PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
J Am Chem Soc. Author manuscript; available in PMC 2010 December 9.
Published in final edited form as:
PMCID: PMC2796621
NIHMSID: NIHMS155724

Hyperpolarized [2-13C]-Fructose: A Hemiketal DNP Substrate for In Vivo Metabolic Imaging

Abstract

Hyperpolarized 13C labelled molecular probes have been used to investigate metabolic pathways of interest as well as facilitate in vivo spectroscopic imaging by taking advantage of the dramatic signal enhancement provided by DNP. Due to the limited lifetime of the hyperpolarized nucleus, with signal decay dependant on T1 relaxation, carboxylate carbons have been the primary targets for development of hyperpolarized metabolic probes. The use of these carbon nuclei makes it difficult to investigate upstream glycolytic processes, which have been related to both cancer metabolism as well as other metabolic abnormalities, such as fatty liver disease and diabetes. Glucose carbons have very short T1s (< 1 sec) and therefore cannot be used as an in vivo hyperpolarized metabolic probe of glycolysis. However, the pentose analogue fructose can also enter glycolysis through its phosphorylation by hexokinase and yield complimentary information. The C2 of fructose is a hemiketal that has a relatively longer relaxation time (≈ 16 s at 37° C) and high solution state polarization (≈ 12%). Hyperpolarized [2-13C]-fructose was also injected into a transgenic model of prostate cancer (TRAMP) and demonstrated difference in uptake and metabolism in regions of tumor relative to surrounding tissue. Thus, this study demonstrates the first hyperpolarization of a carbohydrate carbon with a sufficient T1 and solution state polarization for ex vivo spectroscopy and in vivo spectroscopic imaging studies.

Keywords: DNP, hyperpolarized carbon, hemiketal, fructose, metabolism

INTRODUCTION

Recent development of techniques to retain highly polarized spins in solution via dynamic nuclear polarization (DNP) has enabled hyperpolarized 13C NMR spectroscopy and MR spectroscopic imaging studies with signal enhancements of over 10,000 fold in short acquisition times 1,2. This new technique has powerful applications for metabolic imaging of cancer. The in-vivo metabolism of [1-13C]-pyruvate and its metabolic products, [1-13C]-lactate, [1-13C]-alanine and [13C] bicarbonate, have been shown to correlate with disease progression 3 and response to therapy 4 in animal models. DNP substrates require a long T1 relaxation to facilitate efficient spin diffusion during the process of hyperpolarization 5,6. Carbonyl carbons, which lack attached protons and limit the relaxation as a result of dipolar cross relaxation, have been the standard species to label and polarize with T1’s on the order of 40–60 seconds, depending on the field strength 79.

Although a number of molecules of interest have been polarized and observed through their carbonyl carbons 1,3,911, a great number of important metabolic intermediates do not contain a carbonyl. Specifically, changes in carbohydrate metabolism occur with the evolution and progression of cancer 1214 as well as a number of other human diseases such as non-alcoholic fatty liver disease 1517. Glucose carbons have very short T1s (< 1 sec) and therefore uniformly 13C labelled glucose, a mainstay of current metabolic studies 18,19, cannot be used as an in vivo hyperpolarized metabolic probe of glycolysis.

Fructose, occurring as an isomeric mixture of five and six membered rings, has as its most stable isomer β-fructopyranose with a hemiketal in the C2 position. Fructose can enter glycolyis via hexokinase or fructokinase 2024. The one-step metabolism via hexokinase to the phosphorylated fructose-6-phosphate is analogous to the first step of glycolysis, in which glucose is phosphorylated to glucose-6-phosphate. The metabolic flux to fructose-6-phosphate in the cell is related to the downstream glycolytic metabolic events as well as activity of the pentose phosphate pathway (PPP) 2527. The pentose phosphate pathway is responsible for the predominant amount of nucleotide synthesis (which is increased at high turnover rates) and has been postulated to be a source of regeneration of NADPH in cancer cells 14 making them more resistant to oxidative stress and allowing them to replenish glutathione.

