Bacterial Metabolism and Transport Genes Are Associated with the Preference of Drosophila melanogaster for Dietary Yeast

ABSTRACT

Many animal traits are influenced by their associated microorganisms (“microbiota”). To expand our understanding of the relationship between microbial genotype and host phenotype, we report an analysis of the influence of the microbiota on the dietary preference of the fruit fly Drosophila melanogaster. First, we confirmed through experiments on flies reared bacteria-free (“axenic”) or in monoassociation with two different strains of bacteria that the microbiota significantly influences fruit fly dietary preference across a range of ratios of dietary yeast:dietary glucose. Then, focusing on microbiota-dependent changes in fly dietary preference for yeast (DPY), we performed a metagenome-wide association (MGWA) study to define microbial species specificity for this trait and to predict bacterial genes that influence it. In a subsequent mutant analysis, we confirmed that disrupting a subset of the MGWA-predicted genes influences fly DPY, including for genes involved in thiamine biosynthesis and glucose transport. Follow-up tests revealed that the bacterial influence on fly DPY did not depend on bacterial modification of the glucose or protein content of the fly diet, suggesting that the bacteria mediate their effects independent of the fly diet or through more specific dietary changes than broad ratios of protein and glucose. Together, these findings provide additional insight into bacterial determinants of host nutrition and behavior by revealing specific genetic disruptions that influence D. melanogaster DPY.

IMPORTANCE Associated microorganisms (“microbiota”) impact the physiology and behavior of their hosts, and defining the mechanisms underlying these interactions is a major gap in the field of host-microbe interactions. This study expands our understanding of how the microbiota can influence dietary preference for yeast (DPY) of a model host, Drosophila melanogaster. First, we show that fly preferences for a range of different dietary yeast:dietary glucose ratios vary significantly with the identity of the microbes that colonize the fruit flies. We then performed a metagenome-wide association study to identify candidate bacterial genes that contributed to some of these bacterial influences. We confirmed that disrupting some of the predicted genes, including genes involved in glucose transport and thiamine biosynthesis, resulted in changes to fly DPY and show that the influence of two of these genes is not through changes in dietary ratios of protein to glucose. Together, these efforts expand our understanding of the bacterial genetic influences on a feeding behavior of a model animal host.

INTRODUCTION

Animal dietary preferences are behaviors where animals choose to consume one diet more abundantly than others. In the model organism Drosophila melanogaster, dietary preference has been measured in many different ways. These include feeding with capillaries to measure the volume of fly diet consumed (CaFe assay) (1), analysis of consumed colored or radioactive diets (2–4), manual or automated scoring of proboscis extensions into the diet (5, 6), or quantitative PCR of trace oligonucleotides that were added to consumed diets in specified amounts (7). Use of these assays has shown that fly dietary preference is influenced by a variety of factors, including mating status (8), temperature (9), essential amino acid deprivation (10), chronic hypoxia (11), and associated microorganisms (“microbiota”) (12). Together, these results have established that fly dietary preference can be robustly assessed by a variety of methods and that this trait is influenced by or responds to numerous external factors (i.e., in addition to host genetics). Our purpose in this work was to better understand the role of the microbiota in influencing D. melanogaster dietary preference.

The fruit fly is an established model for understanding the interactions between a host and its microbiota. The D. melanogaster microbiota is of relatively low numeric abundance and taxonomic diversity, with individual flies usually bearing 10⁵ to 10⁶ individual bacteria, most of which are readily culturable and are from the family Acetobacteraceae (acetic acid bacteria [AAB]) or orders Lactobacillales (lactic acid bacteria [LAB]) and Enterobacteriaceae (13–15). Flies are readily reared in the absence of colonizing bacteria by sodium hypochlorite treatment or with defined bacterial communities by inoculating bacteria-free embryos with live bacterial cultures (16). Numerous studies using mono- or poly-associated flies have confirmed the bacterial influences on diverse host phenotypes are strain specific. For example, fly hosts inoculated with individual bacterial species display variation in mating behaviors (17), foraging behaviors (12), longevity (18), starvation resistance (19), development (20–22), and life history strategy (23). Interactions between community members have also been demonstrated to be important determinants of these and other host traits (10, 24–28). Recent work has established that the D. melanogaster microbiota can strongly influence fly dietary preferences. For example, in the absence of essential amino acids (eAAs), flies normally prefer an AA-rich diet (10), but this preference can be alleviated by manipulation of the microbiota (10). Follow-up experiments explained how, on a defined, holidic fly medium (29), a two-species microbiota composed of the AAB strain Acetobacter pomorum and one of at least two LAB strains is sufficient to spare the flies of this dietary response (25). LAB act indirectly by producing lactic acid that A. pomorum uses directly to produce an unknown molecule or molecules that suppress the compensatory yeast preference (25). Separately, an analysis of fly dietary preferences in flies that were reared on different geometries of sugar and protein on rich, undefined media (i.e., without yeast deprivation) showed that members of the microbiota can influence fly foraging behaviors (12). Flies inoculated individually with A. pomorum preferred a 2:1 ratio of glucose (G) to yeast (Y) in their diets, while flies inoculated with the LAB strain Lactobacillus plantarum showed no such preference (12). Thus, with or without eAA deprivation, flies colonized with A. pomorum have lower preference for dietary yeast relative to flies reared with other microbial treatments and the bacterial genetic mechanisms driving yeast preference under eAA deprivation are partially but incompletely described (10, 25). In particular, attempts to identify the microbial influence on fly dietary preference for yeast (DPY) have ruled out roles for specific amino acids but still have not identified a specific effector. The influence of the microbiota on DPY in the presence of abundant yeast is also incompletely defined.

