Perilipin-2 modulates dietary fat-induced microbial global gene expression profiles in the mouse intestine

Abstract

Background

Intestinal microbiota are critical determinants of obesity and metabolic disease risk. In previous work, we showed that deletion of the cytoplasmic lipid droplet (CLD) protein perilipin-2 (Plin2) modulates gut microbial community structure and abrogates long-term deleterious effects of a high-fat (HF) diet in mice. However, the impact of Plin2 on microbiome function is unknown.

Results

Here, we used metatranscriptomics to identify differences in microbiome transcript expression in WT and Plin2-null mice following acute exposure to high-fat/low-carbohydrate (HF) or low-fat/high-carbohydrate (LF) diets. Consistent with previous studies, dietary changes resulted in significant taxonomic shifts. Unexpectedly, when fed a HF diet, the microbiota of Plin2-null and WT mice exhibited dramatic shifts in transcript expression despite no discernible shift in community structure. For Plin2-null mice, these changes included the coordinated upregulation of metabolic enzymes directing flux towards the production of growth metabolites such as fatty acids, nucleotides, and amino acids. In contrast, the LF diet did not appear to induce the same dramatic changes in transcript or pathway expression between the two genotypes.

Conclusions

Our data shows that a host genotype can modulate microbiome function without impacting community structure and identify Plin2 as a specific host determinant of diet effects on microbial function. Along with uncovering potential mechanisms for integrating how diet modulates host and microbial metabolism, our findings demonstrate the limits of 16S rRNA surveys to inform on community functional activities and the need to prioritize metatranscriptomic studies to gain more meaningful insights into microbiome function.

Electronic supplementary material

The online version of this article (10.1186/s40168-017-0327-x) contains supplementary material, which is available to authorized users.

Background

Obesity is a substantial public health concern with increasing prevalence worldwide [1]. Since 1980, worldwide obesity has more than doubled and in 2014, 13% of adults were considered obese [2]. A critical determinant of obesity risk is the intestinal microbiota [3–5], which has been linked to a wide range of co-morbidities, including metabolic syndrome [4], gastrointestinal disease [6], type-2 diabetes (T2D), [7] and non-alcoholic fatty liver disease (NAFLD) [8, 9]. Metabolic functions encoded by the intestinal microbiome have a significant impact on the host [10]. In addition to the extraction of key nutrients such as amino acids and vitamins, fermentation by the intestinal microbiota produces short-chain fatty acids (SCFAs), such as butyrate, which are the primary energy sources for colonic epithelia and essential to the development of villus morphology within the GI tract [11, 12]. Although previous studies have shown that diet can have a dramatic impact on the composition and function of the gut microbiome [10, 13, 14], much less is known concerning the role of the host in shaping diet-microbiome interrelationships.

Cytoplasmic lipid droplets (CLDs) are intracellular organelle-like structures that play critical regulatory roles in lipid homeostasis [15–21]. Unlike other lipoprotein particles that enter circulation (e.g., LDL, HDL), CLDs form exclusively intracellular, cytoplasmic organelles storing neutral lipids [15–20]. Within enterocytes, CLDs are thought to function as sites of temporary storage for dietary lipids, which eventually are incorporated into chylomicrons and secreted into the lymphatic capillaries draining the villi of the small intestines [15, 22]. Underlying the formation and regulation of CLDs are members of the perilipin family which act as protein scaffolds [23–26]. Within this family, perilipin-2 (Plin2) is among the most highly expressed, particularly in the liver and small intestine [21, 27–31]. In a previous work, we documented in a transgenic whole-body Plin2 knockout mouse model (Plin2-null) that Plin2 modulates rapid (< 4 days) effects of diet on fecal lipid levels, enterocyte CLD contents, and fuel utilization properties of mice that correlate with differences in their gut microbial communities [32]. What is not known is how interactions between diet and Plin2 genotype modify the functional, as opposed to compositional, properties of the gut microbiome.

To date, most microbiome investigations have relied on the use of 16S rRNA surveys. Although such surveys provide details of community structure, they provide only limited functional insights. Algorithms such as PICRUSt [33] can be applied to deduce the functional capacity of a microbiome based on taxonomic abundances; however, such capacity does not directly translate to functional activity. Consequently, whole microbiome RNASeq (metatranscriptomics) has emerged as a powerful technology to interrogate microbiome function and define members of microbial communities in terms of their functional activities [34]. Here, we apply metatranscriptomics to compare the impact of two dietary regimes (high-fat/low-carbohydrate vs. low-fat/high-carbohydrate) on the structure and function of the intestinal microbiomes of Plin2-null and WT mice. Note that throughout, we define community structure on the basis of expressed transcripts rather than, for example, marker genes such as 16S rDNA genes.

Results

Knockout of Plin-2 has minimal impact on the biodiversity of the mouse intestinal microbiome in the context of low-fat and high-fat diets

In previous work, we used 16S rRNA surveys and metagenomics to show that both dietary fat content and Plin2-null genotype could significantly and independently impact gut microbiome composition, diversity, and function [32]. To derive a more mechanistic understanding of the relationship between dietary fat content, Plin2 genotype, and microbiome function, we undertook a metatranscriptiomics analysis of the colon contents of both wild type (WT) and Plin2-null (Plin2) mice shifted from chow to either a low-fat/high-carbohydrate (LF) or high-fat/low-carbohydrate (HF) diet for 4 days. Bulk RNA was prepared from colon contents and sequenced, resulting in ~ 419 million sequence reads from 16 mice (4 replicates of each diet/genotype combination) of which ~ 234 million reads (55.8%) were of putative bacterial mRNA origin (Table 1; see Additional file 1: Table S1 for additional information on sequence reads).

Table 1.

