Increases in bioactive lipids accompany early metabolic changes associated with β-cell expansion in response to short-term high-fat diet

Abstract

Pancreatic β-cell expansion is a highly regulated metabolic adaptation to increased somatic demands, including obesity and pregnancy; adult β cells otherwise rarely proliferate. We previously showed that high-fat diet (HFD) feeding induces mouse β-cell proliferation in less than 1 wk in the absence of insulin resistance. Here we metabolically profiled tissues from a short-term HFD β-cell expansion mouse model to identify pathways and metabolite changes associated with β-cell proliferation. Mice fed HFD vs. chow diet (CD) showed a 14.3% increase in body weight after 7 days; β-cell proliferation increased 1.75-fold without insulin resistance. Plasma from 1-wk HFD-fed mice induced β-cell proliferation ex vivo. The plasma, as well as liver, skeletal muscle, and bone, were assessed by LC and GC mass-spectrometry for global metabolite changes. Of the 1,283 metabolites detected, 159 showed significant changes [false discovery rate (FDR) < 0.1]. The majority of changes were in liver and muscle. Pathway enrichment analysis revealed key metabolic changes in steroid synthesis and lipid metabolism, including free fatty acids and other bioactive lipids. Other important enrichments included changes in the citric acid cycle and 1-carbon metabolism pathways implicated in DNA methylation. Although the minority of changes were observed in bone and plasma (<20), increased p-cresol sulfate was increased >4 fold in plasma (the largest increase in all tissues), and pantothenate (vitamin B₅) decreased >2-fold. The results suggest that HFD-mediated β-cell expansion is associated with complex, global metabolite changes. The finding could be a significant insight into Type 2 diabetes pathogenesis and potential novel drug targets.

INTRODUCTION

β-Cell mass expansion in response to macronutrient stimuli, such as glucose (2, 66) or high fat diet (HFD), and/or insulin resistance is observed in rodent models (33, 55, 60) and is also thought to occur in humans; autopsy studies have revealed that nondiabetic obese individuals have 50% greater β-cell mass compared with lean individuals (71). Inadequate β-cell proliferation in the face of insulin resistance can lead to diabetes (47). Pancreata from Type 2 diabetes (T2D) patients have diminished β-cell mass compared with nondiabetic body mass index (BMI)-matched individuals (8), suggesting failed β-cell compensation in the face of insulin resistance. However, lack of reliable methodology to longitudinally measure β-cell mass in living humans makes it difficult to assess their true dynamics. It is possible that some individuals are born with less β-cell mass than others, thus making them more susceptible to metabolic stressors such as HFD later in life.

In adult mice, β-cell mass increases via proliferation in response to several stimuli including pregnancy (5, 64, 70, 76), obesity (8, 10, 65), and HFD (25, 57, 78). Evidence for proliferation in human β-cells in response to pregnancy or altered nutrition in vivo has been equivocal (9, 34) and β cells from human cadaver donor islets ex vivo often show inconsistent responsiveness to the same factors effective in mice (79). Importantly, several recently identified and widely dissimilar stimulators of adult mouse β-cell proliferation induce significant proliferation in adult human β cells ex vivo or when transplanted into mice. These include the serine protease serpin B1 (61), the DYRK1A inhibitor harmine (87), antagonists of the PGE₂ receptor EP3, and agonists of the PGE₂ receptor EP4 (11, 42), soluble factors released from CD4⁺/CD8⁺ T cells (24), and the bone-related decoy receptor osteoprotegerin and its related osteoporosis drug, denosumab (44). Thus identifying the key regulators of β-cell mass expansion in response to HFD-feeding in mice could help us understand regulation of compensatory β-cell proliferation in response to metabolic and nutritional cues, and develop new strategies for augmenting functional β-cell mass in T2D patients.

Whereas long-term HFD consumption induces glucose intolerance, insulin resistance, and enhanced β-cell mass and proliferation in rodents (1, 6, 52, 67, 85), it is clear that β-cell proliferation increases within 1 wk in response to HFD, in the absence of insulin resistance (57, 82). Metabolomic assessment of α-hydroxybutyrate (αHB), a plasma biomarker increased with insulin resistance in humans (18, 30, 32, 86), in conjunction with hyperinsulinemic-euglycemic clamps, confirmed the absence of insulin resistance in multiple tissues of 1 wk chow diet (CD)- and HFD-fed mice, thereby validating this biomarker in mouse models (57). Together, these results suggest that in the setting of HFD, β cells initially proliferate in response to a change in diet composition or an unknown acutely-upregulated HFD-induced factor. Thus adult pancreatic β cells respond robustly to proliferative cues without the complications of insulin resistance. These previous studies laid the groundwork for defining the early HFD-induced proliferative signals, the molecular nature of which are unknown. There has been a relative lack of information regarding changes in metabolite profiles following HFD feeding. Here we used metabolomic analysis on plasma and metabolically-important tissues to characterize the changes that occur within 1 wk of HFD feeding in mice. These studies reveal candidate factors that change during β-cell proliferation and mass expansion, which may correlate with promotion in β-cell compensation in response to overnutrition in the short term.

