Proton MR Spectroscopy Measurements of White and Brown Adipose Tissue in Healthy Humans: Relaxation Parameters and Unsaturated Fatty Acids

Abstract

Background

Activation of brown adipose tissue (BAT) in rodents increases lipolysis in white adipose tissue (WAT) and improves glucose tolerance. Adult humans can have metabolically active BAT. Implications for diabetes and obesity in humans require a better characterization of BAT in humans.

Purpose

To study fat depots with localized proton MR spectroscopy relaxometry and to identify differences between WAT and fluorine 18 fluorodeoxyglucose (FDG) PET/CT proven cold-activated BAT in humans.

Materials and Methods

Participants were consecutively enrolled in this prospective study (ClinicalTrials.gov identifiers: NCT01568671 and NCT01399385) from August 2016 to May 2019. Supraclavicular potential BAT regions were localized with MRI. Proton densities, T1, and T2 were measured with localized MR spectroscopy in potential BAT and in subcutaneous WAT. FDG PET/CT after cold stimulation was used to retrospectively identify active supraclavicular BAT or supraclavicular quiescent adipose tissue (QAT) regions. MR spectroscopy results from BAT and WAT were compared with grouped and paired tests.

Results

Of 21 healthy participants (mean age, 36 years ± 16 [standard deviation]; 13 men) FDG PET/CT showed active BAT in 24 MR spectroscopy–targeted regions in 16 participants (eight men). Four men had QAT. The T2 for methylene protons was shorter in BAT (mean, 69 msec ± 6, 24 regions) than in WAT (mean, 83 msec ± 3, 18 regions, P < .01) and QAT (mean, 78 msec ± 2, five regions, P < .01). A T2 cut-off value of 76 msec enabled the differentiation of BAT from WAT or QAT with a sensitivity of 85% and a specificity of 95%. Densities of protons adjacent and between double bonds were 33% and 24% lower, respectively, in BAT compared with those in WAT (P = .01 and P = .03, respectively), indicating a lower content of unsaturated and polyunsaturated fatty acids, respectively, in BAT compared with WAT.

Conclusion

Proton MR spectroscopy showed shorter T2 and lower unsaturated fatty acids in brown adipose tissue (BAT) than that in white adipose tissue in healthy humans. It was feasible to identify BAT with MR spectroscopy without the use of PET/CT or cold stimulation.

© RSNA, 2021

See also the editorial by Barker in this issue.

Online supplemental material is available for this article.

Summary

Relaxation measurements with proton MR spectroscopy showed that cold-activatable brown adipose tissue had shorter T2 and lower unsaturated fatty acids than did white adipose tissue in healthy humans.

Key Results

• ■ In a study of 21 healthy participants, the T2 of the lipid methylene resonance of cold-activatable brown adipose tissue (BAT) (69 msec ± 9, 24 regions) was shorter than that of white adipose tissue (WAT) (83 msec ± 3, 18 regions, P < .001).

• ■ The difference in methylene T2 can be used to differentiate cold-activatable BAT, as confirmed with cold exposure PET/CT examination, from adipose tissues nonresponsive to cold stimulation with 85% sensitivity and 95% specificity.

• ■ Proton densities showed 33% lower unsaturated fatty acid contents in BAT (P = .004) and 24% lower polyunsaturated fatty acid contents in WAT (P = .03).

Introduction

Obesity- and diabetes-related health problems have fueled a wide interest in the biochemistry and physiologic characteristics of adipose tissue (1). Brown adipose tissue (BAT) has been the subject of numerous studies after it was recently identified in adult humans (2). In humans, metabolically active BAT occurs near the clavicles and in the neck (2). The primary method for detection of active human BAT is fluorine 18 fluorodeoxyglucose (FDG) PET/CT (3,4) after activation by cold (2) or pharmaceuticals (5). However, the FDG image has moderate spatial resolution compared with MRI scans and probably underestimates the portion of activated BAT metabolism that uses free fatty acid as an energy source (6).

Progress has been made in identifying active BAT in humans with MRI (7), often with water-fat MRI (8,9). Most MRI methods quantify fat in organs with fat contents less than 50%, using assumptions that may not be applicable to BAT. A single T2* is often reported for the total MRI signal, making it codependent on the fat content and different if most of the total signal is from fat (10).

Localized MR spectroscopy relaxometry allows for the extraction of T2 for each of the MRI-observable resonances of triglyceride molecules and water. Proton MR spectroscopy can also help measure the relative amounts of total unsaturated fatty acids (UFAs) and polyunsaturated fatty acids (PUFAs) in human adipose tissue (11,12). A study of excised rodent tissue samples examined with proton MR spectroscopy (13) showed differences in water content and UFA and PUFA content between interscapular BAT and white adipose tissue (WAT). Animal studies (14) show that the triglyceride composition of BAT can be lower in UFA and PUFA when deriving fatty acid content in large parts from de novo synthesis from glucose. If similar mechanisms exist in humans, then the fatty acids in triglyceride stores in human cold-activatable BAT may also show a difference in UFA and PUFA compared with other adipose tissues.

The purpose of this prospective study in healthy humans was to obtain in vivo detailed information on the proton MR spectroscopy relaxation parameters, composition, and unsaturated fatty acid content in adipose supraclavicular regions, proven with FDG PET/CT to be cold activatable, and to compare this information with not cold-activated, quiescent adipose tissue (QAT) in similar locations and in distal subcutaneous WAT. We examined the potential of these parameters to improve BAT identification with MR spectroscopy–based methods and to further our understanding of the biochemical and biophysical characteristics of BAT in humans.

Materials and Methods

Participant Characteristics

This prospective study was approved by the institutional review board and was compliant with the Health Insurance Portability and Accountability Act. The study included non-Hispanic White participants or African American participants who were 18–35 years old (men or women) or 55–75 years old (men) and who were consecutively enrolled between August 2016 and May 2019 at the National Institutes of Health (ClinicalTrials.gov identifier: NCT01568671). They also volunteered for MRI and had no contraindications to MRI (ClinicalTrials.gov identifier: NCT01399385). All participants provided previous written informed consent. The study was conducted in compliance with the Declaration of Helsinki. Data generated or analyzed during the study are available by request from the corresponding author.

MRI Protocol

Participants were scanned in the supine position with a Verio 3.0-T scanner (Siemens Healthcare) with a neck coil and a flex array coil placed over the clavicular area. After B0 field mapping (15) and shimming, T2-weighted images were obtained of the neck and shoulder area. High-spatial-resolution variable projection water-fat MRI (16) was used to record orthogonal sections and oblique sections aligned with supraclavicular adipose regions to find continuous potential BAT regions for placement of MR spectroscopy volumes (median size, 14 × 8 × 8 mm³) and to measure the water and fat content of those regions (Fig 1).