In this study we sought to better understand how the fly microbiota shapes fly DPY under non-yeast-limited conditions by asking two distinct questions: (i) which fly dietary preferences are most responsive to colonizing microorganisms; and (ii) what are some of the bacterial genetic bases for these influences? We answered these questions by screening a range of dietary preferences in flies that were monoassociated with different microorganisms, performing a metagenome-wide association (MGWA) to predict the bacterial genes responsible for these influences, and screening bacterial mutants for MGWA-predicted genes to confirm the predicted bacterial influences. The main results identified transposon insertions in specific bacterial genes that can shape the fly DPY. Furthermore, even though it is established that some AAB can significantly lower the glucose content of the fly diet (20), we show that bacterial influence on fly DPY can be independent of glucose catabolism, identifying multiple potential mechanisms whereby the bacterial partners can shape the feeding patterns of the fly host.

RESULTS

There are bacterial strain-specific influences of the microbiota on D. melanogaster dietary preference.

As a baseline to demonstrate we could detect similar influences of the microbiota on fly dietary preference as has been reported previously, we compared the dietary preference of flies that were axenic or monoassociated with one representative AAB strain, Acetobacter tropicalis, and one representative LAB strain, Levilactobacillus brevis. To measure fly dietary preference, we used a flyPAD device, which tracks electrical capacitance to determine when a fly closes a circuit between two electrodes. When a fly stands on one electrode and extends its proboscis into a second electrode that is a food trough, then the change in capacitance is recorded as a “sip” (6). In our initial screen, we compared the dietary preference of flies given a choice between a diet with 100 g yeast (Y) and 100 g glucose (G) L⁻¹ (100Y-100G) with a test diet that decreased only the yeast or only the glucose in increments of 25 g L⁻¹. (Fig. S1; Table S1). Also, we synchronized fly appetite by starving each fly for ~3 h, as in our previous work (30) before placing it into the flyPAD device; the rationale is shared with other studies that measure feeding rate although the approach and duration vary (6, 12). In most trial conditions at least one microbial treatment significantly and substantially influenced the flies’ preference for the trial diet relative to axenic flies, but there were no universal patterns that varied directly and linearly with decreased sugar or protein content of the diet. For example, decreasing the abundance of yeast in the trial diet by 25, 50, or 75% led to no major changes in dietary preference in axenic or L. brevis-treated flies and to decreased, increased, and little to no preference for the trial diet in A. tropicalis-reared flies, respectively. Also, in most cases the microbiota treatments had no effect on the total number of fly sips, the average duration of each sip, or the number of sips taken on each diet in a given diet challenge (Fig. S2; Table S2). When there was a microbiota treatment difference on the total number of fly sips, the most common effect was that elimination of the microbes led to fewer sips, consistent with previous reports that axenic flies feed less than bacteria-colonized flies (30) (Fig. S2A). The most obvious evidence of a microbiota by diet effect was that flies colonized with A. tropicalis, but not axenic flies or flies colonized with L. brevis, increased their feeding of a 50% yeast-restricted, relative to the control, diet. Equal ratios of glucose to yeast in the trial diet conferred similar preference, but not absolute feeding, phenotypes regardless of absolute abundance of glucose and yeast (Fig. S1 and S2C, compare the 50Y-100G values with 50Y-50G and 25Y-50G values). These findings reveal that some but not all of the microbiota by diet effect is linked to the relative abundances of nutrients in the diet. Taken together, these data corroborate the expectation of species-specific influences of the microbiota on D. melanogaster preference for diets that restrict glucose or yeast, consistent with the general trends of previous reports (10, 12).

To extend the results of our first experiment to a broader range of bacteria, we compared the DPY of flies given a choice between two diets when the flies were raised individually with different species of bacteria. We focused on yeast preference because of established work in this area and because the strain-specific, microbiota-dependent variation in fly dietary preference between these two diets in our initial survey was particularly variable (Fig. 1; Table 1). Under these conditions we detected a range of preference phenotypes in flies reared individually with 32 different bacterial strains, including some bacteria that conferred significant preferences for each of the diet choices (Fig. 2; degrees of freedom = 31; sum of squares = 4,663.6; mean squares = 150.44; F value = 150.44). The differences in preference index did not mask or obscure underlying feeding patterns. For example, there was no difference in the number of fly sips between treatments, including when the number of sips on the two test diets were analyzed separately (Fig. S3A and C). Also, sip duration varied with bacterial treatment and diet type (50Y-100G versus 100Y-100G), but the diet effect usually did not vary within a bacterial treatment (Fig. S3B). The one bacterial treatment that did affect sip duration (Weissella paramesenteroides DmW_107 caused longer sips on the diet with less yeast) had a similar effect on the number of sips (W. paramesenteroides DmW_107 caused more sips on the diet with less yeast). Finally, there was no obvious phylogenetic trend to explain the variation in preference; e.g., AAB did not confer preference for one diet and LAB for the other. Thus, the data demonstrate that the magnitude and targets of host DPY can be modulated by the microbiota and does not substantially depend on the number of fly sips, duration of fly feeding, or bacterial phylogeny.

FIG 1.

Microbiota-dependent fly dietary preference under yeast limitation. (A and B) Flies reared bacteria-free (gray) or with A. tropicalis (red) or L. brevis (blue) when feeding in choice arenas where both wells contained our standard lab diet (“control diet,” 100Y-100G) (A) or flies could choose between the control diet and a 50Y-100G diet, the trial diet (B). Different letters next to the bars represent statistically significant differences between treatments as determined by a generalized linear mixed effects model with a binomial family with a post hoc Tukey test. The preference index was extracted from the model as the linear prediction of the inputted values; n = 19 to 55 measures per treatment (median = 28.5) spread across 3 (A) or 4 (B) different experiments.

TABLE 1.

Statistical results for models whose results are shown in Fig. 1

Diet	g yeast L−1	g glucose L−1	No. days	No. per treatment			Statistics
				Lba	Atb	Axc	Dfd	SSe	MSf	F g
Y-G (control)	100	100	3	19	31	55	2	592.9	296.5	296.5
50Y-100G	50	100	4	30	19	27	2	2,225.2	1,112.6	1,112.6

^aLactobacillus brevis.