METHODS

Experimental animals.

Seven-week old male C57Bl/6J mice (Jackson Laboratory, Bar Harbor, ME) were delivered to the AAALAC accredited Vanderbilt Division of Animal Care and allowed to acclimatize for 1 wk. Mice were weighed and randomly assigned to one of two groups: 1) chow diet (CD; 11% kcal from fat, Laboratory Diet 5LJ5, Purina, St. Louis, MO) and 2) HFD (60% kcal from fat, BioServ F3282, Frenchtown, NJ). Mice were housed on a 12:12-h light-dark cycle receiving water and food ad libitum. Body weight was obtained at 8 wk of age, before starting the diets and at 1 wk after diet initiation. Euthanasia was performed at 9 wk of age using isoflurane until unconscious followed by cervical dislocation, consistent with the Panel on Euthanasia of the American Veterinary Medical Association. All procedures were conducted in accordance with Vanderbilt Division of Animal Care and Use Committee approved protocols and under the supervision of the Division of Animal Care.

Intraperitoneal glucose tolerance test (IP-GTT).

Animals were fasted for 16 h and fasting blood glucose measured from 2 μl of tail vein blood using an Accu-Chek Aviva glucometer and glucose test strips (Roche Diagnostics, Indianapolis, IN). Animals received an intraperitoneal (IP) injection of filter-sterilized glucose (2 mg dextrose/g body wt) and blood glucose was measured at 15-, 30-, 60-, 90-, and 120-min intervals following injection. Plasma insulin was not measured during the IP-GTT in this cohort. For historical data using this model the reader is referred to our previous publication (57).

Plasma collection.

Mice were placed under anesthesia following an IP-GTT. Whole blood was drawn from the vena cava. The blood was placed on ice before being centrifuged for 15 min at 13,000 rpm at 4°C. The plasma layer was collected and frozen. Other tissues used for metabolomics were also collected at this time. Use of isoflurane in healthy young female C57Bl/6 mice has been shown to impair glucose tolerance during an oral glucose tolerance test (OGTT), without affecting insulin secretion (89). Isoflurane produces a mild hyperglycemic effect in healthy young male C57Bl/6 mice, without affecting insulin secretion (19). Although all animals received the same treatment, and thus any effects of isoflurane would be expected to be present, we acknowledge the possibility that the response to anesthesia may be affected by diet.

Ex vivo β-cell proliferation assays.

The β-cell proliferation assay was performed as previously described (56). Islets were isolated from 8- to 10-wk-old wild-type C57Bl/6J male and female mice by collagenase digestion through the pancreatic duct and then hand-picked until free from acinar tissue. Islets from at least three different mice were combined and the isolated mouse islets were aliquoted at 40 islets per well into a 96-well plate and incubated with mouse proliferation media (56) supplemented 10% with plasma from CD- or HFD-fed mice collected from animals used for metabolomics studies. Separate cohorts of control islets were incubated with the following compounds: vehicle 1 [phosphate-buffered saline (PBS)], vehicle 2 [sodium bicarbonate (NaHCO₃)], 0.5 μg/ml recombinant human placental lactogen (PL; Harbor-UCLA Research and Education Institute) in NaHCO₃, 1–10 μM α-hydroxyisobutyrate (Sigma) in PBS, 50–250 μM p-cresol sulfate (ApexBiotech) in PBS. Islets were incubated for 3 days, with one media change after 36 h. For combined treatment 5 μM α-hydroxyisobutyrate and 50 μM p-cresol sulfate were used. Following incubation, islets were mildly dispersed using 0.025% trypsin and cells were cytospun onto charged slides. Slides were fixed and immunolabeled with guinea pig anti-insulin (1:400; Dako), rabbit anti-Ki67 (1:400; Abcam), Cy2-conjugated anti-guinea pig IgG (1:300; Jackson ImmunoResearch Laboratories), and Cy3-conjugated anti-rabbit IgG (1:300; Jackson ImmunoResearch Laboratories). Nuclei were visualized with 4′,6′-diamidino-2-phenylindole (DAPI, 1 μg/ml; Molecular Probes). Images were obtained using a ScanScope FL slide scanner (Aperio Technologies). β-Cell proliferation was determined by quantifying the number of insulin-Ki67 dual-positive cells using a macro generated with the CytoNuclearFL algorithm in eSlide Manager (Aperio Technologies). Data are represented as fold change in proliferation compared with vehicle-treated islets.

Metabolite profiling by mass spectrometry.

