Glial and axonal changes in systemic lupus erythematosus measured with diffusion of intracellular metabolites

CNS involvement is common in systemic lupus erythematosus (SLE). Using diffusion-weighted imaging, Ercan et al. explore the link between cell-specific microstructural changes and neuropsychiatric symptoms. Increased diffusivity of predominantly glial metabolites is observed in SLE patients with previous CNS symptoms, and may be a marker for inflammation-related glial reactivity.

CNS involvement is common in systemic lupus erythematosus (SLE). Using diffusion-weighted imaging, Ercan et al. explore the link between cell-specific microstructural changes and neuropsychiatric symptoms. Increased diffusivity of predominantly glial metabolites is observed in SLE patients with previous CNS symptoms, and may be a marker for inflammation-related glial reactivity.

Abstract

Systemic lupus erythematosus is an inflammatory autoimmune disease with multi-organ involvement. Central nervous system involvement in systemic lupus erythematosus is common and results in several neurological and psychiatric symptoms that are poorly linked to standard magnetic resonance imaging outcome. Magnetic resonance imaging methods sensitive to tissue microstructural changes, such as diffusion tensor imaging and magnetization transfer imaging, show some correlation with neuropsychiatric systemic lupus erythematosus (NPSLE) symptoms. Histological examination of NPSLE brains reveals presence of cerebral oedema, loss of neurons and myelinated axons, microglial proliferation and reactive astrocytosis, microinfacrts and diffuse ischaemic changes, all of which can affect both diffusion tensor imaging and magnetization transfer imaging in a non-specific manner. Here we investigated the underlying cell-type specific microstructural alterations in the brain of patients with systemic lupus erythematosus with and without a history of central nervous system involvement. We did so combining diffusion tensor imaging with diffusion-weighted magnetic resonance spectroscopy, a powerful tool capable of characterizing cell-specific cytomorphological changes based on diffusion of intracellular metabolites. We used a 7 T magnetic resonance imaging scanner to acquire T₁-weighted images, diffusion tensor imaging datasets, and single volume diffusion-weighted magnetic resonance spectroscopy data from the anterior body of the corpus callosum of 13 patients with systemic lupus erythematosus with past NPSLE, 16 patients with systemic lupus erythematosus without past NPSLE, and 19 healthy control subjects. Group comparisons were made between patients with systemic lupus erythematosus with/without past NPSLE and healthy controls on diffusion tensor imaging metrics and on diffusion coefficients of three brain metabolites: the exclusively neuronal/axonal N-acetylaspartate, and the predominantly glial creatine + phosphocreatine and choline compounds. In patients with systemic lupus erythematosus with past NPSLE, significantly higher diffusion tensor imaging mean and radial diffusivities were accompanied by a significantly higher intracellular diffusion of total creatine (0.202 ± 0.032 μm²/ms, P = 0.018) and total choline (0.142 ± 0.031 μm²/ms, P = 0.044) compared to healthy controls (0.171 ± 0.024 μm²/ms, 0.124 ± 0.018 μm²/ms, respectively). Total N-acetylaspartate, total creatine and total choline diffusion values from all patients with systemic lupus erythematosus correlated positively with systemic lupus erythematosus disease activity index score (P = 0.033, P = 0.040, P = 0.008, respectively). Our results indicate that intracellular alterations, and in particular changes in glia, as evidenced by increase in the average diffusivities of total choline and total creatine, correlate with systemic lupus erythematosus activity. The higher diffusivity of total creatine and total choline in patients with NPSLE, as well as the positive correlation of these diffusivities with the systemic lupus erythematosus disease activity index are in line with cytomorphological changes in reactive glia, suggesting that the diffusivities of choline compounds and of total creatine are potentially unique markers for glial reactivity in response to inflammation.

Introduction

Systemic lupus erythematosus (SLE) is a female predominant autoimmune disease that affects multiple organs (Jeltsch-David and Muller, 2014). CNS involvement in SLE is common and results in several neurological and psychiatric symptoms. These symptoms are poorly characterized by standard MRI, which appears normal in ∼50% of patients with neuropsychiatric systemic lupus erythematosus (NPSLE). Focal lesions and vascular infarcts, visible on MRI of patients with NPSLE, are non-specific and often do not correlate with clinical outcome and with symptom severity (Luyendijk et al., 2011).

MRI methods sensitive to tissue microstructural changes, such as diffusion tensor imaging (DTI) and magnetization transfer imaging (MTI), show diffuse white matter changes that correlate with the clinical status of patients with NPSLE (Bosma et al., 2000; Emmer et al., 2008; Jung et al., 2010, 2012; Ercan et al., 2015). Histological examination of NPSLE brains has revealed the presence of cerebral oedema, loss of neurons and myelinated axons, microglial proliferation and reactive astrocytosis, microinfarcts and diffuse ischaemic changes, all of which can affect the image contrast in DTI and MTI (Sibbitt et al., 2010). Therefore, although clinically informative, due to the lack of specificity, these imaging modalities provide limited insight into the microstructural deficit in NPSLE.

