Skeletal Muscle Quantitative Nuclear Magnetic Resonance Imaging and Spectroscopy as an Outcome Measure for Clinical Trials

LIST OF ACRONYMS

THE MOTIVATION FOR USING NMR AS AN OUTCOME MEASURE IN NEURO-MUSCULAR DISORDERS

The medical approach to neuro-muscular diseases has radically changed during the last two decades or so. A majority of these diseases is of genetic origin and were so far impossible to cure. Treatments were essentially supportive and palliative. With no exception, they are rare disorders. This fact contributed to their confinement far aside from the main avenues of research of the pharmaceutical industry. Rare diseases are now receiving full attention, with major structural programs launched and funded bypublic institutions at a multi-national level. And, more important than any other consideration, the progress of gene therapy and pharmacogenetics has or is about to revolutionize the course of many of these diseases [7, 23, 32, 33, 45, 62, 76, 174, 184, 217, 248, 256].

Because of these innovative therapies, new needs have emerged over a relatively short period of time, with the necessity to monitor muscle response to interventions. To this end, novel tools had to be developed. In an ideal world, they should be non-invasive, quantitative, cost effective and generate results that are simple to interpret. They can be grouped in 3 main categories: the functional tools, the biological fluid biomarkers and the imagery. The functional tools occupy the foreground, with a variety of devices and protocols, many optimized for the investigation of specific movements, but also with others aiming at a global evaluation of the patient activity [54, 109, 158, 167– 169, 171, 191, 216, 218– 220]. These ones belong to a rapidly growing new discipline, named actimetry, and offer the exclusive possibility to evaluate the patient in his personal environment over extended periods of time. The biological fluids constitute also a new class of biomarkers, with promising results, in particular since the discovery and now use of μRNAs. Imaging, the third main player, is increasingly being used as an outcome measure. It requires expensive equipment, high-end ultrasound devices making no exception to this. In addition, tomographic methods, computed tomography (CT) and nuclear magnetic resonance (NMR), suffer from absence of portability. Nevertheless, NMR is the only technique that can assess individual muscle anatomy, composition and function in a single examination. These unique properties are more and more largely understood and appreciated at their exact value, as illustrated by the imaging requirements issued by the regulatory authorities, EMA and FDA, for new drug registration.

With this review, the authors tried to offer a comprehensive overview of the data currently published on the use of muscle NMR as an outcome measure, in actuality or as a potential tool for future studies. Instead of listing studies in an objective but impersonal mode, they chose deliberately to insert comments and express opinions with the main goal of guiding and helping readers who might have in mind the introduction of quantitative imaging in their practice. The authors acknowledge the subjective nature of parts of this review.

QUANTITATION IS A PREREQUISITE TO THE USE OF NMR AS A BIOMARKER

To be utilized as a biomarker [233, 250], any surrogate indicator of disease or condition status must be expressed as a quantified measure and demonstrate its capacity to reflect specific pathological events with accuracy and precision. To fulfill these requirements, medical imaging had to go through a long evolution, adding a quantitative dimension to the standard qualitative description of pathology. This was possible thanks to a steady flow of technological innovation, which has resulted in spectacular improvements in hardware stability and versatility, and thanks also to the patient and careful development of dedicated protocols [244]. This process, which has affected all systems and organs, was no exception for muscle imaging. For muscle disorders as for other ailments, the benefits of quantitative imaging are multiple: a better appreciation of disease severity, the possibility to monitor pathological changes over time and, most importantly, to evaluate skeletal muscle responses to therapeutic intervention.

Similarly to what was achieved in most organs, NMR has become a key-player for quantitative imaging of the skeletal muscle [112]. Quantitative variables and indices derived from NMR imaging but also spectroscopy constitute the most serious imaging candidates as biomarkers or outcome measures in clinical trials focusing on musclepathology.

NMR OUTCOME MEASURES CURRENTLY AVAILABLE

NMR imaging and spectroscopy can generate a wealth of information on skeletal muscle anatomy, structure and composition, physiology and biochemistry. While a great many NMR variables are being tested and have their potential role as biomarkers being investigated, only 3 NMR imaging outcome measures are widely accepted, if not fully validated, for skeletal muscle longitudinal monitoring. They are almost systematically included in the setting of new clinical trials. They aim at:

SKELETAL MUSCLE TROPHICITY

Due to the combination of high spatial resolution, good contrast, 3D capabilities, and efficient correction algorithms for source of distortions, mainly gradient non-linearities, NMR imaging is the reference method for measuring organs volume and dimensions [120, 125, 177, 277]. The statement applies to skeletal muscle at least as much as to other organs. Accuracy is seldom assessed because it requires autopsy specimens but was excellent when it could be performed [178]. Furthermore, reproducibility and discriminant power have systematically been reported as very high [19, 80, 241, 257], at least as good as with ultrasound [68] or computed tomography [231] and with the advantage of superior tissue characterization (see § below). The following examples illustrate NMR capability to identify subtle changes in muscle trophicity. After botulinic toxin injection into a gastrocnemius muscle of children with cerebral palsy, a 4% decrease in mass was measured, compensated by an increase of 4% of the soleus [268]. Forearm muscles volume determination was repeated with coefficient of variation comprised between 0.8 and 5.7% for the different muscles [227]. The effect of detraining after weeks of flow-restricted concentric and eccentric exercise protocols were evaluated, revealing different responses to detraining depending on the training protocol [282].

In most instances, muscle trophicity rather than volume is the variable of interest. When volumes are determined, normalization to muscle length, and more often to adjacent bone length is performed in order to obtain trophicity indices.

In diseased muscle, true muscle mass and not total muscle volume is a preferred trophicity index. For this reason, the contractile mass index is calculated as the muscle volume (or cross-sectional area)×(1– fat fraction), with the fat fraction taken from the water-fat image analysis (see next §) [273].