^bAcetobacter tropicalis.

^cAxenic.

^dDegrees of freedom.

^eSum of squares.

^fMean of squares.

^gF statistic.

FIG 2.

Species-specific effects of microbiota on D. melanogaster dietary preference. Flies were raised with each of 32 different bacterial strains, and their dietary feeding for a 100Y-100G control diet versus a 50Y-100G trial diet was measured. Bars are color coded by high-level taxonomic assignments. (Acetobacter spp.: red; non-Acetobacter AAB: orange; proteobacteria: green; Lactobacillus spp: blue; non-Lactobacillus LAB: lavender; non-LAB Firmicutes: purple). Different letters next to the bars represent statistically significant differences between treatments as determined by a generalized linear mixed effects model with a binomial family with a post hoc Tukey test. The preference index was extracted from the model as the linear prediction of the inputted values. The phylogenetic tree is based on a clustalW alignment built with MUSCLE and was edited in FigTree; n = 8 to 22 measures per treatment (median 14) randomized across 18 different experiments.

The broad genetic variation in the strain panel and phenotypic variation conferred on the flies by the strains enabled us to perform MGWA (20). For the MGWA we compared the DPY conferred by the different bacterial species with the bacterial gene presence-absence patterns for each gene in the strain panel. We first defined the gene presence-absence patterns across the 32 bacterial strains by clustering each gene into clusters of orthologous groups (OGs). We detected 13,474 OGs across the strain panel (Table S3), which included all of the 32 strains used in this study (Table 2) plus other strains from our laboratory stocks of bacteria isolated from flies (Table S4). When we performed the MGWA, 3,832 OGs in 1,208 distinct groups of the 32 strains (a phylogenetic distribution group [PDG]) were significantly associated with a change in fly DPY (max, 219; median, 1; mean, 3.17 PDGs OG⁻¹) or about one-third of all tested OGs (Table S5). This large number of OGs resulted from P-value inflation that could not be eliminated by using other statistical models or including as covariates in the models, principal components that represented genetic relatedness of the strains (Fig. S4). Even though the data were not well fitted to the model, we proceeded to perform a mutant analysis based on some top predictions with the rationale of using the MGWA as a surrogate genetic screen to identify bacterial gene candidates that might influence fly DPY.

TABLE 2.

Strains used in this study^a

Strain code/gene name	Strain name/well position in mutant library (White et al. [31])	WGS ID	Media	Oxygen condition
aace	Acetobacter aceti NBRC 14818	SLZP01	mMRS	Oxic
aci5	Acetobacter sp. DmW_043	JOMN01	mMRS	Oxic
amac	Acetobacter malorum DmCS_005	JOJU01	mMRS	Oxic
aori	Acetobacter orientalis DmW_045	JOMO01	mMRS	Oxic
apa3	Acetobacter pasteurianus 3P3	CADQ01	mMRS	Oxic
apan	Acetobacter pasteurianus NBRC 101655	AP014881.1	mMRS	Oxic
apnb	Acetobacter pasteurianus NBRC 106471 or LMG1262t	BDER01	mMRS	Oxic
apoc	Acetobacter pomorum DmCS_004	JOKL01	mMRS	Oxic
asl5	Acetobacter sp. SLV-7 DmW_125	VHOZ01	mMRS	Oxic
atrc	Acetobacter tropicalis DmCS_006	JOKM01	mMRS	Oxic
atrn	Acetobacter tropicalis NBRC 101654	BABS01	mMRS	Oxic
bsub	Bacillus subtilis subsp. subtilis str.168	NC_000964.3	LB	Oxic
cint	Acetobacter fabarum DmL_052	JOPB01	mMRS	Oxic
ecok	Escherichia coli str. K-12 substr. MG1655	NC_000913.3	LB	Oxic
galb	Gluconobacter sp. DsW_056	JOPF01	Potato	Oxic
gfra	Gluconobacter frateurii NBRC 101659	BADZ02	Potato	Oxic
kmed	Komagataeibacter medellinensis NBRC 3288	NC_016037.1	Potato	Oxic
lbrc	Lactobacillus brevis DmCS_003	JOKA01	mMRS	Microoxic
lbuc	Lactobacillus buchneri NRRLB-30929	NC_015428.1	mMRS	Microoxic
lc37	Leuconostoc citreum DmW_137	JADAXK01	mMRS	Microoxic
lfal	Leuconostoc fallax KCTC 3537	AEIZ01	mMRS	Microoxic
lfrc	Lactobacillus fructivorans DmCS_002	JOJZ01	mMRS	Microoxic
lfrk	Lactobacillus fructivorans KCTC 3543	QSLO01	mMRS	Microoxic
llcs	Lactococcus lactis DmW198	NEQN01	mMRS	Microoxic
lmli	Lactobacillus mali KCTC 3596 = DSM 20444	AYYH01	mMRS	Microoxic
lpar	Lactobacillus paracasei DmW181	NDXH01	mMRS	Microoxic
lplw	Lactiplantibacillus plantarum WCFS1	NC_004567.2	mMRS	Microoxic
lsui	Leuconostoc suionicum strain DmW_098	JADAXL01	mMRS	Microoxic
pput	Pseudomonas putida F1	NC_009512.1	LB	Oxic
wp07	Weissella paramesenteroides DmW_107	VHPP01	mMRS	Microoxic
wp15	Weissella paramesenteroides DmW_115113	VHPE01	mMRS	Microoxic
wp18	Weissella paramesenteroides DmW_118	VHPB01	mMRS	Microoxic
NitT/TauT	Plate13-f12		mMRS kan50	Oxic
thiE	Plate96-g11		mMRS kan50	Oxic
oprB	Plate6-b9		mMRS kan50	Oxic
permease	Plate60-a1		mMRS kan50	Oxic
iscS	Plate95-b2		mMRS kan50	Oxic
nuoA	Plate63-G7		mMRS kan50	Oxic
ALDH	Plate4-h5		mMRS kan50	Oxic
Hypothetical	Plate96-b11		mMRS kan50	Oxic

^aWGS, Whole Genome Shotgun; LB, lysogeny broth.

