Highlights

Different types of 2D MRS can offer different types of information to understand the complexities underlying pathophysiology of disease. Technical developments specific to 2D method development and advanced post-processing methods will allow for the identification of biomarkers of diseases at an early stage. Acceleration of signal acquisition, as well as automated data processing algorithms are essential to introduce 2D MRS methods into the clinic.

Target Audience

Imaging specialists and clinicians interested in spectroscopy, MRS technologists, basic scientist and translational researchers interested in learning of new methods to study in vivo metabolism.

Outcome/Objectives

The goal of this talk is to discuss emerging 2D MR spectroscopy methods and their application in the human body. Participants will be provided with a brief overview of the emerging new methods ranging from promising approaches that may be several years away from clinical use to novel development of existing technologies that focus on addressing the unmet needs of patients.

Purpose

MRI has an unprecedented power in detecting a showing morphological and structural changes inside the human body, with many new contrast mechanisms to show difference between different types of tissue environments. MRI, however, generates images based on signal arising from water or fat; the most abundant species in the body. MRS, however, shows the chemical makeup of tissue without the need to remove a biopsy. One–dimensional (1D) spectroscopy has been around for more than 3 decades with important breakthroughs. A shortcoming of 1D spectroscopy is the narrow spectral bandwidth (0-4.7) leading to overlapping signals and subsequent analytical and processing challenge. Two-dimensional (2D) was proposed to reduce the signal overlapping that 1D suffers from and thus increase spectral sensitivity. Of course, three is no free lunch, and 2D comes with its challenges of long acquisition times and complex analysis. However, many advances in the field lead to the increasing reliability of this emerging method. Therefore there is a strong need to develop new techniques that are sensitive to the underlying pathophysiological changes that occur in disease and potentially to diagnosis conditions early on to allow for treatments. The purpose of this session is to learn of different 2D methods and their application in the human body in vivo.

Methods

The first localized in vivo human brain 2D correlated spectroscopy (COSY) experiment was collected in 1994, with an acquisition time of 102 min and region of interest (ROI) of 240 cm3 [1] on a 2 T whole body magnet. In 1995, Ryner et al. implemented a J point-resolved spectroscopy ( JPRESS) in the white matter of human brain in 27 min (ROI 27 cm3), using a ‘‘90–180–t1–180–t1–ACQ’’ pulse sequence, where t1 is the incremental delay, and all RF pulses are spatially selective [2]. A longer list of 2D methods followed sooner after by the same authors [3]. Ziegler [4] and Thomas [5] implemented in vivo cosy in human brain at 3T and 1.5T, respectively. The majority of current in vivo 2D data are collected on commercially available clinical MR scanners at a field strength of 1.5 or 3 T, with some emerging at 7T [6, 7]. The COSY sequence uses a minimum of three RF pulses (Figure 1), where the first two RF pulses generate a spin-echo, and then the third RF, a 90 RF pulse, acts as a spatially selective and coherence transfer pulse. A different 2D pulse sequenced known as JPRESS, with a minimum of three spatially selective RF pulses applied in the presence of slice-selective gradients along each of the three orthogonal axes, is used to generate a double-refocused echo. RF shapes can be sinc, Shinnar-Le Roux, adiabatic or any slice-selective pulse with suitable spectral bandwidth. Even though the sequence localizes in a single shot, a phase cycle that eliminates all signal except double-refocused echo is recommended.

