Abstract
BACKGROUND AND PURPOSE: White matter lesions are commonly found in patients with Fabry disease. Existing studies have shown elevated diffusivity in healthy-appearing brain regions that are commonly associated with white matter lesions, suggesting that DWI could help detect white matter lesions at an earlier stage This study explores whether diffusivity changes precede white matter lesion formation in a cohort of patients with Fabry disease undergoing yearly MR imaging examinations during a 5-year period.
MATERIALS AND METHODS: T1-weighted anatomic, FLAIR, and DWI scans of 48 patients with Fabry disease (23 women; median age, 44 years; range, 15–69 years) were retrospectively included. White matter lesions and tissue probability maps were segmented and, together with ADC maps, were transformed into standard space. ADC values were determined within lesions before and after detection on FLAIR images and compared with normal-appearing white matter ADC. By means of linear mixed-effects modeling, changes in ADC and ΔADC (relative to normal-appearing white matter) across time were investigated.
RESULTS: ADC was significantly higher within white matter lesions compared with normal-appearing white matter (P < .01), even before detection on FLAIR images. ADC and ΔADC were significantly affected by sex, showing higher values in men (60.1 [95% CI, 23.8–96.3] ×10–6mm2/s and 35.1 [95% CI, 6.0–64.2] ×10–6mm2/s), respectively. ΔADC increased faster in men compared with women (0.99 [95% CI, 0.27–1.71] ×10–6mm2/s/month). ΔADC increased with time even when only considering data from before detection (0.57 [95% CI, 0.01–1.14] ×10–6mm2/s/month).
CONCLUSIONS: Our results indicate that in Fabry disease, changes in diffusion precede the formation of white matter lesions and that microstructural changes progress faster in men compared with women. These findings suggest that DWI may be of predictive value for white matter lesion formation in Fabry disease.
ABBREVIATIONS:
- ERT
- enzyme replacement therapy
- IQR
- interquartile range
- LME
- linear mixed effect
- MNI
- Montreal Neurological Institute
- NAWM
- normal-appearing white matter
- TP
- time point
- WML
- white matter lesion
Fabry disease is a rare, X-linked lysosomal storage disorder caused by a mutation in the galactosidase alpha (GLA) gene.1 This mutation causes a deficiency of α-galactosidase A activity, resulting in the accumulation of glycosphingolipids, mainly globotriaosylceramide, primarily in the cardiovascular system, brain, and kidneys.2 Accumulation in the arterial walls is believed to contribute to cerebrovascular and cardiac diseases, which are among the leading causes of death in patients with Fabry disease. Among other neurologic findings, white matter lesions (WMLs) are found in nearly one-half of all patients.3⇓-5 WML burden increases with age and is associated with cognitive decline.5,6 Although the prevalence, severity, and progression of WMLs are comparable in men and women, men develop WMLs at a younger age.4,7 Moreover, a higher prevalence of WMLs in men with classic compared with nonclassic Fabry disease has been reported.8
The pathophysiology of WMLs in Fabry disease is complex and incompletely understood. Cell damage in the cerebrovascular system may impair the autoregulation of cerebral perfusion, resulting in a hyperdynamic circulation and endothelial dysfunction caused by shear stress and incompliant vessel walls.4,9,10 Combined with glycosphingolipid storage in endothelial cells, these effects might cause the release of reactive oxygen species, further compromising endothelial cells and accelerating vascular dysfunction.11,12 Patients with Fabry disease with and without WMLs have shown increased CBF but decreased glucose metabolism in the WM, indicating microangiopathy.13 Last, increased CSF protein levels have been reported, indicating blood-brain barrier dysfunction.14
MR imaging is the criterion standard imaging technique to detect brain alterations in Fabry disease, including WMLs. DWI can accurately quantify microstructural WM changes and has been used in patients with Fabry disease.5,15,16 These studies showed elevated mean diffusivity in healthy-appearing brain regions that are commonly associated with WMLs.5,15,16 Moreover, WM mean diffusivity was found to be correlated with plasma globotriaosylceramide levels.5 These findings suggest that DWI could help to detect WML formation at an earlier stage and potentially provide a tool to evaluate treatment options like enzyme replacement therapy (ERT). However, these studies were of a cross-sectional design, prohibiting the assessment of longitudinal changes required to investigate whether changes in diffusion precede WML formation. Therefore, this study aimed to explore whether DWI can detect microstructural changes before WMLs appear on FLAIR-weighted images in a cohort of patients with Fabry disease undergoing yearly MR imaging examinations during a 5-year period.