Furthermore, metabolism of fructose is implicated in non-alcoholic steatohepatitis (NASH) 15, and in the pathogenesis of specific types of cancer. Fructose can also be metabolized to the fructose-1-phosphate via fructokinase, a reaction that takes place primarily in the liver 20. Hepatic uptake is via the GLUT5 transporter, that demonstrates relative specificity for fructose. Expression of this transporter may be an important biomarker for disease in extrahepatic tissues. For example, the human fructose transporter, GLUT5 (as shown in Figure 1), is highly expressed in breast cancer cell lines but not by normal breast tissue 28. A recent study of metabolites in the prostate gland has also shown a relationship between fructose and benign versus cancer tissues 29. Thus the goal of this study was to investigate a new non-carbonyl hyperpolarized 13C probe, [2-13C]-fructose for the study of metabolism in vivo.

Figure 1
The mechanism for transport by GLUT5 and the first step of metabolism of fructose to fructofuranose-6-phosphate by hexokinase.

MATERIALS AND METHODS

Hyperpolarized [2-13C]-Fructose

A 4.0M solution of [2-13C]-fructose (Isotec, Miamisburg, OH) in water containing 15mM OX063 trityl radical (Oxford Instruments) was hyperpolarized on a Hypersense instrument (Oxford Instruments) as previously described 1. The frozen sample was dissolved in 1X phosphate buffered saline (PBS), with a resultant pH of 7.6, and transferred immediately to a 10mm NMR tube.

11.7T NMR Studies

NMR studies were performed on an 11.7T Varian INOVA spectrometer (125MHz 13C, Varian Instruments) using a 10mm 15N/31P/13C broadband direct detect probe and temperature controlled at 37°C. Initially, a thermal spectrum was acquired for a natural abundance fructose sample in 1X PBS buffer at 37°C (nt=9000, sw=30000, np=30000, TR=3.5s, acq time= 0.5s) using a 45-degree pulse. Figure 2 demonstrates the natural abundance 13C spectrum of fructose. The C2 carbon resonances are denoted by the blue brackets and correspond to the isomeric distribution of the two ring forms (pyranose and furanose forms) of the fructose molecule. For the acquisition of hyperpolarized 13C spectra eighty pulse and acquire hyperpolarized 13C NMR spectra (1 scan, spectral window = 20000 Hz, number of points= 40000, TR=3s, total acquisition time = 2 min 55s) were acquired using a 5° pulse and proton decoupled using a waltz-16 decoupling scheme. Hyperpolarized studies were followed by acquisition of thermal data with nearly identical parameters, using a 90° flip angle and a repetition time of greater than four T1’s (TR=300s, nt=16). For T1 measurements hyperpolarized solution was placed into a NMR tube pre-heated to 37° C and this temperature is maintained using the variable temperature unit of the NMR spectrometer. T1’s were determined by collecting a series of spectra with 3 sec temporal resolution, starting 12 secs after dissolution. These spectra were then fit to a mono-exponential function to determine the spin-lattice relaxation time as previously described9,30,31. Percent polarization in solution was calculated by comparing the first hyperpolarized spectrum acquired with its thermal spectrum, correcting for differences in tip angle (5 versus 90), and the number of transients (1 versus 16) obtained. Solution state polarizations were calculated by correcting the enhancement for the T1 relaxation during the transfer time (12 secs) from the polarizer to the spectrometer, and the thermal polarization at 11.7T (9.6 ppm).

Figure 2
The natural abundance spectrum of fructose (a) and DNP spectrum of [2-13C]-fructose (b). The linear form is present in the DNP spectrum, but at a very low level analogous to the thermal spectrum. (Top) Structures of each of the isomers are shown with ...

For NMR studies of the enzymatic conversion of Fructose to fructose-6-phosphate, hyperpolarized [2-13C]-fructose was reacted with 400U of hexokinase (Sigma Aldrich) in the presence of 15mM ATP, 50mM TRIS and 13mM MgCl2. The labelling and mechanism for transport and metabolism is shown below (Figure 1), though in this enzymatic study the transport element has been removed and the enzyme activity was independently measured. Peaks corresponding to fructose-6-phosphate were identified using a natural abundance carbon spectrum, using a similar set of experimental parameters.

