A Review of MR Spectroscopy Studies of Pediatric Bipolar Disorder

Pediatric bipolar disorder is a severe mental illness whose pathophysiology is poorly understood and for which there is an urgent need for improved diagnosis and treatment. MR spectroscopy is a neuroimaging method capable of in vivo measurement of neurochemicals relevant to bipolar disorder neurobiology. MR spectroscopy studies of adult bipolar disorder provide consistent evidence for alterations in the glutamate system and mitochondrial function. In bipolar disorder, these 2 phenomena may be linked because 85% of glucose in the brain is consumed by glutamatergic neurotransmission and the conversion of glutamate to glutamine. The purpose of this article is to review the MR spectroscopic imaging literature in pediatric bipolar disorder, at-risk samples, and severe mood dysregulation, with a focus on the published findings that are relevant to glutamatergic and mitochondrial functioning. Potential directions for future MR spectroscopy studies of the glutamate system and mitochondrial dysfunction in pediatric bipolar disorder are discussed.

W ith an estimated lifetime prevalence of up to 5.1%, 1 bipolar disorder (BD) is a disabling and often fatal brain disease characterized by recurrent episodes of depression and mania. In pediatric BD, the rate of attempted suicide is 40 times that of healthy adolescents, 2 and BD is the diagnosis imparting the greatest risk for completed suicide. 3 Adding to the morbidity and mortality imposed on patients and their families, the annual economic burden of BD in the United States is at least $151 billion. 4 Despite decades of research, the underlying pathophysiology of BD across the life span is yet to be elucidated. 5,6 The neurobiology of pediatric BD is of particular interest because up to 65% of patients with BD experience its onset before 19 years of age. 7,8 Adolescence is the peak period for the first episode of mania, 9 the mood state that defines the illness. In fact, the World Health Organization ranks BD as the fourth most disabling disease worldwide in persons between 10 and 24 years of age. 10 Leverich et al 11 found that childhood-onset BD is associated with a delay of first treatment of Ͼ16 years. Expert consensus has identified the improved definition and diagnosis of pediatric BD, based on its underlying pathophysiology, as a critical barrier to progress in the field. 12 The National Institute of Mental Health Strategic Plan 13 and A Research Agenda for DSM-V 14,15 both advocate attempts to discover neuroimaging biomarkers of BD. Thus, neuroimaging has an important role to play in translational research in pediatric BD. [16][17][18] Converging lines of evidence implicate 2 related systems in the pathophysiology of BD: 1) alterations in glutamatergic neurotransmission, [19][20][21] and 2) cerebral mitochondrial dysfunction. [22][23][24] They are interdependent because Ͼ80% of all synapses are glutamatergic 25 and approximately 85% of the energy derived from glucose in the brain is used to support glutamatergic neurotransmission and the conversion of glutamate (Glu) to glutamine (Gln). 26,27 Mood-related alterations in cerebral bioenergetics would presumably have an impact on the glutamate system. Support for this is provided by the fact that inhibition of mitochondrial respiratory chain complexes I, III, and IV in an animal model of depression is reversed by administration of the N-methyl-D-aspartate glutamate receptor antagonist ketamine, 28 a novel intervention for refractory BD. 29,30 MR spectroscopy is a neuroimaging method capable of noninvasive interrogation of specific brain metabolites in vivo. Because it allows measurement of the chemical status of specific brain regions, MR spectroscopy is one potential method for establishing quantitative correlates of illness and treatment response in psychiatric conditions such as BD. 31,32 Accordingly, MR spectroscopy has been used extensively in BD research to study both the glutamate system and brain bioenergetics. Two systematic reviews 33,34 and 1 meta-analysis 35 have each concluded that MR spectroscopy studies provide convincing evidence for glutamatergic abnormalities in BD. As reviewed by Stork and Renshaw, 36 the MR spectroscopy literature in BD also provides consistent support for mitochondrial dysfunction. The promising nature of these findings has led to the conjecture that MR spectroscopy studies may represent "a pathway to diagnosis, novel therapeutics, and personalized medicine" in mood disorders. 37 There is an urgent need for translational pediatric BD research, for the reasons enumerated above and because the data suggest that juvenile BD is continuous with adult BD. [38][39][40] The MR spectroscopy literature in pediatric major depressive disorder was recently reviewed, 41 but a review of MR spectroscopy studies of child and adolescent BD is lacking. The purpose of this article is to provide a companion review in pediatric BD, with particular attention paid to evidence for alterations in the glutamatergic system and mitochondrial dysfunction, and to discuss opportunities for further study.

