H NMR spectra from human adipose tissue were obtained routinely at 7T. The features of the 1
H spectra at 7T were consistent with spectra of triacylglycerols obtained under high-resolution conditions (8
) or in mice at 7T (20
). Perhaps the most significant observation was that chemical shift dispersion in humans is greatly improved compared with 1.5 T or 3.0 T. The ability to resolve protons adjacent to double bonds allows noninvasive estimation of the fatty acid composition of adipose tissue.
C NMR of human adipose tissue (15
) allows more-detailed characterization of fat composition, compared with a 1
H spectrum. For example, the outer unsaturated carbons in a bisallylic group (-CH2
-CH = CH-CH2
-CH = CH-CH2
-) are resolved from the inner carbons in spectra obtained in vivo at 1.5 T. Natural abundance 13
C NMR is limited by relatively low sensitivity, low spatial resolution and additional technical requirements, all disadvantages compared with 1
H spectroscopy for human study. Conversely, 7T instruments are not widely available. However, because the number of 7T instruments in medical sites is increasing steadily and 1
H spectroscopy is routine on all systems, it seems reasonable to anticipate use of 1
H spectroscopy at 7T for clinical research.
At 7T, ten proton resonances in the chemical shift span of 0.9–5.3 ppm are resolved. The three resonances in the narrow range of 2.03–2.77 ppm provide an internal standard plus information about the abundance of protons between or adjacent to double bonds. This information, after correction for relaxation, allows calculation of the relative concentration of saturated, monounsaturated, and diunsaturated fatty acids. The proposed analysis of lipid composition was confirmed by comparison to authentic standards.
The fraction of saturated fats in triglycerides as determined by 1
H NMR spectroscopy at 7T, about 27–29%, is in excellent agreement with literature values. However, the fraction of fatty acids that are monounsaturated was somewhat less (about 46–50%) in this study, compared with biopsy studies, where about 55–60% monounsaturated fatty acids are typically found (22
). Consequently, the fraction of fatty acids estimated to be diunsaturated was about 23–24% in this study compared with about 17–18% in biopsy studies.
At least three technical issues may contribute error in this analysis. First, macromolecules or aqueous metabolites with chemical shifts overlapping the lipid resonances may alter the estimates of relative intensity. Second, the signals in vivo are the sum of very complex 1
H spin-coupled multiplets from several different fatty acids with slightly different chemical shifts. Consequently, a single Lorentzian-Gaussian line does not properly represent the observed lineshape. Even when high-resolution analytical 1
H NMR was used to measure marrow fat composition from samples extracted in chloroform, thereby removing aqueous species and other complicating factors that may contribute error in the study of intact tissue, the NMR method overestimated the fraction of saturated fatty acids, compared with “gold standard” gas chromatography of the same samples (12
). Third, as emphasized earlier (20
), correction for T2
is also essential and more important than correction for T1
differences. For a typical TE of 11 ms, the relative difference in T2
correction among the resonances D, E, and F accounts for ~4%, but it reaches 8% at TE = 20 ms, and is as large as 18% at TE = 40 ms. This compares to only about 1% difference after correction for T1
effects for the same three resonances at TR = 2 s. The T1
correction is not needed when the spectrum is collected with TR of 4 s. It is found that the calf subcutaneous fat has relatively shorter (~7%) proton relaxation times than the marrow (), indicating a larger local Bo field inhomogeneity on subcutaneous tissue. This can be understood from the difference in the microscopic structure between these two types of adipose tissues, as shown in the MRI images (, insert), in which subcutaneous tissue is seen as packed with large fat cells of different sizes, embedded with rich vasculature, and curvedly shaped, whereas the bone marrow appears as fine uniform structure, aligned straight inside the tibial bone, and nearly parallel to the Bo field. Because of this, the spectral resolution from bone marrow is generally superior to that from subcutaneous fat, as shown in .
