A short exposure to a semi-natural habitat alleviates the honey bee hive microbial imbalance caused by agricultural stress

Measurement of colony strength traits and Varroa destructor load

Hives located in the natural area (henceforth natural hives or colonies) weighed the most (hive weight, Kruskal–Wallis test, p < 0.0001) (Fig. 1a) yet had the smallest bee population (ANOVA test, p < 0.0001) (Fig. 1d), excluding one colony with higher bee loads (i.e. outlier). Pollen was nearly absent in hives situated in the semi-natural area (i.e. semi-natural hives or colonies) compared with the pollen-rich agricultural (henceforth agricultural hives or colonies) and natural colonies (pairwise Dunn test, p < 0.0001) (Fig. 1b). Brood presence was highest in agriculture (ANOVA test, p < 0.0001) and very similar for natural and semi-natural ambients (pairwise Tukey test, p > 0.05) (Fig. 1e). Mean Varroa load was null and equal in all apiaries (Kruskal–Wallis test, p > 0.05) but two natural colonies had high mite loads (Fig. 1c).

Comparison of bacterial community richness and taxonomic composition between hive niches

In total, 158 samples were adequately amplified and sequenced. After pair-end sequence assembly, quality filtering and singletons removal, our final dataset consisted of 7,962,468 reads. The total frequency of reads was 4,040,931 for gut (mean = 93,975.139 ± 29,760), 1,089,501 for bee bread (mean = 26,573.195 ± 13,298), 2,471,615 for hive entrance (mean = 58,847.976 ± 19,523), and 360,421 for internal air (mean = 11,263.156 ± 9,867).

Hive niche had a significant impact on bacterial community biodiversity (Faith Phylogenetic Diversity, PD), Shannon’s diversity index (H) and Pielou’s evenness index (J′) (Kruskal–Wallis p < 0.0001). Gut samples presented the least diverse (lowest PD and H, p < 0.0001) but most evenly distributed bacterial microbiome of all sample types (highest J′ p < 0.05) (Supplementary Fig. S1).

Proteobacteria were the most abundant taxa (Fig. 2a), being Alphaproteobacteria the most abundant class (in all niches except for gut), followed by Gammaproteobacteria, Bacilli and Actinobacteria (Fig. 2b). Internal hive air samples were overrun by Alphaproteobacteria [86%, represented by genera Sphingomonas (54.9%), Methylobacterium (13.2%), Bradyrhizobium (2.66%), and Phyllobacterium (1.82%)] (Fig. 2c, Supplementary Fig. S2), and only Gammaproteobacteria (5.66%), Bacilli (2.2%) and Actinobacteria (1.4%) classes had relative abundances over 1% (Fig. 2b). Bee bread microbiotas were rich in Acidobacteria (0.83%), Verrucomicrobiae (0.80%), Thermoleophilia (0.69%) and Oxynphotobacteria (0.27%) families (Fig. 2b), and Methylobacterium (15.06%), Acinetobacter (6.79%) and Pseudomonas (1.06%), and shared similar abundances of Sphingomonas, Bradyrhizobium and Phyllobacterium with air samples (Fig. 2c, Supplementary Fig. S2). Hive entrance presented the highest relative abundances of the Actinobacteria (6.81%), Bacteroidia (3.69%) and Blastocatellia (0.21%) families among all hive niches, as well as unknown bacteria (2.92%) (Fig. 2b) and the Arsenophonus (15.64%), Curtobacterium (1.81%) and Hymenobacter (1.21%) genera (Fig. 2c, Supplementary Fig. S2). The bacterial microbiome composition of gut samples was skewed towards Gammaproteobacteria (55.7%) class and the Firmicutes phylum (31.47% Bacilli) presence, and was rich in classes with < 0.1% relative abundance (Fig. 2b). At genus level, over 70% of the gut microbiome was formed by the genera Gilliamella, Lactobacillus and Snodgrasella (Fig. 2c, Supplementary Fig. S2).

