Inversion recovery ultrashort echo time imaging of ultrashort T2 tissue components in ovine brain at 3 T: a sequential D2O exchange study

Inversion recovery ultrashort echo time (IR‐UTE) imaging holds the potential to directly characterize MR signals from ultrashort T2 tissue components (STCs), such as collagen in cartilage and myelin in brain. The application of IR‐UTE for myelin imaging has been challenging because of the high water content in brain and the possibility that the ultrashort T2* signals are contaminated by water protons, including those associated with myelin sheaths. This study investigated such a possibility in an ovine brain D2O exchange model and explored the potential of IR‐UTE imaging for the quantification of ultrashort T2* signals in both white and gray matter at 3 T. Six specimens were examined before and after sequential immersion in 99.9% D2O. Long T2 MR signals were measured using a clinical proton density‐weighted fast spin echo (PD‐FSE) sequence. IR‐UTE images were first acquired with different inversion times to determine the optimal inversion time to null the long T2 signals (TInull). Then, at this TInull, images with echo times (TEs) of 0.01–4 ms were acquired to measure the T2* values of STCs. The PD‐FSE signal dropped to near zero after 24 h of immersion in D2O. A wide range of TInull values were used at different time points (240–330 ms for white matter and 320–350 ms for gray matter at TR = 1000 ms) because the T1 values of the long T2 tissue components changed significantly. The T2* values of STCs were 200–300 μs in both white and gray matter (comparable with the values obtained from myelin powder and its mixture with D2O or H2O), and showed minimal changes after sequential immersion. The ultrashort T2* signals seen on IR‐UTE images are unlikely to be from water protons as they are exchangeable with deuterons in D2O. The source is more likely to be myelin itself in white matter, and might also be associated with other membranous structures in gray matter.

obtained with these techniques have not been strongly correlated with clinical manifestations. [10][11][12][13][14][15][16][17] It is possible that this lack of correlation is because these measures cannot differentiate demyelination and remyelination from other pathological substrates, such as axonal loss and gliosis, which are associated with heterogeneous clinical manifestations of MS. 18,19 Methods for direct myelin imaging in vivo provide a more specific and sensitive evaluation of myelin, and thus might be of considerable value.
Previous nuclear magnetic resonance (NMR) spectroscopy studies have measured tissue myelin T 2 ex vivo and have reported values of approximately 50 μs in fixed human WM 13 and 50 μs to 1 ms in freshly excited frog sciatic nerve. 4 Inversion recovery ultrashort echo time (IR-UTE) sequences have the potential to selectively image short-lived MRI signals in vivo through the efficient suppression of long-lived water signals using an adiabatic inversion recovery pulse. 20, 21 Wilhelm et al 2 first demonstrated IR-UTE imaging of myelin signals in excised rat spinal cord on a 9.4-T microimaging system. A more recent study explored the use of IR-UTE sequences to detect small MS lesions in cadaver brains, as well as to measure ultrashort T 2 * values in WM both ex vivo and in vivo on a 3-T clinical scanner. 22 The reported T 2 * values (~110-330 μs) were comparable with those of myelin lipid powder when measured either alone or in a mixture with H 2 O or D 2 O. 22 However, because brain tissue contains different water pools (termed as long T 2 tissue components in this article) that might have different T 1 values, [5][6][7][8]23 it can be challenging to completely null water signals. Therefore, there is a possibility that the ultrashort T 2 * signals seen on IR-UTE images are contaminated by water signals.
