Abstract
Objective
Extensive blood–brain barrier (BBB) leakage has been linked to cognitive impairment in SLE. This study aimed to examine the associations of brain functional connectivity (FC) with cognitive impairment and BBB dysfunction among patients with SLE.
Methods
Cognitive function was assessed by neuropsychological testing (n = 77). Resting-state FC (rsFC) between brain regions, measured by functional MRI (n = 78), assessed coordinated neural activation in 131 regions across five canonical brain networks. BBB permeability was measured by dynamic contrast-enhanced MRI (n = 61). Differences in rsFC were compared between SLE patients with cognitive impairment (SLE-CI) and those with normal cognition (SLE-NC), between SLE patients with and without extensive BBB leakage, and with healthy controls.
Results
A whole-brain rsFC comparison found significant differences in intra-network and inter-network FC in SLE-CI vs SLE-NC patients. The affected connections showed a reduced negative rsFC in SLE-CI compared with SLE-NC and healthy controls. Similarly, a reduced number of brain-wide connections was found in SLE-CI patients compared with SLE-NC (P = 0.030) and healthy controls (P = 0.006). Specific brain regions had a lower total number of brain-wide connections in association with extensive BBB leakage (P = 0.011). Causal mediation analysis revealed that 64% of the association between BBB leakage and cognitive impairment in SLE patients was mediated by alterations in FC.
Conclusion
SLE patients with cognitive impairment had abnormalities in brain rsFC which accounted for most of the association between extensive BBB leakage and cognitive impairment.
Keywords: SLE, cognitive impairment, blood–brain barrier, functional connectivity, neuroimaging
Rheumatology key messages.
Functional MRI demonstrated abnormalities in brain functional connectivity (FC) in SLE patients with cognitive impairment.
Extensive blood–brain barrier (BBB) leakage is associated with cognitive impairment in SLE patients.
FC accounted for 64% of the association between BBB leakage and cognitive impairment.
Introduction
Cognitive impairment is one of the most common neuropsychiatric manifestations of SLE (NPSLE) [1], with a prevalence of 33–43% in unselected SLE patients [2]. The majority of SLE patients with cognitive impairment experience a fluctuating course of impairment [3], suggesting CNS dysfunction that is potentially reversible.
The neuropathological origins of cognitive impairment in SLE are unclear. Structural brain abnormalities detected by MRI in patients with SLE have weak associations with cognitive impairment [4]. Brain activation assessed with functional MRI (fMRI) provides an indirect index of neuronal activity. Functional connectivity (FC) is calculated from correlation between simultaneous neuronal activity in different anatomical brain sites. Mutually synchronized activation in different regions, indicated by positive correlations, represents positive FC, while reciprocal de-activation between regions, indicated by negative correlations, represents negative FC. Synchronization of neuronal activity is the basis of integrating brain regions into networks that perform higher brain functions, including cognition.
Previous studies demonstrated differences in FC between SLE patients and healthy subjects [5–8] and an association between altered FC and cognitive impairment [9, 10]. We reported an association between extensive BBB leakage with cognitive impairment in SLE patients [11], but no study has concurrently measured FC, BBB permeability and cognition. We hypothesized that in SLE patients the association between extensive BBB permeability and cognitive impairment is mediated by alterations in FC.
We compared whole-brain FC in SLE patients with and without cognitive impairment to identify differences in patterns of connectivity between groups. We assessed the associations between FC patterns and extent of BBB leakage. Finally, we performed causal mediation analysis to determine whether the association between BBB leakage and cognitive impairment is mediated by alterations in FC.
Methods
Participants
Seventy-eight patients fulfilling revised ACR criteria for SLE [12] were consecutively recruited from the Dalhousie Lupus Clinic, Division of Rheumatology, Queen Elizabeth II Health Sciences Center, Halifax, Nova Scotia, Canada. Patients were not pre-screened for cognitive impairment. Healthy controls (HC) recruited for fMRI scanning (n = 71) were 18–75 years old with no history of neuropsychiatric disease, SLE, diabetes mellitus or chronic pain. The study was approved by the Nova Scotia Health Authority (NSHA) Research Ethics Board, and participants provided written informed consent.
For SLE patients, demographic variables, previous NP events and attribution [13], medications, lifestyle habits and comorbidities (smoking, diabetes mellitus and hypertension) were documented. Global SLE disease activity [SLEDAI-2000 (SLEDAI-2K) [14]] and cumulative organ damage [SLICC/ACR Damage Index (SDI) [15]] were recorded. Laboratory variables included complete blood count, serum creatinine, urinalysis, anti-dsDNA, aPL (aCL, LA), C3 and C4 levels. Clinical and cognitive assessments, blood collection and MRI scanning were performed on the same day.
Cognitive function
Neuropsychological tests based on ACR recommendations [16], were completed by 77 SLE patients (one non-native English speaker was omitted) to assess cognitive domains commonly affected in SLE [17]. Information processing speed and executive abilities were assessed by the Symbol Digit Modalities Test [18] and Design Fluency test [19], respectively. Components of the California Verbal Learning Test II [20] provided indices of attention span, new learning and delayed recall. Raw scores were standardized based on published normative data and converted to Z-scores [19]: Z-scores ≤–1.5 in one or more domains indicated cognitive impairment (SLE-CI), with the remainder having normal cognition (SLE-NC).
Statistical analysis of clinical data
Summary group data were expressed as percentages, mean (s.d.), or median and interquartile ranges as appropriate. For categorical variables, comparisons between groups were performed with Fisher’s exact test. For continuous variables that were normally distributed, t-tests and ANOVAs were used for group comparisons. For continuous variables not normally distributed, Wilcoxon rank sum tests were used.
