The relative abundances of yeasts attractive to Drosophila suzukii differ between fruit types and are greatest on raspberries

Fungal metabolic volatiles attract Drosophila suzukii which oviposits in ripening fruits, but there are few data describing the fungal microbiomes of commercial fruits susceptible to this insect pest. We tested the hypothesis that fruit type and ripening stage have a significant effect on fruit surface fungal communities using DNA metabarcoding approaches and found strong support for differences in all three fungal community biodiversity metrics analysed (numbers, types, and abundances of taxa). There was an average fivefold greater difference in fungal communities between sites with different fruit types (strawberry, cherry, raspberry, and blueberry) than across fruit developmental stages, demonstrating site and/or fruit type is the greater factor defining fungal community assemblage. The addition of a fungal internal standard (Plectosphaerella cucumerina) showed cherry had relatively static fungal populations across ripening. Raspberry had a greater prevalence of Saccharomycetales yeasts attractive to D. suzukii, including Hanseniaspora uvarum, which aligns with reports that raspberry is among the fruits with greatest susceptibility and attraction to D. suzukii. Greater knowledge of how yeast communities change during fruit maturation and between species or sites may be valuable for developing methods to manipulate fruit microbiomes for use in integrated pest management strategies to control D. suzukii.


Results
Six biological replicates each were sampled from four fruit species (blueberries, cherries, raspberries, and strawberries) at four developmental stages. Developmental stages were based on fruit pigmentation ranging from unripe (green) to fully ripe (red/purple/navy; Fig. S1) throughout June to September in 2018. Ten fruits (except blueberries N = 20) were collected for each species per replicate, and this was replicated six times for each ripening stage for each fruit at different sites.
Quantitative analysis of fungal communities. Metabarcoding analysis is generally not quantitative, but the addition of 265 P. cucumerina cells to sub-samples prior to DNA extraction served as an internal standard to attempt an estimation of the size of fungal populations. One replicate spiked with the internal standard of the strawberry stage 3 samples was removed due to poor sequence quality leaving 96 non-spiked and 95 spiked samples which produced a total of 38,445,395 reads that clustered into 1712 > 97% identity Amplicon Sequence Variants (ASV), which from here-in we call phylotypes (Table S1). Blast searches across all phylotypes for matches to the P. cucumerina internal standard's ITS sequence generated from Sanger sequencing revealed one phylotype that matched with 100% identity. Plectosphaerella cucumerina was naturally present in 21 of the 95 non-spiked samples and comprised of a total of 444 reads. Cherry was the only fruit where the internal standard was reliably recovered: 23 of 24 spiked samples and only one of 24 non-spiked samples contained the internal standard phylotype. After internal standard DNA read normalisation, the mean (± SE) number of fungal cells from each of the useable 23 pairs of cherry replicates was 307,323 (± 39,090) cells. The range of phylotype cell abundance across all cherry samples was 3.9 million for an Aureobasidium phylotype to 3 cells for a phylotype taxonomically assigned no higher level than kingdom. There was no significant change in total fungal cell numbers across cherry maturation stage (Kruskal-Wallis, chi-squared = 2.63, P = 0.45; Fig. S2), but fruit surface areas also increased significantly (Kruskal-Wallis, chi-squared = 19.70, P = 0.0002, Fig. S2). When cell numbers were normalised for surface area this revealed that absolute fungal population sizes remained static across cherry maturation stages (Kruskal-Wallis, chi-squared = 2.49, P = 0.48; Fig. 1A). However, there was a significant change in absolute Saccharomycetales cell numbers when normalised for cherry surface area across maturation (Kruskal-Wallis, chi-squared = 15.30, P = 0.002): stage 1 had significantly greater absolute Saccharomycetales cell numbers than stage 4 (P = 0.0007; Fig. 1B). Six individual Saccharomycetales yeast phylotypes from the genera Debaryomyces, Saccharomyces, Kodamaea, one from the family Pichiaceae, and phylotypes with > 97% www.nature.com/scientificreports/ then Pichia (5.2%), with the remaining genera contributing fewer than 3% each. Candida was the most diverse genus within the order Saccharomycetales accounting for 21.8% of phylotypes, despite only comprising 2.4% of reads, followed by Metschnikowia (11.5%), Hanseniaspora (8.0%) and Pichia (6.9%), with each of the remaining genera contributing fewer than 3.5% of phylotypes each (Fig. S3B). The most common Saccharomycetales yeast across all samples was a phylotype from the genus Hanseniaspora with > 97% homology to H. uvarum and comprised 38.2% of the total Saccharomycetales reads (Fig. S3B).
