The objective of this laboratory is to use nuclear magnetic resonance (NMR) or magnetic resonance spectroscopy (MRS) as a probe of intracellular conditions and metabolic reactions, and as an imaging tool that is sensitive to the physiological status of the tissue under investigation. Ultimately, the correlation of the metabolic information obtained from MRS and the proton NMR images should provide new insights into normal and pathological human physiology.

In vivo ³¹P MRS has focused on non-Hodgkin's lymphomas (NHL) because usually these tumors are histologically homogeneous, grow fast, and rarely become cystic or necrotic. A number of NHL tumors appear clinically as large superficial masses that can be studied using surface coil technology. These NHL characteristics plus technical advances such as dual-tuned radiofrequency (rf) probes, ¹H-decoupling, and three dimensional (3D)-localization of ³¹P signals have allowed us to acquire non-contaminated localized ³¹P spectra from superficial human NHL's with high reproducibility and signal-to-noise ratios (S/N). The main characteristics of the ³¹P spectral pattern of superficial, treatment-naive NHL are large levels of phosphomonoesters (PME), mainly phosphoethanolamine (PE), and low inorganic phosphate (Pi) and phosphocreatine (PCr) relative to the b signal of nucleotide triphosphates (NTP).

We have extended these in vivo studies to the human spleen using our ¹Hdecoupled ³¹P MR spectroscopy protocol to determine whether it was possible to differentiate between NHL signals and apparent background signals from normal spleen. Our results suggest that the main background signals from normal spleen arise from the two phosphates in 2,3 diphosphoglycerate (DPG), one within the PME region and one overlapping with Pi. Furthermore, by using ¹H decoupled ³¹P spectra, the DPG signal that is present in the PME region is separated from PE thereby decreasing the overlap between spectral patterns in NHL, benign conditions and/or normal spleen.

In vivo, 3D-localized ¹H-decoupled ³¹P spectroscopy from four treatment-naive NHL patients with clinically determined spleen involvement, one patient with benign splenomegaly and four normal volunteers was acquired using our Siemens Helicon clinical imager (Siemens Erlangen, Germany) at 1.5 T; a custom-made dual-tuned ¹H/³¹P surface coil and the 3D chemical shift imaging protocol was implemented. Post-processing consisted of Fourier-transforming the data in time and the three spatial dimensions, as well as voxel shifting to obtain at least one voxel with uncontaminated spleen signals. The voxel shifting was assessed by both image and spectral analysis. The spectra from the obtained spleen voxels were summed and the chemical shifts and areas of the peaks in the resultant spectra determined. The results were compared with previously analyzed localized spectra from superficial NHL masses.

Figure 1 shows representative in vivo ¹H-decoupled ³¹P localized spectra. The results are: 1) A common characteristic of all spleen spectra (A to C) is the presence of two strong signals of almost the same intensity around 3.5 and 2.5 ppm (arrows). The peak at 3.5 ppm arises from the overlap of phosphocholine and one of the signals of DPG, while the peak at 2.5 ppm is from the overlap of DPG and Pi. 2) There are no spectral pattern differences between normal spleen (A) and benign splenomegaly (B), except a larger S/N due to the larger volume of spleen in B. 3) Although PE is present in normal spleen and benign splenomegaly, the relative amounts of PE are smaller in these conditions than in NHL in spleen (compare A and B with C). 4) The ratio of PE versus the b signal of NTP (PE/NTP ratio) of the NHL in spleen (C) and superficial nodes (D) is larger than the one in benign spleen conditions (A and B). 5) Spectra from larger spleen NHL were identical to superficial NHL spectra (data not shown). These results demonstrate that the use of ¹H decoupled ³¹P MR spectroscopy increases the specificity of the measurement in differentiating between NHL and benign conditions in the spleen.

³¹P MAGNETIC RESONANCE SPECTROSCOPY OF HUMAN TUMORS. ARIASMENDOZA, BERARDOCCO, STOYANOVA, BROWN in collaboration with CHAPMAN,^§ SCHILDER,^§ SMITH,^§ MOVSAS,^§ MILLENSON^§

Prior work by others has shown that analysis of ³¹P MR spectra of human cancer cell lines in vitro demonstrate relatively high levels of PME; a reduction of these metabolites is associated with early indication of treatment response. Our previous work, using anatomically localized (image-guided) ³¹P MR spectroscopy in human Hodgkin's lymphomas and NHL, breast carcinomas, soft-tissue sarcomas, multiple myeloma, chronic lymphocytic leukemia, colon carcinoma and carcinomas of the head and neck, has shown similar spectral patterns in malignant tissues in vivo. High PME and low PCr levels relative to NTP, as well as alkaline intracellular environment, are constant in these spectral patterns. However, these patterns differ in their levels of phosphodiesters and Pi, which in turn depend on tumor type and/or surrounding (contaminant) tissues (i.e., muscle, brain, and spleen).