Transposon insertions in bacterial thiamine biosynthesis and glucose transport genes influence D. melanogaster DPY.

To focus on a subset of the many OGs that were significantly associated with variation in fly DPY, we chose eight genes for follow-up mutant analysis, based on two additional criteria: (i) the genes could be assigned a Kyoto Encyclopedia of Genes and Genomes (KEGG) Orthology (KO) number (32); and (ii) a mutant for the gene was present in our arrayed freezer library of transposon-insertion Acetobacter fabarum mutants. The flies were separately mono-associated with each A. fabarum mutant, and fly dietary preference was measured in each of 15 separate experiments that contained all bacterial treatments (n per treatment per experiment was 2). There was no difference in total feeding or sip duration between the treatments and usually no difference in sip duration or total feeding on each diet within treatments (Fig. S5). However, there were significant differences in fly DPY: transposon insertions in four of the eight genes promoted or suppressed the preference for decreased dietary yeast content relative to the wild-type A. fabarum strain (Fig. 3; Table S6). Insertions in transporters oprB (a glucose permeable porin) and a purine/thiamine permease gene suppressed fly preference for yeast beyond the natural levels conferred by A. fabarum. Conversely, insertions in the genes iscS, encoding a cysteine desulfurase that is necessary for the production of thiamine from cysteine, and thiE, which encodes a thiamine biosynthesis gene, enhanced the fly’s preference for dietary yeast. Whereas the iscS and thiE genes were the last genes in possible A. fabarum polycistrons, the insertions in the purine permease and oprB were not (Fig. S6). Therefore, while we cannot rule out polar effects of any insertions, off-target effects can be a consequence of transposon insertions and would be more likely for the oprB and purine permease insertions than the insertions in iscS and thiE. Thus, this approach identified specific transposon insertions that contribute to the DPY of the flies.

FIG 3.

Transposon insertions in bacterial thiamine biosynthesis and glucose transport genes influence D. melanogaster DP. Significant differences between the wild type (WT) and each of the mutants was performed using a generalized linear mixed effects model with a binomial family, followed by a Dunnett test (***, P < 0.001). Gene name abbreviations and their meaning are shown in Table 2; n = 13 to 21 measures per treatment (median 16) spread across 15 different experiments. The preference index was extracted from the model as the linear prediction of the inputted values.

Bacteria need not alter dietary protein to sugar ratios to influence D. melanogaster DPY.

As a follow-up to the mutant analysis, we tested if a subset of the mutants that influenced D. melanogaster DPY could have done so by altering the glucose content of the flies or their diet. It was previously established that some Acetobacter strains can consume dietary glucose, which we reasoned could alter the dietary ratios of protein and glucose and possibly explain some of the influence of the mutants. Contrary to this expectation, there was no significant variation in the ratio of protein to glucose in fly diets inoculated with the different bacterial mutants with (Fig. 4A) or without (Fig. 4B) flies or in whole bodies of the adult flies (Fig. 4C, see also Fig. S7). These results confirm that bacterial mutants that influence fly DPY must do so through a mechanism other than modifying the protein and glucose ratios of the flies or their diets. This idea is consistent with previous findings that bacteria do not modify fly feeding preferences by changing amino acid levels in flies or their diets (10, 25).

FIG 4.

Bacteria need not alter dietary protein to sugar ratios to influence D. melanogaster DP. We measured the glucose and protein content in the diet cultured with bacteria with flies (A), the diet cultured by bacteria without flies (B), and flies isolated from diet vials used in panel A (C). In each panel, different letters over the bars (mean ± SE) represent significant differences between treatments as determined by a Kruskal-Wallis test. Unless values were dropped, data are based on 6 replicate pools of 5 flies in each of three separate experiments.

DISCUSSION

Our results confirm and extend previous findings that the microbiota can influence fly dietary preference for yeast. For example, our current experiments corroborate previous findings that some, but not all, bacteria spare the flies of a preference for dietary yeast (10, 12, 25). Our work also extends these findings by demonstrating the varied influence of a broad panel of bacterial strains on fly DPY, including that some individual strains amplify or suppress preference for dietary yeast. Among the influences conferred by these singly-associated bacterial strains was that A. fabarum-colonized flies have a relatively low preference for dietary yeast and that this effect could be enhanced by disrupting A. fabarum transport genes or amelorioted by disrupting thiamine biosynthesis genes. Finally, we showed that despite their discordant influence on fly dietary preference, transposon insertions in a subset of these genes did not lead to differences in the protein to glucose ratio in flies or their diets. This last result confirms these genes do not affect fly dietary preference for yeast through modification of macronutrients (protein, glucose) in the diet, a known mechanism for other Acetobacter genes that influence D. melanogaster life history traits. Together, these findings identify bacterial genes that, when disrupted, affect the influence of associated microorganisms on fly dietary preference. The patterns of dietary preference change in the mutants also implicate thiamine, in particular, as a possible effector of fly DPY.

In our original experiments (Fig. 1; Fig. S1), we saw species-specific influences of Acetobacter tropicalis and Lactobacillus brevis on fly dietary preference. We originally selected these two bacterial species for analysis because of their distinct influences on D. melanogaster nutritional and life history traits (18, 19, 23, 24, 30). In previous experiments the influences of LAB and AAB strains were sometimes more similar to those conferred by other conorder, rather than heteroorder, strains. Also, order-level geographic variation in microbiota composition, where the LAB:AAB ratio in wild flies increased with latitude, was associated with the order-level influences of the bacteria on fly traits (23). In this study the broad differences in the strain-specific influences of the bacteria on fly DPY are not patterned with high taxonomic level (Fig. 2). Also, the effect of the microbiota on fly DPY is not associated with the influences of the microbiota on fly life history traits (Fig. S8). Together, these results suggest that microbiota-dependent variation in fly DPY is unlikely to be a mechanism that maintains order-level variation in the Drosophila microbiota composition or is related to control of microbiota-dependent fly life history traits.

