The importance of environmental microbes for Drosophila melanogaster during seasonal macronutrient variability

Experiments manipulating the nutritional environment and the associated microbiome of animals have demonstrated their importance for key fitness components. However, there is little information on how macronutrient composition and bacterial communities in natural food sources vary across seasons in nature and on how these factors affect the fitness components of insects. In this study, diet samples from an orchard compost heap, which is a natural habitat for many Drosophila species and other arthropods, were collected over 9 months covering all seasons in a temperate climate. We developed D. melanogaster on diet samples and investigated stress resistance and life-history traits as well as the microbial community of flies and compost. Nutrient and microbial community analysis of the diet samples showed marked differences in macronutrient composition and microbial community across seasons. However, except for the duration of development on these diet samples and Critical Thermal maximum, fly stress resistance and life-history traits were unaffected. The resulting differences in the fly microbial community were also more stable and less diverse than the microbial community of the diet samples. Our study suggests that when D. melanogaster are exposed to a vastly varying nutritional environment with a rich, diverse microbial community, the detrimental consequences of an unfavourable macronutrient composition are offset by the complex interactions between microbes and nutrients.


Results
Microbial community structure of diet samples. The diet microbiotas were dominated by a few genera, such as Acetobacter, Lactobacillus, Gluconobacter and Brevundimonas ( Supplementary Fig. S1A).
Nonmetric multidimensional scaling (NMDS) analysis based on Bray-Curtis dissimilarity matrices of amplicon sequence variants (ASVs) showed clear differences in the microbial community structure among diets (Fig. 1a). This was confirmed by an ANOSIM test (R = 0.98, P < 0.001). Underlying this difference in microbial community structure was an influence of the protein to carbohydrate (P:C) ratio (Mantel test: r = 0.15, P < 0.05) as well as an influence of the average temperature of the month the diets were collected (Mantel test: r = 0.49, P < 0.001). Using the Shannon diversity measure, we found that the bacterial diversity of the diets was significantly different across the collection days (F 1,22 = 41.78, P < 0.001) (Fig. 1b). The diversity of the diet collected on 2nd May was significantly lower than that of all the other diets, while the diversity of the diets collected in late autumn to winter was significantly higher than that of those collected in summer (Fig. 1b, Supplementary Tables S2, S3).
In the top 20 most abundant ASVs present, we found that the relative abundance of eight microbial strains significantly decreased as the year progressed; seven significantly increased, and the remaining microbial strains did not change significantly (Fig. 1c, Supplementary Table S4).
Microbial community structure of fly samples. The microbiotas of the flies were highly dominated by the genus Acetobacter ( Supplementary Fig. S1B) and differed significantly across the diets they consumed (ANOSIM test: R = 0.76, P < 0.001, Fig. 2a), although this distinction was not as strong as for the microbial community structure of the diet samples. Nevertheless, we still detected an influence of the P:C ratio of the diet on the microbiota of the flies (Mantel test: R = 0.13, P < 0.001) and the average monthly temperature when samples were collected (Mantel test: R = 0.16, P < 0.01). The Shannon diversity of the microbiota of flies did not differ across the collection day of the diets (F 1,44 = 1.07, P = 0.4), nor did it differ across the P:C ratios of the diet (F 1,44 = 1.09, P = 0.3 (Fig. 2b, Supplementary Table S2).
The Shannon diversity of the microbiota of the flies was significantly lower than that of the diets (F 1,67 = 108.58, P < 0.001, Fig. 2b, Supplementary Table S2). The microbiota of the flies was also significantly different from that of their diets (ANOSIM test: R = 0.64, P < 0.001, Supplementary Fig. S2).
As with the diet samples, we next took a closer look at the top 20 relatively most abundant ASVs. Half of the top microbial strains did not change significantly across the collection dates (Fig. 2c, Supplementary Table S4). The remaining ASVs showed variations depending on the collection date of the diet (Supplementary Table S4).