Magnetic resonance spectroscopy (MRS) reports on concentrations of cell-specific metabolites, and MRS studies have shown differences in the concentrations (relative to total creatine) of several brain metabolites, including significantly lower N-acetylaspartate (NAA) and significantly higher choline and myo-inositol levels in patients with SLE and NPSLE compared to healthy controls (Sibbitt et al., 1997; Axford, 2001). In addition, one study reported significantly lower NAA in SLE patients with high disease activity compared to those with low disease activity (Appenzeller et al., 2005). Although MRS provides cell-type specific information, it does not provide any structural information.

Diffusion-weighted magnetic resonance spectroscopy (DW-MRS) combines the cell-type specificity of MRS with the microstructural sensitivity of diffusion-weighted imaging (DWI), and allows studying of cell- and compartment-specific properties of tissue microstructure by probing the diffusion of intracellular brain metabolites (Nicolay et al., 2001; Kan et al., 2012; Ronen et al., 2013). Of these metabolites, N-acetylaspartate typically co-measured with N-acetylaspartylglutamate (NAAG) (NAA + NAAG = tNAA) resides almost exclusively in neurons/axons; creatine and phosphocreatine (Cr + PCr = tCr), pivotal in aerobic cell energetics, are found in all neural cells, but their astrocytic concentration is twice their neuronal one, and soluble choline-containing compounds (tCho) are predominantly glial, with a glial/neuronal concentration ratio of 3:1 (Urenjak et al., 1993; Choi et al., 2007). The diffusion properties of these metabolites are strongly dictated by the structural and physiological features of their respective intracellular space, and thus provide a unique in vivo probe for pathology affecting intracellular structures, such as ischaemia (Harada et al., 2002; Zheng et al., 2012), tumours (Harada et al., 2002; Colvin et al., 2008), and axonopathy in multiple sclerosis (Wood et al., 2012), as well as making accurate in vivo cell-specific characterization of tissue microstructure possible (Ronen et al., 2013, 2014).

In this study we use, for the first time, the sensitivity of DW-MRS to selectively report on axonal and glial microstructure (i) to investigate the underlying microstructural alterations in a normal appearing portion of the corpus callosum in the brain of SLE patients with and without history of NPSLE; and (ii) to assess the relationship between DW-MRS indices and SLE activity in the patient population in this study. These studies were performed at ultrahigh field (7 T) to obtain the sensitivity required for robust DW-MRS measurements.

Materials and methods

Human subjects

Twenty-nine patients with SLE (one male, 28 females, age: 43 ± 10 years) and 19 age- and sex-matched healthy volunteers (one male, 18 females, age: 41 ± 11 years) were included in the study. The study adhered to the Declaration of Helsinki and was approved by the institutional review board of our institution. Written informed consent was obtained from all subjects prior to the study. Of 29 patients with SLE, 13 had a history of NPSLE and 16 had no history. For convenience, patients with past NPSLE incidence are hereafter referred to as ‘patients with NPSLE’. All patients with SLE were diagnosed according to the 1982 revised American College of Rheumatology criteria (Tan et al., 1982; Hochberg, 1997). All patients with NPSLE were diagnosed at the Leiden NPSLE clinic after a standardized multidisciplinary medical examination (Zirkzee et al., 2012). Neuropsychiatric diagnoses were classified according to the 1999 American College of Rheumatology case definitions for NPSLE syndromes (ACR, 1999). In the NPSLE group, only patients with at least one CNS NPSLE syndrome were included: cerebrovascular disease (five patients), seizures (three patients), cognitive disorder (three patients), movement disorder (one patient), headache (two patients), acute confusional state (three patients), psychosis (one patient), transverse myelitis (one patient), and anxiety (one patient). To categorize the patients according to SLE disease activity, we calculated the systemic lupus erythematosus disease activity index 2000 (SLEDAI-2K) for each patient (Gladman et al., 2002). Permanent and irreversible damage due to SLE was assessed with the systemic lupus international collaborating clinics (SLICC)/American College of Rheumatology damage index (SDI) (Gladman et al., 1997, 2000). SLE patients with a SLEDAI-2K ≥ 8 were considered to have high SLE activity and were categorized as SLE-active (Appenzeller et al., 2005), while the remaining SLE patients were categorized as SLE-inactive. The demographics of the study and the clinical characteristics of the SLE patients are shown in Table 1.

Table 1.

