Characterization of white matter alterations in phenylketonuria by magnetic resonance relaxometry and diffusion tensor imaging

Authors


Abstract

A multimodal MR study including relaxometry, diffusion tensor imaging (DTI), and MR spectroscopy was performed on patients with classical phenylketonuria (PKU) and matched controls, to improve our understanding of white matter (WM) lesions. Relaxometry yields information on myelin loss or malformation and may substantiate results from DTI attributed to myelin changes. Relaxometry was used to determine four brain compartments in normal-appearing brain tissue (NABT) and in lesions: water in myelin bilayers (myelin water, MW), water in gray matter (GM), water in WM, and water with long relaxation times (cerebrospinal fluid [CSF]-like signals). DTI yielded apparent diffusion coefficients (ADCs) and fractional anisotropies. MW and WM content were reduced in NABT and in lesions of PKU patients, while CSF-like signals were significantly increased. ADC values were reduced in PKU lesions, but also in the corpus callosum. Diffusion anisotropy was reduced in lesions because of a stronger decrease in the longitudinal than in the transverse diffusion. WM content and CSF-like components in lesions correlated with anisotropy and ADC. ADC values in lesions and in the corpus callosum correlated negatively with blood and brain phenylalanine (Phe) concentrations. Intramyelinic edema combined with vacuolization is a likely cause of the WM alterations. Correlations between diffusivity and Phe concentrations confirm vulnerability of WM to high Phe concentrations. Magn Reson Med 58:1145–1156, 2007. © 2007 Wiley-Liss, Inc.

Phenylketonuria (PKU; McKusick 261600) is the most frequent inherited disorder of amino acid metabolism (1, 2). A severe lack of phenylalanine (Phe) hydroxylase (Enzyme Commission 1.14.16.1) induces hyperphenylalaninemia, which causes disturbed brain development and impairments of myelin in untreated PKU patients. Early dietary treatment with restricted Phe intake is essential and blood Phe levels serve as indicator for the dietary strictness needed. Many PKU patients loosen or even stop the diet as adolescents. As dietary treatment of PKU was introduced in the early 1960s, the oldest treated patients have only reached midadulthood. Thus, the consequences of severely elevated Phe concentrations on brain structure and function are not yet clear. It was noted that single patients developed severe neurological symptoms years after ending dietary treatment (2).

Despite early treatment, MRI has demonstrated diffuse white matter (WM) hyperintensities on T2-weighted and fluid attenuation inversion-recovery (FLAIR) images (3–5). The origin of the partly reversible WM changes is not well understood since to date there is no histologic study in early-treated PKU patients. Histopathologic findings in WM specimens from untreated or late-treated PKU patients, who are usually characterized by severe neurologic deficits due to disturbed brain development, included hypomyelination, spongy changes, gliosis, and demyelination (Ref.6 and references therein). To gain insight into the nature of alterations in the microstructure of brain tissue in vivo, diffusion-weighted MRI and diffusion tensor imaging (DTI) studies have been performed. Diffusion-weighted MRI demonstrated reduced apparent diffusion coefficients (ADCs) in WM lesions (7–10). Initial conference abstracts of DTI studies are inconclusive, reporting the fractional anisotropy (FA) to be mostly normal in myelinated fibers in classical PKU (10, 11).

The WM abnormalities on T2-weighted MRI and diffusion changes have been discussed in relation to the finding of increased myelin turnover in an animal model of PKU (12) and the neuropathologic findings in untreated PKU (2, 8, 9). They were mostly attributed to some form of dysmyelination including defective (re)myelination and reversible WM edema with intramyelinic vacuole formation (6). Demyelination would be expected to increase ADCs instead of reducing them. Accordingly, demyelination in multiple sclerosis demonstrates increased diffusion coefficients (13) and increasing myelination in developing brain goes along with reduced diffusivity values (14).

The interpretation of ADC alterations with changes in myelination remained somewhat speculative since the myelin content is not assessed directly. An alternative and more direct approach to assess myelin integrity is through a detailed analysis of the different water compartments by MR relaxometry. Separate cerebral water compartments lead to multiple relaxation time components, mostly measured as T2-dispersion (15–17). Based on T2 relaxometry, the brain water signal can be separated into at least three contributions: the main component arising from intra- and extracellular water, a long T2 component from cerebrospinal fluid (CSF), and a contribution with short T2 and T1 from water trapped between myelin bilayers (myelin water, MW) (18). There is strong evidence that MW is a marker for myelin (19) and in vivo measurement of MW content can thus provide a quantitative measure of myelin loss or abnormal formation and maintenance of myelin. Relaxometry can therefore substantiate DTI findings that are attributed to changes in myelin. Imaging of MW has very recently demonstrated that PKU patients have reduced MW content in various WM structures (20).

A limitation of relaxometry based on T2 dispersion is that gray matter (GM) compartments (GMC) and WM compartments (WMC), i.e., the portion of WM that excludes MW, cannot be differentiated (16). Thus MW can be determined solely relative to GMC plus WMC or relative to the total signal including CSF, and no information can be obtained on changes of WMC and GMC. To segregate the latter, T1 information can be used. In this study, relaxometry data was recorded using an inversion-recovery (IR) multiple-echo sequence to characterize and map four brain compartments (including WMC/GMC differentiation): MW, WMC, GMC, and signals with long T2 (CSF-like signals, referred to as “free water” [FW]). The additional information comes at the expense of largely reduced spatial resolution, which leads to compartment mixing due to the pixel size and the point spread function. The low resolution was accepted because the aim of the method is mainly to characterize brain compartments in large lesions and in normal-appearing brain tissue (NABT), where one would average over large areas anyway, and also because the large pixel size assures a good signal-to-noise ratio (SNR), which is crucial for relaxometric analyses.