The present study aimed to explore myelin as a source of the ultrashort T 2 signals seen in WM on IR-UTE images, and to assess the feasibility of using IR-UTE to image the ultrashort T 2 tissue components (STCs) in GM on a 3-T clinical scanner in an ovine brain D 2 O exchange model. Water protons in these specimens were sequentially replaced by deuterons in D 2 O. If non-exchangeable protons, such as those in myelin, are the source of the ultrashort T 2 signals seen on IR-UTE images, the measured ultrashort T 2 * value should remain constant before and after sequential D 2 O exchange. Figure 1 shows the two-dimensional IR-UTE pulse sequence used in this study ( Figure 1A) and its contrast mechanism ( Figure 1B, C). The sequence employs a half radiofrequency (RF) pulse excitation (pulse duration, 472 μs; bandwidth,~2.7 kHz), followed by two-dimensional radial ramp sampling. 20 An initial adiabatic inversion pulse (duration, 8.64 ms; bandwidth,~1.5 kHz) was used to invert and null the longitudinal magnetization of long T 2 tissue components. To maximize the signal-to-noise ratio (SNR) and minimize eddy currents, slice-selective gradients were turned off to image the~3-5-mm-thick specimens used in this study.

| Pulse sequences and contrast mechanisms
To measure the ultrashort T 2 * signals in WM, the inversion time (TI) is chosen to null the long T 2 tissue components in WM (WM L ) ( Figure 1B).
At the time at which UTE acquisition starts (TE = 10 μs, i.e. TE 10μs ), the STCs in WM (WM S ) and GM (GM S ) have positive magnetization, whereas the long T 2 tissue components in GM (GM L ) have negative magnetization. The addition of the negative magnetization from GM L and the positive magnetization from GM S on the TE 10μs image leads to a smaller magnitude of the total GM signal, relative to that of GM L alone. At a relatively later TE (e.g. TE = 0.6 ms, i.e. TE 0.6ms ), the positive magnetization of GM S decays to zero or near zero (so does that of WM S ), whereas the negative magnetization of GM L decays very little. As a result, the magnitude of the total GM signal on TE 0.6ms is nearly the same as that of GM L alone.
Consequently, subtraction of the magnitude signal on the TE 0.6ms image from that on the TE 10μs image leads to a negative signal for GM and positive signal for WM S , i.e. WM S is selectively imaged.
With TI chosen to null GM L ( Figure 1C), both WM S and GM S have positive magnetization at TE 10μs , which decays to zero or near zero at TE 0.6ms . WM L has positive magnetization at TE 10μs , which only decays slightly at TE 0.6ms . Subtraction of the magnitude signal on the TE 0.6ms image from that on the TE 10μs image highlights both WM S and GM S , with WM generally showing higher signal intensity, possibly as a result of its higher myelin content than GM. 4,21 2.2 | Ovine brain specimen preparation Six specimens containing cerebral hemisphere (n = 5) and cerebellum (n = 1) were prepared from three freshly frozen ovine brains. All specimens were stored at 4°C and pretreated with broad-spectrum antibiotics in saline before the initial imaging (an initial image for cerebral hemisphere specimen no. 4 was not available).
Each cerebral hemisphere or cerebellum specimen (~3 mm thick) was immersed in 10 mL D 2 O (99.9%, Sigma-Aldrich, St. Louis, MO, USA) in a 6.4-cm covered and sealed plastic dish to allow exchange. Four cerebral hemisphere specimens were imaged with IR-UTE sequences before and after 90-min, 150-min, 8-h and 24-h immersion intervals in D 2 O (for details, please see Table 1). One cerebral hemisphere specimen was imaged with both UTE and IR-UTE sequences before and after 24 h of immersion in D 2 O. The cerebellum specimen (~3 mm thick) was imaged after 8-h and 32-h immersion intervals in D 2 O. All specimens were allowed to reach room temperature before imaging. At each immersion time point, specimens were removed from the D 2 O incubation container and flushed with fresh D 2 O for 5 s prior to imaging. They were immersed in fresh D 2 O again after each imaging experiment.

