Postmortem brain studies of bipolar disorder (BD) and schizophrenia (SZ) have revealed extensive abnormalities in the number, packing density, and size of neurons and glia in the cerebral cortex (1, 2). Cell body size reductions of 16-18% are reported in both BD and SZ (3). These findings suggest that BD and SZ are characterized by an inability to promote the development, survival, and plasticity of existing cells and to generate functional cells de novo through neurogenesis and gliogenesis (4). Although neuroimaging studies have reported volume reductions in specific macroscopic brain structures in BD and SZ, there are no in vivo reports specifically examining abnormalities in cell size and density in these conditions.
Magnetic resonance spectroscopy (MRS) can provide a window on the brain's cellular microenvironment through the measurement of T2 (transverse) relaxation times. T2 relaxation is a result of nuclear spin-spin interactions and is sensitive to changes in molecular motion, mainly through interactions of small molecules (metabolites) with structural or cytosolic macromolecules. T2 relaxation leads to MRS signal decay as echo time (TE) increases. Signals from both water molecules in brain tissue and MRS-visible metabolite molecules are affected by T2 relaxation; we will call these water T2 relaxation and metabolite T2 relaxation, respectively. T2 relaxation times can convey valuable neurobiological information. For example, there is a dramatic reduction in brain water T2 relaxation time during early postnatal brain development as water molecules increasingly interact with rapidly proliferating macromolecules (lipid membranes, myelin components, cytosolic proteins) (5). T2 relaxation times of major brain metabolites likewise reflect their differential tissue distribution and local molecular interactions (6, 7). For example, total creatine (tCr) has a shorter T2 relaxation time than either total N-acetylaspartate (tNAA) or total choline (tCho) for poorly understood reasons, possibly because it “interacts more closely with a macromolecular assembly” (8).
Water T2 relaxation times are prolonged in patients with SZ, particularly in frontal and temporal gray matter (GM) (9, 10), and fornix (11). One study showed that this prolongation extends into both gray and white matter (12). Because water T2 relaxation times are determined mainly by water-macromolecule interactions, these studies suggested that SZ may be associated with neuronal hypofunction and possibly neurodegeneration, leading to an impoverished macromolecule compartment and prolonged water T2 relaxation times.
T2 relaxation times for intracellular metabolites such as tNAA, tCr, and tCho (13) have not been studied in psychiatric conditions. In this study, we quantified brain metabolite T2 relaxation times in a J-resolved MRS dataset from acutely ill inpatients with BD or SZ, and matched healthy controls (NC) at 4 T. We collected data from two voxels: one in the anterior cingulate cortex (ACC), a brain region where postmortem studies have found cellular abnormalities in BD and SZ (1, 2), and one in the parieto-occipital cortex (POC), a region that has not been implicated in these disorders. We hypothesized that SZ and BD patients would have shorter metabolite T2 relaxation times, caused by cellular atrophy and reduced intracellular molecular mobility in these conditions.
MATERIALS AND METHODS
Following approval by the McLean Hospital institutional review board, 21 NC, 39 BD manic, and 18 SZ subjects were recruited. Patients were recruited from a psychiatric inpatient unit. Subjects with significant neurologic or medical problems, current substance abuse, or history of substance dependence were excluded. Tobacco abuse or dependence was not assessed, and tobacco smokers were not excluded from the study (five BD and five SZ patients identified themselves as habitual smokers). All subjects had negative urine toxicology tests on admission. Following provision of informed consent, we used an Informed Consent Survey, which asks 10 simple questions about the study, such as “What illness is being studied?” and “If you don't want to, do you have to be in this study?” Every subject answered these questions correctly, and answers were documented. Subjects were assessed using the Structured Clinical Interview for DSM-IV (14), Young Mania Rating Scale, Montgomery-Asberg Depression Rating Scale, and the Positive and Negative Syndrome Scale on scan day. Demographic and clinical variables and statistical comparisons are reported in Table 1.