MR Spectroscopy Measures Relevant to Bipolar Disorder
A technical description of MR spectroscopy methods for data acquisition and analysis is beyond the scope of this article, but excellent technical reviews are available. 42,43 MR spectroscopy can be used to study a range of atomic nuclei that possess magnetic properties, including hydrogen ( 1 H), phosphorus ( 31 P), lithium ( 7 Li), fluorine ( 19 F), and carbon ( 13 C). 31 To date, the published pediatric BD literature has largely focused on 2 of these: 1 H and 31 P. 1 H-MR Spectroscopy. The most common spectroscopic imaging method used in BD research is 1 H-MR spectroscopy because the scans can be obtained on standard low-field clinical systems. Glutamatergic 1 H-MR spectroscopy metabolites include Glu, Gln, ␥-aminobutyric acid (GABA) and N-acetyl aspartylglutamate. 44 At the magnetic field strengths used in clinical research, separation of the Glu and Gln resonance is unreliable 42 ; however their combined peak (Glx) can be accurately quantified and therefore is most commonly reported. Although by convention Glx is defined as GluϩGln, GABA may also contribute to the total Glx signal. 45 However, when conventional MR spectroscopy methods are used, the contribution of GABA to Glx is considered very small. 46 Significant findings in 1 H-MR spectroscopy measures considered indicators of mitochondrial dysfunction in BD include the following: decreased NAA, decreased total creatine, increased total choline (tChol), increased Glx, and increased mIns. 36 NAA is synthesized inside neuronal mitochondria from L-aspartate ϩ acetyl coenzyme A by the enzyme L-aspartate-N-acetyl transferase in an energy-dependent process, suggesting that decreased NAA concentrations are consistent with impaired mitochondrial bioenergetics. 36,47 The concentration of NAA by in vivo 1 H-MR spectroscopy methods is consistently higher than that found by careful 1 H-nuclear MR analysis of freeze-clamped animal brain tissue, suggesting additional contributions to the "NAA" in vivo peak. 48 The total creatine peak is composed of phosphocreatine (PCr), a temporal and spatial buffer of adenosine triphosphate (ATP), and creatine (PCrϩCre). tChol is a trimethylamine peak that is composed of phosphocholine, a membrane phospholipid precursor; glycerophosphocholine, a membrane phospholipid breakdown product; and choline, acetylcholine, carnitine, and acetyl-L-carnitine. The replicated finding of increased tChol in adult BD is hypothesized to be due to increased phospholipid turnover resulting from mitochondrial dysfunction. 36 The Glx peak contains contributions from glutamate, glutamine, and ␥-aminobutyric acid. The largest contributors to the Glu resonance are Glu in metabolic pathways and, to a much lesser degree, the neurotransmitter Glu. Increased Glx in BD is hypothesized to reflect Glu-induced neuronal hyperactivation, 36 which places abnormally large demands on neuronal and glial energy metabolism. 46 Notably, the classic medication lithium has a significant normalizing effect on Glx in BD. 49 The mIns resonance consists primarily (Ͼ95%) of the cyclic sugar alcohol mIns, with minor contributions from inositol sugar phosphate compounds and glycine. 50 The relevance of mIns to BD stems from its status as a potential indicator of altered membrane metabolism resulting from mitochondrial dysfunction 36 and the fact that a decrease in mIns is associated with administration of the BD medication lithium. 36 A representative 1 H-MR spectroscopy spectrum is shown in Fig 1. 31 P MR Spectroscopy. Another MR spectroscopy technique used in BD research is 31 P-MR spectroscopy, which requires specialized hardware and software, such as a dual-tuned 1 H-31 P radiofrequency coil, a broadband radio-frequency power amplifier, and customized pulse sequences. Investigator experience with 31 P-MR spectroscopy pulse-sequence design and data processing/ analysis is also essential because MR imaging system manufacturers do not typically supply these tools. Yet 31 P-MR spectroscopy may provide unique insights into BD neurobiology because it is a validated method for in vivo measurement of the ultimate mitochondrial process: ATP synthesis. 51 Few 31 P-MR spectroscopy studies of adult BD have been reported, 52 and to date, there are just 3 published 31 P-MR spectroscopy investigations in the pediatric BD literature. 53-55 31 P-MR spectroscopy measures relevant to mitochondrial function include phosphomonoesters, phosphodiesters, ␤-nucleoside triphosphate, PCr, inorganic phosphate (Pi), and intracellular pH. 52 The phosphomonoesters signal contains contributions from the membrane precursors phosphoethanolamine and phosphocholine, in addition to sugar and inositol phosphates. 36 Phosphomonoesters are the building block precursors of neuronal membrane phospholipids. Both increased and decreased phosphomonoesters has been observed in studies of BD; thus, changes in the phosphomonoesters resonance may be state-dependent-with increased phosphomonoesters in depression and mania reflecting increased membrane phospholipid turnover, with decreased phosphomonoesters in subjects with euthymic BD possibly reflecting an opposite quiescence. 36 The phosphodiester signal, made up of contributions from glycerophosphocholine, glycerophosphoethanolamine, and mobile phospholipids, represents the breakdown products of phospholipid membranes. 52 PCr is the buffer storage form of ATP and serves as the substrate reservoir for the creatine kinase reaction. 56 In the mitochondria, this reaction reversibly converts PCr into ATP ϩ creatine in a 1:1 molar ratio. 57,58 Neuronal energy demands are met through a shift in reaction equilibrium, which is designed to maintain constant ATP concentrations. 59-61 ␤-nucleoside triphosphate predominately measures ␤-ATP levels and is, therefore, used as a proxy measure of relative ATP concentrations. Pi and adenosine diphosphate are the products of the ATP hydrolysis reaction and are released when ATP is consumed. Intracellular pH can be calculated by using a modified Henderson-Hasselbalch equation and the resonance Pi relative to PCr. 62 Figure 2 presents a representative 31 P-MR spectroscopy spectrum.