Finally, we assessed the share of ASVs between the flies and their respective diets. Based on Sørensen's dissimilarity measure 24 , the microbiome of the flies was significantly different from that of the diet samples (ADONIS: R 2 = 0.15, P < 0.001) ( Supplementary Fig. S2). By merging the fly samples with their respective diet and creating nine groupings, we found that 3-17% of the total ASVs in each diet group were shared ( Table 2); 20% of these were found in at least two of the fly diet groups (Supplementary Table S5). Impact of diet on fly fitness components. The developmental time across the natural diets was significantly different (F 1,100 = 184.5, P < 0.001). Flies developed on diets collected in summer had significantly longer developmental time, with flies developing on the diet collected on June 2nd having the longest developmental time (Fig. 3a (Fig. 3a). Conversely, flies maintained constant egg-to-adult viability across the diets from different seasons (χ 2 = 166.63, P = 0.99) and P:C ratios (χ 2 = 166.63, P = 0.53) (Fig. 3b). There was also no significant difference between the  (Fig. 3c). The sensitivity of the flies to low temperatures appeared to have a peak when the P:C ratio was at its highest ( Fig. 3c); however, this was not statistically significant (F 1,126 = 1.98, P = 0.16). In contrast, for tolerance towards high temperatures, the Critical Thermal maximum (CTmax) appeared to decrease when the P:C ratio was high, and this negative correlation was statistically significant (F 1,131 = 8.26, P < 0.05) (Fig. 3d). There was, however, no significant difference between the diet collection times (F 1,131 = 1.22, P = 0.27) (Fig. 3d). No significant difference was found in starvation resistance across the collection time (F 1,149 = 0.56, P = 0.82), and there was no correlation with the P:C ratio (F 1,149 = 0.32, P = 0.57) (Fig. 3e).

Discussion
Changes in diet composition throughout the year. In nature, D. melanogaster acquires much of their carbohydrate and protein from decaying fruit and yeasts growing on the fruit 25 . During fruit decay, yeasts break down the fruit tissue and its sugar components, which should cause temporal fluctuations in the amount of protein and carbohydrate available to other organisms 10 . In diets collected in late spring to summer, carbohydrate was more prevalent than protein. This changed in the colder months, where protein levels increased and carbohydrate levels decreased, but only until November, where protein levels decreased again, probably due to the fruit being too decomposed for yeasts to successfully metabolize it (Table 1). We also found very distinct changes in microbial community structures of the natural diet samples collected across the year (Fig. 1a). The change in relative abundance of some of the most prevalent ASVs, Gluconobacter, Acetobacter and Komagataeibacter, is consistent with these genera's preference for carbohydrate-rich environments [26][27][28][29] . Thus, their decrease in relative abundance followed the decrease in carbohydrate in the Danish winter (Fig. 1c). This illustrates that throughout the year, flies (and other arthropods affiliated with compost heaps) are exposed not only to nutritional fluctuations but also to corollary changes in the microbial community structure of their diet.
Characteristic differences between microbial communities in diets and flies. The microbial diversity and the most abundant genera of the flies differed from those of the diet samples (Fig. 1c, 2c, Supplementary Fig. S3). The top 20 ASVs found in the flies were dominated by ASVs from Acetobacter (Fig. 2c, Supplementary Fig. S1B) and ASVs not found in the diet samples ( Supplementary Fig. S3). D. melanogaster obtains most of its microbiota from bacteria on rotting fruit and faecal matter in their environment or inherits it via egg surfaces [30][31][32][33][34] . As eggs from our laboratory stock were not washed when transferred to the diet samples, the microbial community structure of the flies was a mix of compatible microbes from the diet and the laboratory stock.
We found only a small number of microbes that were shared between the diet and the fly samples (Table 2,  Supplementary Table S5). Even fewer microbes were shared as the year progressed, and the diversity of the diet samples increased (Fig. 1a, Table 2), which could be a result of several factors. A number of physiological and immune response features of the fly gut are known to contribute to microbial regulation 35 . Only a small number of Table 1. Natural diet sample information: collection day (counted from the date of the first sample collected), date of sample collection, average temperature of the month the diets were collected, protein value, carbohydrate value and protein to carbohydrate ratio of each sample.