In DMD boys, individual muscle contractile volume determination has revealed complex relation with strength decrements. It was proportional to the muscle loss in the quadriceps but it was found considerably more important than expected from the contractile tissue loss in the hamstrings and in the tibialis anterior [273].

In normal muscle or when disease affects homogeneously the muscle, cross-sectional areas measured at well-selected levels, f.i. mid-distance of the thigh, will provide pertinent trophicity indices [110, 183, 234] while considerably shortening acquisition and post-processing steps. However, this approach implies rigorous positioning relative to anatomical landmarks. These landmarks can be identified externally, such as the superior iliac spine or the upper border of patella, or in scout scans acquired prior to the actual measurement [81]. Preference is usually given to external landmarks because of possible complexities in determining the scanner table position when switching between coil configurations. However, correction for slice mis-positioning on successive exams is much easier when the bony structures have been entirely scanned. Better inter-scan reproducibility of cross sectional area determination has been reported with scout-image than with external anatomical landmarks [81] When pattern of muscle destruction by disease is known to be peculiar, f.i. progressing from proximal to distal or when it is simply unknown, the acquisition of entire volumes or at least thick stacks of slices spanning most of the muscle length is highly preferable.

The pediatric population poses additional problems related to the growth process itself. On examinations repeated at intervals allowing growth effects to manifest, it is not easy to decide how slice registration has to be adapted. For lack of better solution, slice registration is usually adjusted in proportion to the length increment. For 3D acquisitions, it can be done during the data processing step. When stacks of 2D slices are acquired, slice gaps have to be increased also in proportion to the measured growth and this is often skipped because it requires an intervention at the time of the acquisition itself.

In practice, except for specific studies, thorough evaluation of muscle trophicity has been very seldom performed so far. It requires patient contouring of each individual muscle that, to this date, has to be performed manually. The task is long, tedious and little rewarding, which explains why the evaluation of muscle trophicity is most often left aside. Very few longitudinal studies of muscle trophicity are available for neuromuscular patients. A positive effect of enzyme replacement therapy on lower limb muscle volume was observed in Pompe patients already after 6 months of treatment [204]. After a one-year observation period, calf muscle cross sectional area of inclusion body myositis patients lost 6.5% while it did not change in Charcot-Marie-Tooth 1A patients [181]. In a small population sample, and using an arguable methodology, no change in muscle trophicity was detected in the lower limbs of spinal muscular atrophy patients [228].

Attempts have been made to develop software for automatic segmentation of muscle images. Very few of the proposed solutions have been able to generate reliable results, and particularly no solution has yet been successful when diseased, fatty infiltrated muscles have to be analyzed. Very recently, a model-based automatic segmentation software was reported to generate results in close agreement with those of manual segmentation [130]. The naive suggestion has been made to determine global estimates of muscle mass based on intensity histogram of entire segments of limbs [166]. It would offer little advantage over simple external measurements and ignores the fact that usually not all muscles are affected simultaneously by disease. In the thigh of GRMD dogs, depending on the muscle considered, trophicity estimated by the muscle volume normalized to body mass was found to increase, stay normal, or decrease with age [138]. Focusing attention on the ones that are moderately to severely affected but not destroyed will increase the capacity to detect inflexion in disease progression whilst global measurement by a dilution effect over less affected muscles would decrease the capacity to detect changes. A more realistic approach seems to turn to interactive software solutions. Here, operator contribution remains mandatory, but it is also the case in practice when using automatic methods with the need to scroll through the segmented volume to check for errors. At least one of such software is under development and is based on the random-walker algorithm [42]. With this method, the operator has just to strike each muscle not on all but only every five to ten slides. In case contours are incorrectly detected by the algorithm, the operator can immediately apply a correction by another pencil strike in the erroneously assigned region. Tests performed on thigh muscles show comparable results to manual segmentation, in terms of muscle volume determination and of inter-operator variability, with the advantage of a processing time cut by a factor of ten at least. This can be further accelerated by optimizing the software reactivity to intervention, also possibly by injecting some prior knowledge of muscle shapes and relativepositions.

The difficulty in segmenting muscle images and the main raison for failure of automatic detection is the frequent absence of visible muscle contours on part of their circumference. This is particularly true with standard T1w or T2w spin echo sequences. One way to improve situation is to perform the segmentation on images that present an improved contrast between muscle tissue and fascia, as it is the case on gradient echo images, acquired at specific echo times when water and fat signals are in opposed phases (see below). In our laboratory, trophicity measurements are systematically performed on this type of images. Attempts to further reinforce contrast of fascia and aponeuroses, with dedicated susceptibility imaging sequences, have so far been unsuccessful.

Another motivation for a fine assessment of muscle trophicity is the frequent absence of fatty degenerative changes in neuromuscular pediatric patients at the first stage of disease. It has been hypothesized by pediatric neurologists that relatively subtle alteration of muscle trophicity might be an early indicator of muscle involvement (Robert Carlier and Susana Quijano-Roy, personal communication). It remains to be proven and the availability of fast, user-friendly segmentation methods will allow thorough testing of this idea.

CHRONIC DEGENERATIVE CHANGES

Chronic damages to myocytes as well as long-lasting fiber structural anomalies will at some point result in contractile tissue replacement by fat and/or connective tissue. Imaging of fibrosis remains a challenging issue, which will be discussed in another paragraph. To the contrary, fatty degenerative changes are easily detected and quantified, taking advantage of the differences in resonance frequencies (chemical shift) and also possibly in T2, or even in T1, relaxation rates between water and lipids hydrogen components. For a comprehensive technical review, read Bley et al. or Hu and Kan [24, 119].