Processing of in vivo 2D MRS data

The raw matrix is usually linearly predicted to 1.5 its size in F1. This is followed by zero filling to double the size of the linear-predicted data. Apodization with phase shifted sine-bell squared functions along t1 and t2 is then applied before Fourier transform. The software package, known as Felix [8] is used by many workers in this field, and is equipped with suitable processing functionalities. L-COSY spectra are usually asymmetrical about the diagonal due to the implementation of water suppression, which partially suppresses resonances close to water frequency. This in turn makes the process of coherence transfer between coupled spins non-equivalent[4]. In the absence of water suppression, L-COSY spectra are symmetrical. ProFit (Prior-Knowledge Fitting) [9] is a 2D MRS fitting program with an approach that combines ‘linear combination of model spectra’ (LCModel) [10] and ‘variable projection’ (VAPRO) [11] methods. Thus, the combination of frequency domain fitting with prior knowledge constraints as well as a parameterized time domain fitting with linear and non-linear fitting algorithms is implemented in ProFit. The ProFit software produces concentration ratios to resonances in the spectrum, with Cramer-Rao lower bounds (CRLBs). In addition, the number of identifiable and quantifiable metabolites in human brain in vivo in ProFit is 17 compared to 8 in LCModel, mostly due to data spread on another spectral domain in JPRESS [12]. ProFit was first attempted [9] on a 2D JPRESS phantom spectrum containing standard brain metabolites at physiological concentrations [13]. Reproducibility was checked by acquiring the JPRESS data at different times on a 3T scanner, and CRLBs were calculated for each fit yielding acceptably low CRLBs. In addition, JPRESS and PRESS [14] data were collected from 27 healthy volunteers to evaluate ProFit in vivo and to compare to LCModel. The concentration ratios with respect to creatine, and the CRLBs were calculated using ProFit and LCModel. Results obtained by ProFit were slightly overestimated, probably due to macromolecular contamination. Intra-subject variability, obtained by acquiring and analyzing data more than one, was lower in ProFit than LCModel. Recently, in vivo data acquired with L-COSY sequence [5] was analyzed with ProFit and the resulting CRLBs compared to CRLBs obtained using JPRESS sequence [15]. It was found that the mean CRLBs are lower with L-COSY than with JPRESS.

Results/ Discussion

Human Muscle

Localized 1D MRS was applied successfully to human skeletal muscle in vivo to detect the two lipid pools referred to as intramyocellular lipids (IMCL) and extramyocellular lipids (EMCL) [16-19]. A typical L-COSY spectrum from healthy soleus of a male volunteer is shown in Figure 2. It was proposed by Schick et al that the resonances from the lipids in the muscle spectra are seen as two signals due to the geometrical arrangement and anisotropic susceptibility of these lipid compartments [16]. Velan et al. [18, 19] implemented L-COSY in 10 volunteers and was able to determine IMCL/Cr and EMCL/Cr as 10.2±1.9 and 15.4±2.9, respectively. The measure of unsaturation was achieved by measuring the ratio of the C2/C1, where C2 and C1 are volume of cross peaks due to diallylic and allylic protons, respectively. The measure of unsaturation for IMCL and EMCL were 1.1± 0.11 and 0.87±0.12, respectively. Thomas et al. [20] used a STEAM-like pulse sequence, localized exchange spectroscopy with a long mixing time to detect proton exchange between different molecular species or fragments of the same molecule. Short tm values (<100 ms) did not show exchange cross peaks. Recently, L-COSY was used to characterize lipids in soleus muscle and abdominal fat in type 2 diabetes (T2D) [21]. Analysis of the soleus muscle 2D L-COSY spectral data showed significantly elevated IMCL ratios with respect to Cr and decreased IMCL unsaturation index in patients when compared to healthy subjects (P < 0.05). In T2D, Pearson correlation analysis showed a positive correlation of IMCL/Cr with EMCL/Cr (0.679, P < 0.05). Characterization of muscle IMCL and EMCL ratios, unsaturation index, and abdominal fat, may be useful for the noninvasive assessment of the role of altered lipid metabolism in the pathophysiology of type 2 diabetics, and for assessment of the effects of future therapeutic interventions designed to alter metabolic dysfunction in T2D. In an attempt to reduce the effect of limited spatial coverage and long acquisition times, faster and more covering methods were developed. A significant reduction in the total scan duration using the multi-echo based echo-planar correlated spectroscopic imaging (ME-EPCOSI) [22, 23] sequence was accomplished using two bipolar readout trains with different phase-encoded echoes for one of two spatial dimensions within a single repetition time (TR). In agreement with an earlier report using single-voxel based 2D MRS, significantly increased unsaturated pools of IMCL and EMCL and decreased IMCL and EMCL unsaturation indices (UIs) were observed in the soleus and tibialis anterior muscle regions of subjects with type 2 diabetes compared with healthy controls. In addition, significantly decreased choline content was observed in the soleus of T2D subjects compared with healthy controls. 1D and 2D MR correlation spectroscopy at 7T from human soleus muscle were analysed leading to T1 and T2 relaxation time constants due to their importance in sequence design and spectral quantitation[6]. Additionally, the L-COSY spectra obtained from the soleus muscle provided information on lipid content and chemical structure not available, in vivo, at lower field strengths. All molecular fragments within multiple lipid compartments were chemically shifted by 0.20-0.26ppm at this field strength. 1D and 2D L-COSY spectra were assigned and proton connectivities were confirmed with the 2D method.