| Imaging experiments
All specimens were imaged using a GE 3-T Signa TwinSpeed MR scanner (GE Healthcare Technologies, Milwaukee, MI, USA) and a 7.6-cm receiveonly coil. A conventional proton density-weighted fast spin echo (PD-FSE) sequence (TR/TE = 8000/13.5 ms) was used with a fixed receiver gain at all imaging time points to measure long T 2 proton signals and thereby to assess exchange between tissue protons and deuterium in D 2 O. An IR-UTE sequence with TR/TE = 1000/2.2 ms and variableTIs (20, 100, 300, 500 and 800 ms) was used to determine the optimal TI for nulling long T 2 signals in WM (WM L TI null ). Other parameters included a bandwidth of 166.6 kHz, a field of view (FOV) of 12 cm and an acquisition matrix of 96 × 96, providing a nominal voxel size of 0.94 × 0.94 × 3 mm 3 , with a scan time of 96 s per acquisition. The same IR-UTE sequence was then used with TI = WM L TI null and variable TEs (0.01, 0.1, 0.2, 0.4, 0.6, 2 and 4 ms) to measure T 2 * of WM S in all specimens. The T 2 * value of GM S was also measured in the cerebellum specimen using the same IR-UTE sequence with TI set to null long T 2 signals in GM (GM L TI null ). A UTE sequence without inversion preparation was performed (TR = 1000 ms; flip angle, 65°; 11 TEs ranging from 0.01 to 15 ms) in one cerebral hemisphere specimen to quantitatively estimate the fraction of STCs (f S ) before and after a 24-h immersion in D 2 O.

| Data analysis
Regions of interest (ROIs) were carefully chosen in WM and GM to avoid partial volume effects. T 1 and TI null were measured in each ROI offline using Matlab (Mathworks Inc., Natick, MA, USA) by fitting the IR-UTE DICOM images obtained with variable TIs using a single-component, , with an inversion time (TI) that is chosen to null WM L and GM L , respectively. The adiabatic inversion recovery pulse provides robust inversion of the longitudinal magnetization (M z ) of long T 2 tissue components. M z of the ultrashort T 2 tissue components experiences significant relaxation on the transverse plane during the long adiabatic inversion process, and is not inverted but partially saturated. At the time at which UTE acquisition starts (TE = 10 μs, i.E. TE 10μs ) and with TI chosen to null WM L B, GM L have negative and GM S have positive magnetization, leading to a smaller magnitude of the GM net magnetization. At a relatively longer TE (e.G. TE = 0.6 ms, i.E. TE 0.6ms ), the magnitude of the GM L signal is essentially the same as that at TE 10μs , whereas the WM S and GM S signals decay to zero or near zero. Subtraction of the magnitude signals seen in the image acquired at a longer TE from those seen on the TE 10μs image provides positive contrast for WM and negative contrast for GM, enabling exclusive visualization of WM S . With a TI chosen to null GM L C, both WM L and WM S have positive signals at TE 10μs , and the subtraction image provides positive contrast for both WM and GM, with WM generally showing a higher signal three-parameter fitting model. The T 2 * of STCs was quantified by fitting the IR-UTE DICOM images obtained with variable TEs using a single-component, three-parameter fitting model, 20 or fitting the UTE DICOM images with a previously reported bi-component fitting model. 24 Changes in PD-FSE signals, T 1 and T 2 * values were plotted against the immersion time in D 2 O. SNR was calculated as the ratio of the mean signal intensity inside an ROI to the standard deviation of the background noise. All images shown and used for analysis were magnitude and not phase sensitive.