Visual grading of fat infiltration on T1w spin echo images, for example the Lamminen-Mercuri scale [140], is satisfactory for muscle diagnostic purposes but is absolutely inappropriate to monitor the more subtle chronic degenerative changes progression. Supposing that the human eye would be good enough to classify lesions between 1 and 4 without error, which is certainly not the case, a change in classification would happen on average for a 17.6% increase in fat fraction. Even the most severe forms of dystrophy do not achieve such an annual rate of muscle destruction. Failure of Lamminen-Mercuri grading to assess adult limb-girdle muscle dystrophy type I progression was patent in the recent study of Willis et al. It is sometimes proposed as a solution to perform on the same screen a head to head comparison of series of T1w images acquired at different points in time during follow up. Looking that way at repeated series of images certainly helps to detect changes but remains highly observer dependent, with unknown sensitivity threshold and without the indispensable quantitation needed to compare patients or interventions. Trying to take advantage of the apparent simplicity of the approach, attempts have been made quite some time ago [148] and more recently [166, 198] to separate water and fat by a threshold applied to routine T1w images. Some earlier efforts have tried to take into account the possible coexistence of fat and water in the same voxel and have calculated the fat fraction using a linear combination of pure fat and pure muscle signals [148]. The latest studies have just operated a binary separation between fat and muscle voxels, which is totally inadequate for the evaluation of fatty infiltrated muscles in the context of chronic muscle diseases [166, 198]. All these approaches based on standard T1w image analysis assume perfect or almost perfect transmitter and receiver homogeneity, which is never the case. It was possibly an acceptable postulate when it was common to work with low field systems, 0.5T; it is not the case anymore, and certainly wrong at high field, 3T and more, or when reception is made by multiple arrays of surface coils. Until technical solutions are found and implemented to guarantee very high homogeneity of transmitter and receiver fields and/or perfect post-processing corrections of imperfections, one should discourage intensity-based separation of muscle and fat, which is particularly misleading for the non-experts who can be abused by the apparent simplicity of the method.

\spaceskip =.22em plus.1em minus.1emFor the monitoring of muscle chronic degenerative changes, preference is nowadays largely given to water-fat imaging sequences, most often referred to as Dixon sequences [159]. They take advantage of the phase shift that progressively develops during the evolution time of a gradient echo to separate the water and fat resonances. The major advantage of this approach is that the separation between water and fat is to the second order independent of the main magnetic field homogeneity and, as a consequence, extended fields of view covering large volumes can be investigated.

In the standard Dixon version, only the methylene resonance of lipids is considered and two or, better, three images with fat and water successively in-phase and out-of-phase are acquired and allow extraction of the water and fat components. Two-point (extended) Dixon may generate satisfactory results in the liver. However, for muscle applications, swaps between water and fat components can occur between limbs or inside limb segments, the latter being considerably more of a problem [112]. Three-point Dixon is almost always immune to this issue.

Ignoring the other lipid resonances introduces some degree of inaccuracy [271]. This can be improved by a better modelling of the lipid spectrum, usually 3 or 4 main resonances, which requires the collection of 6 echoes and relatively intensive computation [118]. This approach, known as Iterative Decomposition of water and fat with Echo Asymmetry and Least-square estimation (IDEAL) and T2*-IDEAL, is currently the most advance method for water-fat imaging. The multiple echo acquisition implies relatively long echo collection and repetition time, with the need of T2* effect corrections. The IDEAL method measures more accurately the hydrogen fat fraction and has the potential to identify differences in the lipid spectrum, which might be caused by nutrition or disease. There is little evidence of this in the skeletal muscle, or if present, there are likely to be of small amplitude and hardly detectable in routine conditions.

If the relative intensities of the lipid spectrum infiltrating the muscle can be considered as being independent of the patient condition, which seems a reasonable assumption, there is no need to go through a 6-echo acquisition. A linear correction factor can be applied to the fat signal and accurate fat fraction will be obtained from standard 3-point measurement. The correction factor was determined in our lab to be 1.82 [14].

How the muscle fat content is expressed can also vary. One can simply look at the percentage of the actual NMR signal in the voxel or in the muscle that can be attributed to fat. Depending on the image repetition time and echo times chosen, corrections for T1 and T2* effects can be applied. One can go further and try to express the fat content as g of lipids per unit muscle mass or volume. Such procedures have been developed for the liver and require additional assumptions or measurements of the tissue lipid composition [155].

As already stressed, consistency is key to success in clinical trials. Simplicity is a vital factor to ensure consistency. Bearing this in mind, at the time of this writing, we recommend 3-point, 3D when possible, Dixon, with proton density weighting (for example, TR of 10 ms and flip angle of 3°) for water-fat imaging of diseased muscle [112]. The standardized correction factor for lipid spectrum may or not be applied and simple percentage of fat signal parametric maps generated. Accuracy may be slightly affected but not precision nor discriminant power, which is what matters in the context of longitudinal studies, with or without intervention.

When intramuscular fat content is low, there is no need to deploy convoluted hazardous processes [12, 136]. Lipid detection can easily be sensitized by shortening TR in 2D Dixon or increasing the flip angle in 3D Dixon while exactitude of relative fraction will be preserved by applying correction factors for water saturation.

While water-fat separation based on chemical shift differences is the accepted state-of-the-art method to evaluate fatty infiltration of tissues, many clinical groups are still measuring the muscle mono-exponential T2 decay [83, 92, 134, 135, 267]. In absence of mobile lipids in the tissue, an elevated T2 points towards inflammatory, oedematous changes. The T2 increase due to inflammation or oedema rarely exceeds 5 to 10 ms. But when fatty degenerative changes are present, because of the much longer T2 of lipids as compared to water, the mono-exponential fit of T2 decay is largely driven by the degree of fat infiltration and the muscle global T2 becomes essentially a measure of tissue lipid content [37], as demonstrated by the tight correlation between global T2 and fat fraction calculated from Dixon images [14] or lipid fraction measured by ¹H localized spectroscopy [135].