Whole liver, skeletal muscle (gastrocnemius, soleus, and plantaris muscles), bone (fibula and tibia), and plasma were collected from 9-wk-old C57Bl/6J mice fed either a HFD (n = 8) or maintained on a CD (n = 8) for 1 wk. Tissues were dissected, flash frozen in liquid nitrogen, crushed by mortar and pestle, and stored at −80°C before being sent to Metabolon (Durham, NC) for both GC-MS and LC-MS-based untargeted metabolic profiling. Sample preparation, instrument analysis, and data processing analysis performed by Metabolon are as detailed in previous publications (28, 59). Metabolites were identified by their Metabolon identification number, which were converted to Human Metabolome Database (HMDB) or Kyoto Encyclopedia of Genes and Genomes (KEGG) metabolite numbers for subsequent analysis.

Metabolomic analysis and statistics.

Metabolite data sets from each tissue type were assessed separately. Metabolites undetected in >50% of samples within a tissue were filtered from analysis. Any remaining missing values were imputed with half the minimum detected value. Following imputation, metabolite intensities were mean centered but otherwise untransformed. Means for control and HFD groups were compared using either a 2-sample t-test with Welsh’s correction or nonparametric Mann-Whitney as appropriate following a test for equal variance using Metaboanalyst 3.0 (91). The false discovery rate (FDR, Q value) was calculated to account for multiple comparisons according to the Benjamini-Hochberg procedure (38). FDR values ≤ 0.1 were considered significant, as appropriate for large biological data sets (40). Multivariate analysis was carried out using partial least-squares discriminant analysis (PLS-DA) as a reductive method (29). The data were projected using Plotly for Python (https://plot.ly) with a modified 3-D plotting script (v2.7.13). Random Forest analysis was undertaken to better assess the importance of the metabolite changes to group differences (84). Analysis of relative enrichment of KEGG pathways was also carried out with Metaboanalyst 3.0. Enrichments were determined by analysis of covariance (ANCOVA) method. The outputted data files were matched for common pathway enrichments between tissues (FDR<0.1), and covarying metabolites contributing to enrichment matched to significantly changing metabolites (FDR<0.1) using custom python scripts. The results were then projected as a networked hive plot (46), using a modified common license HTML script, which makes use of the open source D3.js. Metabolites contributing to significant pathway enrichment were only plotted if they were also determined to be changing between groups (by mean comparison above). Complete enrichment data are included as a supplement (Supplemental Table S2, available at the Journal website with the online version of this article).

RESULTS

HFD feeding for as little as 1 wk leads to β-cell expansion.

To determine the overall physiological metabolic impact of an acute switch to HFD, and to identify potential factors that promote β-cell proliferation in response to short-term HFD in the absence of insulin resistance, 8-wk-old male C57Bl/6J mice were either maintained on CD or fed HFD for 1 wk. The HFD-fed group weighed on average 3.6 g more than the CD-fed group at euthanasia (HFD = 25.2 ± 0.8 g, CD = 21.6 ± 0.5 g, P = 0.0014) (Fig. 1A). Plasma from short-term HFD-fed mice was assessed to determine whether it contained factors capable of stimulating β-cell proliferation. Intact islets isolated from wild-type CD-fed mice were cultured for 3 days as previously described with media containing plasma collected from mice fed either CD or HFD (56). Following treatment, islets were dissociated and β-cell proliferation quantified using insulin and Ki67, a nuclear marker of actively proliferating cells (Fig. 1, B–D). β-Cell proliferation was increased 1.75-fold in islets cultured with HFD-fed mouse plasma compared with islets cultured with CD-fed mouse plasma (Fig. 1B). Blood glucose levels were modestly, but not significantly, elevated in the HFD group relative to the CD group, at fasting, during the IP-GTT and 2 h post glucose injection (Fig. 1, E–G). Thus it is unlikely that elevations in blood glucose in the HFD-fed mice explain the observed increase in β-cell proliferation.

Fig. 1.

Significant β-cell proliferation occurs with high-fat diet (HFD) feeding for as little as 1 wk. A: mice fed HFD for 1 wk were significantly heavier compared with chow diet (CD)-fed mice. B: increased β-cell proliferation in control mouse islets cultured in the presence of HFD-fed (n = 6) mouse plasma compared with CD-fed plasma (n = 8). C and D: representative images used for B. Arrows mark proliferating β cells. E: blood glucose measurements during intraperitoneal glucose tolerance test (IP-GTT), at fasting (F) and at the 2-h time point (G). **P = 0.001, ***P = 0.0001.

Unique metabolite profiles of liver, muscle, bone, and plasma are altered in HFD.

To identify metabolites and metabolic pathways associated with HFD-induced β-cell expansion, we performed global metabolic profiling on liver, muscle, bone, and plasma. The number of compounds identified in each tissue and the number of biochemical metabolites that were significantly elevated in each diet group are reported in Fig. 2. Overall plasma and bone had the fewest changes in response to 1-wk HFD, with a total of seven and nine statistically relevant changes, respectively (FDR < 0.1) (Fig. 2A). Liver and muscle were more metabolically responsive with a total of 83 and 60 statistically significant changes, respectively (FDR < 0.1) (Fig. 2A). The significant changes were split largely evenly between increasing and decreasing, indicating compensatory metabolic effects beyond just the detection of dietary fats and other diet-related components. Moreover, the changes were distinct for the different tissues, with 126 of 159 total significant changes being discrete metabolites, and 99 of the metabolite changes taking place in only one of the four tissues assessed (Fig. 2A).