Figure 1:

A–C, Proton density fat fraction (PDFF) images used for MR spectroscopy planning and, D, PET/CT image for retrospective brown adipose tissue activity confirmation. High-spatial-resolution water-fat MRI-derived coronal (A), oblique transverse (B), and oblique coronal (C) PDFF maps (resolution, 1.4 × 1.4 × 3.0 mm) in 24-year-old man with location of proton MR spectroscopy volume cross-section with images shown as black boxes (at gray arrow). Oblique transverse map (B) was prescribed from coronal image (A). Object on top is saline bag used for coil loading and stabilization. Oblique coronal map (C) was prescribed from image B. D, Fluorine 18 fluorodeoxyglucose (FDG) PET/CT image recorded in separate session after cold stimulation (not available to MR spectroscopy operator at time of examination). SUV = standard uptake value. E, Graph shows CT attenuation as function of FDG PET standard uptake value (SUV) within regions of interest retrospectively matched to proton MR spectroscopy volume locations. Symbols and error bars represent means and ranges for CT attenuation in Hounsfield units (vertical axis) and FDG uptake (horizontal axis, logarithmic scale) for region of interest matched to proton MR spectroscopy volumes. Blue symbols are for regions classified as quiescent adipose tissue (n = 5), and black symbols represent regions classified as brown adipose tissue (n = 24). Dashed line is drawn at a glucose uptake of 1.5 SUV, the Brown Adipose Reporting Criteria in Imaging Studies 1.0 criterion for counting FDG PET image pixels as brown adipose tissue in lean participants.

A series of point-resolved spectroscopy (17) localized spectra was recorded at a repetition time of 0.8 seconds using 16 logarithmically spaced echo times (23.7–200 msec) and one extra spectrum for saturation estimates (18–20). This was repeated three times for each volume placement. Subsequently, scanning time permitting, a matching set of relaxation time series of WAT was recorded in the same individual. A more detailed description is given in Appendix E1 (online).

Spectra were analyzed with the advanced method for accurate, robust, and efficient spectral fitting of MR spectroscopy data (21) in the Java MR user interface (version 6.0 beta; http://www.jmrui.eu/) (22). A sample spectrum with fitted resonances is shown in Figure 2. Echo time series data were corrected for saturation according to the study by Gambarota et al (18). Relaxation-corrected spin densities and T2s were extracted with linear regression using Matlab software (version 2019a; Mathworks).

Figure 2a:

Analysis of proton MR spectra of brown adipose tissue. (a) Triglyceride molecule with one of the fatty acids shown in detail with peak labels A–H for individual lipid components. A is methyl proton −CH₃. B1 and B2 are methylene proton −(CH₂)_n. C is methylene protons β to COO. D is methylene protons α to HC=CH. E is methylene protons α to COO. F is diallylic methylene protons. G1, G2, and G3 are glycerol. H is vinylic protons HC=CH. (b) Chemical shift spectra, in parts per million (ppm), of individual fitted components of fit model used in advanced method for accurate, robust, and efficient spectral fitting of MR spectroscopy data color coded and labeled as in a. Overlapping peaks for glycerol protons G1, G2, and G3 and for water are offset for clarity. H₂O peak was fitted with combination of Lorentzian and Gaussian line shape H₂O_L and H₂O_G, and resulting peak areas were summed for estimates of relaxation parameters and water content. (c) Detail of fitting of two Gaussian peaks (B1 and B2) to asymmetric methylene signal. Matlab reconstruction of the (CH₂)_n signal from two fitted components was used to find line width (fwhh) and true fitted peak maximum position (fhmax). Exact frequencies at half height were determined by linear interpolation between closest in points in 4096 point and 1500 Hz digitization of model spectrum. (d) Results of time domain data fitting for representative brown adipose tissue spectrum (repetition time sec/echo time msec, 0.8/24; signal-to-noise ratio, 444; methylene line width, 31 Hz) recorded at echo time of 24 msec with measured data (black line), fitted model spectrum (red line), and fit residual (green line) (all shown at same scale) and were recorded at 1500 Hz spectral width of 1024 points and reconstructed with 4096 points and 2 Hz line broadening.

Figure 2c:

Analysis of proton MR spectra of brown adipose tissue. (a) Triglyceride molecule with one of the fatty acids shown in detail with peak labels A–H for individual lipid components. A is methyl proton −CH₃. B1 and B2 are methylene proton −(CH₂)_n. C is methylene protons β to COO. D is methylene protons α to HC=CH. E is methylene protons α to COO. F is diallylic methylene protons. G1, G2, and G3 are glycerol. H is vinylic protons HC=CH. (b) Chemical shift spectra, in parts per million (ppm), of individual fitted components of fit model used in advanced method for accurate, robust, and efficient spectral fitting of MR spectroscopy data color coded and labeled as in a. Overlapping peaks for glycerol protons G1, G2, and G3 and for water are offset for clarity. H₂O peak was fitted with combination of Lorentzian and Gaussian line shape H₂O_L and H₂O_G, and resulting peak areas were summed for estimates of relaxation parameters and water content. (c) Detail of fitting of two Gaussian peaks (B1 and B2) to asymmetric methylene signal. Matlab reconstruction of the (CH₂)_n signal from two fitted components was used to find line width (fwhh) and true fitted peak maximum position (fhmax). Exact frequencies at half height were determined by linear interpolation between closest in points in 4096 point and 1500 Hz digitization of model spectrum. (d) Results of time domain data fitting for representative brown adipose tissue spectrum (repetition time sec/echo time msec, 0.8/24; signal-to-noise ratio, 444; methylene line width, 31 Hz) recorded at echo time of 24 msec with measured data (black line), fitted model spectrum (red line), and fit residual (green line) (all shown at same scale) and were recorded at 1500 Hz spectral width of 1024 points and reconstructed with 4096 points and 2 Hz line broadening.

