High-fructose feeding suppresses cold-stimulated brown adipose tissue glucose uptake independently of changes in thermogenesis and the gut microbiome.

Diets rich in added sugars are associated with metabolic diseases, and studies have shown a link between these pathologies and changes in the microbiome. Given the reported associations in animal models between the microbiome and brown adipose tissue (BAT) function, and the alterations in the microbiome induced by high-glucose or high-fructose diets, we investigated the potential causal link between high-glucose or -fructose diets and BAT dysfunction in humans. Primary outcomes are changes in BAT cold-induced thermogenesis and the fecal microbiome (clinicaltrials.gov, NCT03188835). We show that BAT glucose uptake, but not thermogenesis, is impaired by a high-fructose but not high-glucose diet, in the absence of changes in the gastrointestinal microbiome. We conclude that decreased BAT glucose metabolism occurs earlier than other pathophysiological abnormalities during fructose overconsumption in humans. This is a potential confounding factor for studies relying on 18F-FDG to assess BAT thermogenesis.

Introduction

The consumption of fructose, a food additive highly prevalent in ultra-processed foods in the form of sucrose or high-fructose corn syrup, has risen precipitously over the past few decades. Its overconsumption has paralleled the rapid rise in obesity and the associated metabolic complications, including insulin resistance, non-alcoholic fatty liver disease, and cardiovascular disease. Animal studies, primarily performed in rodents, have consistently demonstrated that diets high in sucrose or fructose induce hypertriglyceridemia and hyperinsulinemia, providing a causal link between fructose consumption and dysmetabolism.1, 2, 3, 4 Such direct evidence in humans, however, has remained more elusive. Several pivotal short-term intervention studies in humans have shown that consuming fructose-sweetened beverages contributing 10% to 25% of energy intake increases circulating triglyceride levels, visceral adiposity, hepatic lipid accretion, and insulin resistance compared with isocaloric glucose or sweeteners.5, 6, 7, 8, 9 Conversely, restricting sugar-sweetened beverages in children with obesity, lowered fasting triglyceride levels and hepatic fat, while also improving postprandial insulin kinetics. While these studies have been critical to the characterization of the impact of consuming modest levels of fructose on metabolic risk factors, the mechanisms by which isocaloric fructose may promote metabolic disease and its association with insulin resistance remain unknown. This has been particularly difficult to ascertain given the limited number of studies using appropriate controls and directly comparing the effects of fructose-sweetened beverages with isocaloric glucose.

One emerging mechanism that has been proposed as a driving factor in fructose-induced dysmetabolism is the progressive impairment in brown adipose tissue (BAT) metabolic activity. Retrospective studies commonly report that high spontaneous BAT activity is associated with a favorable metabolic profile including lower circulating triglycerides and higher insulin sensitivity,12, 13, 14, 15 whereas impairments in BAT glucose uptake, non-esterified fatty acid (NEFA) uptake, blood flow, and insulin sensitivity have been noted in aging, obesity, and type 2 diabetes (T2D).,16, 17, 18 Although several recent studies have demonstrated the ability to activate human BAT through acute and intermittent cold exposure and acute and chronic beta-adrenergic stimulation,19, 20, 21, 22, 23, 24, 25 our understanding of the factors that suppress its activity with aging, obesity, or T2D remains incomplete. Recent evidence in rodents suggests that fructose ingestion increases intestinal microbiota-derived acetate production, which may not only provide a precursor for hepatic lipogenesis but could incidentally suppress BAT function, providing a prospective causal role for BAT in fructose-associated dysmetabolism.

We hypothesized that fructose could impair BAT metabolic activity and thermogenesis and may be an early feature of the fructose-induced deterioration in metabolic profile in humans. To test this hypothesis, we conducted a randomized crossover 2-week fructose versus glucose overfeeding (+25% of energy intake) study in 10 healthy young men, interspersed by a 4-week washout period and preceded by a metabolic assessment performed under an isocaloric condition. The primary outcomes were (1) changes in the fecal microbiome and short-chain fatty acid (SCFA) composition, and (2) BAT cold-induced thermogenesis, assessed via measures of oxidative metabolism and intracellular triglyceride depletion. Based on previous studies, we expected that fructose would change the taxonomic composition of the microbiome, that fecal acetate would increase, and that BAT oxidative metabolism would decrease. Consequently, the decrease in oxidative metabolism would lead to less BAT intracellular triglyceride depletion during BAT activation. Whole-body energy metabolism and BAT glucose metabolism were also investigated as secondary outcomes. Based on previous studies of pharmacological BAT inhibition, we did not expect to see a detectable change in whole-body heat production, but a significant decrease in BAT oxidative metabolism and glucose uptake was expected. A complete list of the primary and secondary outcomes for this study is available on clinicaltrials.gov (NCT03188835).

Results

Short-term fructose overfeeding does not alter adiposity, the gut microbiome, or fecal SCFAs

The high-fructose and high-glucose diets were designed to be hypercaloric, with the carbohydrate-enriched drink representing an additional 25% of the total caloric intake. Each diet was administered for 2 weeks in an outpatient setting based on previous studies showing effects of this level of high-fructose consumption on plasma lipid profiles., The short duration of each diet minimized weight gain and changes in body composition, which are potential confounding factors. Participants confirmed drinking the entire diet supplement and were instructed to keep their other dietary intake the same as in the baseline isocaloric period. Changes in total body mass, fat mass, and lean mass (after the diet versus before the diet) were highly variable between individuals for all diets (Figure 1A). On average, there was no difference in body composition between diets. The high-fructose diet elicited a more reproducible increase in total body weight, although this seemed to be related to a gain in lean body mass, rather than in fat mass. Physical activity levels and resting energy expenditure did not change between diets (Figures 1B and 1C). Similarly, there were no differences in tissue-specific fat content assessed by magnetic resonance imaging (MRI) for BAT, subcutaneous or visceral adipose tissue, or the liver (Figure 1D). These results are coherent with those of Stanhope et al. reporting changes in body weight of less than 0.25 kg per week of diet.

Figure 1

Body composition at baseline and following the high-carbohydrate diets

No significant change in whole-body or tissue-specific composition was noted with the different diets.

(A) Changes in body mass and composition assessed (after the diet - before the diet) by dual-energy X-ray absorptiometry (n = 10).

(B) Energy expenditure from exercise assessed by uniaxial accelerometer at baseline (n = 8), during high fructose consumption (n = 10), and high glucose consumption (n = 9).

(C) Whole-body energy expenditure at rest assessed by indirect calorimetry (n = 10).

(D) Fat content of tissues assessed by Dixon MRI (n = 10). All p values are >0.05. BAT, brown adipose tissue; FM, fat mass; PDFF, proton-density fat fraction; LM, lean mass; scWAT, subcutaneous white adipose tissue; VAT, visceral adipose tissue. Participant 13 was excluded from the analyses due to antibiotic use (potential effects on the microbiome) before the fructose and glucose diets and is shown with a square symbol shape.

We investigated the effects of the high-carbohydrate diets on the fecal microbiome composition through 16S ribosomal RNA gene variable region 3–4 sequencing. Participants provided stool samples at the end of every 2-week diet period (at baseline, high glucose, and high fructose). No significant differences in fecal microbiome composition were observed in terms of taxonomic composition, α-, and β-diversity (Figures 2A–2C). We also measured the fecal SCFA content (acetate, butyrate, and propionate), important metabolites produced by microbiota fermentation of indigestible carbohydrates, by gas chromatography–tandem mass spectrometry. There was no effect of the diets on SCFA content or content ratios (e.g., acetate/propionate ratio), although there was a high variability between subjects (Figures 2D and 2E). By comparison, large changes in the gut microbiome and SCFA were reported as early as 5 days following the onset of plant-based or animal-based diets.

Figure 2

Fecal microbiome analyses and fecal short-chain fatty acids for the different diets

(A) Relative abundance taxonomic summaries divided by diet and ordered by participant for all participants included in statistical analyses (n = 9). Top 12 genera in the data were colored and are identified in the legend. All other genera are in “other.”

(B) Participants’ fecal microbial α-diversity assessed by Shannon and Simpson indices at baseline (n = 10), following high fructose consumption (n = 9) and high glucose consumption (n = 9). Each dot represents the value of the diversity metric from each sample. The boxes show the median and interquartile range. A generalized mixed-effects model found no significant differences by α-diversity.

