Introduction

The high rate of attrition of drug projects through the pharmaceutical pipeline is a significant contributor to the increasing R&D costs seen in recent years. In 2004, the FDA released a report entitled “Innovation or Stagnation, Challenge and Opportunity on the Critical Path to New Medical Products” in which the alarm was raised that only 8% of the molecules that enter clinical development were successfully registered (http://www.fda.gov/oc/initiatives/criticalpath/whitepaper.html). Recent data suggests that this figure had fallen to 4% by 2010 (Bunnage 2011). Many more fail in the preclinical stages of development. There is an urgent need for new tools to improve drug development and the critical path document specifically highlights imaging as one of the new technologies that has a potential to contribute. One quote from the report is particularly telling, “Often, developers are forced to use the tools of the last century to evaluate this century’s advances.”

Despite the explosion of potential biomarkers due to the “-omics” approaches, there is an acknowledged need to find and establish more sensitive, specific, and predictive biomarkers (Wehling 2006). ICI (now AstraZeneca) and Sandoz (now Novartis) introduced MRI into the pharmaceutical industry in 1983, and the use of imaging biomarkers to accelerate drug discovery and development has been well documented (Chandra et al. 2005; Pien et al. 2005; Beckmann et al. 2007). MRI has been successful in the pharmaceutical industry for the same reasons that it is popular in clinical practice; it is a noninvasive imaging technique with superb soft tissue contrast capable of delivering quantitative 3D information on organ anatomy and function (Beckmann et al. 2004; Maronpot et al. 2004). Because it is noninvasive, aside from the need to anesthetize animals to immobilize them during image acquisition, animals can be imaged on multiple occasions and studies can be designed so that each animal serves as its own control increasing the statistical power of experiments and allowing group sizes to be reduced. However, despite penetration into preclinical and clinical drug efficacy studies, there are relatively few reports of the use of MRI in drug safety studies. Toxicology accounts for approximately one third of attrition in development and is thus a major cost in the pharmaceutical industry. An informal survey of a number of preclinical imaging groups in the pharmaceutical industry showed that approximately 5% of effort (range 0–20%) was devoted to safety imaging studies. This seems a disproportionally small effort considering that MRI is a powerful tool that could potentially be used to reduce attrition in the late pipeline where it is most expensive. It is important to understand why MRI has not been more widely used in the drug safety arena before describing in detail a few of the MRI assays appropriate for preclinical safety studies.

There are three types of safety pharmacology studies conducted in the pharmaceutical industry: (1) single-dose core portfolio preclinical safety studies conducted to good laboratory practice (GLP), (2) supplemental studies of compound specific effects after chronic dosing that are conducted when results from the core battery of tests raise concern, and (3) “front-loading” safety studies conducted in the drug discovery function with the aim of designing safety liabilities out of the lead compound series.

The first type of study forms part of the legally required activities toward the registration of a pharmaceutical product. The International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use (ICH) safety pharmacology guidelines recommend the use of unanesthetized animals, which is incompatible with the standard MRI experiment in which animals are anesthetized to prevent motion interfering with image quality. It is certainly feasible to habituate animals to the MRI environment; however, in practice, the results may not warrant the effort involved. In addition, it is unlikely that MRI assays will replace conventional endpoints or shorten the study duration. Thus, there is little incentive to routinely incorporate MRI assays in the core package.

The second type of study is investigational and is conducted when results from the core battery of tests raise concern (Ettlin et al. 2010). In almost all cases, the pharmaceutical industry prepares a comprehensive package of studies for the regulatory authorities that include effects after chronic dosing. The chronic-dosing regimen encourages the design of imaging experiments in which each animal acts as its own control, increasing statistical power with smaller group sizes and allowing longitudinal studies without the need to kill groups of animals at each time point; two factors that both separately and combined offer dramatic sparing of laboratory animals. In these investigational studies, there are no guidelines against anesthesia, although clearly one must consider the impact of anesthesia on each individual experiment. One of the most significant obstacles in incorporating MRI assays into investigational safety studies is that these studies are often on the critical path for drug development, and therefore there is an urgency that leaves little time to develop and evaluate sophisticated new assays. Thus, MRI is most appropriate to investigate adverse events that recur regularly in safety assessment departments so that the appropriate validation work with positive and negative controls can be in place before the technique is needed in earnest.

