FID signals of tendon and cortical bone samples before and after treatment with a procedure to remove the water signal.
The magnitude has been normalized by the maximum signal observed in the unprepared tendon. The shaded areas around the lines indicate the 90% central ranges of the averaged data, which are generally very narrow and only large for the phases of the treated samples. A rapidly decaying signal, attributed chiefly to collagen, is observed on top of the slower-decaying collagen-bound water signal for untreated samples. Much longer decaying components from free water and fat are also present but hardly differentiable at this timescale. The collagen component appears to have completely decayed after ∼40 μs. The treated bone signal decays into the noise floor, as also seen in the phase of the signal. The treated tendon signal remains above the noise floor due to the presence of longer-living fat signal in the tissue. The distinct bump observed in the magnitude data (for treated and untreated bone, and treated tendon) is attributed to an oscillation stemming from dipolar coupling of the collagen protons. The reduced amplitude of the rapidly decaying component after treatment is believed to be a result of H-D exchange of exchangeable protons on collagen and other molecules with short-lived signals. The variation in phase accrual observed before and after treatment indicates chemical shift differences of the signal contributors that persist post-treatment. The marked collagen decay region indicates the maximum time range available for spatial encoding in direct collagen imaging. In contrast, imaging based on collagen-bound water can utilize the full plotted range.
Direct MRI of collagen.
From an image series of collagen-rich tissue samples with increasing echo times (TEs), four examples at early TEs are shown. The magnitudes are normalized by the maximum signal in the shortest-TE image of the respective sample. A decrease in image intensity is observed with increasing TE, reflecting the decay of collagen signal. In the treated samples, the signal has virtually completely decayed by TE = 35.4 µs. The effect is less obvious in the untreated samples due to strong background signal from water contained in the samples. In the treated tendon sample, the bright fat signal appears constant due to negligible signal decay over the given timescale.
Decay of collagen signal observed with spatial localization by imaging at multiple echo times (TEs).
The magnitude of the mean signal in a region of interest (ROI) in the bone and tendon samples is plotted as a function of TE. The rapidly decaying collagen signal is observed in both the treated and untreated samples. These plots are equivalent to the FID signal magnitudes shown in Fig. 1, including the dipolar oscillation in bone and treated tendon at the interval of ∼20-40 µs. However, with imaging, the signal can be observed at specific locations. The shaded areas around the lines indicate the 90% central ranges of the averaged data points.
Selective imaging of collagen in tendon and bone samples.
Isolation of the collagen signal is achieved by subtraction of the shortest-TE image (10.4 µs) and an image with slightly longer TE (24.4 µs). The image intensities have been normalized by the maximum signal observed in the shortest TE image of the respective sample. The bright fat signal in the tendon has been clipped to visualize the collagen in the non-difference images. The images have been scaled, as indicated, for visualization purposes. In the difference images, long-lived water and fat signals are suppressed, leaving only the short-lived components, which are attributed chiefly to collagen. For the treated samples, the effect of subtraction is negligible. The fat signal in treated tendon persists and is removed after subtraction, indicating that the treatment does not impact the fat and that longer-lived signals are suppressed by the subtraction procedure. The difference images of the untreated samples exhibit less T2 blurring than the treated samples due to residual signal contamination from the collagen-bound water signal (Supplementary S.IV). In the untreated tendon, the subtraction yields a bright region that is not visible in the treated counterpart. This suggests an effect of the treatment on the samples that is observable at the collagen timescale.
Direct collagen MRI of a right human forearm in vivo.
Three views of the forearm are shown at two short-TE acquisitions along with the collagen-selective difference image. Images have been normalized according to the maximum signal observed in the earliest TE image of the displayed slices and scaled as indicated. Note that the difference in contrast of the two raw transverse views is due to variations in transmit sensitivity of the RF coil. The difference images show collagen-rich anatomy such as a) cortical bone, b) tendon, c) skin and subcutaneous tissue, whereas longer-lived signals such as from d) trabecular bone are suppressed. Signal from padding, e), is also captured. A slight ringing artifact (typical of high-bandwidth radial acquisitions) is observed in the transverse view.
The short-T2 PETRA protocol employed for direct collagen MRI.
a) Basic zero echo time (ZTE) pulse sequence. After the RF dead time (DT), radial encoding is performed to collect either ZTE data along radial spokes in k-space or single-point-imaging data to fill the central k-space gap caused by the DT. b) Corresponding t(k) plot, showing the time after excitation when each k-space data point is acquired. Inside the gap, the SPI data forms a plateau, whereas outside, the ZTE data forms a linear increase. To observe signal decay, the DT is increased in successive imaging experiments. The k-space gap (kGap ∝ DT × G) is kept constant by reducing the gradient strength, leading to a steeper slope of t(k).
