In Vivo Measurement of Phosphorous Markers of Disease

Phosphorus Magnetic Resonance Spectroscopy (31P-MRS) has been utilized to study energy, carbohydrate, and phospholipid metabolism in vitro and in vivo in live tissues non-invasively. Despite its lack of sensitivity, its application has extended to in situ human tissues and organs since proper signal localization was devised. Follow-up of phosphocreatine in neuromuscular diseases and schizophrenia and follow-up of phospholipid-related molecules in tumors are described here to demonstrate the value of 31P-MRS as an imaging technique to determine in vivo markers of disease and in the diagnosis, prognosis, and follow-up of human diseases.


Historical background
In the past three decades,nuclear magnetic resonance (NMR), an analytical technique initially utilized to elucidate the structure and composition of molecules, has become a methodology that allows the study of live samples. The basic principle of NMR is that nuclei with a nuclear spin in a magnetic field show resonances at various frequencies proportional to the magnetic field. The strengths of these signals are proportional to the amount of their originating nuclei, and their frequencies depend on the molecule in which they are locat-ed. Many of these signals can be observed from living systems because the microenvironment inside cells and tissues resembles free solutions on a microscopic scale. Initial observations of living systems consisted of cells and isolated organs in high field, small bore magnets [1,2]. With the development of magnetic resonance imaging (MRI) technology, large magnets at intermediate magnetic fields have become available, and these studies have been extended to humans. Parallelling the developments in magnetic technology were similar improvements in spectral localization techniques. Localization began with MRI, which primarily uses the strong 1 H signal of water in biological tissues and occasionally the 1 H signal from the methylenes of fatty acids. This development led to localized in vivo magnetic resonance spectroscopy (MRS) a few years latter with the development of single-and double-volume localization techniques. The early work was well re-viewed by Bottomley and colleagues [2]. More complex multiple-volume methodology for spectroscopy, chemical shift imaging (CSI), capable of isolating spectra in three dimensions, was developed by Brown and colleagues in 1983 [3]. Recently, even higher magnetic fields (4-7 T) have become available, which suggests substantial improvement in spectral and image resolution in vivo in the future.
The most extensively used nuclei in the study of live samples by MRS have been phosphorus-31 ( 31 P), carbon-13 ( 13 C), and hydrogen-1 ( 1 H). In every case, the nuclear spins that can be observed are those in relatively small molecules (<5-10 kD) that can tumble in solution sufficiently quickly to produce a sharp resonance. The natural abundance of 31 P is 100% of the available pool of phosphorus molecules, but its spectroscopic sensitivity is lower than 1 H, which allows visualization of phosphorus-containing molecules with a concentration in the millimolar range. However, the molecules in living matter that can be visualized by 31 P MRS are very important because they are linked to bioenergetics, the metabolism of carbohydrates, and the turnover of phospholipids and membranes. Note that, though phospholipids, membranes, polynucleotide molecules (i.e., DNA, RNA), and even bone constituents have a large amount of phosphorus, they are not observed as sharp lines becaus they are not able to tumble rapidly enough. On the other hand, 13 C has a natural abundance of only 1.1% and its sensitivity is slightly lower than 31 P. The natural abundance signals of 13 C in living matter show only storage molecules with repetitive units (i.e., glycogen and fatty acids). However, experiments with 13 C-enriched molecules have demonstrated intermediates and end products of metabolic pathways from which many inferences about metabolic regulation of such pathways can be drawn. Finally, 1 H is the most sensitive of the resonant nuclei. Its natural abundance is almost 100%, and almost all biological molecules have hydrogen. The high sensitivity and abundance issues could allow visualizing many biologically important molecules but a smaller range of resonant frequencies makes it harder to distinguish signals from different compounds. The very large signals from hydrogen in water and lipids in living samples that obscure visualization of other 1 H signals from metabolites of much lower concentration are another drawback. Nevertheless, the fact that MRI is based on 1 H signals from water and lipids has made the clinical imagers accessible to collect 1 H-MRS. This is why proton spectroscopy has been more available for in vivo applications in humans. Although the nuclei listed above (i.e., 31 P, 13 C, 1 H) have been the dominant ones applied to studies in vivo, other nuclei also have demonstrated potential to be used in vivo (i.e., nitrogen-15, fluorine-19, and sodium-23) [4]. In this review, we focus on the applications and advancements of 31 P MR spectroscopy in the study of biological samples in vivo, with special regard for the possible role of this technique in identifying markers of disease in humans.

