As the skeleton ages, the risk of fracture increases. This increasing propensity to fracture is not solely due to the well-documented loss of bone mass with aging but also to a deterioration in strength and toughness of bone (bone quality) (1). Clinical measurements of bone mass by gold-standard dual-energy X-ray absorptiometry are relatively insensitive to soft tissue characteristics in bone such as water molecule distribution, which has been conclusively linked to the energy dissipation mechanisms of the bone matrix (2–8) and likely plays a pivotal role in bone quality. Unlike X-ray-based imaging modalities, MRI is highly sensitive to the water microenvironment in tissue, as well as other molecular sources of 1H, and offers the potential to noninvasively evaluate bone quality. While conventional MRI methods are not suitable for imaging cortical bone because of its low proton density and short T2 constants, so-called ultrashort echo time and related MRI methods have opened clinical MRI to hard tissue imaging (9–11). Recently, these methods have produced high-quality images of human cortical bone (HCB) (10, 12) and raised the question of what bone characteristics can be derived from such images. Presented herein is a phenomenological and biophysical characterization of the ultrashort echo time MRI-visible 1H NMR signal of HCB. Specifically, unique contributions to the net HCB NMR signal are determined for both bound and mobile water, lipids, and bone matrix collagen. Knowledge of these signal contributions and their NMR properties is critical for future development and interpretation of cortical bone MRI and may form the foundation for a novel diagnostic of bone health.
Cortical bone is a complex composite of nano- and microstructural components, including type-I collagen, calcium phosphate mineral deposits, interosteonal lipids, and porous spaces such as haversian canals and the lacunar-canalicular system, which generally occur in repeating units of osteons (Fig. 1) (13–15). It is likely that each of these tissue components includes one or more distinct proton microenvironments, which differ according to their local chemical, magnetic, and electrical milieu. For example, the collagen component is host to both macromolecular-adsorbed water and covalently bound backbone protons, while mineral and porous spaces contain water with varying degrees of motional restriction, and lipids can be found in the cement line spaces between osteons (14). These proton microenvironments are too small to be spatially resolved with clinical MRI, but their 1H NMR signal contributions may be resolved by decomposing the net NMR signal into chemical shift and relaxation components. Presented here is a series of NMR measurements at 4.7 T, relevant signal decompositions, and physical manipulations of HCB specimens from cadaveric donors to identify common signal components and their microanatomic origins.
MATERIALS AND METHODS
The Musculoskeletal Tissue Foundation (Edison, NJ), a nonprofit tissue allograft bank, and the Vanderbilt Donor Program (Nashville, TN) supplied human femurs from six cadaveric donors (four male, two female, aged 21–94 years old, mean ± standard deviation: 60 ± 31 years) under instruction to not provide tissue from donors who had tested positive for a blood-borne pathogen (e.g., human immunodeficiency virus or hepatitis C). One HCB specimen was extracted from the medial midshaft of each donor's right femur and was machined to 15 × 5 × 5 mm dimensions via end mill. During dimensioning, care was taken to remove endosteal and periosteal surfaces such that the final specimens for NMR measurement were pure cortical bone. Specimens were stored in phosphate-buffered saline at −20°C between processing and NMR measurements, and specimens were thawed at 4°C approximately 18 h prior to NMR measurements. Immediately prior to NMR measurements, thawed specimens were removed from phosphate-buffered saline and blotted dry to remove the large amount of pooled surface water that remained. Specimens endured no more than three freeze-thaw cycles, and separate experiments found that up to six freeze-thaw cycles had negligible impact on the NMR properties. Final HCB dimensions were measured with digital caliper for volume determination.
μCT was performed on 2 × 4 × 4 mm HCB specimens (harvested adjacent to NMR specimen femoral sites) with a Scanco μCT 40 scanner (Scanco Medical, Brüttisellen, Switzerland) at 6-μm isotropic resolution, which was sufficient to image the Haversian canal and vascular porosity. The osteons of the cortical specimen were aligned with the scanning axis using the standard tube. At a resolution of 6-μm voxel size (isotropic), a portion of each specimen (2.4 mm in length) was scanned, acquiring 1000 projections per 180°. X-ray source settings did not vary among the specimens. After reconstruction, the Scanco thresholding procedure segmented bone tissue from air or soft tissue. Applying the Scanco evaluation software to the cortical bone, we quantified porosity, defined as one minus cortical bone volume per total volume.