To obtain a better understanding of WM changes and lesions in PKU patients, which appear to be different from lesions in many leukodystrophies, we performed a combined study employing DTI, relaxometry, and MR spectroscopy (Ref.21 and references therein). The main objective was to test the hypothesis that reductions in ADC go along with changes in myelin content determined by relaxometry, possibly indicating dysmyelination with intramyelinic vacuole formation. The results from DTI and relaxometry were correlated and compared to Phe content in blood and brain. In particular, we tested whether ADC values and Phe content are correlated in PKU.

List of abbreviations

ADC = apparent diffusion coefficient, CSF = cerebrospinal fluid, CV = coefficient of variation, DTI = diffusion tensor imaging, FA = fractional anisotropy, FW = free water (CSF-like signals), GMC = gray matter compartment, MW = myelin water, NABT = normal appearing brain tissue, Phe = phenylalanine, PKU = phenylketonuria, ROI = region of interest, TI = inversion time, WM = white matter, WMC = white matter compartment

MATERIALS AND METHODS

Study Population

The study protocol was approved by the local ethical committee and written informed consent was obtained from all subjects.

The study population comprised nine early-treated male patients (mean age, 32.5 ± 2.5 years; age range, 27–35 years) with classical PKU. Patient characteristics are summarized in Table 1. DTI and MR spectroscopy were performed in all nine PKU subjects, with separate measurements performed two or three times in seven subjects. Relaxometry was performed in eight of the nine subjects with measurements repeated two or three times in five subjects. The first and last measurement for each subject was performed within 13 months (average, six months).

Table 1. Patient Characteristics
CodeAgeSexMutations, 1/2Severity of mutationaPKU, phenotypebDiet including, amino acid (AA) supplementDiet onset (days)IQ, WAIS-RMRI, gradecPlasma Phe (mM)Brain Phe (mmol/kgww)
  • a

    Severity according to Ref.21.

  • b

    Phenotype according to Ref.22.

  • c

    MRI-grade according to Ref.23.

  • WAIS-R = Wechsler Adult Intelligence Scale–Revised, m = male.

P131.9mIVS12nt1G>A, R261QSevere, moderateModerateOnly protein, restriction356461.5530.384
P233.1mR408W, IVS12nt1G>ASevere, severeSevereNo PKU diet3710591.9790.426
P327.3mIVS10nt11, L48SSevere, moderateMild to moderateNo PKU diet288361.5770.274
P432.1mR408W, not identifiedSevere, presumably severeSevereNo PKU diet, vegetarian2810551.5290.306
P534.9mT238P, R158QNot identified, severeSevereOnly protein restriction42105nd1.8100.407
P633.0mR261Q, G272XModerate, severeModeratePKU diet with AAs63118111.1820.325
P730.5mR408W, R158QSevere, severeSeverePKU diet with AAs2111571.4140.334
P833.9mY166X, R261QSevere, moderateModeratePKU diet, with AAs2412650.9230.206
P935.3mR408W, IVS12nt1Severe, severeSevereNo PKU diet319881.8800.345

The patients were compared to data from seven healthy male volunteers (mean age, 29.4 ± 6.6 years; age range, 23–39 years), without a history of neurological disease.

Laboratory Examinations

Laboratory examinations were performed as described previously (21): blood plasma Phe concentrations were determined with an enzymatic assay (Quantase Phe assay, Porton Cambridge, UK). Clinical evaluation (Table 1) included the identification of phenylalanine hydroxylase mutations. They were classified according to Ref.22. All patients were examined with a standardized neurological investigation and intelligence testing (Wechsler Adult Intelligence Scale–Revised). MRI-visible WM abnormalities were graded according to Ref.23.

MR Examinations

MRI and spectroscopy was performed on a 1.5T clinical scanner with a standard quadrature head coil (GE Signa version 5.8; General Electric Medical Systems, Milwaukee, WI, USA).

Relaxometry

An IR fast-spin-echo sequence was modified to yield one image per echo (32 images with individual echo times [TE] of multiples of 11 ms). The sequence included GE's product option of using adapted flip angles for the refocusing pulses to stabilize initial echo amplitudes and coaddition of spin echoes and stimulated echoes (the so-called tailored RF option: 90°, 173.9°, 158.3°, 160.1°, 160.1°, …). The sequence was applied with five different IR times (inversion time [TI] = 10, 260, 510, 760, 2010 ms), yielding 160 intensity values with different TE/TI combinations for each pixel. In order to keep imaging time short, only 16 phase encoding steps (one excitation each) were applied, resulting in an image matrix of 512 × 16. Imaging time for one slice was ∼5 min. Other parameters included slice thickness = 10 mm and field of view (FOV) = 180 × 180 mm2. The repetition time (TR) = 2500 ms + TI. The slice was acquired in transverse orientation at the level of the corpus callosum including the known T2-hyperintensities. Repeated scans were performed at the same slice position.