Spectral information contained in phosphorus MRS of live samples
Phosphorus MRS ( 31 P-MRS) initially was believed to have the best potential for use in biological research due to the nature of the biological information obtained. An example of this is shown in the spectrum at the bottom of Fig. 1, which is an 1 H-decoupled 31 P MR spectrum of a perchloric acid extract of human lymphocytes from a normal volunteer acquired at 9.4 T (see figure caption for acquisition parameters) [5]. Four expanded regions of the same spectral area are shown in the lower portions of insets A-D. This well-resolved spectrum is due to the higher magnetic field strength and the highly homogeneous aqueous solution of the extract. In the lower spectrum of the main panel of Fig. 1, the signals of the phospholipid-related phosphomonoesters, phosphoethanolamine (Etn-P) and phosphocholine (Cho-P), are shown in extracted normal lymphocytes. The larger signal downfield from Etn-P (∼3.0 ppm) in the spectrum corresponds to inorganic phosphate (Pi). The Pi concentration in this particular spectrum is biologically meaningless due to contamination by the phosphate-based buffers utilized in the extraction procedures [5]. The smaller signals between Cho-P and Pi present in the spectrum have not been identified. Inset A shows the region where phosphodiesters (mainly glycerophosphoethanolamine [Gro-P-Etn] and glycerophosphocholine [Gro-P-Cho]) resonate. Inset B shows the resonant area of the terminal phosphates of nucleotide polyphosphates. In living tissues, these signals mainly are the resonances of the γ phosphate of nucleotide triphosphates (γ-NTP) and the ß phosphate of nucleotide diphosphates (ß-NDP). These signals are split in two peaks (doublets) due to their interaction with the neighboring phosphorus in the anhydride chain. Inset C in Fig. 1 shows the doublet signals of the initial phosphates (a) of the anhydride chain in NDP and NTP and the phosphates of several diphosphoadsesters (i.e., UDP-sugars, nicotine adenine dinucleotide) not fully identified (region X).
Finally, inset D shows the triplet signal of the middle phosphate of the anhydride chain in NTP (ß-NTP). In contrast, spectrum A of Fig. 2 (from [6]) shows the 31 P MR signals from a whole human head in vivo at 1.5 T (see figure caption for acquisition parameters). As expected, the more inhomogeneous conditions of study and the lower magnetic field strength resulted in lower spectral resolution. Although in different ratios, the signals in Fig. 1 also are present in Fig. 2. Conversely, spectrum A in Fig. 2 shows signals that are not present in Fig. 1. Phosphocreatine (PCr) is selectively present, depending on the particular tissues involved (i.e., brain, muscle, and heart), but is not present in lymphocytes. The broad phospholipid (PL) signal under the sharp signals from spectrum A in Fig. 2, although present in variable concentrations in all tissues, is lost in an acid extract such as the one used to obtain the spectra shown in Fig. 1. Furthermore, spectrum B in Fig. 2, acquired from the same human head, does not show the PL signal because an off-center Gaussian pulse as described elsewhere [6] was used to saturate this broad signal, which in turn has improved the baseline of the sharp signals.
The signals from the normal tissue spectra shown in Figs 1 and 2 exemplify the information that 31 P MR spectroscopy can provide: Cho-P, Etn-P, Gro-P-Cho, Gro-P-Etn, and PL are metabolites in the phospholipid and membrane turnover; UDP-sugars are intermediates in the carbohydrate metabolism; and NTP, NDP, Pi, and PCr participate in bioenergetics. Furthermore, due to the interaction of Pi with free protons (H + ) and NTP with H+ and magnesium (Mg +2 ) and the nearequilibrium of these interactions in biological tissues, the Pi and NTP signals modify their resonant frequencies according to the concentrations of H + (pH) and Mg 2+ in vivo [7,8]. However, these interactions are lost when a tissue is extracted to obtain an aqueous solution.