Environmental and anthropic effects on diversity in different hive niches

Microbial communities were in general more evenly composed in the natural environment (Pielou’s index, Kruskal–Wallis test, p < 0.05 for gut, p < 0.01 for hive entrance), with semi-natural samples showing intermediate values (Supplementary Fig. S3a). Phylogenetic diversity changed for hive entrance samples, with natural hives having the highest values and anthropic the lowest (Kruskal–Wallis p < 0.05) (Supplementary Fig. S3b). There were also significant differences in Shannon’s diversity when comparing hive entrances located in agricultural versus natural landscapes (Kruskal–Wallis p < 0.01) (Supplementary Fig. S3c).

Bacterial community composition significantly differed by environment for bee bread, hive entrance and gut samples (PERMANOVA, p ≤ 0.01 for all, pseudo-F > 4, 11, 14 respectively) but not for internal air (Supplementary Table S1). Bee bread community differences between environments were significant but weak, with a slight clustering in the PCoA (Supplementary Fig. S4a) and a lack of clustering in the UPGMA (Supplementary Fig. S4b). On the contrary, hive entrance samples showed a differentiated cluster formed by natural samples and a significant division between anthropic and semi-natural environments (pairwise PERMANOVA, pseudo-F = 7.63, p = 0.002) (Supplementary Table S1) reflected in clustering of environments on UPGMA, but not clearly visible in the PCoA (Fig. 3a). However, heterogeneous dispersion by environment was found to be significant (betadisper, F = 9.941, p = 0.001) with the natural group showing intra compositional variance. Worker guts also clustered separately for natural samples and were the only sample type showing a clear microbial clade divergence between agricultural and semi-natural environments (PCo2 = 12.81% and PERMANOVA pseudo-F = 8.86, p = 0.001) (Fig. 3b) (Supplementary Table S1). However, the largest Bray–Curtis distance (an index that measures between samples microbial compositional dissimilarity) was found when comparing either anthropic or semi-natural against natural samples (PCo1 = 42.87% and PERMANOVA pseudo-F > 10, p = 0.001) (Supplementary Table S1).

Functional and bacterial community profiles across environments

Honey bee worker gut samples

All three environments had a strong representation of Firmicutes (Lactobacillus), Gammaproteobacteria (the Orbaceae Gilliamella and Frischella), Betaproteobacteria (Snodgrasella), Actinobacteria (Bifidobacterium) and Alphaproteobacteria (Bartonella, Methylobacterium, Sphingomonas and Bradyrhizobium). However, there were clear differences between natural and agricultural environments. Enrichment of the Enterobacteriaceae (Gammaproteobacteria) and Rhizobiaceae (Alphaproteobacteria) families was detected in the agricultural environment (LDA > 3.6) (Fig. 4a). The Gammaproteobacteria genera Pantoea and Arsenophonus were also enriched in the agricultural environment compared to the other environments, albeit not significantly (Supplementary Table S2). The genera Lactobacillus (Bacilli), Commensalibacter (Alphaproteobacteria) and Snodgrasella (Gammaproteobacteria) represented the natural environment (LDA > 4.6) (Fig. 4a). Interestingly, the Gilliamella genus was underrepresented in natural samples, with overall lower relative frequencies than in the other two environments (median 16.94% in natural, 34.28% in semi-natural and 37.16% in agricultural) (Supplementary Table S2). Semi-natural samples were characterised by a higher presence of Bartonella (sparse in the other environments) and Frischella (median 8.24% versus agricultural 2.91% and natural 3.54%) (Supplementary Table S2). No differently enriched taxa were found in the semi-natural habitat (Fig. 4a), but importantly most agricultural and natural representatives had intermediate abundances in this environment (Fig. 6a).