Patient characteristics

	Patients with NPSLE n = 13	Patients with SLE n = 16	P
Age, years (mean ± SD)	43 ± 8	42 ± 11	0.721
SLE disease duration, years) (mean ± SD)	12 ± 9	8 ± 5	0.178
SLEDAI-2K**, median (range)	5 (2 to 22)	2.5 (0 to 8)	0.006
SDI, median (range)	1 (0 to 5)	1 (0 to 4)	0.241
Antiphospholipid syndrome	3/13 (23%)	1/16 (6%)	0.223
Presence of auto-antibodies
Antinuclear antibody	12/13 (92%)	14/16 (88%)	0.580
Anti-ENA	9/13 (70%)	7/16 (44%)	0.160
Anti-DNA	1/13 (8%)	6/16 (38%)	0.074
Anti-RNP	3/13 (23%)	3/16 (19%)	0.565
Anti-SSA	4/13 (31%)	7/16 (44%)	0.372
Anti-SSB	1/13 (8%)	2/16 (13%)	0.580
Anti-Smith	2/13 (15%)	2/16 (13%)	0.617
Anticardiolipine auto-antibodies	3/13 (23%)	1/16 (6%)	0.223
Lupus anticoagulant	6/13 (46%)	2/16 (13%)	0.055
Anti-β2 glycoprotein IgG	2/13 (15%)	0/16 (0%)	0.192
Presence of ACR criteria ever in disease course
Malar rash	5/13 (38%)	8/16 (50%)	0.404
Discoid lupus	2/13 (15%)	0/16 (0%)	0.192
Photosensitivity	5/13 (38%)	7/16 (44%)	0.537
Ulcers	7/13 (54%)	7/16 (44%)	0.434
Arthritis*	12/13 (92%)	9/16 (56%)	0.038
Serositis	4/13 (31%)	6/16 (38%)	0.507
Lupus nephritis	4/13 (31%)	5/16 (31%)	0.647
Neurological disorder	4/13 (31%)	0/16 (0%)	0.448
Haematological disorder	6/13 (46%)	8/16 (50%)	0.566
Immunological disorder	9/13 (70%)	13/16 (81%)	0.374
Antinuclear antibodies	13/13 (100%)	16/16 (100%)	0.374
Current medication
Prednisone	9/13 (70%)	9/16 (56%)	0.372
Azathioprine	3/13 (23%)	6/16 (38%)	0.336
Methotrexate	1/13 (8%)	1/16 (6%)	0.704
Hydroxychloroquine	11/13 (85%)	12/16 (75%)	0.435
Mycophenolate mofetil	3/13 (23%)	2/16 (13%)	0.396

ACR = American College of Rheumatology; IgG = immunoglobulin G; SDI = systemic lupus international collaborating clinics (SLICC)/American College of Rheumatology damage index.

*P < 0.05; ^**P < 0.01.

Data acquisition

All subjects were scanned on a 7 T Philips Achieva MRI scanner (Philips Healthcare) equipped with a 32-channel receive head coil (Nova Medical Inc.). The scan protocol consisted of a short survey scan and a sensitivity encoding reference scan followed by: (i) sagittal 3D T₁-weighted images (field of view = 246 × 246 × 174 mm³, resolution: 0.85 × 0.85 × 1 mm³, repetition time/echo time = 4.00/1.84 ms, total scan time = 1.59 min); (ii) axial multislice diffusion tensor images (field of view = 224 × 224 × 150 mm³, resolution: 1.75 × 1.75 × 2.20 mm³, repetition time/echo time = 10 000/65 ms, 15 diffusion-weighting directions with b = 1000 s/mm², total scan time: 3 min); and (iii) single-volume diffusion-weighted spectroscopy scans (detailed protocol below).