MATERIALS AND METHODS
For this retrospective, observational study, we adhered to the guidelines of the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) statement. Data collection has been described before.17 The Amsterdam University Medical Centers (location in the Academic Medical Center) are the national referral center for patients with Fabry disease in the Netherlands. Follow-up at the outpatient clinic ranged from 6 months to once every 2 years and included brain MR imaging scans. Follow-up data from patients with a confirmed Fabry disease diagnosis were collected in a local database after patients provided written informed consent. From this database, follow-up data collected between 2013 and 2019 were extracted. All patients were classified as having the classic or nonclassic phenotype using predefined criteria.8,18,19 Patients were included if good-quality FLAIR-weighted, T1-weighted, and DWI scans from at least 3 visits were available. According to Dutch law, no approval of the study protocol was needed because this was a retrospective study and patients were not subjected to procedures or rules of behavior in addition to regular clinical follow-up.
Image Acquisition
MR imaging scans were acquired on 3T scanners (Ingenia; Philips Healthcare). T1-weighted anatomic scans were acquired with the following parameters: TR/TE: 9/4.1 ms, flip angle: 8°, resolution: 0.5 × 0.5 × 0.9 mm3, FOV: 256 × 256 × 170 mm3, and number of averages: 1. DWI scans were acquired with 2 b-values, 0 and 1000 s/mm2. Other acquisition parameters were the following: TR/TE: 6340/83 ms, flip angle: 90°, resolution: 0.9 × 0.9 × 3.0 mm3, FOV: 229 × 229 ×147 mm3, and number of averages: 2. FLAIR acquisition parameters were as follows: TR/TE: 4800/356 ms, flip angle: 90°, resolution: 1.04 × 1.04 × 1.12 mm3, FOV: 250 × 250 × 360 mm3, and number of averages: 2. MR imaging software upgrades introduced some variation in acquisition parameters of the DWI scan (Online Supplemental Data). Periodic image-quality tests using a vendor-provided phantom were performed during the course of the study, including before and after software upgrades. Any observed abnormalities were addressed by the vendor before continuation. Patient positioning and planning of the scans were performed according to standard clinical procedures. FLAIR scans were assessed by a neuroradiologist for WMLs according to the Fazekas scale. The Fazekas scale rates WMLs in the periventricular and deep brain separately from zero (no WMLs) to 3 (severe confluent WMLs) and adds both scores, resulting in a total score that can range from zero to 6.
Postprocessing
Postprocessing was performed in Matlab (Version 2019b; MathWorks). An overview of the processing steps is schematically shown in Fig 1.
Overview of postprocessing steps. T1-weighted scans were segmented in native space and registered to MNI space (A). WMLs were automatically segmented on FLAIR images, thresholded, and manually corrected (B). ADC maps were fitted from DWIs with b-value = 0 and 1000 s/mm2 (C). WML masks and ADC maps were registered to MNI space using the transformation matrix from the T1 registration. NAWM masks were created by subtracting the WML masks from the WM segmentation (D). WML masks were eroded with 2 voxels. Last, GM and CSF masks were dilated with 3 voxels and subtracted from the resulting WML mask and NAWM mask.
T1-weighted scans were segmented into GM, WM, and CSF probability maps using CAT12 (https://neuro-jena.github.io/cat12-help/) and nonlinearly registered to standard Montreal Neurological Institute (MNI) space using SPM12 (http://www.fil.ion.ucl.ac.uk/spm/software/spm12) (Fig 1A). After registration to MNI space, tissue probability maps were thresholded at 0.8.
WMLs were automatically segmented on the FLAIR images using the lesion segmentation toolbox (https://www.applied-statistics.de/lst.html) lesion-prediction algorithm20 and manually corrected (Fig 1B). This segmentation was performed only on FLAIR images that were scored >0 on the Fazekas score to avoid false-positive detections. WML probability maps were thresholded at 0.2, which was deemed appropriate on the basis of visual inspection (Online Supplemental Data). Corrected WML maps were registered to MNI space.
ADC maps were fitted from the DWI data using the FMRIB Software Library (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/) (Fig 1C) and registered to MNI space using the previously obtained transformation matrix.
