View as PDF

Rosa T. Branca , Le Zhang, Warren S. Warren, Edward Auerbach, Arjun Khanna, Simone Degan, Kamil Ugurbil, Robert Maronpot

View as PDF

The recent discovery of active Brown Adipose Tissue (BAT) in adult humans has opened new avenues for obesity research and treatment, as reduced BAT activity seem to be implicated in human energy imbalance, diabetes, and hypertension. However, clinical applications are currently limited by the lack of non-invasive tools for measuring mass and function of this tissue in humans. Here we present a new magnetic resonance imaging method based on the normally invisible intermolecular multiple-quantum coherence 1H MR signal. This method, which doesn’t require special hardware modifications, can be used to overcome partial volume effect, the major limitation of MR-based approaches that are currently being investigated for the detection of BAT in humans. With this method we can exploit the characteristic cellular structure of BAT to selectively image it, even when (as in humans) it is intimately mixed with other tissues. We demonstrate and validate this method in mice using PET scans and histology. We compare this methodology with conventional 1H MR fat fraction methods. Finally, we investigate its feasibility for the detection of BAT in humans.


Obesity is rapidly spreading across most developed countries and is thought to be more harmful to health than alcohol or smoking because of its association with many other medical conditions

[1]. At a fundamental level, obesity is the result of an imbalance between energy intake and energy expenditure. The latter is very difficult to quantify and recent work suggests that it can be altered by the function of Brown Adipose Tissue (BAT) [2]. BAT [3], [4] is a type of fat that modulates both basal (cold exposure) and inducible (overeating related) energy expenditure in mammals, thereby affecting whole-body metabolism and modifying susceptibility to weight gain [5]–[8]. It is considered to be the “good fat” that, unlike the white “bad fat”, burns calories to produce heat through a process called non-shivering thermogenesis.

While in small animals malfunction of this tissue is known to cause obesity, in humans the role and the incidence of tissue is less clear. In fact until recently, BAT was thought to exist in humans only in infancy and early childhood.[9] However, combined 18F-fluorodeoxyglucose positron emission tomography (18F-FDG-PET) and computed tomography (CT) scans have identified active BAT in adults and shown a strong correlation between BAT activity and the basal metabolic rate [3], [10]–[12]. As it turned out, BAT was missed in adult humans because of its diffuse anatomical distribution: this tissue is present only in scattered amounts in the neck and chest areas, around major blood vessels [13], muscles [14] or white fat [15]. Nevertheless, it is estimated that BAT activity could account for up to 20% of daily energy expenditure in an adult human [16].

Although this tissue is a clear target for obesity treatments, the current modality of choice for imaging metabolically active BAT in humans, PET/CT, presents significant limitations. BAT metabolism relies on fatty acid consumption, not glucose consumption, so 18F-FDG-PET is highly nonspecific. Moreover, confounding factors such as blood glucose levels and room temperature conditions [11], [17], [18] may affect glucose uptake in BAT. Other, more specific, PET metabolic tracers have been used to estimate BAT oxidative capacity and fatty acid uptake [19] during thermogenic activity in humans. Still, radiation exposure from PET imaging, although considerably smaller than PET/CT imaging [20], precludes the repetitive BAT screening in healthy and young subjects needed to determine the physiological relevance of BAT in humans [21], [22].

MRI is very attractive for BAT studies since it is non-invasive, does not deliver mutagenic radiation, and has no limitations in imaging penetration depth. More interestingly, the difference in chemical shift between water protons and fat protons makes it possible to differentiate lean from fatty tissues as well as normal white fat (WAT) from brown fat. While brown fat is characterized by multilocular brown adipocytes with average water content of about 50%, white fat is characterized by unilocular adipocytes with water content of less than 10%. As a consequence, fat fraction measurements performed with 1H MR can be used to differentiate these two tissues, at least in rodents [23]–[27]. However, in adult humans this tissue is present only in scattered amounts, and partial volume effects caused by different types of cells in a single voxel (for example WAT, BAT, and muscle) lead to both false positive and false negative findings. Although, in principle, MR resolution could easily be increased to reduce partial volume effects, in practice this is unfeasible. For example, an isotropic increase in image resolution by a factor of two (i.e. a 8-fold reduction in pixel volume) results in a factor of 8 loss in SNR, which can be compensated only by a 64-fold increase in acquisition time.