Figure 2d:

Analysis of proton MR spectra of brown adipose tissue. (a) Triglyceride molecule with one of the fatty acids shown in detail with peak labels A–H for individual lipid components. A is methyl proton −CH₃. B1 and B2 are methylene proton −(CH₂)_n. C is methylene protons β to COO. D is methylene protons α to HC=CH. E is methylene protons α to COO. F is diallylic methylene protons. G1, G2, and G3 are glycerol. H is vinylic protons HC=CH. (b) Chemical shift spectra, in parts per million (ppm), of individual fitted components of fit model used in advanced method for accurate, robust, and efficient spectral fitting of MR spectroscopy data color coded and labeled as in a. Overlapping peaks for glycerol protons G1, G2, and G3 and for water are offset for clarity. H₂O peak was fitted with combination of Lorentzian and Gaussian line shape H₂O_L and H₂O_G, and resulting peak areas were summed for estimates of relaxation parameters and water content. (c) Detail of fitting of two Gaussian peaks (B1 and B2) to asymmetric methylene signal. Matlab reconstruction of the (CH₂)_n signal from two fitted components was used to find line width (fwhh) and true fitted peak maximum position (fhmax). Exact frequencies at half height were determined by linear interpolation between closest in points in 4096 point and 1500 Hz digitization of model spectrum. (d) Results of time domain data fitting for representative brown adipose tissue spectrum (repetition time sec/echo time msec, 0.8/24; signal-to-noise ratio, 444; methylene line width, 31 Hz) recorded at echo time of 24 msec with measured data (black line), fitted model spectrum (red line), and fit residual (green line) (all shown at same scale) and were recorded at 1500 Hz spectral width of 1024 points and reconstructed with 4096 points and 2 Hz line broadening.

The total lipid spin density, Sf, was estimated from the fatty acid methyl (Fig 2a, 2b) plus methylene resonance signals (B1 + B2), assuming that these two comprised 73% of the total lipid spin density (23). The MR spectroscopy–derived proton density fat fraction (PDFF_S) was then calculated as follows: PDFF_S = Sf/(Sf + Sw) × 100%, where Sw is the spin density of water. For image pixels within intersections of the MR spectroscopy volumes and water-fat images, an image-derived proton density fat fraction (PDFF_i) was calculated from fat and water pixel intensities (If and Iw, respectively) as follows: PDFF_i = If/(If + Iw) × 100%.

Figure 2b:

Analysis of proton MR spectra of brown adipose tissue. (a) Triglyceride molecule with one of the fatty acids shown in detail with peak labels A–H for individual lipid components. A is methyl proton −CH₃. B1 and B2 are methylene proton −(CH₂)_n. C is methylene protons β to COO. D is methylene protons α to HC=CH. E is methylene protons α to COO. F is diallylic methylene protons. G1, G2, and G3 are glycerol. H is vinylic protons HC=CH. (b) Chemical shift spectra, in parts per million (ppm), of individual fitted components of fit model used in advanced method for accurate, robust, and efficient spectral fitting of MR spectroscopy data color coded and labeled as in a. Overlapping peaks for glycerol protons G1, G2, and G3 and for water are offset for clarity. H₂O peak was fitted with combination of Lorentzian and Gaussian line shape H₂O_L and H₂O_G, and resulting peak areas were summed for estimates of relaxation parameters and water content. (c) Detail of fitting of two Gaussian peaks (B1 and B2) to asymmetric methylene signal. Matlab reconstruction of the (CH₂)_n signal from two fitted components was used to find line width (fwhh) and true fitted peak maximum position (fhmax). Exact frequencies at half height were determined by linear interpolation between closest in points in 4096 point and 1500 Hz digitization of model spectrum. (d) Results of time domain data fitting for representative brown adipose tissue spectrum (repetition time sec/echo time msec, 0.8/24; signal-to-noise ratio, 444; methylene line width, 31 Hz) recorded at echo time of 24 msec with measured data (black line), fitted model spectrum (red line), and fit residual (green line) (all shown at same scale) and were recorded at 1500 Hz spectral width of 1024 points and reconstructed with 4096 points and 2 Hz line broadening.

Relative fatty acid spin densities were calculated by dividing them by the sum of all fatty acid resonance peak areas (peaks A–H, excluding glycerol peaks G1–G3 that may be subject to contamination from water signals).

PET/CT Protocol

In a separate examination, participants were tested for cold-induced BAT activity with PET/CT using an established protocol for quantification of human BAT (Brown Adipose Reporting Criteria in Imaging Studies 1.0) (4). Each participant was given 10 mCi of FDG by intravenous injection 60–90 minutes before scan acquisition with a Biograph mCT PET/CT scanner (Siemens Healthcare).

Image Analysis

Two radiologists (A.M.G., with 20 years of experience, and A.H., with 15 years of experience) in consensus, who were blinded to MR spectroscopy results, matched the PET/CT images to water-fat MRI scans with the position of the MR spectroscopy volume marked (Fig 1). If no consensus could be reached, a third reader (J.F.E., with 30 years of experience), would have independently reviewed the data. The readouts were completed in three sessions, 1 week apart. Co-registration of PET/CT and MRI scans was assisted with specialized software (Virtual Place; Aze).

The mean CT attenuation in Hounsfield units and FDG PET standard uptake value were determined within regions of interest drawn onto the PET/CT image corresponding to the proton MR spectroscopy regions of interest. MR spectroscopy regions coinciding with PET/CT-positive regions following Brown Adipose Reporting Criteria in Imaging Studies 1.0 criteria were classified as BAT. Clavicular MR spectroscopy volumes that did not show PET activity were labeled as QAT (Fig 1e). Subcutaneous adipose from distal regions were identified as WAT.

Statistical Analysis

The PDFF calculated from fat and water pixel intensities and repeat MR spectroscopy measurements were averaged after exclusion of extreme outliers (24). The repeat results and results per location that differed from the median by more than three times the median absolute deviation were removed. Statistic calculations were performed with Matlab software (Statistics and Machine Learning Toolbox, version 11.5; Mathworks) and with MedCalc software (version 19.1; https://www.medcalc.org). Comparisons were performed with paired and unpaired t tests if data were normally distributed (per the D’Agostino-Pearson test) and with the Wilcoxon (paired) or Mann-Whitney ranked sum tests.

The MR spectroscopy–derived PDFF values and PDFF values calculated from fat and water pixel intensities for clavicular regions (BAT and QAT) were compared with linear regression and a mean difference plot. BAT and non-BAT areas were analyzed with a receiver operating characteristic analysis for the separation of WAT and non-BAT using the MR spectroscopy–derived PDFF, or the methylene T2 was compared (25) using all echo time series as separate data and also after averaging the repeated observations per region of interest measured.

Results

Participant Characteristics

Twenty-one healthy participants (mean age, 36 years ± 16 [standard deviation]; 13 men) were evaluated. Participant characteristics are listed in Table 1. The flowchart for this study is shown in Figure 3.