We successfully identified bacterial mutants that influence fly DPY despite using a model for the MGWA that was not ideally fitted to the data. The most obvious consequence of the data’s poor fit to the model was dramatic inflation of P values in the MGWA. We tried unsuccessfully to reduce P-value inflation by using different statistical tests (linear mixed effects models, Wilcoxon tests, and generalized linear mixed effects models) and response variables (the preference index and the raw sip data) in the MGWA (Fig. S4). Previous association studies between bacterial gene presence-absence patterns and Drosophila traits, performed by our laboratory and others, have not always directly reported the degree of P-value inflation but have in some cases revealed no obvious signs of P-value inflation (20, 31, 33, 34); modest to substantial P-value inflation (18, 19); or P-value deflation (35). A shared characteristic between this study and the two previous studies with P-value inflation was that all three had relatively large n. The previous two studies measured individual fly survival in starvation (19) or life span (18) assays, with an n of hundreds of flies per treatment. In this experiment, we targeted approximately 20 replicates per sample, based on the recommendation by the FlyPad device manufacturers that an n of 20 individual flies is a likely minimum appropriate level of replication (P. Itskov, personal communication). When we used the actual numbers of fly sips directly, usually scores to hundreds of measures per fly, 68% of gene predictions were significant (8,156 of 12,167 OGs; Fig. S4C). The main approach we reported in the text instead used as a response variable a preference index, which was derived as the linear prediction from a model run on the raw data. Using the preference index as a response variable instead of the raw sip values incompletely reduced some P-value inflation. We cannot rule out that other efforts to model the data could reduce the P-value inflation, or that permutational bootstrapping approaches could help to identify or eliminate false positives. We also attempted to narrow the list of candidate genes by a KEGG enrichment analysis. In our past efforts, KEGG enrichment analyses have be useful to identify one or a few candidate pathways that are predicted to influence the fly trait (18, 19). However, in this study no KEGG pathways were significantly enriched in the top predictions (data not shown). Altogether, even though none of these approaches reduced the analysis to a small number of predictions, we continued with our goal to use the MGWA as a surrogate genetic screen and we directly tested if a few top gene predictions influenced fly dietary preference. The rate with which we confirmed effects was 50%. This relatively high validation frequency may suggest that many bacterial genes influence fly dietary preference as measured in our assay. A recent analysis of pupal length, a quantitative D. melanogaster trait, showed that many D. melanogaster genes influence pupal length regardless if they are predicted or not by a corresponding screen, in this case a GWA (36). Overall, we emphasize that it would be overinterpreting to conclude that any gene with a significant P value influences the trait unless there is an accompanying mutant analysis.

We used transposon mutagenesis to explore if eight genes influenced fly dietary preference and confirmed that transposon insertions in four of eight candidate genes influence fly dietary preference. Three of the four disrupted genes, thiE, iscS, and one encoding a purine/thiamine permease, are implicated in thiamine (vitamin B₁) biosynthesis or transport (37–39). ThiE catalyzes the formation of thiamine phosphate from two precursor molecules and IscS transfers a sulfur from cysteine to a thiamine precursor in an earlier step in thiamine biosynthesis. The permease is involved in transport of purines, which share key biosynthetic steps with thiamine (40). Connections between the gut microbiota, fly dietary preference, and thiamine are interesting because in the absence of dietary thiamine gut microbiota provisioning of thiamine is essential for D. melanogaster survival (41). In our experiments, the flies presumably received sufficient thiamine from the yeast in their diet, so we did not detect any issues with fly survival when reared with these mutants. However, if there is variation in thiamine content and the flies can sense it, this could be one explanation how the mutants influence fly dietary preferences. For example, if the different transposon insertion mutants influence thiamine content of the fly diet and the flies can sense such a change, then, e.g., insertions in thiamine biosynthesis genes could turn the strains into thiamine-consuming thiamine auxotrophs that lower dietary thiamine levels. Conversely, knocking out a thiamine permease could prevent the bacteria from importing and consequently lowering dietary thiamine levels. All of this is conjecture, unified by the idea that the influences of the different thiamine mutants on fly DPY varies with a predicted (but unmeasured) influence of those strains on thiamine content of the diet. There is precedence for animals detecting variation in thiamine contents of their diets. For example, prior thiamine deficiency can be sufficient to cause thiamine preference in rats (42) and rats with thiamine deficiency can be trained to prefer foods with either supplemental thiamine or high-fat, thiamine-sparing diet (43). Therefore, bacterial mutants that lower dietary thiamine levels could be expected to affect fly preference for diets made with varying amounts of yeast, which contains thiamine. Previous efforts have been unable to identify the molecule or molecules responsible for microbiota-dependent variation in fly dietary preference under yeast deprivation but have ruled out that the bacteria do so by changing the amino acid content of the flies (10, 25). As such, we suggest that thiamine could be a potential small molecule that contributes to the change in dietary preference of the flies. Also, the link between thiamine metabolism genes and fly DPY in this study adds to a substantial body of literature showing B vitamins and B vitamin metabolism genes play a major role in Drosophila-microbiota interactions (18, 19, 27, 41, 44–46). However, these conclusions must be tempered. We did not verify by complementation that B vitamin metabolism is the key function disrupted by the transposon insertions and cannot rule out that the mutations affected other, undetected bacterial roles.