Fig. 2.

Summary of metabolite profiles identified for liver, muscle, bone, and plasma from high-fat diet (HFD)-fed compared with chow diet (CD)-fed mice. A: overall, 159 significant changes were observed across all tissues to 126 discrete species of metabolite in HFD-fed animals compared with controls [false discovery rate (FDR) < 0.1]. Changes were primarily in highly metabolically active tissues (muscle and liver), whereas bone and plasma were relatively resistant to changes associated with high-fat diet. B: principal component (PC) analysis: sample profiles were assessed by partial least squares-discriminant analysis (PLS-DA) regression analysis of HMDB metabolites. Permutation analysis revealed significant differentiation of the tissue clusters (P < 0.001). Metabolites not detected in at least two tissues were removed from the plot. Liver and muscle exhibited the largest spread across PC1 which accounted for the majority of the variation (73.7%). When assessed individually by PLS-DA, there was no difference in overall profile by diet for the four tissues.

Overall the changes led to distinct separation of metabolite profiles by tissue type as assessed by multivariate PLS-DA reduction, which was significant by permutation analysis (P < 0.001), indicating strong tissue-specific metabolite level profiles (Fig. 2B). Separation of these metabolite profiles was due to differences in metabolites normally associated with the particular tissue as determined by loadings plot assessment (data not shown). For example, glucose level was highest in liver and plasma, whereas phosphate and lactic acid were highest in muscle. This verifies that the profiles are a true reflection of the metabolism of the various tissues, which we later subjected to pathway analysis for tissue-specific changes.

The variance observed as a result of the HFD for any given tissue was much smaller than the variance observed between tissue types as assessed by the spread of the clusters in Fig. 2B (light vs. dark vs. different colors). When separately assessed by PLS-DA, there was no difference between HFD and CD profiles for any of the tissues. Plasma exhibited the smallest amount of variance (Fig. 2B, green), while liver was the most variable (Fig. 2B, blue).

Although the PLS-DA did not differentiate CD or HFD metabolite profiles (Fig. 2B), when we assessed the data sets by RandomForest there were significant differences between the groups, with a classification accuracy of 100% for the liver and muscle, and 87.5% for bone and plasma. This indicates that HFD profiles were statistically distinguishable from CD profiles at least seven of eight times for all four tissues. An analysis of the metabolites that were most distinguishing between CD and HFD, as calculated by the mean decrease accuracy, are discussed, tissue-by-tissue, below, and are listed in Supplemental Table S1, available at the Journal website with the online version of this article.

Metabolite changes in liver, muscle, and bone associated with HFD.

The overall changes of metabolites (FDR < 0.1) are indicated in Figs. 3–5 as fold change from CD organized by metabolite classification. For lipids, fatty acids, the principal diet composition difference in HFD compared with CD, were assessed separately (Fig. 3). Overwhelmingly, fatty acid metabolites were significantly increased in liver, except for the glycerolphosphocholine and glycerophosphoethanolamine lysophospholipids (Fig. 3). Significant fatty acid changes in the liver overlapped with the nine most important metabolite classifiers by RandomForest, inclusive of branched-chain, medium-chain, long-chain, and polyunsaturated fatty acids (Supplemental Table S1; Fig. 3); malonylcarnitine, a biomarker of disrupted fatty acid oxidation, appears as the 10th-most-changed metabolite in liver.

Fig. 3.

Heat map summary of fatty acid metabolite changes in high-fat diet (HFD)-fed mice compared with chow diet (CD)-fed controls. Fold change indicated with significant changes (P < 0.05, FDR<0.1) are designated by bold, whereas nonsignificant are annotated by gray. Heat map is colored by relative scale of change across all tissues and metabolites. Darker shades of red show increases while darker shades of blue show decreases.

Fig. 5.

Heat map summary of other metabolite changes in high-fat diet (HFD)-fed mice compared with chow diet (CD)-fed controls. Fold change indicated with significant changes (P < 0.05, FDR < 0.1) are designated by bold, whereas nonsignificant are gray. Heat map is colored by relative scale of change across all tissues and metabolites. Darker shades of red show increases while darker shades of blue show decreases.

