The study was conducted in three geographically distinct regions in Germany: the UNESCO Biosphere Reserve area Schorfheide-Chorin (SCH) in the northeast, the National Park Hainich-Dün (HAI) in the center, and the UNESCO Biosphere Reserve Swabian Alb (ALB) in the south. We selected 27 grassland plots (9 per region) as part of the long-term Biodiversity Exploratories project (Fischer et al., 2010) (Supplementary Figure SM1). Each 50 m × 50 m plot represented different land use intensity types, including meadows, mowed pastures, grazed pastures, and fertilized and unfertilized areas, with management types extending beyond plot borders (Blüthgen et al., 2012) (Supplementary Table SM1). Detailed framework information is available in Fischer et al. (2010).

Bees were sampled according to legal requirements with permits ALB: AZ: 55-8/8848.02-07, HAI: AZ: 63.02/15.02.11-bio_expl2017.2 & AZ: 1011-17-301, SCH:AZ: 4743/128+5#69122/2018. We used vegetation records from BExIS public datasets (IDs 23,586 and 24,247: Vegetation Records for 150 Grassland EPs 2008–2018) to assess plot vegetation Shannon diversity (Schäfer et al., 2018; Bolliger et al., 2020), as well as data on management and land use intensity (LUI) from dataset IDs 25,086 and 31,514 (Lorenzen et al., 2023; Ostrowski et al., 2023). Land use intensity was categorized into low (LUI < 1.5), intermediate (LUI 1.1–2.3), and high (LUI > 2.3). Additionally, we used plant pollen dataset ID 27229 from our previous study on O. bicornis larval pollen provisions to analyze the effects of land use intensity on pollen plant diversity (Peters et al., 2022).

To collect various sample types (larval pollen provisions, soil nest enclosures, bee larvae and pupae) of Megachilid solitary bee nests, we used artificial perpendicular trap nests made of plastic tubes with 60–80 hollow reed sticks (length ~ 20 cm, width 4–12 mm). In early spring 2017, the trap nests were placed at the fence of a weather station in the center of each grassland plot. From March to July in 2017 and 2018, reed sticks were regularly checked for solitary bee occupation. Sticks with closed entrances were carefully removed and replaced with empty ones. The collected sticks were then transported to the laboratory (see Peters et al., 2022 for more details).

We classified solitary bee species based on reed nest closures and bee morphology, following the methodology outlined by Amiet, (2017). Reed cane internodes were opened, and solitary bee larvae, pupae, larval pollen provisions and soil nest enclosure materials were collected separately using sterile spatulas and forceps. We specifically focused on nests of Osmia bicornis, which was the only species present across all three bioregions and the entire LUI gradient. Well-developed larvae and pupae, as well as pollen provisions, soil nest enclosures from the reed cells were transferred into autoclaved tubes and a total of 144 samples immediately frozen at −20°C for preservation. Afterwards reeds were carefully re-closed to allow further larval development of remaining bees to facilitate more detailed classification after enclosure.

Genomic DNA extraction for bacterial analysis the ZymoBIOMICS^™ 96 DNA Kit (Zymo Research) was utilized following the manufacturer’s protocol. To generate a pooled amplicon library for the 16S rRNA V4 region, we employed a dual-indexing strategy following the methods described in Kozich et al. (2013) and Illumina (2017). To minimize amplification biases all PCRs were performed in triplicates and with a proofreading Phusion High-Fidelity PCR Master Mix with HF Buffer according to manufacturer’s instructions (Thermo Fisher Scientific, Waltham, United States). Negative controls, including (i) DNase/RNase Free Water (Zymo Research) and (ii) DNA/RNA Shield^™ (Zymo Research) and a Microbial Community Standard (Zymo Research) as a positive control, were included for quality control purposes. All controls underwent the same workflow as the other samples. To prevent the amplification of chloroplast related sequences in pollen samples, pPNA blocking primer (PNA Bio Inc., Newbury Park, United States) were applied at a final concentration of 0.3 μM during the PCR reactions. Following the method described by Lundberg et al. (2013). PCR conditions were as follows: initial denaturation at 95°C for 4 min, followed by 30 cycles of 95°C for 40 s, annealing at 55°C for 30 s (including PNA clamping at 75°C for 10 s), extension at 72°C for 60 s and a final extension step at 72°C for 5 min. Triplicates were pooled per sample, checked by gel electrophoresis on 1.5% agarose gels for successful amplification and stored at 4°C. Samples were normalized in DNA amounts using the Invitrogen SequalPrep Plate Normalization Kit (Thermo Fisher Scientific) and purified with AMPure beads (Agilent, Santa Clara, United States). The normalized library was pooled and its fragment length distributions assessed using High Sensitivity DNA Chips on a Bioanalyzer 2200 (Agilent). The final library was quantified using the Qubit II Fluorometer with the dsDNA High-Sensitivity Assay Kit (Thermo Fisher Scientific), diluted to Illumina MiSeq requirements (Illumina, 2013, 2016, 2017), complemented with 5% of Illumina PhiXv3 and then loaded into a 500 cycle Illumina MiSeq cartridge following the manufacturer’s protocol (Illumina, 2013, 2017). Sequencing was performed on an Illumina MiSeq device (Illumina Inc., San Diego, United States) at the Department of Human Genetics of the University of Würzburg, Germany.