3T Studies

T1 studies were performed using a 3T GE Signa™ scanner (GE Healthcare, Waukesha, WI) equipped with the MNS (multinuclear spectroscopy) hardware package similar to studies at 11.7T, with temperature maintained using a heating pad calibrated to 37°C. Solution spectra were acquired using a 5° non-localized pulse, TR=3s and fit to a mono-exponential. The RF coil used in these experiments was a dual-tuned 1H-13C coil with a quadrature 13C channel and linear 1H channel construction based on an earlier design and also used in 13C-pyruvate mouse imaging studies. For animal studies, T2-weighted fast spin echo images were acquired prior to MRSI studies to denote anatomy and place voxels on the region of interest. In vivo MRSI studies were carried out using a compressed sensing double spin 3D MRSI acquisition scheme as previously published 32 with a TE = 140 ms, TR=215 ms, FOV = 8 cm × 8 cm, and 16 × 8 resolution. 500 μl of of 80 mM [2-13C]-fructose (0.0013 mmols/kg) was injected similar to previously described methods for [1-13C] pyruvate in a transgenic model of prostate cancer (TRAMP) 3. High dose infusions of fructose (0.5g/kg) can lead to hyperuricemia33, but this is well above the dose given in these studies (0.24mg/kg). These injections were compared to the standard [1-13C]-pyruvate injection for the same voxel in a tumor region of interest. Maps of resonance distributions were generated from the peak heights in each voxel and overlaid on the corresponding T2-weighted image.

RESULTS AND DISCUSSION

Calculated T1’s for the C2 fructose carbon are tabulated in Table 1 for the cyclic isomers of fructose (β-fructofuranose, β-fructopyranose, α-fructofuranose) at both 11.7T and 3T, the field stenghts of the ex vivo and in vivo hyperpolarized studies. The open chain (linear) isomer of fructose is present in very small amounts (0.4%) 21,34 and not observed in the hyperpolarized NMR spectra. There was no significant difference in the C2 T1 between the cyclic isomers of fructose, most likely due to the fast chemical of the isomeric forms 34. There was a small decrease in T1 relaxation (≈ 2 sec) of the C2 carbon with decreasing magnetic field strength from 11.7T to 3T (Table 1). To date, hyperpolarized 13C agents have involved labelling at carbonyl positions, such as the C1 position of pyruvate, due to their relatively long T1’s 1,2,7,9,30,35. In contrast to the quaternary hemiketal carbon of the fructose isomers, carbonyl carbons decrease in T1 with increasing field strengths36. This difference is predominantly due to chemical shift anisotropy dominating the T1 relaxation of carbonyl carbons at higher field strengths36. This does not hold for the hemiketal of fructose, leading to the typical lengthening of T1 with increasing field strength. Percent polarizations (Table 1) show similar values for the isomers of fructose with an average solution state polarization at 37° C of 12%. These polarization values are comparable, although somewhat lower, than those reported for other compounds of interest1,37,38 such as pyruvate which has been reported to be polarized from 21–30%3,39. There was no T1 dependence on pH observed for pH ranges 5.9–7.8 for the fructose isomers.

Table 1
T1 relaxation times at 11.7T and 3T (N=3 for both, ± s.d.) are shown for each of the isomers as well as percent polarization (N=3), which have been corrected for the time from dissolution to measurement. All studies were conducted at 37°C. ...

The polarization levels were calculated relative to thermal signals from the same sample. The DNP polarizer sits in the fringe field of the 11.7T spectrometer in order to eliminate the possibility of passing the hyperpolarized sample through a zero field and loosing all polarization. For carbonyls such as the C1 of pyruvate, we know that the T1 at lower magnetic field strengths, and that the T1 in the fringe field of the magnet should be longer, on the order of 80 secs as compared to 52 secs at 11.7T. This is due to the relationship between CSA and field strength (CSA1B0 2), which dominates the carbonyl T1 at high field strengths 36. However, as demonstrated in this publication, the T1 of the quaternary C2 carbon of fructose is only slightly longer at higher field strengths (Table 1). Therefore the T1 in the fringe field should be slightly shorter than 3T, which would result in a small underestimation of the % polarization at time zero.

The reaction of hyperpolarized C2-fructose with hexokinase after addition of fructose to the hexokinase in buffer within the NMR yields the phosphorylated pentose within 5 seconds (Figure 3). An expansion of the downfield region of the spectrum (Figure 3a) shows the split in the 105.5ppm resonance, which is a combination of both the β-fructofuranose and the β-fructofuranose-6-phosphate (the predominant isomeric form of fructose-6-phosphate). Figure 3 also compares the first scan of the hyperpolarized acquisition (Figure 3b) versus the thermal spectrum acquired over 85 minutes post DNP (Figure 3c). It is apparent that the enzyme has now fully converted the fructose to fructose-6-phosphate and there is no longer a resonance corresponding to β-fructopyranose.

