Introduction
Brown adipose tissue (BAT) is a thermogenic tissue which contributes to energy homeostasis in human adults. Upon cold exposure, it is activated by the sympathetic nervous system (SNS) and converts chemical energy stored as lipids within the adipocytes directly into heat [
1]. Brown adipocytes differ significantly from white adipocytes: they contain a high amount of mitochondria and the intracellular triglycerides are stored in multiple small lipid droplets allowing for rapid lipolysis. The mitochondria in brown adipocytes express high levels of uncoupling protein 1 (UCP1) which is unique to this cell type [
2]. When activated by fatty acids, UCP1 allows protons to flow across the inner mitochondrial membrane along the proton gradient which has been built up by the respiratory chain. This is partly uncoupled from ATP synthase. As the proton-motive force driving the ATP synthase is reduced higher amounts of ADP accumulate in turn activating the citric-acid cycle and the respiratory chain. Thus, the activation of UCP1 reduces the efficiency of oxidative phosphorylation and energy stored in the mitochondrial proton-gradient is dissipated as heat [
3]. Activation of thermogenic adipocytes increases energy expenditure (EE) and facilitates uptake of glucose and lipids into the tissue [
4]. Therefore, BAT is an appealing potential therapeutic target for treatment of obesity and associated metabolic diseases. Since the increase of EE results upon cold exposure, the raise in EE is called cold-induced thermogenesis (CIT). CIT can be determined by measuring the difference EE during warm and cold conditions using indirect calorimetry [
5]. In order to reliably evaluate interventions that target energy homeostasis and BAT activity, accurate techniques for quantification are needed. Brown adipocytes are found in the cervical, supraclavicular, axillary, and retroperitoneal region with the cervical and supraclavicular regions being the most predominant [
6]. In these adipose tissue depots, thermogenic adipocytes can emerge from white adipocytes in response to cold stimulation. This lineage of brown adipocytes is different from the classical BAT found in human newborns. Upon cold stimulation, these brown-like (brown in white, brite) adipocytes transdifferentiate from white adipose tissue (WAT) [
7]. This transitional process from white to brown is highly dynamic, since conversion can be reverted when cold stimulus disappears [
8]. This plasticity of the tissue can also be observed in humans in temperate climate zones in whom the amount of active BAT is considerably higher during the cold season [
9‐
11].
Currently, BAT activity is most accurately quantified by [
18F]fluoro-2-deoxy-D-glucose ([
18F]FDG FDG) positron emission tomography/computed tomography (PET/CT) as it allows to quantify and visualize and precisely localize metabolically active BAT. However, [
18F]FDG PET/CT has several limitations: since the radiolabeled tracer [
18F]FDG PET/CT is taken up preferentially by metabolically active tissue, BAT has to be activated prior to PET/CT scanning. Furthermore, it exposes individuals to ionizing radiation and is expensive [
12]. Magnetic resonance imaging (MRI) has been proposed as an alternative to PET/CT [
13]. The imaging properties differ significantly from those of WAT since BAT contains more mitochondria and therefore a higher amount of iron and a lower amount of fat [
14]. Therefore, the fat fraction (FF) of BAT is generally lower than in WAT. Recent MRI studies described an inverse correlation of the tissues FF and the metabolic activity in [
18F]FDG PET/CT [
13]. However, other studies could not find such correlations [
15‐
18].
In this study, we aimed to quantify the predictive value of MRI FF for BAT by comparing it to the current standard imaging method which is [18F]FDG PET/CT.
Discussion
In the present study, we evaluated whether measurement of the fat fraction (FF) within the supraclavicular fat depot (scAT) in vivo can reliably predict the metabolic activity of brown adipose tissue (BAT) as assessed by [
18F]FDG PET/CT. Within each individual, the FF differed significantly between the PET-negative and PET-positive area of the scAT which is in line with the notion that human BAT is highly heterogeneous, consisting of brown (thermogenic) and white adipocytes and that highly active BAT contains a higher amount of mitochondria and less fat [
8]. However, the predictive power of the FF in the scAT depots was relatively low, accounting only for 16 to 36% of the variance in BAT activity. Previous studies in this field had already evaluated the relation of BAT FF and its metabolic activity in smaller cohorts or as proof-of-principle studies. While some studies demonstrated a relatively good correlation between MRI determined FF and BAT activity, others did not find significant associations.
In a group of 13 healthy volunteers, Holstila et al. found a good association between glucose uptake and FF both during cold exposure and warm ambient conditions [
13]. Remarkably, the measured FF varied substantially among individuals and FF of WAT and BAT depots were even sometimes within the same range. FF, as measured in this study, accounted for approximately 40% in the variation of FDG uptake. In a similar study including ten healthy individuals from the same research group, FF was shown to predict BAT activity (
R2 = 0.41,
p = 0.04) [
24]. In a cohort of twelve healthy volunteers studied by [
18F]FDG PET/MRI, the authors found a good correlation between FF and BAT glucose uptake rate in cold-stimulated supraclavicular BAT (
R2 = 0.52) [
25]. A study in 20 healthy volunteers using cold exposure and capsinoid administration to activate the SNS assessed different imaging modalities and found a significant correlation (
R2 = 0.39
p = 0.012) between SUV
mean and FF [
26].