The third type of study is not designed with regulators in mind or conducted within dedicated safety assessment functions. It is now widely recognized that the pharmaceutical industry can no longer afford to start safety evaluation only after candidate selection, knowing that many candidates will quickly fail due to safety issues. The pressure is on to reduce attrition in the late pipeline by introducing safety screens in the early pipeline when it is still possible to design known safety liabilities out of the lead series. In future, we expect to see increased numbers of these early pipeline non-GLP safety pharmacology studies of pre-candidates conducted in the drug discovery functions for purely internal decision-making purposes. These “front-loading” studies are likely to be the most amenable to MRI as the drug project lifetime is sufficient to discover and develop the appropriate MRI assays.

Good laboratory practice is often considered a major hurdle in the use of MRI in safety assessment studies. GLP ensures that the data produced from nonclinical studies are of high quality, reliable, and valid. Since regulators use these data to authorize clinical trials and marketing of the end product, it is important that they are correctly recorded and reproducible. An experienced, multidisciplined, and dedicated function is needed to ensure that such work is in compliance with legal requirements. The current generation of preclinical MRI scanners are not equipped with GLP software tools that would guarantee consistent spectrometer operation or data transfer in compliance with GLP. In principle, there is no reason why collaboration between MRI scanner manufacturers and the pharmaceutical industry could not produce GLP compliant MRI assays; however, the burden of GLP documentation makes compliance for complex and innovative assays impractical. In practice, regulatory agencies do accept investigatory studies not to GLP if the work is critical to a scientifically based risk assessment and has been conducted to an acceptable standard. In this case, there is still a definite advantage if protocols, data acquisition, transfer, archival, staff records, and so on are in accord with GLP principles.

Despite the obstacles mentioned above, the advantages of noninvasive imaging techniques to drug safety studies are obvious. It is possible to design longitudinal studies in which the same animal is studied at baseline and then at several time points while on study. Changes in individual animals can be quantitated and compared with baseline measurements either in simple percentage terms or, for example, using more sophisticated linear mixed-effect models (Brown and Prescott 1999), leading to a reduction in group size and obviating the need to sacrifice animals at each time point. Baseline data can be used either to select or deselect animals to be included in a study or as a basis for randomization between groups. And, of course, at the end of the study, the animal is still available for other, complimentary, analysis techniques. In general, the readout time for MRI endpoints is faster than that for histology leading to faster management go/no-go decisions. Some biomarkers are only amenable via MRI, for example, quantitation of intramyocellular lipid (IMCL) is straightforward with MR spectroscopy (MRS) but time-consuming with traditional microdissection techniques.

Further, the imaging biomarkers identified in preclinical safety assessment studies can also be used in clinical drug safety studies, as MRI is widely available and safe to use in volunteer studies. This can be an advantage for the preclinical safety assessment function as it provides feedback on translation of animal safety assessment studies to humans using the same endpoint. Clearly, one would not run MRI on all clinical safety studies but in those cases where there is no cheaper, simpler safety biomarker available and there is doubt about the degree of risk posed in man, for example, because of species differences or because the effect size in the placebo group is expected to be very high.

The conclusion is perhaps best put in a recent posting on the FDA website, “Imaging technologies provide powerful insights into the distribution, binding, and other biological effects of pharmaceuticals. As part of its Critical Path initiative, FDA has joined the National Cancer Institute (NCI), the pharmaceutical industry, and academia in a number of activities that will facilitate the development of new imaging agents and the use of medical imaging during product development. We believe that with a little effort on the part of all of us, imaging agents and technologies can contribute important biomarkers and surrogate endpoints during disease progression and contribute to the development of new therapies to treat disease” (http://www.fda.gov/cder/regulatory/medImaging/default.htm).

Liver Volume Measurement

PURPOSE AND RATIONALE

Liver hypertrophy is a frequent side effect in drug development caused by a wide variety of compounds. Because it is often the first indication of the hepatocarcinogenic potential of a drug candidate, liver weight is routinely monitored in safety assessment studies (Ou et al. 2001; Shoda et al. 2000). This is necessarily a terminal procedure so that longitudinal evaluation of hypertrophy must involve serial kills of groups of animals at the time points of interest. Assuming that the compound administered does not significantly change liver density, liver volume changes should be at least as sensitive as liver weight changes. Noninvasive serial MRI measurements of liver volume can reduce animal usage by following the same groups of animals over the time points of interest. In addition, the ability to measure difference from baseline instead of a single time-point liver volume usually increases the precision of treatment measurements, and the resulting increase in statistical power can be used to reduce group sizes.