The effect of fruit species and ripening stage on epicarp fungal communities. We analysed differences in three biodiversity metrics to evaluate the effect of fruit species and maturation stage on fungal communities: differences in the absolute numbers of phylotypes (richness); differences in the types of phylotypes (i.e. presences/absences); and differences in the relative abundances of phylotypes (community composition) following Morrison-Whittle et al. 14 and Morrison-Whittle and Goddard 37 .
The similarities and differences of fungal phylotypes. The core fruit fungal microbiome. Analyses thus far have focussed on differences in fruit microbiomes, but it is valuable to contrast this with quantifying fruit microbiome similarity. The core fruit fungal microbiome (i.e. those phylotypes present across all fruits) consisted of 199 (11.6%) of the 1712 fungal phylotypes and comprised 97.6% of DNA reads (Table S11). Approximately 12-22% of the 1712 phylotypes were only found associated with specific fruits: 216 with blueberry, Figure 2. Number of observed phylotypes across fruit types and maturation stages. Number of fungal phylotypes across four ripening stages (1, unripe/green fruit; 2, de-greening fruit; 3, ripening fruit; and 4, fully ripe/harvest fruit) for blueberry, cherry, raspberry and strawberry (N = 12 except N = 11 for strawberry stage 3). Numbers of fungal phylotypes differ across ripening stages for cherry, raspberry and strawberry but not blueberry (ANOVA, P values shown). Where significant, different lowercase letters represent significant differences in phylotype numbers within each fruit (P < 0.028) with separate Dunn's comparisons post-hoc (with Benjamini-Hochberg multiple comparison correction). Different letter groups show any significant differences between ripening stages within each fruit separately.  (Table S11), with 3 unique to blueberry, 5 to cherry, 25 to raspberry and 15 to strawberry (Fig. 5B, Table S12).
The phylotypes that are most differentially abundant. Analyses across all biodiversity metrics show fruit type had a greater effect on fungal communities than maturation stage. Overall, 195 (11.4%) indicator phylotypes (spanning 76 families) had significantly differential abundances between fruit types: 33 phylotypes were significantly overrepresented on blueberry, 70 on cherry, 39 on raspberry and 53 on strawberry (FDR corrected P values ranging from P = 0.011 to P = 0.044). The complete list of significantly differentially overrepresented phylotypes is shown in Table S13 but the two most significantly differentially overrepresented phylotypes on each fruit are listed here: Polyphialoseptoria species and Ramularia (most likely Ramularia endophylla) on blueberry; Exobasidium species and a phylotype from the poorly described order Leotiomycetes on cherry; phylotypes with > 97% homology to Metschnikowia kunwiensis and H. uvarum on raspberry; and phylotypes with > 97% homology to Kalmanozyma fusiformata (Ustilaginaceae smut fungi) and Podosphaera aphanis on strawberry. Twenty-four of the 195 indicator phylotypes belonged to the Saccharomycetales budding yeasts (Table S13). There were no Saccharomycetales indicator phylotypes for cherry, and just one for blueberry, a fungal phylotype . NMDS plots representing the differential presences of fungal phylotypes. Nonmetric Multidimensional Scaling (NMDS) plots of binary Jaccard measures of community dissimilarity of (A) total fungal communities and (B) Saccharomycetales budding yeasts on blueberry (blue), cherry (purple), raspberry (green) and strawberry (red) at four ripening stages (1, unripe/green fruit; 2, de-greening fruit; 3, ripening fruit; and 4, fully ripe/harvest fruit; denoted by shade of colour, lightest shade for green fruit and moving through to darkest shade for fully ripe/harvest). Both total fungal and Saccharomycetales yeasts communities significantly differ in the presences of phylotypes across all fruit types (FT) and ripening stages (RS) by PermANOVA (values shown top right).  Fig. 6.
