Orchard Management and Landscape Context Mediate the Pear Floral Microbiome

Our study sampled bacterial and fungal communities associated with pear flowers across 15 orchards with three management types (conventional, bio-based IPM, and organic; 5 sites of each). After bacterial (16S) and fungal (internal transcribed spacer [ITS]) gene sequencing, quality filtering, and processing, we detected 142 bacterial and 1,703 fungal amplicon sequence variants (ASVs) from the pear flowers. The bacterial community was dominated by members of Bacillaceae, Enterobacteriaceae, Lactobacillaceae, and Pseudomonadaceae (Fig. 1A), with each family comprising, on average, 22% (standard deviation [SD], ±0.32), 15% (SD, ±0.03), 9% (SD, ±0.4), and 9% (SD, ±0.05) of sequences, respectively. Beneficial bacteria associated with disease suppression in this system (i.e., Bacillus, Pantoea, and Pseudomonas) comprised 11% of taxa (ASVs) observed and 41% of the relative abundance. The fungal community was dominated by members of Aureobasidiaceae, Cladosporiaceae, Mycosphaerellaceae, and Sclerotiniaceae (Fig. 1B), with each family comprising, on average, 16% (SD, ±0.10), 8% (SD, ±0.02), 14% (SD, ±0.03), and 7% (SD, ±0.05) of sequences, respectively. Of the Aureobasidiaceae, four ASVs were identified to the species level as Aureobasidium pullulans, a beneficial fungus used for biological control of fire blight. Twenty-one additional ASVs were identified as belonging to genera Botrytis, Cladosporium, Monilinia, Mycosphaerella, or Penicillium, potentially important agents of pre- and postharvest disease.

Orchard pest management practices were significantly associated with pear flower bacterial and fungal diversity (Shannon index) (Table 1). Considered alone, conventional and biological-based integrated pest management (bIPM)-managed orchards were found to have ∼60% higher bacterial diversity than those managed organically (Fig. 2A), while organically managed orchards exhibited the highest fungal diversity (Fig. 2A). Yet the positive effects of organic management on fungal diversity were not significant in the multiple variate linear model when controlling for land cover and climate (Fig. 2B). In these linear regression models (Table 1), both organic and bIPM management styles reduced bacterial and fungal diversity, although the negative influence of organic management on fungal diversity was weak. Land cover was also associated with bacterial and fungal diversity: bacterial diversity declined with increasing proportion of habitat containing forest or pear, while fungal diversity increased with pear crop cover (Table 1). Microclimatic conditions were also associated with both bacterial and fungal diversity, though minimum temperature was the only variable of significant effect on fungal diversity and minimum vapor pressure deficit (VPD) for bacterial diversity in the top Akaike information criterion (AIC)-selected model (Table 1).

Focal bacterial and fungal genera of concern were first investigated to assess the scale of spatial autocorrelation, as well as potential associations with aspects of landscape context. Positive spatial autocorrelation was exhibited for each of the nine taxa examined, but only at the shortest distances of less than 1 km. Using canonical correlation analysis to assess how landscape and management variables were associated with the microbial species composition, we found significant associations between predictors and bacterial (Fig. 3A; Table S1 in the supplemental material; Pillai’s trace P = 0.014) and fungal (Fig. 3B; Table S2; Pillai’s trace P = 0.005) communities. The three bacterial genera associated with disease suppression were distributed very differently in association with the factors of interest. More specifically, the relative abundance of Bacillus, bacteria commonly applied to suppress disease in pear, was most strongly associated with organic management (Fig. S2), followed closely by the amount of surrounding forest and then geographic distance. These top factors, aligned with axis 1, were negatively associated with Pseudomonas, while bIPM was the most important predictor of Pantoea (more aligned with axis 2). Similar to Pantoea, bIPM (+) and organic management (−) best predicted the presence of Aureobasidium, a beneficial fungus aligned with axis 1, and Monilinia to a lesser degree. Minimum temperature (+) and minimum VPD (+) best predicted Botrytis, Cladosporium, and Mycosphaerella, as well as Monilinia (−), all pathogenic fungi of concern for pears. Finally, the proportion of forest in the landscape and geographic distance were associated with the distribution of these fungal genera of interest (Table S2).