Diffusion-weighted magnetic resonance spectroscopy

Protocol

The DW-MRS sequence was based on the PRESS (Point Resolved Spectroscopy) sequence with bipolar diffusion-weighting gradients added on both sides of the 180° pulses. A 3 cm³ volume of interest [25 (AP) × 15 (RL) × 8 (FH) mm³] was positioned on the anterior body of the corpus callosum as shown in Fig. 1. The diffusion-weighting gradients were applied in two directions: a right-left direction in the volume of interest frame, mostly parallel to the direction of the callosal fibres [direction (1,0,0)], and a direction mostly perpendicular to the callosal fibres [direction (0,−1,1)], as shown in Fig. 1A and B. The centre frequency was set to the tNAA singlet peak at 2.0 ppm. Water suppression was performed using two frequency-selective excitation pulses, each followed by a dephasing gradient before metabolite excitation. Pencil beam second-order shimming was performed, resulting in a typical tNAA line width of 10 Hz. A peripheral pulse unit was used to gate data acquisition to the cardiac cycle, thereby minimizing signal fluctuations due to cardiac pulsation. The parameters for DW-MRS acquisitions were: echo time = 121 ms, repetition time = three cardiac cycles (∼3000 ms), cardiac trigger delay = 300 ms, number of time-domain points = 1024, spectral width = 3000 Hz, gradient duration (δ) = 37 ms, bipolar gap = 16 ms, diffusion time (Δ) = 60.5 ms with five different gradient amplitudes resulting in b-values of 212, 651, 1335, 2262, and 3462 s/mm² in the (1,0,0) direction and 440, 1336, 2718, 4586, and 6945 s/mm² in the (0,−1,1) direction. The total number of spectra per diffusion condition was 32, resulting in a total scan time of 10–15 min. Following this scan, a shorter scan (fewer signal averages) with identical volume of interest position and diffusion conditions was performed without water suppression and with the center frequency set at the water resonance frequency. These spectra were used for eddy-current correction in the post-processing stage.

Figure 1.

The position of the volume of interest in sagittal (A) and coronal (B) views. Gradients applied in directions approximately perpendicular (A) and parallel (B) to the callosal fibres are shown in solid lines. Typical spectra acquired with diffusion weighting in the (0,−1,1) and the (1,0,0) directions are shown as a function of b-value in C and D, respectively. Line broadening of 5 Hz was applied for display purposes. The attenuation of the tNAA peak is more pronounced in the direction approximately parallel to the axonal fibres, along which diffusion of intra-axonal molecules is less restricted than in the direction perpendicular to the fibres.

Processing

All spectral preprocessing was performed with custom codes in MATLAB® release R2014b (Mathworks, Natick, MA, USA). Spectral preprocessing consisted of correcting DW-MRS data for eddy currents, zero-order phasing, correction of frequency drift for individual acquisitions, and removal of the residual water peak: averaged spectra were generated for each condition (Wood et al., 2015). Figure 1C and D show typical sets of diffusion-weighted spectra obtained with diffusion-weighting in the (1, 0, 0) direction (Fig. 1C) and the (0, −1, 1) direction (Fig. 1D), respectively and from a healthy control subject. The resulting spectra were quantified with LCModel (Provencher 1993). Cramér–Rao lower bound (CRLB) values were used to evaluate the quality of the spectra for each diffusion condition, and the acceptance threshold for DW-MRS data inclusion was set at CRLB <20%. Based on this, datasets from one NPSLE subject and one SLE subject were excluded from the tNAA analysis.

The LCModel spectral estimates were used to calculate the diffusivity (D_par) along the (1,0,0) direction (roughly parallel to the callosal fibres) and diffusivity (D_perp) along the (0,−1,1) direction (roughly perpendicular to the callosal fibres) for tNAA, tCr and tCho. These were calculated by performing a linear fit of the natural logarithm of the DW-MRS signal amplitudes as a function of the diffusion weighting value b, assuming a monoexpoential decay of the signal as a function of b in each direction:

$$ln\left(\frac{S_{b,i}}{S_{b_0}}\right)=-b_i{D}_i$$ (1)

where $S_{b,i}$ is the measured signal in direction i, $S_{b_{0,i}}$ is the signal without diffusion weighting, b_i is the value of b in the direction i, and D_i is the calculated diffusion coefficient for direction i. Even though it is possible that the metabolite diffusion-weighted signal decay displays non-monoexponential behaviour at very high values of b, our previous work has shown that in the range of b-values used in this study the assumption of mono-exponentiality is valid and diffusivity values are reproducible (Kan et al., 2012; Wood et al., 2015). An average of D_par and D_perp was calculated to assess the average diffusivity (D_avg) for tNAA, tCr and tCho (Ronen et al., 2013). The quality of the linear fittings was evaluated via calculation of the coefficient of determination and an acceptance threshold was set at 75%, leading to exclusion of D_avg(tCr) values obtained from two patients with NPSLE and one patient with SLE.

Image processing

DTI volumes were motion-corrected with ExploreDTI (Leemans et al., 2009) and further processed with the DTI toolbox (Behrens et al., 2003) of the FMRIB Software Library (FSL release 5.0, http://www.fmrib.ox.ac.uk/fsl/) to obtain the following DTI measures for each subject: fractional anisotropy, mean diffusivity, axial diffusivity and radial diffusivity. These DTI metrics were further analysed with tract-based spatial statistics (TBSS) (Smith et al., 2006). Statistical differences between patients with NPSLE, patients with SLE and healthy control subjects were assessed on white matter tract voxels in FA-MNI152 standard space using 5000 permutations and ANOVA tests were performed with a significance threshold of P < 0.05 and were corrected for multiple comparisons based on threshold-free cluster enhancement (Winkler et al., 2014). One SLE patient dataset was excluded due to poor registration to FA-MNI152.