Although severity of mania and psychosis cannot be directly compared, all patients were hospitalized with acute psychotic episodes, suggesting that all had high degrees of psychopathology. All BD subjects met full DSM-IV criteria for current manic episode; none were identified as rapid cycling. Eight SZ subjects were diagnosed with schizoaffective disorder and were not currently in a mood episode, according to the Structured Clinical Interview for DSM-IV. Usable data were obtained from 20 NC, 15 BD, and 15 SZ subjects. Inability to tolerate the scanning environment due to psychiatric condition was the primary reason for study dropout, especially for BD subjects. Symptom levels among BD subjects who did not complete the study (N = 24) were somewhat higher than for BD subjects who provided usable data (Young Mania Rating Scale: 26.9; Montgomery-Asberg Depression Rating Scale: 18.4; Positive and Negative Syndrome Scale: 82.0; compare with Table 1). Chlorpromazine equivalents were calculated for patients taking antipsychotic medications (15).
All subjects underwent a structural MRI scan in a Siemens 3 Tesla Trio MR scanner (Erlangen, Germany) with a quadrature radiofrequency coil. TE/pulse repetition time/inversion time times were 2.74 ms/2.1 sec/1.1 sec, with an echo spacing of 6.3 ms, 12° flip. Field of view (FOV) was 256 mm, with 1 × 1 mm pixels, and 1.33-mm slices. Subjects with structural abnormalities were excluded from the study.
All 1H MRS acquisitions and related brain imaging were conducted on a 4-T full-body MR scanner (Varian/UnityInova; Varian Inc., Palo Alto, CA), using a 16-rung, single-tuned, volumetric-birdcage coil. A rapid two-dimensional gradient-recalled echo sequence (12 sec) acquired sagittal, coronal, and axial images to ensure optimal patient positioning. Manual global shimming of the unsuppressed water signal yielded an unfiltered global water line width of ≤30 Hz. T1-weighted sagittal images (TE/pulse repetition time = 6.2/11.4 ms, field of view = 24 × 24 × 8 cm, in-plane resolution = 0.94 × 1.88 mm, slice thickness = 5 mm, readout points = 512, flip angle = 11°), and axial T1-weighted images (similar parameters as above, except for in-plane resolution = 0.94 × 0.94 mm, slice thickness = 2.5 mm) provided clear differentiation between cortical GM and white matter and served as anatomic guides to MRS voxel positioning.
A 2 × 2 × 2 cm single voxel was placed on the ACC, along the midline, just superior and anterior to the genu of the corpus callosum (Fig. 1). Pilot studies indicated that susceptibility artifacts preclude the acquisition of usable data from more inferior voxels. A standard point-resolved spectroscopy sequence modified for the current multi-echo 1H MRS protocol and employing a four-pulse water suppression enhanced through T1-effects sequence (16) was used. Manual shimming of the magnetic field within the prescribed voxel was carried out, with resulting water line widths of less than or equal to 11 Hz. Following tip angle, water-suppression, and RF pulse power optimization using automated methods, 48 TE-stepped spectra were acquired from the voxel, with the echo-time ranging from 30 ms to 500 ms in 10-ms increments (Fig. 2). Specifically, the second TE period in the point-resolved spectroscopy sequence was progressively prolonged by symmetrically flanking the last 180° pulse prior to signal acquisition with two identical delay periods. Acquisition parameters were pulse repetition time = 2.0 sec, acquisition bandwidth = 2 kHz, repetitions = 16; nominal voxel volume = 8 cc; approximate scan duration = 28 min.
The identical process was repeated on a 2 × 2 × 2 cm POC voxel, with one exception: because the quality of spectra was superior, there were eight repetitions leading to an acquisition time of approximately 14 min. The POC voxel was placed along the midline, posterior to the splenium of the corpus callosum and superior to the calcarine sulcus, encompassing as much GM as possible (Fig. 1). In total, time in the magnet was 75-90 min.
The full width at half maximum of the water resonance was: 9.9 ± 1.2 and 8.7 ± 0.8 Hz for NC, 9.3 ± 1.5 and 8.6 ± 0.7 Hz for BD, and 9.4 ± 2.3 and 8.9 ± 0.9 Hz for SZ subjects (in ACC and POC, respectively). There were no between-group differences for this measure.
MRS Data Processing and Analysis
All MRS processing was carried out fully automated and blinded to diagnosis. An MR physicist (J.E.J.) evaluated all spectra, blind to diagnosis and brain region, and identified those inappropriate for quantification (due to low signal to noise and/or spectral resolution). Thus, two ACC and one POC spectrum in NC, one POC spectrum in BD, and one ACC spectrum in SZ were excluded (from five different subjects). Quantification of metabolite T2 relaxation times in either voxel was not possible for one additional subject from each group. These subjects were excluded from the statistical analysis, leading to a total of 19 NC, 14 BD, and 14 SZ usable datasets.