PROCEDURE

Rats are anesthetized with isoflurane (1.5–2%, 0.6–1 l/min), and then MR images are acquired. A high-resolution 3D FISP scan is acquired (TE/TR 1.7/3.3 ms, FOV 50 × 50 × 50 mm, reconstruction size 256 × 192 × 192, NA 1, FA 20°). Individual 3D images take approximately 6 min to acquire. Spectrometer triggering is set such that data acquisition occurs during the expiratory phase of the respiratory cycle.

EVALUATION

Images can be evaluated with Analyze (Biomedical Imaging Resource, MN, USA). Liver volume is determined by manual segmentation of each slice using the ROI spline tool (Fig. 17.1). There is no loss in accuracy in liver volume estimation if only a subset of at least 6 evenly spaced slices are segmented instead of the full 60 slices through the liver. To improve liver segmentation at later time points, follow-up scans can be registered to the baseline scans (Hajnal et al. 1995).

Fig. 17.1
figure 00191figure 00191

MRI coronal section through a rat liver showing good contrast from surrounding tissues (a) and correlation with ex vivo wet weight (b) (Abdel Wahad Bidar, AstraZeneca, personal communication 2007)

Full size image

MODIFICATIONS OF THE METHOD

Cockman et al. (1993) used a multislice spin-echo method and reported that respiratory triggering increased the accuracy of rat liver volume measurements. Hockings et al. (2002) and Hockings et al. (2003a) reported rat liver volumes obtained with a respiratory triggered segmented 3D fat suppressed inversion recovery snapshot readout sequence at both 7 T and 2 T and reported a correlation coefficient between in vivo MRI liver volume and post-mortem liver wet weight of 0.96 and 0.99, respectively. Tang et al. (2002) used a non-respiratory triggered multislice spin-echo method in rats and reported a correlation coefficient of 0.9 against liver wet weight with a systematic overestimation of MRI liver volume. The coefficient of variability of MRI precision was 2.3% and operator reliability for segmentation 2.9%. Garbow et al. (2004a) measured liver volume in mice with MRI at 4.7 T using an intraperitoneal injection of contrast reagent to increase contrast between liver and surrounding organs. The correlation coefficient between MRI volume and wet weight was 0.94.

CRITICAL ASSESSMENT OF THE METHOD

The correlation between MRI liver volume and liver weight has been established by a number of groups using a variety of MRI methods indicating the robustness of the technique. Its advantage over the direct measurement of liver weight is the dramatic sparing of animals as groups of animals no longer need to be sacrificed at each time point and because the ability to make within animal comparisons leads to greater precision and a reduction in group sizes. Hockings et al. (2002) reported a reduction in animal usage from 120 to 6 with the same level of precision. In order to measure liver volume with precision, it is necessary to produce good contrast between liver and surrounding tissues such as intercostal muscle, fat, spleen, stomach wall, and kidney. This can be done through judicious optimization of the MRI pulse sequence and timings. In addition, some researchers have used fat suppression pulses to null the signal from fat to enhance contrast to surrounding organs. Image acquisition normally takes several minutes so motion from breathing and peristalsis in the GI tract can produce artifacts and blurring of the images. Fast imaging, averaging, breath holding, or respiratory triggering strategies can reduce motion artifacts from respiration. The respiratory-triggering strategy synchronizes data acquisition to the respiratory cycle and is the most widely applied strategy for preclinical liver volume determination. Peristaltic motility can be reduced by overnight starvation or the application of antispasmodics such as Buscopan; however, neither approach is usually necessary.

One possible confound for this experiment is that liver weight changes by up to 15% during the day as glycogen levels drop (Latour et al. 1999), and so care must be taken in longitudinal studies that animals are always imaged at the same time of day to reduce within animal variance. In addition, care must be exercised with the choice of anesthetic as anesthetics such as halothane are hepatotoxic and may influence the outcome of the study when there are several imaging sessions.