For muscle, 3-dehydrocarnitine, an intermediary in the production of carnitine (a molecule central to fatty acid oxidation), was the most important by mean decrease accuracy (MDA). This was followed by the fatty acids palmitoleate (16:1n7) and dihomo-linoleate (20:2n6) (Supplemental Table S1). In bone, p-cresol sulfate, the largest significantly changing metabolite in all tissues (Fig. 5), was the most important RandomForest classifier and was also the third most important in plasma (Supplemental Table S1). p-Cresol sulfate is a metabolite derived from the catalysis of tyrosine by the colonic microbiota. As with liver and muscle, fatty acids and carnitines demonstrated significant changes, including 3-dehydrocarnitine, acetylcarnitine, and hexanoylcarnitine as the 4th, 7th, and 8th highest MDA, respectively (Supplemental Table S1). Interestingly, although the greatest number of overall metabolite changes took place in the liver, the greatest amount of changes to metabolites associated with carbohydrate and energy metabolism took place in muscle (Fig. 4).

Fig. 4.

Heat map summary of metabolite changes associated with energy, DNA, and protein metabolism in high-fat diet (HFD)-fed mice compared with chow diet (CD)-fed controls. Fold change indicated with significant changes (P < 0.05, FDR<0.1) are designated by bold, whereas nonsignificant are gray. Heat map is colored by relative scale of change across all tissues and metabolites. Darker shades of red show increases while darker shades of blue show decreases.

Other notable compounds that were altered in liver included metabolites implicated in one-carbon metabolism, as well as bioactive lipids including endocannabinoids, eicosanoids, and various toxins (Fig. 5). The sulfur-containing AAs, which are indicative of oxidative stress and redox homeostasis, exhibited large alterations when mice were placed on HFD for 1 wk (Fig. 4). Significant alterations in purine and pyrimidine metabolism were observed following 1 wk HFD-feeding, particularly in the muscle, and may indicate alterations in oxidative stress due to a short-term exposure to HFD (Fig. 4). Purine metabolites such as inosine, hypoxanthine, and xanthine were increased in short-term HFD-fed mouse muscle tissue by RandomForest analysis, and there was a striking increase in 3-ureidoproprionate in the liver (Fig. 4).

Analysis of metabolic pathway enrichment following HFD.

Changes in metabolically active tissues can have distal effects via plasma intermediaries. To determine which pathway changes may contribute to or associate with the β-cell mass expansion, we conducted a pathway enrichment analysis. We assessed tissues individually for covarying metabolites (ANCOVA) within KEGG pathways. Due to the small number of changes and low variance in the bone and plasma in HFD compared with CD (Fig. 2), significant enrichment was not detected for these tissues. However, a total of 15 pathways in liver and 17 pathways in muscle exhibited significant enrichment, including nine that were common to both tissues, for a total of 23 distinct pathways (Fig. 6A). The pathways are colored by their KEGG classification. A complete list of the covarying metabolites within these pathways is included in Supplemental Table S2.

Fig. 6.

Summary of enriched Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways with 1-wk high-fat diet (HFD). A: pathway analysis performed based on global analysis of covariance (ANCOVA) of metabolites revealed 31 separate significant enrichments of 23 discrete Mus musculus KEGG pathways in either liver or muscle (FDR < 0.1). Pathways are colored according to their metabolic classification in KEGG Orthology. B: graphical representation of the significance of the pathway enrichments as well as the significantly changed metabolites within those enriched pathways for muscle (left axes) and liver (right axes). Top axes: pathways in the table are plotted by increasing significance of enrichment (P < 0.05) from the center outward. Pathways enriched in both muscle and liver are linked. Bottom axes: metabolites within these KEGG pathways that were significantly changed with HFD are plotted by increasing magnitude of change (absolute fold change) from the center outward. Links are colored by KEGG pathway classification. A list of the covarying metabolites associated with the pathway enrichment are included as Supplemental Table S2).

Pathways were plotted by increasing significance of their enrichment (outermost plotted values corresponding to lowest P) (top two axes, Fig. 6B). The bottom two axes of Fig. 6B show the metabolites identified by ANCOVA as covarying metabolites within the linked pathways and plotted by the magnitude of that change (outermost plotted values corresponding to largest fold changes) (Fig. 6B). The data highlight the profound changes in fatty acid and lipid metabolism that are associated with processing the extra lipids in the HFD, with 18 enriched lipid pathways (Fig. 6, blue). These pathways are among the most significantly enriched (blue plots, Fig. 6B, top axes), due to the large number of covering metabolite changes that are mostly fatty acids (blue plots, Fig. 6B bottom axes; and Supplemental Table S2).

Enrichment of lipid pathways was common to both muscle and liver, likely as a result of increased fatty acid uptake and processing. Notably, however, for non-lipid pathway enrichments, only two carbohydrate pathways were common to both tissues (orange links, Fig. 6B, top axes). In other words, shared pathway enrichments were almost exclusively lipid pathways, whereas nonshared enrichments were almost exclusively non-lipid pathways (Fig. 3, Supplemental Table S2). First, this implies that there are profound regulatory effects in both muscle and liver that far exceed the pathways necessary to directly process these lipids. Second, muscle and liver adapt and respond metabolically very differently to a HFD. Significant non-lipid pathways enriched include important regulatory pathways such as inositol phosphate metabolism (Fig. 6), central to which is myo-inositol and associated compound changes, which are known to be both a marker of, and a treatment for, metabolic disease (17, 22, 60). Other pathways of interest that could potentially affect distal β-cell proliferation include steroid hormone biosynthesis, and folate biosynthesis, which has been implicated in DNA methylation and one-carbon metabolism. Broad metabolic rearrangements may influence β-cell expansion through intermediaries from these tissues secreted into the plasma.