Correlations with fruit host potential index (HPI) scores. Finally, Bellamy et al. 38 generated fruit host potential index (HPI) scores from interactions of D. suzukii with commercial ripe fruit including the fruit species analysed here. The combined relative abundances (i.e. the total number of reads on each fruit across replicates) of yeast phylotypes empirically shown to be attractive to D. suzukii (Hanseniaspora, Pichia, Saccharomyces, Candida and Metschnikowia 27-31 ) across these different ripe fruits at the last sample point are positively correlated with fruit HPI scores (Pearson's correlation r = 0.38), as are the relative abundance of just H. uvarum (r = 0.62). The relative abundance of B. cinerea was negatively correlated to HPI scores (r = − 0.65) (Fig. S11); however, none of these correlations were significant (P > 0.35) likely due to the low number of comparisons (N = 4 due to just one HPI score per fruit type).

Discussion
Drosophila suzukii is attracted to fungal volatile chemicals (e.g. 27 ); however, little is known about the fungal microbiomes of commercial fruit, with a paucity of information for D. suzukii susceptible fruit. Here we tested the hypothesis that both fruit type and maturation stage have a significant effect on total fruit fungal communities as well as Saccharomycetales yeasts and found strong support for this for all three community biodiversity metrics analysed (numbers, types, and abundances of phylotypes). Raspberry had the greatest relative abundance of yeasts known to be attractive to D. suzukii. Overall, there was a fivefold greater difference in fungal www.nature.com/scientificreports/ communities between fruit types than maturation stages, showing fruit type was the greater factor defining fruit fungal community assemblage, and cherry had the most distinctive fungal microbiome (Fig. 5). However, there are two main caveats to these conclusions which need consideration. First, we note that fruits matured across different absolute time periods meaning the absolute timing of sampling the various maturation stages differed between fruits, and there is some evidence that other fruit microbiomes can differ through time 39,40 . Second, fruit fungal communities have been shown to differ by geographic location across hundreds of kilometres 6,10,12 and at smaller scales: for example, fungal community dissimilarity increased with distance on grapes from six vineyards separated by a maximum of 35 km 11 . Therefore, as different fruit sampled in this study were from separate locations up to 19 km apart, it is possible that the variance in fruit-associated fungal microbiomes was also influenced by geographic location. There is support for greater microbiome differentiation by distance from a simple correlation of geographic and community differentiation distances (P = 0.001; Mantle test on distance in km and Jaccard distance); however, distance does not completely explain the variation in fruit microbiomes as cherry and raspberry fungal communities have the greatest dissimilarity (are most separated on NDMS 1 in Fig. 4) but derive from some of the most closely geographically situated sites (Fig. S12). Different fruits would have to be sampled from immediately adjacent sites to completely discount any effect of geographic location on Figure 6. Dynamics of changes in the proportion of budding yeast indicator phylotypes. Mean proportion of reads for the Saccharomycetales budding yeast indicator phylotypes that are significantly overrepresented on (A) raspberry and (B) strawberry (P < 0.04) across the four ripening stages (1, unripe/green fruit; 2, de-greening fruit; 3, ripening fruit; and 4, fully ripe/harvest fruit). Indicator phylotypes are reported to the taxonomic level assigned: lower case letter refers to the taxonomic hierarchy of respective taxa (g = genus; f = family; k = kingdom). Where possible, assignment to genus taxonomic levels is shown in parentheses from matches to deposits in Genbank with > 97% homology identified by manual Blast searches. www.nature.com/scientificreports/ fungal microbiomes. Additionally, while all sites were under conventional management approaches, the precise details of spray programmes are commercially confidential, so it is possible that sites with different fruits were treated with differing spray programmes to control for pest and diseases including fungicides, which may have influenced fungal microbiomes. Some evidence from wine grapes in New Zealand suggests the differences between conventional and biodynamic management only has a small effect on fruit fungal microbiomes 14 . However, other studies on grapes and pear flowers reported that management practices had a significant effect on total culturable yeasts as well as on community structure 13,41 . Taken together, the effect of fruit-type detected in this study is likely to be a composite