Sigma) was added automatically. Online measurements of acid and base addition, carbon dioxide evolution rate, dissolved oxygen and stirring speed were recorded for each reactor. The feed medium comprised 250 g l^–1 glycerol. Feeding was automatically initiated once the carbon dioxide evolution rate dropped below 50% at the end of the batch phase. Feeding was set to a constant rate of 0.9 ml h^–1. Samples were taken from each reactor every 6 h for the first 24 h and then every 12 h and immediately frozen until preparation for analyses. DCW and sterol content were determined from 1 ml samples as described above for small-scale cultivation.

For larger scale fermentation, strains were cultivated by fed-batch fermentation in a 5-litre bioreactor (BIOSTAT B-DCU, Sartorius). All fermentations were carried out in duplicate. For each of the strains W29, the TET strain ST11005 and the mixed-sterol strain ST12178, the strain was re-streaked from glycerol stocks onto a YPD agar plate and incubated at 30 °C for 24 h. The preculture was prepared by inoculating strain biomass from the plate into 50 ml YPD medium in a 250 ml shake flask and incubating at 30 °C for 24 h with 250 r.p.m. agitation. The volume of preculture required to inoculate a 2-litre batch medium to a starting OD₆₀₀ of 2.5 was centrifuged for 10 min at 4,000g and concentrated to 10 ml volume. This cell suspension was used to inoculate the bioreactors. The bioreactors were equipped with pH, pO₂ and temperature probes. The temperature was held constant at 30 °C. Dissolved oxygen was maintained above 20% by adjusting stirring between 600 and 1,200 r.p.m. and aeration (using a horseshoe sparger) between 0.5 and 3 standard-litre per min. The pH was kept at 5.5 through the automatic addition of 5 M NaOH. Antifoam A (Sigma) was added as required.

The batch medium comprised mineral medium supplemented with yeast extract (20 g l^–1) and peptone (40 g l^–1). The mineral medium was prepared with 0.5 g l^–1 MgSO₄⋅7H₂O, 14.4 g l^–1 KH₂PO₄, 0.1% (v/v) vitamin solution and 0.2% (v/v) trace metal solution as previously described⁴⁵, 40 g l^–1 glucose and 1 ml l^–1 antifoam A (Sigma). The feed contained 5 g l^–1 MgSO₄⋅7H₂O, 30 g l^–1 KH₂PO₄, 1% (v/v) vitamin solution and 2% (v/v) trace metal solution as previously described⁴⁶, with 300 g l^–1 glucose. An exponential feeding profile was programmed, and feeding was initiated 24 h after inoculation. The feed rate, F (ml h^–1), followed the profile F = 10 × e^{(0.05 × t)}, where t is the time (h) from the start of feeding. After 36 h of exponential feeding, the feed was switched to a constant rate of 75 ml h^–1 until the end of fermentation.

Duplicate samples from each reactor were taken every 8 h for the first 24 h and then every 12 h to measure DCW, sterol content, OD₆₀₀ and glucose concentration. DCW and sterol content were determined from 1-ml samples as described above for small-scale cultivation. During fermentation, 1 ml of culture broth was centrifuged, and the supernatant was used to measure the glucose concentration using a glucose HK assay kit (Sigma). The supernatant was then filtered and frozen until further analyses. Glucose was later quantified using a Dionex Ultimate 3000 HPLC system equipped with a RI-101 refractive index detector (Dionex). An Aminex HPX-87H column (7.8 × 300 mm, Bio-Rad) with a Micro-Guard Cation H⁺ guard column (4.6 × 30 mm) heated to 30 °C was injected with a 10-µl sample. The mobile phase consisted of 5 mM H₂SO₄ with an isocratic flow rate of 0.6 ml min^–1, which was held for 15 min. HPLC data were processed using Chromeleon software (v.7.2.9, Thermo Fisher Scientific). Glucose was identified and quantified using authentic standards. Glucose concentrations were calculated from the peak area by extrapolation from a six-point calibration curve regression.