It is probably unrealistic to anticipate perfect correlation between this MR analysis of extremity fat and marrow compared with literature data obtained largely from the abdomen, chest, or buttocks. There are complex effects of diet, season, gender, and anatomical site on fatty acid composition (24
). For example, extremity adipose tissue has a greater fraction of monounsaturated fat and less saturated fat compared with subcutaneous fat from the chest and abdomen (25
). These factors were not controlled in the current study. Nevertheless, the average ratio fdi
(47% for subcutaneous fat and 53% for bone marrow, ) obtained from this 1
H MRS study for 20 adult subjects (average age 34), is in excellent agreement with a previously reported value (50%) for the ratio of polyunsaturated to monounsaturated fatty acid by in vivo 13
C study (19
) of calf subcutaneous tissue of 17 adult subjects (average age 30). The obvious advantage of 1
H MRS is that it is fast (less than 1 min), has a low specific absorption rate (no need for broadband decoupling as in 13
C), is easily localized (not a volume detection), and is typically more easily implemented (no special scanner hardware or coil required). The high resolution and sensitivity of the current STEAM-based 1
H MR analysis also enable the detection of fat composition variation among uncontrolled healthy subjects (), which might be attributed to the difference in individuals’ long-term diet and metabolic genetics. Therefore, it is anticipated that the method may be useful in rapid detection of substantial changes in the composition of fatty acids in response to diet, exercise, and fat-metabolic diseases. This method, in our view, would help advance noninvasive human lipid research, which has been focused on studying fat mass distribution by MRI and other techniques rather than fat molecular structural composition. In addition, an analysis of our data (not shown) indicates no obvious difference in fat composition between male and female subjects, although, on average, the calf subcutaneous fat in female is approximately two-fold thicker than in male.
The decision to simplify the description of fatty acid composition to saturated, monounsaturated, and diunsaturated fats was based on two considerations. First, the amount of polyunsaturated fats (three or more double bonds) in human adipose tissue contributes very little, less than 1–2%, to the total 1
H NMR signal. Second, earlier studies that were able to quantify triply unsaturated fatty acids separate from other fatty acids required high-resolution analytical conditions (10
). For practical purposes in humans, this achievement would mean the capacity to separately quantify oleic (18:1), linoleic (18:2), and linolenic (18:3) acid. However, at the resolution achieved in these studies, a 50:50 mixture of triolein:trilinolenin is indistinguishable from a spectrum of trilinolein. For these reasons, the description of fatty acids as saturated, monounsaturated, or diunsaturated seems a reasonable simplification.
Frequency-dependent spatial localization introduces a chemical shift displacement artifact in the 1H MR spectrum. Quantitation may be inaccurate if it is based on ratios of two resonances when the region of interest is close to the inter-tissue boundary. Although this effect was not critically evaluated in this study, the use of a small voxel (~0.1 ml) well within the bone marrow plus analysis of resonances over a narrow chemical shift range reduces susceptibility to chemical shift displacement effects. Unlike the methyl resonance, the resonance of protons α to the carbonyl is not affected by the neighboring large bulk CH2 (resonance B), which often induces an uneven baseline at nearby resonances at shorter TEs. Although this baseline problem can be improved at longer TEs, the faster decay of protons α to a double bond (resonance D, T2 = 42 ms for marrow, and 39 for subcutaneous fat) than the reference resonance E (T2 = 59 ms) makes long TE scan conditions less favorable. STEAM sequence has the advantage of shorter TE than PRESS and less dependence on proton J modulation due to the smaller TE/T2 ratio, and therefore was chosen for the quantification of fat composition in this method.
In addition to composition analysis, the well-resolved resonances of human tibial bone marrow and calf subcutaneous fat at 7T could also be very useful as an internal standard to quantify intramuscular lipids (26
), which have attracted great research interest over the last few years, owing to their relation to insulin sensitivity, diabetes, and obesity (27
In conclusion, this 1H MRS study at 7T indicates that the rapid acquisition of high-quality 1H NMR spectra from subcutaneous fat and from bone marrow is straightforward in healthy volunteers, and that the spectra may be analyzed to assess triglyceride composition. This may facilitate longitudinal monitoring of changes in lipid composition in response to diet, exercise, and disease.