effect of complex interactions of fruit-type × location × management practices and further study on the same fruit species across multiple locations would be necessary to confirm the extent to which fruit species impact microbial communities. Another caveat is that the inference of fungal biodiversity here is derived from the analysis of DNA, and this may not necessarily correlate with phylotypes which are active in communities, and the complementary analyses of RNA may provide an insight into this. Regardless of the above caveats, these findings show differences in fungal communities on commercial fruit in space and time, and this also holds for species implicated in the attraction of the D. suzukii insect pest. These findings are in line with the few other studies in this field which have shown that fungi differ significantly between apples and blackcurrants 4 , as well as between sea buckthorn, black chokeberry, red and white currants 3 . While differences in fungal communities across ripening stages were smaller here, they still changed significantly, especially in the types and abundances of phylotypes, and this agrees with the very limited data from a few other studies evaluating the dynamics of microbiomes as fruit matures 8,9,36 . However, the temporal dynamics differed between fruit types: numbers remained constant for blueberry but increased with ripening for cherry, and the intermediate ripening stages of raspberry and strawberry had more phylotypes (Fig. 2). Ripening fruit represents a changing habitat which undergoes several physiological changes, including an increase in size and sugar content as well as changes in firmness, colour, and other secondary metabolites which may contribute to fungal community composition. Despite revealing differences in fungal communities between fruit types/sites and maturation stages, there was a large core microbiome which was present across all fruits/sites: this comprised only a fraction of diversity at just 199 of the 1712 fungal phylotypes (Table S11) but was the majority in terms of abundance as it comprised 97.6% of the DNA reads.
Our attempt to quantify changes in absolute fungal cell numbers was only successful for cherry. The total fungal population load per mm 2 remained constant across ripening, but there are no other published quantitative DNA based estimates from fruit for comparison. Further optimisation of the levels of added internal standard cells may have allowed quantitative estimates across all fruits; alternatively, adding a synthetic chimeric DNA spike to samples before DNA extraction may be a better strategy 42 . Using synthetic sequences as an internal standard has the added benefit of this not occurring in environmental samples 35 . It is also worth noting that including an internal standard in the form of live cells added before DNA extraction assumes that DNA extraction and amplification will be the same across all fungal cells present.
The nature of differences observed for total fungal communities generally held when just Saccharomycetales yeasts were analysed. Specific Saccharomycetales yeast genera which have been empirically shown to be attractive to D. suzukii in field and lab assays 27,30 were more prevalent at the raspberry site. Further, the species which has been implicated most in D. suzukii attraction, H. uvarum 32 , was highly abundant on raspberry. In addition, B. cinerea has been shown to have a repellent effect for D. suzukii 34 . Of the four fruit sites, raspberry had the lowest amounts of B. cinerea showing an inverse correlation with yeasts attractive to D. suzukii. Raspberry was also the fruit with the greatest host potential index scores for D. suzukii attraction by Bellamy et al. 38 , and together these observations are in line with the hypothesis that H. uvarum plays a role in D. suzukii attraction to fruit. However, it must be noted that the observed correlation between yeasts shown in other work to be attractive to D. suzukii and the abundances of these yeasts on fruit shown in this study cannot be taken as a causative correlation at this stage. There are other factors like fruit acidity, sugar content and firmness that have been shown to influence in fruit preference of D. suzukii [43][44][45] . Further work is needed to directly empirically determine the extent to which yeast communities affect D. suzukii preference for fruit. As it stands, these are general correlations, and one may not yet conclude that these abundant yeast phylotypes necessarily drive attraction. The observations here may be compared with a study showing that greater numbers of D. suzukii larvae developed on strawberries than raspberries which where greater than on blueberry; however, this study did not control for fruit associated microbiomes, and factors other than yeast communities may have caused these differing observations 46 .