Either based on true water-fat separation or estimated from global T2 changes, a large body of evidence indicates that muscle chronic degenerative changes can be very precisely assessed, disease progression and response to therapy can be finely monitored. It has been demonstrated in a number of neuro-muscular diseases.

In the thigh of Duchenne patients, average fat deposition was reported to be 5% /year, with 50% of fat total content being predictor of loss of ambulation in the year to come [80]. In a small cohort of 3 adult Becker patients, thigh fatty infiltration rate was measured at 3.7% /year [27]. Combining percentage of fat maps with manual segmentation of muscle, contractile cross-sectional areas were determined in Duchenne patients [273]. Confrontation of percentage of fat maps with Lamminen grading showed a systematic overestimation of the fatty degenerative changes with the qualitative methods [271].

Forearm investigation revealed more severe involvement of flexor than extensor muscles, and a much faster progression of fatty infiltration in non-ambulant than in ambulant Duchenne patients [111, 263].

One year administration of corticosteroids to young Duchenne boys stopped the fatty infiltration process in the thighs and legs while fatty infiltration rate was respectively of 7% and 3% in non-treated children [10]. The severity of the disease allowed the identification of fatty degenerative changes over 18 months, based only on T1-weighted signal intensity measures, with the demonstration of inter-individual and inter-muscle variability [113]. Such an imaging approach is obsolete according to current standards of quantitative imaging and cannot be recommended for future investigation. The same method had been used earlier, in combination with global T2 measurements, to describe differential involvement of muscles in five Duchenne patients [92].

A one-year follow up multi-centric study of LGMDI patients unambiguously established the superiority of quantitative water-fat imaging, pinpointing statistically significant fat content differences of 1 to 4% in lower limb muscles, whilst Lamminen-Mercuri gradings did not detect any trend and standard functional assessment fail to reach statistical significance, to the exception of the respiratory function tests [269].

Quantitative water-fat imaging revealed a peculiar bimodal distribution of fatty degenerative changes across muscles of fascio-scapulo-humeral dystrophy patients, as well as a distal to proximal progression within affected muscles [123].

In oculo-pharyngeal muscular dystrophy patients, lower limb fat content increased by 1.5% over 13 months, while it was unchanged in age-matched control subjects [79]. Neither standard extensive functional evaluation (MFM), nor visual grading did show any change during the same observation period.

The high sensitivity of Dixon methods was demonstrated in adult Pompe patients, most of them presenting with a disease with much slower progression than dystrophinopathies. In the lower limb of these patients, the average annual infiltration rate was less than 1% but was detected with a high degree of statistical significance [39].

Whole muscle fat fraction increased significantly during the 12-month follow-up at calf level (1.2% ) but not thigh level (0.2% ) in patients with Charcot-Marie- Tooth disease 1A, and at calf level (2.6% ) and thigh level (3.3% ) in patients with inclusion body myositis [181].

In patients with rotator cuff lesions, quantitative assessment of fatty degenerative changes more than muscle atrophy in affected muscles showed a strong relation with the extent of the muscle tears [185, 188].

It has also been reported in conditions affecting secondarily the skeletal muscle.

In patients with amyotrophic lateral sclerosis, leg global T2 measurements increased significantly over a 4-month observation period, signing the progression of the fatty degenerative changes and was correlated with the decline in foot dorsiflexion maximal voluntary isometric contraction [30].

While leg muscle total fat content was unaffected, type 2 diabetic patients displayed a preferential distribution of lipids intramuscularly [127].

Whole-body Dixon identified an increased fat content in muscles of hyperkalemic periodic paralysis [146].

Changes in skeletal muscle composition have been systematically and consistently found in elderly subjects, even though they are of limited amplitude until a very advanced age. Percentage of intramuscular lipid signal typically doubles, from 2 to 4% , between the second and the seventh decade [3, 13, 46, 182, 214]. For its largest part, it represents a true increase in muscle fat content and it only marginally reflects the contractile tissue loss with aging [46].

In a diseased muscle, the intramuscular fat content can be seen as an integrator of the damages sustained by the tissue over the patient life time, which makes it a robust biomarker. One would then like to use the increase or the levelling-off of fatty infiltration over a defined period of time as a quantitative index of disease progression or response to treatment. In muscle dystrophy, and particularly in Duchenne boys, disease severity can be extremely variable from one patient to another, which translates in annual fatty transformation rates ranging between 3 and 15+ percent [26, 80, 263]. This will seriously complicate interpretation of any therapeutic intervention. If a 5 percent increase in fat content is measured after one year of treatment, does it have to be interpreted as a positive response in a particularly severely affected patient or as a negative response in a mildly affected patient? At a group level, one could obtain the answer by conducting a standard placebo controlled study. Such classical approach still would not resolve the question at an individual level and raise ethical consideration on delaying a potentially effective treatment in a group of patients suffering from a deadly disease. One could also propose to use each patient as its own control and, after an observation period, determine if the slope of the fatty degenerative progression abates after treatment is initiated. It would allow to appreciate each patient’s response but would cause the same ethical dilemma. Because the intramuscular fat content is a cumulative index of all damages to the muscle over the patient’s life span, it is in itself a potent indicator of disease severity if the age of the subjects is taken into consideration. At a particular age, the higher the fat content, the higher will be the anticipated fatty transformation rate.

Even though limited in number, forearm data from the Genethon DMD natural history study support this contention [262]. References tables could be built by pooling all data collected in the different natural history studies.