Three-dimensional spectroscopic localization
Since its original publication in 1982 [3], chemical shift imaging (CSI) has been the technique of choice to carry out multivolume MRS localization irrespective of the nucleus studied. Since then, more than 250 MRS works have been published that based their spatial localization on CSI techniques. CSI is a phase-encoding technique that avoids the spatial misregistration of spectral signals that affect localization techniques based on frequency selective radiofrequency (RF) pulses in a constant gradient [9]. Aside from the improved signal localization, the CSI experiment can be designed to obtain the spatial distribution of spectra in localized volumes (voxels) in one, two, or three dimensions [3,9,10]. When 31 P MRS is acquired by means of three-dimensional (3D) CSI, the spectroscopic data can be prescribed and aligned in the three dimensions with anatomic features present in MR images by means of the voxel-shifting capability of the Fourier transform [10]. This means that exact preselection of the volume of interest is not necessary, which in turn simplifies data acquisition and reduces machine usage. An example of localized 31 P MRS of the human brain using 3D-CSI is shown in Fig. 3 (from [11]). After collecting reference images and adjusting the magnetic field shims, the 3D-CSI 31 P dataset was acquired and processed as described in the caption of Fig. 3. On the left of Fig. 3, a transverse MR image of the head has been overlaid with the corresponding localization grid of the CSI. In addition, spectra localized entirely in brain volumes (voxels) were subsampled and overlaid on the image. On the right of Fig. 3, the rightmost spectrum of the grid has been magnified. The quality of the localized spectra is indicated by the flat baseline, separation of the signals in the phosphomonoesters (Etn-P and Cho-P) and phosphodiesters (Gro-P-Etn and Gro-P-Cho) regions, and a signal-to-noise ratio (SNR) larger than 2 for the Pi signal.

Double-tuned coils
The low sensitivity of 31 P-MRS makes it important to obtain the highest SNR for its in vivo applications. This can be achieved, at least in part, by obtaining a highly homogeneous static magnetic field (B 0 -field), which usually is adjusted using the 1 H water signal as a reference. Further, we have proved previously that some heteronuclear schemes enhance the in vivo sensitivity of 31 P signals by 1 H excitation (i.e., 1 H decoupling, nuclear Overhauser enhancement [NOE], and polarization transfer) [12,13]. We also have proved that, to obtain the most benefit from these heteronuclear schemes, the adjustment of the B 0 -field needs to be thorough. We have developed a fast and reliable automated method for this based on 1 H CSI localization [14]. As stated before and shown in Figs 3 and 4, it also is necessary to collect 1 H MR images to anatomically place the in vivo localized 31  Examples of 31 P spectra from normal human lymphocytes (bottom spectrum; corrected pellet weight = 0.16 g) and chronic lymphocytic leukemia (CLL) lymphocytes (top spectrum; corrected pellet weight = 0.65 g) acquired and processed as described in [5]. 1 H-decoupled 31 P MR spectra were obtained at 25 • C and 162 MHz (9.4 T). A 16 K free-induction decay (FID) was acquired in 1.02 s using a 45 • pulse (8.6 µs), broadband composite-pulse 1 H-decoupling, and a total repetition time of 2.7 s. One to three 225-minute acquisitions with 5,000 transients were accumulated for each sample as needed to achieve an adequate signal-to-noise ratio. Chemical shifts (δ) were referenced to glycerophosphocholine (Gro-P-Cho; 0.494 ppm at pH 8.0). Considering that the concentration of nucleotide triphosphates (NTP) is constant in CLL and normal lymphocytes, the triplet of the β phosphate of NTP was used to scale the spectra (inset D) to ease comparisons. Due to the lower pellet weight and the scaling of the spectra, the noise in the normal lymphocyte spectrum appears larger. In the main panel, the signals of phosphoethanolamine (Etn-P) and phosphocholine (Cho-P) are labeled. The insets show expanded regions of the spectra. Inset A shows the region where the phosphodiester signals resonate (1.2-0.3 ppm). Glycerophospho-ethanolamine (Gro-P-Etn) and glycerophosphocholine (Gro-P-Cho) are labeled in the inset. Inset B shows the spectral region where the doublets of the terminal phosphates of the NTP (γ-NTP) and nucleotide diphosphates (NDP) moieties (β-NDP) resonate (−5.3 and −6.0 ppm), while inset C shows the region where the doublets of the α-phosphates from the NTP and NDP moieties resonate (−9.8 and −12.8 ppm). Several derivatives of diphosphonucleosides and diphospho-dinucleotides also resonate in the area shown in inset C (X and Y regions). Inset D shows the spectral region where the triplets of the β-phosphate of the NTP moieties resonate (−20.8 and −21.4 ppm).