For sequence analysis we utilized VSEARCH v.2.15.1 (Rognes et al., 2016) according to the pipeline available at https://github.com/chiras/metabarcoding_pipeline (Leonhardt et al., 2022). Forward and reverse reads were merged (with a minimum overlap of 10 bp), and the sequences were filtered based on length (>250 bp) and quality (E_max < 1, no ambiguous base pairs). Singleton reads were excluded, and de-novo chimera filtering was performed (Edgar and Flyvbjerg, 2015). Sequences were denoised and dereplicated into amplicon sequence variants (ASVs) using the Unoise3 algorithm (Edgar, 2016b). Taxonomy assignment of 16S rRNA gene sequences was conducted using the RDP v18 reference database using SINTAX with a threshold of 0.8 (Edgar, 2016a). Additionally, individual ASVs at the species level were double checked against GenBank (Zhang et al., 2000) using the NCBI BLASTn (Altschul et al., 1990). ASVs were filtered that showed conspicuous distributions in positive and negative controls, as well as sequences related to mitochondria or remaining chloroplasts. Samples with less than 1,000 reads were excluded from further analysis.

Data was analyzed in R 4.0.2 (R core, 2017) using the packages phyloseq v1.22.3 (McMurdie and Holmes, 2013), vegan v2.5–2 (Oksanen et al., 2013), lme4 v1.1–21 (Bates et al., 2015), multcomp v1.4–10 (Hothorn et al., 2008), corrgram v.1.14 (Friendly, 2002) and ggplot2 v3.0.0 (Wilkinson, 2011).

Sequencing and bioinformatics yielded an average of 10,062 filtered reads per sample (range from 1,412 to 55,183, SD = 9122.183). Bacterial alpha diversity in terms of Shannon, richness and evenness varied between sample types, except the Larvae – Pupae comparison (Table 1; Supplementary Table SM3; Supplementary Figure SM2). We observed the highest ASV richness and Shannon diversity in soil nest enclosures compared to all other sample types (larval pollen provisions, larvae and pupae) (Supplementary Table SM3; Supplementary Figure SM2).

The bacterial community composition within the microbiome of O. bicornis bee nests exhibited significant differences across samples types (PERMANOVA: F = 9.98, df = 3, R² = 0.17, p < 0.001, Figure 1A), resulting in sample type-specific microbiome compositions (Figure 1A, PERMANOVA between all sample type pairs: p = 0.001). Furthermore, NMDS analysis at the ASV-level revealed a higher similarity in microbiome composition between O. bicornis bee pupae and bee larvae compared to larval pollen provisions and soil nest enclosure microbiomes (Figure 1A). Moreover, the variability in microbiome composition, quantified as the distance to the sample type centroid, and reflecting microbiome variations within sample types, varied among sample types (Kruskal Wallis test: χ² = 10.46, df = 3, p-value < 0.01, Figure 1B), with less variable microbiomes observed in O. bicornis bee pupae (distance to centroid = 0.41 ± 0.15) and soil nest enclosures (distance to centroid = 0.40 ± 0.08) compared to bee larvae (distance to centroid = 0.44 ± 0.15) and larval pollen provisions (distance to centroid = 0.47 ± 0.11) (Figure 1B). Between sample type comparisons revealed significant differences in microbiome variation between groups (Dunn-Test: bee_larvae – larval pollen provisions: p < 0.05, bee_pupae – larval pollen provisions: p < 0.01, and larval pollen provisions - soil nest enclosures: p < 0.01).