Concerning bacteria-bacteria interactions, the natural representatives Commensalibacter and Lactobacillus were positively correlated (R = 0.50, p < 0.01). Several genera from the Enterobacteriaceae family (Hafnia-Obesumbacterium, Pantoea and one non-assigned) possessed high negative correlations with Commensalibacter (R < − 0.62, p < 0.001) and more moderate correlations with Bartonella (R < − 0.46, p < 0.05), which was more abundant in semi-natural colonies. Other Enterobacteriaceae (Morganella, Serratia) presented moderate negative correlations with Commensalibacter alone (R < − 0.48, p < 0.05). Overall, Enterobacteriaceae bacteria promoted the presence of other Enterobacteriaceae, especially between bacteria negatively correlated with Commensalibacter. Two Rhizobiaceae were also positively correlated with several Enterobacteriaceae (R > 0.4, p < 0.05). High positive correlations were also observed for several Acetobacteraceae genera, with the highest correlations present among Asaia versus Gluconacetobacter (R = 1, p < 0.0001); and Gluconobacter versus Komagataeibacter (R = 0.70, p < 0.001) (Fig. 4b).

In the predictive functional profile, natural and agricultural environments presented either the lowest or highest recruitment of features. Semi-natural samples generally had intermediate values, indicating a possible “intermediate microbiome” found in semi-natural gut samples. Similarity analysis via PCoA demonstrated clustering of environments along the PCo1 axis (Fig. 4c). All environments had similar biodiversity values, with agricultural samples having the highest H values against both natural and semi-natural (p < 0.001) and no significant differences found between semi-natural and natural samples (Supplementary Fig. S5a). Natural samples indicated significant recruitment of several anabolic reactions for the generation of precursor metabolites, nucleosides and nucleotides, while Arginine biosynthesis and β-D-glucuronoside degradation pathways were more representative of agricultural samples (Fig. 4d). The semi-natural environment retained no significant pathways, even though it had intermediate abundances for every agricultural- and natural-significant pathway detected.

Hive entrance samples

Environmental effects were clear, with only Sphingomonas, Bradyrhizobium and Methylobacterium abundantly present in all environments. Agricultural and semi-natural apiaries were overrun by Proteobacteria and more enriched than natural samples for Firmicutes (Lactobacillus, Staphylococcus, Streptococcus, Paenibacillus). The non-natural apiaries differed in bacterial abundances, rather than presence/absence of bacteria. Gammaproteobacteria (mainly Arsenophonus, Stenotrophomonas and Pseudomonas) were representative of agricultural samples, as was Lactococcus (Firmicutes) (Fig. 5a). Natural samples had a divergent microbial profile, with abundance of both Actinobacteria and Bacteroidia classes. The Aureimonas and Deinococcus genera were significantly enriched in these natural colonies (LDA > 3.5) (Fig. 5a) while Massilia was slightly enriched (Supplementary Table S2), Arsenophonus was absent and Sphingomonas, Phyllobacterium and Bradyrhizobium had overall lower abundances (Supplementary Table S3). The Sphingomonas genus (LDA > 5.0), the Bacilli class, several Proteobacteria genera, as well as the phyla Firmicutes, Gemmatimonadetes and Actinobacteria were the most significant for semi-natural samples (Fig. 5a). Two genera significant for agriculture, Arsenophonus and Lactococcus, showed intermediate abundances in semi-natural samples (Fig. 6c). Species wise, Paenibacillus larvae and Lactobacillus kunkeei (both Bacilli) plus Corynebacterium afermentans sub. sp. afermentans (Actinobacteria) were semi-natural representatives (LDA > 3.0), with P. larvae present in natural samples and practically absent in agricultural hives, while L. kunkeei and C. afermentans were present in agricultural hives but absent in natural hives (Supplementary Table S4).

Concerning interactions among bacteria, correlations were mostly grouped by taxa. Actinobacteria and Bacteroidia such as Jatrophihabitans, Curtobacterium or Hymenobacter promoted mutual presence and vice versa, while the presence of Alphaproteobacteria was promoted by different Firmicutes (Spearman correlation, R > 0, p < 0.05). Exclusion, marked by negative correlation, was detected between several Actinobacteria or Bacteroidia versus genera such as Arsenophonus, Corynebacterium 1, Micrococcus and Gaiella (Fig. 5b).