(C) Participants’ fecal microbial β-diversity assessed by Aitchison distance and visualized using a Principal-Coordinate Analysis plot. Each dot represents a sample and samples are colored by diet (baseline [n = 10], high fructose [n = 9], and high glucose [n = 9]). A permutational multivariate ANOVA found no significant differences by β-diversity. Percent variation explained by each axis is noted in square brackets.

(D) Fecal short-chain fatty acid composition for baseline (n = 10), high fructose (n = 10), and high glucose (n = 9). No significant difference was observed between diets.

(E) Fecal short-chain fatty acid ratios. No significant difference was observed between diets. All p values are >0.05. Participant 13 was excluded from the analyses due to antibiotic use (potential effects on the microbiome) before the fructose and glucose diets and is shown with a square symbol shape.

Cold-induced BAT oxidative metabolism and intracellular triglycerides depletion remain normal after the carbohydrate-enriched diets

Participants underwent mild cold exposure (18°C) to activate BAT metabolism. Primary outcomes were BAT oxidative metabolism assessed with 11C-acetate positron emission tomography (PET), as well as intracellular fatty acid content and composition assessed by MRI and magnetic resonance spectroscopy (MRS).

During the intervention, skin temperature, shivering activity, heat production, circulating energy substrates, whole-body substrate oxidation, and hormones were also measured (Table 1). As expected, cold exposure elicited a decrease in mean skin temperature and increase in heat production, and this was fueled mainly by an increase in fatty acid oxidation. Both blood glucose and insulin levels decreased during cold exposure, while circulating levels of non-esterified fatty acids increased. This is consistent with an increase in glucose uptake by tissues and an increase in whole-body adipose tissue intracellular lipolysis.

Table 1

Whole-body response to diets and cold

	Baseline		Fructose		Glucose
	RT	18°C	RT	18°C	RT	18°C
EE (kJ·min−1)	5.5 (5.0–5.9)	8.5 (7.2–9.9)d	5.4 (4.8–6.1)	8.7 (7.7–9.6)d	5.5 [5.1–5.8]	8.7 (7.4–9.9)d
Skin temperature (°C)	34.5 (33.8–35.2)	28.6 (27.9–29.4)d	34.2 (33.8–34.7)	29.0 (28.0–30.0)d	34.4 (33.9–35.0)	28.6 (27.8–29.3)d
Shivering intensity (%MVC)f	Not measured at RT	1.29 (0.69–1.88)	Not measured at RT	1.50 (0.72–2.28)	Not measured at RT	1.43 (0.71–2.15)
RaGlucose (μmol·min−1)	1902 [1642–2227]	1457 [1332–1492]d	2134 (1716–2552)	1529 (1163–1894)d	1989 (1710–2269)	1553 (1334–1772)d
RaNEFA (μmol·min−1)	540 (311–770)	1195 (765–1626)b	657 (374–940)	1132 (732–1532)a	849 (435–1263)	1430 (879–1982)a
RaGlycerol (μmol·min−1)	276 [195–334]	859 [758–1431]b	305 [254–369]	941 [589–2302]c	327 [219–396]	986 [725–1647]b
Glucose (mmol/L)	4.9 (4.7–5.1)	4.6 (4.2–5.0)a	4.8 [4.7–5.0]	4.5 [4.3–4.8]b	4.5 [4.4–5.1]	4.6 [4.1–4.9]a
Insulin (pmol/L)	60.7 (43.9–77.6)	38.6 (27.1–50.1)b	53.9 (37.3–70.4)	40.5 (32.2–48.7)	55.8 (36.1–75.5)	39.7 (25.4–53.9)
Glucagon (pg·mL−1)	55.3 (25.9–84.8)	46.4 (24.5–68.3)	48.9 (26.0–71.7)	40.9 (19.6–62.1)	59.7 (24.5–94.9)	48.3 (18.7–77.8)
TG (mmol/L)	0.60 [0.43–1.34]	0.60 [0.46–1.54]	0.63 [0.49–1.60]	0.68 [0.43–1.58]	1.14 (0.47–1.81)	1.02 (0.50–1.54)
VLDL-TG (mmol/L)	0.24 [0.14–0.40]	0.15 [0.11–0.29]	0.32 (0.15–0.49)	0.36 (0.14–0.59)e	0.36 (0.23–0.49)	0.38 (0.18–0.57)e
NEFA (μmol/L)	290 (164–416)	612 (389–836)b	321 (190–452)	538 (377–699)b	316 (177–454)	558 (405–711)c
TSH (IU·L−1)	1.9 (1.1–2.6)	1.5 (0.9–2.2)a	1.8 (1.1–2.6)	1.4 (0.8–2.0)b	1.6 (1.1–2.1)	1.4 (0.9–1.9)
Free T3 (pmol/L)	5.2 (4.9–5.6)	5.4 (4.8–5.9)	5.7 [5.2–6.2]e	5.8 [5.1–6.3]	5.5 [5.1–6.1]	5.5 [5.2–6.3]
Free T4 (pmol/L)	17.1 (14.0–20.2)	18.1 (14.6–21.6)	16.6 (15.1–18.0)	17.7 (16.0–19.5)a	15.5 [15.1–16.9]	16.3 [15.7–17.5]a
Cortisol (nmol/L)	230 (201–259)	201 (168–234)	219 (162–276)	230 (195–264)	223 (139–307)	223 (196–251)e
Leptin (ng·mL−1)	0.8 [0.5–1.8]	0.6 [0.3–1.3]b	1.4 [0.4–1.9]	0.9 [0.3–1.5]b	1.2 [0.4–2.0]	0.7 [0.3–1.4]b
C-peptide (nmol/L)	0.25 (0.18–0.31)	0.18 (0.14–0.22)b	0.24 (0.18–0.29)	0.18 (0.14–0.22)c	0.21 (0.16–0.26)	0.16 (0.12–0.20)b
GIP (pmol/L)	5.4 [4.4–14.4]	7.4 [5.1–9.0]	9.5 (5.7–13.3)	9.1 (6.8–11.3)e	8.5 (5.0–11.9)	7.6 (5.2–10.0)
Total GLP-1 (pmol/L)	35.3 (27.3–43.3)	34.4 (26.2–42.5)	32.9 (24.2–41.6)	33.2 (24.4–42.0)	35.2 (25.6–44.9)	35.6 (26.7–44.4)
HOMA-IR	1.91 (1.36–2.45)	1.15 (0.75–1.55)b	1.73 (1.15–2.30)	1.20 (0.93–1.47)a	1.75 (1.05–2.45)	1.15 (0.72–1.57)

Values are mean (95% confidence interval) for normally distributed data, and median [interquartile range] for nonparametric data. %MVC, percent of maximum voluntary contraction; EE, energy expenditure; GIP, glucose-dependent insulinotropic polypeptide; GLP-1, glucagon-like peptide-1; HOMA-IR, homeostatic model assessment for insulin resistance; NEFA, non-esterified fatty acids; Ra, appearance rate; RT, room temperature; T3, triiodothyronine; T4, thyroxine; TG, triglycerides; TSH, thyroid stimulating hormone; VLDL, very low-density lipoproteins; GIP, glucose-dependent insulinotropic polypeptide; GLP-1, glucagon-like peptide-1; HOMA-IR, homeostatic model assessment for insulin resistance. Participant 13 was excluded from the analyses and is not included in the mean/median calculations due to antibiotic use (potential effects on the microbiome) before the fructose and glucose diets.

Two-way repeated measures ANOVA or repeated measures mixed-effects model with Bonferroni correction for multiple comparisons.

Effect of cold intervention, p < 0.05.

p < 0.01.

p < 0.001.

p < 0.0001.

Different from baseline diet, p < 0.05.

Shivering data were not collected for participant 15 due to COVID-19 restrictions.

The time-activity curves of 11C-acetate in BAT showed a higher uptake and washout during cold exposure than at room temperature (Figure 3A). These changes were quantified with a four-compartment, two-tissue model developed by our group., With this model, the uptake parameter (K1) can be used to estimate blood flow changes because first-pass extraction of the tracer is very high in most tissues. The washout parameter (k2) represents oxidation (i.e., BAT thermogenesis). There was a significant increase in blood flow (Figure 3C; p = 0.007) and oxidation (Figure 3D; p = 0.003) during cold exposure, consistent with increased mitochondrial respiration. However, there was no significant difference between diets in blood flow (fructose: p = 0.3; glucose p = 0.5) or oxidation (fructose and glucose: p = 0.6) during cold exposure, suggesting that mitochondrial respiration is not affected by 2 weeks of high-carbohydrate diet. Measures at room temperature were only acquired during the baseline protocol to limit radiation exposure to participants. Therefore, the magnitude of change between room temperature and cold exposure is reported only at baseline.