itate all of these 1 H requirements during a 31 P study, our team designed several doublee-tuned RF antennas with highly sensitive 31 P and 1 H channels. Moreover, with these double-tuned antennas, 31 P and 1 H spectroscopy can be carried simultaneously, thereby minimizing adjustment and image requirements [15][16][17]. A comparison of the double-tuned probes with conven-tional, single-tuned ones showed that, despite double tuning, there is no significant loss in 31 P sensitivity when the 1 H channel provides the needed performance. For studies at 1.5 T, volume probes using a four-ring birdcage design were developed with the resonators operating in quadrature mode to provide improved sensitivity, excellent B1 homogeneity, and reduced pow- P MR spectroscopy from a normal human head. Proton-decoupled, NOE-enhanced 31 P spectra from a normal human head were collected at 1.5 T (TR = 1,000 ms, spectral width = ± 1,000 Hz, 256 acquisitions, and a 45 • rectangular pulse). The chemical shift (δ) is expressed in parts-per-million (ppm), using phosphoric acid as the reference at 0 ppm (internal reference P α of NTP at −10.01 ppm). Assignments: PCr, phosphocreatine, the rest as in Fig. 1. In spectrum A, the broad signal under the sharp resonance corresponds mainly to phospholipids. This broad signal was cancelled out during acquisition in spectrum B by applying an off-center Gaussian pulse (−300 Hz) that saturated this signal as described elsewhere [6]. er deposition at both frequencies [12]. Double-tuned surface coils with a flexible 1 H element also have been developed and used successfully by our group [18,19] and cooperatively by several institutions [20] to study superficial cancers in the trunk and extremities and triggered the development of a double-tuned surface probe specifically for the neck [21].

Proton decoupling and nuclear Overhausser effect
Simultaneous application of 1 H decoupling and NOE of in vivo 31 P signals significantly improves SNR and enhances spectral resolution [12]. We have acquired 1 H-decoupled, NOE-enhanced 31 P spectra localized to defined regions of different human tissues in vivo [12,15,18,22]. The usual scheme to obtain 1 H-decoupled and fully NOE-enhanced 31 P spectra from human tissues in vivo includes the use of Waltz-4 modulation for proton decoupling and continuous wave bi-level excitation for NOE [12]. An example of the application of 1 H-decoupling and NOE to 31 P signals is shown in Fig. 4. Localized spectra of a non-Hodgkin's lymphoma tumor in vivo are shown before (lower spectrum) and after (upper spectrum) 1 H-decoupling and NOE were applied. Higher SNR and better resolved peaks in the phosphomonoester (Etn-P and Cho-P) and NTP regions were obtained in the 1 H-decoupled, NOE- enhanced spectrum without processing for resolution ehancement. This demonstrates that 1 H-decoupling and NOE of 31 P signals permits obtaining more information about the in vivo metabolism of human tissues than was possible previously and should enhance the utility of this technique for studying human disorders.
In the following sections, specific examples of the study of markers of disease using 31 P MR spectroscopy in MC, schizophrenia, non-Hodgkin's lymphoma, and chronic lymphocytic leukemia are described.

31 P MRS studies in mitochondrial cytopathies
Due to their initial non-specific symptoms, differential diagnosis of MC must be done against the rest of the ailments grouped as neuromuscular diseases (ND). The ND group includes a variety of pathologies of acquired and genetic origin, common among children, with complicated identification and characterization due to overlapping clinical manifestations that delay instituting specific treatment. In the case of MC, the usual underlying alteration is a genetic mutation that produces malfunctions of the oxidative phosphorylation, with concomitant reductions in the cellular energy state. This energy deficit is demonstrated by decreased synthesis capacity for adenosine triphosphate (ATP, the most abundant intracellular NTP) by the respiratory chain, which in turn affects all tissues; especially those that are highly active (i.e., muscle, heart, nervous system). Although measurement of respiratory and phosphorylating activities in intact mitochondria isolated from the muscle of patients with ND is a very powerful in vitro method of identifying MC, the heterogeneity of these diseases can prevent their diagnosis using this methodology. It has been recognized that the non-invasive nature of 31 P MRS and its ability to visualize NTP, NDP, Pi, and PCr and their changes during exercise make it a promising technique to aid MC diagnosis and to monitor therapy [24,27,31,32,34,36,40,42,50,94,99,196,198].