In this paper we show that we can overcome these weaknesses of conventional MRI by using the non-linear MR signal generated by intermolecular zero-quantum coherences (iZQCs) that originates from closely spaced water and fat spins [28]. These coherences are generated by simultaneous and opposite transitions of water and fat spins separated by a user-controlled “correlation distance”, a distance defined as half the modulation created in the nuclear magnetization by the applied pulsed field gradients: dc = 1/(2γGT), typically 20–100 µm. A critical feature is that the correlation distance can be much finer than the image resolution, allowing us to probe structure at a much smaller scale [29], without loss of sensitivity [30]. More specifically, by choosing a correlation distance much smaller than the image resolution and comparable to the cellular size, we can suppress coherences between water and fat spins that reside in different tissues within an image voxel (WAT and muscle, for example), while retaining and enhancing coherences between water and fat spins that reside in the same cells or tissue (BAT). This is briefly outlined in figure 1.

Building upon our previous spectroscopy studies in rodents [28], where we found a strong association between the mass of brown adipose tissue and the intensity of the iZQC signal from closely spaced water and fat spins, here we demonstrate and validate, with both histology and18FDG-PET scanning, that this signal does indeed originate from brown adipose tissue and can therefore be used to specifically detect and map this tissue in rodents. Finally, we investigate the feasibility to use this signal, which we call BATSCI (BAT-Specific Coherence Imaging), to detect human BAT, using both in vitro and in vivo experiments.

Detection of Brown Adipose Tissue-fig1

Figure 1. Origin of the water-fat iZQC signal from BAT (BATSCI). (A–B) Histological haematoxylin and eosin (H&E) staining of BAT (A) and white fat (B) from a mouse showing different cell morphologies. Brown fat cells present multiple smaller lipid vacuoles and higher hydration level and are usually smaller than white fat cells, which are made by a single large lipid droplet. (C) Cartoon showing the different cellular structures and the origin of the iZQC signal in BAT: unlike in white fat, in BAT water and fat spins are mixed together at the cellular level such that the selection of a small correlation distance can select the BATSCI signal only from BAT (D) Scheme of the radio frequency pulse sequence used in the experiments for the detection of BAT. doi:10.1371/journal.pone.0074206.g001


Phantom Results

Figure 1.D shows the sequence used to acquire the BATSCI signal. In this sequence, as described in [28], the BATSCI signal coming from water-fat iZQCs evolves first as a water-fat double quantum coherence (iDQC) during the delay tau, as zero quantum coherence during the second evolution delay t1, and finally as single quantum coherence during the acquisition time. This specific coherence pathway is selected by a GT/2 GT gradient combination that, during the acquisition time, allows us to refocus only single quantum coherences that have evolved as double quantum coherences during the tau delay. The extra gradient pulse G’T, on the other hand, selects the evolution of the water-fat iZQC signal during the t1 delay and suppresses both homomolecular (water-water and fat-fat) coherences and heteromolecular water-fat iDQCs.

This sequence is used to collect 2D iZQC spectra as well as to map the BATSCI signal in both animals and humans. BATSCI maps are collected by a CSI type acquisition scheme. Briefly, a series of 2D iZQC images are acquired with a different iZQC evolution times, t1. A Fourier transform along the t1 dimension allows us to display the iZQC spectrum from which a BATSCI map can be obtained.

In figure 2 we compare the performance of this sequence to the more established MR fat fraction measurement method for the detection of excised BAT [23], [31]. While the fat-fraction measurement clearly differentiates samples made by 100% BAT from samples made by 100% WAT, it does not distinguish BAT from the WAT/muscle mixture. At the relative coarse resolution (0.72 mm3) used in this experiment, which is still higher than the resolution typically used for human brown fat MRI studies (~1 mm3), the BAT and the WAT/muscle mixture look exactly the same since they have a similar water/fat faction (2.B). On the other hand, the water-fat iZQC signal for these samples is very different (2.C). The water-fat iZQC signal was present only in the BAT sample, while it was close to the noise level in the WAT-muscle mixture, despite the similar water-fat content of these two samples and the much lower resolution used for this imaging experiment (8 mm3). This is because the intensity of the water-fat iZQC signal depends not only on the relative concentration of water and fat spins, but also on their relative distribution over a distance smaller or equal to the selected correlation distance. The correlation distance for these experiments was selected to be ~80 µm, much smaller than the nominal image r