Altered plasma metabolites as factors for β-cell proliferation.

Muscle and liver are active secretors of bioactive metabolites into plasma. Since plasma from 1-wk HFD-fed mice stimulated β-cell proliferation, we anticipated that peripheral tissues might be a source of circulating stimulatory metabolites in HFD plasma. Most metabolites were decreased in plasma, with only two showing a significant increase: 2(α)-hydroxyisobutyrate (Fig. 4) and p-cresol sulfate (Fig. 5). We exposed isolated mouse islets to each of these compounds independently and in combination ex vivo but did not detect a significant increase in β-cell proliferation (not shown). Plasma serves as a reservoir of metabolite biomarkers, and thus it was surprising to find that the top ten RandomForest classifiers in plasma were not primarily fatty acids or acylcarnitines (Supplemental Table S2). In fact, no statistically significant fatty acid changes were observed in plasma (Fig. 3), which demonstrates the homeostatic resilience of plasma compared with highly metabolic tissues (Fig. 2). The only amino acid (AA) altered exclusively in HFD plasma was γ-glutamylisoleucine, a catabolic product of γ-glutamyltransferase, which was decreased (Fig. 4).

When assessed by Random Forest the most important classifier was α-hydroxyisovalerate (Supplemental Table S1), a marker for ketosis, followed by coprostanol, a gut microbial-derived derivative of cholesterol that was detected exclusively in plasma and demonstrated a 5-fold decrease with HFD (Fig. 5). Additionally among them, a structurally similar plant-derived phytosterol, campesterol, which is also a stoichiometric competitor to cholesterol, was decreasing, as was 3-dehydrocarnitione, a gut microbial metabolite (Fig. 5). Equol sulfate, another metabolite that is gut-derived, significantly changed, was exclusively detected in plasma, and was also among the most important classifiers (Fig. 5). It is metabolized from the soy compound daidzein, and is an estrogen receptor β agonist.

Other important RandomForest classifiers included chenodeoxycholate, a primary human bile acid that is less common in mice, and tauroursodeoxycholate (Supplemental Table S1), which is also gut microbially derived. These bile acid changes are consistent with increased demand for gut absorption of dietary fats via plasma in the enterohepatic circulation. Pantothenate (vitamin B5), essential for coenzyme A synthesis and fat metabolism, was reduced in plasma (as well as muscle and liver) with HFD (Fig. 4). Derivatives of pantothenate, including phosphopantetheine (Fig. 5), were diminished in the liver and not detected in the other tissues. Others are phenol sulfate, another gut microbial-derived compound, and myoinositol (Fig. 5), which was significantly decreased in liver, and is known to have insulin signaling cross-talk and to be associated with metabolic disease. These metabolites were also important RandomForest classifiers (12th and 20th greatest MDA, respectively) (see Supplemental Table S1).

DISCUSSION

The main findings of the current study were to demonstrate for the first time an overall broad reshaping of metabolites and metabolism in mice after only 1-wk exposure to high-fat diet. Changes in metabolically active tissues, liver and muscle in particular, demonstrate broad rearrangement of fatty acid-associated pathways to accommodate the enriched diet. Beyond direct effects there were significant changes associated with carbohydrate metabolites in all tissues, but particularly muscle. Enrichment of other metabolites and pathways included purines and pyrimidines, AAs, and derivatives, as well as cofactors and vitamins. These changes, together with several alterations that could be derived only from gut microbiota, demonstrate the breadth and universality of metabolic change associated with even a short exposure to HFD. Changes in particular to metabolites implicated in DNA methylation and 1-carbon metabolism, oxidative stress, toxicity, and bioactive lipids, also show the enormous potential for regulatory change in diffuse tissues associated with diet. These changes could underpin broader more permanent biological changes in β-cell proliferation and/or long-term metabolic health.