Our interest now has focused on the use of in vivo ³¹P MR spectroscopy to determine if the PME reduction as an early treatment response indicator found in human tumor models in vitro can be reproduced in vivo. Image-guided, 3D-localized ¹H-decoupled ³¹P MR spectroscopy at 1.5 T using sequences and hardware developed at Fox Chase was used to test this hypothesis. Sixteen patients (three were studied in two separate occasions for a total of 19 cases) diagnosed with new or recurrent NHL receiving treatment at Fox Chase were included in the protocol. Each patient had one MR study performed no more than 20 days prior to treatment and at least one more study no more than 20 days post-treatment; nine patients were studied at more than one time post-treatment for a total of 46 MR studies. Data analysis of localized spectra was done using software also developed in our laboratory, while blinded to the clinical outcome of the patients. The analysis comprised the identification of the tumor on MR images, extraction of the least contaminated tumor spectrum by voxel-shifting the 3D spectral data set, and calculation of the PME and b-NTP areas and their ratio on extracted spectra. The signal from b-NTP was used as a scaling internal standard because, as we have shown in previous in vitro studies, it is proportional to tumor cell volume. One case (2 studies) was excluded due to extremely low S/N ratio in the spectra. When cases had more than one post-treatment study an average of all the studies was used in the analysis.

Figure 2 shows the PME/NTP ratio of the post-treatment study expressed as the percentage of the pre-treatment value versus the clinical responses to treatment. Six of the 18 final cases showed no clinical response, 3 a partial response, and 9 showed a complete response to treatment. These results indicate that the value of the PME/NTP ratio after treatment is inversely proportional to treatment success; statistical significance between the groups with partial and no response and the group with complete response (p=0.02 and p=0.0004, respectively) was demonstrated.

A post-treatment PME/NTP ratio below 60% of the pre-treatment value was considered the cut-off for MR response. Eight out of 9 patients who responded clinically had post-treatment PME/NTP values below 60%, while 7 out of 7 patients with no clinical response had values above 60% (Table 1). Moreover, the 7 patients with no clinical response showed values above 100%. The overall accuracy of the test was 0.93 (sensitivity=0.89; specificity=1.00). The significance of the results was also tested with the R^*C analysis for independence using the G-test to include the group with partial clinical response (data not shown). Using the same cut-off for the MR response (60%), the significance of the R^*C test was also high (p=0.0008) demonstrating dependency of the PME/NTP values to treatment response. These results strongly suggest significant clinical value of ³¹P MR spectroscopy to predict early tumor response to treatment in NHL.

Based on our finding of relatively large levels of PME in human cancers in vivo and our preliminary results that show PME reduction as an indicator of early treatment response in human NHL, nine institutions around the world have started to develop infrastructure to reproduce these findings in NHL, breast carcinomas, soft-tissue sarcomas, and head and neck carcinomas. Preliminary analysis of in vivo tumor ³¹P spectra from these institutions showed good inter-institutional reproducibility, demonstrated by the expected large PME and low PCr levels relative to NTP in all tumors, as well as alkaline intracellular milieu (Figure 3).

However, before further in vivo analysis can be done, strict intra- and inter-institutional standardization is needed. Steps taken to minimize the problems and to increase reliable comparisons have been: 1) A dual-tuned probe designed at our institution with a flexible ¹H coil and a fixed surface ³¹P coil was supplied to all institutions. Each probe has a similar B₁ field for ³¹P, and has MRI-visible markers to assess coil position in quantification analysis. 2) A 2 ml bulb with a known amount of tri-phenylphosphite (TPP; 1.9 M solution in chloroform doped with copper-acetoacetate; T1<0.2 s) was placed inside the probe housing, concentric with the ³¹P coil to allow rapid sampling of a quantifiable ³¹P external reference. 3) The TPP concentration was calibrated against a commercially available triphenylphosphate standard (Isotec, Inc. USA; accuracy ± 0.05%) and in a second independent laboratory (Galbraith Laboratories, Inc., USA). The commercially available standard was not suitable for these studies due to its long T₁ value.