Figure 3
(a) Spectrum of fructose reacted with 400U of hexokinase, the zoomed in region demonstrates the resonances corresponding to the fructose and fructose-6-phosphate. (b) The dynamic spectrum after 5 secs of reaction with hexokinase. (c) The thermal spectrum ...

Figure 4 demonstrates the metabolism following separate injections of 80mM hyperpolarized fructose (Figure 4d) and pyruvate (Figure 4e) in the same TRAMP mouse. As previously published (3), the primary TRAMP tumor demonstrates high levels of hyperpolarized lactate, as well good signal to noise spectra of hyperpolarized fructose and it’s metabolite β-fructofuranose-6-phosphate. Injections of 80mM hyperpolarized fructose yielded an average total signal-to-noise ratio of 21.0 ± 2.1 in tumor slices (an average of 70 voxels in 2 slices was used for N=3 TRAMP mice). Because the isomeric ratio of the fructose pyranose to furanose in solution is approximately 77/23 21 the in vivo peak at 105.5ppm (and if visible the downstream 108.25ppm resonance) is not solely due to β-fructofuranose-6-phosphate but also has a small contribution from β-fructofuranose. In this relatively large TRAMP tumor, both high levels of LDH activity and hexokinase activity were observed 15 secs post-injection in the same 0.035cc voxels. Figure 4b demonstrates the distribution of the total fructose signal throughout the slice, a measure of combined delivery and uptake of hyperpolarized Fructose. Slightly higher levels of total hyperpolarized fructose were observed in the tumor region relative to surrounding muscle in the same slice. In comparison, Figure 4c demonstrates the distribution of the resonance at 105.5ppm, which includes the metabolite β-fructofuranose-6-phosphate. Importantly, this resonance has a good signal-to- noise (≈ 5:1) and is co-localized in regions of high lactate within the tumor (Figures 4d and 4e).

Figure 4
(a) T2-weighted image of a moderate to late stage TRAMP mouse prostate tumor. Metabolite image overlays of the resonances corresponding to total hyperpolarized fructose (b) and composite β-fructofuranose-6-phosphate and β-fructofuranose ...

Figure 5 demonstrates a case where there was tumor only in the left side of the murine prostate, providing a direct comparison of hyperpolarized fructose uptake/delivery and metabolism between benign and malignant prostate tissues. The MRSI data demonstrated that the resonance corresponding to the composite β-fructofuranose and β-fructofuranose-6-phosphate were higher in the regions of tumor as compared to the benign prostate tissues (Figure 5d). Interestingly, there is no difference in total hyperpolarized fructose in regions of malignant versus benign prostate tissue. However, the composite β-fructofuranose and β-fructofuranose-6-phosphate resonance (Figure 5c) was higher in malignant left lobe of the prostate as compared to the benign right side (Figure 5c and d).

Figure 5
(a) T2-weighted image of a TRAMP mouse with tumor only on the right side of the prostate. Metabolic images of total hyperpolarized [2-13C] fructose resonances (b) and the composite β-fructofuranose-6-phosphate and β-fructofuranose (c) ...

CONCLUSION

In this study, [2-13C]-fructose was hyperpolarized using the DNP method and shown to have sufficiently long T1’s and polarizations sufficient for hyperpolarized 13C NMR spectroscopic and MRSI studies. The hemiketal C2 of fructose demonstrates the first non-carbonyl to be hyperpolarized for use as a metabolic probe and suggests the potential of using other hyperpolarized probes involving quaternary carbons even those in ring structure. Enzymatic conversion of hyperpolarized [2-13C]-fructose, to fructose-6-phosphate has been demonstrated in vitro and in vivo. While the composite β-fructofuranose and β-fructofuranose-6-phosphate resonance was associated with tumor regions, it was also present in some surrounding benign tissues, and additional studies are necessary to fully understand the composition of this resonance and its relationship to malignancy. The enzymatic conversion of hyperpolarized fructose allows the probing of important changes in glycolytic metabolism upstream of pyruvate, including upregulated hexose uptake 28, hexokinase activity and changes in flux through the pentose phosphate pathway 14. Although, this study was focused on prostate cancer models, upstream glyolytic processes have been the basis of a number of cancer studies including the HIF-1 and PI3K related processes 14. Therefore changes in fructose metabolism may be important in the assessment of therapies that target these pathways. A potential link between fructose metabolism and non-alcoholic fatty liver disease has also been demonstrated 40 and thus hyperpolarized fructose could become a valuable metabolic imaging agent to study this and other diseases both ex vivo and in vivo. Moreover, the dose of fructose given in this murine study translates into a very safe patient dose.