Summary of sequence read processing

Genotype	Diet	Total reads	Putative mRNA reads	% putative mRNA	Annotated mRNA reads	% of putative mRNA reads annotated	Unique transcriptsa	Unique enzymesb
Plin2KO	HF	23,259,351	12,202,390	52.46	8,324,287	68.22	354,623	1467
Plin2KO	HF	26,835,502	14,899,443	55.52	12,247,229	82.20	421,650	1499
Plin2KO	HF	30,521,645	16,599,182	54.38	13,388,961	80.66	403,966	1464
Plin2KO	HF	26,597,134	14,650,846	55.08	11,925,997	81.40	373,227	1465
Plin2KO	LF	22,294,860	12,361,287	55.44	10,610,304	85.83	331,728	1421
Plin2KO	LF	24,177,170	13,568,281	56.12	11,565,876	85.24	367,713	1453
Plin2KO	LF	27,308,384	15,414,942	56.45	13,153,267	85.33	337,142	1467
Plin2KO	LF	30,451,443	17,236,702	56.60	14,227,281	82.54	361,329	1481
WT1	HF	21,588,643	11,895,187	55.10	8,943,569	75.19	321,451	1462
WT2	HF	18,204,894	9,395,383	51.61	7,349,179	78.22	303,773	1453
WT3	HF	25,596,552	14,678,726	57.35	12,343,681	84.09	349,102	1466
WT4	HF	26,696,428	15,035,608	56.32	11,779,646	78.34	361,607	1468
WT5	LF	26,214,964	14,929,947	56.95	12,324,432	82.55	333,978	1480
WT6	LF	25,606,683	15,475,773	60.44	14,246,619	92.06	210,470	1407
WT7	LF	31,198,548	17,212,742	55.17	13,361,302	77.62	359,230	1471
WT8	LF	32,317,247	18,081,900	55.95	14,541,413	80.42	380,172	1474

HF high fat, LF low fat

^aDefined as distinct transcripts identified in each sample

^bDefined as distinct enzyme classification (EC) numbers

Applying our previously developed metratranscriptome analysis pipeline [35], reads were filtered and subsequently assembled into contigs (see “Methods”). Contigs and unassembled reads were then passed through a tiered set of sequence similarity searches using BWA, BLAT, and DIAMOND [36–38] against databases of microbial genomes and bacterial non-redundant proteins, resulting in taxonomic and functional annotations for each contig/read (see Additional file 1: Table S1 for detailed statistics of assembly and annotation). A total of 200,007 distinct bacterial transcripts were identified across all samples (Additional file 2: Table S2). With the exception of Fisher’s Alpha between Plin2-HF and Plin2-LF samples, no significant difference was observed in measures of alpha or beta diversity across the four sample types (Table 2). To investigate changes in specific taxonomic groups, initial assignments were placed into one of 17 pre-defined taxa that encompass most mammalian gut bacterial diversity (Fig. 1a). Consistent with our previous study [32], reads associated with Firmicutes and, to a lesser extent, Bacteroidetes dominated each sample, with Lachnospiraceae (of the phylum Firmicutes) being the most abundant family. As before, we also found that the Plin2 mice which shifted to a LF diet (Plin2-LF) exhibited a lower incidence of reads assigned to Firmicutes relative to other samples. Principal components analysis (PCA) revealed no significant separation between four pairwise comparisons (Plin2-HF vs. Plin2-LF; Plin2-HF vs. WT-HF; Plin2-LF vs. WT-LF; and WT-HF vs. WT-LF—non-parametric multiple analysis of variance (PERMANOVA); Additional file 3: Figure S1A). In addition, PERMANOVA tests revealed no significant differences in taxonomic abundance when we compare the two genotypes as a function of diet, or conversely when we compare the two diets as a function of genotype. Furthermore, when we focused on specific taxonomic groups, PERMANOVA tests identified significant differences in read abundance only when comparing HF and LF diets independent of genotype (i.e., combining WT and Plin2 samples); specifically, of the 17 pre-defined taxa, Clostridiaceae, Eubacteriaceae, Oscillospiraceae, and Ruminococcaceae display significantly elevated levels of gene expression under a HF diet (Fig. 1).

Table 2.

Biodiversity analysis across four mice samples. Presented are five indices of alpha diversity and one for beta diversity

Sample	Species richness	Chao 1	F-alphaa	Shannon	Simpson	Beta diversityb
Plin2-HF	1517 ± 2.9	1561 ± 1.3	134.7 ± 2.66	4.56 ± 0.04	0.97 ± 0.002	0.96 ± 0.002
WT-HF	1513 ± 9.8	1557 ± 0.35	136.0 ± 2.62	4.55 ± 0.08	0.97 ± 0.004	0.96 ± 0.006
Plin2-LF	1492 ± 20.2	1563 ± 4.18	131.3 ± 0.58	4.61 ± 0.16	0.98 ± 0.004	0.94 ± 0.013
WT-LF	1512 ± 26.4	1570 ± 0.25	131.7 ± 3.23	4.56 ± 0.36	0.97 ± 0.023	0.96 ± 0.017

Values given are means with standard deviations

^aApplying the two sample t test for the four types of sample comparisons, only the comparison of Fisher’s Alpha between Plin2-HF and Plin2-LF samples revealed a statistically significant difference in diversity (p < 0.05)

^bBeta diversity was calculated as the number of species in a sample/total_number of species across all 16 samples

Fig. 1.

Abundance distribution of microbiome across four mice samples. a Abundance distribution of 17 predefined bacterial taxa for all four mouse genotype/diet groups (Plin2-HF, WT-HF, Plin2-LF, and WT-LF). Blue asterisks indicate the significant shifts in taxon abundance in comparisons of both Plin2 and WT mice fed different diets (p value < 0.05). b–d Venn diagrams of number of mapped microbial transcripts for four samples, Plin2-HF (red), WT-HF (green), Plin2-LF (blue), and WT-LF (yellow) with different cutoffs: b RPKM > 0, c RPKM ≥ 10, and d RPKM ≥ 100

In summary, these broad taxonomic analyses indicate that across the four sample types (diet x genotype), only the dietary regime, and not the Plin2 genotype, had a significant impact on microbiome composition at the phylum/family level. These results corroborate our previous 16S rRNA-based analyses in which Plin2 genotype had minimal impact on microbiota composition in animals fed either HF or LF diets.