In the predictive functional profile, natural hives clustered together (explaining 19.53% of dissimilarities) while agricultural samples scattered across the PCo1 axis (Fig. 5c). Diversity of functional profiles was highest in natural hives and lowest in semi-natural (Shannon’s index, p < 0.05) (Supplementary Fig. S5b). None of the significantly recruited functions were exclusive to one environment. The only non-ubiquitous function was lipopolysaccharide (LPS) biosynthesis (absent in natural colonies) (Fig. 6d), although certain pathways were also scarce in the other environments (pyrimidine biosynthesis III in agricultural gut and (Kdo)2-lipid A biosynthesis in natural hive entrance) (Fig. 6b,d). Natural colonies exhibited increased frequency of the Bifidobacterium shunt, l-methionine biosynthesis (mostly mediated by transsulfuration occurring from oxaloacetate), and a tricarboxylic acid cycle specific to acetate-producers (TCA cycle VII) (Fig. 5d). Agricultural hives possessed enriched synthesis of nucleotides, cofactors, nicotinamide adenine dinucleotide (NAD), membrane components (Kdo2 lipid A, LPS, mycolate) and fatty acids. The tRNA processing pathway resulting in tRNA activation, and the stringent specific ppGpp metabolism were also significantly expressed (Fig. 5d). Semi-natural samples had significant recruitment of the tryptophan 7-halogenase enzyme (Fig. 5e) and of anaerobic gondoate biosynthesis (Fig. 5d) and intermediate values for other pathways enriched in the more extreme environments (Fig. 6b,d).

Bee bread samples

Contrasting environments shared similar microbiomes, with differences primarily found in abundances of scarce taxa. Sphingomonas and Methylobacterium (Alphaproteobacteria) were overall the most abundant genera, followed by Acinetobacter in the natural environment and Bradyrhizobium in the other two (Supplementary Table S2). The gut core genus Bombella was enriched in natural hives, Arsenophonus represented the agricultural environment, and no taxa was augmented in semi-natural samples (LDA > 3.0) (Supplementary Fig. S4c).

Internal hive air

Environmental effects were scarce for internal air, and only seen in the enrichment of Enterobacteriaceae (mostly Arsenophonus), Curtobacterium and Massilia in agricultural samples (Supplementary Fig. S6).

Intermediate functional and bacterial community profiles in semi-natural hives

Beta-Diversity results of gut bacterial communities and predicted functions showed that semi-natural samples clustered between natural and agricultural hives (Figs. 3b, 4c). This effect was also apparent for predicted functions of hive entrance (Figs. 3a, 5c) samples. In concordance, semi-natural hives showed intermediate relative abundances for several taxa and functional pathways, while natural and agricultural environments exhibited either lowest or highest relative abundances (Fig. 6). This trend was more evident at a functional (Fig. 6b,d) than at the taxonomic level (Fig. 6a,c). Bacterial representatives showing intermediate abundances for gut samples included Comensalibacter, which was enriched in natural, scarce in semi-natural and mostly absent in agricultural samples. An unknown Rhizobiaceae genus exhibited the opposite tendency (enrichment in agriculture and absence or nearly absence in natural samples) (Fig. 6a). Same pattern was observed for Lactococcus in hive entrance (Fig. 6b) and Arsenophonus in hive entrance (Fig. 6b), bee bread (Fig. 6e) and gut samples (Supplementary Table S2). A less pronounced transition was detected for Lactobacillus in gut samples and Pseudomonas in the hive entrance. Both were present in all environments but augmented in natural and agricultural environments, respectively, while the semi-natural environment had intermediate abundances. Bee bread and gut samples also showed overall intermediate abundances for some non-environmentally relevant bacteria, such as Bradyrhizobium in bee bread and Gilliamella in gut (Supplementary Table S2). Besides the aforementioned bacteria showing intermediate abundances, all significantly recruited functions in gut and hive entrance samples had highest or lowest recruitment in natural and agricultural colonies, and intermediate values in the semi-natural environment (Fig. 6b,d). This behaviour was clear in the agricultural recruitment of NAD and Kdo2-lipid A synthesis, both displaying relative mean abundances under 0.001% in natural colonies (Fig. 6d). The exceptions to the rule were the gondoate anaerobic synthesis enriched in semi-natural hive entrance and the Bifidobacterium shunt with equally low relative abundances in both anthropized locations (Fig. 6d).

Source: Ecology - nature.com