Imaging
MRI acquisition
Neuroimaging used a 3T MRI scanner (Discovery MR750, GE Healthcare, Waukesha, WI, USA), with a 32-channel head coil (MR Instruments, Inc., Minneapolis, MN, USA). T1-weighted images provided anatomical reference [specifications: 1 × 1 × 1 mm isometric voxels, 184 × 256 × 256 voxel resolution, echo time (TE) 1.33 ms, repetition time (TR) 4.0 ms, flip angle 9°], followed by resting state fMRI (rsfMRI) signals (3 × 3 × 3 mm voxels, 72 × 72 × 51 voxel resolution, TR 0.95 s, 500 TRs) and dynamic contrast-enhanced T1-weighted images (1.25 × 1.25 × 6 mm voxels, 192 × 192 × 34 voxel resolution, TE/TR = 2/4 ms, flip angle 15°, averages 1, Δt = 20 s) acquired between 6 and 20 min after i.v. injection of gadobenate dimeglumine (0.1 mmol/kg, MultiHance, Bracco Imaging Canada, Montreal, QC, Canada).
BBB leakage analysis
Dynamic contrast enhanced MRI (DCE-MRI) scanning for quantitative assessment of BBB permeability was available in 61/78 patients who had rsfMRI scans. Contraindication to contrast use, patient refusal or technical difficulties preventing interpretation accounted for missing scans. The analysis was performed as reported [11, 21, 22]. In brief, this involved normalization to the Montreal Neurological Institute standardized brain image, and identification of brain voxels with BBB leakage. The percentage of voxels with BBB leakage determined whole-brain BBB leakage.
Abnormal BBB leakage threshold was based on our previous study of DCE-MRI, which included 57/61 SLE patients in the current study and 9 healthy controls. The prior study identified extensive BBB leakage with an outlier analysis using the Median Absolute Deviation approach. Participants with >9.11% of brain tissue affected by BBB leakage were considered to have extensive BBB leakage [11].
Resting-state fMRI
Data from 78 SLE patients were pre-processed with FSL, AFNI and FreeSurfer [23–25] based on scripts by the 1000 Functional Connectomes Project [26] to mitigate artefacts due to head motion and off-resonance fields, to filter out temporal and spatial noise, and to remove non-brain data.
Pre-processing involved cropping the first five time points to let the blood oxygen level-dependent (BOLD) signal reach steady state and motion correction to the mean of the remaining time series; this also allowed calculation of head motion parameters for later use. A sample frame was extracted from the mean-aligned data for image registration. Images were spatially smoothed with a Gaussian kernel (full width at half-maximum: 6 mm), intensity-normalized, temporally filtered with a band-pass filter (0.005–0.3 Hz cut-off frequencies), then detrended. Measures of motion outliers based on framewise displacement (FD) and the derivative of variability across voxels (DVARS) were calculated using previously published methodologies [27].
fMRI data were re-scaled, re-oriented and registered to the Montreal Neurological Institute standardized brain [28]. Images were segmented into grey matter, white matter, and cerebrospinal fluid, and mean signals from the latter two were calculated. The white matter and cerebrospinal fluid signals, along with the global signal and head motion parameters, were regressed out of the BOLD signals for each voxel as nuisance data, leaving residuals that better represented resting-state functional activity. Finally, motion outliers calculated previously were extracted for each participant: maximum FD ≥3 mm and a motion outlier rate >30% in either FD or DVARS [29] were additional exclusion criteria, although no subjects met them.
Brain network analysis
From pre-processed data, BOLD signal time series were calculated for 131 regions of interest (ROIs), using an optimized version of the Harvard-Oxford anatomical parcellation [29–31]. These ROIs were assigned to one of five canonical resting-state sub-networks (subcortical, sensory, default mode, attention/executive and language/memory) using a meta-analytic framework (Neurosynth) described previously [31].
To calculate the resting-state functional connectivity (rsFC) between ROIs, a zero-lag Pearson correlation matrix was calculated between mean BOLD signals for each ROI and every other ROI. The correlation coefficient between BOLD signals in two ROIs is referred to as the rsFC value for that connection: large positive or negative rsFC values denote strong positive or negative connectivity between two ROIs, while small rsFC values denote weak to non-existent connectivity. ROIs were sorted by resting-state network [31] and the rsFC values for each connection within or between networks were summed and normalized based on the total number of possible connections for visualization purposes. rsFC values were compared between SLE-CI and SLE-NC using pairwise independent t-tests and corrected for multiple comparisons using the Benjamini–Hochberg false discovery rate (FDR) method at the ROI level [32]; connections that survived correction were designated connections of interest (COI).
The mean rsFC values of the COIs were compared groupwise between HC, SLE-NC and SLE-CI using a one-way analysis of variance with Bonferroni correction. This allowed comparisons of HC vs SLE-NC and HC vs SLE-CI with appropriate statistical rigour; the comparison of SLE-NC vs SLE-CI was ignored, as a significant difference between them is intrinsic to the input variable.
Analysis of BBB leakage in relation to brain networks
For associations between rsFC and BBB leakage, the mean rsFC values for the COIs were compared between SLE patients with and without extensive BBB leakage using unpaired samples t-tests with FDR correction. Mean COI rsFC was also correlated with %BBB leakage using Spearman’s rho. Then, %BBB leakage within each ROI was calculated using the same method used for the whole brain but restricted to voxels of each ROI.
