Quantifying the Metabolic Signature of Multiple Sclerosis by in vivo Proton Magnetic Resonance Spectroscopy: Current Challenges and Future Outlook in the Translation From Proton Signal to Diagnostic Biomarker

Abstract

Proton magnetic resonance spectroscopy (¹H-MRS) offers a growing variety of methods for querying potential diagnostic biomarkers of multiple sclerosis in living central nervous system tissue. For the past three decades, ¹H-MRS has enabled the acquisition of a rich dataset suggestive of numerous metabolic alterations in lesions, normal-appearing white matter, gray matter, and spinal cord of individuals with multiple sclerosis, but this body of information is not free of seeming internal contradiction. The use of ¹H-MRS signals as diagnostic biomarkers depends on reproducible and generalizable sensitivity and specificity to disease state that can be confounded by a multitude of influences, including experiment group classification and demographics; acquisition sequence; spectral quality and quantifiability; the contribution of macromolecules and lipids to the spectroscopic baseline; spectral quantification pipeline; voxel tissue and lesion composition; T ₁ and T ₂ relaxation; B₁ field characteristics; and other features of study design, spectral acquisition and processing, and metabolite quantification about which the experimenter may possess imperfect or incomplete information. The direct comparison of ¹H-MRS data from individuals with and without multiple sclerosis poses a special challenge in this regard, as several lines of evidence suggest that experimental cohorts may differ significantly in some of these parameters. We review the existing findings of in vivo ¹H-MRS on central nervous system metabolic abnormalities in multiple sclerosis and its subtypes within the context of study design, spectral acquisition and processing, and metabolite quantification and offer an outlook on technical considerations, including the growing use of machine learning, by future investigations into diagnostic biomarkers of multiple sclerosis measurable by ¹H-MRS.

Keywords: absolute quantification; biomarker; in vivo proton magnetic resonance spectroscopy 1H-MRS; longitudinal relaxation T1; macromolecules; multiple sclerosis; spectroscopic baseline; transverse relaxation T2.

Figure 1

Relapsing-remitting multiple sclerosis is associated with a number of metabolic changes as measured by proton magnetic resonance spectroscopy in both white and gray matter. (A) Previous applications of proton magnetic resonance spectroscopy (¹H-MRS) in the normal-appearing white matter (NAWM) of relapsing-remitting multiple sclerosis (RR-MS) patients have demonstrated decreases in N-acetyl aspartate (NAA) and glutamate-glutamine (Glx), increases in creatine (Cr), and inositols (Ins), and either decreases or increases in choline (Cho) relative to control. (B) Similar analyses in gray matter (GM) have demonstrated decreases in N-acetyl aspartate and glutamate-glutamine with either decreases or increases in choline.

Figure 2

Secondary progressive multiple sclerosis is associated with a number of metabolic changes as measured by proton magnetic resonance spectroscopy in both white and gray matter. (A) Previous applications of proton magnetic resonance spectroscopy (¹H-MRS) in the normal-appearing white matter (NAWM) of secondary progressive multiple sclerosis (SP-MS) patients have demonstrated decreases in N-acetyl aspartate (NAA) with increases in creatine (Cr) and inositols (Ins). (B) Similar analyses in gray matter (GM) have demonstrated decreases in NAA with increases in inositols.

Figure 3

Primary progressive multiple sclerosis is associated with a number of metabolic changes as measured by proton magnetic resonance spectroscopy in both white and gray matter. (A) Previous applications of proton magnetic resonance spectroscopy (¹H-MRS) in the normal-appearing white matter (NAWM) of primary progressive multiple sclerosis (PP-MS) patients have demonstrated decreases in N-acetyl aspartate (NAA) with increases in creatine (Cr) and inositols (Ins). (B) Similar analyses in gray matter (GM) have demonstrated decreases in NAA, creatine, and glutamate-glutamine (Glx).

Figure 4

Cross-sectional analyses of relapsing-remitting and mixed multiple sclerosis subtypes published between 1991 and 2018 have reported comparing patient groups with age-matched or younger control groups. Shown are the standardized mean differences between MS patient and control groups of 106 ¹H-MRS studies published between 1991 and 2018 that report means as well as standard deviations and/or range for group ages examining metabolic differences in the brain and/or spinal cord between individuals with and without relapsing-remitting, unspecified, or mixed subtypes of multiple sclerosis. A random-effects model exhibited a significant effect of group on subject age across the 106 comparisons, with an overall Hedges' g of +0.28 ± 0.09 (mean difference +2.5 ± 0.8 years) and significance level of p < 0.0001. Many analyses attempted to compensate for such age differences in their statistical modeling procedures. Marker area is weighted by group size. MS: multiple sclerosis; MD: standardized mean difference, reported as Hedges' g; CI: confidence interval.