Diet and Plin2 deletion alter microbiome gene expression

We next examined if genotype influenced the functional distribution of reads (Fig. 1b–d). Of the 200,007 unique microbial transcripts identified across all samples, a core set of 156,289 transcripts (78.1%) were identified in all four diet/genotype groups (i.e., Plin2-HF, Plin2-LF, WT-HF, WT-LF). Of the remaining 43,718 transcripts absent in at least one of the four samples, only a small proportion of transcripts were uniquely associated with either Plin2 genotype or diet (Fig. 1b). Interestingly, samples from the same genotype (i.e., Plin2-HF and Plin2-LF; WT-HF and WT-LF) shared a greater proportion of expressed transcripts (1.7 and 2.7%, respectively) than samples sharing diets (0.8% for Plin2-HF and WT-HF, and 0.9% for Plin2-LF and WT-LF), suggesting that Plin2 genotype was a factor in modifying transcript profiles between treatment groups. When only transcripts with at least moderate or high levels of expression (defined as ≥ 10 and ≥ 100 reads per kilobase of transcript per million mapped reads (RPKM), respectively) in at least one sample type were considered, the proportion of shared, core transcripts dropped to 16.4 and 10.3%, respectively, with many transcripts displaying elevated expression only in a single sample type (Fig. 1c, d). For example, across the 6121 highly expressed transcripts (≥ 100 RPKM), 16.6% were specific to Plin2-HF samples, 13.4% were specific to WT-HF samples, 16.9% were specific to Plin2-LF samples, and 11% were specific to WT-LF samples. Given that their high expression is limited to specific samples, such transcripts might reflect key responses to changes in environmental conditions imposed by different diets and/or genotypes.

To further examine the impact of Plin2 genotype and diet on gene expression, we used DESeq [39] to identify 3777 bacterial transcripts that exhibited significant differences in expression across the four sample types (Additional file 4: Table S3). PCA based on the expression of these transcripts revealed a significant separation between Plin2 and WT mice fed a HF diet (q value < 0.05, PERMANOVA). Significant differences were also observed in HF vs. LF dietary regimes fed to Plin2 mice (q value < 0.05, PERMANOVA), indicating altered community transcriptional activity (Additional file 3: Figure S1B). Hence, through metatranscriptomics, PCA revealed that although the taxonomic distribution of the gut microbiota was unaffected by host Plin2 genotype, the microbiota of Plin2 and WT mice exhibited distinct gene expression profiles when fed a HF diet.

We next explored whether the 3777 differentially expressed transcripts were significantly enriched in any of the 17 pre-defined, gut-dominant bacterial taxa. Of the 200,007 unique transcripts identified in our analysis, the three largest represented taxa were Lachnospiraceae, Clostridiaceae, and “Other Firmicutes” (Fig. 2a, top panel and Additional file 5: Table S4). Transcripts exhibiting significant differential expression were identified in 6 of the 17 pre-defined taxa, including both abundant (e.g., Lachnospiraceae) and rarer taxonomic groups (Fig. 2a). For example, the increased abundance of Lachnospiraceae reads and enrichment in differentially expressed transcripts in three of four comparisons (i.e., Plin2 vs. WT on HF diet, Plin2 vs. WT on LF diet, and Plin2 on HF vs. LF diet; Geno-HF, Geno-LF, and Diet-Plin2 panels, respectively, in Fig. 2a) is consistent with our previous 16S-based study, which suggested an increased abundance of this taxon, albeit without statistical significance, under a HF diet for both Plin2 and WT mice [32]. Notably, we did not identify differential expression in this taxon between WT mice fed LF and HF diets, suggesting that loss of Plin2 function alters the expression of Lachnospiraceae genes, irrespective of dietary factors (Diet-WT panel, Fig. 2a). Furthermore, Deltaproteobacteria was enriched in differentially expressed transcripts in comparisons between Plin2 and WT animals raised on a HF diet, in addition to dietary (HF vs. LF) comparisons for both Plin2 and WT mice. Porphyromonadaceae was also enriched in differentially expressed transcripts in comparisons between Plin2 and WT mice fed HF diets, in addition to comparisons of the two genotypes fed a LF diet and the two diets fed to WT mice.

Fig. 2.

Microbial transcripts expressed differentially across taxa for different sample-wise comparisons. a Bar charts, from top to bottom, show the total number of mapped microbial transcripts and the enrichment of significant differentially expressed transcripts (significant transcripts) identified from each of four pairwise comparisons (see inset key) across 17 predefined taxa. Detailed q values are provided in Additional file 5: Table S4. b Venn diagram of number of significant transcripts for all four pairwise comparisons. c Boxplots indicating distribution of expression fold change (eFC) of significant transcripts from each pairwise comparison. Only the eight taxonomic groups enriched in significant transcripts are shown

Contrasting significantly differentially expressed transcripts across the four sample-wise genotype/diet comparisons (Fig. 2b) revealed that most significant differentially expressed transcripts were unique to each group. The WT vs. Plin2 comparison under a LF diet exhibited the fewest significant transcripts (612 total, 394 specific to this comparison), whereas the WT vs. Plin2 comparison under a HF diet exhibited the most (1447 total, 1163 specific to this comparison). Three transcripts were identified in all four comparisons (i.e., Plin2 vs. WT on HF vs. LF diet): a SSU ribosomal protein S3P (EC1_07050) putatively expressed by Eubacterium cylindroides T2–87; an environmental response regulator, CLS_10500, putatively expressed by Clostridium cf. saccharolyticum, a member of the Lachnospiraceae; and a flagellin (WP_031391632.1) putatively expressed by Clostridium sp. KNHs209. CLS_10500 encodes a CheY-like receiver domain and a winged-helix DNA-binding domain. Together, these two elements form part of a modular OmpR-like two-component signaling system [40] which has been shown in Escherichia coli to have roles in osmoregulation [41], chemotaxis [42], and the sensing and transport of nutrients, including carbohydrates [43] and fatty acids [44]. Under a LF diet, CLS_10500 exhibited almost 90-fold higher expression in Plin2 mice relative to WT (95.6 ± 112.6 vs. 1.10 ± 1.32 RPKM for Plin2 vs. WT; Additional file 4: Table S3), but was diminished over 200-fold (1.49 ± 1.96 vs. 345.2 ± 685.7 RPKM for Plin2 vs. WT) under a HF-diet. The WP_031391632.1 transcript is also of interest since in some organisms, and flagellin is regulated by the OmpR system [45]. Here, we found that WP_031391632.1 was again significantly upregulated in Plin2-null mice relative to WT under a LF diet, but was significantly downregulated in Plin2-null mice under a HF diet (RPKM = 124.9, 1.7, 0.4, and 47.9 for Plin2-LF, Plin2-HF, WT-LF, and WT-HF, respectively). These findings illustrate the interlinked effects of Plin2 and diet on bacterial transcript expression. Of note, previous studies have reported that flagellins secreted by motile bacteria are important determinants of host-microbiome interactions and gut homeostasis [46–48]. Further analyses of the 417 transcripts representing flagellin from different taxa found 37 transcripts that exhibited significant differential expression, although no clear pattern of direction of change in expression was observed (Additional file 4: Table S3).