We calculated the total number of brain wide connections between each ROI and every other ROI, termed degree [33], using the Brain Connectivity Toolbox for MATLAB [34]. Correlation of the degree of each ROI with its %BBB leakage value was calculated using Spearman’s rho to determine which areas of the brain were most affected by BBB dysfunction; P-values were corrected for multiple comparisons using FDR. For the ROIs in which degree was significantly associated with %BBB leakage, the average degree of affected regions was calculated and correlated with whole-brain BBB leakage using Spearman’s rho, then compared groupwise between HC, SLE-NC and SLE-CI using a one-way analysis of variance with Bonferroni correction. In all cases, corrected P < 0.05 was regarded as significant.
Causal mediation analysis
To determine whether alterations in brain network connectivity mediate the association between extensive BBB permeability and cognitive impairment among SLE patients, a causal mediation analysis was performed using the ‘medeff’ package in STATA-IC Version 14 (StataCorp, College Station, TX, USA). Only SLE patients with complete data for mean rsFC, BBB permeability and cognitive function were included. Of the 78 patients with rsfMRI scans, 17 were excluded due to missing DCE-MRI scans and one did not undergo cognitive testing, leaving 60 patients for this analysis. We fit regression models for the mediator and the outcome. The mediator model was a linear regression model for mean rsFC with BBB leakage (extensive vs non-extensive) as the independent variable. The outcome model was a logistic regression model for cognitive status (cognitive impairment vs NC) with BBB leakage and mean rsFC as independent variables. Both models were adjusted for patient age as a covariate.
Using these models and a standard mediation approach [35], we estimated the total effect of BBB permeability on cognitive status, the average causal mediated effect (ACME) through mean rsFC and the average direct effect (ADE) via pathways outside of the mediator of interest. Nonparametric bootstrapping (1000 replications) was used for the quasi-Bayesian approximation of 95% confidence intervals for the effect estimates. A 95% confidence interval for the ACME that did not cross zero was evidence that mean rsFC mediated the association between BBB permeability and cognitive status at P < 0.05. The proportion of the total effect mediated by mean rsFC was reported as a percentage.
A key assumption underlying causal mediation analysis is sequential ignorability, which implies that there is no unmeasured confounding of the exposure–mediator, exposure–outcome or mediator–outcome relationships. To evaluate the robustness of our results to violations of this assumption, a sensitivity analysis was performed using the ‘medsens’ package in STATA-IC. This sensitivity analysis is based on the correlation coefficient (ρ) between the error term in the mediator model and the error term in the outcome model. More extreme values of ρ represent larger departures from the sequential ignorability assumption and greater likelihood that unmeasured confounding will lead to a biased estimate of ACME. While the true value of ρ is unknown, it is possible to calculate the value of ρ for which the ACME would be zero: a large ρ at ACME = 0 indicates a robust result.
Results
Demographic and clinical characteristics
The 78 SLE patients and 71 controls for rsfMRI differed in the proportion of females (89.7% vs 69.0%; P = 0.002), mean (s.d.) age [49.4 (14.3) vs 38.9 (12.9) years; U = 3865, P < 0.001], race/ethnicity (91.0% vs 77.6% Caucasian; P = 0.040) and years of education [15.5 (3.6) vs.16.5 (2.9) years; U = 2100, P = 0.014]. The mean SLE disease duration was 15.8 (11.0) years. Cumulative ACR classification criteria for the 78 SLE patients were malar rash (47%), discoid rash (6.4%), photosensitivity (53.8%), oral/nasal ulcers (56.4%), serositis (34.6%), arthritis (78.2%), renal disorder (28.2%), neurological disorder (10.3%), haematological disorder (89.7%), immunological disorder (83.3%) and ANA (100%). Medication utilization and autoantibodies reflected that of a general lupus population [13] with low generalized disease activity and modest organ damage (Table 1).
Table 1.
Total SLE patients and comparison of those with and without cognitive impairment
| All SLE patients |
SLE and Cognitive imparment |
SLE and Normal cognition |
P | |
|---|---|---|---|---|
| (N = 78) | (N = 37) | (N = 40) | ||
| Female, N (%) | 70 (89.7) | 34 (91.9) | 36 (90.0) | 1.0 |
| Age (years), mean (s.d.) | 49.4 (14.3) | 50.2 (14.4) | 48.7 (14.4) | 0.658 |
| Race/ethnicity, N (%) | ||||
| Caucasian | 71 (91.0) | 31 (83.8) | 39 (97.5) | 0.051 |
| Other | 7 (9.0) | 6 (16.2) | 1 (2.5) | |
| Years of education, mean (s.d.) | 15.5 (3.6) | 14.7 (2.9) | 16.2 (4.0) | 0.093 |
| Current smokers, N (%) | 8 (10.3) | 6 (16.2) | 2 (5.0) | 0.144 |
| Ever smokers, N (%) | 29 (37.2) | 17 (45.9) | 11 (27.5) | 0.104 |
| SLE disease duration (years), mean (s.d.) | 15.8 (11.0) | 18.7 (11.8) | 12.8 (9.3) | 0.027 |
| Prior NP events (excluding cognitive impairment), N (%) | 46 (59.0) | 22 (59.5) | 24 (60.0) | 1.0 |
| Prior NP events attributed to SLE (excluding cognitive impairment), N (%) | 18 (23.1) | 11 (29.7) | 7 (17.5) | 0.282 |
| SLEDAI-2K score, median (IQR) | 2 (0-4) | 2 (1-4) | 2 (0-3.5) | 0.131 |