A simple protocol of in vivo 31 P MRS of active muscle can be used in the differential diagnosis of MC in pediatric patients with ND symptomatology. Under this protocol, we studied 23 ND-diagnosed patients (4-17 years-old) and 12 age-matched healthy volunteers with 100% compliance. Twelve of the ND patients already were confirmed as having MC based on isolated mitochondria studies and/or formal genetic determina-tion of their mutation. The surface dual ( 1 H/ 31 P) antenna with flexible 1 H flaps described elsewhere [18] was wrapped around the right leg of the subjects using the sensitive field of the coil as the sole means for localization. Fifteen-second 31 P spectra of the right calf muscle acquired as described in the caption of Fig. 5 were collected serially before, during, and after calf activation (a 1-second plantar flexion/extension cycle against a mild resistance carried out continuously for 3 to 7 min). Intensity changes in Pi and PCr and chemical shifted changes in Pi related to pH always were recorded in all subjects, whereas no intensity changes in NTP were evident. The lack of changes in NTP was important for two reasons. One reason was to prevent strenuous exercise that could utilize NTP that in turn could generate pain and other discomforts; the other reason was to ensure that the muscle activation was not modifying the position of the RF antenna.
Under these exercise conditions, the PCr variations correspond to the usage of high-energy phosphates during the exercise period and their re-synthesis during post-exercise. Furthermore, the PCr re-synthesis postexercise is proportional to the ATP synthesis by oxidative phosphorylation, which allows determination of the speed at which this pathway is working. Figure 5 shows an example of the spectroscopic results in one of the MC patients studied. While in this patient the PCr recovery curve took about 60 seconds from the exercise plateau to the baseline level, in normal volunteers this recovery was immediate (> 15 sec). Slower recovery curves after exercise (< 45 sec), such as the one shown in Fig. 5, were found in 9 of the 12 formally diagnosed MC patients, and fast (normal) recoveries (> 25 sec) were found in 100% of normal volunteers. Furthermore, two previously undiagnosed ND patients also showed slower PCr recovery curves, despite normal results in their isolated mitochondria study. The possibility that these two ND patients could be suffering from MC but were missed by the isolated mitochondria study demonstrates the possible use of 31 P MR spectroscopy in the differential diagnosis of MC. [11] Changes in schizophrenia have suggested that the physiopathology of the disease probably is due to a low energy state in several areas of the brain. Accordingly, coenzyme Q 10 (CoQ 10 ) has been hypothesized as improving brain energy states and cognitive functions in schizophrenia. We conducted a double-blind, placebocontrolled crossover study to examine the correlation between cognitive results and the in vivo signals in brain-localized 31 P MRS, testing the effect of CoQ 10 on impaired cognitive functions in schizophrenia patients under neuroleptic treatment. Ten schizophrenia patients were selected randomly to receive either placebo or CoQ 10 during a period of 8 weeks, with the counterpart in the following 8 weeks. At the end of each arm, both cognitive studies and localized brain 31 P MRS were performed. However, only 7 of the 10 patients had the 31 P MRS studies done in both arms. Six matched normal volunteers also were used as controls for the brain 31 P MRS study.