DTI

DTI was performed using a line scan diffusion imaging sequence (24) with the following parameters: for each section, six images with high b values (1000 s/mm2) in six noncollinear directions (relative gradient amplitudes: [Gx, Gy, Gz] = {[1,1,0], [0,1,1], [1,0,1], [–1,1,0], [0,–1,1], [1,0,–1]}) and two images with low b-value (5 s/mm2); FOV = 220 × 165 mm2; section thickness = 5 mm, spacing = 1 mm; TR/TE = 3264 ms/87 ms; image matrix size = 128 × 96; sections = 6.

Single-Voxel Spectroscopy

Brain Phe concentrations were determined by single-voxel spectroscopy as described earlier (25). Spectra were recorded from a large volume of 70 cm3 placed superior to the ventricular system using a point resolved spectroscopy (PRESS) sequence with outer volume suppression. Three spectra were recorded for each subject with TR/TE = 2020 ms/20 ms, averages = 256, data points = 1024, dwell time = 0.512 ms, water presaturation, and phase rotation. For quantification and determination of brain compartments, unsuppressed spectra with different TEs were acquired.

Processing

Relaxometry

The data were analyzed offline employing in-house software, written in IDL (Boulder, CO, USA). Processing included coregistration to T2-weighted images, Gaussian spatial apodization (weight equal to 0.6 at the diameter of a circle inscribed in the k-space square) and reduction of the matrix size to 128 × 16. The signal intensities (Stot) were fitted pixel-by-pixel, employing a modified two-dimensional (2D) Levenberg-Marquardt least-squares fitting routine (MPFIT, from a public-domain IDL library written by CB Markwardt, NASA/GSFC; http://cow.physics.wisc.edu/∼craigm/idl/idl.html) to a four-compartment model according to:

equation image(1)

with S0i being the signal contributions of the four components (MW, WMC, GMC, FW), n the number of echoes, i.e., 32 in our case, and TEmin the minimal TE. This model assumes complete longitudinal saturation after the repeated refocusing pulses.

In Vitro Calibration and Validation

The proposed method was validated in various phantom measurements:

  • 1To determine whether the sequence produces accurate T2 decay curves and to correct for signal variations (e.g., odd–even echo variation typical for Carr Purcell Meiboom Gill [CPMG] sequences) (26), a phantom containing a 0.12 mM aqueous MnCl2 solution was measured with the IR multiple spin-echo sequence. The deviation between the measured and monoexponentially fitted T1- and T2-maps was calculated. From this, correction factors (identical for all TI and <2% in size) were determined (Fig. 1) and later applied to the first three echoes in each TI series. Deviations for later echoes were neglected. Measured monoexponential relaxation times agreed well with expectations based on calculations using published relaxivity values of Mn.
  • 2The accuracy for determination of compartment sizes was tested using two equally-sized and -positioned water phantoms containing different concentrations of MnCl2 (0.12 mM and 0.41 mM, respectively, to match relaxation times of GM/WM and MW (27)), which were scanned separately with the IR multiple spin-echo sequence. The resulting data sets were added up in various proportions, thus mimicking two-compartment models with known contributions. Multicomponent fitting yielded almost perfect linear regression (slope 0.98, offset 1%) between the proportions of the combined data sets and the fitted S0 contributions (Fig. 2a).
  • 3Further validation measurements were performed on six phantoms with dairy cream of various milk fat percentages (3.5, 7, 11.5, 17, 25, and 35%). Dairy cream constitutes an ideal two-compartment phantom with relaxation times similar to WM (28). Biexponential fitting according to Eq. [1] was performed without any constraints. Linear regression between nominal fat percentage and the fitted fraction of the short T2-component yielded a very tight correlation (R = 0.998) with a slope of 0.86 and an offset of –0.6% (Fig. 2b). The deviation from unity for the slope is most likely due to the physical properties of the two component phantom. It could be due to differences in relaxation times between differently concentrated cream phantoms, or it could be due to the very small lipid droplets in our cream with possibly altered relaxation behavior at the surface of the droplet compared to the center.
Figure 1.

Phantom measurements for validation and calibration of the IR multiple spin-echo sequence. a: MR signal behavior for a single pixel from 160 images with different TI and TE times obtained from an aqueous MnCl2 solution with the IR multiple spin-echo sequence and corresponding monoexponential fits (vertical intensity stretching from –S0tot to +S0tot). b: Deviation of the fit from the measured data (relative to the intensity of the first echo), individually for the five TI times and averaged. Note the similar trend for the different TI times.

Figure 2.

Phantom measurements to determine the accuracy of the relaxometry method to estimate compartment sizes. a: Comparison of different nominal contributions of two combined phantom datasets (mimicking two-compartment models) with the compartment contributions obtained from multicomponent fitting (see Materials and Methods for details). An almost perfect linear regression between the proportions of the combined data sets and the fitted contributions was obtained. b: Comparison of nominal fat contributions of different dairy creams comprising various milk fat percentages (constituting two-compartment phantoms) with compartment contributions obtained from biexponential fitting without any constraints (see Materials and Methods for details). Linear regression between nominal fat percentage and the fitted fraction of the short T2-component yielded a very tight correlation.