Honeybee diet preparation

The yeast strains W29, the TET strain ST11005 and the mixed-sterol strain ST12178 were cultivated using 5-litre fed-batch fermentation as described above. At the end of fermentation, the yeast biomass was recovered from the culture by centrifugation (4,000g, 20 min) and washed with deionized water. The biomass was heat-inactivated and dried (60 °C for a minimum of 24 h). The dried material was ground to a fine powder and stored at −20 °C until further use.

The yeast biomass cannot be subject to inactivation by autoclave or chemical treatment, as this will degrade the sterols present in the yeast. Incubation at 60 °C is commonly deemed sufficient to irreversibly inactivate yeast, and heat-inactivation of genetically modified yeast followed by feeding the inactivated yeast to live animals is a method that has been previously used in the United Kingdom⁴⁷. Irreversible inactivation of the yeast was confirmed using a standard colony-forming unit assay. The heat-inactivated dried yeast was dissolved at 10 mg ml^–1 in water. The suspension was plated in serial dilution (100 μl plated of 10, 1, 0.1, 0.01 and 0.001 mg ml^–1 suspensions) on YPD agar and the plates were incubated at 30 °C for at least 7 days. No growth of Y. lipolytica colonies was observed. The detection limit is one organism per mg material or 10⁶ viable organisms per kg of material.

The yeast biomass was then incorporated into a meridic artificial diet at 20% w/w. Four diet types were prepared: a mixed-sterol yeast diet that contained the mixed-sterol strain ST12178; a wild-type yeast diet that contained strain W29; a platform yeast diet that contained the TET strain ST11005; and a base diet control without yeast supplementation. The base diet control was formulated to maintain total protein, sugar, sterol and fat content at the same level as in the yeast-supplemented diets. The content of this diet was a modified version of a previously described diet⁴⁸. Specifically, the base diet contained 17% soy protein isolate (Soysol, MyVegan), 69.4% sugars (fructose, glucose, sucrose and maltodextrin), 6% lipids, 6.50% deionized water, 0.100% vitamin and mineral supplement (Latshaw Apiaries), 0.6% commercial phytosterol mix (BulkSupplements; Supplementary Table 7) and 0.400% carrageenan kappa. The diets were divided into 50 g patties and stored at −20 °C until use. The yeast-supplemented diets had the same proportion of protein (17%), carbohydrates (70%) and fats (6%) adjusted from the reagents of the base diet to accommodate nutrients present in the yeast. Specifically, the yeast-supplemented diets contained the following components: 20.0% dried yeast powder, 7.80% soy protein isolate (Soysol, MyVegan), 63.4% sugars, 0.1% commercial phytosterol mix (BulkSupplements, approximately 55% purity containing a mixture of sterols and stanols; Supplementary Table 7), 4% lipids, 4.20% deionized water, 0.100% vitamin and mineral supplement (Latshaw Apiaries) and 0.400% carrageenan kappa (Special Ingredients).

Yeast-feeding trials

For the sterol analysis of honeybee brood, which is used as a proxy for the natural sterol profile of honeybee pupae, we sampled worker, drone and queen pupae from naturally fed colonies in our apiary (Buckfast queens, John Krebs Field Station, Oxford). Worker pupae were directly collected from capped brood frames. Drone pupae were collected from capped drone comb (larger cell size). Queen pupae were reared by grafting young larvae (2–3 days after hatching) into Nicot Queen Rearing Cups (Paynes Bee Farms). These were placed in queenless colonies in repurposed Styrofoam mini-nucleus hives (APIDEA) for up to 8 days until development to the capped brood stage. Tissues (3 pupae per replicate, n = 5) were sampled into preweighed 20 ml glass vials and the fresh weight was recorded before storing at −80 °C until further analysis.