Figure 3

Brown adipose tissue oxidative metabolism and intracellular triglyceride depletion

(A) Mean time-activity curves (with standard deviation) showing the uptake and elimination of 11C-acetate or its metabolites at room temperature (baseline) and during cold exposure (18°C) for baseline, fructose, and glucose diets.

(B) Proton-density fat fraction shift in brown adipose tissue with cold exposure (18°C - room temperature) for the different diets (n = 10). There is no difference in shifts between diets.

(C) Blood flow index at baseline (n = 10; room temperature and cold), high fructose (cold: n = 9), and high glucose (cold: n = 10). A repeated measures mixed-effects model found a significant difference between room temperature and cold exposure (all conditions).

(D) Oxidative metabolism at baseline (n = 10; room temperature and cold), high fructose (cold: n = 9), and high glucose (cold: n = 10). A repeated measures mixed-effects model found a significant difference between room temperature and cold exposure (all conditions). p values are shown only for significant differences (i.e., p < 0.05). BAT, brown adipose tissue; PDFF, proton-density fat fraction; K1, 11C-acetate uptake rate; k2, 11C-acetate oxidation rate; RT, room temperature; SUV, standardized uptake value. Participant 13 was excluded from the analyses due to antibiotic use (potential effects on the microbiome) before the fructose and glucose diets and is shown with a square symbol shape. 11C-acetate was not performed for participant 13 fructose due to problems with tracer synthesis.

Intracellular triglyceride content, as assessed by Dixon MRI proton-density fat fraction, decreased in BAT during cold exposure (Figure 3B). The reduction in BAT proton-density fat fraction following cold exposure was consistent with previous studies35, 36, 37, 38 and with the previously reported increase in BAT radiodensity,,39, 40, 41 which both reflect a progressive decrease in intracellular triglyceride content. This is also consistent with previous findings by our group indicating that intracellular triglycerides are the preferred energy source for BAT thermogenesis. Diets did not modify the BAT proton-density fat fraction at room temperature (Figure 1D) nor the shift with cold exposure (Figure 3B). These findings are in line with the previous observations that diets did not alter body composition, and that BAT oxidative metabolism was preserved.

Furthermore, the lipid composition of BAT and subcutaneous white adipose tissue investigated by MRS (Table 2) were not affected by diets or cold exposure. However, as expected, BAT contained more saturated fatty acids than subcutaneous white adipose tissue, demonstrated by a lower number of double bonds (median for all diets and temperatures: 3.5 [IQR: 3.0–3.8] versus 4.1 [IQR: 3.4–4.5]; p = 0.002) and methylene-interrupted single bonds (median for all diets and temperatures: 1.4 [IQR: 1.0–2.0] versus 3.3 [IQR: 2.6–3.9]; p < 0.0001).

Table 2

Fat composition of different tissues assessed by MRI and MRS

	Baseline		Fructose		Glucose
	RT	18°C	RT	18°C	RT	18°C
BAT PDFF (%)	71.8 (66.6–76.9)	70.3 (64.3–76.4)	73.2 (66.5–79.9)	71.5 (63.7–79.3)a	71.4 (63.8–79.0)	68.3 (60.0–76.6)a
scWAT PDFF (%)	87.2 [83.1–91.2]	88.3 [81.4–91.2]	89.4 [81.1–90.2]	89.6 [80.9–90.9]	89.7 [81.7–91.6]	89.9 [82.9–90.5]
VAT PDFF (%)	87.8 (84.8–90.7)	87.9 (85.5–90.3)	87.9 (83.9–91.9)	88.2 (85.2–91.1)	88.2 (85.5–90.9)	87.0 (83.2–90.8)
Liver PDFF (%)	4.1 [2.5–5.3]	4.5 [2.4–5.3]	4.5 [2.6–5.1]	5.1 [2.8–6.1]a	3.6 (2.9–4.2)	3.5 (2.6–4.4)
BAT ndb	3.6 (2.5–4.7)	3.9 (3.1–4.6)	3.7 (2.7–4.7)	3.1 (2.6–3.6)	3.6 (2.7–4.5)	3.2 (2.8–3.7)
scWAT ndb	4.1 (3.0–5.2)	3.5 (2.7–4.2)	3.9 (3.2–4.5)	4.1 (3.4–4.8)	4.2 (3.4–5.1)	3.7 (3.0–4.5)
BAT nmidb	2.5 [1.7–3.4]	2.5 [2.2–3.2]	2.2 [1.8–3.3]	2.0 [0.4–2.8]	3.2 (1.7–4.7)	2.0 (1.4–2.6)
scWAT nmidb	3.5 (2.3–4.6)	3.2 (2.6–3.8)	3.2 (2.0–4.3)	2.9 (2.1–3.7)	3.9 (2.9–5.0)	2.4 (1.4–3.5)

Values are mean (95% confidence interval) for normally distributed data, and median [interquartile range] for nonparametric data. BAT, brown adipose tissue; PDFF, proton-density fat fraction; ndb, number of double bonds; nmidb, number of methylene-interrupted double bonds; RT, room temperature; scWAT, lower back subcutaneous white adipose tissue; VAT, visceral adipose tissue. Participant 13 was excluded from the analyses and is not included in the mean/median calculations due to antibiotic use (potential effects on the microbiome) before the fructose and glucose diets.

Two-way repeated measures ANOVA or repeated measures mixed-effects model with Bonferroni correction for multiple comparisons.

Effect of cold intervention, p < 0.05.

Short-term fructose overfeeding reduces BAT glucose uptake

Tissue-specific glucose metabolism was assessed by 18F-fluorodeoxyglycose (18F-FDG) PET during cold exposure. The time-activity curves of 18F-FDG in BAT (Figure 4A) show a quick initial increase in signal, corresponding to the first-pass extraction of 18F-FDG by the tissue. Afterward, the signal continues to increase at a slower pace as 18F-FDG in the blood is taken up and trapped into BAT. The initial uptake phase appears similar for all diets. However, the subsequent trapping rate of 18F-FDG is slower after the high-fructose diet. To quantify this phenomenon, BAT fractional glucose uptake was assessed with a Patlak graphical model. The tissue metabolic rate of glucose was then calculated using fractional glucose extraction, glucose concentration, and a lumped constant of 1.14 for fat and 1.16 for muscle., Both the fractional extraction (Figure 4B; p = 0.044) and metabolic rate of glucose (Figure 4C; p = 0.036) decreased in BAT following the high-fructose diet, but not following the high-glucose diet. We did similar measures and modeling in different skeletal muscles as well as subcutaneous adipose tissue and conclude that this decrease in glucose uptake was specific to BAT (Figure 4D). The reduction in cold-stimulated BAT glucose uptake following the high-fructose diet was not associated with differences in blood glucose, insulin, homeostatic model assessment for insulin resistance (HOMA-IR), glucagon, cortisol, c-peptide, glucose-dependent insulinotropic polypeptide, or glucagon-like peptide-1. Furthermore, no difference in whole-body carbohydrate oxidation was observed with diets (Table 1).

Figure 4

Glucose metabolism of tissues with cold exposure

(A) Mean brown adipose tissue time-activity curves (with standard deviation) showing 18F-fluorodeoxyglucose uptake during cold exposure (18°C) for baseline, fructose, and glucose diets.

(B) Brown adipose tissue fractional glucose uptake at baseline (cold: n = 10), high fructose (cold: n = 9), and high glucose (cold: n = 10). A repeated measures mixed-effects model found a significant difference between baseline and fructose.

(C) Brown adipose tissue metabolic rate of glucose at baseline (cold: n = 10), high fructose (cold: n = 9), and high glucose (cold: n = 10). A repeated measures mixed-effects model found a significant difference between baseline and fructose.