Because of the association of specific taxa with specific differentially expressed transcripts, we were interested in examining the magnitude and direction of change of all transcripts, broken down by a taxonomic group (Fig. 2c). Focusing on the six taxa we previously identified as being enriched in these transcripts, for the dietary comparisons (i.e., Plin2-HF vs. Plin2-LF and WT-HF vs. WT-LF), only modest changes in expression were observed, with most taxa demonstrating an overall decrease in expression under the HF diet (Fig. 2c, top panels). For the genotype comparisons, we found that three of the six taxa (Rikenellaceae, other Bacteroidales, and Deltaproteobacteria) exhibited a general increase in transcript expression in Plin2-null mice relative to WT under a HF diet. Furthermore, under a LF diet, differentially expressed transcripts from all six taxa were upregulated in Plin2 mice relative to WT. Together, these findings suggest that, unlike taxonomic composition, Plin2 genotype and not diet, had a greater impact on the magnitude of change in transcript expression.

Among the transcripts that were significantly differentially expressed in the HF comparison (Plin2-HF vs. WT-HF), 44 displayed fold-changes greater than 300; most of them derived from Lachnospiraceae, Bacteroidaceae, and “other Firmicutes” (Additional file 2: Table S2). These transcripts included CK5_14390, a flavoprotein from Ruminococcus obeum, which was highly expressed only in the Plin2-HF samples (RPKM = 2412 ± 2340; compared to < 2 for the other three sample types). Flavoproteins are involved in butyrate synthesis pathways, a key metabolite for maintaining gut homeostasis and epithelial integrity [49]. Conversely, RHOM_00695, a UDP-4-dehydro-6-deoxy-2-acetamido-d-glucose 4-reductase from Roseburia hominis, displayed considerably reduced expression only in the Plin2-HF samples (RPKM = 29.7 ± 35.5; compared to > 2000 for the other three sample types). RHOM_00695 belongs to the short-chain dehydrogenase/reductase family and is essential for Helicobacter pylori pathogenesis, with roles in the biosynthesis of flagella and lipopolysaccharide [50–52].

Plin2 genotype and dietary fat/carbohydrate balance modulate microbial functional pathways operating in the gut

To further explore the functional consequences of changes in microbial community gene expression profiles, differentially expressed transcripts were mapped to metabolic enzymes and subsequently grouped into pathways as defined by the Kyoto Encyclopedia of Genes and Genomes (KEGG; [53]). Gene set enrichment analysis identified 42 of 144 pathways that were enriched in at least one of the four pairwise comparisons of genotype/diet groups (Fig. 3 and Additional file 6: Table S5). Twenty-three of these pathways were associated with the production and/or degradation of amino acids, energy, carbohydrates, or nucleotides (Fig. 3a). Although we note considerable overlap in enriched pathways in the pairwise comparisons of the genotype/diet groups, the comparison of Plin2-null vs. WT mice fed a LF diet exhibited the fewest number of enriched pathways (20 pathways). Consistent with the taxonomic analysis of differentially expressed transcripts (Fig. 2a), the greatest abundances of differentially expressed enzymes were observed in the comparisons of (1) Plin2-null vs. WT mice fed a HF diet (153 enzymes in 27 pathways) and (2) Plin2-null mice fed HF vs. LF diets (164 enzymes in 26 pathways).

Fig. 3.

Distribution of enzymes in metabolic pathways for different sample-wise comparisons. For each of the 36 KEGG metabolic pathways listed, the number (a) and average expression fold change (b) of significantly differentially expressed enzymes (significant enzymes) are indicated for each of the four pairwise comparisons. Detailed q values are provided in Additional file 6: Table S5. Significant enzymes are defined as enzymes to which at least one significantly differentially expressed transcript is mapped (see “Methods”). Numbers in brackets after pathway name indicate the total number of enzymes associated with each pathway. Twenty-one of the 36 pathways belong to one of the four KEGG-defined super families: (1) amino acid metabolism, (2) energy metabolism, (3) carbohydrate metabolism, and (4) nucleotide metabolism. The ratio of expressed enzymes indicates the proportion of enzymes associated with that pathway that was expressed in at least one of the samples. Asterisks indicate pathways that were statistically significantly enriched in significant enzymes (p value < 0.05) for each genotype/diet comparison, e.g., Geno-HF (red), Geno-LF (blue), Diet-Plin2 (green), and Diet-WT (purple)

Analyses of expression fold-change revealed that WT mice fed LF and HF diets exhibited the least change in microbial pathway expression, with differentially expressed genes in 25 of 42 pathways exhibiting < 10 fold-change (Fig. 3a). In comparison, microbiota of Plin2 mice exhibited a global upregulation of most enriched pathways under a HF diet with 24 of 42 pathways exhibiting > 10-fold change, suggesting that the microbiota of Plin2-null mice either were more sensitive than WT to increased dietary fat or that Plin2 deletion altered the physiological and/or nutritive landscape of the gut. This is further shown in the comparison of Plin2 and WT mice fed a HF diet (Fig. 3b), in which differentially expressed enzymes tended to be upregulated in Plin2-null relative to WT animals. Perhaps reflecting the relatively low number of differentially expressed enzymes, comparisons between Plin2-null and WT mice fed a low-fat/high-carbohydrate diet exhibited more extreme distributions of fold change of expression.

PCA analysis of differentially expressed enzymes revealed significant differences in clustering between the Plin2-null and WT mice fed a HF diet, as well as WT mice fed HF and LF diets (q value < 0.05 for both; Additional file 3: Figure S1D). However, only the former exhibited significant differences in PCA clusters at the level of the pathway (Additional file 3: Figure S1E). Together, these results reveal that differential expression of microbial transcripts between Plin2 and WT mice extends to changes in the expression of multiple enzymes having the potential to alter metabolic activity in the microbiome.