We used the commercial spectral-fitting package, LCModel (version 6.0-1) (17) to measure metabolite peak integrals. For spectral fitting with LCModel, we utilized GAMMA-simulated (18) theoretical basis sets for N-acetylaspartate (NAA), N-acetylaspartylglutamate, alanine, γ-aminobutyric acid (GABA), aspartate, choline (Cho), creatine (Cr), glucose, glutamate (Glu), glutamine, glutathione, glycerophosphocholine, glycine, myo-inositol, scyllo-inositol, lactate, phosphocreatine, phosphocholine, serine, and taurine. We used GAMMA to generate 48 theoretical, TE-stepped spectra ranging from 30 ms to 500 ms in 10-ms increments, with a hard pulse assumption. Each GAMMA spectrum was modeled at 2-kHz spectral bandwidth, with 1024 complex pairs and a 2-Hz Lorentzian line shape, zero and first-order phased with no baseline roll. We also modeled a formate peak at 8.45 ppm and a tetrameth-silla (TSS) reference peak at 0.0 ppm to mimic the standard LCModel stock solution required for basis-set generation (19). These TE-specific metabolite free induction decay (FIDs) were converted into separate LCModel basis sets for each metabolite. This process resulted in artifact-free basis sets for each TE.
Macromolecules were not included in the LCModel basis sets. It would be desirable to have an estimation of macromolecule contributions to our data, but we do not think that these could account for our findings because any between-subject difference in macromolecule resonances would be present at all TEs and would not affect the T2-related signal change we observe. In addition, macromolecule resonances have short T2 relaxation times, so that any T2-related change in them would have taken place long before we obtain our first spectrum at 30 ms.
LCModel provides Cramer-Rao lower bounds, an estimate of the variance associated with fitting for each TE-specific spectrum. The mode, median, and mean Cramer-Rao lower bounds ranged between 3 and 6% for tNAA, 4 and 5% for tCr, and 6 and 8% for tCho. Metabolite-specific T2 relaxation times were obtained from a two-parameter exponential fit of the metabolite peak integral vs TE decay curve using an iterated Marquardt-Levenberg algorithm (using Origin 6.0 software; Microcal Inc., Northampton, MA) (Fig. 3). There was no clear evidence of biexponential T2 decay for any metabolite.
Tissue-segmentation of T1-weighted images into GM, white matter, and cerebrospinal fluid used FMRIB's Automated Segmentation Tool (Oxford, UK). The percentage of GM in ACC and POC was 80 ± 4% and 70 ± 9% (NC), 76 ± 9% and 69 ± 8% (BD), and 79 ± 2% and 71 ± 8% (SZ); that of white matter was 16 ± 4% and 29 ± 9% (NC), 17 ± 9% and 29 ± 8% (BD), and 17 ± 4% and 28 ± 8% (SZ); and that of cerebrospinal fluid was 5 ± 3% and 2 ± 2% (NC), 7 ± 4% and 2 ± 1% (BD), and 4 ± 2% and 1 ± 1% (SZ), respectively. There were no statistical between-group differences in these measures.
We were also concerned about data stability during the lengthy acquisitions. We reasoned that NAA full width at half maximum line width would vary significantly if motion or other sources of variance caused instability during the scan. Therefore, we measured the mean NAA line width for each subject. This measure ranged from 6.1 to 13.5Hz. Although there was some variation in this measure as TE was modulated, the mean within-subject coefficient of variation (= standard deviation/mean across the spectra for that subject) was 14.4% for the full dataset. Coefficients of variation for the BD and SZ groups were comparable to that for the control group, and there were no significant between-group differences in this measure (not shown).