Brain studied by 31 P MRS in schizophrenia
The schizophrenia patients (n = 10) studied showed a slightly significant improvement in attention (p 0.05) when treated with CoQ 10 compared to placebo, but other cognitive tests failed to show differences for the group as a whole. When the patients were divided by below-normal (BN) and near-normal (NN) memory scores, a regular practice in schizophrenia research, the patients with BN scores (n = 4) showed significant improvement in verbal learning (p 0.01), graph motor speed (p 0.04), and attention (p 0.004) during CoQ 10 treatment. Patients with NN scores (n = 6) did not show significant changes in any of the cognitive studies during CoQ 10 treatment. Table 1 shows the mean brain PCr values obtained by 31 P MRS in the volunteer group (n = 6) and in the schizophrenia group (n = 7). As shown in the table, while on placebo, the schizophrenia group showed significantly lower brain PCr values than did normal volunteers (p 0.001). The mean brain PCr value for schizophrenia patients increased significantly during the CoQ 10 period (p 0.03 against placebo), but this value still was significantly below that of the normal volunteers (p 0.03). When the schizophrenia group was divided into BN (n = 3) and NN (n = 4) memo-ry score subgroups, there was a significant increase in mean brain PCr value during CoQ 10 treatment for the BN group (p 0.03) and a non-significant increase for the NN group. We also analyzed these data on an individual basis, comparing the levels of PCr in the different voxels in each patient's brain during placebo or CoQ 10 treatment. These results are shown in Fig. 6. In this analysis, two of the three patients in the BN subgroup showed statistically significant increases in PCr during CoQ 10 treatment (p 0.001 in both) with no changes in the third patient. The NN scores subgroup showed non-significant variations in PCr values.
The correlation of the cognitive studies with our 31 P MRS results suggests that the cognitive alterations found in schizophrenia may be explained at least in part by a problem in bioenergetics. In addition, the correlation between improvements in cognitive parameters and increased brain PCr levels in schizophrenia patients after CoQ 10 treatment seen in this work also supports the link between a bioenergetic impairment and the low cognitive scores found in schizophrenia, a possible therapeutic use of CoQ 10 for patients afflicted with schizophrenia, and the value of 31 P-MRS to follow up on these therapeutic changes.

Non-Hodgkin's lymphoma studies by in vivo 31 P MRS
Identification of metabolic or genetic variables from a tumor that correlate with success or failure of an individual treatment is of great clinical interest. Early knowledge of the likelihood of treatment failure would be of great use, because alternative therapies could then be considered. Currently, the commonly used indicator of treatment prognosis in non-Hodgkin's lymphomas (NHL) is the international prognostic index (IPI) [199,200]. This index is based on clinical parameters related to tumor extent and host status and not to specific genetic or metabolic information about individual patients or their tumors.
Several groups have attempted to find tumor-specific genetic markers of disease in NHL to correlate with outcome in an attempt to determine whether such information improves prognostic accuracy beyond the IPI [201][202][203][204][205][206][207]. Non-genetic markers of disease in cancer also may be important to correlate with treatment response and outcome. In vivo 31 P MR spectroscopy of human tumors has shown a characteristic 31 P spectral pattern: the tumor levels of phospholipid-related phosphomonoesters, particularly Etn-P and Cho-P, have been known to be elevated in human tumors compared to most normal tissues since the first in vivo 31 P MRS observation of a rhabdomyosarcoma in the hand [208]. Several general reviews have been published on the subject [167][168][169][170][171][172], on specific tumor histopathologies [69,173,190], and on specific treatments [191][192][193][194][195].
We have studied patients with NHL using localized 31 P MRS before treatment to demonstrate whether pretreatment levels of Etn-P plus Cho-P correlate with clinical parameters related to treatment response and outcome [19]. Three-dimensional, 1 H-decoupled, NOEenhanced CSI of 31 P spectra was acquired from each patient's major tumor area in a 1.5 T clinical imager [10] using custom-built, dual-tuned ( 31 P/ 1 H) RF antennas, as described elsewhere [18,209]. The tumorcontaining voxels in the CSI dataset were selected using the MR images as reference. Voxel-shifting [10] was used to reduce the amount of contamination in the tumor. An example of the procedure is shown in Fig. 4. Tumor voxels were extracted from the CSI datasets and summed when necessary to produce one spectrum per patient. The sum of the phosphate signals of the phospholipid-related phosphomonoesters, Etn-P and Cho-P (Etn-P+Cho-P), and the P ß signal of NTP in the resulting spectra were integrated manually to obtain the tumor [Etn-P+Cho-P]/NTP ratio. Each patient was studied in the 30 days prior to instituting a new treatment regimen. Treatments were at the discretion of the clinician. The response to treatment was determined following the World Health Organization criteria using bi-dimensional radiological measurements in serial CT scans.