A high-fat diet in the context of Plin2 knockout results in coordinated patterns of expression among consecutive enzymes involved in energy metabolism

In the previous section, we identified several metabolic pathways exhibiting differential expression across genotype/diet groups. Of particular interest were comparisons between Plin2-null and WT mice fed a HF diet, given the role of Plin2 in lipid homeostasis. Although both groups exhibited similar taxonomic distributions, 33 KEGG-defined pathways were enriched in differentially expressed genes in either WT or Plin2 groups on HF diet (Fig. 3a). Here, we further explore changes in expression in several pathways involved in energy metabolism and the production of co-factors, amino acids, and butyrate (Fig. 4 and Additional file 7: Figure S2, Additional file 8: Figure S3, Additional file 9: Figure S4, Additional file 10: Figure S5, Additional file 11: Figure S6, Additional file 12: Figure S7).

Fig. 4.

Comparison of enzyme expression between Plin2-HF and WT-HF mice. Two pathways are shown: the TCA cycle (a) and glycolysis and gluconeogenesis metabolic pathway (b). For each pathway, circular nodes indicate enzymes, with size indicating a relative difference in expression between sample types (Plin2-HF vs. WT-HF) and color indicating the direction of change (see inset key). Associated heatmaps indicate global changes in expression for each enzyme, in addition to taxon-specific changes in expression for each of the 17 defined taxa colored according to phylum. Key metabolites mentioned in the text are indicated with bold blue text. The following abbreviations are used: S-adhlam-E (S-acetyldihydrolipoamide-E), 2H–ThPP (2-hydroxyethyl-ThPP), dhlam-E (dihydrolipoamide-E), S-sdhlam-E (S-succinyldihydrolipoamide-E), and 3C-1H-ThPP (3-carboxy-1-hydroxypropyl-ThPP)

Focusing on energy metabolism, most of the enzymes involved in the TCA cycle (i.e., including those not identified through DESeq as differentially expressed transcripts) were upregulated in Plin2 mice (Fig. 4a and Additional file 13: Table S6). All of these enzymes, except isocitrate dehydrogenase (EC:1.1.1.41), catalyze consecutive reactions linking the glycolytic metabolites, pyruvate and phosphoenolpyruvate (PEP), to key intermediates used in the synthesis of fatty acids and amino acids, as well as the utilization of reducing equivalents [54]. Such “coordinated” patterns of expression may result in potential metabolic channeling that connect both acetyl-CoA, a key intermediate in fatty acid metabolism (through pyruvate), and PEP to oxaloacetate and subsequently to 2-oxoglutarate (α-ketoglutarate) and the synthesis of amino acids, through NADH-dependent fumarate reduction, a part of the bacterial redox system [54].

Analysis of the microbial taxa in the context of this pathway revealed several taxa with relatively consistent upregulation of the enzymes that may drive channeling of PEP and/or Acetyl CoA, with at least 6 of the 12 enzymes involved displaying some level of upregulation (Fig. 4a inset heatmap). These include both taxa displaying an overall upregulation of TCA cycle enzymes (Clostridiaceae, Actinobacteria, “other Proteobacteria,” and “other Clostrida”) and those displaying an overall downregulation of TCA cycle enzymes (Deltaproteobacteria and “other Firmicutes”). In contrast, the three enzymes: citric synthase (EC:2.3.3.1); aconitase (EC:4.2.1.3), and NADP-dependent isocitrate dehydrogenase EC:1.1.1.42), linking acetyl-CoA to 2-oxoglutarate through isocitrate were downregulated in Plin2 compared with WT. Such downregulation was possibly driven by changes in the expression of these three enzymes in Lachnospiraceae, Deltaproteobacteria, and “other Bacteroidiales.” Thus, although similar taxonomic profiles were found in WT and Plin2 mice, notably different taxa exhibited divergent responses in TCA cycle regulation, which presumably reflect inter-taxa metabolic optimizations related to microbial and host responses to HF diet. For example, Bacteroidaceae downregulated the expression of six enzymes and upregulated the expression of five enzymes in the TCA cycle in Plin2 compared with WT mice.

Among the other three pairwise comparisons of diet/genotype groups (i.e., Geno-LF, Diet-Plin2, and Diet-WT), coordinated patterns of differential expression across consecutive enzymes were also observed in both the Geno-LF comparison and the Diet-WT comparison (Additional file 7: Figure S2). In the former, enzymes linking succinate to isocitrate and acetyl-CoA via oxaloacetate are upregulated in Plin2 mice, while in the latter enzymes linking citrate to fumarate via 2-oxaglutarate are consistently downregulated in WT mice fed a HF diet.

Interestingly, analysis of the glycolysis pathway also revealed consistent differential regulation of enzymes performing consecutive reactions (Fig. 4b). For example, enzymes involved in the production of PEP from β-d-fructose-6-phosphate were largely downregulated in Plin2 compared with WT animals on HF diet, whereas those acting directly on various forms of glucose (e.g., phosphoglucomutase—EC:5.4.2.2 and glucose-1-phosphatase—EC:3.1.3.10) were upregulated. As with the TCA cycle, specific taxa yielded distinct patterns of regulation across glycolysis, with 8 of the 17 taxa displaying an overall regulation of this pathway and the other 9 displaying an overall downregulation of this pathway. Focusing on the ten enzymes linking β-d-fructose-6-phosphate to PEP, we observe the greatest contributions to the coordinated pattern of downregulation from Lachnospiracea (8 of 10 enzymes downregulated), “other Firmicutes” (7 enzymes), “other Bacteroidales” (7 enzymes), Bacteroidaceae (6 enzymes), and “other bacteria” (6 enzymes). Other taxa display more heterogeneous patterns of expression; for example, Clostridiaceae, Eubacteriaceae, and Bacilli respectively feature four, five, and four downregulated enzymes and six, three, and five upregulated enzymes.

For the other diet-genotype comparisons, although other parts of the pathway displayed heterogeneous responses to diet and Plin2 genotype, enzymes involved in the production of pyruvate from β-fructose-6-phosphate were largely downregulated under a HF diet in both WT and Plin2-null animals (Additional file 8: Figure S3).