Orchard sites that were closer in proximity, or had the same management scheme, tended to be more similar in terms of bacterial community composition (Fig. 4; Table 2). In contrast, abundance-related turnover across sites was affected mainly by the proportion of landscape under fruit cultivation, namely, apple (Table 2). With respect to fungi (Fig. 4; Table 3), turnover of fungal communities across sites was associated with the amount of pear production in the landscape, temperature, and vapor pressure deficit (VPD). Temperature and VPD, along with surrounding forest, were important drivers of taxon-related turnover (Table 3). In contrast, abundance-related community turnover was associated with geographic distance and the proportion of landscape represented by forest around orchards (Table 3).

Bacterial and fungal alpha diversity responded differently to orchard management schemes. Bacterial diversity was significantly higher in conventional and bIPM orchards than organic orchards; however, the opposite pattern was observed for fungi. Organic orchards had a high relative abundance of Bacillus, likely because of its application as a biological control agent. The strong effect of orchard management on bacterial diversity suggests that application of Bacillus reduced bacterial diversity, which may occur through resource competition, priority effects, or mass effects. Bacillus species have shown promise in limiting the establishment and development of the bacterial pathogen E. amylovora, the causal agent of fire blight (22, 23), and may also affect other floral microbes. Indeed, increased fungal diversity in organically managed orchards could be a consequence of Bacillus application, although we were unable to directly assess if fungal abundance was affected in our study. In contrast to bacteria applied for biological control, we observed that Aureobasidium had a higher relative abundance in conventional and bIPM orchards than organic ones (where it was applied in one orchard for biological control). Background levels of some microbial taxa may be high and more prevalent in the presence of particular landscape and climate conditions (e.g., higher precipitation and high proportion of forest; Tables S1 and S2 in the supplemental material). These patterns may represent preferential use of these biological treatments across orchards in our survey. Though unable to confirm whether ASVs recovered in our data set are these exact commercial strains, biologicals applied to pear flowers often have a high recovery rate in surveys (24, 25).

Our results show that habitat patches with alternate tree fruit crops (apple, cherry) were negatively associated with both bacterial and fungal diversity on pear flowers and appeared to be primary drivers of microbial community structure (Tables 2 and 3). Pear orchards in the Wenatchee River Valley are primarily located in narrow, intermountain areas with highly variable elevation and land cover, including forest, additional pear orchards, and those dedicated to production of other deciduous fruits, namely, apple. Vegetation in and around orchards can be an important source of inocula via airborne dispersal (26, 27). Furthermore, previous work on apple and pear flowers has revealed considerable overlap in the identity of microbes associated with each host species (28–30). Such overlap, in addition to a reduction in diversity with increasing land cultivation, suggests a role for several key processes in shaping floral microbiomes in tree fruits. First, there is a high degree of shared usage of disease and pest management practices employed in pear and apple production systems, as both can suffer greatly from fire blight disease. Inputs applied in conventional and bIPM orchards, including antibiotics and fungicides (Table S3), can act as strong environmental filters on potential floral colonists (17, 18) or serve as a source for inocula when applied as biologicals, as observed in organic orchards. Second, both apple and pear systems rely considerably on honeybees (Apis mellifera) for pollination, which are known to leave a distinct imprint on floral microbiome diversity (31). Increased reliance on a single-pollinator species, combined with chemical and nonchemical inputs, are likely important contributors to patterns observed.