Table 1:

Characteristics of Participants, Sampled Regions, and Observations for Active BAT, QAT, and WAT

Figure 3:

Flowchart of study. Adipose tissue was targeted for localized MR spectroscopy (MRS) using water-fat MRI in left and right clavicular areas or distal subcutaneous white adipose tissue. After MRI examination, participants were exposed to cold and scanned with fluorine 18 fluorodeoxyglucose PET/CT for glucose uptake to identify brown adipose tissue (BAT) as reference standard. FN = false-negative, FP = false-positive, NP = number of participants, TN = true-negative, TP = true-positive.

Localization and Identification of BAT

FDG PET/CT showed cold-activated metabolic activity in MR spectroscopy–targeted BAT clavicular areas in 16 of the 21 participants. In one man, FDG uptake was observed in the clavicular area, but the MR spectroscopy volume did not coincide with this activity focus; thus, this result was excluded. Six men were in the age 55–75 years group. Two had cold-activatable BAT (ages 59 and 62 years), and four did not show glucose uptake with PET/CT. MR spectroscopy results from the five regions scanned in these four participants were labeled as QAT (Fig 1e).

In 17 participants, MR spectroscopy was also performed in subcutaneous WAT on the back of the torso. Thirteen of these participants had metabolically active BAT to yield 20 locations of BAT matched for paired comparisons with WAT volumes in the same individual. The numbers of participants, separate MR spectroscopy locations, and repeated series are listed in Table 1 as are the mean standard uptake value and CT attenuations recorded for the BAT and QAT regions.

MR Spectroscopy

MR spectroscopy series from the BAT and WAT from the same individual are shown in Figure 4. The signal-to-noise ratio and line widths of the (CH₂)_n signal in BAT and in WAT are listed in Table 2. Spectra from subcutaneous WAT had a higher signal-to-noise ratio than spectra from clavicular areas mainly because of the more than threefold larger size of WAT volumes. The line width differences were due to the challenges of shimming the periclavicular fat regions, as well as the shorter T2 of the methylene resonances in BAT.

Figure 4a:

Localized proton MR spectroscopy relaxation measurement series of brown adipose tissue (BAT) and white adipose tissue (WAT). TE = echo time, TR = repetition time. (a) Echo time series spectra in arbitrary signal intensity units against chemical shift in parts per million (ppm) from BAT in volume shown in Figure 1. This representative BAT series had methylene peak line width of 31 Hz, in the shortest echo time spectrum. This was close to median value of 32 Hz (range, 20–63 Hz) for all BAT volumes. Methylene signal-to-noise ratio in time domain of 479 versus median for all BAT of 414 (range, 238–887). (b) Echo time series spectra recorded from distal subcutaneous WAT. For each series, short repetition time spectra and fully relaxed single-shot spectrum are scaled to the same noise level. Noise level after averaging four signals is half that of single shot fully relaxed acquisition. All spectra were recorded at 1024 point resolution and Fourier transformed at 4096 point resolution with 2-Hz line broadening.

Table 2:

Comparisons of Methylene LWs, SNR, Relaxation Properties of Spectral Components, and Spin Density Fractions of PET/CT-Proven Active BAT and Subcutaneous WAT Volumes

Figure 4b:

Localized proton MR spectroscopy relaxation measurement series of brown adipose tissue (BAT) and white adipose tissue (WAT). TE = echo time, TR = repetition time. (a) Echo time series spectra in arbitrary signal intensity units against chemical shift in parts per million (ppm) from BAT in volume shown in Figure 1. This representative BAT series had methylene peak line width of 31 Hz, in the shortest echo time spectrum. This was close to median value of 32 Hz (range, 20–63 Hz) for all BAT volumes. Methylene signal-to-noise ratio in time domain of 479 versus median for all BAT of 414 (range, 238–887). (b) Echo time series spectra recorded from distal subcutaneous WAT. For each series, short repetition time spectra and fully relaxed single-shot spectrum are scaled to the same noise level. Noise level after averaging four signals is half that of single shot fully relaxed acquisition. All spectra were recorded at 1024 point resolution and Fourier transformed at 4096 point resolution with 2-Hz line broadening.

Relaxation Properties

Relaxation properties of spectral components of BAT and WAT are listed in Table 2. The T1 for water was longer in BAT compared with WAT. The summed CH₃ + (CH₂)_n signal and various fatty acid chain protons had longer T1 in BAT. The T2 of water was not different in BAT and WAT, but most fatty acid chain proton signals had a shorter T2 in BAT than in WAT.

Paired comparison of BAT and WAT from the same individuals showed that all protons along the fatty acid chain, except peaks C and E, have shorter T2 in BAT, and most tended to have longer T1 in BAT than in WAT (Table 3). The T2 of methylene protons, shown in Figure 5a, was always greater in WAT than in BAT for a given individual.

Table 3:

Paired Comparisons of BAT and WAT from Same Participants for Methylene LWs, SNR, Relaxation Properties of Spectral Components, and Spin Density Fractions of PET/CT Proven Active BAT and Subcutaneous WAT Volumes

Figure 5a:

Matched comparisons of 20 fluorine 18 fluorodeoxyglucose PET/CT active brown adipose tissue (BAT) regions in 13 individuals with white adipose tissue (WAT) regions in same individuals. (a) Graph shows T2 of (CH₂)_n fatty acid chain. (b) Graph shows ratio of relaxation-corrected proton densities of diallylic methylene peak F (unique for polyunsaturated fatty acid [PUFA]) with peak E of methylene α to fatty acid carboxyl group (of which there is one per fatty acid chain). F/E = ratio of peak F to peak E.

Fatty Acid Composition

Table 2 lists proton densities of the spectral components of fatty acid as a fraction of all fatty acid protons (ie, excluding the triglyceride glycerol protons). Differences were found in proton density fractions unique to UFA and PUFA (labels D, F, and H). Peak D at 2.1 ppm is present in all UFA. The diallylic signal F at 2.8 ppm is unique to PUFA. Thus, BAT appears to have lower total UFA and PUFA contents compared with WAT. The PUFA content was also characterized by the ratio of peak F to the methylene α to the carboxyl group (peak E). In the group comparison in Table 2, this was lower in BAT compared with WAT. In the paired comparison within same individuals, the difference was more pronounced, with a mean F/E ratio of 0.25 ± 0.21 in BAT versus 0.47 ± 0.24 in WAT (n = 19, P < .01) (Fig 5b).

Figure 5b:

Matched comparisons of 20 fluorine 18 fluorodeoxyglucose PET/CT active brown adipose tissue (BAT) regions in 13 individuals with white adipose tissue (WAT) regions in same individuals. (a) Graph shows T2 of (CH₂)_n fatty acid chain. (b) Graph shows ratio of relaxation-corrected proton densities of diallylic methylene peak F (unique for polyunsaturated fatty acid [PUFA]) with peak E of methylene α to fatty acid carboxyl group (of which there is one per fatty acid chain). F/E = ratio of peak F to peak E.