(D) Metabolic rate of glucose for brown adipose tissue compared with skeletal muscles and subcutaneous fat at baseline (cold: n = 10), high fructose (cold: n = 9), and high glucose (cold: n = 10). Data are shown as mean and 95% confidence interval. There was no significant difference for tissues other than BAT (repeated measures mixed-effects model baseline versus fructose). p values are shown only for significant differences (i.e., p < 0.05). 18F-FDG, 18F-fluorodeoxyglycose; BAT, brown adipose tissue; Ki, fractional glucose uptake; MRGlu, metabolic rate of glucose; SCM, sternocleidomastoid muscle; scWAT, lower back subcutaneous white adipose tissue; SUV, standardized uptake value. Participant 13 was excluded from the analyses due to antibiotic use (potential effects on the microbiome) before the fructose and glucose diets and is shown with a square symbol shape. Participant 2 fructose 18F-fluorodeoxyglucose was excluded due to excessive participant motion.

These observations point to a very localized reduction in glucose uptake specific to BAT and independent of changes in oxidative metabolism or intracellular triglyceride depletion. Because 18F-FDG PET provides macroscopic data (with 4 mm3 image voxels) and does not discriminate between metabolic pools (i.e., 18F-FDG and 18F-FDG-6-phosphate produce the same signal), we cannot determine the exact cause of this reduced uptake. Based on the 11C-acetate data, blood flow is not affected by the fructose diet. Therefore, the reduction in glucose uptake is likely due to a deficit in transport into the brown adipocytes (e.g., fewer glucose transporters or reduced transporter activity), reduced phosphorylation rates, or a combination of both. Whether this is related to a local effect of circulating acetate, as suggested by Sun et al., cannot be determined with our experimental design. There was no change in the acetate content of the stool samples, and plasma acetate could not be measured accurately by our gas chromatography–tandem mass spectrometry method.

Fructose overfeeding increases liver triglyceride content as well as serum VLDL-TG in the cold

The liver triglyceride content assessed by MRI was significantly increased after cold exposure compared with room temperature only after the high-fructose diet (Figure 5A and Table 2; p = 0.037). This increase was not observed in the neighboring visceral or subcutaneous adipose tissues (Figure 5A), and as noted previously, BAT triglyceride content decreased in the cold exposure condition (Figure 3B). This suggests that fructose consumption promotes hepatic lipid storage under cold-induced increase in adipose tissue intracellular lipolysis, even though the magnitude of this increased lipolysis is not affected by diets (Figures 5B and 5C).

Figure 5

Effect of cold exposure on the liver and whole-body lipid metabolism

(A) Proton-density fat fraction of the liver, visceral adipose tissue, and white adipose tissue at room temperature and in the cold (18°C) for each diet (n = 10). A two-way repeated measures ANOVA found a significant difference between room temperature and cold in the liver with fructose.

(B) Changes in the rate of appearance of glycerol with cold exposure assessed by stable isotope dilution (n = 9).

(C) Changes in the rate of appearance of non-esterified fatty acids with cold exposure assessed by stable isotope dilution at baseline (n = 9), high fructose (n = 10), and high glucose (n = 10).

(D) Circulating very-low-density lipoprotein-triglycerides at room temperature and during cold (n = 10). A two-way repeated measures ANOVA found a significant difference between the baseline and high-carbohydrate diets in the cold. p values are shown only for significant differences (i.e., p < 0.05). PDFF, proton-density fat fraction; NEFA, non-esterified fatty acids; Ra, rate of appearance; RT, room temperature; scWAT, subcutaneous white adipose tissue; VAT, visceral adipose tissue; VLDL-TG, very-low-density lipoprotein-triglycerides. Participant 13 was excluded from the analyses due to antibiotic use (potential effects on the microbiome) before the fructose and glucose diets and is shown with a square symbol shape.

We also measured circulating very low-density lipoprotein-triglycerides (VLDL-TG), which are secreted by the liver. An increase in de novo lipogenesis due to excess glucose, and especially, fructose has been linked to increased liver fat content and circulating VLDL-TG., We did observe a trend toward elevated VLDL-TG with the carbohydrate-supplemented diets compared with isocaloric diet, but this increase was only significant during cold exposure (Figure 5D; fructose: p = 0.023; glucose: p = 0.013). Whether this is caused by increased production of VLDL-TG by the liver or by reduced tissue clearance cannot be determined with the current data.

The possible link between changes in whole-body lipolysis and lipogenesis in the liver and BAT metabolism requires further investigation. There was no correlation between baseline liver triglyceride content, or cold-induced changes in BAT triglyceride content, and BAT glucose uptake or oxidative metabolism. However, if there was an increase in VLDL-TG secretion by the liver, it could serve as an alternative energy substrate for BAT instead of glucose. Indeed, we have previously shown a direct competition between the uptake of triglyceride-rich lipoprotein-derived fatty acids and circulating glucose, with the former reducing BAT glucose uptake by more than 50% in the postprandial state.

Discussion

Over the past decades, there has been a significant shift in our food supply system that was accompanied by the addition of food additives such as fructose. These changes have paralleled the rise in obesity and its associated metabolic complications. This study aimed to examine whether 2-week overconsumption of fructose could modify the gut microbiome and the production of SCFA, especially acetate, and elicit changes in BAT metabolism. None of the diets, supplemented with either glucose or fructose for 2 weeks, had any effect on body composition, BAT composition, the gastrointestinal microbiome, or fecal SCFA levels. Although fructose overconsumption did not change cold-induced BAT oxidative metabolism or the depletion of intracellular triglycerides, it significantly reduced BAT glucose uptake. Interestingly, this BAT metabolic disturbance occurred without any significant changes in tissue blood flow or systemic insulin resistance. Our findings demonstrate that reduced BAT glucose uptake occurs very early after the onset of a high-fructose diet in young and otherwise healthy participants.

We found BAT glucose metabolism to be very sensitive to fructose overconsumption, with BAT metabolic rate of glucose decreasing by 43% (38% decrease in fractional glucose extraction) within 2 weeks of fructose overfeeding. The exact cause of this reduction in BAT glucose uptake remains unclear. Increased caloric intake alone is unlikely to have caused this BAT metabolic abnormality, as consuming glucose did not change BAT glucose uptake. Neither the glucose nor the fructose-supplemented diet affected whole-body insulin sensitivity based on the HOMA-IR, suggesting that the decreased uptake of glucose by BAT precedes whole-body insulin resistance during the development of fructose-induced metabolic dysfunction. BAT samples were not taken during this study, making it impossible to investigate changes in cell morphology or function, which may explain the decrease in glucose uptake.

The combined reduction of BAT glucose uptake with preserved BAT oxidative metabolism and intracellular triglyceride mobilization during cold exposure observed with fructose overfeeding in the healthy participants is reminiscent of our previous observation in cold-exposed individuals with T2D. Thus, 2 weeks of fructose overfeeding provides another example of the uncoupling of BAT oxidative metabolism and glucose uptake.,, Our results highlight once again the limitations of 18F-FDG PET as a marker of BAT thermogenesis. Based on these results, the possible effect of diets should be considered in future studies assessing BAT function with 18F-FDG.

Since we do not currently have a reliable non-invasive method of measuring active BAT volume in humans independent of 18F-FDG PET, we cannot determine if the impairment in glucose metabolism is accompanied by a reduction in BAT mass. For this reason, changes in the contribution of BAT to whole-body thermogenesis and glucose clearance cannot be assessed. However, we did not observe an effect of fructose overfeeding on whole-body energy expenditure, suggesting that, if there was a decrease in BAT activity, it was compensated by other thermogenic processes or it was too small to measure.

Finally, in parallel with the impairment of BAT glucose metabolism by fructose, we observed early signs of hepatic dysmetabolism with the carbohydrate-supplemented diets, and especially with high fructose. These early changes included increased lipid storage and circulating VLDL-TG levels, both early characteristic features in the development of fructose-induced fatty liver disease. Interestingly, these effects only emerged under cold stress, as the high-glucose or -fructose diet did not affect hepatic lipid content or VLDL-TG levels at room temperature. The high adipose tissue intracellular triglyceride lipolytic rate stimulated in response to cold exposure resulting in increased circulating NEFA flux likely exacerbated the impact of increased hepatic de novo lipogenesis, and possibly reduced fatty acid oxidation that may have been induced by the high-fructose diet., It is also possible that the changes in hepatic lipid content are linked to modifications of bile acids composition in the gut or in the circulation., Quantification of bile acids should be part of future studies to investigate this mechanism and its possible interactions with BAT thermogenesis.