In Vivo Application

To reliably separate the compartments, relaxation time parameters were constrained for MW and FW and fixed for WMC and GMC, based on values from numerous relaxometry studies at 1.5 T and our own work (18). In detail: MW: T1 = 300–400 ms, T2 = 5–35 ms; FW: T1 > 1500 ms, T2 = 500–5000 ms; WMC: T1 = 650 ms, T2 = 100 ms, GMC: T1 = 1028 ms, T2 = 97 ms. As an example for the in vivo data, the intensity of one pixel in dependence of TI and TE, as well as its fitting residue are shown in Fig. 3. The robustness and reproducibility of the method had been established previously with 22 measurements on seven volunteers and four PKU patients (18), in whom repeated scans showed a small variability of 9% for the MW compartment size (14.2 ± 1.2% relative to the WMC component).

Figure 3.

MR signal behavior of a single pixel obtained in vivo with the IR multiple spin-echo sequence. The surface plots illustrate the 160 intensity values (5 TI × 32 TE times) for one pixel from WM. Left: measured data; Right: Fit residue. Note that the scale is extended by a factor of 10.

Maximal attainable precision for the estimated compartment sizes was calculated in the form of Cramer Rao minimal variance bounds using typically achieved SNR and typically found relaxation times for MW and FW, while respecting the constraints for T1 and T2 of WMC and GMC (see Results).

Tissue Selection

From the fitted in vivo data, relative compartment contributions were calculated for NABT and lesions.

NABT

Based on the original IR multiple spin-echo and T2-weighted images, masks were created to exclude from analysis the pixels in the skull, the ventricles, and the hyperintense lesions. For the remaining pixels, compartment contributions in percent were calculated relative to the total signal (MWsum, WMCsum, GMCsum, and FWsum) and relative to WMC (MWWMC). Because of the known colocalization of MW and WMC only pixels with predominantly WM (SWMC > 0.8 × Stot) were analyzed for MWWMC and MWsum to reduce errors in areas with low WMC. Average values were calculated from all pixels in NABT.

Lesions

Variably-shaped regions of interest (ROIs) were placed on coregistered T2-weighted images within hyperintense areas adjacent to the right and left posterior horns of the lateral ventricles (Fig. 4). For control subjects, the ROIs were placed at equivalent positions. Similar to NABT, compartment contributions were calculated relative to the total signal, however, no constraint for MWWMC and MWsum was employed. Mean values were determined for these ROIs, and values from right and left ROIs were averaged.

Figure 4.

Exemplary results for tissue compartmentation from relaxometry in a PKU and a control subject. T2-weighted transverse MR images of a healthy volunteer (a,b) and a PKU patient (g,h). The PKU subject demonstrates the known hyperintensities in periventricular WM. ROIs were placed within the lesions for PKU and similar regions in controls as indicated on (b) and (h). Images of the four compartments for the control and PKU subject are presented: (c,i) MW, (d,j) WMC, (e,k) GMC, and (f,l) FW. The images are scaled to maximum intensity in each image.

DTI

The DTI measurements were processed using “Xphase” (S. Maier, Boston, MA, USA, unpublished software). ADC and FA values were calculated on a pixel-by-pixel basis after interpolation to a matrix size of 256 × 256. Furthermore, the eigenvalues L1–L3 of the diffusion tensor were determined. In strictly parallel fibers, L1 corresponds to diffusivity in the direction of the axons, while L2 and L3 correspond to diffusivity perpendicular to the main axis.

ROIs were placed on coregistered T2-weighted images within hyperintense areas adjacent to the right and left posterior horns of the lateral ventricles (Fig. 5). In addition, ROIs were placed in the splenium of the corpus callosum (Fig. 5) because of dominantly unidirectional fiber structures. Mean diffusion parameters from right and left ROIs were merged, yielding one value per subject.

Figure 5.

Exemplary results for DTI scans in a PKU patient. a:T2-weighted transverse MR image of a PKU patient. The corresponding ADC and FA maps are shown in (b) and (c). The ADC map features hypointense areas colocalizing with T2 hyperintensities. Visibly, anomalies are less obvious in the FA map since FA values are always lower in these particular areas because of fiber crossings. ROIs are overlaid and were placed in areas adjacent to the right and left posterior horns of the lateral ventricles. In addition, ROIs were placed in the splenium of the corpus callosum.

Single-Voxel Spectroscopy

The data processing and quantification scheme has been described in detail earlier (25). It is based on a three-compartment model (brain tissue, FW, and blood). The determination of the Phe peak area was performed using an iterative nonlinear least squares fitting algorithm (29), referenced to an aqueous 25 mmol/liter Phe model solution (pH 7.05). Conversion to absolute concentration units was based on a proton density taken from the literature and on the unsuppressed water signal from the brain tissue compartment.

Statistical Analysis

Frequency histograms of the relative compartment sizes in NABT were created from the relaxometry measurements; i.e., of MWWMC, MWsum, WMCsum, GMCsum, FWsum, and of (WMC + GMC)/sum. First, histograms were created by individually calculating from all pixels the relative frequency of compartments in each bin. After averaging for repeated measurements, the individual histograms were averaged over all subjects. Thus, each subject is equally weighted, independent of the number of pixels in NABT.

For testing of reproducibility, residual coefficients of variation between subjects (CVb) and within subjects (CVw) were calculated as the square root of the residual mean square and are given in percent of the mean value (30).