T₁-weighted images were used for tissue segmentation within the volume of interest (Wood et al., 2015). Fractional anisotropy maps were registered to the T₁-weighted image of the same subject first by affine transformation using FSL FLIRT (Jenkinson and Smith, 2001; Jenkinson et al., 2002) and subsequently by non-rigid transformation using FNIRT (Andersson et al., 2010). The inverse transformation matrices generated were used to register the DW-MRS volume of interest to the DTI space. Subsequently, the registration procedure in TBSS was applied to transform each volume of interest to MNI152 space.

T₁-weighted volumes were further processed in FreeSurfer (http://surfer.nmr.mgh.harvard.edu/) and the intracranial volume, total brain volume, centre corpus callosum volume and mid-anterior corpus callosum volume were calculated for each subject. To evaluate whole brain and callosal atrophy due to SLE and NPSLE, total brain volume, centre corpus callosum volume and mid-anterior corpus callosum volume were normalized according to the intracranial volume of the same subject.

Statistical analyses

Patients with NPSLE and those with SLE were compared with respect to demographic characteristics, presence of autoantibodies and ACR criteria and current medication using chi-square tests and ANOVA or Mann-Whitney tests when appropriate. Primary dependent measures included in the statistics were D_avg(tNAA), D_avg(tCr), D_avg(tCho). Shapiro-Wilk test was used to examine the distribution of variables, and resulted in all being normally distributed. Equality of variances of the different groups was assessed using Levene’s test. Between-group differences on all D_avg values were evaluated using one-way-ANOVAs (pairwise comparisons). P < 0.05 was considered to represent statistically significant differences. Bonferroni’s correction was used to correct for multiple comparisons. Analysis of covariance was performed to analyse the influence of age and SLEDAI-2K on the differences of mean D_avg values between groups. As SLEDAI-2K and SDI scores were not normally distributed, correlations between the metabolite D_avg measurements and SLEDAI-2K score and SDI score were evaluated with Spearman’s rank correlation. Correlations between SLE duration and D_avg values were assessed with Pearson’s correlation test. All statistical analyses were performed with SPSS version 20.0 for Windows (IBM SPSS statistics, Chicago, IL, USA). Scattered plots were generated using GraphPad Prism 5 for windows, version 5.01, GraphPad Software, USA.

Results

Diffusion-weighted magnetic resonance spectroscopy

Average metabolite D_avg values for the NPSLE, SLE and healthy control groups are shown in Table 2 and group D_avg data for the three population groups are displayed in Fig. 2. When all three groups were compared with ANOVA with age included as a covariate, significant differences were found in D_avg(tCho) (P = 0.006) and D_avg(tCr) (P = 0.030). No significant differences were found in D_avg(tNAA) among the groups. Pairwise comparisons showed that in patients with NPSLE, D_avg(tCr) and D_avg(tCho) were significantly higher than in healthy controls after correction for age (mean D_avg(tCr) in NPSLE = 0.202 ± 0.032 μm²/ms, mean D_avg(tCr) in healthy controls = 0.171 ± 0.024 μm²/ms, P = 0.018 and mean D_avg(tCho) in NPSLE = 0.142 ± 0.031 μm²/ms, mean D_avg(tCho) in healthy controls = 0.124 ± 0.018 μm²/ms, P = 0.044). No significant difference was found in D_avg(tNAA) values between patients with NPSLE and healthy control subjects. No significant differences were observed in any metabolite D_avg between NPSLE and SLE or between SLE and healthy control groups. SLEDAI-2K was used as a covariate when patients with SLE and those with NPSLE were compared. When all SLE patients, with and without past CNS involvement, were grouped together and compared to healthy control subjects, D_avg(tCr) and D_avg(tCho) remained significantly higher in patients with SLE than in healthy control subjects after correcting for age [mean D_avg(tCr) in all SLE patients = 0.190 ± 0.034 μm²/ms, mean D_avg(tCr) in healthy controls = 0.171 ± 0.024 μm²/ms, P = 0.060 and mean D_avg(tCho) in all SLE patients = 0.135 ± 0.028 μm²/ms, mean D_avg(tCho) in healthy controls = 0.124 ± 0.018 μm²/ms, P = 0.008].

Table 2.

Metabolite D_avg values for patients with NPSLE, patients with SLE and healthy control subjects

	NPSLE n = 13	SLE n = 16	Healthy control n = 19	% Difference NPSLE versus healthy control
Davg(tNAA) μm2/ms	0.241 ± 0.019	0.227 ± 0.025	0.226 ± 0.019	7%
Davg(tCr) μm2/ms	0.202 ± 0.032*	0.181 ± 0.034	0.171 ± 0.024	18%
Davg(tCho) μm2/ms	0.142 ± 0.031*	0.128 ± 0.024	0.124 ± 0.018	14%

tCho = choline compounds; tCr = total creatine + phosphocreatine; tNAA = N-acetylaspartate.

