Functional traits of the gut microbiome correlated with host lipid content in a natural population of Drosophila melanogaster

Abstract

Most research on the nutritional significance of the gut microbiome is conducted on laboratory animals, and its relevance to wild animals is largely unknown. This study investigated the microbiome correlates of lipid content in individual wild fruit flies, Drosophila melanogaster. Lipid content varied 3.6-fold among the flies and was significantly correlated with the abundance of gut-derived bacterial DNA sequences that were assigned to genes contributing to 16 KEGG pathways. These included genes encoding sugar transporters and enzymes in glycolysis/gluconeogenesis, potentially promoting sugar consumption by the gut microbiome and, thereby, a lean fly phenotype. Furthermore, the lipid content of wild flies was significantly lower than laboratory flies, indicating that, as for some mammalian models, certain laboratory protocols might be obesogenic for Drosophila. This study demonstrates the value of research on natural populations to identify candidate microbial genes that influence ecologically important host traits.

1. Introduction

Lipid, particularly triacylglycerol (TAG), is a key energy reserve for animals, with important consequences for animal health and fitness [1,2]. Insufficient stored lipid can depress starvation survival, impair reproductive output and jeopardize critical life events, such as diapause and migration [3–6]. Equally, excessive lipid storage associated with over-feeding and inactivity in humans and other animals is causally linked to metabolic disease and some cancers [7,8].

There is growing evidence that the allocation of dietary inputs to lipid can be influenced by the composition and activity of gut microorganisms (collectively, the microbiome) [9–13]. Many of the underlying mechanisms are conserved across the animal kingdom [1] and have been investigated in the laboratory mouse and also simple, experimentally tractable animal models [14], including the fruit fly Drosophila melanogaster [15]. The TAG content of Drosophila in laboratory culture is dramatically increased by perturbation or elimination of the gut microbiome [16–19]. This effect has been attributed to both microbial-mediated modulation of insulin-like signalling [16,20] and microbial competition for dietary carbohydrate, which reduces the calories available for assimilation by the animal [21]. However, interpreting these studies beyond the laboratory is constrained by the evidence that the gut microbiome of long-term laboratory Drosophila cultures is taxonomically impoverished and functionally different from natural Drosophila populations [22–24].

The purpose of this study was to investigate the relationship between the host lipid content and the gut microbiome in wild D. melanogaster. The experimental design comprised parallel analysis of carcass TAG content and 16S rRNA gene amplicon sequencing (bacterial taxonomy) and shot-gun sequencing of the total DNA (i.e. the metagenome, an index of the functional potential of the microbial community) of guts dissected from individual flies in a single natural population. We also compared the TAG content of the wild flies and Drosophila in standard laboratory culture. The analysis was restricted to males because female D. melanogaster are morphologically indistinguishable from co-occurring D. simulans. We established that the wild flies are significantly leaner than our standard laboratory Drosophila, and that their TAG content is significantly correlated with several functional traits, but not taxonomic composition, of the gut microbiome.

2. Material and methods

(a). Collection of wild Drosophila melanogaster

Adult drosophilid flies were collected on September 21, 2017 from a dumpster containing rotting fruits at Cornell Orchards, Ithaca, NY, USA. Each of 41 male D. melanogaster was washed thrice with 0.5 ml chilled, sterile phosphate-buffered saline (PBS) and dissected, using 70% ethanol-sterilized forceps, in 50 µl PBS to separate the gut (including crop and Malpighian tubules) and eviscerated carcass. The two samples were transferred to separate, sterile 2 ml screw-cap tubes tubes (Laboratory Product Sales Inc.), flash frozen in liquid nitrogen and stored at −80°C.

(b). Lipid assays

The eviscerated carcass of the 41 wild flies and five male laboratory flies (strain Canton S, of undefined age and mating status) taken from routine culture on a diet of 10% dried yeast/10% glucose (w/v) at 25°C with 12 L:12D was assayed for TAG content (Triglyceride Assay Kit, Sigma TR0100-1KT) and protein content (DC kit, BioRad 500-0111) (details in electronic supplementary material, text S1).