The possible link between liver and BAT impairment by fructose warrants further investigation. Notably, the increase in circulating VLDL-TG may be involved in a shift in BAT energy substrate consumption from glucose to lipids. However, this shift was not observed with the high-glucose diet despite a similar increase in circulating VLDL-TG. Therefore, if a change of energy substrate occurred, it must have been driven by other factors than just an increase in circulating lipids. Possible mechanisms include early development of insulin resistance in BAT and alterations in adrenergically mediated synthesis or translocation of the glucose transporter, GLUT1. Although peripheral blood fructose levels are generally low, a direct effect of fructose on brown adipocytes is also possible, for example, via an increase in glucose 6-phosphate production, which could limit 18F-FDG uptake through feedback inhibition on hexokinase. Further studies on BAT tissue samples and cell lines will be required to elucidate these mechanisms. Despite its apparent importance in fueling BAT thermogenesis in rodents,59, 60, 61 the role of BAT lipid uptake to fuel thermogenesis in humans appears to be less prominent.,,, More studies are needed to document the role of change in BAT lipid metabolism in the development of metabolic dysfunction in humans.

Limitations of the study

The small sample size of our cohort is the main limitation of this study. The number of participants was voluntarily limited due to time constraints and the high cost of imaging studies. Recruitment was also difficult because of the demands placed on participants (e.g., following diets and undergoing three long and complex imaging visits). For these reasons, of the 16 participants recruited, only 10 participants completed the study. The initial sample size calculation (see STAR Methods) assumed larger changes in oxidative metabolism than those observed in this study. Therefore, if there were small effects of the diets on the microbiome or other outcomes, they could not be detected. Despite the small number of participants, we nevertheless demonstrated the reduction in BAT glucose uptake with fructose overfeeding without any detectable trend toward a reduction in BAT thermogenic activity. The exclusion of one subject who underwent antibiotic treatment during his participation in the study (participant 13) from the analyses did not change the results of this study.

Compliance with the diets is another possible limitation of this study. Participants were not required to fill out daily food diaries, and we thus relied on participants maintaining the same baseline diet throughout the study. However, some participants reported a loss of appetite because of the large quantities of added fructose or glucose or the large volume of beverage supplement consumed (1–1.5 L divided over three separate meals). This may have resulted in the consumption of smaller meals during fructose and glucose overfeeding, thus limiting the intended increase in caloric intake. Also, the supplementation with high quantities of glucose or fructose was designed to maximize potential effects on the microbiome, BAT, and whole-body metabolism during a short 2-week period. This is not representative of normal dietary habits in which smaller quantities, of added sugar (generally a combination of glucose and fructose) are consumed over long periods of time.

However, this study also includes several important strengths, including the crossover design that controlled for inter-individual variation in baseline characteristics and possible lifestyle differences, the use of a 2-week isocaloric control condition performed before the fructose/glucose conditions, and the direct comparison between the overconsumption of fructose and glucose.

Conclusion

Two-week fructose overfeeding did not elicit detectable changes in the microbiome or BAT thermogenesis, but impaired BAT glucose uptake, promoted liver fat storage, and increased circulating VLDL-TG levels during cold exposure in healthy young men. Although the cause and clinical significance of this reduction in BAT glucose metabolism are unclear, our study rules out the potential role of changes in the gut microbiome or the production of SCFAs. It also reveals that reduction in BAT glucose uptake is a very early feature of fructose-induced metabolic disturbance in humans. Its early occurrence in young and otherwise healthy participants makes it a potential confounding factor for studies relying on 18F-FDG to measure changes in BAT thermogenic activity.

STAR★Methods

Key resources table

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, André Carpentier (andre.carpentier@usherbrooke.ca).

Materials availability

This study did not generate unique reagents.

Experimental model and subject details

Human subjects

The protocol received approval from the Human Ethics Committee of the Centre de Recherche du Centre hospitalier universitaire de Sherbrooke. Written informed consent was obtained from all participants in accordance with the Declaration of Helsinki. The study was prospectively registered on ClinicalTrials.gov (NCT03188835) on June 13, 2017. A total of 16 male participants were recruited for this study (Figure S1) with the first participant enrolling on July 6, 2017. However, 5 withdrew before undertaking any experimental procedures. Of the remaining 11 participants, 1 withdrew following the baseline imaging protocol and 1 was excluded from analyses of carbohydrate-enriched diets (but not baseline) due to antibiotic use potentially affecting the microbiome. Metabolic data for this last participant were still analyzed and shown on graphs using a different symbol shape. Characteristics of participants can be found in Table S1. The inclusion criteria were as follows: (i) male; (ii) age between 20 and 40 years-old; (iii) BMI <30 kg/m2; (iv) normal glucose tolerance (blood glucose <7.8 mmol/L 2 h post 75 g oral glucose tolerance test); (v) no first-degree familial history of type 2 diabetes (parent or sibling). Exclusion criteria comprised: (i) fasting plasma triglycerides >5.0 mmol/L or history of total cholesterol >7 mmol/L; (ii) consuming more than 2 alcoholic beverages per day; (iii) smoking more than 1 cigarette per day; (iv) history of total cholesterol level >7 mmol/L, of cardiovascular disease or hypertensive crisis; (v) treatment with fibrates, thiazolidinedione, insulin, beta-blocker or other drugs with effects on insulin resistance or lipid metabolism (exception for anti-hypertensive drugs, statins or metformin); (vi) presence of a non-controlled thyroid disease, renal or hepatic disease, history of pancreatitis, bleeding disorder, cardiovascular disease or any other serious medical condition; (vii) history of serious gastrointestinal disorders (malabsorption, peptic ulcer, gastroesophageal reflux having required a surgery, etc.); (viii) contraindication to MR imaging (e.g., metal implant) or PET/CT imaging (e.g., having undergone a scan in the last 12 months); (ix) chronic administration of any other medication.

Method details

Study design

Figure S2 summarizes the study protocols. Briefly, participants were asked to follow three 14-day dietary protocols. The first protocol (A) consisted of a balanced diet for baseline measures. The other two protocols consisted of the same baseline diet supplemented with a high fructose (B) or glucose (C) drink, each representing 25% of energy intake. Protocols B and C were randomized and separated by a 1-month washout period during which participants followed the baseline diet. After each diet protocol, participants provided stool samples for microbiome analysis and underwent a metabolic study day with cold exposure to stimulate BAT thermogenesis. During the metabolic study days, body composition was assessed by dual-energy X-ray absorptiometry (DXA) and magnetic resonance (MR). Whole-body energy expenditure and substrate metabolism were measured by indirect calorimetry and stable isotope tracers. Tissue-specific glucose uptake was assessed by 18F-FDG PET, and oxidative metabolism was assessed by 11C-acetate PET. Participants were fasted for 12 h and abstained from strenuous physical activity for 48 h prior to metabolic study days.

Measures of body composition and energy expenditure

Body composition was assessed by DXA (Hologic) and bioelectrical impedance (Tanita). Whole-body energy expenditure at rest was determined by indirect respiratory calorimetry with correction for protein oxidation for a duration of 20 min per measure. These measures were taken on visits 2, 3, 4, 5 and 6. On imaging visits (visits 3, 4 and 6), the indirect calorimetry (SensorMedics Vmax) was performed once at room temperature (before cooling) and twice during cold exposure (the reported cold exposure data is the average of both measures). The molar rate of fatty acid oxidation was determined by converting the rate of triglyceride oxidation (mg/min) to its molar equivalent (μmol/min), assuming the average molecular weight of human triglyceride is 860 g/mol and multiplying the molar rate of triglyceride oxidation by three, as each mole of triglyceride contains three moles of fatty acids. Energy potentials of 16.3, 40.8 and 19.7 kJ/g were used to calculate the amount of heat produced from glucose, lipid, and protein oxidation, respectively, and converted to a rate in kJ·min−1. The level of physical activity of each participant was assessed with a uniaxial accelerometer (Caltrac) on 3 consecutive days during each diet protocol (before visits 3, 4 and 6).

Diets

The baseline, isocaloric, meal plans were designed by a registered dietician (S.D.) following the recommendations of the 2007 Canada’s Food Guide (∼50% carbohydrates, ∼30% lipids and ∼20% proteins) as well as the dietary preferences and restrictions of each participant. Daily caloric needs were estimated based on resting metabolism (indirect calorimetry) and physical activity (uniaxial accelerometer). Alcohol and dietary supplements likely to affect the gastrointestinal microbiome were prohibited. Participants were asked to refrain from changing their dietary or exercise habits during the study. However, they were not asked to fill out food diaries or otherwise supervised to ensure compliance.