Cardiac Hypertrophy

PURPOSE AND RATIONALE

Measurement of cardiac function and morphology is a key part of the preclinical evaluation of experimental medicinal compounds. Blood pressure, heart rate, and electrocardiogram evaluation are part of the core portfolio of safety pharmacology studies carried out in conscious telemetry dogs. If results from the core battery of tests raise concern, then supplemental studies are conducted to measure endpoints such as left ventricular pressure, pulmonary arterial pressure, heart rate variability, baroreflex, cardiac output, ventricular contractility, and vascular resistance. However, many of these endpoints involve invasive surgery and so are only appropriate for acute single time-point studies. To date, there have been relatively few preclinical studies using MRI to measure cardiovascular function, especially in the dog which is a large animal species widely used in toxicology. MRI can be used to determine myocardial volume, wall thickness, and left ventricular (LV) and right ventricular (RV) end-diastolic and end-systolic lumen volumes (EDV and ESV, respectively). These parameters can be subsequently used to derive functional indices such as wall stress, degree of eccentric hypertrophy, stroke volume (SV), cardiac output (CO), and ejection fraction (EF). MRI studies are particularly suited to chronic-dosing regimen with multiple imaging time points in the same animals.

PROCEDURE

Adult male beagle dogs (Harlan UK) weighing between 9 and 14 kg are used. On days prior to scanning, food is withheld from approximately 4 p.m. Dogs are anesthetized with a bolus intravenous dose of propofol (approx. 10 mg/kg) followed by propofol (32–42 mg/kg/h) maintenance anesthesia and ventilated with medical air via an endotracheal tube. The dorsal metatarsal or femoral artery is cannulated for blood pressure measurements and to enable sampling of arterial blood for monitoring blood gasses to ensure adequate ventilation. ECG, capnography, pulse oximetry, body temperature, and arterial blood pressure are monitored throughout the scanning sessions on a Bruker Maglife C (Wissembourg, France). Body temperature is maintained with the aid of a thermostatically controlled heating blanket.

MRI scanning is performed in a 1 meter bore 2 T Bruker Medspec (Ettlingen, Germany) using a 28-cm transmit/receive birdcage resonator. ECG triggered segmented gradient-echo cine images are acquired during the expiration phase of the respiratory cycle as measured directly from the ventilator. An average of 16 frames per heart cine traverses approximately 80% of the cardiac cycle starting from end diastole. Other relevant imaging parameters are gradient-echo flip angle 20°, TE 3 ms, TR 8 ms, 1–3 averages, SW 100 kHz, image matrix 128 × 128, in-plane field of view 200 mm, four phase-encoding steps per frame, and linear traverse of k-space. Hence, the time resolution per cine frame is 32 ms. Each individual slice cine is acquired in about one to one and a half minutes depending on heart rate, so each set of multislice cines takes about 15–20 min.

To obtain true short axis views, scout imaging commenced with a mid-ventricular coronal slice allowing the vertical long axis (VLA) to be located by aligning another scout through the apex and mid-mitral valve, thus allowing for the leftward angle of the heart. From the VLA, the downward inclination of the heart is allowed for by taking a further scout lining up the apex and mid-mitral valve to generate the horizontal long axis plane (HLA). The scouts are acquired at end diastole (0 ms delay after the QRS wave) so that the atrioventricular (AV) ring, which descends apically in systole, is in its most basal position. The first short-axis cine is then placed just forward of the AV ring on the HLA image, to cover the most basal portions of the right and left ventricles. Approximately 15 contiguous 5-mm-thick segmented gradient echo cines with no interslice gap are then sequentially acquired moving toward the apex and including the apical tip. In this way, the entire ventricle is imaged.

EVALUATION

Frames corresponding to end diastole and end systole are identified from each cine sequence and regions-of-interest (ROI) drawn around the left ventricular (LV) epi- and endocardial borders using ParaVision software (Bruker). The area of the ROIs is summed and multiplied by the interslice distance (5 mm) to calculate the end-diastolic and end-systolic volumes (EDV and ESV) of the whole ventricle and lumen. Other cardiac parameters are calculated as follows:

Stroke volume: \( {\text{SV }} = {\text{ ED}}{{\text{V}}_{\text{Lumen}}} - {\text{ES}}{{\text{V}}_{\text{Lumen}}} \)

Cardiac output: \( {\text{CO }} = {\text{ SV}} \times {\text{heart rate}} \)

Ejection fraction: \( {\text{EF }} = { }\left( {{\text{SV}}/{\text{ED}}{{\text{V}}_{\text{Lumen}}}} \right) \times {1}00 \)

Left ventricle myocardial mass at end systole is calculated as:

$$ {\text{MassLV }} = { }\left( {{\text{ES}}{{\text{V}}_{\text{Ventricle}}} - {\text{ES}}{{\text{V}}_{\text{Lumen}}}} \right) \times {\text{D}} $$

where D is the density of the myocardium (1.05 g/mL) (Hoffmann et al. 2001).