(c). Molecular analyses

The full molecular procedures are provided at electronic supplementary material, text S1. Briefly, DNA was extracted from each dissected gut using the phenol–chloroform method [25], with two fly-free negative controls (0.5 ml PBS) processed in parallel. For bacterial taxonomy, the V3–V4 region of bacterial 16S rRNA genes was amplified with primers 16SF 5′-CCTACGGGAGGCAGCAG-3′, 16SR 5′-GACTACHVGGGTWTCTAAT-3′ and dual indexing protocol [26]. Libraries were submitted for 250 bp paired-end sequencing by MiSeq (Illumina). Sequences were processed using FastQC (www.bioinformatics.babraham.ac.uk/projects/fastqc/), QIIME2_v2018.4 (qiime2.org) [27] and DADA2 pipeline to resolve amplicon sequence variants (ASVs). Taxonomic assignments for bacteria used a machine learned naive Bayes classifier trained on the SILVA128_99% reference sequences [28,29]. In total, 8 463 636 reads were assigned to bacterial ASVs, with 46 427–484 095 reads per fly (Dataset S1a); and the sequence coverage plots indicated saturation of bacterial diversity (electronic supplementary material, figure S1a). Negative control samples yielded no bacterial ASVs. Bacteria with less than 50 reads across the full dataset were removed (Dataset S1b), yielding 8 461 015 reads with 206 366 ± 17 369 (mean ± s.e.) reads per fly. SILVA 16S rRNA designations were verified via NCBI BLAST.

For gut metagenomes, multiple displacement amplification was performed on 10 ng DNA per fly, utilizing the REPLI-g Mini Kit (Qiagen). Libraries were prepared with TruSeq DNA PCR-Free Low Throughput Library Prep Kit (Illumina) and barcoded set A and B indexed adapter sequences. Paired-end (2 × 150 bp) sequencing was conducted on a HiSeq 2500 platform (Illumina). The Fastq sequence files were uploaded to the Metagenomics Rapid Annotation using Subsystems Technology (MG-RAST) server (version 4.0.3) [30,31] and, following quality controls (Dataset S2a), the protein-coding gene calling was performed via the FragGeneScan machine learning approach. Sequence coverage plots indicated saturation of KO identifiers (electronic supplementary material, figure S1b). Functional annotation of clusters assigned to domain Bacteria on the NCBI RefSeq Database (ncbi.nlm.nih.gov/refseq/) used the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (genome.jp/kegg/), and annotations used the Best Hit Classification approach with e-value ≤1 × 10⁻⁵ and 99% identity.

(d). Statistical analyses

Statistical analyses were performed in R [32]. Patterns in the datasets were visualized using Bray–Curtis dissimilarity with principal coordinates analysis (PCoA) for bacterial ASVs and principal components analysis (PCA) for KEGG identifiers in the metagenome datasets. (Presence/absence analyses using Jaccard matrices were not pursued because a low proportion of the variance was explained by the first two axes for both the 16S and metagenome datasets.) Pairwise correlations of fly TAG content with each of ASV abundance and KEGG pathways used an adapted Kendall's τ_b (mazeinda package in R) [33,34], suitable for datasets including many zero abundance levels, with Benjamini–Hochberg false discovery correction for multiple comparisons and threshold q-value of 0.2 [35]. The relationship between fly TAG and Wolbachia relative abundance was tested using Spearman's rank correlation. The TAG content of wild flies and laboratory flies was compared by t-test for samples with equal variance, following confirmation that the data conformed to normal distributions with homogeneous variances.

3. Results

(a). Lipid content of flies

The triglyceride (TAG) of the eviscerated carcass of the 41 male flies collected from a natural population varied 3.6-fold, from 88 to 318 µg mg⁻¹ protein (figure 1a). The mean TAG content of wild flies (184.9 ± 7.36 µg mg⁻¹, mean ± s.e.) was significantly lower than laboratory flies (266.9 ± 23.7 µg mg⁻¹) (t₄₄ = 3.63, p = 0.001) (electronic supplementary material, table S1).

Figure 1.

The wild Drosophila melanogaster. (a) TAG content. (b) Taxonomic composition of the gut bacterial community (minus intracellular symbionts). (c) Principal coordinates analysis (PCoA) of Bray–Curtis distance between samples based on ASV relative abundance. (d) Bacterial metagenome represented as the relative abundance of KEGG categories. Data represent all genes assigned to domain Bacteria apart from those for intracellular symbionts. (e) Principal component analysis (PCA) of Bray–Curtis dissimilarity based on relative abundance of gene pathways.

(b). Taxonomic composition of the gut bacterial communities

The 16S rRNA gene reads amplified from wild fly guts were assigned to 195 ASVs, including seven ASVs with predicted intracellular lifestyles: four Wolbachia ASVs and one ASV of Holosporaceae, Paracaedibacter and Diplorickettsia (Dataset S1b, electronic supplementary material, figure S2). Of the four Wolbachia ASVs, one was detected in every fly and accounted for 90–100% of all the reads of putative intracellular bacteria. The remaining 188 ASVs were likely localized to the gut lumen, and we refer to them as the gut microbiome.