For comparisons of pairs of groups, two-tailed unpaired t-tests were performed if the value distribution was approximately Gaussian. Otherwise (for WMCsum and GMCsum), nonparametric Mann-Whitney U tests were performed. A P-value of less than 0.05 with Bonferroni correction for multiple comparisons was considered to indicate a statistically significant difference for comparisons of relaxometry and DTI results between patients and controls.

Pearson's linear regression analysis was used to correlate results from relaxometry and DTI and to compare both with blood and brain Phe.

Statistical analysis was performed using SPSS 12.0 (SPSS Inc, Chicago, IL, USA) and Excel 2002 (Microsoft Corp., Redmond, WA, USA).

RESULTS

T2-weighted images demonstrated the typical hyperintense areas adjacent to the frontal and posterior horns of the lateral ventricles in all PKU subjects.

Relaxometry

Compartment Images

Distinct compartment images of WMC, GMC, and FW were obtained for PKU and control subjects (Fig. 4). For controls, the images of the myelin component show intense signals only in areas with high WMC content. The MW fraction in GM is not higher in PKU patients; it only seems so because of image scaling to maximum intensity and reduced MW content in WM of patients.

Reproducibility of Relaxometry

All compartment estimations in NABT showed good reproducibility within subjects for the PKU patients, with a within-subjects variation of less than 5% for MW, WMC, and GMC and a slightly higher CVw for FW (Table 2; Fig. 6). In contrast, between subjects, the variation was much greater than CVw for all relaxometry parameters. Reproducibility in lesions was lower with CVw of 3% to 14%. However, the variability within subjects was still much lower than between subjects (Table 2; Fig. 6). Cramer Rao minimal variance bounds of compartment amplitudes, which are independent of relative compartment sizes, were found to be 0.9, 1.8, 2.5, and 1.0 for MWsum, WMCsum, GMCsum, and FWsum, respectively.

Table 2. Reproducibility of Results From Relaxometry and DTI*
5 subjects/13 scansRelaxometry
LesionsNABT
MWWMCMWsumWMCsumGMCsumFWsumMWWMCMWsumWMCsumGMCsumFWsum
  • *

    Coefficients of variation within subjects (CVw) and between subjects (CVb) for the four compartments: myelin water (MW), white matter (WMC), gray matter (GMC), and “free water” (FW) determined in lesions and in normal-appearing brain tissue (NABT) from relaxometry and for apparent diffusion coefficient (ADC), fractional anisotropy (FA), and the three eigenvalues (L1, L2, L3) from diffusion tensor imaging (DTI).

CVw (%)13.19.33.014.210.54.03.74.74.48.5
CVb (%)26.228.423.9100.036.229.629.215.913.519.8
R0.700.890.980.980.950.990.990.930.890.84
7 subjects/17 scansDTI
LesionsCorpus callosum
ADCFAL1L2L3ADCFAL1L2L3
CVw (%)6.15.86.05.78.42.03.12.37.517.5
CVb (%)12.36.711.612.115.57.33.46.612.318.9
R0.850.650.850.820.780.940.120.880.570.21
Figure 6.

Reproducibility of compartmentation results from relaxometry illustrated by correlation plots. Correlation between the first and second measurement of MWWMC and the four water compartments relative to the sum (MWsum, WMCsum, GMCsum, FWsum) for the PKU patients, measured two or three times in NABT (a) and in lesions (b). Most values are close to the identity line.

Histograms

The compartment histograms of NABT demonstrated strong differences between PKU and control subjects for all relative compartment sizes, except GMCsum and (WMC + GMC)/sum (Fig. 7). For MWWMC and MWsum, a shift of a relatively narrow Gaussian-like distribution toward lower values was observed. While for (WMC + GMC)/sum only a slight decrease and broadening toward lower values was obtained, the WMCsum histogram is strikingly different for patients and controls. For controls, it features a relatively homogeneous distribution with a sharp Gaussian-like peak at approx. 0.82. This peak is completely absent in PKU, while the remaining WMCsum distribution appears unchanged. For FWsum a rather sharp peak at ∼0.04 in controls is considerably broadened and shifted to higher values in patients. Because of the low image resolution in one dimension (matrix: 512 × 16), which leads to signal blurring in addition to within-voxel compartment mixing, no pixels with pure WMC or GMC were observed.

Figure 7.

Overall changes in tissue compartmentation illustrated by plots of relative frequency of occurrence for individual brain components. Frequency histograms of the relative water compartment sizes (MWWMC, MWsum, WMCsum, GMCsum, FWsum) and (WMC + GMC)/sum in NABT of PKU and control subjects from the relaxometry measurements. (The vertical scales for GMCsum and WMCsum are confined to the range <2.0% and 2.5%, respectively, for better visibility of detail, resulting in an intensity cutoff for the lowest bin.)

Comparisons of Mean Values Between PKU and Control Subjects

Overall, a content of 13.7 ± 1.9% MWsum was found in NABT for control subjects (Table 3), while a significantly lower value of 9.0 ± 2.5% was determined (P < 0.02) for PKU patients. Similarly, MWsum was reduced in lesions of PKU subjects compared to controls (6.6 ± 1.6% vs. 10.9 ± 2.7%, P < 0.05). MWWMC in NABT was significantly lower for PKU subjects than for controls, while in lesions the difference did not reach statistical significance. Furthermore, the FW compartment was strongly increased (P < 0.002) in PKU, and the WMC tend toward lower values (P < 0.15) in both NABT and lesions.