Parameters for multi-TE imaging of collagen-rich samples and in vivo human forearm.
Abbreviations: TE, echo time, TR, repetition time, BW, image bandwidth
Dominant compounds in tendon and cortical bone with relevant stoichiometric parameters.
The collagen chain of Glycine-Proline-Hydroxyproline is used since it is the most commonly occurring chain in collagen type I (68).
Estimated signal fractions for tendon and cortical bone tissues.
FID of untreated cortical bone sample with the accompanying model fit.
The bump in the magnitude is attributed to an oscillation from dipolar coupling of the collagen protons. The model fits the data well using three terms, of which two components (C1 and C2) are rapidly decaying dipolar coupling terms.
The amplitudes are normalized according to the maximum signal observed in the FID. Notably, the short-lived components make the largest contribution. The values shown in bold were fixed during the fitting procedure.
FID of treated tendon sample with the accompanying model fit.
Again, the bump in the FID from dipolar coupling is preserved. The model fits using 4 terms with C2 and C3 being rapidly decaying components.
The amplitudes are normalized according to the maximum signal observed in the FID. Here, the short-lived components dominate. The values shown in bold were fixed during the fitting procedure.
Exponential T2 as a function of time t according to Eqn. S4 for Gaussian-like decay for parameters µs and E = 1.8 (the fastest decaying component in Table S1).
One observes that the equivalent exponential T2 decreases with increasing time. At the indicated t = 40 µs, T2 is approximately 12 µs.
Parameters used for calculation of T2 blurring.
PSF of PETRA pulse sequence for collagen signal with exponential T2 decay of 12 µs, compared to that of a non-decaying signal, using imaging parameters listed in Table S3.
Imaging simulation of four spheres using collagen, bound-water, and bulk-water signal characteristics.
(i) a “Collagen” component with dipolar coupling (T2 = 10 μs, f = 10 kHz, E=1, Δω = 0.5 kHz), (ii) a “Bound water” component (T2 = 100 μs, f = 0.7 kHz, E=1), (iii) a “Bulk water” component (T2 = 1.3 ms), and (iv) all components combined (adding all components into a single sphere). The spheres were simulated with a PETRA sequence with TE ranging from 10-320 μs. The displayed images are scaled with respect to their own maximum intensity. a) Simulated image at TE = 10 μs. The collagen sphere shows reduced signal and significant blurring. b) Simulated image at TE = 25 μs. The collagen sphere has vanished while the other spheres persist. c) Difference image of TE = 10 μs and TE = 25 μs. Only the collagen component remains. In addition, a residual ring of the intermediate component is observed, indicating a small signal contamination of ∼4%. This ring also appears in the combined sphere, making it slightly sharper than the pure collagen sphere. d) Plotted mean signal intensities over the ROIs drawn over the different spheres as a function of TE. The behavior of the underlying FID signal is reproduced, including the rapid initial signal decay as captured by the shortest TEs. e) and f) Intensity profiles along spheres, illustrating effects of T2 blurring. The profiles show that the collagen sphere is notably more blurred than the bound-water and combined spheres. Furthermore, one observes that the contamination from the bound water adds to the difference profile of the combined signal sphere, resulting in a sharper-looking image. This explains why the T2 blurring from the collagen-only sphere appears more apparent.
Mean signal intensity as a function of echo time (TE) over regions of interest (ROIs) in different parts of the untreated tendon sample.
The signal in the bright region of the difference image d) (ROI shown in panel c) for the TE = 10.4 µs image) starts at a larger amplitude, initially decays faster, and shows a change in bound-water behavior after ∼150 µs as compared to the darker region (ROI shown in panel b) for the TE=10.4 µs image). This change in signal characteristics is hypothesized to be a result of variation in tissue structure. The shaded areas around the lines indicate the 90% central ranges of the averaged data points.
ROIs drawn on collagen image of in vivo forearm for tendon (left) and cortical bone (right) structures.
Anatomical ROIs are shown in yellow, and the noise ROIs are shown in pink. The SNR is calculated to be 24.9 and 16.6 respectively, including the three-slice averaging used for visualization.
Comparison of TEs between Petra and SPI acquisitions.
a) ROI plot for simulated SPI acquisition using the same four spheres presented in Fig. S5. The simulated SPI acquisition yields uniform T2 weighting to all points in k-space and its TE is assigned as TE = DT + Δ. b) difference between signals of different components in the SPI and PETRA acquisitions. The signal difference is small, and this supports the same assignment of TE for the PETRA pulse sequences as for the SPI pulse sequence.