A fast and easy quality control protocol has also been implemented to monitor acquisition at each institution without compromising observation time. Quality control of signal quantification has been achieved using a 2 ml bulb with a known amount of phosphoric acid (PA; 1 mM solution in water doped with 7 mM NiCl2; T1<0.2 s) mounted in a fixed support containing a 0.2% NaCl loading solution that also supports the probe.

Images and shimming were performed while in ¹H mode; determination of the 90 degree pulse and spectral collection of TPP and PA were performed while in ³¹P mode (TR=1 s; pulse length=250 ms; 512 points). When 9 PA samples were tested in our institution using this protocol, the root mean square (RMS) error was 2.8%. One sample was then distributed to each participating institution. The multi-institutional RMS value recorded from the 5 institutions responding thus far is 3.6%. Special emphasis has also been placed on the correct performance of the signal localization scheme (3D chemical shift imaging), B₁-field insensitive excitation pulses (adiabatic pulses), and ¹H decoupling of ³¹P signals. Quality control of these techniques has been stressed due to the inherent in vivo spectral improvement when using them. The careful and systematic performance of all the established quality control tests will ensure the best spectral localization and resolution with comparable results between different institutions, thereby decreasing the possible problems generated by a multi-institutional study and increasing the accuracy of the data analysis.

A fundamental problem in sharing data among research groups working on MR data has been the lack of a standard data format and a readily available multipurpose application. Even within our own research group, the creation of software has been a historical process, resulting in a number of programs that must be run sequentially using outmoded command line-driven programs. To address these problems, a new analysis program designed to run on multiple platforms, to allow easy sharing of data among research groups, and to permit easy addition of user defined variables and operations, is being created.

The new application, CSImage, is written in Java using Swing components, which will permit it to run with the same interface on any platform supporting Java 1.1 or later. The graphical user interface is easy to learn and uses standard menu driven commands and dialog boxes. To make the software useful to the community as a whole, we have designed this application to allow easy addition of any variables and operations, to read any externally formatted file, and to permit easy sharing of data files. We also have used separate user accessible classes to define all onscreen text, allowing easy conversion to languages other than English.

Standard analysis methods for processing inversion recovery MR images traditionally have used single pixel techniques. In these techniques, each pixel is independently fit to an exponential recovery, and spatial correlations in the data set are ignored. By analyzing the image as a complete dataset, improved error analysis and automatic segmentation can be achieved. We apply principal component analysis (PCA) and Bayesian Decomposition (BD) to decompose the 3D data set into three separate images and corresponding recovery times.

The relaxographic ¹H₂O image data (64x64 pixels) was acquired from an axial slice (Field of View [FOV] 47x47 mm) at 4T at Brookhaven National laboratory. The T₁ recovery was sampled at 64 times post-inversion, non-linearly spaced from 30 ms to 17.1 s. The image data was zero-filled twice, thus the entire dataset size was 256x256x64 data-points. PCA was applied to the entire image data and four significant PCs were determined. Presenting their scores as images, it was apparent that the 4th PC was related to artifacts in the posterior region of the skull. To avoid this artifact, we focused our investigation on the central FOV: 128x128 pixels located entirely within the brain; only three significant PCs were identified from this data. Since BD is computationally intensive, we analyzed the 32x32 pixels (prior to the zero-filling), which corresponds to this FOV. BD was programmed to search for three exponential time solutions and their corresponding spatial images. Figure 4 presents the resultant images. The spatial distributions shown are the ones expected for gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF).

The T₁ values determined by this approach agree closely with those determined by other methods, but it is important to note that BD is unbiased (there are no expected T₁ values or favored pixels) and that BD finds a global solution. Furthermore, the T₁ values of the recovery curves are simultaneously determined with the distribution of the recovery curves, so that the segmentation into GM, WM, and CSF shown in Figure 4 is automatic and does not require that T₁ values for the tissues be predetermined.