Feeding trials were conducted using honeybee colonies maintained in repurposed Styrofoam mini-nucleus hives made up of one brood box with five frames and a top feeder with a hole for patty delivery. Hives were maintained in a closed glasshouse environment designed to prevent bee escape during the period between July and October 2022. Hives were distributed across two glasshouse rooms with varying entrance orientation. Feeders with 30% w/v sugar solution and water were distributed inside the glasshouse and replenished as required. The in-hive and ambient temperature and humidity were recorded every 30 min using autonomous in-hive sensors (Supplementary Data 3). A misting system was installed to cool the temperatures in the glasshouse.

Initially, each hive contained 900–1,200 adult bees, 2–3 frames of brood, larvae and eggs and 1–2 frames of honey stores, but no bee bread. Newly mated queens were introduced in cages with beekeeping candy (Candito, PIDA), for slow release, 3 days before the start of the experiment.

Feeding trials were conducted over 3 months from June to September 2022 at the John Krebs Field Station, Oxford. Six hives were randomly assigned to each treatment group (n = 6). At the start of the experiment, diet patties were added through the top feeder and replaced throughout the experiment as required. Hive weight (after removal of the diet patty) and patty weight were measured. The number of bee seams (one seam defined as a continuous line of bees between adjacent frames, observed after initial hive opening) and frames filled with honey (sugar stores) were estimated by visual inspection. The presence of the mated queen, sugar stores, eggs, larvae and capped brood were checked, and brood frames were photographed for subsequent counting. Eggs, larvae and capped brood were counted using the Adobe Photoshop count tool. Full assessments of the hives were conducted every 15 days.

Six days after each full assessment, hives were partially assessed with minimal disruption to the colony. Hive weight and patty weight were measured, and bee seams and sugar stores were estimated by visual inspection. The presence of the mated queen, eggs, larvae and capped brood were briefly checked. On days 21 and 45, hives with low populations (fewer than four bee seams) were topped up with orphanized nurse bees from mixed, naturally fed colonies.

At every assessment, nurse bee and pupae samples were taken from three hives from each treatment group. Six nurse bees and six pupae were collected from each of the sampled hives. Samples were collected into preweighed 20 ml glass scintillation vials. The fresh weights of the samples were measured, and the vials were stored at −80 °C. Nurse bees were dissected to separate the guts and gut contents from the rest of the tissues. This was done by partially thawing the samples on ice and pulling the guts from the abdomen by the stinger. The gut contents were transferred to a 2 ml microcentrifuge tube and the remaining tissues were returned to the 20 ml vial. Dissected samples were stored at −80 °C until further analysis.

Pollen-starvation trial

A semi-field pollen starvation trial was conducted from August to October 2023 at the John Krebs Field Station, Oxford. Colonies were housed in mini-nucleus hives set up in an identical manner to the yeast-feeding trial and were maintained in one room of a mesh polytunnel purpose-built to prevent bee escape. One week before the start of the treatment, colonies were topped up with nurse bees from mixed, naturally fed colonies so that each box contained at least five bee seams and re-queened where necessary. We used a mix of pre-existing colonies and newly established colonies, but all were fed pollen for at least 1 month before the start of the experiment and were producing brood.

Buckets of water and feeders with 30% w/v sugar solution were distributed inside the polytunnel. Colonies were supplied with pollen or candy patties. Pollen patties consisted of 80% multifloral pollen and 20% high-concentrated sugar syrup (around 70% w/v). Candy patties consisted of about 80% beekeeping candy and around 20% maltodextrin, which was added to slow consumption and reduce melting of the patty in the hive. After 1 month of a feeding period, colonies in treatment groups 1, 2 and 3 were deprived of pollen for the corresponding number of weeks and fed with candy only. The control group (0) was fed pollen throughout. Treatment groups were balanced across hive entrance orientations, colony strengths (bee seams) and position in the polytunnel.

We performed a partial assessment every week to measure patty weight and hive weight and estimate bee seams. Every 2 weeks, a full assessment recorded the presence of the mated queen, eggs, larvae and capped cells, and the amount of sugar and pollen stores, and every frame was photographed. The photographs were used to count the number of cells with eggs, larvae and pupae in each hive using ImageJ⁴⁹.