We evaluated the pretreatment [Etn-P+Cho-P]/NTP levels in tumors for their ability to predict the patient's long-term (6-month) treatment response. Clinically, patients were classified into complete (CR) and notcomplete responders (NCR). The difference between CR and NCR was significant (CR: 1.45 ± 0.15, mean ± standard error, n = 10 vs. NCR: 2.28 ± 0.15, n = 17, p 0.001). A Fisher test for [Etn-P+Cho-P]/NTP also was significant (sensitivity = 70%, specificity = 71%, positive predictive value [PPV] of 58%, and overall accuracy = 70%, p 0.04). When the patients were divided into risk groups depending on their IPI, the [Etn-P+Cho-P]/NTP ratio was significant for the low-risk (CR: 1.60 ± 0.21, mean ± standard error,   . Treatment response to coenzyme Q 10 in schizophrenia. The brain PCr obtained using in vivo 31 P MR spectroscopy in each patient during placebo and Coenzyme Q 10 (CoQ 10 ) as the mean ± standard error of the brain voxels selected are shown. Patients were segregated according to those with below-normal memory scores and those with near-normal memory scores. The two asterisks denote those brain PCr values during CoQ 10 treatment that are significantly higher than those during placebo (p 0.001 in both). As shown in the figure, the two significant PCr increases during CoQ 10 treatment happened in patients with below-normal memory scores who were the patients with significant responses on the cognitive studies (see text). P]/NTP were recognized that proved to be better predictors of long-term treatment response in a Fisher test. These variable cutoffs are such that, for patients at higher risk, with larger IPI values, a lower [Etn-P+Cho-P]/NTP is needed to predict positive outcome. Using these cutoffs, the Fisher test significance of the whole group increases to p 0.0002, and its sensitivity and specificity improve to 80% and 94%, respectively.
These results suggest that we have identified an independent metabolic marker for disease. As far as we are aware, this is the first observation of an association between in vivo metabolite levels in tumors and clinical outcome, although genetic variations have been documented [207]. In any case, it is likely that, the more closely a variable can predict treatment outcome, the more likely the variable is causally connected to the underlying mechanism(s) responsible for the response to the treatment. This is true particularly when the variable is a specific intra-tumor metabolite or group of metabolites, as is the case in the present work. Thus, our results suggest that part of the metabolic pathway involved in a tumor being triggered into apoptosis is associated with the phospholipid precursors Etn-P and Cho-P. Although a detailed analysis of the causal possibilities is not feasible at present, the use of [Etn-P+Cho-P]/NTP levels in a clinical setting to guide therapy in individual patients appears to be possible once they have been confirmed in a larger patient cohort. In this work, we concluded that the pretreatment levels of Etn-P and Cho-P in NHL in vivo correlate with individual responses to treatment and time to treatment failure. This suggests that an independent factor in the variability of NHL tumor responses to therapy was identified and that it can be used as a predictor of treatment response. 31 P MRS [5] Chronic lymphocytic leukemia (CLL) is a unique disease because it offers a link between in vivo and in vitro studies. CLL cells can be isolated easily from fresh blood so that cellular conditions are determined by the patient's clinical and treatment status, and cell extracts suitable for 31 P MRS analysis at high fields can be prepared immediately after isolation. Thus, the advantages associated with 31 P MRS studies of cultured cell lines can be obtained in a human tumor through ex vivo studies of CLL cell extracts. Furthermore, CLL and NHL both are B-lymphoid malignancies. The similar cellular origin of both tumors offers the possibility of comparing the ex vivo CLL results with our ongoing studies of NHL in vivo. In this preliminary CLL study, levels of phospholipid-related metabolites (Cho-P, Etn-P, Gro-P-Cho, Gro-P-Etn) of extracts from CLL leukemia lymphocytes and normal human lymphocytes were quantified using phosphorus MRS. Figure 1 shows an example of the comparison between normal (lower spectrum) and CLL lymphocytes (upper spectrum) in this study. The CLL cells vs. normal lymphocytes showed significant increases of Etn-P (8.11 ± 2.10 mean ± standard error, µmol/g wet weight, n = 12 vs. 3.63 ± 1.10, n = 3; p 0.002), Cho-P (2.10 ± 0.37, n = 12 vs. 0.36 ± 0.09, n = 3; p 0.01), Gro-P-Cho (0.26 ± 0.03, n = 10 vs. 0.11 ± 0.05, n = 3; p 0.004), and Gro-P-Etn (0.33 ± 0.03, n = 10 vs. 0.17 ± 0.05, n = 3; p 0.003). Further, the phospholipid precursor ethanolamine was studied in blood and was found to be reduced significantly in CLL patients (4.6 ± 1.6 µM, n = 25) compared to normal volunteers (7.7 ± 2.5, n = 12; p 0.001). Increased intermediates with depletion of precursors suggest the presence of sustained phospholipid metabolism activation in CLL.