Prostate

Due to the location of prostate gland in the body, MRS studies are usually carried out with an endorectal coil as receiver and body coil as RF transmitter to optimize SNR [26]. Metabolites such as polyamines (spermidine and spermine), citrate, and creatine are prominent markers in healthy prostate whereas these are reduced in malignant tissue and replaced by strong choline and lipid resonances [27] and their relative concentrations reflect disease and pathology. Localized 2D JPRESS was implemented on a 1.5 T whole body scanner [28]. The prostate spectrum acquired from 2 cm3 voxel from a healthy volunteer showing citrate, polyamines and lipid resonances was compared to a corresponding spectrum from a benign prostatic hyperplasia (BPH) and prostate cancer patients. In BPH, spermine+spermidine levels remain high as does the citrate. In the 2D spectrum from a prostate cancer, there is a sharp reduction in polyamines and citrate, along with an increase of choline. In 2006, Lange et al introduced strong-coupling point-resolved spectroscopy (S-PRESS) sequence which addressed the strong coupling of citrate and allows for an accurate characterization of the citrate signal [29]. This can be achieved by keeping TE constant but varying the two partial echo times TE1 and TE2 simultaneously. Preliminary evaluations of 2D JPRESS, S-PRESS and L-COSY sequences have demonstrated unambiguous detection of citrate, creatine, choline, spermine and more metabolites in human prostates. Automated quantitation (e.g. ProFIT-based) of JPRESS and L-COSY data clearly shows the superiority of 2D MRS over conventional one-dimensional (1D) MRS. These sequences have been evaluated in a small group of prostate pathologies and pilot investigations using these sequences show promising results in prostate pathologies. Recent advances on the temporal capability of prostate metabolic scanning was achieved with echo-planar correlated spectroscopic imaging (EP-COSI) as well as J-resolved spectroscopic imaging (EP-JRESI)[30]. These methods reduce the long acquisition time required for spatial encoding by using echo-planar spectroscopic imaging (EPSI) technique combined with correlated spectroscopy to give four-dimensional (4D) and the multi-echo (ME) variants. Further acceleration can be achieved using non-uniform undersampling (NUS) and reconstruction using compressed sensing. Reviews covering application of 2D in prostate have been written by Ramadan [31] and Thomas [32]. The feasibility 2D MRS has been shown but a more systematic study on a larger cohort, along with quantification of the cross peaks is required.

Human Bone Marrow

Many 1D spectroscopic studies reported that marrow lipid metabolism might provide important diagnostic information for tumors [24]. A typical L-COSY spectrum from tibial bone marrow of a male volunteer is shown in Figure 3. Single voxel L-COSY and double-quantum filtered (DQF)-COSY were implemented on a 1.5 T whole body magnet in healthy and diseased tibial bone marrow [25]. Six healthy and six patients with acute leukemia were studied using this technique. An unsaturated lipid proton index, was obtained from the DQF-COSY by dividing the sum of all unsaturated cross peaks in L-COSY by the (CH2)n diagonal peak, but was unable to discriminate between healthy marrow and bone marrow in a leukemia patient. Increased water content was detected in leukemia patients. It remains to be seen if diagnostic information is available at the higher field strengths of 3, 4 and 7T.