The HCB specimens were initially assessed with measurements of their 1H NMR resonance spectra and T2 decays, probing NMR signal decay features that dominate ultrashort echo time MRI contrast. NMR measurements were performed in a Varian 4.7-T horizontal-bore magnet with a direct drive receiver (Varian Medical Systems, Palo Alto, CA). An in-house loop-gap style radiofrequency coil was used with Teflon structural support, which gave negligible background 1H signal (≈1% of net HCB signal). Pulse sequences used 90° excitation pulses of ≈8-μs duration and (where relevant) 180° refocusing/inversion pulses of ≈16-μs duration. Free induction decays (FIDs) were collected at 5-MHz bandwidth and 20-Hz nominal spectral resolution from a pulse-acquire sequence, with an 8-μs dead time after excitation to allow for coil ringing to subside. Four excitations were averaged, using a cyclops phase-cycling scheme. Resonance spectra were calculated from FIDs by discrete Fourier transform and manual phasing. To characterize T2, Carr-Purcell-Meiboom-Gill (CPMG) (16) measurements with a total of 10,000 echoes were collected at 100-μs echo spacing (first echo at echo time = 100 μs), yielding data that were fitted with a T2 spectrum (details below). Four excitations were averaged using a 90(x,-x,x,-x)/180(y,y,-y,-y) phase cycling scheme. Echo spacing was minimized while avoiding spin locking effects, which manifested as prolonged apparent T2s when echo spacing was <100 μs. In all measurements, HCB specimens were placed with osteonal direction orthogonal to the static field to avoid altering any potential magic angle effects across measurements.
Further investigation of both the FID and CPMG signal included three additional NMR studies: (1) inversion-recovery-prepared CPMG (IR-CPMG), (2) T2-T2 relaxation exchange spectroscopy (REXSY) (17), and (3) dynamic study of FID and CPMG signal responses to deuterium oxide (D2O) immersion. The IR-CPMG measurements involved an IR preparation with variable recovery time preceding a CPMG acquisition, as described above. Twenty-four recovery times, log-spaced between 1 ms and 10 sec, were used for each sample and the resulting data were fitted with a two-dimensional T1-T2 spectrum (18) (details below). The REXSY measurements involved a CPMG-like preparation with variable number of echoes, followed by a storage/mixing period then a CPMG acquisition, as described above. Thirty-two different CPMG preparations were used, with echo spacing of 100 μs and number of echoes pseudo-log spaced between 1 and 1000. For all acquisitions, the mixing period was 200 ms, and the resulting data were fitted with a T2-T2 spectrum. REXSY measurements used an additional phase cycle of the storage pulse (to store T2-prepared magnetization on the ± z axis during the mixing period), resulting in a total of four averaged excitations. (For further details on using REXSY for exchange measurements see Dortch et al. (19)). Finally, after all aforementioned NMR measurements, specimens were placed in an isotonic D2O solution to study effects of proton-deuteron exchange on NMR properties for determination of proton origins. This solution was made by mixing a volume of 99.9% isotopic purity D2O (Sigma-Aldrich Corp., St. Louis, MO) with the appropriate mass of phosphate-buffered saline electrolyte tablets (MP Biomedicals, Solon, OH) to yield 1× phosphate-buffered saline. Each bone specimen was immersed in a volume of isotonic phosphate-buffered saline that was 100× larger than the specimen volume, and immersed specimens were maintained on a shaker table at 60 revolutions per minute and 4°C until steady state was achieved (as determined by monitoring NMR resonance spectra). Final NMR measurements were then collected for comparison to initial measurements.
NMR Data Fitting
After phasing, real-valued NMR resonance spectra were fitted to a sum of three Lorentzian line shapes with a constrained Levenberg-Marquardt algorithm. For each spectrum, visual inspection provided initial guesses of line widths and frequency offsets so as to improve algorithm convergence, and final inspections of residuals and mean square errors verified proper fitting.
The 10,000 CPMG echo magnitudes were down-sampled in a pseudo-log fashion to 1024 echoes and fitted to a sum of 128 decaying exponential functions (with time constants log-spaced between 20 μs and 10 sec) in a nonnegative least-squares sense, subject to a minimum curvature constraint, which produced a so-called T2 spectrum (20). IR-CPMG data were reduced by singular-value decomposition (21) prior to two-dimensional nonnegative least-squares fitting (18) to the aforementioned range of decaying exponentials, producing a so-called T1-T2 spectrum. REXSY data were processed in a similar manner, yielding T2-T2 spectra. All data processing was performed with MATLAB (The MathWorks, Natick, MA). Results are presented as mean ± 1 standard deviation across samples.