Along the same line, a strong correlation was observed between the actual fat content in a muscle and the annual intramuscular fat increase in a population sample of adult Pompe patients, who have a relatively mild and homogenous disease course, but also a scattered distribution of fatty degenerative changes across the lower limb muscles [39].

The relation between fat fraction and its progression will inevitably plateau for high levels of fat fraction. One cannot expect fact fraction to increase by 15 percent when fatty replacement reaches 80% , even though it was the case when it was only 40% . A simple solution to circumvent the sigmoid relation between fat fraction and its progression, is to normalize the fatty replacement progression to the remaining contractile fraction, i.e the true muscle transformation rate into fat. Taking the example above, the 15 percent fat increase when contractile tissue was 60 % would, if disease severity remains constant, correspond to a 5% increase when contractile tissue was down to 20% .

DISEASE “ACTIVITY”

More than two decades ago, animal studies have identified that skeletal muscle tissue of mouse models of muscular dystrophy had an elevated T2 [172, 235]. The same observation was made in a larger animal model, the GRMD dog [238, 239, 260]. Interestingly, muscle T2 normalized when gene therapy was successful [190, 259, 283].

Recent reports have confirmed and expanded our knowledge on the muscle T2 alterations in animals with muscle dystrophy. Temporal changes of muscle T2 was precisely determined in mdx mice with a peak between 4 and 8 months of ages, then progressively declining [104, 201, 254], T2 abnormalities were described in other models of dystrophy [163, 252], with different patterns of intra-muscle distributions in Large and in mdx mice. The particular sensitivity of dystrophic muscle to eccentric exercise was illustrated by the higher elevation of T2 in mdx mice submitted to downhill running [165].

When response to micro-dystrophin gene expression was assessed in mdx mice, T2 mapping showed superior discriminant capacity over magnetization transfer ratio and diffusion tensor imaging [192]. Successful U7-exon skipping in GRMD dogs was associated with a decrease in T2 of the treated limbs [142]. Losartan treatment normalized muscle T2 of laminin deficient congenital muscular dystrophy mice [252].

The T2 or the spin-spin relaxation time of muscle water can be interpreted as an indicator of disease activity in the skeletal muscle. Disease activity is a purposively vague designation because a change in T2 is a non-specific event that can be caused by many mechanisms, inflammation, necrosis, muscle dystrophy, acute denervation, any circumstance that can create an intracellular oedema, extracellular oedema or a combination of both. This has been extensively demonstrated in animals [31, 99, 101, 272] and in humans. Physical exercise of moderate to high intensity will increase muscle T2 through intra-myocytic fluid accumulation in contracting myocytes. The process is transient and should recess within hours as opposed to pathological increases of T2. It is important to realize that it may bias the evaluation of patients if they are examined shortly after exercise. For this reason, we recommend to perform imaging before the functional evaluation studies, and this particularly for patients with muscles fragilized by disease.

As mentioned in the previous section, animal models of muscle disorders in general do not develop fatty degenerative changes that infiltrate their musculature and a global increase in muscle T2 as measured via a mono-exponential fit to the NMR signal decay will unambiguously indicate an increase in myocytic water T2. In humans however, fatty replacement of myocytes is a frequent outcome in chronic neuromuscular disorders. Since lipids have a much longer T2 than intra-myocytic T2, a few percent of intramuscular fat content will push mono-exponential T2 up to values comparable to the ones measured in an inflamed or damaged muscle tissue.

Many groups have measured the mono-exponential T2 decay of fatty infiltrated muscles and have reported that it was increasing with age in muscles of Duchenne boys [9, 83, 92, 134, 135, 267]. Others, to the contrary, have observed a decrease of muscle T2 of Duchenne patients over time [83, 263], as observed in animal models. These apparently contradictory results can confuse the reader. In the first instance, global T2 was determined and the increase in T2 reflected the progression of the fatty degenerative changes. The information is redundant with the one provided by Dixon sequences and results obtained with the two techniques are tightly correlated [14]. In the second case, muscle water T2 was specifically measured and the progressive decrease in T2 might either be related to growth since it can be observed in normal subjects, at least in the dog [238], or to the progressive exhaustion of the muscle inflammatory and regenerative capacities.

For the sake of clarity, we advocate for the use of precise denominations: muscle global T2, whose alterations reflect mainly the fatty degenerative changes and muscle water T2, which assesses the involvement of the muscle tissue itself by the pathological processes [37]. Because it mixes information that can be obtained separately, we, and these are personal views of the writers, recommend to abandon global T2 measurements of fatty infiltrated muscles. It generates too much confusion, that can be avoided by more elaborated analysis of the muscle T2 decay (see below § on new developments).

The use of muscle water T2 as in indicator of disease activity is not only possible in humans, it is a highly relevant variable, at least as much as in animal models. Water T2 is abnormally elevated in muscles of Duchenne boys [9, 83, 263], but not in Becker patients [274]. In late-onset Pompe patients, water T2 was found moderately abnormal in approximately one third of all muscles examined [39]. In inflammatory myopathies, muscle T2, as a general rule, is increased [161, 193, 279]. In diagnostic imaging, this is classically detected using fat-saturated (STIR) T2w sequences [55, 258]. In patients with fascio-scapulo-humeral muscular dystrophy, a proportion of muscles are STIR positive and have signs of inflammation at biopsy [236]. However, with this qualitative approach, one looks visually for contrast between abnormal and normal muscles or areas. If all muscles in a segment are inflamed, the examination will likely be considered falsely negative. Such situation was encountered in juvenile dermatomyositis patients [38]. For this reason, and even if not widely accepted by the medical community yet, we systematically use quantitative water T2 maps for the assessment of inflammatory myopathies, particularly for the evaluation of the response to treatment.