Table 3. Mean Results From Relaxometry and DTI for PKU Patients and Control Subjects Grouped According to Brain Region
 Relaxometry
LesionsNABT
MWWMCMWsumWMCsumGMCsumFWsumMWWMCMWsumWMCsumGMCsumFWsum
  • Mean values (±1SD) for MWWMC, MWsum, WMCsum, GMCsum, and FWsum from relaxometry and for ADC, FA and the eigenvalues, L1–L3 from DTI. See Table 2 for abbreviations.

  • a

    P-values determined from a Mann-Whitney U test. Otherwise, P-values were obtained from two-tailed unpaired t-tests. P-values include Bonferroni corrections for multiple comparisons.

  • CVw = coefficients of variation within subjects, CVb = coefficients of variation between subjects, MW = myelin water, L1, L2, L3 = three eigenvalues, ns, not significant.

PKU12.1 ± 2.36.6 ± 1.657.5 ± 14.411.2 ± 8.924.8 ± 8.312.1 ± 3.59.0 ± 2.536.0 ± 5.844.9 ± 5.713.3 ± 2.1
Control14.7 ± 3.210.9 ± 2.780.3 ± 4.74.0 ± 5.03.3 ± 1.717.6 ± 2.613.7 ± 1.944.7 ± 2.639.3 ± 3.17.5 ± 0.9
P-valuesns<0.05nsansa<0.002<0.05<0.02<0.1ansa<0.0005
 DTI
LesionsCorpus Callosum
ADCFAL1L2L3ADCFAL1L2L3
PKU591 ± 630.34 ± 0.02804 ± 83564 ± 60404 ± 50650 ± 690.79 ± 0.051425 ± 173345 ± 49181 ± 37
Control762 ± 270.39 ± 0.041099 ± 42714 ± 38474 ± 35754 ± 300.79 ± 0.031658 ± 110408 ± 31198 ± 31
P-values<0.0001<0.04<0.0001<0.0004<0.06<0.01ns<0.03<0.04ns

DTI

Parameter Images

ADC images showed hypointensities in lesions colocalizing with hyperintense areas in T2-weighted images (Fig. 5). Visual inspection of FA images did not demonstrate obvious abnormal intensities in the lesions.

Reproducibility of DTI Measurements

CVw of ADC, FA, and the eigenvalues L1 and L2 were low (CVw < 7.5%) in lesions and also in the corpus callosum of PKU patients (Table 2), demonstrating good reproducibility. In contrast, CVb were larger than the corresponding CVw, especially for ADC, L1, and L2 in both lesions and the corpus callosum.

Comparisons of Mean Values Between PKU and Control Subjects

Highly significantly lower ADCs were found in lesions of PKU patients compared to matched positions in controls (P < 0.0001; Table 3). In addition, FA values were also lower in lesions of PKU subjects (P < 0.04), which was primarily due to strongly reduced L1 values (27%). L2 was significantly reduced by 21% and L3 by 15%, but both were less pronounced than L1. In the corpus callosum, ADCs and the eigenvalues L1 and L2 were significantly lower in PKU, while FA was comparable (Table 3).

Single-Voxel Spectroscopy

Brain Phe concentrations (Table 1) in PKU patients were much above the normal concentration of 0.050 mmol/kg, similar to previously reported results (21). Brain Phe was significantly and linearly correlated with blood Phe (R = 0.78). Because of the small sample size no correlations with the clinical parameters are presented.

Comparisons Between Results From Relaxometry and DTI

Relative MW, WMC, and FW compartments were compared with ADCs and FAs, both determined in lesions at similar positions. No significant correlation was observed between MW content and ADCs or FAs in PKU subjects. However, lower FAs were accompanied by lower WMC content (R = 0.71, P < 0.05; Fig. 8a) and by higher FW (R = –0.78, P < 0.03; Fig. 8b). Figure 8 also shows that the control subjects had higher FA values, higher WMC, and lower FW content than all but one PKU subject. Similar trends were detected in lesions for a comparison of ADC values with WMC and with FW content (Fig. 8c and d), however without reaching significance.

Figure 8.

Interdependence of results from compartmentation and diffusion analysis. Correlation plot between parameters from relaxometry and DTI for PKU and control subjects: (a,b) FA vs. WMCsum and vs. FWsum; (c,d) ADC vs. WMCsum and vs. FWsum. For the regression lines only values from PKU subjects were used. Lower FA and ADC values were accompanied by lower WMCsum and by increased FWsum in PKU subjects. Control values are clearly separated from PKU data.

Correlations of Results From Relaxometry and DTI With Phe Concentrations

No significant correlations between parameters from relaxometry and blood or brain Phe concentrations were observed in PKU subjects. In contrast, ADC values in lesions were significantly negatively correlated with blood Phe and, as a trend, with brain Phe (R = –0.78, P = 0.01 and R = –0.61, P = 0.08, respectively; Fig. 9a). Similarly, lower ADC values in the corpus callosum were accompanied nonsignificantly by higher blood Phe and, in a significant manner, by higher brain Phe (R = –0.39, P = not significant [ns], R = –0.75, P = 0.02, respectively; Fig. 9b).