Note that these data represent a different aspect of the analysis of the same experiments we reported previously (20). However, the data were analyzed differently in the current report. In the previous work, we reported metabolite concentrations. For that purpose, we used a fast Fourier transform in the f2 dimension, leading to a truly two-dimensional (“J-resolved”) dataset, which was then resampled and fitted using LCModel (20). This approach allows more reliable quantification of coupled spin resonances such as glutamine and myoinositol. In the current manuscript, we did not use the same approach because we were interested in fitting the exponential decay curves for each of the major singlet resonances tNAA, tCr, and tCho, with increasing TE in order to obtain T2 relaxation times. Therefore, we took the original 48 FIDs without the second f2 fast Fourier transform and fitted them with LC Model.
We created a general linear regression model to explore the relationship between T2 relaxation times and relevant parameters. Because T2 relaxation is determined partly by the intrinsic characteristics of each metabolite, we examined each metabolite in a separate statistical model. The regression parameters included diagnostic group, brain region, sex, and age. Based on a priori predictions, we added to the model an interaction term for diagnosis × brain region. Finally, we explored diagnostic differences in metabolite T2 relaxation times through post hoc t tests for BD vs NC and SZ vs NC in each voxel. Because percentage of GM within voxels may have an impact on metabolite T2 relaxation times, we repeated the same analysis using GM as a covariate.
Demographic and clinical variables were compared across groups using one-way analysis of variance and χ2 tests (Table 1). We computed a series of correlation coefficients between T2 relaxation times and chlorpromazine equivalents, Montgomery-Asberg Depression Rating Scale, Young Mania Rating Scale, and Positive and Negative Syndrome Scale scores (not corrected for multiple comparisons in order to detect any potential relationship).
We also examined medication effects on T2 relaxation times using one-way analyses of variance with medication as a between-subjects factor. Because there were different numbers of subjects on each medication in the SZ and BD groups, group differences in T2 measures could confound the medication effect. We therefore conducted this analysis separately for BD and SZ.
Both patient groups had shorter metabolite T2 relaxation times than NCs in every voxel (Table 2; Fig. 4). The BD group had significantly shorter T2 relaxation times than the NC group for tCr and tCho, and the shortening approached significance for tNAA (Table 3). When the analysis was repeated with voxel GM percentage as covariate, the same pattern of findings emerged, with one exception: the shortening of T2 relaxation time for tNAA became significant for the BD group compared with control (P = 0.017). Controlling for age, gender, and brain region, the β parameter estimates indicate that BD T2 relaxation times were about 21, 16, and 29 ms shorter for tNAA, tCr, and tCho, respectively. The SZ group also showed shorter than normal metabolite T2 relaxation times by about 12, 2, and 26 ms for tNAA, tCr, and tCho, respectively; only tCho reached statistical significance (Table 3). In voxel and diagnosis-wise post hoc t tests, both BD and SZ groups had significantly lower tCho T2 relaxation times than the NC group in the ACC (t = 2.72; P = 0.01 and t = 2.71; P = 0.01, respectively). The BD group also had a significantly lower T2 relaxation time for tCr in the ACC (t = 2.13; P = 0.04).
Table 2. Metabolite T2 Relaxation Times (Mean ± Standard Deviation in Milliseconds)
Normal control (N = 19)
Bipolar disorder (N = 14)
Schizophrenia (N = 14)
Significantly different from control group; P < 0.05. See text for details.
The BD group showed a more pronounced shortening of tCho T2 relaxation times in the ACC than POC as compared with NC or SZ groups (diagnosis × brain region interaction; P = 0.053) (Table 3). The SZ T2 relaxation time shortening was comparable in the ACC and POC (no significant diagnosis × brain region interaction).
In addition to diagnosis, the effect of brain region, gender, and age were explored in the linear regression models: T2 relaxation times were significantly shorter in the POC than ACC for all three metabolites; men had significantly shorter T2 relaxation times than women for tNAA; age was not significantly associated with T2 relaxation times (Table 3).
The effects of second-generation antipsychotic medications could not be assessed in subgroup analyses because all but one BD and SZ patient were on these medications (one patient was on a first-generation antipsychotic). There were no significant differences in any metabolite T2 relaxation times when BD patients taking lithium (N = 9) were compared with those not taking this medication (N = 6). Similarly, no differences emerged with anticonvulsant medications (not shown). Half of all patients in the study were taking benzodiazepines. These patients showed a trend toward longer T2 relaxation times (opposite to the diagnosis-associated trend) for all three metabolites as compared with those not taking benzodiazepines (P value range 0.064-0.089; between-group T2 relaxation time differences 12 ms or less) but none of these differences reached statistical significance.