The (B1 + B2)/A ratio, indicating the number of CH₂ per fatty acid chain or (saturated) fatty acid chain length, was higher in BAT. Paired comparison of BAT and WAT yielded similar mean results (mean, 5.1 ± 1.8 in WAT and 6.9 ± 3.4, in BAT; n = 20; P < .05) in a paired one-tailed comparison signifying either a shorter fatty acid chain length or more unsaturated carbons for a given fatty acid chain length in WAT.

Fat Fractions of WAT and BAT

The MR spectroscopy–derived PDFF correlated with PDFF values calculated from fat and water pixel intensities from the same regions (R² = 0.791, P < .01) with a slope of 1.07 and an offset of −9%. MR spectroscopy–derived PDFF is plotted as a function of the methylene T2 from all individual measurements in BAT and non-BAT regions in Figure 6a. The MR spectroscopy–derived PDFF showed some overlap between the groups with a mean MR spectroscopy–derived PDFF for BAT of 78% ± 7 compared to 73% ± 16 found previously for adults (8). Methylene T2 in BAT regions (mean, 69 msec ± 7; n = 67; range, 46–85 msec) was shorter than that in QAT (mean, 78 msec ± 3; n = 15; range, 73–82 msec; P = .01) and shorter than in WAT (mean, 83 ± 4 msec; n = 44; range, 76–92 msec; P < .01). Box-and-whisker plots of these groups of discrete measurements are shown in Figure 6b.

Figure 6a:

Separation of brown adipose tissue (BAT), quiescent adipose tissue (QAT), and white adipose tissue (WAT) on basis of fat content (proton density fat fraction [PDFF_s]) or T2 of fatty acid methylene peak. (a) Two-dimensional scatter plot shows MR spectroscopy–derived PDFF_S as function of T2 of (CH₂)_n fatty acid chain protons for fluorine 18 fluorodeoxyglucose PET/CT active BAT, supraclavicular adipose regions in participants showing no cold-induced activity in PET (QAT), and distal subcutaneous white adipose tissue (WAT). Plot shows all observations for BAT (n = 67, red circles), QAT (n = 15, blue squares), and WAT (n = 46, black triangles). Means and standard deviations for these three groups are shown as larger open symbols and error bars in their corresponding colors and shape. (b) Box-and-whisker plot of all observations of T2 of fatty acid methylene peak for BAT, QAT, and WAT. Upper and lower bounds in box are quartiles, and notch indicates median. Whiskers show data limits without outliers. Data points have same markers and colors as in a, and markers for data points beyond 1.5 times the interquartile range are filled. P values were calculated with three-way comparison analysis of variance. (c, d) Receiver operating characteristic curves for identification of adipose tissue as BAT or as not cold-activatable adipose tissue (QAT or WAT) using T2 of fatty acid methylene resonance (T₂ CH₂) (c) and fat content (MR spectroscopy–derived PDFF_S) (d). The same 67 data points for BAT and the 61 data points for QAT and WAT shown in a were used to generate receiver operating characteristic curves for T2 of (CH₂)_n fatty acid protons (area under the curve, 0.95; 95% CI: 0.90, 0.98) and for MR spectroscopy–derived proton density fat fraction (area under the curve, 0.83; 95% CI: 0.75, 0.89). Youden optimum cut-off point for T2 is 76 msec for sensitivity of 85% (95% CI: 74, 93) and specificity of 95% (95% CI: 86, 99) and is indicated by red dot. Receiver operating characteristic curve for MR spectroscopy–derived PDFF has an optimum cut-off at PDFF equal to 89% for sensitivity of 100% (95% CI: 95, 100) and specificity of 52% (95% CI: 38, 65). (e, f) Receiver operating characteristic curves for T2 of fatty acid methylene resonance (T₂ CH₂) (e) and MR spectroscopy–derived PDFF_S (f) with reduced number of data points by averaging repeat measurements per MR spectroscopy location. Receiver operating characteristic for methylene T2 (e) had area under the curve of 0.98 (95% CI: 0.98, 1.00); cut-off was 76 msec for sensitivity of 88% (95% CI: 68, 100) and specificity of 100% (95% CI: 85, 100). Receiver operating characteristic for MR spectroscopy–derived PDFF (f) had area under the curve of 0.86 (95% CI: 0.73, 0.95); cut-off was 84% for sensitivity of 79% (95% CI: 58, 93) and specificity of 77% (95% CI: 55, 92). Again, area under the curve for methylene T2 was better than for MR spectroscopy–derived proton density fat fraction (P < .001).

Figure 6b:

Separation of brown adipose tissue (BAT), quiescent adipose tissue (QAT), and white adipose tissue (WAT) on basis of fat content (proton density fat fraction [PDFF_s]) or T2 of fatty acid methylene peak. (a) Two-dimensional scatter plot shows MR spectroscopy–derived PDFF_S as function of T2 of (CH₂)_n fatty acid chain protons for fluorine 18 fluorodeoxyglucose PET/CT active BAT, supraclavicular adipose regions in participants showing no cold-induced activity in PET (QAT), and distal subcutaneous white adipose tissue (WAT). Plot shows all observations for BAT (n = 67, red circles), QAT (n = 15, blue squares), and WAT (n = 46, black triangles). Means and standard deviations for these three groups are shown as larger open symbols and error bars in their corresponding colors and shape. (b) Box-and-whisker plot of all observations of T2 of fatty acid methylene peak for BAT, QAT, and WAT. Upper and lower bounds in box are quartiles, and notch indicates median. Whiskers show data limits without outliers. Data points have same markers and colors as in a, and markers for data points beyond 1.5 times the interquartile range are filled. P values were calculated with three-way comparison analysis of variance. (c, d) Receiver operating characteristic curves for identification of adipose tissue as BAT or as not cold-activatable adipose tissue (QAT or WAT) using T2 of fatty acid methylene resonance (T₂ CH₂) (c) and fat content (MR spectroscopy–derived PDFF_S) (d). The same 67 data points for BAT and the 61 data points for QAT and WAT shown in a were used to generate receiver operating characteristic curves for T2 of (CH₂)_n fatty acid protons (area under the curve, 0.95; 95% CI: 0.90, 0.98) and for MR spectroscopy–derived proton density fat fraction (area under the curve, 0.83; 95% CI: 0.75, 0.89). Youden optimum cut-off point for T2 is 76 msec for sensitivity of 85% (95% CI: 74, 93) and specificity of 95% (95% CI: 86, 99) and is indicated by red dot. Receiver operating characteristic curve for MR spectroscopy–derived PDFF has an optimum cut-off at PDFF equal to 89% for sensitivity of 100% (95% CI: 95, 100) and specificity of 52% (95% CI: 38, 65). (e, f) Receiver operating characteristic curves for T2 of fatty acid methylene resonance (T₂ CH₂) (e) and MR spectroscopy–derived PDFF_S (f) with reduced number of data points by averaging repeat measurements per MR spectroscopy location. Receiver operating characteristic for methylene T2 (e) had area under the curve of 0.98 (95% CI: 0.98, 1.00); cut-off was 76 msec for sensitivity of 88% (95% CI: 68, 100) and specificity of 100% (95% CI: 85, 100). Receiver operating characteristic for MR spectroscopy–derived PDFF (f) had area under the curve of 0.86 (95% CI: 0.73, 0.95); cut-off was 84% for sensitivity of 79% (95% CI: 58, 93) and specificity of 77% (95% CI: 55, 92). Again, area under the curve for methylene T2 was better than for MR spectroscopy–derived proton density fat fraction (P < .001).