The fructose (protocol B) or glucose (protocol C) supplements were provided in powder form (Sigma-Aldrich) that participants dissolved in 1–1.5 L of water and drank throughout the day. Addition of pure lemon or orange juice (no artificial sweeteners) was suggested to make the drink more palatable. To improve tolerance, the amount of glucose or fructose consumed was gradually increased during the first 3 days of the study (15%, 20 and 25% of estimated daily energy requirements) and remained stable for the remaining 11 days (174 ± 37 g/day, equivalent to 655 ± 138 kcal/day).

Finally, participants fasted (water allowed) from 20 h 00 on the evening before the metabolic study until the end of procedures at approximately 16 h 30 the following day.

Stool sample collection and analysis

Stool samples were provided by participants after each diet protocol. Samples were kept frozen at −80°C until processed. Before analysis, samples were thawed either at 4°C (faecal short-chain fatty acid content) or room temperature (microbiome composition), and remaining portions were refrozen. Stool samples underwent analysis for SCFA (acetate, propionate, and butyrate) at Université de Sherbrooke, as well as microbiome composition at McMaster University.

Faecal SCFA content was measured by gas chromatography-tandem mass spectrometry (GC-MS/MS, Agilent GC model 7890A and Waters Quattro Micro) based on methods developed for blood samples., The column was an Agilent DB-5ms Ultra Inert (30 m, 0.25 mm, 0.25 μm) with helium as the carrier gas. Each sample was analyzed in triplicate. All samples from a same participant were analyzed at the same time to limit variability. Approximately 150 mg of sample was transferred to a 2 mL polypropylene microtube with glass beads and weighted to obtain wet weight. Then, 1 mL of 0.2% HCL was added, and the sample was shaken on ice for 20 min before centrifugation (9400 g, 5 min). The supernatant was transferred to a 2 mL polypropylene tube and recentrifuged (9400 g, 2 min). The pellet was conserved for estimation of dry weight. For sample analysis, 20 μL of diluted internal standard (acetate-d3 25 mM, propionate-d5 5 mM or butyrate-d5 5 mM) was added to a 2 mL crimp top vial, followed by 250 μL of 0.4% HCL, 180 μL of deionised water, 50 μL of the supernatant, and 1 mL of dichloromethane. The tube was closed with the crimp cap and mixed with a shaker (10 min, room temperature) before being centrifuged upside down (800 g, 5 min). Approximately 600 μL of the dichloromethane phase was drawn from the upside-down vial with a glass syringe (avoiding any water). The dichloromethane solution was transferred to a 2 mL glass vial containing 400 mg of dry sodium sulfate. The glass vial was closed and allowed to sit for 1 h at room temperature. Finally, 20 μL of the derivatization reagent (MTBSTFA +1% t-BDMCS reagent diluted 1:100 in dichloromethane) was added to a 2 mL screw-tread vial with a 200 μL, 6 × 29 mm glass insert (mandrel precision point, 33 glass, with plastic spring; Chromatographic Specialties), followed by 200 μL of sample extract (dichloromethane phase). The sample was then injected (3 μL) in the GC-MS/MS in split mode and using positive electron impact ionization for mass detection.

Genomic DNA was extracted as described previously with some modifications. Samples were transferred to screw cap tubes containing 2.8 mm ceramic beads, 0.1 mm glass beads, GES and sodium phosphate buffer. Samples were bead beat and centrifuged as described, and the supernatant was further processed using the MagMAX Express 96-Deep Well Magnetic Particle Processor from Applied Biosystems with the multi-sample kit (Life Technologies #4413022).

Purified DNA was used to amplify the v34 region of the 16S rRNA gene by PCR. 50 ng of DNA was used as template with 1U of Taq Polymerase, 1× Taq buffer, 1.5 mM MgCl2, 0.4 mg/mL BSA, 0.2 mM dNTPs, and 5 pmol each of 341F (CCTACGGGNGGCWGCAG) and 806R (GGACTACNVGGGTWTCTAAT) Illumina adapted primers, as described by Bartram et al. The reaction was carried out at 94°C for 5 min, 5 cycles of 94°C for 30 s, 47°C for 30 s, and 72°C for 40 s, followed by 25 cycles of 94°C for 30 s, 50°C for 30 s and 72°C for 40 s, with a final extension of 72°C for 10 min. Resulting PCR products were visualized on a 1.5% agarose gel. Positive amplicons were normalized using the SequalPrep normalization kit (ThermoFisher #A1051001) and sequenced on the Illumina MiSeq platform at the McMaster genomics facility.

Reads were processed using DADA2. First, Cutadapt was used to filter and trim adapter sequences and PCR primers from the raw reads with a minimum quality score of 30 and a minimum read length of 100 bp. Sequence variants were then resolved from the trimmed raw reads using DADA2, an accurate sample inference pipeline from 16S amplicon data. DNA sequence reads were filtered and trimmed based on the quality of the reads for each Illumina run separately, error rates were learned, and sequence variants were determined by DADA2. Sequence variant tables were merged to combine all information from separate Illumina runs. Bimeras were removed and taxonomy was assigned using the SILVA database version 1.3.8.

We filtered to all ASVs assigned to the kingdom bacteria and removed all ASVs assigned to the kingdom eukaryota, the order chloroplast and the family mitochondria. After filtering, 1,720,471 total reads (average: 61,445; range: 11,449–124,952) with 2517 total ASVs (average 199 ASVs/sample; range: 61–341) remained.

Please refer to the Quantification and statistical analysis section for details on the analysis methods.

BAT stimulation by cold exposure

On metabolic study days, subjects underwent a 3-h cooling protocol. After the morning MRI session, subjects, wearing only shorts, were instrumented with autonomous wireless temperature sensors (Thermochron iButton model DS1922H, Maxim) placed on 12 sites to measure mean skin temperature: forehead, upper back, lower back, abdomen, forearm, quadricep, hamstring, gastrocnemius, anterior tibialis, chest, bicep, and hand. Surface electromyography electrodes (Delsys, EMG System) were placed on the belly of eight muscles known to contribute significantly to shivering during cold exposure76, 77, 78, 79: m. trapezius superior, m. sternocleidomastoid, m. pectoralis major, m. deltoideus, m. biceps brachii, m. rectus abdominis, m. vastus lateralis, m rectus femoris, and m. vastus medialis. To normalize the shivering measures, subjects performed a series of isometric maximal voluntary contractions for each muscle. Finally, participants were asked to put on a liquid conditioned suit (Three Piece, Med-Eng).

Cooling was started ∼12 h 00 and maintained for 180 min. During this period, water at 18°C was perfused through tubing in the suit with a temperature- and flow-controlled circulation bath (Isotemp 6200R28, Fisher Scientific). Shivering and skin temperature were monitored continuously throughout the experiment. Shivering intensity and shivering pattern were determined as previously described.76, 77, 78, Mean skin temperature was calculated using an area-weighted mean from the wireless temperature sensors.

Energy substrate turnover

On metabolic study days, after the morning MRI, participants were asked to empty their bladders and indwelling catheters (18G) were placed in the antecubital vein of both arms for blood sampling (right arm) and tracer infusion (left arm). A primed (3.3 × 106 dpm) continuous infusion (0.56 × 106 dpm min−1) of [3-3H]-glucose was initiated 150 min before the start of cold exposure to determine the plasma glucose appearance rate (Raglucose)., A bolus of [1-13C]-NaHCO3 (1.2 μmol kg−1), followed by continuous infusion of [U-13C]-palmitate (0.01 μmol kg−1·min−1 in 100 mL 25% human serum albumin), and a primed (1.6 μmol kg−1) continuous infusion (0.1 μmol kg−1·min−1) of [1,1,2,3,3-2H]-glycerol were started 60 min before the start of cold exposure to determine plasma NEFA appearance rate (RaNEFA) and plasma glycerol appearance rate (Raglycerol).

RaNEFA was calculated by multiplying the plasma palmitate appearance rate by the fractional contribution of palmitate to total plasma NEFA concentrations. Blood samples were collected for dosing of tracer enrichment and other metabolites before starting the infusion, before cooling, and every hour thereafter. Samples of expired air were also taken every hour to measure the enrichment in 13CO2. Urine was collected at the end of the baseline period, 75 min into cold exposure and at the end of cold exposure to measure urea production. Rates of appearance for glucose, non-esterified fatty acids, and glycerol were reported for room temperature (t = 0, just before cold exposure) and the steady state of the cooling period (t = 120 and 180 min).