The majority of changes seen after acute HFD-diet feeding, and in the absence of insulin resistance, align with what is known in the literature, specifically being an overall decrease in carbohydrate metabolism within muscle tissue, and an increase in lipid utilization and metabolism within the liver (23, 41, 43). Similarly, while widely reported and statistically significant, the metabolite αHB did not rank among the highest changing metabolites in this analysis. This is likely explained by the lack of accompanying insulin resistance at this early stage (57). Pantothenate (vitamin B₅) was uniformly decreased in all tissues examined following HFD-feeding, supporting similar findings in a model of diet-induced non-alcoholic steatohepatitis (NASH) (50), and which may be affected further by compensatory ketogenesis (20). Conversely, the uremic toxin p-cresol sulfate (pCS) was equally increased across all tissues, aligning with known elevations in insulin-resistant states, such as chronic kidney disease (CKD) (36, 45, 77). Thus elevations in p-cresol sulfate may precede and predict insulin resistance. Interestingly, alterations in pCS have been shown in urine and plasma of children with Type 1 diabetes, who otherwise demonstrated good glycemic control and were taking insulin (4). This suggests that pCS may be a general biomarker of metabolic function and may warrant further examination, although we could not directly attribute these changes to increases in β-cell proliferation following acute HFD exposure in the present study.

These data also support the involvement of other metabolites including glucose, mannose, unsaturated fatty acids, and lactate, which were shown to decrease in serum significantly following Roux-en-Y gastric bypass (RYGB) surgery, improving glycemic profiles and insulin sensitivity in both obese and Type 2 diabetes patients (58). That previous study also showed an elevation in lipid metabolites including sphingosine following RYGB, as well as increased p-cresol sulfate in obese and diabetic subjects (58). Our data reveal a decrease in sphingosine presence in the liver and muscle following HFD-feeding; interestingly, sphingosine-1-phosphate (S1P) phosphatase 2 (Sgpp2) has been reported to be a specific regulator of β-cell mass and endoplasmic reticulum stress (ER) homeostasis. Sgpp2^−/− mice exhibit ER stress, and show defects in β-cell proliferation and function following long-term HFD feeding (80). The metabolism of S1P is critically required for regulating relative levels of three related molecules in the sphingolipid pathway, S1P, sphingosine, and ceramide, and alterations in the sphingolipid pathway have been implicated in HFD models of disease (15). We also observed a decrease in the 1-carbon metabolite pathway following HFD-feeding. Previous studies have demonstrated that supplementation along the 1-carbon pathway can alter DNA methylation, gene expression, and heritability of epigenetic traits in rodent models, including insulin resistance and metabolic syndrome (35, 74).

Within the liver specifically, the purine/pyrimidine related metabolite 3-ureidopropionate, a urea derivative of β-alanine and intermediate in uracil metabolism, was significantly increased. In disparate disease etiologies, HFD feeding shares characteristics with the parasitic infection schistosomiasis, causing disruption in AA metabolism (90), and broadly mimicking other liver diseases, including hepatitis (14). In a separate study, gene mutations in UBP1 coding for β-ureidopropionase gathered from plasma proteomic assessment of patients at risk for cardiovascular disease demonstrated inborn errors in metabolism (69). Thus ureidopropionate may be a novel factor to investigate the impetus for β-cell proliferation in response to HFD. Indeed, decades-old evidence has implicated the liver as the mechanistic driver by which high-fat diets, even briefly, impair glucose tolerance (13), and which has been reinforced by more recent and elegant genetic knockout studies (31, 39, 53, 61, 62, 75). One purported hepatic target includes insulin-like growth factor-1 (IGF-1), with one study using an inducible liver-specific insulin receptor knockout model (iLIRKO). The authors report IGF-1 to be acting through the insulin-receptor A isoform (27), putatively via mTORC1/p70S6kinase signaling on β-cells in response to insulin resistance (26).

Pharmacological evidence supports the impact of glycemic perturbation resulting from altered lipid profiles within the liver. Dimethyl biguanide (metformin) is a first-line treatment for Type 2 diabetes, particularly in overweight individuals (51), affecting blood glucose reduction by inhibiting hepatic glucose generation and promoting adipose- and muscle-glucose uptake (3). Metformin has been shown to inhibit β-cell proliferation induced by 1-wk HFD feeding (82), independent of its effects on insulin resistance, and putatively acting by reduction of mTOR signaling, via the activation of AMPK (82). Previous work from a different group showed that the addition of a dipeptidyl peptidase-IV (DPP-IV) inhibitor, TS-021, with metformin treatment had the opposite effect, increasing β-cell mass and lowering glucose, although β-cell proliferation directly was not examined (81). Furthermore, the incretin-effect of DPP-IV inhibition on mouse β-cell mass expansion following HFD feeding has not been validated by other studies (21). Similarly, although in an entirely different class of drugs, β-hydroxy-β-methylglutaryl (HMG)-CoA reductase inhibitors (statins), which act by lowering endogenous cholesterol synthesis in the liver, have been linked with a 9% increased risk of Type 2 diabetes, particularly in older patients (73). The authors’ explanations of this link was not causal, however, suggesting confounding factors; furthermore, there is no consensus regarding the effect of statins on insulin resistance or sensitivity in humans (33, 63) or rats (48). Although one report suggested that atorvastatin increased mouse “pancreatic proliferation,” this conclusion was reached by assessment of β-cell area, and not direct quantitation of proliferation (12), and studies performed by others were either inconclusive in human islet ex vivo conditions (7) or showed no direct effect by the addition of a statin on rat β-cell proliferation following HFD feeding (54). However, one recent report seemed to reconcile these findings, indicating a dual (and opposing) effect of improved insulin sensitivity with impaired β-cell function in mice following HFD feeding (72). Clearly, the widely ranging outcomes on β-cell mass and proliferation following treatment with hepatic-targeted drugs warrant further investigation, and reinforce the liver-endocrine-pancreas link.