The gut microbiota was dominated by bacteria of the family Acetobacteraceae (α-proteobacteria) (figure 1a), accounting for 87 ± 2.5% (mean ± s.e.) of the total bacterial reads per fly. Other bacteria included γ-proteobacteria (especially Tatumella, Pseudomonas and Orbaceae) and, generally at low abundance, Lactobacillales. Three ASVs of Commensalibacter (Acetobacteraceae) were either detected in every fly (Commensalibacter #1 and #2) or in 33 (80%) of the 41 flies (Commensalibacter #3) and, together, accounted for greater than 50% of the reads in all but three flies (11, 38 and 123). The first two axes of the PCoA accounted for 54.9% of the total variance, largely attributable to the three dominant Commensalibacter ASVs (figure 1b).

(c). Metagenome analysis of the functional composition of bacterial communities

The metagenomic analyses yielded 12.7 ± 0.72 million reads (mean ± s.e) per fly (Dataset S2a), 62 ± 3.1% of which were assigned to bacteria (Dataset S2b; electronic supplementary material, figure S3a). Other reads were predominantly eukaryotic (37 ± 3.1% of total reads) and mapped largely to Arthropoda (21 ± 2.5% of total reads, all assigned by BLAST to D. melanogaster), with fungi of the phylum Ascomycota accounting for 4.3 ± 0.6% of total reads (electronic supplementary material, figure S3b). Following removal of sequences assigned to intracellular bacteria, the samples yielded 37 579.3 ± 4913 bacterial protein-coding genes per fly (Dataset S2a), which we considered as representative of the gut microbiome.

The major KEGG categories were detected in the gut metagenome of every fly (figure 1c; Dataset S3). In the PCA, the first two axes accounted for 48.0% of the variance, with carbohydrate metabolism, translation and membrane transport predominantly along the first axis and amino acid metabolism on the second axis (figure 1d).

(d). Relationship between triacylglycerol content of the flies and bacterial traits

No significant relationship between fly TAG content and bacterial taxa (16S data) was obtained after Benjamini–Hochberg false discovery correction for multiple comparisons, whether using relative abundance (at the taxonomic levels of ASV, species, genus and family) (Dataset S4a) or the scores for the first and second axes of the PCoA analysis of ASVs (Dataset S4b). Recognizing that Wolbachia can influence energy metabolism [36,37], we tested for a relationship between fly TAG content and relative Wolbachia abundance, but this correlation was also non-significant (Spearman correlation: ρ = −0.053, p = 0.74).

When we tested for correspondence between fly TAG content and gut microbiome function, as identified from metagenomic data, significant correlations were obtained for 16 KEGG pathways (table 1). Extending the analysis to the gene level identified individual genes significantly associated with TAG content for 14 of the pathways (table 1; Dataset S5). Two pathways negatively associated with TAG content (phosphotransferase system (PTS); glycolysis/gluconeogenesis) included at least five genes with negative τ_b (correlation coefficient); and two positively correlated pathways (porphyrin and chlorophyll metabolism; alanine, aspartate and glutamate metabolism) also included five or more genes with positive τ_b (Dataset S5). Noteworthy among negatively correlated genes are PTSs annotated to mediate the selective uptake of nine sugars, including fruit sugars (e.g. sucrose, glucose and fructose), the principal insect blood sugar trehalose and N-acetyl glucosamine, the monomer of chitin, which is an abundant component of the acellular layer separating the Drosophila gut epithelium and gut lumen [38]. The negative correlations between genes functioning in glycolysis/gluconeogenesis can be attributed to both glycolytic degradation of sugars in energy production and diversion of TCA intermediates from fatty acid synthesis to the production of glucose and other carbohydrates by gluconeogenesis. The nine gene entries in ‘porphyrin & chlorophyll metabolism' positively correlated with fly TAG content contribute to the synthesis of cobamides (vitamin B₁₂ family of cofactors).

Table 1.

KEGG pathways significantly correlated with fly TAG content.

KEGG pathway	Kendall's τb	q-valuea	gene countb
(a) negatively correlated with fly TAG content
03013 RNA transport [PATH:ko03013]	−0.344	0.085	3 (0)
02060 phosphotransferase system (PTS) [PATH:ko02060]	−0.312	0.085	31 (1)
00010 glycolysis/gluconeogenesis [PATH:ko00010]	−0.276	0.133	9 (1)
00230 purine metabolism [PATH:ko00230]	−0.259	0.181	2 (2)
05111 Vibrio cholerae pathogenic cycle [PATH:ko05111]	−0.150	0.181	3 (0)
00511 other glycan degradation [PATH:ko00511]	−0.135	0.189	1 (0)
(b) positively correlated with fly TAG content
00983 drug metabolism - other enzymes [PATH:ko00983]	0.337	0.085	1 (0)
00910 nitrogen metabolism [PATH:ko00910]	0.329	0.085	4 (0)
00860 porphyrin and chlorophyll metabolism [PATH:ko00860]	0.324	0.085	9 (2)
00361 chlorocyclohexane and chlorobenzene degradation [PATH:ko00361]	0.317	0.085	0
04112 cell cycle - Caulobacter [PATH:ko04112]	0.312	0.085	5 (2)
00600 sphingolipid metabolism [PATH:ko00600]	0.300	0.108	1 (0)
00510 N-glycan biosynthesis [PATH:ko00510]	0.283	0.133	1 (0)
00900 terpenoid backbone biosynthesis [PATH:ko00900]	0.283	0.133	0
00120 primary bile acid biosynthesis [PATH:ko00120]	0.280	0.133	1 (0)
00250 alanine, aspartate and glutamate metabolism [PATH:ko00250]	0.278	0.133	7 (3)