Overall, further work is needed to understand if such fruit microbial patterns hold in other locations at other times and whether this correlation with attractive yeast from laboratory and field assays has any underlying basis for causation for D. suzukii fruit susceptibility in the field. If so, this opens the possibility of manipulating fruit microbiomes to deter D. suzukii. Whether fungal species repulsive to D. suzukii species could be 'seeded' onto fruits to reduce attractiveness is an intriguing question. Similarly, if this could be combined with traps containing attractive baits situated in and around crops to form a push-pull system to push flies away from fruits and attract them into traps. Although it is unrealistic to use B. cinerea in this way due to its phytopathogenic nature, certain yeast species are known to be repulsive to D. simulans and D. melanogaster 30,[47][48][49] . Van Timmeren et al. 50 demonstrated that crop sterilants also impact attractive yeast species growth and reduce D. suzukii larval infestation of fruit. A logical extension of this implies that future data might reveal specific microbes, which are not harmful to fruits or humans and are able to reduce D. suzukii attraction and could therefore be applied for crop protection.

Conclusion
This study demonstrates that for general fungal and more specific Saccharomycetales yeast communities, fruit type or site and maturation stage have a significant impact on fungal diversity, with fruit type/site having a larger effect. This observation also holds for yeast species known to attract D. suzukii, and here these yeasts were most  30 . This knowledge may potentially be applied to better understand what drives D. suzukii susceptibility of different fruit crops at different sites. It is also possible this may inform the engineering of fungal/ yeast communities which could be 'seeded' on fruit to reduce the susceptibility of commercial fruit crops to D. suzukii, or to identify ecologically realistic yeast communities for use as potentially attractive phagostimulant baits to control to D. suzukii and reduce the use of chemical pesticides.

Materials and methods
Fruit sampling and processing. All methods, including fruit collections, were performed in accordance with relevant guidelines and the project was conducted under ethics approval CoSREC388 from the University of Lincoln. Based on fruit pigmentation, blueberries, cherries, raspberries, and strawberries were sampled at four developmental stages ranging from unripe (green) to fully ripe (red/purple/navy) (Table S14, Fig. S1) throughout June to September in 2018. Sampling times differed for each fruit type (Table S14). All samples were collected from commercial fruit growers in the United Kingdom southern county of Kent at a maximum distance of 19 km apart; the same sites were revisited at each ripening stage. All fruit were subject to growers' spray programmes to control pest and diseases. Ten fruits (except blueberries N = 20 as these are smaller) were collected for each species and combined into one sterile bag, and this was replicated six times within each site at each of the four stages for each fruit, totalling 1200 individual and 96 combined fruit samples. Fruits were randomly selected within each field or orchard and were aseptically removed with as little of the stalk or calyx as possible without damaging the fruit. Fruits were briefly inspected for damage before removal with sterile scissors, and fruits were allowed to drop directly into sterile sample bags and thus not handled. Fruits were transported directly to the laboratory where 20 mL of sterile water was added to sample bags. Fruits were then surface-washed repeatedly with this water for 15 s every 5 min for 30 min, after which the contents collected in sterile 50 mL falcon tubes and centrifuged for 30 min at 4500 rpm to collect microbes. No surfactants were used in the surface washing process, as fungi vary in their hydrophobicity this may have affected isolation of certain fungi. The supernatant was reduced to approximately 2 mL, the pellet re-suspended and 1 mL was transferred to microfuge tubes and centrifuged further at 13,000 rpm for 10 min. The supernatant was discarded, and the pellet stored at − 80 °C. After washing, fruit were measured with vernier callipers and surface area estimated using 4πr 2 .