However, other research groups using a similar approach could not find a significant relation between MRI measured FF and BAT activity as assessed by [
18F]FDG PET/CT [
18]. Likewise, a larger study including 28 children and adolescents evaluated MRI parameter changes in respect to BAT activity measured by [
18F]FDG PET/CT [
27]. Using voxel-wise co-registration, they were not able to find any direct correlation between FF and BAT activity as assessed by [
18F]FDG-PET under different thermal conditions. Nevertheless, they could show that changes in FF from cold to re-warming correlated with BAT activity. A retrospective study analyzed MRI and [
18F]FDG-PET in 66 pediatric patients to determine whether MR imaging can reproducibly detect human BAT independent of its activation state. The authors did not find a statistically significant and sufficiently predictive correlation of SUV
mean with FF and hypothesized that this might be due to relative stable FF and highly variable SUV
mean values [
17].
Some of the differences in study outcome might be explained by different methodological approaches. First, the determination of FF by MRI is complex and multi-echo sequences might be more suitable than dual-echo sequences. This could theoretically explain the different outcomes from our cohorts 1 (dual-echo) and 2 (multi-echo). However, a recent study demonstrated similar results for dual-echo and multi-echo sequences [
25].
Second, controversy exists whether to use glucose uptake rate (GUR) and dynamic PET scanning versus standardized uptake values (SUV) and static PET scanning as primary PET/CT readout. One should take into consideration that BAT GUR is less sensitive to confounders such as body weight and meal intake [
12,
28]. However, in the setting of fasted individuals after a defined cold stimulus GUR and SUV
mean seem to correlate very well [
29].
Third, the segmentation approach can also influence the quantification of FF mainly due to partial volume effects that can significantly affect the FF. We tried to reduce this effect by shrinking the segmentation as a final step before quantification [
30]. Moreover, it remains an open question which part of the scAT depot should be segmented. Segmentation of the whole depot might underestimate the amount of thermogenically active BAT. On the other hand, segmentation based on the post hoc knowledge of the highest [
18F]FDG uptake is not an option if MRI-based techniques should substitute [
18F]FDG-PET for evaluation of BAT activity. The studies which could show a robust inverse correlation between BAT activity in PET/CT and FF in MRI often placed the FF measurements into the area of BAT with the highest glucose uptake [
13,
15,
24]. Another study found a significant correlation between FF and glucose uptake only in a subgroup with highly active BAT but failed to do so in individuals with lower GUR [
24].
Since we were interested in developing a predictive tool, we segmented the scAT depot in MRI without prior knowledge of PET results. While our approach resulted in a moderate correlation of FF and BAT activity, the coefficient of determination (
R2) was significantly lower. This is in line with other studies, which based their segmentation on the whole scAT depot [
17,
18,
27]. It should be noted, however, that Anderson et al. had a quite good predictive result using this strategy [
25].
Nevertheless, a more likely explanation for the relatively low predictability might be that the differences in scAT FF are not merely due to the amount of thermogenic adipocytes or the mitochondrial density of the tissue. Our results and data from previous studies [
13] indicate that the inter-individual variability of the FF among the participants is large, ranging from 577.6 to 798‰ in cohort 1 and 581.2 to 763.4‰ in cohort 2. Within the same cohort, we found an average absolute difference of 76‰ and 35‰ respectively of intensity in areas of high SUV activity compared to areas of lower SUV activity. This means that the intra-individual differences in FF between thermogenically active and inactive fat are smaller than the inter-individual range. This might be due to the fact that BAT FF content is affected by age [
31] and body weight [
27]. Additionally, the FF in subcutaneous thermogenically inactive adipose tissue is also variable and decreases during cold exposure indicating lipolysis [
24]. This is underscored by the fact that a recent study was not able to establish a threshold to reliably detect BAT by using FF in MRI [
32]. Also, another explanation could be that leaner and younger subjects who generally possess more active BAT have lower FF in their adipose tissue depots.
It should be noted that, [
18F]FDG PET/CT measures glucose uptake in to BAT and not thermogenic activity as such. The determination of oxygen consumption and blood flow in the tissue have been evaluated previously, but are technically much more challenging [
33]. From a physiological point of view, cold-induced thermogenesis (CIT) seems to be a more important variable to measure. BAT activity correlates with CIT but does not fully determine this phenomenon as probably muscle contributes substantially to CIT also [
34]. Another newly discovered mechanism adding to CIT might be a proton leak in tissues such as skeletal muscle. It is mediated by ATP/ADP-cotransporter, working as a parallel proton channel besides of UCP1 [
35]. Therefore, multiple determinants of SUV
mean might be necessary to be taken into consideration, e.g., CIT, age, BMI, and FF. Using multiple linear regression of these parameters in cohort 2 in which we measured CIT directly before PET/CT, we could reach an improved predictive value of
R2 = 0.58. We think this strategy of integrating FF, age, BMI, and CIT to predict BAT activity may be sufficiently accurate for the use in prospective cohort studies or as a screening tool.
Our study is limited in several aspects: first, we used static PET imaging. Several authors propose instead to measure GUR in dynamic PET scanning. This technique is less prone to errors due to [
18F]FDG PET/CT uptake into muscle [
34] rather than performing static PET and measure SUV
mean. Still, as measurements were only performed on fastened individuals, glucose uptake in muscle is minimized and SUV
mean correlated very well with BAT activity [
29]. Second, our study included only young healthy male participants, making the cohort homogenous but not reflecting the general population. Third, the BAT activation protocols differed in the fact that we used a β3 agonist in cohort 1 in addition to controlled cold exposure. Strengths of our study comprise the relatively large sample size of 32 participants, a systematic analysis of the imaging data which was unbiased by prior knowledge of the PET signal as well as the combined analysis of MRI FF and CIT.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.