Figure 9.

Interdependence of ADC values and Phe concentrations. Correlation of Phe in blood and brain with (a) ADCs determined in lesional areas, and (b) with ADCs from corpus callosum (cc). ADC values in lesions and in the corpus callosum decrease with increasing Phe concentrations, both in blood and brain.

DISCUSSION AND CONCLUSIONS

A combined DTI and relaxometry study was performed on a homogeneous group of early-treated adults with classical PKU, together with determination of brain Phe concentrations by MR spectroscopy. Brain water compartmentation into gray and WM components, a CSF-like compartment, and MW was determined by relaxometry in NABT and in lesions while diffusivity and diffusion anisotropy was measured by DTI.

The major findings of this study are: MW content was reduced in lesions, but also in NABT of PKU patients, while the FW contribution was significantly increased. Lower WMC and higher FW contents in lesions of PKU subjects were accompanied by reduced FA and ADC. Decreased ADC values were also found in the corpus callosum, which hardly appeared to be affected on T2 images. The FA reduction in lesions was due to strongly decreased longitudinal diffusion (L1) and less pronounced decreased transverse diffusion (L2, L3). ADC values in lesions and also in the corpus callosum correlated significantly with blood and brain Phe concentrations.

Determination of diffusion parameters appeared robust and highly reproducible. Similarly, fitting of the simultaneously acquired TE and TI data for relaxometry yielded consistent brain components. Intrasubject variations were low and the novel method appears suitable to characterize focal or diffuse cerebral abnormalities. Although only five PKU patients were scanned two or three times, which is marginal for determination of reproducibility, the results confirm our previous study with repeated measurements in six healthy volunteers (18). Reproducibility of our IR relaxometry method seems to be somewhat better than what was found for MW determination based solely on T2 information, for which Vavasour et al. (31) reported 19% for the within subject variability. The intrinsic uncertainties in estimating compartment sizes with the current methodology, expressed as Cramer Rao minimum variance bounds, are much lower than the intra- and interindividual variations found. However, these bounds are calculated under the premise of a correct model and true parameters, conditions that are certainly not strictly fulfilled, given the intensity variations as a consequence of nonideal echo trains, and considering the strict enforcement of some boundaries in parameter space.

Fitting of 160 images for each case with 12 unconstrained model parameters may lead to minima that are physiologically unrealistic. The allowed range of the parameters was therefore severely restrained by enforced parameter relations and range restrictions to avoid physically impossible results. On the other hand, it is possible that the absolute size of components or relaxation times are influenced systematically by the enforced constraints. In particular, in cases where the true relaxation times of any compartments are changed in pathology, fitting with incorrectly fixed or constrained relaxation time parameters may lead to compensatory changes in compartment sizes. Thus, our finding of altered compartment sizes in PKU might—in principle—be due to changes in relaxation times instead. However, this is unlikely since previous studies did not demonstrate relevant T2 changes for individual compartments in PKU. Furthermore, the observed trend for changes in T2 (20) is actually expected to lead to an apparent increase, not a decrease in the MW component. Thus, the extent of the observed decrease in MW might be underestimated rather than overestimated.

As a simplification, the applied model neglects exchange and cross-relaxation between MW and WMC pools. Despite some earlier work that found multiple T1 components in WM (32, 33) and clear evidence of multicomponent T1-relaxation in peripheral nerve (34, 35), several recent studies have suggested that T1-relaxation is monoexponential in myelinated WM at body temperature, indicating relatively fast exchange on the T1 time scale between MW and WMC, while the exchange is sufficiently slow compared to T2 for multiexponential T2 modeling (36–38). However, in case of compartment mixing of GMC, WMC, and FW within a pixel, T1 relaxation is multiexponential and the applied model with parameter restrictions allows the separation of WM and GM compartments, which is impossible based on TE variation alone. Modeling of the impact of MW content on T1 relaxation times of WM is difficult due to effects from geometry, membrane permeability, or local viscosity (36, 39). Furthermore, it is quite possible that exchange rates are different in lesions. Fitting with different restrictions for T1 of MW demonstrated that allowing separate T1s for MW and WMC has only a relatively small effect on the determined size of the MW component, while WMC and GMC contents are more sensitive.

The spatial distribution of MW was found to match the WMC distribution, and the MW content averaged 13.7% ± 1.9% in control subjects. Both results agree well with the literature (15–17, 32) and support the validity of the method. Unlike previous measurements, the applied method yields additional information on the other compartments, i.e., WMC, GMC, and FW—though at the expense of a largely reduced spatial resolution.

The MW content was found to be reduced in lesions and also in NABT, corroborating a very recent work (20). This reduction of MW is compatible with a diffuse loss of myelin as recently described for multiple sclerosis (17) or it could be due to dysmyelination. Reduced MWWMC demonstrated that the reduction was stronger for MW than for the colocalizing WMC fraction, especially in NABT.