Supplementary Material

1_si_001

Acknowledgments

The authors would like to acknowledge the following funding mechanisms for supporting this research: National Institutes of Health (R21 EB005363, R01 EB007588 R21 GM075941) and NIBIB T32 Training Grant 1 T32 ED001631.

Grant sponsors: National Institutes of Health (R21 EB005363, R01 EB007588); NIBIB T32 Training Grant 1 T32 ED001631

Footnotes

SUPPORTING INFOMRATION: Supporting Information Available: The complete citation of reference 29 (Complete Ref. _) is available as well as the full CSI data spectral matrix for the studies corresponding to Figure 4 (Figure S1) and 5 (Figure S2). This material is available free of charge via the Internet at http://pubs.acs.org.

References

1. Ardenkjaer-Larsen JH, Fridlund B, Gram A, Hansson G, Hansson L, Lerche MH, Servin R, Thaning M, Golman K. Proc Natl Acad Sci USA. 2003;100:10158–10163. [PubMed]
2. Golman K, Zandt RI, Lerche MH, Pehrson R, Ardenkjaer-Larsen JH. Cancer Res. 2006;66:10855–10860. [PubMed]
3. Albers MJ, Bok R, Chen AP, Cunningham CH, Zierhut ML, Zhang VY, Kohler SJ, Tropp J, Hurd RE, Yen YF, Nelson SJ, Vigneron DB, Kurhanewicz J. Cancer Res. 2008;68:8607–8615. [PMC free article] [PubMed]
4. Day SE, Kettunen MI, Gallagher FA, Hu DE, Lerche M, Wolber J, Golman K, Ardenkjaer-Larsen JH, Brindle KM. Nat Med. 2007;13:1382–1387. [PubMed]
5. Abragam A, Goldman M. Rep Prog Phys. 1978:41.
6. Borghini M. Physical Review Letters. 1968;20:419–421.
7. Chen AP, Kurhanewicz J, Bok R, Xu D, Joun D, Zhang V, Nelson SJ, Hurd RE, Vigneron DB. Magnetic resonance imaging. 2008;26:721–726. [PMC free article] [PubMed]
8. Merritt ME, Harrison C, Storey C, Jeffrey FM, Sherry AD, Malloy CR. Proc Natl Acad Sci USA. 2007;104:19773–19777. [PubMed]
9. Wilson DM, Hurd RE, Keshari KR, Van Criekinge M, Chen AP, Nelson SJ, Vigneron DB, Kurhanewicz J. Proc Natl Acad Sci USA. 2009
10. Schroeder MA, Atherton HJ, Ball DR, Cole MA, Heather LC, Griffin JL, Clarke K, Radda GK, Tyler DJ. The FASEB Journal. 2009 [PubMed]
11. Warren WS, Jenista E, Branca RT, Chen X. Science. 2009;323:1711–1714. [PMC free article] [PubMed]
12. Denko NC. Nat Rev Cancer. 2008 [PubMed]
13. Ganapathy V, Thangaraju M, Prasad PD. Pharmacol Ther. 2008
14. Kroemer G, Pouyssegur J. Cancer Cell. 2008;13:472–482. [PubMed]
15. Farrell G, Larter C. Hepatology. 2006;43:S99–S112. [PubMed]
16. Huynh M, Luiken JJ, Coumans W, Bell R. Obesity (Silver Spring) 2008;16:1755–1762. [PubMed]
17. Douard V, Ferraris RP. Am J Physiol Endocrinol Metab. 2008;295:E227–237. [PubMed]
18. Antoniewicz MR, Kelleher JK, Stephanopoulos G. Metab Eng. 2006;8:324–337. [PubMed]
19. Lane AN, Fan TW, Higashi RM, Tan J, Bousamra M, Miller DM. Experimental and Molecular Pathology. 2009;86:165–173. [PMC free article] [PubMed]
20. Bais R, James HM, Rofe AM, Conyers RA. Biochem J. 1985;230:53–60. [PubMed]