Figure 5

Cross-sectional analyses of progressive multiple sclerosis published between 1996 and 2017 have reported comparing patient groups with age-matched or substantially younger control groups. Shown are the mean differences between multiple sclerosis patient and control groups of 37 ¹H-MRS studies published between 1996 and 2017 that report means as well as standard deviations and/or range for group ages examining metabolic differences in the brain and/or spinal cord between individuals with and without primary progressive, secondary progressive, progressive relapsing or chronic progressive, or mixed progressive subtypes of multiple sclerosis. A random-effects model exhibited a significant effect of group on subject age across the 37 comparisons, with an overall Hedges' g of +0.73 ± 0.21 (mean difference +6.7 ± 1.9 years) and significance level of p < 0.0001. Many analyses attempted to compensate for such age differences in their statistical modeling procedures. Marker area weighted by group size. MS: multiple sclerosis; MD: standardized mean difference, reported as Hedges' g; CI: confidence interval.

Figure 6

Cross-sectional analyses of relapsing, progressive, unspecified, and mixed subtypes of multiple sclerosis published between 1992 and 2018 have tended to feature control groups with a lower proportion of women than the patient groups with whom they are compared. An investigation of 134 cross-sectional analyses published between 1992 and 2018 that reported group sex ratios found that, on average, control groups contained 6 percentage points fewer women (range 55% fewer to 30% more) than experimentally compared groups of multiple sclerosis patients, with significant disparities from experimental groups in control for both proportion (t = −2.9, two-tailed p = 0.004) and absolute number (t = −2.6, p = 0.01) of tested women. Distinct analyses from the same publications denoted in parentheses. MS: multiple sclerosis.

Figure 7

¹H-MRS research into multiple sclerosis has employed a plurality of methods and software programs for spectral quantification. Among 183 papers published between 1990 and 2018 that describe some method of proton magnetic resonance spectroscopy (¹H-MRS) quantification for at least one individual with multiple sclerosis (MS), 51 (27.9%) of them used commercial linear combination modeling package LCModel, 36 (19.7%) reported direct integration of key resonances without explicit fitting of basis functions, at least 28 (15.3%) employed software provided by the scanner vendor, 23 (12.6%) reported using other software tools, another 22 (12.0%) fit key peaks by simple analytic functions before integrating, 13 (7.1%) employed alternate semi-parametric and parametric combination modeling and quantification schemes, and 10 (5.4%) quantified spectra by peak height alone. These methodological disparities are significant because different quantification methods may exhibit differential vulnerability to confound by variables like spectral overlap, macromolecule- and lipid-heavy baselines, accuracy of measured or simulated metabolite basis functions, and overall spectral quality.

Figure 8

Effect sizes for disease state by multiple sclerosis phenotype in creatine-referenced N-acetyl aspartate as measured by ¹H-MRS. Among 64 publications reporting comparisons with control in proton magnetic resonance spectroscopy (¹H-MRS) N-acetyl aspartate referenced to creatine for individuals with unspecified or mixed, relapsing-remitting, or progressive MS phenotypes in non-lesion voxels, 22 publications did not reject the null hypothesis for comparisons in at least one tissue type (Table 1). Among those reporting significant between-group effects of disease on this parameter, the largest effect size from meta-analysis of studies conducted on individuals with various MS phenotypes was for comparisons involving progressive MS cohorts (including unspecified progressive, secondary progressive, and primary progressive), with a standardized mean difference (Hedges' g) of −1.50 in 25 comparisons over 16 publications reporting group sizes, means, and labeled standard deviations or errors for this metabolite in voxels that were not predominantly lesions. *Standard deviations calculated from group sizes and standard errors. MD: standardized mean difference, reported as Hedges' g; CI: Confidence interval; NAWM: normal-appearing white matter; WM: white matter; GM: gray matter; SP-MS: secondary progressive multiple sclerosis; PP-MS: primary progressive multiple sclerosis; c.c.: corpus callosum.