After transformation to MNI space, normal-appearing white matter (NAWM) was defined by subtracting the WML masks of every available time point (TP) from the WM masks (Fig 1D). WML and NAWM masks were eroded by 1 voxel, and 3-voxel dilated GM and CSF masks were subtracted from the resulting masks to prevent partial volume effects. Unconnected lesions of <5 voxels were excluded (not shown). The resulting lesion masks were termed “full lesion.” Additionally, masks were created that included only WMLs that developed between 2 study visits, denoted as “new lesion.” New lesion masks were created by subtracting the single-voxel dilated full lesion masks from all previous TPs from the current full lesion mask. Again, lesions of <5 voxels were excluded.
The median ADC within the created masks was calculated at each TP (before and after the first appearance on FLAIR-weighted images; Online Supplemental Data). Naming conventions as in the Online Supplemental Data will be used. The TP of first detection on a FLAIR-weighted image was defined the reference TP (time = 0) for that ROI, to which all ADC values were temporally shifted. The difference between ADC within WMLs and NAWM (ΔADC) was calculated by subtracting the median ADC within the NAWM from the lesion ADC.
Statistical Analysis
Statistical analyses were performed in R Studio (Version 2022.2.3.492; http://rstudio.org/download/desktop). Descriptive statistics are reported as mean (SD) or median and interquartile range (IQR) if not normally distributed. ADC values within full and new lesions were compared with corresponding NAWM ADC values using 2-tailed paired t tests with a Bonferroni multiple comparison correction.
Linear mixed effect (LME) modeling was used to investigate changes in ADC and ΔADC with time.21 This investigation was only for new lesions because the age of lesions originating from before the first MR imaging scan was unknown. Subjects were assigned as the random effect and TP of lesion segmentation as a crossed random effect to account for multiple TPs per patient and multiple lesion segmentations per patient, respectively.22 The random effects structure was determined for each model separately as described in the Online Supplemental Data. Next, a priori–determined fixed and interaction effects were added. Fixed effects included time (in months, respectively, to the reference point), sex, and age. Age at baseline was used and centered around the mean. Finally, interaction effects between time and sex and between time and age were included.
If the model could not converge, the structure was simplified as described in the Online Supplemental Data. Model assumptions were tested by fitting of the residuals against the observed data and inspecting their normality. Estimated effects and their 95% CIs are reported.
Models were also evaluated using only the ADC values before detection on FLAIR-weighted images (time < 0). To investigate NAWM ADC in patients with and without lesions, we included a grouping variable “presence of lesions” as a fixed effect, indicating whether WMLs, at any TP, had been found in a patient.
Data Availability
Data are available on reasonable request but not publicly available. Because of the rarity of the disease, even anonymized data can be linked to a specific individual.
RESULTS
Data from 49 patients with Fabry disease initially fulfilled the inclusion criteria. Five FLAIR scans and three T1 scans were excluded because of artifacts detected during image analysis, leading to the exclusion of 1 patient because <3 complete TPs were available. After exclusion, 48 patients with Fabry disease (23 women, all classic phenotypes, all on ERT) with a median of 5 TPs (range, 3–6 TPs) were left, totaling 229 included scan data sets (Table 1). The median interval between TPs was 364 days (range, 196–1134 days). At baseline, 29 (60%) patients had WMLs (18 women; median age, 49 years; range, 21–69 years). Four patients without WMLs at baseline developed WMLs during the study, totaling 33 (69%) patients (18 women; median age, 47 years; range, 21–69 years) who had WMLs by the end of the study. In patients who had lesions at baseline, lesion volume increased significantly during the study (median, 0.6 [IQR, 1.6] mm3 and 1.05 [IQR, 3.1] mm3; Z = 15, P < .01).
Patient characteristics and WML scores
WMLs were predominantly found in the periventricular WM (Fig 2). At time point 1 (TP1), the mean ADC within NAWM was 765 (SD, 32) × 10−6 mm2/s compared with 1113 (SD, 138) × 10−6 mm2/s (t[24] = −14, P < .01) within the full lesion at that TP (full lesion TP1). ADC within full lesion ROIs from all TPs are shown in the Online Supplemental Data.
Lesion prevalence in patients with Fabry disease. Y, Z, and X coordinates refer to standard MNI space.