DNA extraction. Pellets derived from samples were thawed and re-suspended in sterile water, then split into two equal parts. One half of each sample was spiked with 265 live P. cucumerina (Ascomycete: Sordariomycetes) cells determined using a haemocytometer to act as an internal standard. This constituted pairs of samples which were identical other than the spiked P. cucumerina internal standard cells to allow an estimate of absolute cell numbers in the resulting sequence data. Plectosphaerella cucumerina has rarely been reported on the surface of fruits and the isolate used derived from pumpkins in Lincolnshire (UK) and was grown in potato dextrose broth (ThermoFisher Scientific) at 25 °C for 7 days prior to use. Direct cell counts from these fruit samples indicated that 265 cells would represent approximately 0.5% of the community and thus be detectable. DNA was extracted using the DNeasy Blood and Tissue kit (QIAGEN) following the manufacturer's instructions but with an additional bead beating step before incubation: pellets were resuspended in 750 µL ATL lysis buffer and added to 1 g of sterile glass beads with a 1:1 ratio of < 106 µm: 0. Bioinformatics analysis. DNA sequences were processed with QIIME 2 (2019.4) 53 . Sequence quality was evaluated with FastQC 54 and reads were trimmed, denoised, paired end merged and ASVs identified with DADA2 55 . ASVs were subsequently clustered with a > 97% genetic identity using vsearch 56 , and we term ASVs with > 97% identity 'phylotypes' . Phylotypes assigned to the fungal kingdom were identified using q2-featureclassifier plugin using the unite_ver7dynamic database 57 ; any unassigned phylotypes were subjected to manual Blast searches against the Genbank nucleotide database, and only phylotypes identified as belonging to the fungal kingdom were retained. For non-quantitative analysis, any phylotypes with 100% identity to the P. cucumerina internal standard were removed. Raw sequence counts were subjected to CSS variance-stabilising normalisation using metagenomSeq and phyloseq R packages [58][59][60][61] . Recent work indicates that analyses with equal sample depths by rarefication produces the same general patterns as with CSS variance-stabilising normalisation 62 , and this is especially important for comparisons of species richness among samples. For the quantitative analysis of fungal communities, samples containing the spiked fungal internal standard (P. cucumerina) were separately processed through the bioinformatics pipeline. Quantitative estimates of phylotype cell counts were calculated by normalising the read number of each phylotype to the number of P. cucumerina reads in that sample, and absolute cell numbers estimated from the knowledge that 265 P. cucumerina cells were added. Phylotype assignments at the species level were estimated by Blast searching the Genbank nucleotide database with representative www.nature.com/scientificreports/ sequences and reporting hits with > 97% homology. The order Saccharomycetales was analysed by filtering for all phylotypes assigned to Saccharomycetales at the order level.
Statistical analysis. R version 3.6.1 was used for all statistical analyses 63 . The effect of fruit species and ripening stage on numbers of phylotypes (richness) was assessed using a two-way ANOVA with Tukey HSD for post-hoc pairwise comparisons. A square root transformation was applied where the data did not conform to the assumption of normality as determined by Shapiro-Wilks tests, and Kruskal-Wallis tests applied if transformation did not achieve normality. Omega squared estimates of effect size for two-way ANOVA were calculated with ω 2 = df effect × (MS effect − MS error )/(SS total + MS error ) 64 . Shannon's and Simpson's diversity indexes were analysed using Kruskal-Wallis tests. Differences in presences or absences of fungal phylotypes and relative abundances of phylotypes were analysed with two-way full factorial permutational multivariate ANOVA (PermANOVA) using the 'adonis' function in the vegan package 65 with 10,000 permutations on binary (phylotype presences) and abundance based Jaccard dissimilarity matrices 66 . Pairwise PermANOVAs were conducted to analyse differences within fruit species and ripening stages where required. For quantitative analysis of fungal communities, the effect of ripening stage on cell numbers was analysed using a Kruskal-Wallis test. Indicator analysis was used to determine fungal phylotypes which were over-represented in the different fruit species with the 'indicspecies' package 67 . ASV abundances were correlated to overall Host Potential Index scores taken from Ref. 38 using Pearson's correlation coefficient. Venn diagrams were created with the 'eulerr' package 68 . Mantel test was used to correlate geographic and community difference using vegan 65 .
Ethics approval. This project was approved by the University of Lincoln ethics board (CoSREC388).

Data availability
Raw sequences are available on SRA (project ID: PRJNA732273) and the ASV