T2 Relaxation Effect on Metabolite Concentrations
In order to determine whether the T2 relaxation time abnormalities we observed impact the measured metabolite concentrations, we extracted metabolite concentrations at TEs of 30 ms and 150 ms normalized to the magnitude of the water resonance. For this analysis, we focused on the difference between BD and NC groups in the ACC because T2 relaxation differences were largest and statistically significant there. The respective concentrations of tNAA, tCr, and tCho at 30 ms and 150 ms presented as millimolar concentrations, assuming a water concentration of 46 M (21), were 9.94/3.45, 10.35/2.99, and 2.64/0.85 for the NC group and 11.68/3.45, 12.10/2.81, and 3.08/0.88 for the BD group. Based on these numbers, the BD/NC ratio for tNAA, tCr, and tCho went from 1.18, 1.17, and 1.17 at short TE to 1.00, 0.94, and 1.03 at long TE, suggesting an impact of differential T2 decay on metabolite concentration measures.
We report that T2 relaxation times for intracellular metabolites are significantly shorter in patients with BD and SZ than in NCs in two areas of the cerebral cortex. This effect was strongest in the ACC in BD. Our findings suggest that noninvasive study of the cellular composition of brain parenchyma using T2 relaxation may provide useful biologic information about major psychiatric illnesses. T2 relaxation may thus become a valuable in vivo measure for monitoring the impact of disease progression, as well as treatment response on brain cells.
In BD and SZ, metabolite T2 relaxation times are shortened, and this may be due to increased interactions with intracellular macromolecules. This interpretation is consistent with the reported reductions in cell body size in the cerebral cortex in SZ and BD (3, 22). It is also possible that related cytoskeletal abnormalities in these disorders contribute by modifying the intracellular macromolecule or organelle composition (23). Previous reports of prolonged water T2 relaxation times in SZ can be explained in this context by the expansion of the extracellular compartment due to reductions in cell body size (9, 10). The distribution of T2 abnormalities in our study is consistent with postmortem evidence since some cellular abnormalities are limited to the ACC in BD (1), where T2 abnormalities were more pronounced, while abnormalities are reported throughout the cerebral cortex in SZ (24), where T2 abnormalities were not regionally selective.
Our findings are broadly consistent with previously reported metabolite T2 relaxation times in the healthy human brain, giving us confidence about the validity of our disease-specific findings. For example, the NCs in our study showed a pattern of T2 relaxation times very similar to what has previously been reported at 4.1 T (6). Consistent with the underlying physics, the metabolite T2 relaxation values obtained at 4 T in our study are shorter than those reported at 1.5 T (7). In addition, we found regional variability in metabolite T2 relaxation times (longer in ACC than in POC). Although no published studies compared metabolite T2 relaxation times in these cortical regions, metabolite T2 relaxation time differences have been reported between GM and white matter (5, 6) and the basal ganglia and other brain regions (25). The mean cell size and packing density differences between ACC and POC may explain the difference in T2 relaxation times in these regions. The gender effects on T2 relaxation in our study have not been studied previously; they suggest differences in the cellular make-up of the cerebral cortex with gender. We did not obtain evidence of biexponential decay in any metabolite, as has been previously reported (26), but our ability to resolve small deviations from monoexponential decay may be limited by the number of TEs used to fit the decay curve. Future explorations of biexponential relaxation will likely require more TE steps and increased signal averaging to reduce peak integral measurement uncertainty.