Next, with important roles in the TCA cycle and fatty acid metabolism, we examined the pantothenate and CoA biosynthetic pathway (Additional file 9: Figure S4 and Additional file 10: Figure S5). As before, we note consistent patterns of regulation involving consecutive reactions. For example, enzymes linking valine and uracil were consistently upregulated in Plin2-null mice compared to WT mice fed a HF diet, while those linking pantothenate to CoA were largely downregulated (i.e., upregulated in WT mice). Interestingly, this latter pattern was also observed in WT mice fed a HF diet compared to those fed a LF diet. This again highlights the coordinated regulation of consecutive enzymes in these pathways with the potential to redirect metabolic flux between key metabolites. Furthermore, taxon-specific patterns of expression were observed; enzymes in the pathway associated with Rikenellaceae, Deltaproteobacteria, Clostridiaceae, Eubacteriaceae, and Baciilli were largely upregulated in the Plin2-mice compared to WT mice fed a HF diet while other groups displayed more mixed patterns of expression.

Finally, focusing on butyrate metabolism (Additional file 11: Figure S6 and Additional file 12: Figure S7), we again identify patterns of elevated expression of consecutive enzymes in Plin2 mice relative to WT mice fed either a high- or low-fat diet that may help channel metabolites. In particular, we note the upregulation of many enzymes linking acetyl-CoA with butyrate for both dietary comparisons and fumarate to acetyl CoA for the HF diet comparison. Interestingly, we note little correlation in taxonomic responses to the two dietary comparisons. For example for Plin2-null mice, five taxa (Lachnospiraceae, Ruminococcaceae, “other Clostridia,” Bacteroidaceae, and Porphyromonadaceae) exhibit an overall downregulation of the pathway under a HF diet relative to WT but an overall upregulation of the pathway under a LF diet.

Furthermore, we note that two enzymes that help link acetyl-CoA with butyrate, butyryl dehydrogenase (EC 1.3.8.1), and 3-hydroxybutyryl-CoA dehydratase (EC:4.2.1.55) are down regulated in the HF comparison but not the LF comparison. While these expression differences suggest that Plin2 mice may produce more butyrate under a low-fat/high-carbohydrate diet, the relative impact for Plin2 and WT mice fed a HF diet are less clear. Further insights may be gained through the application of methods such as constraints-based modeling to examine changes in pathway flux as a consequence of enzyme expression [55].

Discussion

Previous studies in humans and animal models have documented that diet plays a major role in driving the taxonomic structure of the intestinal microbiota and that the composition of the microbiota can have a significant impact on host health. For example, through the production of short-chain fatty acids, previous studies have shown that species of Bacteroides, Alistipes, and Parabacteroides can influence host body mass [56]. However, although different taxa within a microbiome can impart unique functionality, metagenomic studies have also revealed the capacity of diverse microbiomes to encode similar functional potential [57]. In this study, we applied metatranscriptomic sequence analysis to examine the potential of host genotype and diet to influence microbial community structure and function. Consistent with previous studies, we found that changes in dietary fat and/or carbohydrates had a dramatic impact on community structure. Crucially, however, we also demonstrated that deletion of a mouse lipid storage gene, Plin2, can significantly alter microbial gene expression, despite having minimal impact on community structure (at least in the short-term). Furthermore, although the intestinal microbiota of WT mice exhibited relatively robust and rapid transcriptional responses to changes in dietary fat/carbohydrates, the microbiota of Plin2 mice displayed even more dramatic responses, as indicated by the greater number and increased magnitude of differentially expressed pathways in this group (Fig. 3). Collectively, our results provide evidence of host gene function and diet interacting to regulate gut microbial metabolism and suggest a specific role of Plin2 host functions in integrating metabolic responses of specific bacterial taxa to HF diets.

Plin2 is a member of the perilipin family that organizes the formation of cytoplasmic lipid droplets and contributes to the coordination of cellular and tissue lipid storage and metabolism [20]. In previous studies, we demonstrated that deletion of Plin2 abrogates long-term deleterious effects of a high-fat diet, at least in part through limiting intestinal lipid uptake [32]. We now report that Plin2 deletion results in altered patterns of expression of microbial enzymes responsible for directing metabolic flux through key pathways that drive the production of energy and components required for cell growth. A global consequence of these responses appears to be a general increase in pathways linked with the production of metabolic intermediates contributing to fatty acid and amino acid synthesis, in addition to other metabolites such as uracil. Importantly, data demonstrating bacterial taxon-dependent effects of Plin2 on bacterial metabolic pathways suggest that Plin2-dependent modulation of host lipid metabolism is a determinant of the syntrophic metabolic optimization of gut bacteria in response to dietary substrate alteration.

The precise mechanism by which Plin2, or other host genes, influence gut microbial metabolism remains to be determined and may be complex. Increased quantities of fecal trigylcerides (TG) are observed in both HF-fed (relative to LF- or chow-fed) and Plin2-null (relative to WT) animals [32]. It is possible that increased availability of lipids in the intestinal lumen provides the microbial community with an excess source of energy in relation to other essential metabolites. By altering enzyme expression, the community may respond by directing metabolic flux away from the production of energy and instead into the anabolic production of biomass constituents, such as fatty acids, amino acids, and nucleotides, which are growth-limiting under conditions of lipid/energy excess. For example, through reductive carboxylation, the TCA cycle may operate in reverse, consuming energy in the form of ATP, GTP, and NADH to produce intermediates for the production of metabolically important fatty acids, such as butyrate, and amino acids, such as glutamate. Additionally, the luminal lipid concentration can influence gut osmolarity [58] and secretion of bile acids [59], both of which regulate environmental signaling systems, such as OmpR, which is involved in nutrient uptake and metabolism, as well as motility and invasion [44, 60]. Our findings that bile acid-sensitive flagellin genes [60] are differentially expressed in some bacteria of HF-fed Plin2 mice may reflect the effects of Plin2 on bile acid production. Similarly, decreased expression of the osmotically sensitive OmpR system in HF-fed Plin2 mice is consistent with suppression of this system under low osmotic conditions [58].