| SLEDAI-2K score without NP variables, median (IQR) | 2 (0-4) | 2 (1-4) | 2 (0-3.5) | 0.131 |
| SLICC/ACR Damage Index, median (IQR) | 1 (0-2) | 1 (0-2) | 0 (0-1) | 0.105 |
| SLICC/ACR Damage Index without NP variables, median (IQR) | 0 (0-2) | 1 (0-2) | 0 (0-1) | 0.180 |
| Medications, N (%) | ||||
| CS | 8 (10.3) | 5 (13.5) | 3 (7.5) | 0.470 |
| Antimalarials | 56 (71.8) | 23 (62.2) | 33 (82.5) | 0.072 |
| Immunosuppressive drugs | 36 (46.2) | 18 (48.6) | 17 (42.5) | 0.651 |
| ASA/clopidogrel | 9 (11.5) | 4 (10.8) | 5 (12.5) | 1.0 |
| Warfarin | 7 (9.0) | 5 (13.5) | 2 (5.0) | 0.251 |
| Psychoactive drugs | 33 (42.3) | 18 (48.6) | 15 (37.5) | 0.363 |
| Autoantibody positivity | ||||
| LA, N/N (%) | 17/76 (22.4) | 10/36 (27.8) | 7 (17.5) | 0.409 |
| aCL, N/N (%) | 9/76 (11.8) | 5/36 (13.9) | 4 (10.0) | 0.728 |
| Anti-β2 GPI, N (%) | 9 (11.7) | 5 (13.5) | 4 (10.0) | 0.731 |
| LA or aCL or anti-β2 GPI, N/N (%) | 23/76 (30.3) | 13/36 (36.1) | 10 (25.0) | 0.326 |
| Comorbidities, N (%): | ||||
| Hypertension | 11 (14.1) | 7 (18.9) | 4 (10.0) | 0.336 |
| Diabetes | 4 (5.1) | 3 (8.1) | 1 (2.5) | 0.346 |
P-values refer to the comparison between SLE patients with and without cognitive impairment. NP: neuropsychiatric; SLEDAI-2K: SLEDAI 2000; IQR: interquartile range; immunosuppressive drugs: MTX, AZA, CYC, LEF, MMF and i.v. gamma globulin; anti-β2 GPI: anti-β2 glycoprotein I. P-value is bolded as it is the only one that reaches statistical significance (P < 0.05).
Prior NP events (excluding cognitive impairment) from all causes occurred in 46/78 (59.0%) SLE patients and NP events attributable to SLE were present in 18/78 (23.1%). The latter included transient ischaemic attacks (N = 4), stroke (N = 3), cranial neuropathy (N = 4), seizure disorder (N = 4), acute confusional state (N = 4), psychosis (N = 1) and aseptic meningitis (N = 1). All NP events had resolved by the time of study. Impairment in one or more cognitive tests occurred in 37/77 (48%) patients and included deficits in information processing speed (9%), attention span (22%), new learning (6%), delayed recall (14%) and executive abilities (27%). Patients with cognitive impairment also had longer SLE disease duration (Table 1).
Brain networks in cognitively impaired and unimpaired SLE patients
Differences in rsFC between SLE-CI and SLE-NC are illustrated in Fig. 1A and summarized in supplementary Tables S1 and S2, available at Rheumatology online. The summed and normalized rsFC values by network are shown in Fig. 1B. Significant intra-network rsFC differences were found in the sensory, attention/executive and language/memory networks, while significant inter-network rsFC differences occurred primarily between other networks and the sensory and default-mode networks.
Fig. 1.
The association between rsFC and cognitive impairment in patients with SLE
(A) Contrast maps showing brain regions (coloured spheres) with significant reductions in functional connectivity (yellow lines) in SLE-CI compared with SLE-NC. Multiple comparisons were corrected using FDR at q < 0.05. Node colours represent membership in each of the five canonical resting-state networks (colour key legend at the left lower corner). Visualization created with BrainNet Viewer [36]. For more information on these regions and connections, see supplementary Tables 1 and 2, available at Rheumatology online, respectively. (B) Summed and normalized rsFC contrast matrix depicting the relative strengths of the total significant functional connections between networks (SLE-CI < SLE-NC). Colour heat map (right) represents the range of normalized sums; abbreviations represent resting-state networks as defined in the key below (A). (C) rsFC values of all significant connections identified in (A) were averaged across connections and compared between HC, SLE-NC and SLE-CI patients. Data are presented as a scatterplot over a boxplot; the mean is shown as a black line. rsFC: resting-state functional connectivity; SLE-CI: SLE patients with cognitive impairment; SLE-NC: SLE patients with normal cognition; FDR: false discovery rate.
Post hoc comparison showed the mean rsFC of the COIs in HC differed significantly from both SLE-CI and SLE-NC [F(2,145) = 42.1, P < 0.001, Fig. 1C]. In pairwise group comparisons, both SLE-NC and SLE-CI had significantly different mean rsFC values from HC. While mean rsFC values were negative for HC and SLE-NC, they were more frequently neutral or positive for SLE-CI.
The significant differences in rsFC between SLE patients and controls persisted after adjusting for age, sex, race/ethnicity and education using a one-way analysis of covariance [F(2,119) = 35.353, P ≤ 0.001]; pairwise tests of HC vs SLE-NC [F(1,83) = 13.175, P = 0.001] and HC vs SLE-CI [F(1,80) = 7.063, P = 0.028] remained significant. Disease duration was not a significant covariate of rsFC between SLE-CI and SLE-NC groups [F(1,74) = 1.935, P = 0.168].
BBB leakage and association with brain networks
On DCE-MRI scanning, 16/61 (26.2%) SLE patients had extensive BBB leakage. The mean COI rsFC showed significantly reduced negative FC in SLE patients with extensive BBB leakage compared with those without (Fig. 2A). There was also a significant correlation between extent of BBB leakage and mean rsFC in the total SLE group (Fig. 2B). Fig. 2C illustrates this association using sample BOLD signals for two ROIs in one HC subject and in four SLE patients with different BBB leakage levels.
Fig. 2.