We also identified that a transposon insertion in an A. fabarum gene encoding the glucose transporter OprB influences fly DPY. OprB is required for carbohydrate uptake in several Psuedomonas species (47, 48). The mechanism by which this gene influences fly DPY remains unknown and is not by modifying the protein to glucose ratio in the fly diet (Fig. 4). We previously showed that Acetobacter pasterianus glucose oxidation is a major mechanism by which bacteria influence fat storage in D. melanogaster (20). The presence of glucose and gluconate dehydrogenase genes can lower the amount of glucose in fly diet by converting dietary glucose to 2-ketogluconate (19). It seemed reasonable to hypothesize that dietary glucose transport could alter the yeast to glucose ratio of the diet and be responsible for the shift in fly DPY. However, our comparison of flies, their diets, and fly-free diets when inoculated with the oprB mutant showed no evidence that the oprB mutation affects dietary or fly glucose composition differently than an iscS mutation that amplifies the microbial influence on fly preference for dietary yeast. Expecting bacterial effects on dietary glucose to affect fly DPY in FlyPad assays may inappropriately conflate bacterial dietary activities before and during the assay. Bacteria could potentially modify the FlyPad diets if they were transferred from flies during the assay, but over a relatively short 60- to 90-min assay window these activities may be minor (consider previous work showing a bacteria-mediated decrease in dietary glucose content after 24 h [49] or the relatively minor accumulation of feces in small diet surfaces over 3 h [50]). Regardless, our assays show that glucose transport by the bacteria does not control fly DPY through changes in fly or dietary glucose content.

Together our data suggest avenues for further investigation, including the idea that microbiota-dependent thiamine biosynthesis, transport, or utilization could influence fly preference for dietary yeast under eAA restriction or deprivation. For example, the creation of clean mutants of the iscS, thiE, or purine permease genes could further clarify the roles of these genes in shaping fly DPY. We performed our analyses using uncomplemented transposon insertion mutants, which suffer from caveats of polarity. Recently, site-directed mutants have been constructed in A. fabarum, the background used for our transposon mutants, suggesting that tools are available for complementation or clean deletions of the target genes in our study (51). Also, whereas we confirmed qualitative colonization of the flies by the bacterial transposon insertion mutants, quantitative analysis of these or site-directed bacterial mutants could reveal if there are patterns between the fly yeast preference phenotype and bacterial abundance in the flies or their diets. For example, changes in bacterial abundance can help to explain other fly phenotypes such as triacylglyceride content, development time, and in some instances, life span (18, 24, 52) and could be one mechanism for how bacteria influence fly DPY. Additionally, our study was performed in flies reared from eggs with the different bacterial strains or mutants and does not reveal the timing of the bacterial effect. For example, is bacterial influence on fly dietary preference developmentally programmed (e.g., reference 53), or does fly dietary preference respond flexibly throughout its life to changes in microbiota composition? Finally, our study used dozens of monoassociations to identify potential bacterial influences from across a broad range of possible microbial partners. However, our study does nothing to investigate how microbial community interactions could influence fly dietary preferences, and such interactions are known to affect these and other fly phenotypes (10, 24–28). Further investigation of each of these areas could help explain more clearly how bacteria influence fly dietary preferences.

In summary, we showed that transposon insertions in bacterial transport and thiamine biosynthesis genes can influence the D. melanogaster preference for dietary yeast. These findings underscore the established role of B vitamins in fly-microbiota interactions, suggest dietary thiamine as a candidate effector of fly preferences for dietary yeast, and raise questions about possible roles for feedback loops between these fly preferences, bacterial activity in and colonization of diets, and bacterial colonization and possible attraction to their hosts. We expect that further exploration of interactions between D. melanogaster dietary preferences and the microbiota will provide clues to the role of fly dietary preferences and the microbiota in the natural history of the flies.

MATERIALS AND METHODS

General fly maintenance and bacterial culture conditions.

A D. melanogaster CantonS stock originally obtained from Mariana Wolfner was used in all experiments. This stock is not colonized by the reproductive tract endosymbiont Wolbachia. For general culture, flies were reared at 25°C on a 12-h light:12-h dark cycle at ambient atmosphere (~25% humidity). The diet for general rearing was 10% Brewer’s yeast (0290331225; MP Biomedicals), 10% glucose (158968-5KG; Sigma), and 1% agar (boiled together and cooled to 60°C), with 0.04% phosphoric and 0.42% propionic acid added after cooling (Y-G diet). Diets in other conditions were slightly modified and are described in the corresponding sections.

Bacterial growth conditions are given in Table 2. Briefly, bacterial strains were streaked for isolation and growth in liquid culture on modified de Man-Rogosa-Sharpe (mMRS) medium (1.25% peptone, 0.75% yeast extract, 2% glucose, 0.5% sodium acetate, 0.2% dipotassium hydrogen phosphate, 0.2% triammonium citrate, 0.02% magnesium sulfate heptahydrate, 0.005% manganese sulfate tetrahydrate, 1.2% agar [24]) or lysogeny broth (11-118; ApexBioresearch Products) medium for 3 to 5 days. Strain grown under oxic conditions was cultured with shaking (liquid) and ambient atmosphere (solid). For microoxic culture, liquid cultures were grown statically and plates were placed in an airtight container flooded with CO₂. All strains were grown at 30°C for 1 to 2 (liquid) or 3 to 5 (solid) days. Acetobacter fabarum transposon insertion mutants were streaked for isolation from a library stored at −80°C on yeast-peptone-glycerol (YPG; 0.5% Y, 0.5% P, 1% G) plates with 50 μg mL⁻¹ filter-sterilized kanamycin and subcultured in kanamycin-supplemented YPG broth before normalizing and inoculating to fly eggs (see below).

Axenic and monoassociated fly culture.