When we combined new lesion ADC values with the same time respective to their reference point, ADC was significantly higher compared with NAWM up to 5 years before detection on FLAIR-weighted images (Fig 3A; t[8] = 3.9, P = .047). Similarly, ΔADC was significantly higher than zero up to 5 years before detection (Fig 3B; t(8) = 3.9, P = .047). Increased ADC values within WMLs before detection on FLAIR-weighted images also seemed apparent on FLAIR scans and ADC maps of individual patients who developed new WMLs during the study (Fig 4).
ADC (A) and ΔADC (B) in new lesion ROIs compared with NAWM. ADC within lesions with the same time (in years) relative to their reference point were combined to calculate mean values (data points) and their SDs (shaded areas). For example, ADCTP1 within the new lesion TP2 and ADCTP2 within new lesion TP3 both have time = −1 but are compared with the HAWM ADC from TP1 and TP2, respectively. Paired t tests showed that the ADC within lesions was significantly higher than the ADC within NAWM and that ΔADC was significantly higher than zero at every time. NAWM indicates normal appearing white matter.
Yearly FLAIR scans and ADC maps. Three of these patients developed WMLs during the study (A–C), while 1 patient did not (D). Scans are shown in chronologic order from left to right. New WMLs are indicated by the orange arrows. The patient in B also had a large WM hyperintensity at the right parieto-occipital sulcus, which was diagnosed as a result of infarction before the study. Color scales are optimized for each patient individually. Z values indicate the Z coordinate in standard MNI space.
LME modeling revealed that sex had a small-but-significant positive fixed effect on NAWM ADC values, showing a 22.1 (95% CI, 3.7–40.4) ×10−6 mm2/s higher ADC intercept for men compared with women (P = .02; Online Supplemental Data). Neither time, age, nor the presence of lesions had a significant effect on NAWM ADC.
ADC and ΔADC within new lesions were significantly affected by sex, with higher intercepts in men compared with women (60.1 [95% CI, 23.8–96.3] ×10–6 mm2/s, P < .01 and 35.1 [95% CI, 6.0–64.2] ×10–6 mm2/s, P = .02 respectively; Tables 2 and 3). ADC but not ΔADC was also higher and increased faster in older patients, indicated by the fixed effect of age (2.1 [95% CI, 0.6–3.7] ×10–6 mm2/s/month, P < .01) and the interaction effect between age and time (0.04 [95% CI, 0.00–0.07] ×10–6 mm2/s/year/month, P = .03). The effects of time, sex, and age on ADC are shown in Fig 5. ΔADC values increased significantly faster in men compared with women (0.99 [95% CI, 0.27–1.71] ×10–6 mm2/month, P < .01). Time itself had a significant fixed effect only on ΔADC (0.49 [95% CI, 0.03–0.95] ×10–6 mm2/month, P = .04; Online Supplemental Data).
Visualization of LME models describing the ADC within new lesions. The graphs show how time and sex would affect the ADC of 36-, 46-, and 56-year-old patients superimposed on the measured data (A–C respectively). ADC is higher and increases faster in men compared with women and increases faster in older patients.
Summary of the obtained LME models describing ADC and ΔADC
Summary of the obtained LME models describing ADC and ΔADC before lesion detection on FLAIR-weighted images
When we evaluated the models using only the TPs before detection, none of the effects describing ADC remained significant, though sex nearly did (35 [95% CI, –3.6–74.5] ×10–6 mm2, P = .075; Tables 2 and 3). For ΔADC, only the significant fixed effect of time remained (0.57 [95% CI, 0.01–1.14] ×10–6 mm2/month, P = .047; Online Supplemental Data).
DISCUSSION
In this 5-year follow-up study, ADC within WMLs of patients with Fabry disease was assessed before and after detection on FLAIR-weighted images in a unique longitudinal data set. ADC values were already increased compared with NAWM in regions that at later TPs were positively identified as WMLs on FLAIR MR imaging. This increase in ADC with time was affected by sex, with men showing a faster increase in ADC than women.
With Fabry disease being an X-linked disease, men are generally more severely affected by it than women. Indeed, this study shows that ADC within WMLs is higher and diverges faster from NAWM values in men compared with women, a finding not observed before.7,23,24 However, 1 study reported a comparable WML load between men and women, despite the men being younger, suggesting an earlier onset of WMLs in men.7 A different study comparing diffusivity was limited by its small sample size (n = 27), cross-sectional setup, and combined inclusion of classic and nonclassic patients.24 Although here only slightly more patients with WMLs were included (n = 33), the longitudinal design resulted in 472 observations for the analysis.