Left ventricle myocardial wall thickness in diastole is calculated from the epi- and endocardial areas at the slice where the epicardial area is maximum as follows:

$$ {\text{LV wall thickness}} = \sqrt {{\frac{{{\text{Are}}{{\text{a}}_{\text{LV}}}}}{\pi }}} - \sqrt {{\frac{{{\text{Are}}{{\text{a}}_{\text{Lumen}}}}}{\pi }}} $$

The two ROIs used are assumed to be concentric and circular.

MODIFICATIONS OF THE METHOD

Markiewicz et al. (1987) examined eight pentobarbital anesthetized dogs and reported that cardiac output and stroke volume measured by ECG-triggered MRI correlated significantly with thermodilution measurements (r = 0.73 and 0.93, respectively). Shapiro et al. (1989) also used ECG-triggered MRI in dogs subjected to myocardial infarction and found excellent correlation between MRI-derived myocardial mass and wet weight (r = 0.97) and that MRI-derived myocardial mass measured in systole and diastole correlated closely (r = 0.95). Bambach et al. (1991) examined carbon monoxide–induced ventricular hypertrophy in rats using scan averaging instead of triggering to reduce artifacts from cardiac motion. They found that the mean outside diameter of the left ventricle plus interventricular septum (LV + S) showed a strong correlation with the duration of CO (r = 0.73, p < 0.01) and to the hematocrit (r = 0.72, p < 0.05). Rudin et al. (1991) used a dual respiratory-gated and ECG-triggered approach in two models of cardiac hypertrophy in rats. The correlation coefficient between LV mass determined by MRI and post-mortem LV weight was 0.99 and LV volume, SV, and EF in control animals showed statistically significant differences from cardiac hypertrophy animals. Siri et al. (1997) applied ECG-triggered MRI to murine hearts and found LV mass determined by MRI correlated well with LV weight (r = 0.87). This data demonstrated the dependence of LV mass estimates in the mouse on the geometric model of the heart used and show that MRI provides more accurate estimates of LV mass in mice than does two-dimensional-directed M-mode echocardiography. Slawson et al. (1998) used a dual respiratory- and cardiac-gated MR sequence in mice and obtained a correlation coefficient of 0.99 between MRI and post-mortem heart weight. Hockings et al. (2003b) used the method described above to measure dobutamine- and minoxidil-induced changes in cardiac function in dogs. They showed good correlation between cardiac output measured by MRI and cardiac output measured by thermodilution (r = 0.94) and that MRI could reliably detect acute changes in cardiac output induced by dobutamine infusion (p = 0.01) in small groups of animals (n = 7). Furthermore, they showed that MRI could detect LV enlargement induced by chronic administration of minoxidil and that the increase in EDV without an accompanying change in LV wall thickness indicated a preload-induced hypertrophy. Interestingly, the MRI technique was able to detect small amounts of pericardial effusion.

CRITICAL ASSESSMENT OF THE METHOD

MRI has become the gold standard imaging technique for the study of the human heart. The main advantages are that it is noninvasive and has pronounced contrast between myocardium and blood and good temporal resolution allowing images to be acquired at any phase of the cardiac cycle. Thus, it is an accurate technique for measuring ventricular volumes independent of geometric assumptions, although clearly the precision with which myocardial geometry can be characterized depends on the number of image slices acquired through the heart and on the in-plane resolution. Image acquisition during end diastole and end systole allows the calculation of functional parameters such as stroke volume, ejection fraction, and cardiac output. One of the most important factors in the acquisition of artifact-free images is the quality of the MRI system’s ECG and respiratory triggering. Cardiac exams in the clinic are usually conducted using breathhold rather than with respiratory gating because of the difficulty of obtaining a regular breathing cycle in conscious volunteers and patients. However, in anesthetized animals, breathing irregularities are not usually a significant problem and complications due to the increase in heart rate with hypercapnia during breathhold usually outweigh the time penalty involved in waiting for the respiratory gate. The studies described above indicate that combined respiratory gating and ECG triggering improve the precision of measurements.