Orchard management scheme was a key determinant of bacterial community similarity across sites; however, other predictors often explained high levels of variance in community structure across sites. In particular, geographic distance explained a significant amount of variance in both whole-community and taxon-related beta diversity of bacteria. In contrast, for fungi, geographic distance was a significant predictor of only abundance-related turnover. Beyond geographic distance, climatic conditions also contributed significantly to explained variance in the beta diversity or community turnover of fungal communities. In particular, VPD and temperature were negatively associated with fungal diversity, suggesting both microclimate variables affect either species-specific patterns of growth and/or competition. Moisture availability is also an important determinant of microbial growth on the surface of plant tissues (32), with free water and humidity often being necessary for conidial germination, germ tube growth, and potential penetration of plant tissues, including floral organs. This has been frequently observed in other flowering systems of commercial value, including blueberries (33), raspberries (34), strawberries (35), and cut roses (36). Within these systems, infection of the gynoecium can be a primary route of disease development. Alternatively, infection of petals and other organs can facilitate secondary infections of fruits (37). Of the fungal genera examined in our study, Botrytis has been documented to successfully infect the mesocarp via stamen filaments (38). For the others of interest, it is unclear if there is a link between flower colonization and resulting development and pre- and postharvest diseases.

More broadly, our results provide insight into local- and landscape-level drivers of floral microbiome diversity in an important tree fruit commodity, pear. Given the critical link between flowers, yield, and disease, identifying such drivers across both spatial and temporal scales could improve the understanding of links between management, host microbiome structure, and potentially disease resistance or susceptibility. With growing appreciation for the role of host microbiota in affecting resistance against disease (3, 39), such information has potential to inform development of sustainable management practices in many different types of agroecosystems.

We surveyed 15 orchards throughout the Wenatchee River Valley of central Washington (Fig. 5) in spring 2018. Within the United States, Washington State is the leading producer of deciduous tree fruit crops such as apples, pears, and cherries. These, as well as other commodities, are grown in variable intermountain river valleys and basins east of the Cascade Mountains. These production areas generally experience temperate, dry conditions, in addition to favorable access to irrigation water originating from streams and rivers fed by snowmelt (40). Given the diverse topography of this region, however, individual orchards range in elevation from 20 to 1,000 m above sea level (40). Key stages of fruit production, such as flower bloom, can thus experience considerable variation in microclimatic conditions among orchards, affecting bloom timing, fertilization, and fruit development (41, 42). As flowers are a habitat for diverse microbiota (43), including a number of pathogenic species that cause pre- and postharvest diseases of tree fruits (44), microclimatic conditions could affect habitat quality, as well as colonization dynamics and the resulting structure of the floral microbiome.

Our survey assesses microbe communities in orchards that used one of three management schemes, with five replicates per scheme, which include organically certified, conventional, and biological-based integrated pest management (bIPM) (45). With each of these broad management types, growers were not restricted to a specific spray schedule, but each used a defined set of tools for pest and disease management (45). Conventional management followed a standard practice (e.g., application of synthetic pesticides), while organic orchards were all managed following USDA‐certified organic standards, which prohibits use of such synthetic chemicals. To control fire blight, organic producers often use Serenade Opti (Bayer Crop Science, St. Louis, MO, USA) at full bloom, a bio-based fungicide and bactericide that leverages Bacillus subtilis (strain QST 713) endospores and its metabolic by-products as active ingredients (22). Serenade is not the only bio-based product leveraged by producers for control of fire blight in pear, however, and other products such as Blossom Protect (Westbridge Agricultural Products, Vista, CA, USA) can be used across organic, bIPM, and conventional schemes. Blossom Protect is derived from air-dried spores of Aureobasidium pullulans (strains DSM 14940 and 14941) (25), an epiphytic or endophytic fungus associated with a wide range of plant species, including many tree fruits. For those orchards that employed the bIPM scheme, growers used a toolbox of cultural controls combined with pesticides with less documented negative impact on natural enemies and other beneficial organisms. Such products included lime sulfur, kaolin, spinosad, and biologicals applied at various stages of bloom (45).