Use of MR Spectroscopy to Identify Cold-activatable BAT

Receiver operating characteristic curves for classifying cold-activatable BAT versus non-BAT were constructed from the MR spectroscopy–derived PDFF and methylene T2 data shown in Figure 6a and b. The methylene T2 can be used to distinguish BAT from non-BAT with a cut-off value of 76 msec or less at a sensitivity of 85% (57 of 67 regions; 95% CI: 74, 93) and a specificity of 95% (58 of 61 regions; 95% CI: 86, 99) (Fig 6c). Using the MR spectroscopy–derived PDFF, BAT can be distinguished from non-BAT at cut-off value MR spectroscopy–derived PDFF of 89% or less with a sensitivity of 100% (67 of 67; 95% CI: 95, 100) and a specificity of 52% (31 of 60 regions; 95% CI: 38, 65) (Fig 6d). The methylene resonance T2 showed better area under the curve than MR spectroscopy–derived PDFF with P < .01 when compared using the method of DeLong et al (25).

Figure 6c:

Separation of brown adipose tissue (BAT), quiescent adipose tissue (QAT), and white adipose tissue (WAT) on basis of fat content (proton density fat fraction [PDFF_s]) or T2 of fatty acid methylene peak. (a) Two-dimensional scatter plot shows MR spectroscopy–derived PDFF_S as function of T2 of (CH₂)_n fatty acid chain protons for fluorine 18 fluorodeoxyglucose PET/CT active BAT, supraclavicular adipose regions in participants showing no cold-induced activity in PET (QAT), and distal subcutaneous white adipose tissue (WAT). Plot shows all observations for BAT (n = 67, red circles), QAT (n = 15, blue squares), and WAT (n = 46, black triangles). Means and standard deviations for these three groups are shown as larger open symbols and error bars in their corresponding colors and shape. (b) Box-and-whisker plot of all observations of T2 of fatty acid methylene peak for BAT, QAT, and WAT. Upper and lower bounds in box are quartiles, and notch indicates median. Whiskers show data limits without outliers. Data points have same markers and colors as in a, and markers for data points beyond 1.5 times the interquartile range are filled. P values were calculated with three-way comparison analysis of variance. (c, d) Receiver operating characteristic curves for identification of adipose tissue as BAT or as not cold-activatable adipose tissue (QAT or WAT) using T2 of fatty acid methylene resonance (T₂ CH₂) (c) and fat content (MR spectroscopy–derived PDFF_S) (d). The same 67 data points for BAT and the 61 data points for QAT and WAT shown in a were used to generate receiver operating characteristic curves for T2 of (CH₂)_n fatty acid protons (area under the curve, 0.95; 95% CI: 0.90, 0.98) and for MR spectroscopy–derived proton density fat fraction (area under the curve, 0.83; 95% CI: 0.75, 0.89). Youden optimum cut-off point for T2 is 76 msec for sensitivity of 85% (95% CI: 74, 93) and specificity of 95% (95% CI: 86, 99) and is indicated by red dot. Receiver operating characteristic curve for MR spectroscopy–derived PDFF has an optimum cut-off at PDFF equal to 89% for sensitivity of 100% (95% CI: 95, 100) and specificity of 52% (95% CI: 38, 65). (e, f) Receiver operating characteristic curves for T2 of fatty acid methylene resonance (T₂ CH₂) (e) and MR spectroscopy–derived PDFF_S (f) with reduced number of data points by averaging repeat measurements per MR spectroscopy location. Receiver operating characteristic for methylene T2 (e) had area under the curve of 0.98 (95% CI: 0.98, 1.00); cut-off was 76 msec for sensitivity of 88% (95% CI: 68, 100) and specificity of 100% (95% CI: 85, 100). Receiver operating characteristic for MR spectroscopy–derived PDFF (f) had area under the curve of 0.86 (95% CI: 0.73, 0.95); cut-off was 84% for sensitivity of 79% (95% CI: 58, 93) and specificity of 77% (95% CI: 55, 92). Again, area under the curve for methylene T2 was better than for MR spectroscopy–derived proton density fat fraction (P < .001).

Figure 6d:

Separation of brown adipose tissue (BAT), quiescent adipose tissue (QAT), and white adipose tissue (WAT) on basis of fat content (proton density fat fraction [PDFF_s]) or T2 of fatty acid methylene peak. (a) Two-dimensional scatter plot shows MR spectroscopy–derived PDFF_S as function of T2 of (CH₂)_n fatty acid chain protons for fluorine 18 fluorodeoxyglucose PET/CT active BAT, supraclavicular adipose regions in participants showing no cold-induced activity in PET (QAT), and distal subcutaneous white adipose tissue (WAT). Plot shows all observations for BAT (n = 67, red circles), QAT (n = 15, blue squares), and WAT (n = 46, black triangles). Means and standard deviations for these three groups are shown as larger open symbols and error bars in their corresponding colors and shape. (b) Box-and-whisker plot of all observations of T2 of fatty acid methylene peak for BAT, QAT, and WAT. Upper and lower bounds in box are quartiles, and notch indicates median. Whiskers show data limits without outliers. Data points have same markers and colors as in a, and markers for data points beyond 1.5 times the interquartile range are filled. P values were calculated with three-way comparison analysis of variance. (c, d) Receiver operating characteristic curves for identification of adipose tissue as BAT or as not cold-activatable adipose tissue (QAT or WAT) using T2 of fatty acid methylene resonance (T₂ CH₂) (c) and fat content (MR spectroscopy–derived PDFF_S) (d). The same 67 data points for BAT and the 61 data points for QAT and WAT shown in a were used to generate receiver operating characteristic curves for T2 of (CH₂)_n fatty acid protons (area under the curve, 0.95; 95% CI: 0.90, 0.98) and for MR spectroscopy–derived proton density fat fraction (area under the curve, 0.83; 95% CI: 0.75, 0.89). Youden optimum cut-off point for T2 is 76 msec for sensitivity of 85% (95% CI: 74, 93) and specificity of 95% (95% CI: 86, 99) and is indicated by red dot. Receiver operating characteristic curve for MR spectroscopy–derived PDFF has an optimum cut-off at PDFF equal to 89% for sensitivity of 100% (95% CI: 95, 100) and specificity of 52% (95% CI: 38, 65). (e, f) Receiver operating characteristic curves for T2 of fatty acid methylene resonance (T₂ CH₂) (e) and MR spectroscopy–derived PDFF_S (f) with reduced number of data points by averaging repeat measurements per MR spectroscopy location. Receiver operating characteristic for methylene T2 (e) had area under the curve of 0.98 (95% CI: 0.98, 1.00); cut-off was 76 msec for sensitivity of 88% (95% CI: 68, 100) and specificity of 100% (95% CI: 85, 100). Receiver operating characteristic for MR spectroscopy–derived PDFF (f) had area under the curve of 0.86 (95% CI: 0.73, 0.95); cut-off was 84% for sensitivity of 79% (95% CI: 58, 93) and specificity of 77% (95% CI: 55, 92). Again, area under the curve for methylene T2 was better than for MR spectroscopy–derived proton density fat fraction (P < .001).

The same analysis using data obtained by first averaging the repeat measurements per MR spectroscopy volume location yielded similar results (Fig 6e, 6f). The T2 cut-off was 76 msec or less for a sensitivity of 88% (21 of 24 regions; 95% CI: 68, 97) and a specificity of 100% (22 of 22 regions; 95% CI: 85, 100) (Fig 6e).

Figure 6e:

Separation of brown adipose tissue (BAT), quiescent adipose tissue (QAT), and white adipose tissue (WAT) on basis of fat content (proton density fat fraction [PDFF_s]) or T2 of fatty acid methylene peak. (a) Two-dimensional scatter plot shows MR spectroscopy–derived PDFF_S as function of T2 of (CH₂)_n fatty acid chain protons for fluorine 18 fluorodeoxyglucose PET/CT active BAT, supraclavicular adipose regions in participants showing no cold-induced activity in PET (QAT), and distal subcutaneous white adipose tissue (WAT). Plot shows all observations for BAT (n = 67, red circles), QAT (n = 15, blue squares), and WAT (n = 46, black triangles). Means and standard deviations for these three groups are shown as larger open symbols and error bars in their corresponding colors and shape. (b) Box-and-whisker plot of all observations of T2 of fatty acid methylene peak for BAT, QAT, and WAT. Upper and lower bounds in box are quartiles, and notch indicates median. Whiskers show data limits without outliers. Data points have same markers and colors as in a, and markers for data points beyond 1.5 times the interquartile range are filled. P values were calculated with three-way comparison analysis of variance. (c, d) Receiver operating characteristic curves for identification of adipose tissue as BAT or as not cold-activatable adipose tissue (QAT or WAT) using T2 of fatty acid methylene resonance (T₂ CH₂) (c) and fat content (MR spectroscopy–derived PDFF_S) (d). The same 67 data points for BAT and the 61 data points for QAT and WAT shown in a were used to generate receiver operating characteristic curves for T2 of (CH₂)_n fatty acid protons (area under the curve, 0.95; 95% CI: 0.90, 0.98) and for MR spectroscopy–derived proton density fat fraction (area under the curve, 0.83; 95% CI: 0.75, 0.89). Youden optimum cut-off point for T2 is 76 msec for sensitivity of 85% (95% CI: 74, 93) and specificity of 95% (95% CI: 86, 99) and is indicated by red dot. Receiver operating characteristic curve for MR spectroscopy–derived PDFF has an optimum cut-off at PDFF equal to 89% for sensitivity of 100% (95% CI: 95, 100) and specificity of 52% (95% CI: 38, 65). (e, f) Receiver operating characteristic curves for T2 of fatty acid methylene resonance (T₂ CH₂) (e) and MR spectroscopy–derived PDFF_S (f) with reduced number of data points by averaging repeat measurements per MR spectroscopy location. Receiver operating characteristic for methylene T2 (e) had area under the curve of 0.98 (95% CI: 0.98, 1.00); cut-off was 76 msec for sensitivity of 88% (95% CI: 68, 100) and specificity of 100% (95% CI: 85, 100). Receiver operating characteristic for MR spectroscopy–derived PDFF (f) had area under the curve of 0.86 (95% CI: 0.73, 0.95); cut-off was 84% for sensitivity of 79% (95% CI: 58, 93) and specificity of 77% (95% CI: 55, 92). Again, area under the curve for methylene T2 was better than for MR spectroscopy–derived proton density fat fraction (P < .001).

Figure 6f:

Separation of brown adipose tissue (BAT), quiescent adipose tissue (QAT), and white adipose tissue (WAT) on basis of fat content (proton density fat fraction [PDFF_s]) or T2 of fatty acid methylene peak. (a) Two-dimensional scatter plot shows MR spectroscopy–derived PDFF_S as function of T2 of (CH₂)_n fatty acid chain protons for fluorine 18 fluorodeoxyglucose PET/CT active BAT, supraclavicular adipose regions in participants showing no cold-induced activity in PET (QAT), and distal subcutaneous white adipose tissue (WAT). Plot shows all observations for BAT (n = 67, red circles), QAT (n = 15, blue squares), and WAT (n = 46, black triangles). Means and standard deviations for these three groups are shown as larger open symbols and error bars in their corresponding colors and shape. (b) Box-and-whisker plot of all observations of T2 of fatty acid methylene peak for BAT, QAT, and WAT. Upper and lower bounds in box are quartiles, and notch indicates median. Whiskers show data limits without outliers. Data points have same markers and colors as in a, and markers for data points beyond 1.5 times the interquartile range are filled. P values were calculated with three-way comparison analysis of variance. (c, d) Receiver operating characteristic curves for identification of adipose tissue as BAT or as not cold-activatable adipose tissue (QAT or WAT) using T2 of fatty acid methylene resonance (T₂ CH₂) (c) and fat content (MR spectroscopy–derived PDFF_S) (d). The same 67 data points for BAT and the 61 data points for QAT and WAT shown in a were used to generate receiver operating characteristic curves for T2 of (CH₂)_n fatty acid protons (area under the curve, 0.95; 95% CI: 0.90, 0.98) and for MR spectroscopy–derived proton density fat fraction (area under the curve, 0.83; 95% CI: 0.75, 0.89). Youden optimum cut-off point for T2 is 76 msec for sensitivity of 85% (95% CI: 74, 93) and specificity of 95% (95% CI: 86, 99) and is indicated by red dot. Receiver operating characteristic curve for MR spectroscopy–derived PDFF has an optimum cut-off at PDFF equal to 89% for sensitivity of 100% (95% CI: 95, 100) and specificity of 52% (95% CI: 38, 65). (e, f) Receiver operating characteristic curves for T2 of fatty acid methylene resonance (T₂ CH₂) (e) and MR spectroscopy–derived PDFF_S (f) with reduced number of data points by averaging repeat measurements per MR spectroscopy location. Receiver operating characteristic for methylene T2 (e) had area under the curve of 0.98 (95% CI: 0.98, 1.00); cut-off was 76 msec for sensitivity of 88% (95% CI: 68, 100) and specificity of 100% (95% CI: 85, 100). Receiver operating characteristic for MR spectroscopy–derived PDFF (f) had area under the curve of 0.86 (95% CI: 0.73, 0.95); cut-off was 84% for sensitivity of 79% (95% CI: 58, 93) and specificity of 77% (95% CI: 55, 92). Again, area under the curve for methylene T2 was better than for MR spectroscopy–derived proton density fat fraction (P < .001).

Using MR spectroscopy–derived PDFF, BAT can be distinguished from QAT or WAT at the cut-off value MR spectroscopy–derived PDFF of 84% or less with a sensitivity of 79% (19 of 24 regions; 95% CI: 58, 93) and a specificity of 77% (17 of 22 regions; 95% CI: 55, 92) (Fig 6f). With these averaged data, the area under the curve for methylene T2 was again better than for MR spectroscopy–derived PDFF (P < .01).

Discussion

MR spectroscopy relaxation measurements revealed differences in the MRI properties of various fatty acid proton signals and water between brown adipose tissue (BAT) and white adipose tissue (WAT). Metabolically active BAT was retrospectively identified with fluorine 18 fluorodeoxyglucose PET/CT after cold stimulation. This showed the feasibility of targeting potential supraclavicular BAT areas for study with localized MR spectroscopy with only multiplanar high-spatial-resolution water-fat MRI as guidance. Only one MR spectroscopy volume was incorrectly placed, and 29 regions were correctly targeted in 20 participants. Twenty-four of these were in metabolically active BAT areas, and five were in quiescent adipose tissue in the four participants who had no clavicular cold-induced BAT activity. We found several differences in MRI relaxation properties of fatty acid resonances. One of them, the T2 of the most prominent fatty acid methylene resonance, differentiated cold-activatable BAT from other adipose tissues that are nonresponsive to cold (quiescent adipose tissue and WAT) with a sensitivity of 85% and a specificity of 95% without activation of the BAT with drugs or cold exposure. Also, the characterization of the fatty acid resonances revealed differences in the content of unsaturated fatty acid between WAT and cold-activatable BAT. The ratio of the spin densities of diallylic protons over α-carboxyl methylene protons, proportional to the amount of polyunsaturated fatty acid, was 38% lower in BAT compared with WAT (P = .03). This difference was even more pronounced in a paired comparison of BAT and WAT in the same individuals where this ratio was 46% lower in BAT (0.25 vs 0.47, P < .001).

A shorter T2 combined with a longer T1, as was observed for most fatty acid chain methylene resonances, can be indicative of a lower fluidity (26). Even small changes of UFA content can influence fluidity in biologic phospholipid bilayers (27), or food oils (28). More saturated fatty acid lowers the fluidity, so this was also consistent with a lower UFA content in BAT.

Studies in animals suggest that the lower UFA and PUFA content in BAT as compared with WAT could be due to differences in glucose uptake and lipogenesis, or free fatty acid metabolism in these adipose tissue types (14). Our study shows that differences in unsaturation between cold-activatable BAT and WAT found with MR spectroscopy in ex vivo murine samples (13) are also found in adult humans in vivo.

Individual differences in metabolism and diet are likely to cause greater variance in water content (29), lipid composition, and MRI properties in human adipose tissues compared with that in laboratory animals. Even so, grouped comparisons showed significant differences in MR relaxation properties and in the unsaturated fatty acid contents between cold-activatable BAT and other adipose tissues. However, these differences were even more pronounced in intraindividual paired comparisons between BAT and WAT.

Most other attempts at methods to identify BAT with MRI are based on water content, or T2* (3,7). Because the T2* is a mix of inhomogeneity, water T2, and fat T2, the T2* and fat fraction are interdependent (10) and much more method dependent than the fatty acid methylene T2. The T2 differences of the fatty acid signal could be exploited to identify cold-activatable BAT with MRI by separating or suppressing the water signal in a spin-echo sequence, but localized MR spectroscopy brings added value in information on the biochemistry of BAT and WAT in humans.

Our study had some limitations. First, our T1 estimates were based on a two-point saturation method. Repeated measurements helped improve accuracy somewhat, but a much longer inversion-recovery T1 measurement would have improved accuracy. This protocol was not focused on T1 but designed to optimize efficiency and accuracy of the T2 measurements by allowing for more observations in the given time. Second, localization of MR spectroscopy volumes in BAT is challenging, and there is a risk of partial volume effects, which we minimized by using high spatial definition water-fat images in multiple orientations. Still, chemical shift artifacts can introduce off-resonance signals from the direct vicinity of the MR spectroscopy volume. However, the adjacent tissues of lean individuals are mainly muscle with a much lower fat content. Thus, these effects should not have had a high impact on the relaxation parameters and fatty acid composition data, but they could have led to errors in MR spectroscopy PDFF estimates in adipose tissue.

In conclusion, the T2 of the strongest fatty acid chain proton signal was lower in brown adipose tissue (BAT), and this proved to be a good parameter for classifying adipose tissue as cold-activatable BAT or not cold-activatable adipose tissue (ie, quiescent adipose tissue or white adipose tissue [WAT]). Proton MR spectroscopy can also help detect differences in unsaturated fatty acid content between BAT and WAT that may relate to important metabolic properties of BAT. These results form the basis for development of MRI-based methods for identification and quantification of BAT in humans without the need for cold activation or radiation. They will be useful for advancing knowledge of human BAT biology and possibly also the development of future medical applications.