Brain

In 2001, Thomas et al [5] reported L-COSY on a 1.5 T scanner with data acquired from the frontal and occipital gray/white matter regions of healthy volunteers. The data was acquired from a 27 cm3 region in 34 minutes. In the same year, Ziegler et al [4] reported a similar experiment on a 3T whole body scanner with data acquired from a similar voxel size in the left parieto-occipital lobe of a healthy volunteer in about the same time. Cross peaks identified included N-acetyl aspartate (NAA), aspartate (Asp), myo-inositol (mI), taurine (Tau), glutamine/glutamate (Glx), Gamma-aminobutyric acid (GABA) and threonine (Thr). A typical brain L-COSY is shown in Figure 4. An in vivo L-COSY spectrum acquired on a 3T magnet [33], shows (1) the improved spectral resolution at 3T compared to similar spectra acquired at 1.5T [5], (2) more reliable detection of Glutathione (GSH), (3) better separation of cross peaks close to diagonal, e.g. cerebral glucose (Glc) and the cross peaks of NAA and Glx, (4) weaker effect of water suppression on close-by peaks (e.g. NAA and GSH), and (5) better separation of threonine (Thr) and lactate (Lac). L-COSY and JPRESS were aslo compared in the above work to realise that increased spectral width along the new spectral dimension in L-COSY resulted in an improved spectral dispersion enabling the detection of several brain metabolites at low concentrations. Due to increased sampling rate, severe loss of metabolite signals due to T2 during t1 was a major drawback of 2D JPRESS in vivo. Two-dimensional J-resolved spectroscopy (JPRESS) in vivo was first implemented in 1994 by Ryner et al [2, 3] on a 1.5 T magnet. More work followed [33] to produce a better quality JPRESS in vivo spectrum at 3T where NAA, Glx, Asp, Cr, Cho and mI were identified. In another study where L-COSY was used study the neurochemistry of late-life depression [34], the mean [Cho]/[Cr] in a sample of 33 elderly subjects was found to be higher in men than women by 10% (P < 0.05). Metabolites like mI, phopshoethanolamine (PE) and Glx were also found to be higher in depressed elderly females than controls, although these differences were not statistically significant. Cerebral metabolites were also measured using L-COSY technique in HIV-infected patients[35]. The concentration of mI and GABA were found to be elevated in the left frontal brain of HIV-patients. Watanabe implemented constant-time correlation spectroscopy for simultaneous observation of glutamate, gamma-aminobutyric acid, and glutamine in human brain at 4.7 T [36]. A constant delay of 110ms was used to produce a decoupled F1 signal displaying the metabolites separately. The use of gradient modulated constant adiabaticity pulses for in vivo localization in correlated spectroscopy and total correlated spectroscopy was first introduced by Andronesi et al [37] for accurate signal localization and lower SAR. These methods were tested on patients with brain tumors on 3T clinical platforms equipped with standard hardware. Adiabatic version of COSY sequence, AL-COSY, was introduced to improve slice selection profile and reduce chemical shift artefacts [38]. It was shown that AL-COSY reduces chemical shift artefacts compared with L-COSY and that slice profiles of adiabatic pulses were found to be sharper and more symmetrical than those of traditional Mao pulses. Comparison of 2D spectra obtained revealed spectroscopic peak volume improvements in AL-COSY and less residual water. L-COSY technique was also used used to compare the biochemistry of the human brain to that of glioblastoma (GBM) [39]. Statistically significant differences were observed (P < .05) between the cross peak volumes measured for healthy subjects and those with GBM which include lipid, alanine, N-acetylaspartate, gamma-aminobutyric acid, glutamine and glutamate, glutathione, aspartate, lysine, threonine, total choline, glycerophosphorylcholine, myo-inositol, imidazole, uridine diphosphate glucose, isocitrate, lactate, and fucose. For faster data collection sequences that cover wider regions in the brain (i.e. multivoxel 2D COSY), four-dimensional (4D) multi-echo COSI (ME-COSI) was implemented in human brain [22]. Diagonal and cross-peaks were identified from several brain metabolites including N-acetyl aspartate (NAA), Ch, Cr, lactate (Lac), aspartate (Asp), glutathione (GSH), and glutamine\glutamate (Glx) with coefficients of variation (CV) in metabolite ratios across repeated measurements were <15% for diagonal and <25% for cross-peaks observed in vivo. This was followed by the implementation of correlation chemical shift imaging with low-power adiabatic pulses and constant-density spiral trajectories [40]. This technique provides minimal chemical shift displacement error, reduced lipid contamination from subcutaneous fat, uniform optimal flip angles, and efficient mixing for coupled spins, while enabling short repetition times due to low power requirements. An 8 x 8 single-slice 2D COSY dataset took 8: 32 min on 3 T clinical scanners, which makes it feasible for in vivo studies on human subjects. In vivo COSY was also successfully used to detect 2-Hydroxyglutarate in isocitrate dehydrogenase-Mutated Glioma Patients, where false positives were obtained by using 1D method [41]. Recently Verma et al implemented L-COSY sequence in a human brain on a 7T magnet [7], where they detected cross-peaks from gamma-aminobutyric acid (GABA), isoleucine (Ile), lysine (Lys) and Ethanolamine (Eth) were quantified, which are not readily resolvable with conventional one-dimensional (1D) MR spectroscopy.