Higher ADC values were found in regions confirmed as WML at later TPs, consistent with earlier research showing increased diffusivity in NAWM areas commonly associated with WMLs16 and indicating that changes in diffusion precede WML detection on FLAIR images. When only the TPs before detection were entered into the model, no effects remained statistically significant for ADC. For ΔADC, only time remained statistically significant. This finding may be explained by a lack of statistical power to detect changes in ADC while still detecting ADC increasing faster compared with NAWM. We did not develop a prediction model for WMLs based on ADC but merely showed an association between increased ADC and WML development at later TPs. To determine the predictive value of DWI, a model should be verified using a separate data set. Nevertheless, our findings add to the existing evidence that changes in diffusion precede the formation of WMLs in patients with Fabry disease.
The increasing ADC values with age in WMLs were not observed when considered relative to NAWM (ΔADC). This finding suggests that the observed effect of age is not WML-specific but rather affects WM globally. However, time had no significant effect on NAWM ADC. These ambiguous results emphasize the need for caution when interpreting the NAWM ADC results from this patient population. On the other hand, heterogeneous age-related microstructural alterations across the WM25 might prevent the observation of a time effect when considering global WM ADC.
A limitation of this study is the absence of an age- and sex-matched control group. Therefore, ADC within lesions could be compared only with NAWM ADC from other brain regions. NAWM ADC values were similar to literature values from healthy control studies (804 [SD, 110] × 10−6 mm2/s).26 However, it is possible that WM is affected globally in patients with Fabry disease and that areas detected as lesions are just more severely affected than NAWM. This possibility would align with the pathogenesis of patients with MS as observed in studies using magnetization transfer imaging27 and DTI.28 A potential continuation of this study could address this limitation by including a carefully matched healthy control group. Nonetheless, the assessment of diffusion within WMLs both before and after their detection on FLAIR images remained unaffected by the absence of a control group.
All patients included here received ERT, hampering the analysis of the effect of ERT. Treatment duration was explored as a parameter in the LME models but did not yield a statistically significant finding, possibly due to selection bias. Earlier studies reported ambiguous results on the effect of ERT on WMLs. Two studies did not find an effect,3,29 while a third study found that WML burden was more likely to remain stable in patients on ERT.30 Although its ability to evaluate ERT is uncertain, this study shows that DWI could still be useful in clinical decision-making. For example, ERT may be delayed until signs of disease progression are observed. In such cases, DWI could serve as a more sensitive marker for detecting early changes. Moreover, if DWI reveals that WMLs continue to progress despite ERT, this finding may be a signal to consider discontinuing the therapy. However, a more thorough validation of the predictive value of DWI must be performed before it can be relied on for such applications.
Some other limitations should be acknowledged. Due to the rarity of the disease, the number of included patients is relatively low, limiting statistical power. However, the longitudinal study design partly overcomes this limitation. Given the low sample size, exclusive inclusion of patients with the classic phenotype, and the primary focus on investigating the potential predictive value of DWI, a comparison among patients with different GLA gene mutations or phenotypes was not conducted, which remains a topic for future research. Second, because of the large number of MR images, a semiautomated approach was used to segment the WMLs rather than manual segmentation by a trained neuroradiologist. Last, the acquisition protocol used in this study was not initially designed for a scientific investigation but rather for initial exploration, explaining why a more comprehensive technique like DTI was not used in this patient group.
CONCLUSIONS
This work presents the analysis of a unique, longitudinal data set of DWI and FLAIR MR imaging in patients with Fabry disease. Adding to the existing body of literature, these results indicate that changes in diffusion precede the formation of WMLs, suggesting that DWI may be of predictive value for WML formation in Fabry disease. Moreover, these results show that WML diffusivity progresses faster in men compared with women.
Acknowledgment
ChatGPT (https://openai.com/chatgpt) has been used to improve the clarity, grammar, and overall quality of the manuscript.
Footnotes
Disclosure forms provided by the authors are available with the full text and PDF of this article at www.ajnr.org.
References
- Received May 3, 2023.
- Accepted after revision August 16, 2023.
- © 2023 by American Journal of Neuroradiology