3 | RESULTS Figure 2B shows representative two-dimensional IR-UTE images acquired with variable TIs (TR/TE = 1000/2.2 ms) from cerebral hemisphere specimen no. 1 before immersion in D 2 O. At TI = 20 and 100 ms, there was little or no contrast between WM and GM. At TI = 300 ms, there were near-zero signals in WM and a relatively high signal in GM, suggesting dramatic suppression of signals from WM L and insufficient suppression of signals from GM L . At TIs of 500 and 800 ms, WM and GM signals both increased, with the WM signal being greater than that from GM. Figure 2C shows the T 1 map of WM L and GM L calculated from the images shown in Figure 2B. Figure 2D shows typical T 1 fitting curves for a WM ROI and a GM ROI (red boxes). TI null values of~300 and~370 ms (TR = 1000 ms) were found for WM L and GM L , respectively ( Figure 2D). Figure 3A shows representative IR-UTE images with variable TEs acquired from cerebral hemisphere specimen no. 1 at WM L TI null , and the T 2 * map calculated from these images. WM had relatively lower signals than the surrounding GM, and these signals dropped progressively with the increase in TE. As shown on the T 2 * map, the WM S T 2 * was 200-300 μs. Figure 3B shows a subtraction image (the TE 10μs image minus the TE 0.60ms image) revealing a positive signal for WM and negative signal for GM. Figure 3C shows a representative mono-exponential T 2 * fitting curve for the ROI shown in Figure 3B (yellow box). Figure 4A shows PD-FSE images from cerebral hemisphere specimen no. 1 before and after sequential immersion in D 2 O for up to 24 h. Signals progressively decreased with increasing immersion time. The SNR decreased from 166 to 11.3 in a WM ROI (the uppermost yellow box in the 0-min PD-FSE image, Figure 4A). Figure 4B shows the corresponding IR-UTE TE 10μs images acquired at WM L TI null . The GM magnitude signals, arising from both short and long T 2 tissue components, progressively decreased with increasing immersion time, whereas the WM magnitude signals were relatively constant. After 24 h of immersion, the specimen, especially the WM, showed near-zero PD-FSE signals, but still had higher IR-UTE signals than the background, with an SNR of 21.4 in the abovementioned WM ROI (the uppermost yellow box in the 0-min PD-FSE image, Figure 4A). This was comparable with the SNR of the same ROI before exchange (SNR~20.9). Figure 4C shows the averaged longitudinal proton density (PD) signal changes in three WM ROIs (yellow boxes in the 0-min PD-FSE image, Figure 4A). Figure 4D depicts the average T 1 changes in these ROIs. T 1 decreased from 527 ± 2 ms at baseline to 417 ± 34 ms after 24 h of immersion in D 2 O. Figure 4E shows T 2 * values plotted against immersion time. As summarized in Table 1, the WM S T 2 * values were measured to be 197 and 248 μs before and after exchange, respectively, in four cerebral hemisphere specimens. Bi-component T 2 * analysis of the UTE images showed a very small f S (3.8%) before exchange and an increased f S (41%) after exchange, with an ultrashort T 2 * value constantly in the range 200-300 μs at both time points (i.e. 277 μs before exchange and 227 μs after exchange). Figure 6A shows a PD-FSE image of the cerebellum specimen before exchange with D 2 O. Figure 6B shows selected corresponding IR-UTE images acquired at variable TEs with TIs of 260 and 350 ms for nulling of WM L and GM L respectively. With TI = WM L TI null , the WM signal (from WM S only) decreased and the GM signal increased with increasing TE. It should be noted that, at TE = 10 μs, GM L and GM S signals had negative and positive values, respectively, leading to signal cancelation. At later TEs (4 ms), GM S signals decayed to near zero, whereas GM L signals were largely unchanged. Therefore, the subtraction image (the TE 10μs image minus the TE 0.6ms image) showed a positive signal for WM and a negative signal for GM ( Figure 6B, top panel). With TI = GM L TI null , both WM and GM signals decreased with increasing TE. The WM signal was generally higher than the GM signal. The subtraction image (the TE 10μs image minus the TE 0.6ms image) showed positive signals for both WM and GM ( Figure 6B, bottom   (TI = 20 and 100 ms), there is little or no contrast between WM and GM. With TI increased to 300 ms, WM L signals were largely suppressed and GM L signals were moderately suppressed, as evidenced by the near-zero signals in WM and dramatic signal reduction in GM. A further increase in TI resulted in a signal increase in both WM and GM, with WM showing a higher signal than GM. C, T 1 map calculated from the images shown in (B). The three purple boxes represent three regions of interest (ROIs) defined in the WM for quantification of the average WM L T 1 . D, T 1 fitting curves of a WM ROI (left, red box) and a GM ROI (right, red box) shows that, with TR = 1000 ms, the TI for optimal nulling of long T 2 signals (TI null ) was 300 ms for WM L and~377 ms for GM L . SI, signal intensity   Figure 6C shows representative T 2 * fitting curves in a WM ROI (red box in the inset) and a GM ROI (yellow box in the inset). Figure 6D

| DISCUSSION
This study employed half-pulse IR-UTE for the first time to estimate the value of TI null needed to null long T 2 water signals in the brain on a 3-T clinical scanner. Long T 2 suppression is critical for the selective imaging of myelin as water protons contribute more than 90% of the detectable UTE signal (our preliminary bi-component analysis of UTE images of fresh WM showed that only about 4% of the UTE signal has a T 2 * of~0.3 ms, as presented in Figure 5). As suggested by the non-zero signal in the spinal cord WM on the IR-UTE image with TE/TI = 1.2/500 ms in Wilhelm et al, 2 an empirically selected TI is unlikely to be adequate for selective myelin imaging. By using the same sequence for TI null estimation and T 2 * measurement, we were able to minimize potential T 1 measurement inaccuracies which could have led to insufficient nulling of long T 2 signals. This is also the most efficient method to obtain the best estimate of WM TI null for the measurement of T 2 * of the STCs using IR-UTE sequences. It is simple and straightforward without requiring complicated fitting models or technically demanding corrections.