Breast

An L-COSY spectrum can be acquired from ~1 cm3 of breast tissue (Figure 5). In vivo breast L-COSY was implemented on eight cancer patients and nine healthy volunteers to evaluate water to lipid (saturated and unsaturated) ratio and to try to correlate this ratio with disease [42]. Three regions were studied; fatty, mixture of fatty and glandular and glandular tissues. No choline resonances were detected in healthy tissue. A 2D L-COSY from a breast benign mass displayed a more prominent water peak and weaker cross peaks of the olefinic protons, when compared to the healthy spectra. Choline resonances where not observed in the spectra from benign breast tumors. The L-COSY spectrum from a patient with invasive ductal/adenocarcinoma showed an elevated water diagonal peak, reduced saturated and olefinic lipid resonances, a prominent choline diagonal resonance at F1=F2=3.25 ppm, and a weak choline cross peak at (4.0, 3.5). The presence of choline and the higher water/fat ratio in cancer patients than in healthy volunteers, is in agreement with 1D spectroscopy results [43]. In 2005, a statistical classification report on the diagnostic efficiency of breast L-COSY in human breast carcinoma was published [44]. Significant decrease of diagonal and cross lipid peaks volumes were observed in carcinoma patients. Choline was only observed in spectra from the carcinomas. Reviews summering application of 2D spectroscopy in breast was published recently [45, 46]. Cohorts with breast benign masses, unclassified cancer, invasive ductal cancer and a healthy group were studied with L-COSY [47]. Peak ratios to Pk2+Pk4 (the sum of cross peaks between aliphatic and olefinic fat fragments) were found to be capable of differentiating, with reasonable statistical significance, between different types of pathologies. A recent study by Ramadan et al used L-COSY technique to study healthy versus BRCA1/2 gene mutation carriers [48]. Statistically significant peak ratios were found (P>0.05, Mann-Whitney two-sided nonparametric test upon comparing data from different cohorts, even though MRI showed no pathologic signs. Localized COSY recorded significant changes in women with BRCA1 and BRCA2 gene mutations when compared with control subjects. If these changes are ultimately proven to be a premalignant stage, this method may prove useful in screening.

Conclusion

The progress of in vivo 2D MRS has been slow but steady and forwardly. Twenty two years after the first 2D in vivo COSY was collected, the clinical application of the technique is still in its infancy. There is no doubt about the utility and usefulness of the technique, however, more development is required to reduce the experimental acquisition time, expand coverage and automate processing. The higher field strength magnets, now available for in vivo studies are providing better spectral resolution and reducing scan time.