Alternatives to MRI include echocardiography to measure LV wall thickness, lumen volume, and cardiac output (Coatney 2001; Collins et al. 2003; de Simone et al. 1990; Zhou et al. 2004), dye-dilution techniques such as bolus thermodilution to measure cardiac output (Siren and Feuerstein 1990), and implanted pressure transducers and flow probes to measure left ventricular pressure and blood flow parameters. Like MRI, echocardiography is noninvasive and has the further advantages that it provides low cost, real-time images with structural, functional, and hemodynamic information. Functional information is usually acquired in M-mode, and hence it is necessary to make geometrical assumptions that may not be applicable if heart morphology changes. In addition, the superior inter-study reproducibility of MRI in comparison with 2D echo leads to better reliability of observed changes and thus greatly reduced patient numbers in clinical trials (Grothues et al. 2002). Both dye-dilution and implanted pressure and flow probes are invasive techniques.

When planning functional studies, it is important to consider that most anesthetics cause cardiac and respiratory depression. For chronic studies, it may only be important to ensure that the depth of anesthesia is reproducible from imaging session to imaging session; however, for acute studies, it is necessary to consider interactions between the anesthetic and the test substance. The complexity of cardiac structure and function needs to be understood to devise a well-planned imaging protocol.

Hepatic Steatosis

PURPOSE AND RATIONALE

Hepatic steatosis is a side effect associated with a number of classes of compounds including some metal compounds, cytostatic drugs, antibiotics, and estrogens. In some cases, drug-induced hepatic steatosis patients can present with a rapid evolution of severe hepatic failure, lactic acidosis, and ultimately death (Diehl 1999). The absence of predictable correlation between abnormalities in liver enzymes and histologic lesions led Clark et al. (2002) to conclude that localized magnetic resonance spectroscopy (MRS) was the best noninvasive way to quantify liver fat in patients. This approach was favored because it avoids the risks associated with invasive liver biopsy. Lee et al. (1984) demonstrated that MRI can detect fatty infiltration of the liver clinically, and Longo et al. (1993) demonstrated that MRS is a reliable noninvasive method, comparable to computerized tomography (CT), for quantifying clinical liver steatosis in humans. Recently, Szczepaniak et al. (2005) used localized MRS to show a strikingly high prevalence of hepatic steatosis in the US population, and Cuchel et al. (2007) showed that treatment with BMS-201038 was associated with hepatic fat accumulation, a potentially serious adverse event. A trend toward increased hepatic fat was also seen by Visser et al. (2010) after treatment with mipomersen.

For 20 years, localized MRS has been used in medicine and biomedical research to obtain noninvasive biochemical information from living tissue (Koretsky and Williams 1992). The spectra obtained possess the very valuable property that the intensity of a given peak is proportional to the number of nuclei contributing to that peak provided that certain experimental precautions are taken. This allows a quantitative determination of a substance if there is an appropriate internal or external reference. In the case of localized in vivo 1 H spectroscopy, the water signal is usually chosen as internal standard as the proportion of body water to ash and protein is relatively invariant. Single-voxel localized MRS allows spectra to be obtained with spatial resolutions down to 8 μL in some circumstances allowing localization of a volume of interest entirely within the liver in animals as small as mice (Fig. 17.2).

Fig. 17.2
figure 00192figure 00192

(a) Coronal view through the liver of a Café diet mouse showing the position of the 2 × 2 × 2-mm-localized MRS voxel in the right lateral lobe of the liver, (b) in vivo–localized PRESS spectrum from three mice with different degrees of hepatic steatosis, and (c) correlation between in vivo MRS and ex vivo triglyceride measurements (Abdel Wahad Bidar, AstraZeneca, personal communication 2007)

Full size image

PROCEDURE

Isoflurane anesthetized mice or rats can be scanned in a dedicated small animal MRI system with a transmit/receive radiofrequency birdcage-design resonator. MRI and MRS acquisition are synchronized with the respiratory cycle to minimize artifacts (Schwarz and Leach 2000; Wilson et al. 1993). Scout multislice spin-echo images through the liver are used to determine voxel placement. Localized 1 H PRESS spectra (Bottomley 1987) with, for example, TE/TR 6/3,000 ms and 64 averages can be obtained. For the mouse, a 2 × 2 × 2-mm cube in the right lateral lobe adjacent to the portal vein and well removed from the surface of the liver and distinct hyperintense fatty deposits is appropriate to provide sufficient signal to noise.