The "glycation hypothesis" is a leading candidate to explain many pathological conditions of diabetes and aging. Increased glycation in the kidney, retina and nerves in diabetes has been observed. The structural changes produced by crosslinking proteins may well lead to the functional changes of aging and of those detected up to 15 years or more after diagnosis with diabetes. An example of structural changes preceding the functional changes is the early changes in the glomeruli and tubules after 8 to 12 weeks of diabetes in which excess glycated proteins are found in diabetic tissue. This is thought to lead to basement membrane thickening and microalbuminuria, which also reflects early changes in the glomerular membranes. Microalbuminuria has been found in several studies to be a significant risk factor to further kidney pathology, namely, degradation and sclerosis of the glomeruli, changes in glomerular filtration rate and decreasing ability to filter plasma properly.

The compounds 3-deoxyglosone (3DG) and 3-deoxyfructose (3DF) are two early intermediates that appear likely to be important in understanding the further reactions of non-enzymatic glycation. This is so because 3DG is one of the most potent early glycating agents formed and 3DF, its reductive detoxification product, is a measure of how well 3DG is removed in an individual. A small number of studies have been conducted on the levels of 3DG and 3DF in plasma and urine of diabetic populations that showed approximately twice as much 3DG and 3DF in the diabetic individuals compared to the non-diabetic individuals. In addition to the differences in the levels of both 3DG and 3DF between the non-diabetic and diabetic populations, we have shown that the ratio of these compounds was much more variable in the diabetic population. This data suggests that certain diabetic individuals were not able to detoxify 3DG as well as others.

We have recently studied a group of patients with initial signs of microalbuminuria who were followed prospectively for six years. The objective was to identify early predictors of the progression of microalbuminuria. Progression is defined as movement from low microalbuminuria (35 to 75 mg/day) in the two-year baseline period to high microalbuminuria (75 to 300 mg/day) or proteinuria (>300 mg/day). We have measured levels of 3DF in 31 "progressors" and 31 "nonprogressors"; the samples were analyzed without knowledge of patient status. The results were normalized to urine creatinine (Cr) to correct for urine volume variations. The 3DF/Cr in the "nonprogressors" and "progressors" was 180± 90 and 240± 130 nmols/mg creatinine (p<0.04), respectively. Aside from HbA1c (glycated hemoglobin) and a weak correlation with blood pressure this is the only significant association with a metabolic variable that has been observed for these patients.

FRANKS, S.E., NEGENDANK, W.G., SMITH, M.R., SHALLER, C., PADAVIC-SHALLER, K., KAPPLER, F., ZHANG, Y., BROWN, T.R., Phosphomonoester concentrations differ between chronic lymphocytic leukemia cells and normal human lymphocytes. Am. J. Hematol. (in press).

LAL, S., SZWERGOLD, B.S., WALKER, M., RANDALL, W., KAPPLER, F., BEISSWENGER, P.J., BROWN, T.R., Production and Metabolism of 3-deoxyglucosone in Humans. The Maillard Reaction in Foods and Medicine (in press).

MURPHY-BOESH, J., JIANG, H., STOYANOVA, R., BROWN, T.R., Quantification of phosphorus metabolites from chemical shift imaging spectra with corrections for point spread effects and B1 inhomogeneity. Magn. Reson. Med. 39:429-438, 1998.

OCHS, M.F., STOYANOVA, R., ARIAS-MENDOZA, F., BROWN, T.R. A new method for spectral decomposition using a bilinear Bayesian approach. J. Magn. Reson. (in press).

^§   Fox Chase researcher

^a   W.C.Randall: 1735 Supplee Road, Lansdale, PA 19446

^b   The Cooperative Group on MRS Application to Cancer: H.C. Charles -- Duke University Medical Center, Durham, NC 27706; J.A. Koutcher -- Memorial Sloan-Kettering Cancer Center, New York, NY 10021; M.O. Leach -- The Royal Marsden Hospital, London, England; J.R. Grifftiths -- St. George's Hospital Medical School, London, England; S.J. Nelson -- University of California at San Francisco, San Francisco, CA 94143; A. Heerschap -- University Hospital Nijmegen, Nijmegen, The Netherlands; J.D. Glickson -- University of Pennsylvania., Philadelphia, PA 19104; J.L. Evelhoch -- Harper Hospital of the Wayne State University, Detroit, MI 48202

^c   W.D. Rooney, X. Li, J-H. Lee, C.S. Springer: Brookhaven National Laboratory/SUNY Stony Brook, Stony Brook, NY 11790

^d   L.Scott, J. Warram, A. Krolewski: Joslin Diabetes Center, Boston, MA 02215

Illustrations or unpublished data in these reports should not be used without permission of the author.

Fox Chase Cancer Center

Scientific Report 1998