In this study we sought to assess molecules that are possible stimulators of β-cell proliferation, which could be plasma-borne, even if their origins are elsewhere. Considering the total number of observed changes that were higher in liver but low in plasma (Fig. 2A), the data reflect a highly homeostatic plasma compared with a very dynamic metabolic response in liver with HFD challenge. Plasma is often considered to be a rich repository of biomarkers and biological effectors reflective of tissue changes. Given the homeostatic resilience observed here, the small list of metabolites that were able to very significantly overcome these mechanisms was of high interest. Neither α-hydroxyisobutyrate nor p-cresol sulfate stimulated β-cell proliferation in mouse islets ex vivo at the concentrations tested here. It is possible that either or both compounds would stimulate β-cell proliferation if tested at different concentrations. Alternatively, it may be that one of the compounds that was decreased in plasma from HFD-treated animals (there were 5) may be an inhibitor of β-cell proliferation. This possibility remains to be tested in future experiments.

Our data demonstrate that the liver primarily, and muscle to a lesser degree, derive the greatest amount of metabolic changes in response to short-term HFD, but that many changes are common to different tissues. Of the rare specific changes in plasma, α-hydroxyisobutyrate elevation has been previously associated with protein ingestion-induced insulin resistance (37), and shares similarities with neonatal murine undernutrition including muscle and fat catabolism, liver injury, and oxidative stress (68). Furthermore, although metabolomic analyses in the Framingham Offspring study showed a positive relationship between elevated plasma branched-chain and aromatic AA presence and the future risk of Type 2 diabetes (88), we found no such AA elevations following HFD feeding. In fact, within the liver, this AA metric was decreased, but ostensibly could be explained in our study by the brief (1 wk) HFD exposure in a model free of insulin resistance.

Forces driving pancreatic β-cell proliferation and its consequent mass expansion have been the subject of intensive scrutiny, both to understand normal physiological changes during response to injury and development, as well as to target pharmacological agents for intervention in diabetes therapy. A limitation of our strategy is that circulating factors with unchanged abundance, but increased or decreased flux, would be missed. Despite that, to our knowledge, this is the first report to use an unbiased platform and describe HFD-induced metabolite changes in metabolically important organs such as liver and muscle. Some of these metabolites are tissue-inherent whereas others are derived from the gut microbiome, which has been shown to be changed by HFD feeding (16, 49). The effects of a Western diet high in fat on these metabolites and their potential correlating impact on the endocrine pancreas warrant further investigation. Given the universality and magnitude of diverse pathway changes, particularly from liver, our findings are consistent with the theory that β cells initially proliferate in response to a HFD-induced factor potentially released from the liver. These data further reveal numerous avenues for interrogation to elucidate the mechanisms affecting β-cell proliferation, as well as potential novel drug targets.

GRANTS

V. Poitout holds the Canada Research Chair in Diabetes and Pancreatic Beta Cell Function and was supported by the National Institutes of Health (NIH)/National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) (Grant R01-DK-58096), the Canadian Institutes of Health Research (MOP77686), and the Juvenile Diabetes Research Foundation (17-2012-26). S. E. Townsend was supported by the Vanderbilt Molecular Endocrinology Training Grant (5-T32-DK-07563). M. Gannon was supported by grants from the NIH/NIDDK (Grant 2R24-DK-090964-06), Department of Veterans Affairs (1I01 BX003744-01), and the Juvenile Diabetes Research Foundation (17-2012-26). K. M. Aagaard was supported by the NIH/NIDDK (Grant 2R24-DK-090964-06).

DISCLOSURES

K. Pappan is an employee and stockholder of Metabolon, Inc., where some of the experiments were performed. None of the other authors has any conflicts of interest, financial or otherwise, to disclose.

AUTHOR CONTRIBUTIONS

R.E.M., S.E.T., V.P., and M.G. conceived and designed research; R.E.M., S.E.T., and K.P. performed experiments; M.D.S., C.A.B., R.E.M., S.E.T., K.P., K.A., and M.G. analyzed data; M.D.S., C.A.B., S.E.T., K.P., K.A., and M.G. interpreted results of experiments; M.D.S., C.A.B., S.E.T., and K.P. prepared figures; M.D.S., C.A.B., and M.G. drafted manuscript; M.D.S., C.A.B., R.E.M., K.P., V.P., K.A., and M.G. edited and revised manuscript; M.D.S., C.A.B., R.E.M., K.P., V.P., K.A., and M.G. approved final version of manuscript.