Besides the lack of specificity of the muscle water T2 changes, there remain uncertainties concerning the exact temporal relation between the pathological events and the T2 changes. It is unclear whether there is a delay and how long it could be. Discrepancies exist between muscle T2 and clinical status in some patients with inflammatory myopathies. Natural fluctuation of T2 in the course of muscular dystrophy is not documented at all.

Interestingly, muscle water T2 can be found elevated in some congenital myopathies, at least in animal models of nebulin, ACTA1 or dynamin gene mutations [93– 95, 162]. This suggests that a certain degree of cellular disorganization can affect intracellular mobility enough to modify T2. The possibility of abnormal muscle T2 in stable myopathies introduces a restriction to the generalization of the use of muscle water T2 as an absolute indicator of disease activity. Also, muscle water T2 changes after denervation is due to the relative increase in extracellular space that accompanies the muscle tissue atrophy [82, 129, 199] and reflects more the structural reorganization of the muscle than disease activity stricto sensu.

It might then be wiser to restrain the use of muscle water T2 for monitoring disease activity to conditions known for being affected by destructive pathological processes at least during certain periods of their evolution. When it is the case, the determination of muscle water T2 appears to be a powerful biomarker of disease progression or response to intervention. What was clearly demonstrated in animal models, is now supported by several studies in humans. The T2 level has been found to predict the progression rate of the chronic degenerative changes. In young Duchenne boys, after initiation of steroid treatment, leg muscle T2 rapidly decreased by a few ms, this reduction was maintained over time and was associated with a blockade of the fatty infiltration process [10]. As already mentioned, in adult Pompe patients, one third of all lower limb muscles investigated presented a mild to moderate elevation of water T2 in at least one of two scans at one year interval. Those muscles experienced a faster progression of the degenerative changes, approximately 35% faster, than the muscles with normal T2 [39]. For the different muscles, there was a significant correlation between mean T2 value and mean fat infiltration rate. In patients suffering from fascio-scapulo humeral dystrophy, it was shown that muscles with hyperintensities on T2w images would subsequently display significant fatty degenerative changes on later follow up scans [84, 123]. Such observations are pivotal in establishing the utility of muscle water T2 as a biomarker of disease activity in neuromuscular diseases.

Many questions remain to be addressed. One of the most intriguing is the muscle water T2 response to dystrophin expression in Duchenne patients already under steroids. Steroids bring water T2 back close to normal values and it is uncertain whether some degree of dystrophin expression as obtained currently with exon skipping will be able to further decrease T2. Some preliminary observation suggests that it might be difficult to detect (author’s personal observation).

Methodological issues make the exact determination of muscle water T2 challenging. We discussed in detail the difficulties generated by fatty infiltration and replacement. Full suppression of the lipid signal is extremely challenging to obtain and residuals of only a few percent will impact on the water T2 estimate. Preference is nowadays given to methods that do not attempt to minimize the lipid signal but rather to separate the water and fat signals, based on chemical shift or T2 differences, during the acquisition itself or by subsequent processing. There are not many and the most frequently used are based on multiple spin-echo collection, in CPMG regime and covering a large range of echo-times. The Dixon water-fat separation scheme has been incorporated in a multi spin echo diagram, giving birth to the IDEAL-CPMG sequence [122]. Another option is to perform a tri-exponential fitting to the signal decay of a multi spin echo sequence. Taking advantage of the large difference in T2 between water and fat, this method was shown to efficiently separate the two components [14]. Since this is obtained by post-processing exclusively, and the sequence used is standard on most NMR scanners, this approach can easily be implemented in multi-center trials and this is actually the case.

Accurate measurement of T2 is very demanding. In a multi spin echo sequence, excitation and refocusing pulses have to be exactly set tot 90° and 180°. Otherwise, the signal decay will be too fast, if efficient gradient spoiling is available, which is rarely the case on clinical systems, or too slow if unspoiled signal arising from stimulated echoes adds to the spin echo [112]. A practical approach has been to complement the multi TE spin echo sequence with another sequence that maps the B1 field, and gives the exact flip angle experienced by each voxel [280]. When it departs from the prescribed value, the voxel in question is withdrawn from the T2 map [14]. It is simple, efficient, but eliminates a significant proportion of voxels in areas of poor B1 homogeneity. Better solutions exist and are in the process of being implemented for clinical research studies. Instead of fitting exponentials to the signal decay, models following a more exact trajectory of both spin echoes and stimulated echoes can be implemented and suppress the need for B1 map. For example, the Extended Phase Graphs (EPG) formalism is a powerful tool to depict and understand the magnetization response of an arbitrary multi-pulse MR sequence, under different T1 and T2 relaxation constraints. Recently, Lebel and Wilman have proposed to model the MSME signal using the EPG formalism for precise T2 relaxometry [143]. Rooney et al. have suggested that EPG modeling could be used to improve agreement and accuracy of muscle T2 mapping across multiple sites [209].

Accuracy is important in clinical research but, what counts most, is the discriminant power of a technique. What is reassuring with water T2 measurements, is that, even if the measurement is inaccurate because of pollution by stimulated echoes, it remains sensitive to the effect of pathologies and interventions. That was the case in a clinical study where the measured control T2 value was almost twice the real one, but the effect of inflammation and its reversal by steroids was nevertheless properly captured [161].