Statistics and reproducibility

All graph plotting and statistical analyses were performed in R⁵⁰ (except Extended Data Fig. 8, which was created in GraphPad Prism and analysed using SPSS). Data were tested for normality when appropriate. GLMs and GLMMs were fitted using stats⁵⁰ and glmmTMB⁵¹ packages. Models with nonsignificant interaction terms were re-run without the interaction term. Post hoc analysis was performed using the car⁵² and emmeans⁵³ packages with Tukey adjustments for family-wise error rates.

The mean relative abundance of each of the major sterols in naturally fed honeybee pupae was calculated as a percentage of the total sterol content. We compared sterol relative abundance across pupal types using a GLM with quasi-binomial distribution (relative abundance modeled as a function of (~) sterol type × pupal type). Sterol concentrations in pupae were calculated from the fresh weights of the pupal tissue. We compared sterol concentrations in pupal tissue across types using GLMs with Gaussian distributions (sterol concentration ~ sterol type × pupal type). We compared the relative abundance of sterols in pollen using a GLM with quasi-binomial distribution (relative abundance ~ sterol type). The coefficient of variation for each sterol was calculated by dividing the standard deviation by the mean relative abundance values of each sterol.

Counts for each brood type were compared across diet treatment groups using GLMMs, fitted to counts from day 45 onwards, with hive identifier (ID) as a random effect and negative binomial distributions (brood count ~ diet × time + (1|hive ID)). For egg counts, the interaction term was excluded (egg count ~ diet + time + (1|hive ID)).

Both the total diet provided to each colony and the total diet consumption by each colony were compared across diet treatment groups by fitting GLMs with Gaussian distribution (diet weight ~ diet). Because the hive weight correlated significantly with bee seams and the consumption rate correlated significantly with the hive weight for all treatment groups (Extended Data Fig. 8), the daily consumption rates in each interval were normalized by hive weight as a proxy for colony size. The normalized consumption rates were compared across diet treatment groups using a GLMM, with hive ID as a random effect and Gaussian distributions (normalized consumption rate ~ diet + time + (1|hive ID)).

The weight of each hive was compared across diet treatment groups using a GLMM, with hive ID as a random effect and Gaussian distributions (hive weight ~ diet + time + (1|hive ID)). The number of bee seams in each hive and the number of frames filled with honey were doubled to give integer values and compared across diet treatment groups using GLMMs, with hive ID as a random effect and Poisson distributions (2 × bee seams ~ diet + time + (1|hive ID); 2 × sugar stores ~ diet + time + (1|hive ID)).

For the sterol contents in the bodies and guts of nurse bees and of pupae (μg per individual), GLMMs were fitted for each sample type in each sterol, with Gaussian distributions and hive ID as a random effect (sterol content ~ diet × time + (1|hive ID)). Interaction terms were not significant in some models (total sterol in pupae, 24-MC in pupae, CAMP in the bodies and guts of nurse bees, ISOFUC in pupae, DESMO in pupae and guts of nurse bees) and were removed as appropriate.

We examined the relationships between variables measured during feeding trials by fitting GLMs (response variable ~ predictor variable × diet). All models used Gaussian distributions, apart from when comparing capped brood counts to the total sterol content of bodies of nurse bees from the same colony. In this case, a negative binomial distribution was used. When no significant interaction was found between diet treatment group and the predictor variable, the interaction term was excluded from the models (response variable ~ predictor variable + diet). For each diet treatment group, linear regressions were fitted between the predictor and response variables of interest.

All experiments in the study were performed once. Apart from bioreactor cultivations, a minimum of three biologically independent replicates were collected for all experiments. For bioreactor cultivations, two biological replicates with two technical replicates each were performed for measurement of all parameters. Replicates gave similar results for all experiments.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.