In order to quantitatively compare the absolute signal amplitudes of T2 spectra across specimens and days, a 20-μL H2O reference sample with long T2 (≈2.5 sec) was included in each CPMG measurement. This reference sample, together with its known molar 1H density and the known HCB specimen volumes, enabled the calculation of proton concentrations in the bulk bone specimens for each CPMG relaxation component by comparing integrated areas of each T2 spectral component to the area of the marker. No such reference was used during resonance spectrum measurements, but the intermeasurement standard deviation of the reference sample's signal intensity was 4.8%, so it was concluded that the absolute amplitudes of the resonance spectral components were reliable to approximately this degree of precision.
FID-derived NMR resonance spectra (Fig. 2a,b) revealed three distinct signal components, as determined by Lorentzian fitting: two relatively narrow-band, slow-relaxing chemically shifted components were consistently observed atop a broadband, fast-relaxing component. In all specimens, the narrow-band component line widths were <5 parts per million (ppm) and the broad-band component line widths were >100 ppm, giving clear discrimination between slow- and fast-relaxing species. The dominant narrow-band component accounted for 57.7 ± 3.9% of the total FID signal. The second narrow-band component was consistently chemically shifted from the first component by −4.0 ± 0.2 ppm and represented 4.6 ± 3.2% of the FID signal. The broadband component accounted for the remaining 33.7 ± 2.5% of the FID signal and had a −33 ± 6.8 ppm chemical shift from the dominant narrow-band component. T*2s of 736 ± 46 μs, 1700 ± 470 μs, and 11.8 ± 1.1 μs for the on-resonant narrow-band, off-resonant narrow-band, and broadband components, respectively, were calculated from the components' line widths.
Figure 2c shows CPMG-derived T2 spectra from each HCB specimen, demonstrating two distinct submillisecond relaxation components and a broad collection of signals spanning the millisecond-second T2 domain, which hereafter is collectively considered a third T2 component for discussion purposes. The two submillisecond components had T2s of 57 ± 4 μs and 416 ± 35 μs, with respective CPMG bone signal fractions of 16.1 ± 1.3% and 60.8 ± 5.3%. The long-lived third T2 component accounted for the remaining 23.0 ± 6.5% of the CPMG signal. Comparing these signal fractions to the concomitantly measured T2 marker (see NMR Data Fitting) gave proton concentrations of 6.6 ± 1.5, 24.8 ± 5.8, and 9.2 ± 2.3 mol 1H/Lbone for the ≈60-μs, ≈400-μs, and long-lived T2 components, respectively. Note that typical CPMG data signal-to-noise ratio was ≈7000 (integrated spectral area divided by the standard deviation of the residuals from the T2 spectrum fit), which was ample to fit the observed 60-μs T2 component. In preliminary studies, general characteristics of both resonance and T2 spectra did not change substantially between 0.5 T and 9.4 T static magnetic field strengths (data not shown), so the three FID and three CPMG signal components should be relevant to all mainstream laboratory and clinical HCB NMR.
For all HCB specimens, T1-T2 spectra from IR-CPMG measurements (Fig. 3) showed T1 values ranging from ≈350 ms for the two short-lived T2 components to ≈1 sec for much of the long-lived T2 component. A biexponential T1 relaxation for the ≈400 μs T2 component was consistently observed, characteristic of magnetization transfer between this component and a shorter-T2 proton pool (22). Observable T2 proton pools participating in this magnetization transfer were identified from REXSY T2-T2 spectra (Fig. 4). The main diagonal in the T2-T2 spectrum is analogous to the one-dimensional CPMG-derived T2 spectrum, while off-diagonal signals arise from an exchange of magnetization between T2 components during the 200-ms mixing period (similar to EXSY experiments in chemical-shift-resolved studies (23)). As such, REXSY consistently indicated exchange between the two short-lived T2 components (≈60 μs and ≈400 μs). No significant exchange involving the long-lived T2 signals was observed in any of the HCB samples.