Chronic lymphocytic leukemia studies by
With respect to their 31 P MR spectra, CLL lymphocytes are characterized by elevated phosphomonoesters (Etn-P and Cho-P) and mobile phosphodiesters (Gro-P-Cho and Gro-P-Etn) in comparison to normal lymphocytes. This pattern of phospholipidrelated metabolites cannot be accounted for by elevated serum ethanolamine. Although steady-state concentrations of Etn-P, Cho-P, Gro-P-Cho, Gro-P-Etn, CDPethanolamine, and CDP-choline are determined by a complex interaction of phospholipid synthetic pathways (from ethanolamine and choline), catabolic pathways (phospholipases A2, C, and D), and reentry of intermediates into synthetic pathways, probable causes for observed patterns can be inferred. For example, an increase of Gro-P-Etn and Gro-P-Cho upon mitogenic stimulation is compatible with a concurrent activation of phospholipase A2 known to occur upon stimulation [210,211]. Similar to mitogenically stimulated normal human lymphocytes, our study shows that CLL cells have higher Gro-P-Etn, CDP-ethanolamine, and Gro-P-Cho than controls, which suggests an activation of phospholipase A2 in CLL lymphocytes. Alternatively, activation of the kinases of ethanolamine and choline could be responsible for the elevated Etn-P and Cho-P seen in CLL cells if this activation is such that it compensates for lower substrate levels (lower serum ethanolamine in CLL patients than in controls).
A high concentration of Etn-P has been observed with stimulation of phospholipases C or D in a variety of cancer cell lines [212][213][214]. Gillham and Brindle suggested a tentative connection between the elevated levels of plasma fatty acids seen in cancer patients and the increased phosphomonoester signals observed in 31 P MR spectra of tumors [214]. Dixon and Tian in a murine lymphoma infiltrating the liver support this hypothesis of phospholipase C or D activation [215,216]. In this tumor model, high Etn-P levels relative to Cho-P, and a lack of elevated Gro-P-Etn or Gro-P-Cho, were found. Etn-P synthesis was increased relative to that of Cho-P, and Etn-P synthesis via the CDPethanolamine synthetic pathway was decreased compared to the normal liver [215]. From these results, the authors suggested that their lymphoma model has high levels of Etn-P due to phospholipid breakdown via phospholipase C or D in combination with rate-limiting activity of CTP:Etn-P cytidylyltransferase [215]. Our results differ from these with regard to the elevation of Gro-P-Etn and Gro-P-Cho and a greater increase of Cho-P compared to Etn-P in CLL than in normal lymphocytes. Hence, we believe that sustained activation of phospholipase C or D is not the mechanism for our findings.
Although our data support phospholipase A2 or kinases activation in CLL, a formal mechanism by which these alterations are associated with malignancy still needs to be elucidated. However, regardless of the mechanism, there is a clear association between the malignant phenotype and elevated levels of Etn-P and Cho-P in both NHL and CLL cells. Moreover, several investigators have observed elevated levels of these compounds in other tumors as well [167,174,176,183,[215][216][217][218][219][220][221][222][223][224][225][226][227][228][229][230][231]. Since there is a reasonable possibility that this elevation is due to pathways that involve the control of fundamental cellular processes (such as apoptosis [232]), further investigation is warranted.

Summary
As illustrated above, there are many diseases in which the observation of in vivo metabolism can provide helpful clinical information, both for diagnosis and treatment. As our abilities to measure metabolites improve, these unique markers of disease, which clearly are related to the disease status in an individual patient, will become more useful in diagnosing and treating each unique patient.