The differences in expression profiles of HF-fed Plin2 mice compared with HF-fed WT mice could have arisen simply through elevated TG reaching the colons of Plin2 animals. However, this conclusion is difficult to reconcile with our finding that the genes and pathways modulated by HF vs. LF diet differed between Plin2 and WT groups (Fig. 3, noting the contrast of Diet-Plin2 with Diet-WT), as well as between HF-fed Plin2 and WT groups (Fig. 3, Geno-HF comparison). In other words, relative to WT animals, Plin2 animals did not merely exhibit an exacerbated response of similar metabolic pathways to HF feeding, distinct patterns of up- and downregulated pathways were observed between the two groups. Thus, we hypothesize that Plin2 deletion alters the intestinal environment, and the microbiota’s transcriptional response to this environment, beyond increasing fecal TG levels; we are currently exploring other Plin2-dependent aspects of lipid metabolism and hepato-intestinal cross-talk, such as bile acid pool size and composition that might affect the gut microbiome in this system.

Interestingly, over the timescale of this experiment (4 day exposure to LF or HF), changes in pathway expression apparently did not selectively benefit the growth of any single taxon because no significant difference in intestinal community composition was observed between the WT or Plin2 mice. In longer experiments involving weeks of dietary exposure, we have noted compositional change in 16S rRNA datasets between WT and Plin2 mice (Frank and McManaman, unpublished). Thus, this study indicates that applying methodologies that more directly interrogate microbiome function, such as metatranscriptomic profiling, can provide a more sensitive means of detecting early or subtle changes in microbiome activity than can marker gene inventories. We note that tools such as PICRUSt [33] that seek to infer microbiota function using 16S rRNA gene abundance data may fail to recognize the potential of communities with similar taxonomic structures to exhibit significant shifts in gene expression and hence functional capacity.

Key outstanding questions include the mechanism by which changes in lipid uptake and/or metabolism by the host influence microbial gene expression patterns. By using metabolomics and stable isotope labeling, we expect that future studies will reveal how changes in enzyme expression affect the flux of metabolites through diverse metabolic pathways and thereby impact microbial growth and community function. Of particular interest would be experiments that allow the delineation of metabolites produced and consumed by the microbiome from those associated with the host. Furthermore, because the two dietary regimes examined in this study differed in both fat and carbohydrate content, additional work is needed to delineate the relative contributions of these two energy sources to the phenotypes reported in this study. Finally, longitudinal follow-up studies are needed to determine the temporal relationships between diet-induced changes in microbiome function and development of metabolic disease. Conducting these studies across developmental stages (i.e., infant, juvenile, adult, and aged mice) would also be of interest to better understand inter-generational transmission of obesity and metabolic disease risk.

Conclusions

Obesity and its co-morbidities, such as fatty liver disease, are major global health concerns. The composition and function of the intestinal microbiome is recognized as a critical determinant of metabolic disease risk. Here, we show that under a high-fat/low-carbohydrate dietary regime, despite sharing similar microbiome community structures, the knockout of the Plin2 gene in mice can give rise to dramatic differences in microbial gene expression profiles. Many of these changes were associated with the coordinated expression of suites of enzymes mediating consecutive reactions within a pathway, directing metabolic flux towards the production of important “biomass” constituents (i.e., amino acids, nucleotides, and other molecules required for growth). To our knowledge, this is the first demonstration of how host genotype can impact microbiome function without altering community composition, emphasizing the need to prioritize metatranscriptomic studies over 16S rRNA surveys to gain more meaningful insights into microbiome function.

Methods

Animal procedures

Eight-week-old male C57BL/6 (WT) and Plin2-null (Plin2) mice on the C57BL/6 background were used for all studies. The generation and characterization of Plin2-null mice have been described in detail previously [61]. The WT mice used in this study were obtained from a breeding colony maintained at the University of Colorado School of Medicine’s Center for Comparative Medicine and housed in the same room as Plin2-null mice. All mice were fed standard mice chow (2020X, Harlan Laboratories, 16% fat calories, 24% protein calories and 60% carbohydrate calories) ad libitum from weaning to 8 weeks of age, at which time they were housed individually in a metabolic monitoring system at 30 °C, the thermoneutral temperature of mice [62], for measurements of energy balance (intake and expenditure), the respiratory exchange ratio (RER), and activity levels (Columbus 8 M Oxymax) [63]. Following a 3-day adjustment period, the mice were fed nutritionally balanced high-fat (HFD; 60% fat calories, 20% protein calories, 20% carbohydrate calories, D12492) or low-fat (LFD; 10% fat calories, 20% protein calories, 70% carbohydrate calories, D12450B) diets from Research Diets Inc. (New Brunswick, NJ) ad libitum for 4 days.

Metatranscriptomic analysis

The RiboPure-bacteria kit (Ambion, Austin, TX, USA) was used to isolate total RNA from colon contents according to the manufacturer’s protocol. Briefly, samples were collected and suspended in RNAwiz (provided in a kit) and bead beaten with zirconia beads using the Roche MagnaLyser (Roche Applied Science, Indianapolis, Indiana). Nucleic acids were recovered from the lysate by adding chloroform, centrifugation, and removal of the aqueous layer. Following ethanol precipitation, the sample were bound to a spin filter, washed, and then eluted with 50 μl of the provided elution solution. The eluted RNA was then treated with DNAse I (provided in the kit) for removal of any contaminating DNA from RNA. The final product RNA was stored in −80 °C and shipped on dry ice to the Donnelly Sequencing Centre of the University of Toronto, Canada. Ribosomal RNA depletion was performed using the Ribo-Zero™ Epidemiology kit (Epicentre Inc, Madison, WI, USA). Sequencing was performed on an Illumina NextSeq500 platform using a single high-output cartridge to generate ~ 419 million 1 × 150 bp reads. With ~ 20–30 million reads generated per sample (Table 1), rarefaction analysis revealed sufficient depth of coverage to identify the vast majority of species and enzymes present in the samples (Additional file 14: Figure S8). Sequence data is available at the NCBI Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra/) with the BioProject identifier: PRJNA379425.