RELATIONS BETWEEN FUNCTIONAL AND NMR OUTCOME MEASURES

A most frequent and recurrent question is the relation between the standard functional outcome measures for the skeletal muscle and the imaging ones. When a clinical investigator is short of imagination, it seems to have become a favorite proposal: does NMR correlate with function and to which extent? The answer is always known in advance: yes, there is a correlation between the two classes of tools, and the correlation coefficient is typically around 0.7. It is indeed obvious that muscle performance depends for a significant part on muscle mass and composition. It was already patent in early studies using qualitative indices of fatty degenerative changes in Duchenne patients [153]. This has been systematically confirmed [80, 91, 134, 246, 253, 273, 275]. Recent additional evidence was obtained for myotonic dystrophy type I [106], Charcot-Marie-Tooth 1a and inclusion-body myositis patients [181] as well as for rotator cuff lesions [185] and aging [46]. Very logically, muscle strength correlates better with residual contractile mass [273] than with the percentage of fat infiltration [275].

An example of complementarity between imaging and function is the use of the contractile cross sectional area to determine whether a decreased muscle strength or torque is due only to muscle hypotrophy and wasting or if intrinsic contractile defects intervene. In Duchenne boys, the decreased muscle strength remained abnormally low when normalized for the remaining contractile tissue mass, pointing to contraction anomalies of the dystrophic muscle [253, 273].

There are other relevant questions. What is the temporal relation between muscle composition changes and the functional alterations? Data from the Imaging-DMD consortium, presented at a WMS meeting [255] but yet unpublished, showed that leg muscles fat fractions increased up to 60% without major impact on the 6 min walk test in DMD boys, indicating that imaging changes might be a good predictor of clinical prognosis. This is of utmost importance and needs to be confirmed. Other studies, also in DMD patients, indicate, that decline in muscle strength occurs in parallel with the progression of fatty degenerative changes [253, 273], or might even precedes them in the forearms [110].

Another important question is the discriminant power of outcome measures. How do biomarkers compare in terms of their capacity to detect the effect of an intervention, for the smallest possible change in the smallest population sample? Because they are independent of subject’s volition, most NMR imaging quantitative indices have superior reproducibility performance and have been proposed as best candidates to detect and monitor effect of therapy [26, 107]. This conclusion relied on the assumption that therapeutic agent will induce changes of identical amplitude for all outcome measures, which is unlikely. The question remains open but, even if not eliminated, the lesser dependence of NMR outcome measures towards patient’s collaboration constitutes a major asset.

Our enthusiasm must be tempered by the skepticism opposed by the FDA to the use of NMR as an outcome measure to assess the response of dystrophic muscle to exon 51 skipping with drisapersen. The regulatory agency expressed concerns on the impact of imaging platforms diversity, but also of differences in imaging schemes and acquisition parameters, and image analysis procedures on the estimated NMR variables. There were also remarks on insufficient quality control of system stability and assessment of reproducibility. It was objected that the expected effect of treatment might be on the same order as the variability of the technique. Small sizes of population samples were also criticized. Uncertainties on the relation between NMR variables and functional measurements were also stressed. Interestingly, there was again some confusion in this report between global muscle T2 and muscle water T2. The FDA request of a complete harmonization of procedures across sites is unrealistic from a technical standpoint. The only way to reassure regulatory agencies on the value of the NMR measurements will be the careful validation of the processes that generate parametric maps and the demonstration of the independence of the output relative to differences in acquisition conditions.

A definitive advantage in favor of NMR would be the identification of biomarkers that would be able to predict the future clinical outcome very early on, and before any functionally or anatomically detectable response. In oncology, NMR spectroscopic indices have been shown to reveal positive response to treatment before any morphological sign of tumor regression [147, 206, 221]. Similar markers have yet to be identified for neuromuscular diseases. Muscle phosphodiester level is a candidate for muscle dystrophy, where it is elevated in relation with the increased membrane phospholipid turnover [69, 132, 262, 272], and seems to respond very early on when dystrophin is expressed [142].

There is a whole lot of reasons to pursue confrontations between functional and imaging biomarkers in patients with muscular disorders. As a general rule, these are the discrepancies and the deviations from tight correlations that should generate interest. Unravelling their mechanisms will advance knowledge whilst the contemplation of perfect agreement is reassuring but is essentially sterile.

THE ROLE OF ³¹P NMR SPECTROSCOPY AS AN OUTCOME MEASURE

In vivo NMR spectroscopy is a very powerful tool to investigate skeletal muscle energy metabolism non-invasively via ¹H, ³¹P, ¹³C nuclei, to name the most important ones [25]. ³¹P NMR spectroscopy identifies and quantifies key energy metabolites, phosphocreatine, ATP, inorganic phosphate [131]. It measures intracellular pH. Free cytosolic ADP concentration and oxidative and non-oxidative ATP production can be determined. With the relatively simple use of surface coils positioned directly on the segment of interest, it has generated a wealth of invaluable information on skeletal muscle energetics, its regulation during exercise and its disturbances by diseases [8, 44, 102]. So far, NMR spectroscopy has not played a significant role as an outcome measure in neuro-muscular diseases, to the exception of muscle dystrophy.

Years before dystrophin was identified and dystrophin mutations were recognized as the cause of Duchenne muscular dystrophy, striking anomalies were described in the 31P spectra of Duchenne boys [44, 69, 187]. Interestingly, they are present at rest, which notably facilitates their identification, and they point towards anomalies at different subcellular levels, the mitochondria, the contractile apparatus and most importantly, the sarcoplasmic membrane [130, 247, 262, 284]. Total amount of P-compounds is low, as a consequence of muscle atrophy and adiposity, phosphocreatine/ATP is low, indicating a loss in contractile tissue, inorganic phosphate/phosphocreatine is elevated, reflecting higher ADP level at rest, which in turn reveals a dysregulation of the mitochondrial oxidative phosphorylation control and/or an abnormal energy demand to maintain ionic homeostasis despite leaky cell membranes; intracellular pH is alkaline, more precisely there is a prominent pool of inorganic phosphate at a higher pH, coming from damaged dystrophic cells and/or an expanded interstitial compartment related to fibrosis [265]; last but not least, the phospho-diester resonance, mainly composed of glycerophosphocholine, is elevated, an indicator of an accelerated membrane phospholipid turnover in dystrophic myocytes.