Collection day
Date of collection Average monthly temperature (°C) Protein (g/100 g) Carbohydrate (g/100 g) P:C ratio www.nature.com/scientificreports/ taxa can therefore survive in the gut of Drosophila species 5,34,36 . This active regulation of residing microbes would also explain the overall lower diversity in our fly samples compared to the diets (Fig. 2b, Supplementary Table S2). Interestingly, we found abundant Streptomyces in all the fly samples but only a trace abundance in just one of the diet samples ( Supplementary Fig. S3, Fig. 2b). As Streptomyces is usually found in soil and has a high growth rate at warmer temperatures 37 , we found it likely that Streptomyces was present in all diet samples in trace amounts but that the relatively high temperature of the laboratory experiments resulted in an exponential increase during exposure to flies. However, we have found no evidence in the literature of Streptomyces being a successful symbiont of Drosophila, but one explanation could be that its presence is transient. This result also highlights that the microbial community of the diet may have changed during the development of the larvae, which could have also resulted in a change in the microbial load of the food. An analysis of the diet after the larval phase could have confirmed this.
Fitness components are constant despite macronutrient fluctuations of the diet. The developmental time of the flies was the only trait significantly impacted by the collection time of the fly diet (Fig. 3a,  Supplementary Table S5). Apart from developmental time and CTmax, none of the other traits significantly correlated with the P:C ratio of the diet (Fig. 3d). This was unexpected, as changes in protein and carbohydrate amounts in the diet during laboratory experiments are known to influence a range of traits in insects 13,17,[38][39][40][41] . This is also the case in D. melanogaster 16 , where low protein content in the diet leads to increased egg-to-adult viability and starvation 17 , and higher protein content leads to increased heat and cold tolerances 21 . In contrast to our findings, we therefore expected the fluctuating protein and carbohydrate contents in the nine natural diets to influence all traits. One reason for this discrepancy could be the use of highly simplified foods with very little microbial diversity in previous laboratory investigations. We suggest that the diverse microbes associated with natural diets play an important role in reducing the impact of nutritional fluctuations on fly phenotypes, and we discuss this further below.
Bacteria to the rescue. The relationship between resident microbes and host physiology is complex and influenced by a range of factors, such as host genotype 42 , immunity 43 and diet 44 . For D. melanogaster, we know that although an associated microbial community is not essential for survival, it can be extremely beneficial in times of macronutrient stress 4,15,18,45 . Association with certain species of Acetobacter, for example, can also alter the nutritional balance of the food ingested by D. melanogaster, as these microbes can selectively consume sugars, preventing the fly from digesting too much carbohydrate 3 . We therefore speculate that the high relative abundance of Acetobacter in both the diets and the flies in the present study ( Supplementary Fig. S1A,B) had a balancing effect on food ingestion when carbohydrate levels were high in the warmer months, stabilizing fitness components of flies across the year. This conclusion, however, must be made with caution: Keebaugh et al. 18 also showed that diets with higher amounts of yeast can provide ideal growth conditions for bacteria 18 , so even though the relative abundance of Acetobacter is high during high carbohydrate periods (Fig. 1b, Supplementary  Fig. S1A,B), the absolute amount of Aceobacter could be exactly the same during colder months. Our results find a positive correlation with the P:C ratio and the diversity of natural food (Fig. 1b), which supports Keebaugh et al. 18 .
Additionally, flies developed on an unfavourable, imbalanced diet have increased fitness when the diet is supplemented with high amounts of bacteria 18,46 . Keebaugh et al. 18 showed that the presence of sufficient quantities of Acetobacter species can ameliorate the negative effects of insufficient protein on lifespan and larval viability. Developing on a microbial rich, natural diet could thus be beneficial for an organism going through macronutrient stresses, although it is unclear if the microbes are a valuable food source or if they provide crucial amino acids 47,48 .