21. Skoog M, Johansson G, Olsson B, Appelqvist R. Mikrochimica Acta. 1988;III:131–142.
22. Funari V, Herrera V, Freeman D, Tolan D. Molecular Brain Research. 2005;142:115–122. [PubMed]
23. Petersen A, Kappler F, Szwergold BS, Brown TR. Biochem J. 1992;284(Pt 2):363–366. [PubMed]
24. Levi J, Cheng Z, Gheysens O, Patel M, Chan CT, Wang Y, Namavari M, Gambhir SS. Bioconjug Chem. 2007;18:628–634. [PubMed]
25. Voet D, Voet JG. Biochemistry. 3. J. Wiley & Sons; New York: 2004.
26. Tong X, Zhao F, Thompson CB. Curr Opin Genet Dev. 2009;19:32–37. [PMC free article] [PubMed]
27. Vizán P, Alcarraz-Vizán G, Díaz-Moralli S, Solovjeva O, Frederiks W, Cascante M. Int J Cancer. 2009;124:2789–2796. [PubMed]
28. Zamora-León SP, Golde DW, Concha II, Rivas CI, Delgado-López F, Baselga J, Nualart F, Vera JC. Proc Natl Acad Sci USA. 1996;93:1847–1852. [PubMed]
29. Sreekumar A, et al. Nature. 2009;457:910–914. [PMC free article] [PubMed]
30. Gallagher FA, Kettunen MI, Day SE, Lerche M, Brindle KM. Magnetic resonance in medicine: official journal of the Society of Magnetic Resonance in Medicine/Society of Magnetic Resonance in Medicine. 2008;60:253–257. [PubMed]
31. Golman K, Ardenkjaer-Larsen JH, Petersson JS, Mansson S, Leunbach I. Proc Natl Acad Sci USA. 2003;100:10435–10439. [PubMed]
32. Hu S, Lustig M, Chen AP, Crane J, Kerr A, Kelley DA, Hurd R, Kurhanewicz J, Nelson SJ, Pauly JM, Vigneron DB. J Magn Reson. 2008;192:258–264. [PMC free article] [PubMed]
33. Wang YM, van Eys J. Annu Rev Nutr. 1981;1:437–475. [PubMed]
34. Goux WJ. Journal of the American Chemical Society. 1985;107:4320–4327.
35. Gallagher FA, Kettunen MI, Day SE, Hu DE, Ardenkjaer-Larsen JH, in’t Zandt R, Jensen PR, Karlsson M, Golman K, Lerche MH, Brindle KM. Nature. 2008;453:940–U973. [PubMed]
36. Ven FJMvd. Multidimensional NMR in Liquids: Basic Principles and Experimental Methods. Wiley-VCH: New York; 1995. pp. 356–358.
37. Merritt ME, Harrison C, Storey C, Sherry AD, Malloy CR. Magnetic resonance in medicine: official journal of the Society of Magnetic Resonance in Medicine/Society of Magnetic Resonance in Medicine. 2008;60:1029–1036. [PMC free article] [PubMed]
38. Kohler SJ, Yen Y, Wolber J, Chen AP, Albers MJ, Bok R, Zhang V, Tropp J, Nelson SJ, Vigneron DB, Kurhanewicz J, Hurd RE. Magnetic resonance in medicine: official journal of the Society of Magnetic Resonance in Medicine/Society of Magnetic Resonance in Medicine. 2007;58:65–69. [PubMed]
39. Schroeder MA, Cochlin LE, Heather LC, Clarke K, Radda GK, Tyler DJ, Shulman RG. Proceedings of the National Academy of Sciences of the United States of America. 2008;105:12051–12056. [PubMed]
40. Donnelly KL, Smith CI, Schwarzenberg SJ, Jessurun J, Boldt MD, Parks EJ. J Clin Invest. 2005;115:1343–1351. [PMC free article] [PubMed]