Search Strategy
A computer-assisted literature search by using PubMed and MEDLINE data bases of the National Library of Medicine was conducted to identify reports focusing on pediatric BD samples studied with MR spectroscopy. The following terms were included in the search: "MR spectroscopy," "bipolar disorder," "pediatric or child or adolescent or juvenile or early-onset." A backward search of bibliographic references from the identified references was performed to ensure inclusion of relevant articles; a forward citation search for identified studies was also performed. We also included reports from MR spectroscopy studies of at-risk youth who were the offspring of parents with BD and subjects with severe mood dysregulation, 63 also known as disruptive mood dysregulation disorder, 64 a new mental illness of childhood published in the Diagnostic and Statistical Manual of Mental Disorders, 5th ed, 65 the diagnostic manual for US psychiatrists, in May 2013. All relevant articles published in English were included, and due to the small number of studies, no MR spectroscopy methodologic exclusion criteria were applied.

Overview of the Literature
The literature searches yielded 55 citations, of which 26 contained original neuroimaging acquired from subjects younger than 18 years of age. The sample characteristics, scanning acquisition methods, voxel location, and MR spectroscopy metabolite results are presented in Tables 1-4, and the key findings related to glutamatergic and mitochondrial function are summarized below. In addition to the modest number of reports and diversity of study methods and samples, this literature is in its infancy: The first MR spectroscopy study of pediatric BD was published in 2000. Despite this, the investigations conducted to date point to numerous important directions for further study.