Conclusion
Artificial laboratory diets where macronutrient amounts are manipulated have provided important insights into the phenotypic responses to varying macronutrients. Sterile practices, however, of autoclaving food and regularly providing fresh food reduce the microbial diversity in the food and compromises the ecological relevance of the diet. Here we quantified the natural fluctuations in macronutrient availability and microbial community structure that D. melanogaster faces in nature. The investigated fitness components were largely unaffected by these natural fluctuations. Because of the complex nature of microbe-nutrient-organism interactions though, many factors could play a role in this phenotypic canalization. To fully characterise the importance of these factors we therefore advocate future work sterilizing the natural diet to separate the effects of environmental microbes and the natural diet.
This study provides an example where varying the protein to carbohydrate ratios does not significantly affect phenotypic traits known to be influenced by diet. Based on our results, we suggest that a diet made up of a diverse range of microbes could offset the detrimental effects of an unfavourable macronutrient composition without directly impacting the associated microbiota of the fly. www.nature.com/scientificreports/ Prior to the experiment, flies were taken from the stock and kept at 25 °C for three generations, the temperature at which the experiment was conducted. Population density was controlled by transferring 30 eggs to vials containing 7 mL of standard laboratory medium. Once emerged, 200 adults were transferred to bottles containing 35 mL of standard laboratory Drosophila medium and left to lay eggs overnight. This was done for three generations.

Methods
After these three generations, the experiment was initiated by transferring 40 eggs laid on medium in 35 mL bottles to vials containing 7 mL of our experimental diets. This was done across 5 × 8 h periods to account for any differences in developmental time 49 . The use of laboratory-reared flies ensured a more similar microbiome across individuals with low diversity 5,32,36 . Natural diets. The natural diets originated from a compost heap at the orchard where the laboratory stock flies were collected, Odder (55° 56′ 42.46″ N, 10° 12′ 45.31″ E), Denmark. The compost heap was established based on windfall apples and pears collected in November 2013. Nine samples (of ca. 1.5 kg each) were collected from the heap across nine months in 2014 (Table 1). Each sample was made up of collections taken from multiple areas of the heap due to the variability of materials in the heap. The samples were stored at − 20 °C, whereby changes in microbial community and microbial viability over the period of storage time would be very minor 50,51 . Before measurements were taken, the diets were blended. The average monthly temperatures of the location of the collected samples were obtained from the Danish Meteorological Institute, www. dmi. dk (Table 1).
During pilot experiments, we observed dominant fungal growth in the natural food when kept at 25 °C in the lab, which significantly impacted the development and survival of the flies. We therefore added nipagin (12 mL/L) to the compost to solve this problem before exposing flies to the diets. The low concentrations have been shown not to change the viability of the bacterial community or affect gut bacterial composition directly 52-55 . Nutritional composition of natural diets. The nutritional compositions of the natural diets were analysed by ALS Food and Pharmaceutical (http:// www. alsgl obal. com/) using the methods described in Supplementary materials Table S1. Protein and carbohydrate (not including indigestible fibre) amounts are displayed in Table 1.
Microbiota molecular procedure and processing of sequence data. DNA extraction and sequencing were performed externally by DNASense, Aalborg, Denmark (http:// www. dnase nse. com). The outside of the flies was washed with ethanol before being sent for sequencing. DNA was extracted from blended samples of the natural diets and whole flies pooled in groups of five. Three diet-free and fly-free negative controls were also included. The extractions were performed using the DNeasy Blood and Tissue kit according to the manufacturer's protocol (Qiagen, Germany).
The DADA2 pipeline (which can be found at https:// github. com/ benjj neb/ dada2) processes Illumina MiSeq raw amplicon sequences into a table of amplicon sequence variants (ASVs), a higher-resolution measure than OTU clustering 58 , which are present and the number of times each ASV is observed in each sample. To obtain our ASV table, we followed the pipeline using the DADA2 "standard" filtering and trimming parameters (https:// github. com/ benjj neb/ dada2) 58 , adjusting only the truncation length to 280 and 260 to suit our chosen primers. The core DADA2 algorithm 58 was then applied to the filtered and trimmed sequence data, and chimeras were removed from the ASV table. Taxonomy was then assigned using the Silva version 132 database 59 . An average of 18,059 reads per diet sample and an average of 24,347 reads per fly sample were assigned to bacterial ASVs (Supplementary Table S2).