Imaging of triglyceride content by magnetic resonance imaging

On metabolic study days, MRI and MRS were performed in the morning (before cooling, at approximately 7:30) and evening (after cooling, at approximately 16:00) on a 3T Philips Ingenia scanner (Philips Healthcare). On both occasions, triglyceride content was assessed in the supraclavicular region and in the abdominal region.

Participants, wearing only the standard hospital gown, were placed on a dedicated table (Philips Healthcare) compatible with the MR and PET scanners. They were immobilized with a vacuum cushion (Vac-Lok; Civco Radiotherapy) to ensure consistent positioning throughout the day. In the morning, participants were wrapped in a warm blanket to ensure participants were not cold.

Images were acquired with an 8-channel torso surface coil (dStream; Philips) covering the neck, torso, and abdomen. Tables S2 and S3 summarize the MRI and MRS sequence parameters. Briefly, an anatomical T1-weighted image of the supraclavicular region was taken to localize BAT. Three 8 mm3 voxels were placed in the supraclavicular fat for BAT proton spectroscopy. Voxel placement aimed at avoiding blood vessels and surrounding muscles while covering the largest volume possible (Figure S3). Since supraclavicular fat depots are bilateral, at least one voxel was placed on each side. To quantify fat content, mDixon Quant images were acquired covering the neck and shoulders. Similar sequences were performed for the liver region, except that a single 25 mm3 voxel was placed in the right lobe of the liver, and a single 10 mm3 voxel was placed in the subcutaneous fat of the lower back. Liver spectroscopy, Dixon imaging, and T1-weighted image of the liver were performed in breath hold (inspiration), other sequences were performed with free breathing.

Imaging of glucose metabolism and oxidative metabolism by PET/CT

A dynamic 11C-acetate PET scan (Philips Gemini TF) was performed at room temperature at approximately 11:30 only for protocol A (baseline). For all protocols, a dynamic 11C-acetate scan was performed at approximately 13:30 after 90 min of cooling. Each scan started with the injection of 11C-acetate (∼185 MBq, over 20 s with a 20 s saline flush) and was performed as a 30-min list-mode acquisition. A dynamic 18F-FDG scan was performed at approximately 14:00 with a 2-min background, injection of the radiotracer (∼125 MBq, over 20 s with a 20 s normal saline flush) and 30-min list-mode acquisition. Finally, a whole-body (head to knee) static PET scan was performed at approximately 15:00, after the 3-h cooling period. This static scan measured 18F-FDG distribution in different tissues without additional radiotracer injection.

Low-dose CT scans were obtained for anatomical reference and attenuation correction. A CT of the cervicothoracic region (120 kVp, 49 mA, 30 mAs) was done before each dynamic 11C-acetate scan, and a whole-body CT (120 kVp, 30 mA, 16 mAs) was done before the whole-body 18F-FDG scan. CT acquisition was not needed for the dynamic 18F-FDG scan because the participant was in the same position as for the previous 11C-acetate scan.

Image processing

MR images (Figure S3) were reconstructed online. Dynamic PET images (Figure S4) were reconstructed with a 3D RAMLA algorithm. Time frames for 11C-acetate were: 24 × 10 s, 12 × 30 s, and 4 × 300 s. Time frames for 18F-FDG were: 1 × 2 min (background), 12 × 10 s, 10 × 45 s, 7 × 90 s, and 2 × 300 s. The static 18F-FDG image was reconstructed with a BLOB-OS-TF algorithm. CT reconstruction was performed with filtered back projection. Standard corrections for decay, attenuation (CT-based), scattering (single scatter simulation) and random coincidences (delayed window) were performed for PET. PET data was expressed in units of standardized uptake value ().

Regions of interest were segmented manually in PMOD (v. 3.7, Bruker). Axial CT images were used as anatomical references for PET segmentation. Axial T1-weighted images were used as anatomical reference for MRI segmentation. PET and MR images were also registered and segmented together, but this method proved less reliable due to small registration errors, and results were not presented in this article. The BAT region included all fat visible in the neck and supraclavicular fossa on both sides. Other BAT depots (e.g., paraspinal) were not segmented because they are not visible for all participants. Other segmented tissues include muscles (m. sternocleidomastoideus, m. trapezius, m. deltoideus, and m. pectoralis major), subcutaneous white adipose tissue (lower back for MRI and cervical for PET), abdominal visceral adipose tissue (excluding perirenal adipose tissue), and liver. For pharmacokinetic modeling, the arterial input function was extracted from a spherical region (diameter: 10 mm) in the aortic arch.

For static images, such as proton-density fat fraction maps from Dixon MRI, CT radiodensities, and static 18F-FDG PET, average values were calculated for each region of interest. For dynamic PET imaging, time-activity curves were extracted for each region of interest. The blood signal for 11C-acetate was then corrected to exclude the contribution of metabolites.

Pharmacokinetic modeling was performed in MATLAB (The MathWorks, R2019a). A four-compartment, two-tissue, model was applied to the 11C-acetate signal to derive the rates of uptake (K1 [ml·g−1·min−1]), oxidation (k2 [min−1]) and lipid synthesis (k3 [min−1]). For 11C-acetate under cold condition, the blood volume fraction was constrained based on data from the 18F-FDG three-compartment model. Fractional glucose uptake (Ki [min−1]) was also obtained from Patlak graphical analysis and converted to metabolic rate of glucose using the plasma glucose level (MRGlu = Ki·glycemia·LC−1·100 [μmol·100 g−1 min−1]). This conversion assumes a lump constant (LC) value for 18F-FDG compared with endogenous plasma glucose which was set at 1.14 for fat and 1.16 for muscles.,

Finally, MR spectra were processed in jMRUI and quantified with AMARES. For BAT and scWAT, 6 gaussian peaks were considered: 0.9 ppm methyl, 1.3 ppm methylene/β-carboxyl, 2.1 ppm α-olefin/α-carboxyl, 4.2 ppm glycerol, 4.7 ppm water, and 5.3 ppm olefin/glycerol. For the liver, only the water and methylene peaks were visible. Correction for T1 and T2 decay was performed for each resonance. For T1, data from the literature was used., For T2, correction was based on the multi-echo data for the liver and on the literature for BAT and scWAT., Voxels with poor shimming or excessive signal contamination by other tissues were rejected. Values obtained for the different BAT voxels in a same participant and condition were averaged. The number of double bonds and methylene-interrupted double bonds were calculated for BAT and scWAT based on the method by Hamilton et al.

Biological assays

Individual plasma NEFA (palmitate, linoleate, oleate) [U-13C]-palmitate enrichment and [1,1,2,3,3-2H]-glycerol enrichment were measured by GC-MS/MS (Agilent GC model 7890A and Waters Quattro Micro). The column was an Agilent DB-5ms Ultra Inert (30 m, 0.25 mm, 0.25 μm) with helium as the carrier gas. The precipitation method was based on that of Patterson et al with some modifications. An aliquot of plasma (40 μL) was mixed with internal standard of 13C18-linoeate (20 μL of 50 μM in acetone) and 13C-glycerol (20 μL of 500 μM in water). Then, acetone (500 μL) was added, and the mixture was cooled at −20°C to allow protein precipitation. After centrifugation, (3500 g, 10 min), the supernatant was extracted, then mixed with deionised water (500 μL) and hexane (500 μL). After another centrifugation (3500 g, 10 min), the two phases were separated, and dried under vacuum. The organic phase residue, which contains free fatty acids was derivatized with a solution of MTBSTFA +1% t-BDMCS (8.2 mg, 34 μmol) and imidazole (54 ng, 0.8 nmol) in DMF. Aqueous phase residue for glycerol quantification was derivatized with a solution of MTBSTFA +1% t-BDMCS (20.7 mg, 86 μmol) and imidazole (136 ng, 2 nmol) in DMF (40 μL). After incubation (80°C, 1 h), samples were cooled down and then injected (1 μL) in the GC-MS/MS in split mode and using positive electron impact ionization for mass detection.

Tritium from [1-3H]-glucose enrichment was detected by liquid scintillation spectrometry (Beckman Coulter LS 6500).