In order to treat the flies individually with different bacterial strains, we created axenic embryos and then inoculated the embryos separately with different bacterial strains as in our previous work (16). We first harvested D. melanogaster eggs, less than 18 h old, from grape-juice agar plates (Y-G diet, omitting the acids and supplemented with 2 tbsp Welch’s frozen grape juice concentrate per 500 mL). Then, we dechorionated the eggs in 0.6% sodium hypochlorite twice for 150 s each. Inside a sterilized biosafety cabinet, which maintained atmospheric sterility, we painted 30 to 50 sterile eggs onto 7.5 mL autoclaved and cooled Y-G diet (acid omitted) in 50-mL centrifuge tubes, using a sterile paintbrush. To rear axenic flies the eggs in the vials were left undisturbed. To rear monoassociated flies, individual bacterial strains were grown in overnight culture, washed once in 1× PBS (8 g NaCl, 0.2 g KCl, 1.44 g Na₂HPO₄, 2.4 g KH₂PO4 [54]), normalized in PBS to optical density (OD) at 600 nm = 0.1, and added to the eggs in a 50 μL volume. In each of the Fig. 1 experiments, 2 to 4 replicate fly vials were picked in each of three separate days, seeding 2 to 4 replicate FlyPad experiments with at least 20 flies per treatment. In the Fig. 2 experiment, 2 to 4 replicate vials were picked in each of 3 separate days, seeding 18 replicate FlyPad experiments performed over 6 days with 0 to 3 flies per treatment. In the Fig. 3 experiment, 4 to 6 replicate vials were picked in each of three separate days, seeding 15 replicate FlyPad experiments performed over 5 days with 2 flies per treatment. For each set of experiments we targeted approximately 20 flies per treatment group total, and the variable number of replicate vials and experimental days varied to accommodate different numbers of treatment groups in each experiment. Together, these approaches enabled us to rear flies with a similar N in each different microbiota by diet combination (treatment).

flyPAD dietary preference assay.

We measured dietary preference of individual 4- to 6-day-old female flies using flyPAD arenas (6). Each arena is an enclosure for a single fly that contains two diet wells, each well composed of two electrodes. When flies stand on one electrode and extend their proboscis into the diet contained in the second electrode, the capacitance changes and a computer records a feeding event, or sip. The creators of the hardware reported that because of the way sips are measured, a minority of feeding activities are missed (7.5%) and about 7.5% of nonfeeding activities such as walking across the diet and platform simultaneously can also be recorded as sips (6). Fly feeding preferences can be compared by placing a different diet in each diet well.

At 4 to 6 days posteclosion (dpe) we aspirated female flies from a single vial and starved them in empty fly containers for 3 h to synchronize them for appetite. We used a comparable duration of starvation time (2 h) in our previous work (30), and flies on our diet can survive on average for 1.5 to 3 days depending on their microbiota composition (19), together suggesting negligible negative health effects of the starvation over a short 3-h period. Immediately before adding flies to the flyPAD arenas, we prepared each arena by placing 6 μL of two types of diets (control Y-G diet or trial diet) into one of the two wells in each arena. For each diet, 500-μL aliquots were prepared previously, stored at −20°C, then briefly melted in a microwave, and kept at 70°C in a heat block while inoculating all flyPAD arenas for a given experiment. We always added the control Y-G diet to the left well and the trial diet to the right well. To control against directional and desiccation bias respectively, we alternated in different experiments the orientation of the flyPAD modules (each module contains 4 arenas) inside the incubator during the tests and which diet was inoculated to the arenas in the module first. Test diets were composed of the same reagents as the Y-G diet, but in different ratios as shown in Fig. S1 and detailed in Table S1. Flies were added to each arena after diet had been added to all arenas. Then, after flies were added to all arenas (usually <10 min between introducing the first and last fly), the flyPAD modules were placed in an incubator set at 25°C, lights on, and 75% humidity, connected to a computer, and readings began. Figure 1 experiments ran for ~90 min. Figure 2 and 3 experiments ran for ~60 min.

During the assay, flies were allowed to feed ad libitum and dietary preference was measured continuously over that time period and recorded as sips. We focused on the number of sips as the main response variable. To ensure the microbiota of the flies was as expected, i.e., against contamination, we homogenized and dilution plated on mMRS a pool of five vial-matched male siblings from each vial. If the siblings displayed greater than 100 CFU fly⁻¹ of a contaminating microbe, dietary preference readings for all test flies derived from that vial were removed before calculating the statistics. Flies were also removed from the analysis if both diet troughs registered 0 sips or if at least 1 trough registered >1,000 sips. We also qualitatively examined the microbial load in the homogenate and confirmed that each bacterial mutant in our analysis colonized the flies at a minimum threshold of 2,000 CFU per fly or >100 CFU per 10-μL dilution.

Differences in fly dietary preference between the different treatments were determined using a generalized linear mixed effects model with a binomial family. The raw number of sips each fly took on the Y-G diet and on the trial diet was used as the response variables, bacterial strain was the main effect, and experimental replicate was a random effect. For Fig. 2, the time of day an experiment was performed was also collected. For each diet, significant differences between treatments were determined by a post hoc Tukey test and are presented as compact letter displays (55). In the figures we present the differences using a preference index, a value extracted from the model as a linear prediction for each inputted value. Together, these approaches allowed us to define the differences between dietary preference of flies that bore different or no colonizing microorganisms.

We also calculated diferences in the total number of sips taken by flies on both diets combined and on each diet separately, and the duration of each sip on each diet, to gain additional insight into how the bacteria influence fly feeding behaviors. To measure total feeding we calculated the sum of sips from both wells during the trial. Then, we compared the total feeding values between treatments using a Kruskal-Wallis test, followed by a Dunn test for post hoc multiple comparisons (Fig. S2A, S3A, and S5A; Table S2). We also used a Kruskal-Wallis test and post hoc Dunn test to compare the average sip duration, which was directly measured by the FlyPAD in seconds, between both microbial treatment and diet type (Fig. S2B, S3B, and S5B; Table S2). Finally, we calculated the average number of sips the flies took on each diet separately. We performed multiple comparisons tests for significant differences between average number of sips on each diet in two ways. First, we used a post hoc Dunn test to determine significant differences between all treatments (Fig. S2C, S3C, and S5C; Table S2 and S6). Second, we tested if fly feeding on each diet increased, or decreased relative to axenic flies (Fig. S3C; Table S2) or the wild-type A. fabarum strain (Table S6; no asterisks are shown in Fig. S5C because there were no significant differences) using pairwise Kruskal-Wallis tests.