Orchards were sampled once at peak bloom, either on 30 April or 1 May of 2018. At each orchard, 10 trees (Bradford variety) were sampled, 5 near the edge of the orchard and 5 in the interior. We chose this approach because previous studies suggest that seminatural habitat in the surrounding landscape can both support and increase rates of visitation by native pollinators such as bees and flies (46). Moreover, pollinators can be important dispersal agents for microbes (31, 47); thus, our aim was to detect potential contributions of pollinator visitation to flower microbiome assembly in orchards. For each site (i.e., edge or interior) and sampling event, 50 open flowers (n = 10 per tree) were collected using aseptic technique and pooled at the site level. Flowers with flat, fully reflexed petals that had been open for ∼3 days were collected. Once collected, flowers were placed in a cooler, transferred to the lab, and then stored at 4°C until processing.

In a laboratory, whole flowers were washed with 20 ml of 1× 0.15% phosphate-buffered saline (PBS)-Tween solution, and samples were sonicated for 10 min to dislodge epiphytic microbes. After sonication, floral tissue debris was removed from sample tubes by pouring samples through autoclaved cheesecloth into a new, sterile Falcon tube. Falcon tubes containing debris-filtered samples were centrifuged at 3,000 rpm for 10 min at 4°C to pellet microbial cells. We poured off the supernatant, resuspended microbial cell pellets in 1 ml of autoclaved PBS solution, vortexed tubes, and then transferred the cell suspensions to new 1.7-ml microcentrifuge tubes.

Genomic DNA was extracted from samples using a ZymoBIOMICS DNA microprep kit (Zymo Research, Irvine, CA, USA) following the manufacturer’s protocol. Extracted DNA was then used as the template for library preparation and amplicon sequencing following Comeau et al. (48), performed at the Centre for Comparative Genomics and Evolutionary Bioinformatics at Dalhousie University (Halifax, Nova Scotia, Canada). There, amplicon fragments were PCR- amplified from DNA in duplicate, using separate template dilutions (1:1 and 1:10) and high-fidelity Phusion polymerase (New England BioLabs Inc., Ipswich, MA, USA). A single round of PCR was performed using “fusion primers” (Illumina adaptors plus indices plus specific regions) targeting either the 16S V4-V5 (bacteria; primers 515FB and 926R [49, 50]) or ITS2 (fungi; primers ITS86 and ITS4 [51]) regions with multiplexing. PCR products were verified visually by running a high-throughput Invitrogen 96-well E-gel (Thermo Fisher Scientific Corp., Carlsbad, CA, USA). Any samples with failed PCRs (or spurious bands) were reamplified by optimizing PCR conditions to produce correct bands to complete a sample plate before continuing with sequencing. The PCRs from the same samples were pooled in one plate, cleaned, and then normalized using the high-throughput Invitrogen SequalPrep 96-well plate kit (Thermo Fisher Scientific Corp.). Samples were then pooled to make one library and then quantified fluorometrically before sequencing. Amplicon samples were then run on an Illumina MiSeq using 2× 300-bp paired-end V3 chemistry.

Demultiplexed sequences were trimmed of trailing low-quality bases using the DADA2 pipeline (v1.8.0) (52) in R (53). Paired-end reads were then quality filtered, error corrected, and assembled into ASVs. Once assembled, chimeras were detected and removed, and taxonomic information was then assigned to each ASV using the Ribosomal Database Project (RDP) naïve Bayesian classifier (54) trained to either the RDP training set (v14) or UNITE general FASTA release (v7.2) for bacteria or fungi, respectively. ASVs that failed to classify to kingdom or identified as chloroplast or mitochondrial sequences were discarded. Further, potential contaminant ASVs were identified through inclusion of negative controls during sample and sequence processing and then removed using the “prevalence” method with the decontam package in R (55). This filtering resulted in samples sequenced at a mean depth of 43,057 sequences per sample for bacteria and 25,890 for fungi. Samples were then rarefied (bacteria, 49; fungi, 14,920), with all but one bacterial sample (19orgedge) retained in the analyses that follow. Such a low cutoff for bacteria is a consequence of a large proportion of reads being identified as plastid DNA, which were removed from the data set. Despite this, we included bacterial data in our study because sampling curves indicate that we were able to identify the majority of bacterial taxa present in samples (Fig. S2 in the supplemental material). Moreover, previous characterization of microbial communities associated with flowers has frequently observed low species richness (43).