Dynamic investigation of muscle energetics with ³¹P NMR seems to have never been performed in DMD boys. In BMD and in DMD/BMD carriers, anomalies of pH regulation during and after exercise have been reported, however not pointing systematically in the same direction [16– 18, 132, 247]. Both increased or defective glycolytic activity has been noted, while oxidative phosphorylation was most often normal [132, 154, 246, 247]. Globally, exercise results were considerably more scattered and did not show the high degree of consistency observed at rest.

The severity of the ³¹P spectral anomalies at rest has been shown to correlate with disease severity, being already detectable in DMD female carriers, present in Becker patients and patent in Duchenne boys [132]. In the fingers flexors of DMD patients, spectral anomalies were shown to be more intense in non-ambulant than in ambulant patients and to relate to the number of years after loss of ambulation [262]. In patients with fascio-scapulo-humeral dystrophy, PCr/ATP ratio was found to decrease at the same time as the muscle entered a phase of rapid destruction and replacement by adipocytes [123].

Again in the fingers flexors of Duchenne patients, the progression of these anomalies was monitored over a one-year period and was significant in the non-ambulant boys [110]. In the GRMD dogs, the forelimbs treated by an adenovirus associated virus mediated U7 exon-skipping, saw the elevated phosphodiester peak resume normal levels [142]. While still relatively anecdotal, these observations strongly suggest that ³¹P spectroscopy and in particular the phosphodiester resonance may provide useful biomarkers of dystrophic muscle response to dystrophin expression. Since ³¹P spectroscopy assesses the sarcoplasmic membrane integrity and stability, it will, by adding some degree of specificity, complement the more global information given by fat fraction and T2 maps. Very interestingly, PDE/ATP was recently reported to be elevated in muscles of BMD patients without any significant increase in fat infiltration [272], suggesting that it might be a very early biomarker of dystrophic processes, whilst water T2 did not seem to be in the same population sample [274].

NMR IMAGING OUTCOME MEASURES IN DEVELOPMENT

Current developments are being pursued in two directions

RECENT TECHNICAL DEVELOPMENTS AND THEIR IMPACT ON THE USE OF NMR AS OUTCOME MEASURE

Muscle imaging, and NMR imaging in particular, is increasingly used as an outcome measure. It still constitutes a relatively young discipline certainly offering a great potential but with a need to grow, and this in several directions. Broadening the range of tools available to further characterize normal and diseased muscles is one of the priority and several of the main options were discussed in the previous paragraphs. Developing protocols dedicated to the investigation of specific muscles, in particular targets located deep in tissues, will also gain importance. Very little has yet been proposed to evaluate diaphragm structure and composition in vivo [237]. Anatomical conformation as well as respiratory motion make diaphragm investigation challenging but its functional role and implication in numerous diseases justify efforts [173]. Another example among many others is the cell therapy for oculo-pharyngeal muscular dystrophy patients [197], with an interest to perform high resolution imaging of laryngeal muscles injected with homologous myoblasts and monitor, if not directly the fate of the grafted cells, at least the possible impact on the degenerative changes progression.

The wealth of additional information on muscle structure and function that NMR will provide in future, and is already largely able to provide, and the multiple anatomical sites to investigate create a major problem, the time required to perform these advanced investigation. The issue is both economical, the examination cost, and practical and ethical, the patient compliance to scans lasting more than one hour. The latter is of course a major concern with pediatric patients. To minimize this problem, a great deal of efforts has been devoted and will continue 1/ to speed up image acquisitions, 2/ to collect several NMR variables, and potentially biomarkers, in one acquisition. The two approaches can in many instances be combined.

CONCLUSIONS

Despite elevated costs, impossibility of bedside operation, technological complexity and sometimes challenging interpretation, NMR imaging and spectroscopy are progressively conquering a significant role as outcome measures for muscle studies. The role of imaging as an outcome measure in muscle studies may still be perceived as lagging behind as compared to the imaging contribution to the evaluation of other organs, like the brain. This situation is largely circumstantial, related to the lesser incidence of muscle disorders, the poor availability of scanners and the lack of therapeutic perspectives. The latter two obstacles are gradually being lifted. Parametric maps of muscle fat fraction and of muscle water T2 provide non-invasive and quantitative measures of skeletal muscle degenerative changes and of disease activity, respectively. They are by far the two NMR biomarkers that are the most widely used in muscle protocols. The perceived added value of such tissue characterization has, despite the limitations and constraints listed above, has prompted the progressive and now quasi systematic integration of NMR imaging as an outcome measure in new therapeutic trials. Current technical developments aim at shortening the sequence acquisitions and at collecting the fat fraction and T2 variables at once and will spectacularly reduce the examination time. Their implementation in multi-center trials can be expected in the next months and years and will significantly improve the NMR examination acceptance by the patients, in particular the pediatric ones. Alternatively, speeding up the acquisition of the fat fraction and T2 maps will free up time to include the determination of other biomarkers in clinical protocols, for instance, the phosphodiester level by 31P spectroscopy, muscle perfusion with arterial spin labeling and/or interstitial fibrosis with UTE sequences. The capability to investigate skeletal muscles, either superficially or profoundly located, and to characterize muscle anatomy, structure, biochemistry and function non-invasively and potentially simultaneously is unique to NMR. Thanks to the steady flow of technological innovation, these properties shall be more and more exploited even in a clinical context in the near future and will reinforce the role of NMR as an outcome measure in skeletal muscle clinical trials.