We found significant overlaps in bacterial communities between larval pollen provisions and soil nest enclosures with those of bee larvae and pupae (Supplementary Figures SM3, SM4, 75 omnipresent ASVs), however also as well unique elements (pollen: 1,096, soil: 445, larvae: 314, pupae: 220 exclusive ASVs). Regarding potential bacterial acquisition pathways from soil and pollen toward bee larvae and pupae, bacterial genera, such as Pseudomonas, Streptomyces, Acinetobacter, and Halomonas, were abundant across bee larvae, pupae, and larval pollen provisions, while they were less abundant in soil nest enclosures. Higher abundances of Bacillus genera were found in bee larvae (19.78%) and soil nest enclosures (8.4%), contrasting with lower abundances in larval pollen provisions (3.5%) and bee pupae (1.4%) (Supplementary Figure SM4).

Pairwise comparisons between bacterial distance matrices across various sample types of O. bicornis nests using Mantel tests revealed a significant positive correlation exclusively between larval pollen provisions and soil nest enclosures (Table 2). Furthermore, bacterial distances between bee larvae and larval pollen provisions, including soil nest enclosures as a covariate via partial mantel tests, also showed a significant positive correlation, while no significant correlations were found for bee pupae (Table 2).

Within O. bicornis bee nests, we found negative effects of LUI on the Shannon diversity of microbiomes of pollen (p < 0.001), soil nest closings (p < 0.01) and bee larvae (p < 0.05), however not on pupae (Figure 2; Supplementary Figure SM5). We additionally tested for the separate effects of mowing, grazing, and fertilization intensities at plot sites, which are all integrated in the LUI index, and found significant decreases in bacterial diversity particularly with increasing mowing intensities and fertilization (Supplementary Figure SM6; Table 3). Notably, no significant impact of any land-use variable on bacterial diversity was observed for bee pupae (Figure 2). Moreover, we observed a very high relative increase in the relative abundance of Bacillus sp. in O. bicornis larvae samples and soil enclosures (Supplementary Figure SM5), respectively from low [4.25% (larvae), 4.31% (soil)] to high land use intensity field sites [37.81% (larvae), 12.02% (soil)], which represents an increase of approximately 8 times for larvae and an increase 3 times for soil enclosures.

Bray-Curtis dissimilarities (multivariate homogeneity of group dispersions, R² = 0.15, p < 0.001) showed significant changes in the variability of bacterial communities in pollen provisions, increasing from low (distance to centroid = 0.40 ± 0.11), intermediate (0.48 ± 0.11) to high intensity field sites (0.50 ± 0.09) (Figure 3). Variability in O. bicornis bacterial larval communities also tended to increase (R² = 0.18, p = 0.068) from low to high intensity field sites (distance to centroid; low: 0.29 ± 0.09, intermediate: 0.45 ± 0.08, high: 0.41 ± 0.12) (Figure 3). We observe neither for soil enclosures nor pupae bacterial communities significant variability differences between low and high land use intensities (multivariate homogeneity of group dispersions, soil: R² = 0.13, p = 0.14, pupae: R² = 0.18, p = 0.91).

When assessing microbial richness and diversity across various O. bicornis sample types, we observed significant differences in bacterial community composition across sample types, suggesting a sample type-specific microbiome of O. bicornis bee nests which aligns with previous studies (Voulgari-Kokota et al., 2018). Moreover, we could show that soil samples, used for the segregation of individual nest chambers, demonstrated the highest bacterial diversity among all the examined nest materials (Keller et al., 2013; Voulgari-Kokota et al., 2020). This is in contrast to honey bee nest walls, which are built of waxes and propolis as a protective measure against microbial colonization and proliferation, and show little bacterial diversity (Anderson et al., 2011). The high soil-derived bacterial diversity in O. bicornis nests contrasts the antimicrobial environments typically found in managed hives where honey bee pupae develop (Gilliam, 1971).