*P < 0.05 versus healthy control subjects.

Figure 2.

Metabolite D_avg values for healthy controls subjects and patients with SLE and patients with NPSLE. D_avg (tNAA), D_avg (tCr) and D_avg (tCho) data are shown in A, B and C, respectively. Statistically significant differences are shown as *P < 0.05 and ^**P < 0.01. No significant differences were found between SLE and healthy control subjects, or between NPSLE and SLE in any of the metabolite D_avg values. HC = healthy controls.

Average diffusion coefficients of all metabolites showed a link to disease activity in both the patients with SLE and those with NPSLE. When all patients with SLE were pooled together, regardless of their neuropsychiatric status, SLEDAI-2K scores correlated positively with D_avg(tNAA) (r = 0.412, P = 0.033), D_avg(tCr) (r = 0.405, P = 0.040) and D_avg(tCho) (r = 0.480, P = 0.008). Scatter plots of the SLEDAI-2K scores of all patients as a function of metabolite D_avg values are shown in Fig. 3. When patients with SLE and those with NPSLE categorized as SLE-active (SLEDAI-2K ≥ 8, n = 5) were compared to healthy control subjects, statistically significant differences in D_avg(tCr) were found [D_avg(tCr) in SLE-active = 0.216 ± 0.038 μm²/ms, D_avg(tCr) in healthy controls = 0.171 ± 0.024 μm²/ms, P = 0.006], as well as in D_avg(tCho) [D_avg(tCho) in SLE active = 0.154 ± 0.037 μm²/ms, D_avg(tCho) in healthy controls = 0.124 ± 0.018 μm²/ms, P = 0.006] after correcting for age. No correlation was found between metabolite D_avg values and SLE duration or SDI scores.

Figure 3.

Correlation of metabolite D_avg values with patient SLEDAI-2K scores. The resulting Spearman’s rank correlation r and significance of the correlations are shown for D_avg (tNAA) (A), D_avg(tCr) (B) and D_avg(tCho) (C).

Diffusion tensor imaging and volumetric results

Significantly lower fractional anisotropy, higher mean diffusivity and higher radial diffusivity were found throughout white matter in the NPSLE patient group compared to the healthy control and SLE groups, including the callosal region within the DW-MRS volume of interest (P < 0.05). Figure 4 shows maps of statistically significant differences (P < 0.05) in DTI measures overlaid on the MNI152 T₁-weighted image, and the cumulative DW-MRS volume of interest (i.e. the sum of all the individual volumes of interest following transformation from the individual subject coordinates to the FA-MNI152 coordinates). The voxels with significantly higher mean and radial diffusivity values in the patients with NPSLE compared to healthy controls are shown in blue and voxels with lower fractional anisotropy values in patients with NPSLE compared to healthy controls are shown in red. No significant differences were found in any DTI measure between SLE patients (with and without past CNS involvement) and healthy control subjects. No significant differences were found in corpus callosum volumes or total brain volumes between patients with SLE or those with NPSLE, and healthy control subjects.

Figure 4.

Tract-based spatial statistics results showing regions with statistically significant differences in DTI measures in the white matter skeleton of patients with NPSLE and healthy control subjects (P < 0.05). Maps are shown for one sagittal (left) and one coronal (right) slice in MNI152 space. The mean fractional anisotropy skeleton is shown in green, regions with higher values in the patients with NPSLE compared to healthy control subjects are shown in blue and regions with lower values in patients with NPSLE compared to healthy control are shown in red. Cumulative volume of interests chosen for DW-MRS of patients with NPSLE and healthy controls are shown in yellow. FA = fractional anisotropy; MD = mean diffusivity; RD = radial diffusivity.

Discussion

This is the first study to address cell-specific microstructural alterations in the brain of SLE patients with DW-MRS at ultrahigh field. This study focused on measuring the diffusion properties of two predominantly glial metabolites, tCr and tCho, and one exclusively axonal/neuronal metabolite, tNAA. The most salient finding in this study is the strong and consistent link between both D_avg(tCr) and D_avg(tCho) and disease state, with respect to disease activity and to past CNS involvement, suggesting glial involvement in the brain of these patients. Two potential pathological mechanisms that can explain the significantly higher tCr and tCho diffusivities found in patients with NPSLE are inflammation-mediated morphological changes in microglia and astrocytes, and intracellular oedema, which would affect both glia and neurons/axons (Brooks et al., 2010; Graeber and Streit, 2010; Sofroniew and Vinters, 2010).