Phenotypic assessment. Developmental time, egg-to-adult viability, starvation resistance and Critical Thermal maximum (CTmax) and minimum (CTmin) were assessed. Developmental time was assessed by counting flies that had emerged in each vial every 12 h (40 eggs per vial, n of vials: 2nd May = 10, 2nd June = 14, 30th June = 9, 30th July = 12, 11th Sept = 13, 2nd Oct = 11, 3rd Nov = 9, 1st Dec = 11, 2nd Jan = 12). Egg-to-adult viability was measured as the percentage of flies that emerged from the 40 eggs in each replicate vial across the nine developmental diets (n of vials: 2nd May = 19, 2nd June = 17, 30th June = 18, 30th July = 18, 11th Sept = 18, 2nd Oct = 18, 3rd Nov = 18, 1st Dec = 18, 2nd Jan = 18). On the day starvation resistance and thermal tolerances were measured, at 00:00, emerged flies were counted and removed. Male flies that had emerged between 0:00 and 8:00 were used for the stress resistance tests. Therefore, at 8:00, the males were collected (by sight, with no CO 2 ) and randomly assigned to a stress resistance experiment. Starvation resistance was measured by placing one fly per vial that contained 3.5 mL of water and agar (2%) and counting how many flies died every 8 h (n of flies: 2nd May = 18, 2nd June = 14, 30th June = 17, 30th July = 18, 11th Sept = 16, 2nd Oct = 16, 3rd Nov = 17, 1st Dec = 17, 2nd Jan = 17). CTmax was measured by methods described in Overgaard, Kristensen and Sørensen 60 . In short, we placed flies from each diet in sealed 5 mL screw cap glass vials with one fly per vial. These vials were then fixed onto racks and immersed in a water tank. The temperature of the water then increased from 25 by 0.1 °C per minute. The CTmax score was the temperature of the water when the fly stopped responding by physical move- Analyses of the microbial communities were performed on the relative abundance of each ASV using the phyloseq v1.34.0 package in R 63 . Patterns of dissimilarities were visualized using nonmetric multidimensional scaling (NMDS) using Bray-Curtis dissimilarity. Further statistical analysis of the dissimilarities within the microbial communities of the diet samples and within the fly samples was performed using the R package vegan v.2.5-7 64 . First, we conducted ANOSIM tests to determine whether the microbial communities within the diet samples and within the fly samples were significantly different. This was followed by Mantel tests, based on Bray-Curtis dissimilarity matrices of the ASVs, to calculate the correlations between the microbial communities and the P:C ratio of the diet and the average temperature of the month the diets were collected. Collection day was additionally used as a continuous variable to conduct Pearson's correlation tests conducted on individual ASVs to investigate their change throughout the year. For the statistical analyses on diversity measures and phenotypic assessments, the day of collection of the diets (Table 1) was used as the predictor variable and was considered a factorial variable. Statistical analyses were also performed on diversity measures and phenotypic assessments using the P:C ratios of the diets (Table 1) as a continuous variable.
Correlations between carbohydrate amount, measures of microbial diversity and relative abundance of certain bacteria were assessed using Pearson's correlation coefficient. For protein amount, a linear model containing protein amount as the response variable was compared with a quadratic model using an F test to obtain the P-value.
The effect of the collection day of the developmental diet on developmental time, CTmin, CTmax and starvation resistance was assessed by constructing separate linear models. The models contained the average developmental time of each vial, CTmin, CTmax or starvation resistance as the response variable. These models were compared to a reduced model using an F test to obtain the P-values.
To investigate the effect of the collection day of the diet on egg-to-adult viability, we used logistic regression in a generalized linear model. The model contained the day of collection as the only predictor variable. We detected no over-dispersion in the model. We compared the full model with a reduced model using a likelihood ratio test to obtain a P-value for the effect of collection day of the diet.