Bacterial communities within solitary bee nests can be influenced by pollen used to provision solitary bee larvae (McFrederick et al., 2017) and thus the spectrum of allocated plant sources (Voulgari-Kokota et al., 2019a; Keller et al., 2021). We also found ASV overlaps and positive correlations between the bacterial communities between the pollen and soil microbiomes with those of the larvae. This suggests that environmentally introduced microbiomes from both pollen and soil influenced the bacterial communities of our larvae (Voulgari-Kokota et al., 2018; Cohen et al., 2020; Steffan et al., 2024). Certain bacterial genera, such as Pseudomonas, Streptomyces, Acinetobacter, and Halomonas, were shared between and enriched in bee larvae, pupae, and larval pollen provisions, while Acidobacteria and Bacillus exhibited higher abundances in soil nest enclosures and larvae as also reported in other studies (Keller et al., 2013; Lozo et al., 2015; McFrederick and Rehan, 2016; Voulgari-Kokota et al., 2018; Fernandez De Landa et al., 2023). While the role of Pseudomonas for bees is not quite understood (Fernandez De Landa et al., 2023), it is known for its diversity and prevalence on plant surfaces and in floral microbiomes (Roberson and Firestone, 1992; Chang et al., 2007; Voulgari-Kokota et al., 2018; Gaube et al., 2020; Steffan et al., 2024). Similarly, Acinetobacter is commonly found in nectar and associated with wild bees and has been shown to contribute to pollen germination and nutrient uptake within the protoplasm (Alvarez-Pérez et al., 2012; Fridman et al., 2012; Christensen et al., 2024). Soil poses an potential alternative bacterial transmission pathway, as shown by other studies for mud/soil nest enclosures (Keller et al., 2018; Voulgari-Kokota et al., 2020) or cut leaves (Rothman et al., 2019). Bacterial hubs or reservoirs are especially relevant during the early stages of bee development (Keller et al., 2018; McFrederick and Rehan, 2019; Dew et al., 2020; Christensen et al., 2023). For example, the bacterial community of pollen/nectar provisions in O. cornifrons brood cells initially exhibits a diverse bacterial composition, which is gradually reduced and altered over time by larval feeding (Kueneman et al., 2023). This process involves the suppression or elimination of less common taxa, while bacterial endosymbionts typically associated with insects and a range of plant pathogens proliferate (Kueneman et al., 2023).

Despite the observed overlaps with environmental materials, we still observed that the microbiomes of bee pupae and larvae were more similar to each other than to those from larval pollen provisions and soil nest enclosures. The overlap between pupae and larvae indicates a strong potential for bacterial proliferation and transmission from before to after metamorphosis (Voulgari-Kokota et al., 2019b). Bee pupae (and soil nest) enclosures exhibited lower variability compared to bee larvae and larval pollen provisions, indicating greater microbiome stability in these sample types. Furthermore, pupal microbiomes did not correlate with such of pollen nor soil in their composition as a potential result of environmental bacteria reduction over time in the nest (Kueneman et al., 2023). This suggests that while transgenerational passthrough is managed by some bacteria, there is likely a potential filter in the transfer of the microbiome from larvae to pupae. How emerging solitary bees recover their microbiome remains unclear. Proposed routes include inoculation by chewing through remaining nest materials or from flower hubs (Keller et al., 2021), which can however be excluded here as pupae did not emerge yet. Unlike social bees, solitary bee species lack adult nursing, which aids in establishing stable transgenerational microbial communities (Turnbaugh and Gordon, 2009; Danforth et al., 2019). Rather than maintaining a similarly consistent or conserved microbiome across individuals, adult solitary bees appear to exhibit microbial compositions that are more influenced by the environments and collected materials of previous generations, in line with our results here (McFrederick et al., 2012; McFrederick et al., 2017; Voulgari-Kokota et al., 2020; Kapheim et al., 2021; Nguyen and Rehan, 2023). Interestingly, a recent investigation conducted within the brood cells of the solitary bee Anthophora bomboides also revealed the presence of individually consistent microbiomes persisted throughout multiple life stages, suggesting a certain degree of individual microbiome stability (Christensen et al., 2023).