Astrocytic and microglial reactivity in response to inflammation and/or ischaemia are both highly consistent with an increase in intracellular diffusivity in glia. Reactivity-related cellular hypertrophy and thickening of the processes near the soma (especially in astrocytes) (Sofroniew, 2009) would result in an increase of the intracellular space, and a decrease in molecular crowding and intracellular tortuosity, leading to increased diffusivity in the cytosol. The pathogenesis of NPSLE is thought to involve various immune and inflammatory processes that can lead to neuronal injury and vasculopathy (Popescu and Kao, 2011). The inflammatory response to injury likely results in glial reactivity and cellular hypertrophy, especially in microglia and astrocytes (Graeber and Streit, 2010; Sofroniew and Vinters, 2010). Histolopathological investigations of brains of patients with NPSLE confirm the widespread presence of reactive microglia and astrocytes, as well as of lipid-laden macrophages among the heterogeneous pathological phenomena (Brooks et al., 2010). Furthermore, the correlation of in vivo MRS results with histological results from the same patients (Brooks et al., 2010) suggests: (i) an association between an increase in tCho concentrations and gliosis, vasculopathy and oedema; (ii) possible association of tCr with gliosis and reduced neuronal/axonal density; and (iii) an association between lower tNAA concentrations and a decrease in neuronal/axonal density (Brooks et al., 2010).

The higher D_avg(tNAA) values found in patients with NPSLE compared to healthy controls may be attributed to changes in cytosolic viscosity in axons due either to neuronal/axonal damage or to cytotoxic oedema, both of which are seen in the histopathology of brains of patients with NPSLE (Sibbitt et al., 2010). Higher D_avg(tNAA) has also been observed in a study of patients with schizophrenia where it was hypothesized that inflammatory processes may play a role (Du et al., 2013). On the other hand, lower tNAA parallel diffusivity values were found in patients with multiple sclerosis compared to healthy controls in a study focused on myelin and axonal changes in the corpus callosum (Wood et al., 2012). It is likely that the different behaviours in tNAA diffusivity seen in multiple sclerosis and in NPSLE reflect different intra-axonal pathological mechanisms associated with these two diseases. Central to multiple sclerosis are demyelination and axonopathy (Petzold et al., 2011; Stadelmann et al., 2011). As demyelination has no direct effect on diffusion in the intra-axonal space, it has been hypothesized that in multiple sclerosis the decrease in tNAA axosolic diffusivity stemmed from axonal damage that included unusual patterns of neurofilament phosphorylation and packing compared to normal tissue, and a less organized axoskeleton and/or problems with axonal transport (Petzold et al., 2008). In contrast to findings in multiple sclerosis, histology of patients with NPSLE have shown that cerebral oedema occurs much more frequently than axonal/neuronal loss (Sibbitt et al., 2010) and is thus more compatible with the increase in axosolic diffusivity, as evidenced by the higher levels of D_avg(tNAA) observed in our study.

The high correlation in all patients with SLE or NPSLE between SLE disease activity, as quantified by the SLEDAI-2K score, and D_avg(tCr) and D_avg(tCho), suggests that the SLE-related peripheral inflammation and autoimmune response may have an effect on the brain, independent of overt clinical CNS involvement in SLE. Additionally, the correlation of D_avg(tNAA) values with SLEDAI-2K scores suggests a permanent or continuous damage to axons correlated with high SLE activity. This is further corroborated by the finding that patients with higher disease activity (those we defined as SLE-active) have higher metabolite diffusivity levels and higher statistical significance in the difference in metabolite diffusivity compared to healthy controls. A previous MRS study in patients with NPSLE/SLE has shown a significantly lower tNAA/tCr level in SLE patients with a high SLEDAI-2K score, and that the level of tNAA/tCr was renormalized in follow-up for patients who were no longer SLE-active, regardless of their neuropsychiatric status (Appenzeller et al., 2005). This finding suggests a pathological mechanism, attributed by the authors to neuronal dysfunction that affects both neurons and axons, the degree of which depends on SLE disease activity, but which is essentially reversible in nature. In our view, and based on our corroborative findings, we attribute this finding to intracellular/intra-axonal oedema. It has been suggested that inflammation outside the brain can prime microglia and result in microglial activation for several weeks (Perry and Holmes, 2014). Our findings, as well as those described by Appenzeller et al. (2006), support the view that systemic inflammation affects the brain in NPSLE, and the underlying mechanism by which this occurs, e.g. potential disruption of the blood–brain barrier in SLE (Abbott et al., 2003), should be further investigated.