Plasma glucose, total NEFA triglycerides, cortisol, TSH, free T3 and free T4 were measured using specific radioimmunoassays and colorimetric assays. Plasma C-peptide, GIP, total GLP-1, glucagon, insulin, and leptin were measured using Luminex xMAP-based immunoassays (Millipore).

Quantification and statistical analysis

Sample size calculation

The sample size was based on a decrease of 50% in BAT oxidative metabolism from previously observed values (K = 1.8 ± 0.5 min−1). For a paired t-test with an α of 0.05 and a power of 80%, the required sample size was 9. This level of oxidative metabolism suppression seemed realistic based on previous data from pharmacological inhibition of BAT activity. We planned to recruit 15 participants to account for potential incomplete data sets or losses to follow-up.

Statistical analysis

All microbiome analyses were conducted in RStudio (version 1.2.5042; April 2020) with R (version 4.0.0; April 2020) using the microViz, tidyverse, phyloseq, vegan, lme4, and emmeans packages for analysis, and ggplot2, ggpubr, and viridis for visualization. For α-diversity calculations, samples were rarefied to 11,449 sequence reads per sample. Differences in α-diversity between our metabolic measures of interest were assessed using generalized mixed-effects model analysis with participants used as the random effect variable and diet and sequence used as fixed effects. To measure β-diversity, samples were normalized by centred log-ratio transformation. Differences in β-diversity were assessed using permutational multivariate analysis of variance (PERMANOVA) assessing for differences by diet.

Metabolic data (including imaging data, blood chemistry, indirect calorimetry, and stable isotopes) are expressed as mean with 95% confidence interval or median with interquartile range. Analyses were performed in Prism (GraphPad; 9.0.2). Normality of data was assessed by the Shapiro-Wilk test. For normally distributed data, the effects of diets and cold on whole-body energy expenditure and substrate turnover as well as specific tissue composition and metabolism were assessed by a repeated measures ANOVA (if no missing values) or mixed-effects model (if missing values) with Bonferroni correction for multiple comparisons. Otherwise, for data that was not normally distributed, the Friedman test was used with Dunn’s correction for multiple comparisons. Correlation and linear regression analyses were applied to investigate the relationships between BAT metabolic parameters and other metabolic and hormonal cofactors. The significance threshold was set at p < 0.05.

Additional resources

ClinicalTrials.gov identifier NCT03188835.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
Human plasma samples	This paper	N/A
Human stool samples	This paper	N/A
Human urine sample	This paper	N/A
Human plasma control	BioIVT	#HMPLEDTA2
Chemicals, peptides, and recombinant proteins
Zinc Sulfate Solution	Z2876	Sigma Aldrich
Barium Sulfate Solution	B4059	Sigma Aldrich
Auto kit Glucose	Wako Diagnostics U.S.A. Corporation	997-03001
NEFA	Wako Diagnostics U.S.A. Corporation	995-34791999-34691993-35191991-34891
Triglyceride	Wako Diagnostics U.S.A. Corporation	461-08992461-09092
C-peptide, GIP, GLP-1 total, Glucagon, Insulin, Leptin, PYY	Millipore-Sigma	HMH3-34K-07
Human Ghrelin total	Millipore-Sigma	EZGRT-89K
Acylated ghrelin human ELISA	bertin pharma	A05106-96
HMW Total adiponectin	ALPCO	80-ADPHU-E01
Taq DNA Polymerase, Recombinant	Life Technologies	10342020
Critical commercial assays
MagMAX™-96 DNA Multi-Sample Kit	Invitrogen	4413021
SequalPrep normalization kit	ThermoFisher	A1051001
Deposited data
Raw and analyzed data	This paper	Zenodo: https://doi.org/10.5281/zenodo.5834789
Pharmacokinetic modeling tool for MATLAB	This paper	Zenodo: https://doi.org/10.5281/zenodo.5834789
16S rRNA v34 Sequencing Data	This paper	BioProject ID: PRJNA793467 (www.ncbi.nlm.nih.gov/bioproject/PRJNA793467)
Oligonucleotides
dNTPs Solution, Mix	New England Biolabs	N0447L
341F (CCTACGGGNGGCWGCAG)	Bartram et al. (https://doi.org/10.1128/AEM.02772-10)
806R (GGACTACNVGGGTWTCTAAT)	Bartram et al. (https://doi.org/10.1128/AEM.02772-10)
Software and algorithms
GraphPad Prism 9 for statistical analysis	GraphPad Software	RRID: SCR_002798
MassLynx V.4.1	Waters	SCN714
PMOD 3.7	PMOD Technologies LLC	RRID: SCR_016547
Adobe Photoshop	Adobe	RRID: SCR_014199
GNU Image Manipulation Program 2.10	www.gimp.org	RRID: SCR_003182
MATLAB R2019a	The MathWorks	RRID: SCR_001622
DADA2 Version 1.14	Callahan et al. (https://doi.org/10.1038/nmeth.3869)
Cutadapt Version 1.18	Marcel Martin (https://doi.org/10.14806/ej.17.1.200)	RRID: SCR_011841
SILVA database Version 1.3.8	Quast et al. (https://doi.org/10.1093/nar/gks1219)Yilmaz et al. (https://doi.org/10.1093/nar/gkt1209)	RRID: SCR_006423
microViz Version 0.7.3	Barnett et al. (https://doi.org/10.5281/zenodo.4644057)
tidyverse Version 1.3.1	Wickham et al. (https://doi.org/10.21105/joss.01686)	RRID: SCR_019186
phyloseq Version 1.32.0	McMurdie and Holmes (https://doi.org/10.1371/journal.pone.0061217)	RRID: SCR_013080
vegan Version 2.5–7	Oksanen et al. (URL: https://cran.r-project.org/web/packages/vegan/index.html)	RRID: SCR_011950
ggplot2 Version 3.3.3	Wickham (URL: https://ggplot2.tidyverse.org)	RRID: SCR_014601
ggpubr Version 0.4.0	Kassambara (URL: https://cran.r-project.org/web/packages/ggpubr/index.html)	RRID: SCR_021139
viridis Version 0.6.0	Garnier et al. (https://doi.org/10.5281/zenodo.4679424)	RRID: SCR_016696
lme4 Version 1.1–27	Bates et al. (https://doi.org/10.18637/jss.v067.i01)	RRID: SCR_015654
emmeans Version 1.6.1	Lenth (URL: https://CRAN.R-project.org/package=emmeans)	RRID: SCR_018734
RStudio Version 1.2.5042	RStudio Team (URL: http://www.rstudio.com/)	RRID: SCR_000432
R Version 4.0.0	R Core Team (URL: https://www.R-project.org)	RRID: SCR_001905
Other
[11C]acetate, radiotracer	CIMS	N/A
[18F]FDG, radiotracer	CIMS	N/A
[3-3H]glucose, stable isotope tracer	PerkinElmer	Cat# NET331A005MC
[1,1,2,3,3-2H]glycerol, stable isotope tracer	Cambridge Isotopes	Cat# DLM-1229
[U-13C]palmitate, stable isotope tracer	Cambridge Isotopes	Cat# CLM-3934
MTBSTFA + 1% t-BDMSC	Millipore-Sigma	M-108
13C- glycerol	Cambridge Isotope Laboratories	CLM-1397
U13C-Linoleic Acid potassium salt	Cambridge Isotope Laboratories	CLM-8835
Palmitic Acid	Millipore-Sigma	P5585
Oleic Acid	Millipore-Sigma	O1008
Linoleic Acid	Millipore-Sigma	L1376
Sodium butyrate	Millipore-Sigma	B5887
Sodium propionate	Millipore-Sigma	P1880
Sodium acetate	Millipore-Sigma	S2889
GC-MSMS	Waters	Quattro Micro-GC
Gemini TF PET-CT scanner	Philips	N/A
Ingenia 3.0T MR system	Philips	N/A
Liquid-conditioned suit	Med-Eng	Cat# 1004340
Isotemp 6200R28 water bath	Fisher Scientific	Cat# 13-874-676
Vmax 29n (indirect calorimetry)	SensorMedics	N/A
Thermochron iButton	Maxim	Cat # DS1922H
Myomonitor IV EMG system	Delsys	N/A
Horizon QDR DXA	Hologic	N/A
Beta Counter LS 6500	Beckman Coulter	N/A
Optima L-100XP ultracentrifuge	Beckman Coulter	N/A
MagMAX Express 96-Deep Well Magnetic Particle Processor	Applied Biosystems	N/A
MiSeq System	Illumina	N/A