Cross-Sectional MR Spectroscopy Studies of Pediatric BD
Cross-Sectional Studies of the Glutamate System. A summary of the cross-sectional MR spectroscopy studies comparing patients with pediatric BD versus healthy controls (HC) is shown in Table 1. Castillo et al 66 were the first to study juvenile patients with BD and controls in 2000, reporting that patients with BD showed elevated Glx in the bilateral frontal lobes and basal ganglia. In a cross-sectional study of stably medicated pediatric patients with BD versus patients with mania, Glx was decreased in mania; this finding, the authors noted, could represent anterior cingulate cortex (ACC) hypometabolism. 67 In another study, decreased ACC Gln was found in pediatric patients with manic BD, compared with both controls and stably medicated patients, 68 while there were no differences in Glu.
Cross-Sectional Studies of Mitochondrial Function. The frontal lobes have been the subject of numerous case-control studies in pediatric BD, with several investigators reporting decreased NAA concentrations in subjects with pediatric BD compared with controls. [69][70][71][72][73] In one report whose findings were in the opposing direction, Patel et al 74 reported increased NAA in the ACC in patients with pediatric BD depression. Two studies reported increased mIns in pediatric BD, compared with both HC 74 and intermittent explosive disorder. 75 One study of pediatric mania found reduced tChol levels compared with HC, 75 while a study of pediatric BD depression found an increase in tChol. 74 As noted above, 3 recent studies of pediatric BD used 31 P-MR   79 treated adolescents with BD depression with lithium and found that mIns concentrations did not significantly change from baseline, though pretreatment cortical mIns was significantly lower in patients who achieved remission. The same investigators later reported a significant decrease in NAA levels in response to treatment with lithium. 80 DelBello et al 81 treated pediatric patients with mania with olanzapine and found that increased tChol was associated with treatment; in addition, treatment remitters demonstrated increased NAA while nonremitters showed a decrease in NAA. Chang et al 82 randomized subjects with pediatric BD depression to quetiapine or placebo and found that decreased mIns levels were associated with a positive treatment response. One repeated-measures MR spectroscopy study has been performed with a high-risk sample of youth with subsyndromal mood symptoms and at least 1 parent with BD: Chang et al 83 found no significant alterations in NAA, tChol, or mIns associated with valproic acid treatment. However, a large effect size was noted for decreased posttreatment NAA in the right dorsolateral prefrontal cortex. 83

Cross-Sectional MR Spectroscopy Studies of Youth at Risk for BD
Cecil et al 69 conducted a study of mood-disordered children with a familial risk for BD and found increased mIns and decreased NAA in the high-risk subjects compared with HC (  cant between-group differences. Singh et al 85 studied an at-risk sample of offspring of parents with BD, focusing first on the ACC and then on the cerebellar vermis. 86 In the ACC, the authors found decreased absolute Glu concentrations in BD compared with both HC and offspring with subsyndromal BD symptoms. 85 Then, studying at-risk offspring without BD and HC, Singh et al 86 reported decreased cerebellar mIns and tChol. Finally, Wozniak et al 87 measured ACC Glu in children with at least 1 parent with BD, dividing their sample into children with high-versus-low scores on a Child Behavior Checklist 88 profile proposed as a correlate of pediatric BD. 89,90 No group differences in ACC Glu were found, but in the high-score group, there was a positive correlation between Glu levels and Child Behavior Checklist profile scores. 87

Cross-Sectional and Repeated-Measures MR Spectroscopy Studies of Severe Mood Dysregulation
The publication of the Diagnostic and Statistical Manual of Mental Disorders, 5th ed 65 in May 2013 introduced disruptive mood dysregulation disorder as a new mood disorder of childhood (Table  4). Designed to differentiate children who present with severe, nonepisodic irritability from those with BD, 91 disruptive mood dysregulation disorder is closely related to the severe mood dysregulation syndrome defined by Leibenluft 63 and Leibenluft et al. 92 Dickstein et al 93 conducted a series of prescient MR spectroscopy investigations of severe mood dysregulation. In a case-control study, the investigators used 1 H-MR spectroscopy to interrogate the frontal, temporal, and parietal cortices in severe mood dysregulation versus HC; their main finding was reduced mIns in the temporal cortex, though female subjects with severe mood dysregulation showed trends toward increased temporal tCr and Glx. The authors then conducted a randomized controlled trial of lithium in severe mood dysregulation, selecting this intervention on the basis of its potential effects on irritability, aggression, and neurometabolism. 94 MR spectroscopy scans were acquired at baseline and repeated following 6 weeks of treatment with lithium or a placebo. A significant treatment Group ϫ Time interaction was found for parieto-occipital Glx, which increased in the lithium group and decreased in placebo group. 94