Reduced negative rsFC in patients with SLE is associated with extensive BBB leakage
(A) In individual SLE patients, the mean rsFC was significantly higher in those with extensive BBB leakage. Data are presented as a scatter plot over a boxplot; the mean is shown as a black line. Normal vs extensive BBB leakage groups were defined using the median absolute deviation approach in Kamintsky et al. [11]. (B) In individual SLE patients, there was a significant association between the mean of all significant functional connections with %BBB disruption. (C) %BOLD signals (Z-scored) for two representative nodes are plotted for five different study participants in order of increasing %BBB disruption (#1 HC; #2 SLE-NC; #3-5 SLE-CI). The plots show a shift from negative (#1) or no connectivity (#2-3) to positive connectivity (#4-5) between the two nodes with increase in %BBB leakage in SLE. SMGp_L (orange line): left supramarginal gyrus posterior, part of the attention/executive network; MPFC_R (blue line): right medial prefrontal cortex, part of the default mode network; FC: functional connectivity; rsFC: resting-state functional connectivity; BBB: blood–brain barrier; BBBL: BBB leakage; BOLD: blood oxygen level dependent; HC: healthy control; SLE-CI: SLE patients with cognitive impairment; SLE-NC: SLE patients with normal cognition; FDR: false discovery rate.
The ROI-level BBB leakage analysis is depicted in Fig. 3. Brain regions with significant correlations between degree and %BBB leakage are shown in Fig. 3A: eight ROIs showed correlations that survived correction, of which seven had negative correlations (i.e. lower degree with greater BBB leakage) and one had a positive correlation. The mean degree for the seven negatively correlated ROIs showed a significant association with whole-brain %BBB leakage (Fig. 3B). Mean degree also differed across HC, SLE-NC and SLE-CI groups [F(2,123) = 5.401, P = 0.006; Fig. 3C], with brain-wide rsFC of the seven ROIs significantly lower in SLE-CI group compared with the SLE-NC and HC groups. Including the single positively-correlated region did not alter these findings, and repeating the analysis for the single positively correlated region alone returned no significant results.
Fig. 3.
The total brain-wide connectivity of brain regions (degree) is linked with %BBB leakage and cognitive impairment
(A) Nodes in brain regions with decreased (red) or increased (green) degree were correlated with %BBB leakage (Spearman ρ, corrected for multiple comparisons with FDR). (B) Higher %BBB leakage is associated with lower average degree in the negatively affected nodes. (C) The average degree of affected nodes is significantly lower in SLE-CI compared with SLE-NC and HC. Data visualized as violin plot overlaid box and scatter plots: red horizontal line, median; white horizontal line, mean. BBB: blood–brain barrier; FDR: false discovery rate; SLE-CI: SLE patients with cognitive impairment; SLE-NC: SLE patients with normal cognition; HC: healthy controls; FC: functional connectivity; ACCrp_R, right anterior cingulate cortex rostral posterior, part of the default mode network; Amyg_L, left amygdala, part of the subcortical network; INSm_L, left middle insula, part of the sensory network; MTGp_R, right middle temporal gyrus, posterior division, part of the language/memory network; pHippp_R, right parahippocampal gyrus, posterior division, part of the language/memory network; PostC_R, right post central gyrus (S1), part of the sensory network; STGp_L, left superior temporal gyrus, part of the language/memory network; vMPFC_R, right ventral medial prefrontal cortex, part of the default mode network.
Causal mediation analysis
Analysis of mean rsFC as a potential mediator of the relationship between BBB permeability and cognitive status is summarized in Fig. 4. As we reported previously [11], a statistically significant total effect of BBB permeability on the probability of cognitive impairment was found after adjusting for age (total effect = 0.352; 95% confidence interval: 0.079–0.571). The mediated effect by mean rsFC was also significant (ACME = 0.235; 95% confidence interval: 0.049–0.405), accounting for ∼64% of the total effect, while the direct effect of BBB permeability on cognitive status was not significant (ADE = 0.118; 95% confidence interval: –0.065 to 0.319). Sensitivity analysis found that ρ would need to exceed 0.7 for the ACME of mean rsFC to become zero or negative, indicating robust findings.
Fig. 4.
Association of BBB permeability with cognitive impairment in SLE, mediated by altered FC
Effect estimates (with 95% confidence intervals) represent the percentage-point change in the probability of cognitive impairment with extensive BBB leakage versus non-extensive BBB leakage via the total, mediated, or direct pathways. Approximately 64% of the total effect of BBB permeability on cognitive status is mediated by mean rsFC. Both the mediator and outcome models are adjusted for patient age in years. BBB: blood–brain barrier; FC: functional connectivity; rsFC: resting state functional connectivity; ACME: average causal mediated effect; ADE: average direct effect; CI: cognitive impairment.
Discussion
Nervous system disease in SLE is challenging due to heterogeneity of clinical manifestations, uncertain attribution of NP events to SLE or non-SLE causes, and low frequency of patients with more severe manifestations. Cognitive impairment is common, amenable to study and may be viewed as an indicator of overall brain health [2]. We report that cognitive impairment in SLE is associated with reduced negative rsFC in several resting-state networks in brain regions associated with sensory functions, memory and attention. Our findings suggest that the previously reported association between cognitive impairment and increased BBB permeability in SLE [11] may be attributed, in part, to changes in rsFC.
Brain injury in patients with NPSLE may occur through two pathways [37]. Ischaemic injury results from intravascular thrombosis mediated by aPL, endothelial activation and complement activation. Inflammation-induced injury is mediated by enhanced BBB permeability which allows circulating proteins, autoantibodies and immune cells to access brain tissue. These in turn bind to neuronal/glial cells or form immune complexes with downstream activation of astrocytes and plasmacytoid dendritic cells that trigger neuroinflammatory processes, and lead to neuronal-network reorganization (aberrant synaptogenesis/pruning). Both pathogenetic mechanisms can fluctuate in intensity over time in keeping with the evanescent clinical course of NPSLE.