Land cover within a 1-km buffer of each study orchard was classified into the following three habitat types: (i) pear orchard, (ii) other fruit orchard (apple and cherry), and (iii) forest. These classifications were performed using the Cropland Data Layer spatial product (https://www.nass.usda.gov/Research_and_Science/Cropland/SARS1a.php). Across our study region, pears were the dominant agricultural crop, although the habitat around individual study sites varied widely from 2% to 66% pear orchards. Other fruit crops had less variability, with 0% to 6% of surrounding land cover, while forest land was highly variable and ranged from 0% to 46%. Forest patches were primarily composed of evergreen trees.

To assess the role of abiotic factors, high-resolution climatic metrics for each site were obtained from publicly accessible PRISM data in April 2018. PRISM data are collected at a spatial resolution of 2.5 arcmin (∼4 km). An arcmin is an angular measurement equal to 1/60 of a degree. PRISM data used included elevation (m), minimum and maximum temperature (°C), minimum and maximum vapor pressure deficit (VPD [hPa]), and precipitation (mm). Vapor pressure deficit is the difference between the amount of moisture in the air and how much moisture the air can hold when saturated, where high VPD indicates drier conditions. As with land cover, the abiotic conditions where sites were located were variable, with elevation ranging from 1,152 to 1,526 m above sea level, April precipitation ranging from 4.2 to 5.3 cm, minimum temperatures ranging from 2.4 to 3.7°C, and maximum temperature ranging from 13.6 to 15.7°C.

We used multivariate linear regression to assess effects of land cover, orchard management, and climate on the alpha diversity of pear flower microbiomes, using both the Shannon diversity and inverse Simpson index. We chose to include the latter metric to specifically isolate the evenness/dominance aspect of community structure from the taxonomic richness, which heavily contributes to the Shannon diversity metric. All analyses were conducted using R v3.6.1 (53). To reduce multicollinearity among predictors, we calculated variance inflation factors (VIFs) and used a threshold of 10 to eliminate variables with problematic covariance. This eliminated temperature, precipitation, and elevation from the alpha diversity models. We calculated multimodel average coefficients based on the 90% confidence interval of top models as well as the importance of each coefficient, which indicated the number of top models in which it appeared.

We also assessed effects of landscape, climate, and farm management on the dominance (relative abundance) of a few focal genera that are highly important for pre- and postharvest diseases of pear, including putative pathogens and beneficial taxa. These included fungal genera Aureobasidium, Botrytis, Cladosporium, Monilinia, Mycosphaerella, and Penicillium and beneficial bacteria, which included Bacillus, Pantoea, and Pseudomonas (24, 56, 57). One ASV (BactSeq29), identified as an Erwinia sp., was detected at a single orchard in our survey. Given such limited detection, we were unable to perform an analysis of links between variables of interest and Erwinia presence and abundance. However, to examine associations between microbial genera and predictors described earlier, we used canonical correlation analysis (CCA), an extension of linear regression that finds linear relationships between combinations of explanatory and response variables which maximize the correlation. Separate models were run on fungi and bacteria of interest.

Differences in species composition among sites could be affected by processes including substitution of taxa and variation in abundance of particular taxa, so we further evaluated the effects of farm management, land cover, and climate variables on abundance-related and taxon-related aspects of community turnover (microbial beta diversity) and the overall community dissimilarity (which incorporates both processes). Beta diversity was partitioned into abundance-related and taxa-related components of Bray-Curtis dissimilarity using the bray.part function in the betapart R package (58). The influence of explanatory variables on these two components of community turnover between sites, as well as their cumulative overall Bray-Curtis dissimilarity, was investigated using restricted distance-based analysis (RDA) and AIC model selection and executed using the capscale and ordiR2step functions in the vegan R package (59). The variance explained by factors included in the top AIC-selected models is included in the results.

Raw amplicon sequences are available on the NCBI Sequence Read Archive (SRA) under BioProject accession no. PRJNA659266.