DISCUSSION
Pediatric BD is a prevalent and disabling illness, for which progress in timely diagnosis and effective treatment is urgently needed. Research in psychiatry is increasingly focused on biomarker discovery, 95,96 and a consensus is emerging that MR neuroimaging investigations, 16 including multimodal approaches that include MR spectroscopy, 17 offer significant promise in elucidating the pathophysiology of BD. 97

Findings Related to the Glutamate System in Pediatric BD
It has been proposed that MR spectroscopy studies of the glutamate system may hold the key to elucidating the pathophysiology of BD and to identifying novel treatment interventions. 37 In adults, a consistent pattern has emerged in meta-analyses of the MR spectroscopy mood disorder literature: increased cerebral Glx levels in BD 33,35 and decreased Glx in major depressive disorder. 33,98 In comparison, there have been relatively few pediatric  Using high-field 4T scans, these investigators were also able to parse the Gln and Glu resonances in the ACC. Most intriguing, they found decreased Gln in untreated youths with BD compared with both stably medicated patients with BD and HC. 67 Taken together with the fact that there was no difference in Glu among the 3 groups, 67 this finding suggests that the Gln/Glu ratio measured in adult BD investigations 99,100 merits further study in pediatric BD. Finally, in a longitudinal valproic acid treatment study, Strawn et al found decreased baseline Glx in treatment remitters and decreases in Young Mania Rating Scale scores correlated with decreased Glu concentrations. 76 The authors concluded that the predictive value of MR spectroscopy neuroimaging "may relate to a disturbance in either glutamine or GABA, or in the homeostatic equilibrium of Glu and glutamine," 76 providing further support for the exploration of the Gln/Glu ratio as a potential biomarker in pediatric BD.

Findings Related to Mitochondrial Dysfunction in Pediatric BD
There are 5 reports of decreased cortical NAA concentrations in pediatric BD, [69][70][71][72][73] 80 ; and no significant change in NAA following treatment with lithium, 78 quetiapine in BD depression, 82 or youth at risk for BD treated with valproic acid. 83 While additional work will be required to confirm the role of NAA in pediatric BD, the 5 casecontrol studies [69][70][71][72][73] showing decreased NAA are in agreement with the findings in the adult literature. Studies of mIns have reported an increase in pediatric BD in both the manic 75,78 and depressed mood state. 74 In addition, Davanzo et al 78 showed that decreased ACC mIns is associated with a positive response to acute lithium treatment, a finding that is consistent with the molecular mechanism common to mood-stabilizing medications. 102 To date, only three 31 P-MR spectroscopy studies of pediatric BD have been published. Shi et al 53 studied 14 depressed subjects with BD and 24 HC and found that unmedicated BD had decreased ACC Pi compared with both HC (17%; P ϭ .038) and medicated BD (24%; P ϭ .022). In a study of 8 subjects with BD and 8 HC, Sikoglu et al 54   production. 78,103 Furthermore, it is thought that the only Pi that is detectable by nuclear MR imaging is involved in oxidative phosphorylation. 104,105 Mammalian cells in which oxidative phosphorylation is impaired can reduce the concentration of free Pi via compartmentation to the inner mitochondrial membrane, which immobilizes the phosphorus ions and renders them invisible to MR spectroscopy. [106][107][108] In addition, Pi has a direct effect in vitro on glucose use in cortical neurons. 109 It has been posited that decreased ATP consumption leads to a fall in cytosolic Pi, to a level that balances ATP synthesis and use, 110,111 thus stabilizing the phosphorylation potential of the cell. 107,108 Additional support for the potential relevance of Pi in the pathophysiology of BD is provided by the observation that the activity of sodium-potassium adenosine triphosphatase (Naϩ / K Ϯ ATPase), an enzyme partially regulated by Pi, [97][98][99][100][101][112][113][114] is altered in patients with BD. 115 In the third and most recent 31 P-MR spectroscopy study of pediatric BD, Weber et al 55 found reduced intracellular pH and decreased adenosine diphosphate concentrations in the ACC. These results are in line with the consistent finding of reduced intracellular pH in studies of adult BD 36 and with the discovery of Chance et al 116 that adenosine diphosphate is one of the principal controllers of oxidative metabolism.