Previous studies using rsfMRI identified altered FC in SLE patients affecting multiple networks [5–10, 38–42], albeit with inconsistent findings [9, 41, 42]. This may, in part, reflect highly variable methodologies for determining rsFC [9, 41, 42]. Most studies have found differences in FC between SLE patients and HC [5–8], but relatively few have reported an association with global SLE disease activity [5–7] or with cognitive impairment [9, 10]. In healthy participants the default-mode network (DMN) has negative rsFC with other networks [43]. In the current study and in others [6, 8], reduction in negative rsFC was found in SLE patients, indicating DMN dysfunction. The DMN is associated with self-referential thinking and memory recall, and DMN function is positively correlated with cognitive performance [44]. Thus, it is plausible that DMN dysfunction could directly relate to cognitive impairment; indeed, one study has suggested that altered FC may be an early indicator of cognitive impairment in SLE patients [38].
No previous studies of rsFC in SLE patients have included concurrent BBB imaging. Our study demonstrates reduced negative rsFC in SLE patients with cognitive impairment compared with those without, and in SLE patients with extensive BBB leakage compared with those with normal BBB function. Causal mediation analysis found that 64% of the total effect of BBB permeability on cognitive status was mediated by changes in rsFC. Note, however, that pairwise group comparisons found that SLE-NC had significantly more negative mean rsFC values compared with HC, while one of the eight significant correlations between degree and %BBB leakage was positive. This emphasizes the heterogeneous nature of NPSLE and potentially a more complex relationship between BBB permeability and cognition.
Although the clinical phenotype of cognitive impairment may look similar among SLE patients, there are likely to be different causal pathogenic mechanisms. Thus, the association of cognitive impairment with objectively defined abnormalities in BBB permeability and FC could identify a specific clinical subset. This use of advanced, multimodal neuroimaging may be one pathway to personalized or precision medicine for diagnosis and treatment of NPSLE. Such an approach has resulted in advances in other fields including prediction of treatment responses in individual patients with depression [45] and in fibromyalgia [46].
One strength of the current study is the use of an unselected sample of adult SLE patients, representative of clinic-attending lupus patients, albeit predominantly Caucasian, with stable and generally quiescent SLE. Additionally, all clinical, cognitive and neuroimaging assessments were performed on the same day, usually within 6 h of each other, thereby minimizing temporal dissociations between cognitive impairment and putative pathogenetic mechanisms detected by neuroimaging. Study limitations include a relatively small sample size that precluded identification of associations with subsets of cognitive impairment. To avoid excessive demands on study participants, cognitive testing was constrained, and neuroimaging did not include task-based FC evaluations. The lack of information on other cardiovascular risk factors in HC limited the ability to fully investigate their potential contribution to group differences in FC and BBB function. Finally, we cannot exclude the possibility of unmeasured confounds in causal mediation analysis, although our sensitivity analysis indicates a robust effect. Regardless, the reported associations will require independent confirmation and longitudinal study.
In summary, we have identified distinct changes in rsFC, in part associated with extensive BBB leakage, in SLE patients with cognitive impairment compared with SLE patients without cognitive impairment and HC. Future studies should establish how these clinical and neuroimaging abnormalities evolve over time and the association, if any, with inflammatory mediators and lupus autoantibodies. Our study, which combined multi-modal neuroimaging with comprehensive clinical evaluation of SLE patients, demonstrates potential to unravel pathogenetic mechanisms underlying cognitive impairment and other NP manifestations in SLE patients, thereby providing a pathway for the development and evaluation of novel therapeutic targets and strategies.
Supplementary Material
Acknowledgements
J.G.H. reports grants from Canadian Institutes of Health Research (CIHR, MOP-159703) during the conduct of the study. J.D.F. receives research grant support from the Canadian Institutes of Health Research, the National Multiple Sclerosis Society, the Multiple Sclerosis Society of Canada, Crohn’s and Colitis Canada, Research Nova Scotia, Nova Scotia Health Authority Research Fund; and consultation and distribution royalties from MAPI Research Trust. L.K. reports grants from MITACS, during the conduct of the study.
Funding: This study was supported by the Canadian Institutes of Health Research (CIHR, MOP-159703 and PJT148896, CIHR project grant 168878), Nova Scotia Health Research Foundation (NSHRF, MED-EST-2015–10067), Natural Sciences and Engineering Research Council of Canada (NSERC, RGPIN 04293–17 & RGPIN/05684–2016) and Brain Canada (BC PSG2015-3780), Mitacs grant IT23320 and Nova Scotia Research Fund. The funders of the study had no role in study design, patient recruitment, data collection, analysis, interpretation or publication.
Disclosure statement: S.D.B. reports grants from GE Healthcare, outside the submitted work. The other authors have declared no further conflicts of interest.
Contributor Information
John G Hanly, Division of Rheumatology, Department of Medicine and Department of Pathology, Queen Elizabeth II Health Sciences Center and Dalhousie University.
Jason W Robertson, Department of Anesthesia, Pain Management and Perioperative Medicine, Dalhousie University, Nova Scotia Health Authority.
Alexandra Legge, Division of Rheumatology, Department of Medicine, Queen Elizabeth II Health Sciences Center and Dalhousie University.
Lyna Kamintsky, Department of Medical Neuroscience, Dalhousie University, Halifax, Nova Scotia, Canada.