The deuterium in D 2 O is an isotope of hydrogen. Its MR frequency (which scales with the gyromagnetic ratio) is 6.5 times lower than that of protons. 25 As a result, deuterium is not detectable using conventional 1 H MRI techniques. After sequential exchange with D 2 O, the specimens in this study showed a gradual decrease in PD-FSE signals. The PD-FSE sequence used in this study had a TE > 10 ms and could only detect signals from long T 2 (several milliseconds or longer) tissue components. After a 24-h D 2 O exchange, the specimens had near-zero long T 2 signals on the PD-FSE image, and the SNR in a WM ROI decreased by about 15-fold. The UTE signal also dropped significantly, and the bi-component T 2 * analyses showed a water fraction (relative to STCs, i.e. 1f S ) of approximately 59% in WM, which suggests that approximately 90% of the WM water had been replaced by D 2 O considering the innately low f S in brain tissue. However, with TI set to TI null for WM L , SNR of the WM S signals seen on the IR-UTE images was relatively unchanged after exchange. Furthermore, T 2 * values obtained with the IR-UTE sequence remained in the range 200-300 μs, and were independent of the duration of exchange. The S T 2 * values obtained with bi-component T 2 * analysis of the UTE images were also constantly in the range 200-300 μs before and after D 2 O exchange. These T 2 * values were comparable with those measured in myelin extract powder, as well as in mixtures of the powder with D 2 O and H 2 O using the IR-UTE sequence in one previous study. 22 Despite the incomplete H 2 O/D 2 O exchange in the present ovine brain D 2 O exchange model, our results suggest a minimal contribution from exchangeable long T 2 protons to the ultrashort T 2 * signals seen on IR-UTE images acquired both before and after D 2 O exchange. These results support the view that the ultrashort T 2 * signals seen on IR-UTE images are unlikely to be generated from water or residual water in the tissue. They are more likely to be associated with non-aqueous protons.