Limitations of Pediatric MR Spectroscopy Studies
Despite the consistent difference in the glutamatergic entity Glx between patients with BD and controls found by 3 independent analyses of the adult MR spectroscopy literature [33][34][35] and the reports of glutamatergic alterations in pediatric BD discussed here, there are uncertainties associated with MR spectroscopy measures of Glu. In the brain, Glu plays at least 3 key roles: It is the major excitatory neurotransmitter, it serves as the precursor for the major inhibitory neurotransmitter GABA, and it is involved in the synthesis of smaller metabolites, including glutathione, as well as larger peptides and proteins. 117 Glu is involved in a variety of metabolic pathways, including the neuronal tricarboxylic acid cycle, the astrocytic tricarboxylic acid cycle, pyruvate carboxylation, and the glutamate-glutamine cycle that links neuronal and astrocytic metabolism. 118 This metabolic compartmentation leads to spatial uncertainty because Glu is present in glutamatergic neurons, GABAergic neurons, and astrocytes, in addition to extracellular spaces. The published studies of pediatric BD do not parse these multiple Glu pools in terms of either function or location.
Another important limitation of the extant pediatric BD MR spectroscopy literature is the static nature of the measurements reported. Static measures are likely to be insufficient in generating a comprehensive picture of BD pathophysiology because brain metabolism is predominantly made up of dynamic processes (ie, enzyme-catalyzed reactions and the transfer of chemical groups through metabolic pathways). Therefore, elucidation of BD etiopathogenesis may require the use of techniques capable of dynamic in vivo measures. Two examples of these that may find application in the study of BD are magnetization transfer, which can be used to measure the K f of the creatine kinase reaction, 56 and dynamic 13 C MR spectroscopy, which has been validated as a method for studying neuronal bioenergetics and Glu neurotransmission and cycling. 119

Future Translational Directions in MR Spectroscopy Studies of Pediatric BD
Improved diagnosis and treatment for pediatric BD are urgently needed: While the evidence suggests that pediatric bipolar illness is continuous with adult BD, 38-40 the delay to first appropriate treatment experienced by individuals with childhood-onset BD averages more than 16 years. 11 To date, MR spectroscopy studies of pediatric BD comprise a small, albeit rapidly expanding, literature. On the basis of its ability to measure neurochemical entities relevant to glutamatergic and mitochondrial function in vivo, MR spectroscopy will play a vital role in future translational studies of pediatric BD diagnosis, treatment development, and personalized medicine. 37,120 Cross-sectional studies are needed to determine whether MR spectroscopy can reliably distinguish pediatric BD from disorders with overlapping symptoms, such as disruptive mood dysregulation disorder, attention deficit/hyperactivity disorder, and, especially, major depressive disorder. 121 Comparison with MR spectroscopy studies of normal brain development 122 will shed light on the natural history of BD and point investigators toward opportunities for intervention. Longitudinal studies may determine the predictors of continuity with adult BD and whether timely treatment is capable of altering the course of BD in at-risk individuals. Novel study designs combining MR spectroscopy with other neuroimaging methods in a multimodal approach can be used to increase the dimensionality of the information gleaned from a study sample. 17,123,124 Finally, future studies would benefit from larger sample sizes because it has been calculated that MR spectroscopy investigations require analyzable data on at least 39 affected subjects and 39 HC to have adequate power to detect a 10% group difference in neurometabolite concentrations. 125,126