Resonance and T2 spectra from a representative HCB specimen undergoing prolonged D2O immersion are shown in Fig. 5a,b. Since the D2O deuteron is undetectable in 1H NMR measurements and may freely exchange with various protons in bone, these spectra demonstrate a combination of 1H signal components that wash out with D2O (i.e., water and chemically exchangeable protons such as amines and hydroxyls) and those that do not (methylene protons). Of the three resonance spectra components, the dominant on-resonant narrow-band component was the only component that was nearly completely washed out (>95%). The off-resonant narrow-band component was unaffected by D2O immersion, as was 83 ± 2% of the broadband component (17 ± 2% washed out). Likewise, the only T2 component effectively removed by D2O was the ≈400 μs T2 (>98% washed out). The ≈60-μs T2 component was unaffected by D2O immersion (to within the tolerances associated with CPMG data fitting), and 58 ± 12% of the long-lived T2 component was removed by D2O.
In animal (24–26) or human (27–31) cortical bones, previous studies have identified various subsets of the six NMR components observed herein but have not entirely agreed upon or thoroughly explored their biophysical origins. In particular, the large, submillisecond T2 component has been attributed to both macromolecular-bound water (25, 27–31) and porous water of the lacunae and canaliculi (12, 26). Because this signal will dominate most standard ultrashort echo time MRI scans of cortical bone, understanding its biophysical origin and relation to all other signal components is important for interpreting cortical bone MRI. By comparing results from our combination of studies, we have reasoned the biophysical origins of all the FID and CPMG signal components. These assignments and relevant data are summarized in Fig. 6 and the rationale is outlined as follows.
Based on its chemical shift, endurance through D2O immersion, and relatively slow relaxation rate, the off-resonant FID component must arise from mobile methylene protons—most likely found on lipids (as similarly concluded by Fernandez-Seara et al. (6)). The line width of this component (≈1 ppm) corresponds to a T*2 ≈1.6 ms; therefore, it must be present as a fraction of the long-T2 component (≈45%) that also survives D2O immersion. Aside from the lipid signal, most of the remaining signal that survives D2O immersion—the broadband FID and short-lived CPMG components—must, by nature of their very short T2/T*2 signals, be derived from relatively immobile methylene protons. The majority of these protons must belong to collagen macromolecules, which represent >90% of the organic mass in HCB (15). Note that both the broadband FID and short-lived CPMG components may arise from the same collagen molecules, but they likely do not originate from the same protons due to the ≈70,000 sec−1 difference in their relaxation rates (1/T*2 and 1/T2), which is too large to be entirely caused by static magnetic field variations and must arise from different molecular dynamics. It is conceivable that the faster-relaxing FID broadband protons occupy methylene sites on the semicrystalline collagen triple helix backbone, while the slower-relaxing ≈60-μs T2 CPMG protons arise from more mobile collagen methylene, such as side chain positions and the amorphous domains between crystalline regions. While previous cortical bone studies have not distinguished these two signal components, a study of dentin—a bonelike material—found two FID components with T*2 ≈12 and 60 μs (32).
In contrast to the signal that survives D2O, the signal that washes out during D2O immersion must be a combination of water and non-water exchangeable protons (NWEPs, i.e., exchangeable protons that are not found on water). NWEPs in HCB arise chiefly from amides/hydroxides in collagen and hydroxides in bone minerals, and NWEP abundance can be estimated as follows: assuming that HCB collagen, on average, can be represented by the well-studied model sequences poly(Gly Pro-Pro) (33) and poly(Pro-Hyp-Gly) (34), only 6–12% of collagen protons are exchangeable with D2O, representing 0.4–0.75% of total collagen mass. Given that 1 mL of HCB contains an average of 0.48 g organic content (35)—greater than 90% of which is collagen (15)—the expected concentration of exchangeable collagen protons in bone is on the order of 1.7–3.2 mol 1H/Lbone. A similar analysis of bone mineral, using 1.2 g mineral/mL bone (35) and a 0.042% weight fraction of exchangeable hydroxide protons (formulated from Cho et al. (36)), gives an expected concentration of 0.5 mol 1H/Lbone. Thus, collagen and mineral sources together give a total NWEP abundance of approximately 2–4 mol 1H/Lbone. Since NWEPs originate from rather immobile proton sites and are expected to be fast relaxing, it is likely that the fraction of the broadband FID component removed by D2O (≈6% of the total FID signal) accounts for the entire NWEP population. However, it is possible that some of the NWEPs may contribute to the longer-lived FID and CPMG spectral components, but given that an average of 29.5 ± 7.2 mol 1H/Lbone signal was removed from CPMG spectra, we can conclude that this longer-lived signal is predominantly water. Unlike NWEPs, the so-called crystal water protons—ultra-fast-relaxing water in HCB mineral crystal lattice vacancies (37)—must be confined to the small fraction of the broadband FID component removed by D2O, although the relative contributions of NWEPS and crystal water to this broadband fraction cannot be determined from the measurements herein. With this interpretation, the range of water volume fractions (assuming 1 g/mL water density) found in our six HCB specimens was 17.3–35.6%, which is similar to that found from a cohort of pre- and postmenopausal volunteers and renal osteodystrophy patients (12).