We found strong negative effects of land use intensity (LUI) on the bacterial diversity of larval pollen provisions, bee larvae and soil nest enclosures, more in detail with increasing mowing and fertilization intensities, but not on the Shannon bacterial diversity nor composition of bee pupae. The negative correlation is likely an indirect effect, since the diversity of flowering plant species on plots also correlated positively with bacterial Shannon diversity of larval pollen provision, soil nest enclosures and bee larvae. The direct link here is likely between the available spectrum of flowering plants and the microbiome associated with O. bicornis bee nests. A study conducted by Nguyen and Rehan (2022) demonstrated that microbial composition in the small carpenter bee, Ceratina calcarata, varies across different urban land use gradients. Specifically, microbes like Acinetobacter and Apilactobacillus are more common in less urbanized areas, while the fungus Penicillium is more prevalent in developed urban areas. Interestingly, in the study of Fernandez De Landa et al. (2023), no connection was found between the overall gut microbiome composition and land use intensity for the solitary bees Xylocopa augusti, Eucera fervens, and Lasioglossum. However, changes were observed for the bacterial symbionts Snodgrassella and Nocardioides, which displayed higher abundances in less anthropogenically impacted sites. Similarly, higher land use intensity also led to flower bacterial communities that were less phylogenetically diverse and more uniform in composition, and to a reduced floral bacterial species pool at high land-use intensity plots (Gaube et al., 2020). This supports the idea that LUI indirectly impacts the microbiome associated with pollen collected by solitary bees and consequently bee offspring via the food and nest resources allocated in nests (Figure 4).

Moreover, we found the variability and heterogeneity of bacterial microbiomes in larval pollen provisions and bee larvae to increase with land use intensification, which might indicate more erratic microbiomes or alternative floral sources used in intensively used areas. The stage-specific vulnerability of developing bees to environmental stressors (Ferguson et al., 2018; Christensen et al., 2024) combined with the lack of microbiome consistency in bee larvae growing up in areas of differing land use intensity (Engel et al., 2016) might increase the risk of the bees’ microbiome being more susceptible to and thus invaded by pathogenic environmental bacteria (Voulgari-Kokota et al., 2020; Fernandez De Landa et al., 2023). Interestingly, we also observed higher variability in the bacterial compositions of soil nest enclosures in high and intermediate than low land use intensity sites, which have been described as barriers for pathogen spillover in solitary bee nests (Voulgari-Kokota et al., 2020). In fact, certain bacteria, e.g., Bacillus sp., were present in higher abundances in O. bicornis bee larvae and soil nest closings at highly intensified areas. This genus was the major suspect in causing mortality in O. bicornis larvae in the study of Voulgari-Kokota et al. (2020). These findings underscore the necessity for in-depth investigations of microbial acquisition routes of certain bacterial genera of the distinct developmental stages of various bee species over time (Weinhold et al., 2024). Interestingly, land use intensity did not affect the bacterial composition of O. bicornis bee pupae. This could mean that either such bacteria cannot pass transgenerational filters or that only those larvae that are capable of maintaining or establishing a healthy microbiome, despite external influences, can develop properly and survive metamorphosis (Voulgari-Kokota et al., 2020; Brar et al., 2023). This may also explain (at least partially) why the number of vital larvae and nesting cells of O. bicornis decreased with increasing land use intensity in a previous study at the same sites (Peters et al., 2022).

As proposed here, land use indirectly affects pollen bacterial communities via alterations in plant spectra and collections of environmental microbes via pollen provisions (Leonhardt et al., 2022; Peters et al., 2022). The resulting decrease in bacterial diversity and shifts in community composition can directly affect the invasion potential of pathogenic bacteria or indirectly the larvae’s ability to effectively uptake nutrients. Thus, microbiome shifts at high land use intensity likely result in less healthy larvae and increasing larval mortality. This highlights the complex interaction between land use practices, microbial communities, and bee health, underscoring the importance of considering microbiome dynamics in agriculturally managed landscapes. Untangeling these connections can in turn provide deeper insights into potential mitigation strategies to reduce the negative impacts of agricultural intensification on pollinator populations and ecosystem health.