EVALUATION

Quantification was accomplished by simulating the water signal (which was used as a chemical shift reference) at 4.7 ppm, and the fat signals at 2.1, 1.3, and 0.9 ppm, with an 80:20 Gaussian–Lorentzian lineshape using the Bruker XWINNMR package. Without knowing the average lipid chain length and degree of unsaturation, it is impossible to calculate a valid molar fat–water ratio, so the intrahepatocellular lipid (IHCL) content is expressed as the percentage of the sum of the fitted peak areas of the three fat peaks to the fitted water peak area.

$$ \% \;{\text{IHCL}} = 100*\left( {\frac{{{\text{ A}}{{ }_{{{\text{(lipid)}}}}}{ }}}{{{{\text A}_{{{{\text{H}}_2}{\text{O}}\;}}} + {\text{A}}{{ }_{{{\text{(lipid)}}}}}}}} \right)\; $$

MODIFICATIONS OF THE METHOD

Hazle et al. (1991) used MRS to follow the time course of ethanol-induced liver steatosis in rats. Spectra were acquired without respiratory triggering, and lipid signal was normalized to signal from an external reference sample. Correlation between MRS normalized lipid signal and biochemically determined lipids was moderate (r = 0.52). Ling and Brauer (1992) used respiratory-triggered MRS to examine the same model and were able to show that a 5.5-fold increase in lipid signal on treatment was matched by ex vivo analysis although a correlation coefficient was not given. Szczepaniak et al. (1999) used two animal models to show a close correlation between hepatic triglyceride measured by in vivo MRS and liver biopsy (r = 0.93). These researchers converted the MRS fat–water signal ratio to micromoles triglyceride/gram wet tissue by correcting for NMR relaxation and triglyceride proton density relative to water. Daubioul et al. (2002) used non-triggered localized MRS to show a reduction in hepatic steatosis in Zucker rats fed a dietary supplement with non-digestible carbohydrates. The spectra presented showed artifacts consistent with respiratory motion during acquisition. Hockings et al. (2003a) measured the MRS fat–water ratio in the livers of Zucker rats. They found a good correlation between MRS fat–water ratio and the fractional volume of intrahepatic fat determined by histology (r = 0.89) and were able to show that rosiglitazone treatment reduced liver fat content. Kuhlmann et al. (2003) reported similar findings in Zucker diabetic rats treated with rosiglitazone. Liver lipid levels in mice were examined by Garbow et al. (2004b). They reported that respiratory-triggered acquisition of spectra was important to remove the deleterious effects of respiratory motion and that the variation in MRS lipid content across the liver was typically less than 10%. The correlation coefficient between in vivo MRS and ex vivo wet chemistry lipid measurements was 0.95. Zhang et al. (2004) reported the use of a respiratory-triggered 3D three-point Dixon MRI method to determine liver fat–water ratio in rats treated with a microsomal transfer protein inhibitor known to produce hepatic steatosis. They reported a high level of reproducibility in in vivo measurements and were able to detect drug-induced steatosis, but the correlation coefficient against liver triglyceride and information on spatial inhomogeneity of lipid accumulation in the liver were not given.

CRITICAL ASSESSMENT OF THE METHOD

A number of both clinical and preclinical studies have shown a robust correlation between liver fat–water signal ratio measured by in vivo–localized MRS and ex vivo analysis. Most groups have used a short echo time PRESS sequence with respiratory triggering to reduce motion artifacts and water as an internal standard. Both liver biopsy and single-voxel localized MRS are hampered by sampling errors if fatty infiltrations are inhomogeneously distributed in the liver. In the clinical setting, alternative MRI or spectroscopic imaging techniques have been used to measure lipid content across the entire liver where there is a risk of fatty infiltrations. Preclinically, Ling and Brauer (1992) have shown that fat is distributed homogeneously throughout the liver in rats with ethanol-induced hepatic steatosis, and Garbow et al. (2004b) reported similar findings for wild-type and two transgenic strains of mice on low-fat or high-fat diets. Most researchers avoided the problem of potential inhomogeneous lipid distribution by selecting one region of the liver and always returning to the same region in serial time-point studies. For preclinical studies, it is clearly possible to kill groups of animals at each time point, but particularly when the within group variability is large in comparison to the measurement precision, the introduction of a noninvasive technology can result in a dramatic sparing of animals.