Metabolite apparent diffusion coefficient values found in healthy control in this study are similar to those reported in previous DW-MRS studies performed on a similar region of the corpus callosum at 7 T and 3 T (Ronen et al., 2013; Wood et al., 2015). A recently published robustness and reproducibility study of DW-MRS in the anterior body of the corpus callosum, aimed to provide guidelines for DW-MRS acquisition for clinical studies such as the one presented here. Using power calculations based on actual data it was estimated that to observe a 10% difference in D_avg values for tNAA, tCr and tCho in a case-control study, nine, four and 12 subjects, respectively, are sufficient (Wood et al., 2015). In this current study we observed a 7% increase in D_avg(tNAA), 19% increase in D_avg(tCr) and 14% in D_avg(tCho) in patients with NPSLE compared to healthy control, with 13 and 19 subjects per group, respectively. This suggests that it is feasible to observe reliably the disease effect on metabolite D_avg reported here. Moreover, based on our power estimation, the number of subjects required to observe reliably a difference in D_avg(tCho) is higher than that required for D_avg(tNAA). This further supports the notion that our findings indicate a larger effect size for glial than for axonal involvement in SLE/NPSLE. The DTI results are also consistent with previous studies on SLE and NPSLE (Emmer et al., 2010; Jung et al., 2010; Ercan et al., 2015). Significantly lower fractional anisotropy values in the anterior body of the corpus callosum of patients with NPSLE compared to healthy controls were previously reported in a DTI study on NPSLE where region of interest analysis was performed (Zimny et al., 2014). This study also included MRS measurements in normal appearing parietal white matter, and no differences in glial metabolite concentrations (tCr, tCho) between patients with NPSLE and healthy controls were reported. Although changes in concentrations of glial metabolites have been linked to glial reactivity and neuroinflammation (Chang et al., 2013) this may suggest that the diffusion properties of glial metabolites are more sensitive to glial reactivity than their cellular concentration. Atrophy measurements did not show any significant callosal or global atrophy in the patient population, possibly stemming from a type II error due to the small size of the cohort.

There were several challenges and limitations to the study. The main (unmet) challenge of this study was to scan patients with active neuropsychiatric symptoms at the time of the scan, as well as to include more patients with high SLE disease activity. Low incidence of active NPSLE was a factor, as well as significant potential discomfort to patients with clinically overt neuropsychiatric symptoms, preventing these patients from being involved in a research study that is not part of the routine clinical procedure. Another limitation of this study is the small number of patients per specific neuropsychiatric syndrome. This is a generally recognized problem in NPSLE studies, mainly related to the low prevalence and the high heterogeneity of NPSLE. Subsequently, definite conclusions concerning the relationship between DW-MRS findings and NPSLE syndromes cannot be drawn. In our patient cohort, the relative low number of patients with antiphospholipid syndrome made it difficult to evaluate the effect of ischaemic/vascular changes on metabolite diffusion. As a consequence, we did not separately analyse changes in ischaemic and inflammatory patients with NPSLE. Future studies will focus on separate evaluation of the effects of neuropsychiatric activity, ischaemic and inflammatory effects. The study would have benefitted from a separate MRS acquisition at short echo time, for an accurate evaluation of the metabolite concentrations in the same volume of interest, together with additional cell-specific metabolites such as glutamate, glutamine and myo-inositol.

In conclusion, the results presented in this study show for the first time that intracellular metabolite diffusion reflective of glial and neuronal/axonal involvement can be measured by DW-MRS in a complex autoimmune inflammatory disorder such as SLE/NPSLE. This technique has great potential for the study of the aetiology of disease-related changes in tissue microstructure of patients with SLE/NPSLE. We believe that if incorporated in a comprehensive diagnostic scanning protocol together with existing microstructural MRI tools, such as DTI, MTI and susceptibility weighted imaging, DW-MRS can contribute to the diagnostic process of these patients and may help unravel underlying pathogenic mechanisms in this complex disease.

Glossary

Abbreviations

DTI =

diffusion tensor imaging

MRS =

magnetic resonance spectroscopy

NAA =

N-acetylaspartate

(NP)SLE =

(neurospsychiatric) systemic lupus erythematosus

SLEDAI-2K =

systemic lupus erythematosus disease activity index 2000

Funding

A.E.E. was supported by a Dutch Arthritis Association (Reumafonds) grant (# 2010_0000135). E.T.W. was supported by Intramural Research Program of the National Institute of Neurological Disorders and Stroke and is the recipient of an NINDS Competitive Intramural Graduate Fellowship award. The C.J. Gorter Center for High Field MRI is supported by grant number 176.010.2005 from the Nederlandse Organisatie voor Wetenschappelijk Onderzoek: Grootschalige Onderzoeksfaciliteiten (Dutch Organization for Scientific Research: Large-Scale Research Facilities).