^aThe significance threshold for q-value is set at 0.2.

^b(a) Displays the number of genes with a significant negative correlation with TAG content and (b) displays the number of genes with a significant positive correlation with TAG content. In parentheses are displayed the number of genes with the reverse significant relationship (positively correlated despite assignment to a pathway negatively correlated with TAG content, or negatively correlated, despite assignment to a pathway positively correlated with TAG content).

4. Discussion

We have demonstrated that the lipid content of Drosophila collected from a natural population is correlated with various functional traits of the gut microbiome (table 1), but not with bacterial taxonomy as identified from 16S rRNA gene sequence data. These field correlations are open to alternative explanations. Specific microbial traits may favour (or disfavour) fly lipid storage; the conditions and resources in the gut of flies with different lipid content may favour microorganisms with different functional traits; and fly lipid or microbiome traits may covary with other traits (of the fly or microbiome) that are causally related. Furthermore, some correlations may not be robust to subsequent surveys of field populations or experimental analysis of causality, given that the age, genotype and diet of the flies studied here were uncontrolled and unknown. Nevertheless, we can infer likely processes from current knowledge. In particular, the demonstration that, under laboratory conditions, bacterial consumption of dietary sugar depresses fly lipid content [21] can explain the negative correlation between bacterial genes encoding sugar-transporting PTSs and enzymes in central carbon metabolism. Relevant to the positive correlation between fly lipid and microbial cobamide-synthesizing enzymes, vitamin B₁₂ deficiency in humans and rodents has been associated with dyslipidaemia and elevated TAG content [39,40]. This raises the testable hypothesis that microbial cobamide production promotes Drosophila lipid content, with the possibility that the leanest wild flies were vitamin B₁₂ deficient.

Several processes may contribute to the non-significant correlation between the taxonomic composition of the microbiota and fly lipid content. Divergent taxa may possess functionally equivalent traits that have been conserved over substantial evolutionary distances (e.g. many reactions in central carbon metabolism) or transferred horizontally between phylogenetically distant taxa, e.g. cobamide synthesis genes [41–43]. In addition, closely related bacterial genotypes with high 16S sequence identity can be functionally different because specific genotypes have acquired genes horizontally or lost genes, especially genes coding costly metabolic functions mediated by other members of the microbial community [44,45]. Supporting a role for these processes, there is considerable microdiversity and high horizontal gene transfer rates in gut-associated bacterial taxa [46–48], and extensive functional differences and environmental specialization among bacterial genotypes with greater than 99% 16S sequence identity in various habitats [49–51]. Although future studies with larger sample sizes and greater statistical power may reveal statistically significant taxonomic associations, our results reinforce the growing appreciation that the taxonomic resolution achievable with 16S sequences can often be inadequate to capture functionally important variation in the gut microbiome.

We conclude by considering how research on natural animal populations can promote understanding host–microbiome interactions. As our study illustrates, this approach can generate gene associations with ecologically important host traits, suitable for subsequent testing of causality by experimental study. Field studies also provide a check for the design of laboratory studies. For example, the significantly lower lipid content of the field flies than flies in our routine laboratory culture questions whether some laboratory flies may bear excessive lipid and that modifications to laboratory cultures (e.g. diet composition) are needed to ensure a ‘healthy' base-line for laboratory study of microbiome–host interactions. These considerations echo concerns that the standard rearing conditions for mammalian laboratory animals may also be obesogenic [8], potentially distorting the host–microbiome interactions in these systems.

Data accessibility

All data are available as electronic supplementary material or at NCBI Short Read Archive under PRJNA57929.

Authors' contributions

A.E.D. conceived the study. D.K. conducted the experimental work and analysis. D.K. wrote the first manuscript draft. Both authors revised it for intellectual content and approved the final version for publication. D.K. and A.E.D. are accountable for all aspects of the accuracy and integrity of this work.

Competing interests

We declare we have no competing interests

Funding

This research was funded by NSF grant Bio 1241099 to A.E.D. and NIH grant R01GM095372.