Guillermo Aristi, Department of Anesthesia, Pain Management and Perioperative Medicine, Dalhousie University, Nova Scotia Health Authority.
Alon Friedman, Department of Medical Neuroscience, Dalhousie University, Halifax, Nova Scotia, Canada; Departments of Cognitive and Brain Sciences, Physiology and Cell Biology, Ben-Gurion University of the Negev, Beer Sheva, Israel.
Steven D Beyea, Biomedical Translational Imaging Centre (BIOTIC), QEII Health Sciences Centre, and Department of Diagnostic Radiology.
John D Fisk, Nova Scotia Health Authority, Halifax, Canada and the Departments of Psychiatry, Psychology & Neuroscience and Medicine, Dalhousie University.
Antonina Omisade, Acquired Brain Injury (Epilepsy Program), Nova Scotia Health Authority.
Cynthia Calkin, Department of Psychiatry and Department of Medical Neuroscience.
Tim Bardouille, Department of Physics, Dalhousie University.
Chris Bowen, Biomedical Translational Imaging Centre (BIOTIC), QEII Health Sciences Centre, and Department of Diagnostic Radiology.
Kara Matheson, Research Methods Unit, Nova Scotia Health Authority, Halifax, Nova Scotia, Canada.
Javeria A Hashmi, Department of Anesthesia, Pain Management and Perioperative Medicine, Dalhousie University, Nova Scotia Health Authority.
Data availability statement
The data underlying this article cannot be shared publicly for the privacy of individuals who participated in the study. The data will be shared on reasonable request to the corresponding author.
Supplementary data
Supplementary data are available at Rheumatology online.
References
- 1. Govoni M, Hanly JG.. The management of neuropsychiatric lupus in the 21st century: still so many unmet needs? Rheumatology (Oxford) 2020;59:v52–62. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Rayes HA, Tani C, Kwan A. et al. What is the prevalence of cognitive impairment in lupus and which instruments are used to measure it? A systematic review and meta-analysis. Semin Arthritis Rheum 2018;48:240–55. [DOI] [PubMed] [Google Scholar]
- 3. Ceccarelli F, Perricone C, Pirone C. et al. Cognitive dysfunction improves in systemic lupus erythematosus: results of a 10 years prospective study. PLoS One 2018;13:e0196103. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Nystedt J, Nilsson M, Jönsen A. et al. Altered white matter microstructure in lupus patients: a diffusion tensor imaging study. Arthritis Res Ther 2018;20:21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Hou J, Lin Y, Zhang W. et al. Abnormalities of frontal-parietal resting-state functional connectivity are related to disease activity in patients with systemic lupus erythematosus. PLoS One 2013;8:e74530. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Lin Y, Zou QH, Wang J. et al. Localization of cerebral functional deficits in patients with non-neuropsychiatric systemic lupus erythematosus. Hum Brain Mapp 2011;32:1847–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Liu S, Cheng Y, Xie Z. et al. A conscious resting state fMRI study in SLE patients without major neuropsychiatric manifestations. Front Psychiatry 2018;9:677. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Yu H, Qiu X, Zhang YQ. et al. Abnormal amplitude of low frequency fluctuation and functional connectivity in non-neuropsychiatric systemic lupus erythematosus: a resting-state fMRI study. Neuroradiology 2019;61:331–40. [DOI] [PubMed] [Google Scholar]
- 9. Nystedt J, Mannfolk P, Jönsen A. et al. Functional connectivity changes in core resting state networks are associated with cognitive performance in systemic lupus erythematosus. J Comparat Neurol 2019;527:1837–56. [DOI] [PubMed] [Google Scholar]
- 10. Bonacchi R, Rocca MA, Ramirez GA. et al. Resting state network functional connectivity abnormalities in systemic lupus erythematosus: correlations with neuropsychiatric impairment. Mol Psychiatry 2021;26:3634–45. [DOI] [PubMed] [Google Scholar]
- 11. Kamintsky L, Beyea SD, Fisk JD. et al. Blood-brain barrier leakage in systemic lupus erythematosus is associated with gray matter loss and cognitive impairment. Ann Rheum Dis 2020;79:1580–7. [DOI] [PubMed] [Google Scholar]
- 12. Hochberg MC. Updating the American College of Rheumatology revised criteria for the classification of systemic lupus erythematosus. Arthritis Rheum 1997;40:1725. [DOI] [PubMed] [Google Scholar]
- 13. Hanly JG, Urowitz MB, Gordon C. et al. Neuropsychiatric events in systemic lupus erythematosus: a longitudinal analysis of outcomes in an international inception cohort using a multistate model approach. Ann Rheum Dis 2020;79:356–62. [DOI] [PubMed] [Google Scholar]
- 14. Gladman DD, Ibanez D, Urowitz MB.. Systemic lupus erythematosus disease activity index 2000. J Rheumatol 2002;29:288–91. [PubMed] [Google Scholar]
- 15. Gladman DD, Urowitz MB, Goldsmith CH. et al. The reliability of the Systemic Lupus International Collaborating Clinics/American College of Rheumatology Damage Index in patients with systemic lupus erythematosus. Arthritis Rheum 1997;40:809–13. [DOI] [PubMed] [Google Scholar]
- 16. The American College of Rheumatology nomenclature and case definitions for neuropsychiatric lupus syndromes. Arthritis Rheum 1999;42:599–608. [DOI] [PubMed] [Google Scholar]
- 17. Kozora E, Ellison MC, West S.. Reliability and validity of the proposed American College of Rheumatology neuropsychological battery for systemic lupus erythematosus. Arthritis Rheum 2004;51:810–8. [DOI] [PubMed] [Google Scholar]
- 18. Smith A. Symbol digit modalities test. 9th edn. Berlin, Germany: Western Psychological Services; 2002. [Google Scholar]
- 19. Walker LAS, Marino D, Berard JA. et al. Canadian normative data for minimal assessment of cognitive function in multiple sclerosis. Can J Neurol Sci 2017;44:547–55. [DOI] [PubMed] [Google Scholar]
- 20. Delis DJ, Kaplan E, Ober BA. California Verbal Learning Test Second Edition Adult Version Manual. California: Springer, 2000.