The major limitation of our study is that all patients were taking medication. Many kinds of medication can modify tissue T2 relaxation times, including diuretics (27) and omega-3 fatty acids (28). The effect of mood-stabilizing and antipsychotic medications on metabolite T2 relaxation times has not been studied, and we cannot rule out the possibility that our findings are secondary to effects shared by antipsychotic and mood stabilizing medications. Nonetheless, several arguments reduce this possibility: first, our findings are specific to diagnosis (BD > SZ) and brain region (ACC > POC in BD), although all patients were taking medications and the medications distribute throughout brain. This pattern could only be produced by a diagnosis- and brain region–specific response to medication, which seems unlikely. Second, although the comparisons are underpowered, subgroup analyses among our patients taking or not taking mood stabilizing medications (lithium or anticonvulsants) did not reveal significant drug effects. Interestingly, patients taking benzodiazepines showed a trend toward prolonged metabolite T2 relaxation times, but this could not explain the shortened T2 relaxation times in patient groups. Third, there was no relationship between chlorpromazine equivalents and metabolite T2 relaxation times. This suggests that the dose of antipsychotic medication did not have an effect on our measures. Fourth, the only study to examine a psychotropic medication reported that omega-3 fatty acids reduce water T2 relaxation times; although metabolite T2 relaxation times were not assessed in that study, they might be expected to change in the opposite direction. Since the mechanism of action of omega-3 fatty acids in BD is unclear, we cannot extrapolate any broader effects from this study. Ultimately, studies of patient groups before and after a well-controlled medication trial are needed.
In this study, we focused on metabolite quantification in an acutely ill patient population. The total time in the magnet was 75-90 min. We decided against collecting water-unsuppressed spectra to measure water T2 relaxation times in order not to prolong time in the magnet further. Measurements of water and metabolite T2 relaxation times in the same experiment are required to substantiate our interpretations and permit direct comparison of our results with previously published SZ studies. Lacking water spectra at each TE, we were not able to correct for eddy currents in each spectrum. Eddy currents change with TE, and this factor likely contributed variability to our data. In addition, quantification difficulties due to overlapping resonances (e.g., by Glu in the case of tNAA) provide another source of variability in our fitting procedure. We would expect any significant contamination of NAA measurements by underlying glutamate moieties with different T2 relaxation rates to show up as a biexponential (or multiexponential) contribution to the primary NAA exponential decay curve. Our attempts to identify biexponential contributions to any of NAA, Cr, or Cho decay curves were unsuccessful. These contributions are therefore small but nonetheless represent a source of uncertainty.
Another significant limitation of this study is that we did not start collecting data until 30 ms. Point-resolved spectroscopy sequence acquisitions with a TE shorter than 30 ms on our system were made difficult by hardware limitations and would require optimization of the sequence for this purpose. There may be substantial information in short TE spectra that we are missing, and collecting data at shorter TEs may reveal multiexponential decay in metabolite signal. In addition, T2 relaxation does not provide any detail on cell type (neurons vs glia), with the exception of NAA, which is enriched in neurons, or subcellular localization (soma vs dendrite/axons). This will await development of methods to measure T2 relaxation times of other metabolites (e.g., glutamate, GABA) with specific neurobiological roles and locations (29).
Another concern is possible differential T1 effects in our study groups. This would indeed lead to differential metabolite saturation levels. However, this possibility would not impact our T2 relaxation time measures. In this study, each individual serves as their own control for T1 differences, since T2 values were measured by varying TE at constant pulse repetition time. Therefore, T2 relaxation could be observed isolated from differential T1 saturation effects.
As demonstrated in our own metabolite concentration data, our findings may have implications for metabolite quantification in BD and SZ using proton MRS. Shortening of metabolite T2 relaxation times by 10% or more, as observed here, implies that metabolite signals may decay faster in some patient groups than in NCs. This may lead to artifactual reductions in patient metabolite levels, especially in longer TE studies where more differential signal decay takes place. A recent analysis of published proton MRS data in SZ suggested that this effect is a concern (30). In that analysis, the choice of TE used for an MRS acquisition (i.e., long vs short TE) was predictive of whether (or not) significant metabolite level differences were found between SZ patients and NCs. This issue is potentially of greater concern at higher magnetic field strengths (e.g., 4 T) where T2 relaxation times are short. In this setting, differential signal decay between patient and control groups may exist even at short TEs, leading to false-positive (or false-negative) findings. Since T2 relaxation effects are independent of concentration, they may either exaggerate or obscure actual metabolite concentration differences. In spite of the multiple limitations of our study, we provide evidence that quantification of metabolite T2 relaxation times is relevant and should be considered in future proton MRS studies in BD and SZ.
We thank Miriam Lundy and Caitlin Stork for technical assistance. This study was funded by grants 5R01MH058681-05 (Dr. Renshaw) and 1K23MH079982-01A1 (Dr. Öngür) from the National Institute of Mental Health, and the Clinical Investigator Training Fellowship from Harvard/M IT (Dr. Öngür).