Sequence reads were processed as described previously [34, 35]. In brief, we processed sequence reads by removing adaptor-contaminated and low-quality reads using Cross_Match v0.990319 (www.phrap.org) and USEARCH v7.0.1001 [64], respectively. Next, rRNA and tRNA reads were filtered using Infernal v1.1.1 [65] and host reads identified through BWA v0.7.5a [66] and BLAT v35 [67] sequence similarity searches against a database of mouse transcripts (ENSEMBL release 78 [68]). Putative mRNA reads were then assembled using the Trinity v2.1.1 de novo assembler [69], and resultant contigs, as well as unassembled reads, were assigned to microbial transcripts using a tiered set of sequence similarity searches against a database of sequenced microbial genomes (downloaded from NCBI June 2015) using BWA and BLAT, as well as the protein non-redundant database (downloaded from NCBI July 2015) using DIAMOND v0.7.5a [38]. The expression level of a microbial transcript is then based on the number of sequence reads mapped to that transcript. Consistent with previous studies [70], we first filtered for transcripts expressed at low levels, defined here as those with < 5 mapped reads (representing ~ 6.5% of all mapped reads), and then normalized expression levels of transcripts as reads per kilobase per million mapped reads (RPKM; [71]). Where no expression was observed, RPKM values were set to 0.

Non-parametric permutational multivariate analysis

For each dietary/genotype comparison, in addition to examining differences in the overall distribution of microbiome composition, we also investigated changes in the abundance of each taxon, defined as the sum of RPKM values of transcripts assigned to that taxon. For the former, the influence of only a single factor was investigated in the analysis, i.e., genotype (Plin2 vs. WT) or diet (HF vs. LF). For each taxon, we estimated the influence of two independent factors, genotype and diet, on one dependent variable, i.e., relative expression as defined by RPKM values. Comparisons were performed using the non-parametric permutational multivariate analysis of variance test (PERMANOVA; [72]) to assess the difference in microbiome composition between different genotypes or diets. PERMANOVA was implemented through the f_npManov function of MATLAB (R2015a, The MathWorks Inc., Natick, MA, USA) toolbox Fathom [73], using 100,000 replicate label permutations and adjusting p values with the Benjamini-Hochberg procedure [74]. The cutoff of the adjusted p value was set as 0.05.

Principle component analysis

To reveal the correlation of the overall expression distributions relating to taxa, transcripts, differentially expressed transcripts, enzymes, and pathways across the 16 samples, we applied principal component analysis (PCA) using the pca function from MATLAB (R2015a, The MathWorks Inc., Natick, MA, USA). In addition, we used PERMANOVA to test how well the first and the second principle components separate sample types (e.g., Plin2-HF vs. WT-HF).

Biodiversity analysis

We used four biodiversity indices, i.e., Shannon entropy index (Shannon), Fisher’s alpha index (F-alpha), Chao1 index (Chao1), and Simpson index (Simpson) to examine biodiversity distributions of our data. Chao1 values were calculated using EstimateS v 9.1.0 [75] with 100 bootstrap replicates. Other indices were calculated using the vegan package v2.4.3 [76] in R v3.4.0 [77].

Gene set enrichment analysis

To test if taxonomic categories and KEGG-defined pathways were enriched with either significantly differentially expressed transcripts or enzymes, gene set enrichment analyses were performed using a hypergeometric test with a minimum of two genes per gene set. In these analyses, to ensure consistency across sample comparison, we examined enrichment relative to the total pool of all transcripts identified across all 16 samples. We used a false discovery rate (FDR) adjustment with the Benjamini-Hochberg procedure to correct p values. Hypergeometric tests were performed using the hygecdf and mafdr functions from MATLAB (R2015a, The MathWorks Inc., Natick, MA, USA) with a FDR cutoff of 0.05.

Expression fold change

Given a transcript or enzyme g, its expression fold change (eFC), for a pairwise comparison between sample 1 and 2, was calculated as:

$$\begin{array}{l}{\mathit{eFC}}_{12}\left\{\begin{array}{c}\hfill sign{\left(r_2-r_1\right)}^{\ast }\frac{max\left(r_1,r_2\right)}{min\left(r_1,r_2\right)},min\left(r_1,r_2\right)>0\hfill \\ \hfill r_2-r_1,min\left(r_1,r_2\right)=0\hfill \end{array}\right.\\ \end{array}$$

where r ₁ is the RPKM value of a transcript or enzyme in sample 1 (note several transcripts may be assigned the same enzyme (EC number)), while r ₂ is the RPKM of the transcript or enzyme in sample 2. eFC ₁₂ > 0 indicates the transcript or enzyme g is upregulated in sample 2, while eFC ₁₂ < 0 indicates g is downregulated in sample 2. The average eFC (aeFC) of an enzyme, for a sample-wise comparison between sample 1 and 2, is calculated as:

$${\mathit{\text{aeFC}}}_{12}={\mathit{eFC}}_{12}/n$$

Where n is the total number of transcripts mapped to this enzyme in that sample. Pathway visualizations of enzyme eFCs were performed using Cytoscape v3.4.0 [78] with pathways downloaded in KGML format from KEGG [79].

Differential expression analysis of transcripts and enzymes

Differential expression analysis of mapped transcripts for different sample-wise comparisons was performed using DESeq [39]. Since our interest is focused on comparisons between different genotypes or diets, only four combinations were explored, i.e., Plin2-HF vs. WT-HF (Geno-HF), Plin2-LF vs. WT-LF (Geno-LF), Plin2-HF vs. Plin2-LF (Diet-Plin2), and WT-HF vs. WT-LF (Diet-WT). For each pairwise comparison, we defined significantly differentially expressed transcripts (sig. transcripts) as those with q values < 0.05 (using the Benjamini-Hochberg procedure to correct p values) and Log₂ (eFC) > 2. Across all comparisons, 3777 unique sig. transcripts were identified. Enzymes are defined as significantly differentially expressed if at least one sig. transcript is mapped to that enzyme.

Authors’ contributions

XX performed sequence processing and data analysis and drafted the manuscript; EB performed the mouse experiments. DI prepared the RNA for sequencing and helped to draft the manuscript. CER participated in the study design and data analysis. JM, DF, and JP designed and supervised the study and helped to draft and edit the manuscript. All authors read and approved the final manuscript.

Ethics approval

All animal protocols adhered to the University of Colorado Denver Institutional Animal Care and Use Committee (IACUC) guidelines.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.