- 21. Veksler RV, Serlin Y, Prager O. et al. Slow blood-to-brain transport underlies enduring barrier dysfunction in American football players. Brain 2020;143:1826–42. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Kamintsky L, Cairns KA, Veksler R. et al. Blood-brain barrier imaging as a potential biomarker for bipolar disorder progression. NeuroImage Clinical 2020;26:102049. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Cox RW. AFNI: what a long strange trip it's been. Neuroimage 2012;62:743–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Fischl B. FreeSurfer. Neuroimage 2012;62:774–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Jenkinson M, Beckmann CF, Behrens TE, Woolrich MW, Smith SM.. FSL. Neuroimage 2012;62:782–90. [DOI] [PubMed] [Google Scholar]
- 26. Biswal BB, Mennes M, Zuo XN. et al. Toward discovery science of human brain function. Proc Natl Acad Sci USA 2010;107:4734–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Power JD, Barnes KA, Snyder AZ, Schlaggar BL, Petersen SE.. Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion. Neuroimage 2012;59:2142–54. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Mazziotta J, Toga A, Evans A. et al. A probabilistic atlas and reference system for the human brain: international Consortium for Brain Mapping (ICBM). Philos Trans R Soc Lond Ser B Biol Sci 2001;356:1293–322. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Saghayi M, Greenberg J, O’Grady C. et al. Brain network topology predicts participant adherence to mental training programs. Netw Neurosci 2020;4:528–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Hashmi JA, Kong J, Spaeth R. et al. Functional network architecture predicts psychologically mediated analgesia related to treatment in chronic knee pain patients. J Neurosci 2014;34:3924–36. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Hashmi JA, Loggia ML, Khan S. et al. Dexmedetomidine disrupts the local and global efficiencies of large-scale brain networks. Anesthesiology 2017;126:419–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Benjamini Y, Hochberg Y.. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B Stat Methodol 1995;57:289–300. [Google Scholar]
- 33. Liu J, Li M, Pan Y. et al. Complex brain network analysis and its applications to brain disorders: a survey. Complexity 2017;2017:1–27. [Google Scholar]
- 34. Rubinov M, Sporns O.. Complex network measures of brain connectivity: uses and interpretations. Neuroimage 2010;52:1059–69. [DOI] [PubMed] [Google Scholar]
- 35. Imai K, Keele L, Tingley D.. A general approach to causal mediation analysis. Psychol Methods 2010;15:309–34. [DOI] [PubMed] [Google Scholar]
- 36. Xia M, Wang J, He Y.. BrainNet Viewer: a network visualization tool for human brain connectomics. PLoS One 2013;8:e68910. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Hanly JG, Kozora E, Beyea SD, Birnbaum J.. Review: nervous system disease in systemic lupus erythematosus: current status and future directions. Arthritis Rheumatol 2019;71:33–42. [DOI] [PubMed] [Google Scholar]
- 38. Mackay M, Vo A, Tang CC. et al. Metabolic and microstructural alterations in the SLE brain correlate with cognitive impairment. JCI Insight 2019;4:1–18. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Niu C, Tan X, Liu X. et al. Cortical thickness reductions associate with abnormal resting-state functional connectivity in non-neuropsychiatric systemic lupus erythematosus. Brain Imaging Behav 2018;12:674–84. [DOI] [PubMed] [Google Scholar]
- 40. Zhang XD, Jiang XL, Cheng Z. et al. Decreased coupling between functional connectivity density and amplitude of low frequency fluctuation in non-neuropsychiatric systemic lupus erythematosus: a resting-stage functional MRI study. Mol Neurobiol 2017;54:5225–35. [DOI] [PubMed] [Google Scholar]
- 41. Cao ZY, Wang N, Jia JT. et al. Abnormal topological organization in systemic lupus erythematosus: a resting-state functional magnetic resonance imaging analysis. Brain Imaging Behav 2021;15:14–24. [DOI] [PubMed] [Google Scholar]
- 42. Liu S, Cheng Y, Zhao Y. et al. Clinical factors associated with brain volume reduction in systemic lupus erythematosus patients without major neuropsychiatric manifestations. Front Psychiatry 2018;9:8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43. Chai XJ, Castanon AN, Ongur D, Whitfield-Gabrieli S.. Anticorrelations in resting state networks without global signal regression. Neuroimage 2012;59:1420–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44. Mak LE, Minuzzi L, MacQueen G. et al. The default mode network in healthy individuals: a systematic review and meta-analysis. Brain Connect 2017;7:25–33. [DOI] [PubMed] [Google Scholar]
- 45. Cash RFH, Zalesky A, Thomson RH. et al. Subgenual functional connectivity predicts antidepressant treatment response to transcranial magnetic stimulation: independent validation and evaluation of personalization. Biol Psychiatry 2019;86:e5–7. [DOI] [PubMed] [Google Scholar]
- 46. Ichesco E, Peltier SJ, Mawla I. et al. Prediction of differential pharmacologic response in chronic pain using functional neuroimaging biomarkers and a support vector machine algorithm: an exploratory study. Arthritis Rheumatol 2021;73:2127–37. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data underlying this article cannot be shared publicly for the privacy of individuals who participated in the study. The data will be shared on reasonable request to the corresponding author.