The water signal can be further decomposed into two subpopulations: collagen-bound water (again, collagen being the predominant macromolecule) and relatively free water in porous spaces. Based on μCT at 6-μm isotropic resolution, Haversian canal porosity of the HCB specimens in this study was 4.0 ± 1.0%, and we estimate a similar volume of lacunar-canalicular pores (14), making the total average pore volume of the samples ≈8%, or ≈9 mol 1H/Lbone. We expect the pore water to have a longer T2 than the bound water, and ≈1/2 of estimated pore water volume can be accounted for by the 5.1 ± 2.2 mol 1H/Lbone that washed out of the long-T2 component. The remaining pore water may be included in the ≈400 μs T2 CPMG component (accounting for no more than ≈15% of it) and/or some of this water may have been wicked inadvertently from the larger pores during sample handling. Thus, including possible contributions of both NWEPs and porous water, the remainder of the ≈400 μs T2 CPMG component must be at least ≈70% collagen-bound water (or 17.4–24.8 mol 1H/Lbone). This assignment is roughly consistent with literature estimates of 0.48 g collagen per mL of HCB (35) and 0.49 g water bound per gram collagen in bone, which together equate to an expected bound-water proton concentration of ≈26 mol 1H/Lbone.
The assignments of the ≈60-μs and ≈400-μs components to collagen methylene protons and collagen-bound water, respectively, are also consistent with the observation of exchange between these components seen in our RESXY measurements (Fig. 4). Interestingly, the observation that REXSY demonstrates exchange between these two components but only the ≈400-μs T2 component is removed by D2O indicates that the observed magnetization exchange is mediated by a through-space, dipole-dipole interaction, rather than chemical exchange, which further supports our biophysical assignments. It also is noteworthy that our biophysical assignments in HCB are similar to previous findings in human dentin (32) (determined through 1H NMR relaxometry and sample dehydration), with one particular exception: the ≈60-μs decay component in dentin was attributed to a mixture of hydroxide and water, in addition to protein. At least one other study has identified very-short-lived water in bone mineral (37); however, it disappeared with dehydration and so we presume it would be washed out with D2O. Herein, we conclude that the 60-μs T2 component is predominantly protein derived because it survives D2O immersion.
Finally, it should be noted that the broad age range of the HCB donors herein allowed the study of common relaxation features across presumably diverse physiologic states. Interestingly, relaxation rates among all the donors were similar, indicating that the chemical nature of various bone proton milieux persist during aging. The dominant source of variability observed among donors was in the relaxation component pool sizes; for example, the oldest and most visibly osteoporotic donor had a larger long-lived T2 component (Fig. 2c, turquoise line) than the other donors. However, a thorough analysis of the interdonor relaxation differences would require a significantly larger number of donors, which is beyond the scope of this work but is the subject of future study.
In summary, 1H NMR signal from HCB was found to exhibit a broad distribution of transverse relaxation components attributable to known bone proton sources (Fig. 6) as follows: i) T*2 ≈12 μs, derived ≈80% from collagen backbone methylene protons and the remainder from NWEPs (amide/hydroxide) and possibly adsorbed mineral water, ii) T2 ≈60 μs, derived predominantly from collagen side-chain or otherwise mobile methylene protons, iii) T2 ≈400 μs, derived predominantly from water bound to the collagen responsible for the 12-μs T*2 and 60-μs T2 components, iv) T2 ≈milliseconds-second, derived ≈60% from pore water and ≈40% from lipid methylene protons. These findings indicate that modern ultrashort echo time MRI of cortical bone is dominated by signal from water bound to bone matrix collagen and that commonly used long-T2 suppression techniques for enhancing bone contrast will suppress signal from lipids and pore space water that may be of biological/clinical interest. Also, while some short-T2 signals in HCB may be difficult to quantify directly, the interaction between their underlying proton pool and the dominant water signal offers the potential to probe prohibitively fast-relaxing HCB signals through magnetization transfer contrast. Finally, there was considerable variability in the lipid and pore water content across the diverse (albeit small number of) HCB samples studied herein, indicating these two metrics may contain the useful information on bone quality and health.