The STC-invisible PD-FSE images always showed higher signal in GM and lower signal in WM, suggesting that there was more long T 2 water in GM than in WM both before and after D 2 O exchange (Figures 4 and 5). Like the PD-FSE images, the UTE images (TR = 1000 ms) acquired at an ultrashort TE (TE = 0.01 ms) were also mainly PD weighted. As a result of the large FA used (FA = 65°), the UTE image also showed T 1 weighting, leading to a higher signal in WM than in GM (if there was no difference in PD) because WM L had a shorter T 1 than GM L (Figures 2D, 6D). Unlike the PD-FSE image, the 'STC-sensitive' UTE image showed weaker WM/GM contrast before D 2 O exchange and largely no WM/GM contrast after D 2 O exchange ( Figure 5). This can be explained as follows. In fresh brain tissue, GM had a higher water PD than WM, and WM had a higher non-aqueous PD than GM. 1,2,4 The total number of protons, including all water protons and solid mass protons, might not be significantly different between WM and GM. 26 However, not all of the solid mass proton signals were detectable with the UTE sequence with a TE of 0.01 ms. 2 Therefore, the WM/GM contrast induced by the water content difference might not be completely canceled by the WM/GM non-aqueous proton content difference and the T 1 weighting of the UTE sequence, leading to a higher signal in GM than in WM on the UTE image ( Figure 5B). After 24-h D 2 O exchange, the signal on the PD-FSE image dropped by~90%, suggesting significant exchange of H 2 O with D 2 O ( Figure 5A, D). The PD-FSE image still showed higher signal in GM than in WM, suggesting that there was more residual water in GM than in WM after 24-h D 2 O exchange ( Figure 5D), but the absolute WM/GM water PD difference should be much smaller than that before D 2 O exchange. Meanwhile, methylene protons in the brain non-aqueous tissue are thought to be the major non-exchangeable protons in nervous tissue. 4 As a result, the difference between WM/GM non-aqueous PDs was largely reserved after D 2 O exchange. In addition, WM L still had a shorter T 1 than GM L after D 2 O exchange ( Figure 6D). These factors, altogether, could have made the contrast between WM and GM undiscernible on the UTE image after GM S also showed ultrashort submillisecond T 2 * values, which were slightly but consistently higher than those of WM S in the same specimen at all imaging time points and, like WM S , showed minimal change with increasing immersion time in D 2 O. The ultrashort T 2 * signals seen in nerve tissues and bovine brain myelin extract after D 2 O exchange are thought to predominantly arise from carbon-bound methylene protons. 4 However, methylene protons also exist in other non-myelin membranous structures of cells. 27 Because GM contains more cellular and subcellular structures than WM, GM S might be composed of a much higher fraction of other methylene-containing macromolecules than myelin. Protons in these macromolecules might have different T 2 * values from those in myelin, which warrants more sophisticated investigation with a larger sample size and the analysis of different WM and GM regions. The IR-UTE sequences may therefore have applications in the characterization of GM abnormalities, and the characterization of GM S may also potentially provide an important biomarker for MS. 28 The T 1 values of GM L and WM L varied significantly with increasing immersion time in D 2 O; therefore, a wide range of TIs (240-330 ms for WM and 320-350 ms for GM) were used to null signals from these components. The T 1 decrease with D 2 O exchange may be a result of gradual tissue water loss. In native tissue, the fast-relaxing non-aqueous protons would accelerate T 1 relaxation of aqueous protons through magnetization transfer. 29 In deuterated tissue, the ratio of exchangeable aqueous protons to non-exchangeable non-aqueous protons was greatly reduced. The magnetization transfer effects could be more prominent and lead to further reduced T 1 of the residual aqueous protons. Another possibility is that the relatively free water might be first replaced by D 2 O; the remaining signal on IR-UTE images would be dominated by bound water, such as that trapped in macromolecules, and could have a relatively short T 1 . 5,23 Prolonged D 2 O exchange for up to 7 days led to a T 1 increase (data not shown here), possibly because of a loss of macromolecular peptides as a result of tissue degeneration. 30 The T 2 * values remained relatively constant despite changes in T 1 of GM L and WM L .
There are several limitations of this study. First, in a clinical setting with slice-selective gradients switched on, eddy currents, which occur commonly with MRI scanners, can cause distortion of the combined k-space signal profile. This was not investigated in this study. If strong eddy currents exist, it might be necessary to take measures, such as dedicated prescans, to mitigate against this problem. 31 Second, different forms of tissue water may have different exchange rates with D 2 O. This was not explored in this study. Nevertheless, the relatively constant T 2 *values suggest minimal water contamination in the IR-UTE T 2 * measurements. Third, myelin T 1 and PD were not measured. It is possible to probe the fraction of exchangeable myelin protons by comparing the IR-UTE signals before and after D 2 O exchange. This needs T 1 correction as the T 1 value of myelin protons may change significantly in a deuterated environment, and this warrants further investigation. Last, although this study was performed within 32 h of tissue thawing, the specimens may still have undergone natural degradation and their myelin properties might have changed after D 2 O exchange. Histology is required in future studies for the assessment of any such effect.