Metagenome-wide association.

To predict bacterial genes that influenced D. melanogaster DPY, we used a MGWA approach as in our previous work (20). We obtained amino acid sequences from the GenBank entries for the exact bacterial strains used in the monoassociation experiments. Then, we clustered the amino acid sequences into OGs using OrthoMCL software with default parameters and an inflation value of 1.5. Differences in fly DPY were determined by a linear mixed effects model. The response variable was the preference index, which was extracted as the linear prediction of a generalized linear mixed effects model with a binomial family run on the Fig. 2 data as described in the previous section. Briefly, the linear predictions normalize the different number of total sips taken and the variation across replicate experiments on different days and times. In addition, OG presence/absence (0 or 1) was the main effect, microbial treatment and principal components representing half of the population structure were additional fixed effects, and the time (1 PM, 4 PM, or 7 PM) and date the flyPAD assay began were random effects. Then P values were Bonferroni corrected for multiple tests. Quantile-quantile plots showed that these efforts did not ideally model the data (Fig. S4A), so we also performed MGWA using two additional models. We used a Wilcoxon test to determine if the preference index (response variable) varied with the OG presence-absence pattern in the bacteria (Fig. S4B). We also used a generalized linear mixed effects model with a binomial family directly, with the preferences for each diet as the response variables, OG presence-absence as the main effect, the principal components representing half of the population structure as additional fixed effecs, and the time (1 PM, 4 PM, or 7 PM) and date of the tests as random effects (Fig. S4C). The MGWA was performed using the MAGNAMWAR R package (56), with some custom scripting to run the generalized linear mixed effects model with a binomial family.

The phylogenetic tree for Fig. 2 was built from ~1,400 nucleotides of the 16S rRNA gene sequence from each sequenced genome. In some cases where the 16S sequence could not be identified in the draft genome sequences by BLAST, a top BLAST match from a conspecific was used. The sequence for Halobacterium jilantaiense JCM 13558 (NR_113425) was included as an outgroup. The sequences were converted to RNA and aligned with the EMBL-EBI online MUSCLE tool (57) using default ClustalW parameters. From these results we downloaded a neighbor-joining tree built without distance correction and used FigTree v 1.4.4 to format and rotate the nodes for presentation in Fig. 2 (58).

Glucose and protein content assay.

To test if bacterial mutants influenced the glucose and protein contents of flies or their diets, we used colorimetric assay kits as in our previous work (20, 59). Briefly, flies were reared in monoassociation with each bacterial mutant to 5–7 day old adults. As controls, fly-free diets were inoculated with 50 μL of bacteria and placed side-by-side in an incubator with the bacteria- and fly-inoculated diets. Then, female flies were sorted into pools of 5 on a CO₂ pad, homogenized for 2 min at 1750 RPM in 125 μL Tris-EDTA-Triton X-100 (TET; 10 mM Tris, pH 8; 1 mM EDTA; 0.1% Triton X-100) and 125 μL 1.4-mm ceramic beads using a 2010 GenoGrinder (Spex) and centrifuged as 13,000 × g for 1 min. Then, 50 μL of cleared lysate was heated to 72°C for 10 min and stored at −80°C for future use in triacylglyceride determination assays. Also, 10 μL of homogenate was directly stored at −80°C for use in measuring protein content. At the same time the flies were sorted, ~20 mg of fly diet was also sampled from the spent fly vials using a wooden stick. The material was removed by twisting the stick into a microcentrifuge tube containing 125 μL beads and 125 μL TET buffer, then preparing for colorimetric assays exactly as the flies were. Later the same day, fly-free diet samples were prepared for colorimetric assays exactly as the spent-fly diets were prepared.

The glucose and protein content of each sample was measured from 5 μL of the prepared homogenate using commercially available colorimetric kits following manufacturer’s instructions. The protein content of each sample was measured using the Bio-Rad protein determination kit (product no. 500-0113 and 500-0114). A standard curve was created by measuring in duplicate wells the OD₇₅₀ of 10 μL of 0, 0.1, 0.2, 0.4, 0.7, and 1 mg mL⁻¹ dilutions of bovine serum albumin. To test wells we sequentially added 5 μL of homogenate, 25 μL of reagent A, and then 200 μL of reagent B, incubated samples for 15 min at room temperature, and measured their OD₇₅₀ on a Biotek Epoch plate reader. The glucose content of each sample was measured using Sigma-Aldrich glucose assay measurement reagents. The standard curve was created by measuring in duplicate wells the OD₅₄₄ of 5 μL of glucose at the following dilutions (158968-5KG, Sigma): 0, 0.02, 0.04, 0.08, 0.14, and 0.2 mg/mL. In 96-well plates we added 5 μL of homogenate, 150 μL of the assay reagent (1 capsule glucose oxidase/peroxidase reagent (G3660; Sigma) dissolved in 39.2 mL ddH₂O, 50 mg of O-dianisidine dihydrochloride (F5803-50MG; Sigma) predissolved in 800 μL ddH₂O), incubated for 30 min at 37° C, added 150 μL of 6.25 M H₂SO₄, and measured OD₅₄₄.

Data availability.

The data analysis was performed in R using various packages (56, 60–70). The code can be accessed as an R tutorial by running ‘devtools::install_github(“emarsh25/TBCdietpref@master”)’ and ‘learnr::run_tutorial(“TBCprefAnalysis,” “TBCdietpref”)’ in an RStudio console where devtools and learnR, plus other packages executed in the tutorial, are installed. After running this code, the tutorial should appear in the “Tutorial” pane, and can be started by clicking “Start Tutorial.” The code can also be accessed as Data set S1. Links to the raw data are provided on github and linked to in the supporting code.