Histogram analysis primarily revealed a shift of peaks for MW and FW, while for WMC the dominant and relatively narrow peak with high WMC content, present in controls, was absent in PKU subjects, and the values shifted nonspecifically toward a wide range of lower values. As discussed above, the finding of an apparent strong decrease in WMC content might in part reflect changes in T1 rather than “vanishing white matter.” Altered WMC relaxation could well be a consequence of the reduction in MW content because the size of the MW compartment may—through exchange and cross-relaxation between MW and WMC pools—codetermine T1 of WMC. Because T1 of WMC was fixed, the compartment might then have been fitted as a mixture of GMC (with inherently longer T1) and WMC. The detected reduction in WMC content would therefore also hint at changes in myelin structure. However, T1-weighted images showed only minor changes in PKU subjects in previous studies (23), indicating that T1 changes probably, at most, account for parts of WMC changes in this study.

The increase in FW compartments was expected, replicating previous results (3) and accounting for the well-known hyperintensities in T2-weighted images (7, 23). The higher proportion of FW indicates an increased pool of extracellular water with relatively low protein content, but otherwise unknown properties and location. As described below, liquid in intramyelinic vacuoles is a potential candidate.

Reduced ADC values in PKU lesions confirm previous findings (7–10). In contrast, in our study, FA was found to be reduced in lesions, which was due to less strongly reduced diffusion in the radial than the longitudinal direction. It should be noted that the lesions lie in areas of crossing fibers and that systematic changes in fiber tract organization could also lead to the observed reductions in FA in these regions. Our findings differ from the results of the few DTI studies in classical PKU published in recent abstracts, where mostly unchanged FA was found for myelinic fibers (10, 11). Conversely, in malignant PKU, reduced diffusion anisotropy was demonstrated (40), but malignant PKU should not be compared with classical PKU. It is known to feature other MRI abnormalities and the cause for the decrease in anisotropy is different: increased radial (with unchanged parallel) diffusivity in malignant PKU compared to strongly reduced parallel (with slightly reduced radial) diffusivity in classical PKU. This argues for clear structural differences in the two PKU variants, supported by the fact that none of the patients with malignant PKU had decreased ADCs. Reduced FAs were accompanied by lower WMC and higher FW content in lesions of our PKU subjects. Values from control subjects corresponded to these relations. Unlike most WM diseases, like multiple sclerosis, amyotrophic lateral sclerosis, or Alzheimer's disease, which feature increased ADC values (41), decreased ADC values of WM lesions have also been observed in Canavans disease, metachromatic leukodystrophy, maple syrup urine disease, and Creutzfeldt-Jakob disease (41, 42). This has been interpreted as cytotoxic edema due to reduction of the Na+K+-ATPase activity in PKU (9) and similarly in maple syrup urine disease (43), or alternatively as dysmyelination with intramyelinic edema in PKU (7, 8) or again in maple syrup urine disease (44). However, our additional finding of reduced myelin water along with the reduced ADC values suggests that both might be associated with status spongiosis (8) or with the swelling/separation of myelin sheaths, helping the formation of vacuoles and producing potentially reversible intramyelinic edema (42, 45). This would affect the bound water trapped in myelin sheaths and possibly increase the amount of “free” water within the sheaths and vacuoles, which may well feature long relaxation times and restricted diffusivity at the same time. This interpretation is supported by a recent abstract that characterized the additional water reservoir in PKU patients with an intermediate T2 that was clearly longer than that of tissue water in GMC and WMC, but shorter than for CSF (46).

The decrease in perpendicular diffusivity may be due to reduced interstitial space because of separation of myelin sheaths and formation of vacuoles. Similarly, reduction of longitudinal diffusion may also be due to vacuolization, but possibly also irregular delamination of myelin sheaths around the fiber periphery (47), thus also increasing longitudinal barriers. The effect of vacuolization can be modeled trivially as a superposition of WM tissue with properties found in control subjects plus an additional component with long T2, small ADC, and isotropic diffusion. The observed diffusion effects in PKU might be explained solely by additional signal contributions from the vacuoles that are of different strength for lesions and NABT. In contrast, demyelination would be expected to increase radial diffusivity (48).

Previously, the corpus callosum has been described to be relatively spared in PKU (6). On the other hand, abnormalities of interhemispheric connections have been suggested in PKU (49) and the corpus callosum was found to be 10% smaller compared to controls (50). Our results of reduced ADC values with strongly decreased longitudinal and transverse diffusion support an involvement of the corpus callosum.

In contrast to a previous conference report in which no correlation between the degree of WM involvement and brain Phe concentration had been found (10), reduced ADC values in lesions and also in the corpus callosum were accompanied by higher blood and brain Phe concentrations in our study. It should be noted that although concurrent Phe concentrations were used to detect correlations with structural properties, this is not necessarily an indication that concurrent concentrations are indeed involved. Rather, since measurements were averaged over multiple time points and single Phe concentrations usually relate to the individual long-term concentrations, it is expected that structural adaptations are a consequence of long-term increases of Phe levels prior to the MR investigations. Similarly, Kono et al. (9) describe a correlation between ADC values in posterior cerebral WM and blood Phe. All these results confirm the hypothesis that Phe levels have a deleterious effect on brain compartments and tissue makeup.

In conclusion, we detected reduced MW along with reduced diffusivity in WM lesions and also in NABT of PKU patients, suggestive of intramyelinic edema—likely in the form of vacuolization. Correlations between diffusivity and blood and brain Phe concentrations indicate vulnerability of WM structures to high Phe concentrations.

Acknowledgements

We thank Dr. Stephan Maier from B&W Hospital, Boston, MA, USA for providing and supporting the line scan DTI sequence and corresponding evaluation software.

Ancillary